Pro

ocee
edin AM 
ng off CIA
 
Confeerence on 
Indusstrial a
and App
plied M
Mathem
matics 
 
6th – 8thh July 201
10 
Institutt Teknolo ogi Bandu
ung 

Editorrs:
L.H. Wiry
yanto
S.R. Pudjjaprasetya

Faaculty of Math hematics and Natural Sciennces

Insstitut Teknoloogi Bandung
Jallan Ganesha 101 Bandung, Indonesia.
I
Ph
hone : +62 222 2502545
Faax : +62 222 250 6450 
IS
SSN: 977-2
208-70510-0
0-8
Proceeding of Conf. on Industrial and Appl. Math., Bandung‐Indonesia 2010 
 

Electronic Proceeding

Conference on Industrial and Applied Mathematics

6 - 8 July 2010 

The Committees of the conference 

Scientific Committee
Larry Forbes (University of Tasmania, Australia)
Robert McKibbin (Messey University, New Zealand)
Susumu Hara (Nagoya University, Japan)
Edy Soewono (ITB, Indonesia)
Chan basaruddin (University of Indonesia, Indonesia)
Roberd Saragih (ITB, Indonesia)

Organizing Committee
L.H. Wiryanto (Chair)
Sri Redjeki Pudjaprasetya
Novriana Sumarti
Andonowati
Kuntjoro Adji Sidarto

Technical Committee
Jalina Wijaya Maulana Wimar Banuardhi
Agus Yodi Gunawan Indriani Rustomo
Nuning Nuraini Pritta Etriana
Janson Naiborhu Adrianus Yosia
Adil Aulia Rafki Hidayat
Lina Anugerah Intan Hartri Putri
Ismi Ridha Yunan Pramesi Haris
Ikha Magdalena Freddy Susanto

i 
 
Proceeding of Conf. on Industrial and Appl. Math., Bandung‐Indonesia 2010 
 

Introduction
This proceeding contains papers which were presented in Conference on Industrial and
Applied Mathematics. The editors would like to express their deepest gratitude to all
presenters, contributors/authors and participants of this conference for their overwhelming
supports that turn this conference into a big success. While every single effort has been made
to ensure consistency of format and layout of the proceedings, the editors assume no
responsibility for spelling, grammatical and factual errors. Besides, all opinions expressed in
these papers are those of the authors and not of the conference Organizing Committee nor the
editors.

The Conference on Industrial and Applied Mathematics is the first international conference
held at Institut Teknologi Bandung-Indonesia, during July 6 – 8, 2010; hosted by Industrial
and Financial Mathematics Research Division, Faculty of Mathematics and Natural Sciences
ITB. The research division has continuing research interests in financial mathematics,
optimization, applied probability, control theory and its application; biological, physical
modeling and the application of mathematics in sciences, fluid dynamics, and numerical
methods and scientific computing. The conference provided a venue to exchange ideas in
those areas and any aspect of applied mathematics, in promoting both established and new
relationships.

Permission to make a digital or hardcopies of this proceeding for personal or classroom use is
granted without fee provided that copies are not made or distributed for profit or commercial
advantage.

Editors:
L.H. Wiryanto
S.R. Pudjaprasetya

Faculty of Mathematics and Natural Sciences

Institut Teknologi Bandung
Jalan Ganesha 10 Bandung, Indonesia.
Phone : +62 22 2502545
Fax : +62 22 250 6450

ii 
 
Proceeding of Conf. on Industrial and Appl. Math., Bandung‐Indonesia 2010 
 

Table of Content
Page

The committees of the conference i

Introduction ii

Table of Content iii

Research Articles:
1 An adaptive nonstationary control method and its application to 1-8
positioning control problems, Susumu Hara

2 Some aspects of modelling pollution transport in groundwater aquifers, 9-16

Robert McKibbin 

3 Jets and Bubbles in Fluids – Fluid Flows with Unstable Interfaces, Larry  17-25
K. Forbes 
 
4 Boundary Control of Hyperbolic Processes with Applications in Water 26-28
Flow, M. Herty and S. Veelken 

5  Isogeometric  methods  for  shape  modeling  and  numerical  simulation,  29-33

Bernard Mourrain, Gang Xu 
 
6  FOURTH‐ORDER  QSMSOR  ITERATIVE    METHOD  FOR  THE  SOLUTION  OF  34-39
ONE‐DIMENSIONAL  PARABOLIC  PDE’S,  J.  Sulaiman,  M.K.  Hasan,  M.  Othman, 
and S. A. Abdul Karim 
 
7  A  Parallel  Accelerated  Over‐Relaxation  Quarter‐Sweep  Point  Iterative  40-43
Algorithm  for  Solving  the  Poisson  Equation,  Mohamed  Othman,  Shukhrat  I. 
Rakhimov, Mohamed Suleiman and Jumat Sulaiman 
 
8  Value‐at‐Risk  (VaR)  using  ARMA(1,1)‐GARCH(1,1),  Sufianti and Ukur A. 44-49
Sembiring
 
9  Decline Curve Analysis in a Multiwell Reservoir System using State‐Space  50-53
Model, S. Wahyuningsih(1), S. Darwis(2), A.Y. Gunawan(3 ), A.K. Permadi 
 
10  Study  of  Role  of  Interferon‐Alpha  in  Immunotherapy  through  54-62
Mathematical Modelling, Mustafa Mamat, Edwin Setiawan Nugraha, Agus 
Kartono , W M Amir W Ahmad 
 
11  Improving  the  performance  of  the  Helmbold  universal  portfolio  with  an  63-66
unbounded learning parameter, Choon Peng Tan and Wei Xiang Lim 
 
12  Optimal  Design  The  Interval  Type‐2  Fuzzy  PI+PD  Controller  And  67-71

iii 
 
Proceeding of Conf. on Industrial and Appl. Math., Bandung‐Indonesia 2010 
 

Superconducting  Energy  Magnetic  Storage  (SMES)  For  Load  Frequency 

Control    Optimization  On  Two  Area  Power  System,  Muh Budi R Widodo, M 
Agus Pangestu H.W 
 
13  Dependence of biodegradability of xenobiotic polymers on population of  72-78
microorganism, Masaji Watanabe  and Fusako Kawai 
 
14  PROTOTYPE  OF  VISITOR  DISTRIBUTION  DETECTOR  FOR  COMMERCIAL  79-85
BUILDING, Sukarman,  Suharyanto, Samiadji Herdjunanto 
 
15  APPLICATION ANFIS FOR NOISE CANCELLATION, Sukarman  86-93
 
16  THE STABILITY OF THE MECD SCHEME FOR LARGE SYSTEM OF ORDINARY  94-99
DIFFERENTIAL EQUATIONS, Supriyono 
 
17  Real  Time  Performance  of  Fuzzy  PI+Fuzzy  PD  Self  Tuning  Regulator  In  100-103
Cascade  Control,  Mahardhika  Pratama,  Syamsul  Rajab,  Imam  Arifin, 
Moch.Rameli 
 
18  APPROXIMATION  OF  RUIN  PROBABILITY  FOR  INVESTMENT  WITH  104-108
INDEPENDENT  AND  IDENTICALLY  DISTRIBUTED  RANDOM  NET  RETURNS 
AND  MULTIVARIATE  NORMAL  MEAN  VARIANCE  MIXTURE  DISTRIBUTED 
FORCES  OF  INTEREST  IN  FIXED  PERIOD,  Ryan  Kurniawan  and  Ukur  Arianto 
Sembiring 
 
19  Stochastic  History  Matching  for  Composite  Reservoir,  Sutawanir  Darwis,  109‐115
Agus  Yodi  Gunawa,  Sri  Wahyuningsih,  Nurtiti  Sunusi,  Aceng  Komarudin 
Mutaqin, Nina Fitriyani 
 
20  PROBABILITY  ANALYSIS  OF  RAINY  EVENT  WITH  THE  WEIBULL  116‐120 
DISTRIBUTION AS A BASIC MANAGEMENT IN OIL PALM PLANTATION,  Divo
D. Silalahi
 
21  A Multi‐Scale Approach to the Flow Optimization of Systems Governed by  121‐126 
the Euler Equations,  Jean Medard T. Ngnotchouye, Michael Herty, and Mapundi 
K. Banda 
 
22  Modelling  and  Simulating  Multiphase  Drift‐flux  Flows  in  a  Networked  127‐133
Domain, Mapundi K. Banda, Michael Herty , and Jean Medard T. Ngnotchouye 
 
23  Calculating Area of Earth’s Surface Based on Discrete GPS Data, Alexander A  134‐137 
S Gunawan , Aripin Iskandar 
 
24  Study  on  Application  of  Machine  Vision  using  Least‐Mean‐Square  (LMS),  138‐144 
Hendro Nurhadi and Irhamah 
 
25  Cooperative Linear Quadratic Game for Descriptor System, Salmah  145‐250 
 

iv 
 
Proceeding of Conf. on Industrial and Appl. Math., Bandung‐Indonesia 2010 
 

26  ARMA Model Identification using Genetic Algorithm (An Application to Arc  151‐155 

Tube  Low  Power  Demand  Data),  Irhamah,  Dedy  Dwi  Prastyo  and  M.  Nasrul 
Rohman 
 
27  A  Particle  Swarm  Optimization  for  Employee  Placement  Problems  in  the  156‐161 
Competency  Based  Human  Resource  Management  System,  Joko  Siswanto 
and The Jin Ai 
 
28  Measuring Similarity between Wavelet Function and Transient in a Signal  162‐166 
with Symmetric Distance Coefficient, Nemuel Daniel Pah 
 
29  An  Implementation  of  Investment  Analysis  using  Fuzzy  Mathematics,  167‐169
Novriana Sumarti and Qino Danny 
 
30  Simulation  of  Susceptible  Areas  to  the  Impact  of  Storm  Tide  Flooding  170‐178 
along Northern Coasts of Java, Nining Sari Ningsih, Safwan Hadi, Dwi F. Saputri , 
Farrah Hanifah, and Amanda P.Rudiawan 
 
31  Fuzzy  Finite  Difference  on  Calculation  of  an  Individual’  Bank  Deposits,  179‐183 
Novriana Sumarti and Siti Mardiah 
 
32  An Implementation of Fuzzy Linear System in Economics, Novriana Sumarti  184‐187
and Cucu Sukaenah 
 
33  Compact  Finite  Difference  Method  for  Solving  Discrete    Boltzmann  188-193
Equation, PRANOWO, A. GATOT BINTORO 
 
34  Natural convection heat transfer with an Al2O3 nanofluids at low Rayleigh  194-199
number, Zailan Siri, Ishak Hashim and  Rozaini Roslan 
 
35  Optimization model for estimating productivity growth in Malaysian food  200-206
manufacturing industry, Nordin Hj. Mohamad, and Fatimah Said 
 
36  Numerical study of natural convection in a porous cavity with transverse  207‐211 
magnetic  field  and  non‐uniform  internal  heating,  Habibis Saleh, Ishak
Hashim and Rozaini Roslan
 
37  THE  DISTRIBUTION  PATTERN  AND  ABUNDANCE  OF  ASTEROID  AND  212‐215 
ECHINOID AT RINGGUNG WATERS SOUTH LAMPUNG,  Arwinsyah Arka, Agus 
Purwoko, Oktavia 
 
38  Low  biomass  of  macrobenthic  fauna  at  a  tropical  mudflat:  an  effect  of  216‐224 
latitude?, Agus Purwoko and Wim J. Wolff 
 
39  Density  and  biomass  of  the  macrobenthic  fauna  of  the  intertidal  area  in  225‐234 
Sembilang national park, South Sumatra, Indonesia,  Agus Purwoko and Wim 
J. Wolff 
 

v 
 
Proceeding of Conf. on Industrial and Appl. Math., Bandung‐Indonesia 2010 
 

40  Intelligent traffic light system for AMJ highway, Nur Ilyana Anwar Apandi,  235‐238 

Puteri Nurul Fareha M. Ahmad Mokhtar, Nur Hazahsha Shamsudin and Anis 
Niza Ramani and Mohd Safirin Karis 
 
41  Goodness  of  Fit  Test  for  Gumbel  Distribution  Based  on  Kullback‐Leibler  239‐245 
Information using Several Different Estimators,  S. A. Al‐Subh , K. Ibrahim , M. 
T. Alodat , A. A. Jemain 
 
42  Impact  of  shrimp  pond  development  on  biomass  of  intertidal  246‐256
macrobenthic fauna: a case study at Sembilang, South Sumatra, Indonesia, 
Agus Purwoko, Arwinsyah Arka and Wim J. Wolff 
 

vi 
 
 An adaptive nonstationary control method and
its application to positioning control problems

Susumu Hara(1)
(1)
Department of Mechanical Science and Engineering, Graduate School of Engineering,
Nagoya University, Nagoya, Japan
(Email: haras@mech.nagoya-u.ac.jp)

ABSTRACT problem based on the vibration system model of a

This plenary talk discusses an adaptive nonstationary controlled object and produces superior transient
control (ANSC) method based on the progressive time performance and vibrationless trajectories by a
calculations of time-varying Riccati equations and simple single-degree-of-freedom control system
shows its application to positioning control problems.
structure. Moreover, NOR does not need the
Some nonstationary control methods generate useful
reference design and generates useful trajectories
motion trajectories. Moreover, ANSC is a practical
control method for actual positioning problems. by the feedback control algorithm. However,
Concretely, this talk selects a positioning problem of a tuning of time-varying weights of LQ criterion
flexible structure installed on X-Y table and a mode function depends on trials and errors in many NOR
switching control problem of a power assist cart. The designs. Then, some controller designers are
effectiveness of the ANSC method is verified via the compelled to solve inefficient controller design
two applications. problems. Hara proposed a semiautomatic NOR
design method based on GA meta-optimization [5].
KEYWORDS But, generally it takes much time to obtain the
Adaptive nonstationary control; motion control; optimized controller. These inefficiencies are
positioning; power assist; time-varying Riccati equation; caused by the difference of calculation time
vibration control. directions of controller design and response
simulation. In NOR designs, at first, a time-varying
I. INTRODUCTION Riccati equation is solved by reversed time
Positioning is one of the most important calculations –(i). Secondly, the responses of a
mechanical systems control tasks. It is very controlled object are solved by progressive time
important for conveyance devices, robots, calculations –(ii). If the responses in (ii) do not
information precision machinery, and so on. satisfy some specifications, we tune the time-
Recently, many mechanical structures have varying weights again and repeat (i). Hence, both
become light and/or large scale structures due to (i) and (ii) cannot be carried out simultaneously.
the expansion of their applications. Recently, the author has proposed an efficient time-
Simultaneously, control performances such as varying gain type controller design method using
traveling period (seeking time) and settling the solutions of time-varying Riccati equations. In
accuracy have to satisfy hard specifications. this method, we require the progressive time
Therefore, we cannot ignore the influences of calculations only [6], [7]. Both the time-varying
vibration modes of controlled objects in many Riccati equation and the responses of a controlled
problems. From such a point of view, positioning object are calculated simultaneously by the
control problems for vibration systems have been progressive time calculations. This method is
increased. different from the optimal control method based
on LQ criterion function. However, the responses
As one of effective control methods for positioning in this method become similar to the NOR case
of vibration systems, “Nonstationary Optimal responses. Moreover, the progressive time
Regulator (NOR)” is well-known [1]–[4]. NOR is calculations of the Riccati equation enable us to
obtained by the finite-horizon LQ optimal regulator
simplify the weight tuning [6]. In [6] and [7], the vibration characteristics of the controlled object.
combination of the above method and a random NOR means the LQ optimal regulator based on
search technique for the weight tuning was applied the following finite-horizon (time: t ∈ ⎡⎣0, t f ⎤⎦ )
to a positioning control problem of a vibration criterion function:
system.
tf
J = ∫ ⎡⎣ x T ( t ) Q ( t ) x ( t ) + r ( t ) u 2 ( t ) ⎤⎦dt . (2)
The progressive time calculations of the Riccati 0

equations mean that the time-varying feedback

gains are obtained by real-time computations in The optimal control input is obtained by the
actual implementations. This feature is convenient following state feedback formula:
for the realization of some kind of adaptive control.
Therefore, this plenary talk discusses an adaptive u ( t ) = − fb ( t ) x ( t ) , fb ( t ) = r −1 ( t ) bT ( t ) P ( t ) (3)
nonstationary control (ANSC) method based on
the feature of the author’s previous study and where f b ( t ) are time-varying feedback gains.
shows its application to positioning control P ( t ) are the solutions of the time-varying Riccati
problems. At first, we select a flexible structure equation:
installed on an X-Y table as a controlled object
example. The ANSC method in this talk offers − P ( t ) = P ( t ) A ( t ) + AT ( t ) P ( t )
single nonstationary controller design realizing the − P ( t ) b ( t ) r −1 ( t ) bT ( t ) P ( t ) + Q ( t ) . (4)
positioning motion, the estimation of the natural
frequency during access motion, and the residual
In NOR designs, at first, the time-varying Riccati
vibration reduction (settling) using the estimated
equation (4) is solved by reversed time ( t f → 0 )
information. Moreover, not only the vibration
calculations under appropriate time-varying
system application but also another ANSC
weights Q ( t ) , r ( t ) and the boundary condition
application to the mode switching control of a
(e.g., P ( t f ) = 0 ) –(i). Secondly, the responses of
power assist cart is shown. In this system, the
the controlled object are solved by progressive time
control mode switches smoothly from the ANSC
based access control to manual positioning with
( 0 → t f ) calculations –(ii). If the responses in (ii)
do not satisfy some specifications, we tune the
nonstationary impedance control (nonstationary
time-varying weights again and repeat (i). NOR
power assist control). The effectiveness of the
produces superior transient performance and
ANSC method is verified via the two applications.
vibrationless trajectories by a simple single-degree-
of-freedom control system structure. NOR does not
The contents of this paper are arranged from the
need the reference design as general servo systems
author’s original papers [8] and [9] for the plenary
and generates useful trajectories by the feedback
talk in CIAM2010, Bandung, Indonesia.
control algorithm.
II. ADAPTIVE NONSTATIONARY
However, both the calculations of (i) and (ii) cannot
CONTROL METHOD
be carried out simultaneously. This point makes
NOR design complicated. Especially, tuning of the
A. Nonstationary Control Method Using the
time-varying weights takes much time in many
Solutions of Time-Varying Riccati Equations
NOR designs. From such a point of view, the author
This subsection summarizes “Nonstationary
proposed an efficient time-varying gain type
Optimal Regulator (NOR)” [1]–[4] and the
controller design method using the solutions of
author’s previous method [6], [7]. Here, we
novel time-varying Riccati equations [6], [7]. In
consider that a controlled object is described as
this method, the following time-varying Riccati
the following single-input time-varying system
equation is utilized instead of (4):
state equation:

x (t ) = A (t ) x (t ) + b (t ) u (t ) dP ′ (τ )
(1) = P ′ (τ ) A (τ ) + AT (τ ) P ′ (τ )
dτ
where x ( t ) and u ( t ) are state variables and control − P ′ (τ ) b (τ ) r ′−1 (τ ) bT (τ ) P ′ (τ ) + Q ′ (τ ) (5)
input, respectively. This model includes the
where Q ′ (τ ) = Q ( t = τ ) and r ′ (τ ) = r ( t = τ ) . The f bd′ ( n ) = ⎡⎣ rd′ ( n ) + bdT ( n ) Pd′ ( n ) bd ( n ) ⎤⎦
−1

time-varying weights are given to (5) following

time progress. The control input is obtained by ibdT ( n ) Pd′ ( n ) Ad ( n ) . (8)

u ( t ) = − fb′ ( t ) x ( t ) C. Adaptive Nonstationary Control (ANSC)

fb′ ( t ) = r ′−1 (τ = t ) bT (τ = t ) P ′ (τ = t ) (6) Method
The use of (7) and (8) implies that the time-varying
where f b′ ( t ) are time-varying feedback gains. The feedback gains can be obtained by real-time
combination of (1) and (5) requires the progressive computations in actual implementations. This
time calculations only. Then, both the time-varying feature is convenient for the realization of some
Riccati equation and the responses of the controlled kind of adaptive control. We assume that the
object are calculated simultaneously. The parameters possessing some information of the
responses in this method become similar to the controlled object can be detected on-line/real-time.
responses of NOR case. This strategy is useful for The real-time identified controlled object model
the weight tuning and realizes efficient controller is described as follows:
design in comparison with NOR. The theory of
this method is quite different from NOR based on xm ( t ) = Am ( t ) xm ( t ) + bm ( t ) um ( t ) . (9)
the criterion function (2). However, it is important
for controller designers to obtain useful trajectories The Riccati equation corresponding to the model
similar to the optimal control results easily in many (9) is
actual applications. Generally, we cannot discuss
the closed loop stability and robustness issues in ′ ( n + 1) = Amd
Pmd T
( n ) Pmd′ ( n ) Amd ( n )
this method by conventional control theories. − AmdT
( n ) Pmd′ ( n ) bmd ( n )
However, the controller designs are settled through −1
′ ( n ) + bmd
i ⎡⎣ rmd T
( n ) Pmd′ ( n ) bmd ( n )⎤⎦
many time-historical simulations. Therefore, these
points are checked by the response calculations
T
ibmd ( n) Pmd′ ( n) Amd ( n ) + Qmd′ ( n ) (10)
[7]. In the author’s previous papers, the above
method with a random search technique for the where Amd ( n ) and bmd ( n ) are discrete-time
weight tuning was applied to a positioning control matrices of Am ( t ) and bm ( t ) by the sampling
problem of a vibration system [6], [7]. period of actual implementation, respectively. The
control input is obtained by
B. Use of Discrete-Time Riccati Equations
In order to reduce load of computers, this
subsection introduces the use of the following um ( t ) = − fbmd
′ ( n ) xm ( t )
−1
discrete-time Riccati equation instead of (5): ′ ( n ) = ⎡⎣ rmd
f bmd ′ ( n ) + bmd
T
( n ) Pmd′ ( n ) bmd ( n )⎤⎦
Pd′ ( n + 1) = AdT ( n ) Pd′ ( n ) Ad ( n ) − AdT ( n ) Pd′ ( n ) bd ( n )
T
ibmd ( n ) Pmd′ ( n ) Amd ( n ) . (11)
−1
i ⎡⎣ rd′ ( n ) + bdT ( n ) Pd′ ( n ) bd ( n ) ⎤⎦ bdT ( n ) Pd′ ( n ) Ad ( n ) The method utilizing (9)–(11) is termed “Adaptive
+ Qd′ ( n ) (7) Nonstationary Control method (ANSC method)”
in this paper.
where Ad ( n ) and bd ( n ) are discrete-time matrices
of A ( t ) and b ( t ) by the sampling period of actual III. CONTROLLED OBJECT
implementation, respectively. n ( n = 0, 1, 2, ) In this paper, we select a controlled object as in
means the sample number of discrete-time. There Fig. 1 as an example. This object consists of an X-
are some equations including both the continuous- Y table and a flexible structure installed on the
time t and the discrete-time sample n in this paper, table. The table possesses the driving system
for convenience. Let us consider that these n consisting of DC servomotors, reduction gears, and
correspond to appropriate t. The control input is timing belts. The motion of single axis only (X-
obtained as follows: axis in Fig. 1) is considered in this paper. When
the table is settled in some point, the natural
u ( t ) = − fbd′ ( n ) x ( t ) frequency of the flexible structure is 4.55 Hz. The
 Acceleration pick-up vibrating part and the base part as shown in Fig. 2.
Aluminum plate
(300×300×10)
The state equation for this model is as follows [2],
Vibration axis [4]:
Aluminum rigid
plate
x p ( t ) = Ap x p ( t ) + bp u p ( t ) + d p f r (12)

780
Aluminum flexible Gap sensor
T
x p ( t ) = ⎡⎣ xs ( t ) xs ( t ) θ m ( t ) θ m ( t ) i ( t ) ⎤⎦
plate (800×50×5)
1270
0 0
127 100
1000
where xs ( t ) , θ m ( t ) , and i ( t ) are the absolute
Encoder (X) displacement of the tip of the vibrating part [m],
Moving axis the angular displacement of the motor [rad], and
X-axis
Encoder (Y)
Timing belt Reduction gears the current of the motor [A], respectively. The
Y-axis DC servomotor with
an encoder (X)
control input u p ( t ) is the motor voltage e ( t ) [V].
DC servomotor with
an encoder (Y)
(Unit: mm)
In (12), the influence of Coulomb friction in the
driving system is taken into account by using sign
Figure 1. Controlled object.
functions in the same manner as in [2], [4].
ks Vibrating part
Base part cs ms If the influences of the motor’s inductance and the
Reduction θb(t) Timing belt
gears
fr mb Coulomb friction are sufficiently small, (12) can
F(t)
r be approximated by the following equation [3]:
rb, Jb xb(t) Pulley
rm, Jm xs(t)
xr(t)≡xs(t)-xb(t) x p 2 ( t ) = Ap 2 x p 2 ( t ) + bp 2 u p 2 ( t ) (13)
DC servomotor: L, R, KT, KE, B, θm(t), i(t), e(t) T
x p 2 ( t ) = ⎡⎣ xs ( t ) xs ( t ) θ m ( t ) θ m ( t ) ⎤⎦
Figure 2. Controlled object model.
u p2 (t ) = e (t ) .
Table 1. Parameters of the controlled object model.
ms Equivalent mass of the vibrating part 4.50 kg
ks Equivalent spring constant of the vibrating part 3.68×103 N/m Generally, the influences of dampings are small in
cs Equivalent damping coefficient of the vibrating pt. 1.28 N･s/m controlled objects like the object in Fig. 1.
mb Equivalent mass of the base part 22.1 kg Moreover, if we neglect the force from the
Jb Moment of inertia (Base side gear and the pulley) 3.57×10-5 kg･m2 vibrating part to the base part, more simple model
Jm Moment of inertia (Motor and the motor side gear) 8.33×10-4 kg･m2 is obtained as follows:
r Radius of the pulley 0.0159 m
α Gear ratio (= rb/rm) 11.0 x p 3 ( t ) = Ap 3 x p 3 ( t ) + bp 3u p 3 ( t ) (14)
T
L Inductance of the motor 2.0 mH x p 3 ( t ) = ⎡⎣ xs ( t ) xs ( t ) θ m ( t ) θ m ( t ) ⎤⎦
R Resistance of the motor 1.0 Ω
u p3 ( t ) = e ( t )
KT Torque constant of the motor 0.215 N･m/A
KE Back electromotive force constant of the motor 0.215 V･s/rad ⎡ 0 1 0 0 ⎤
⎢ ⎥ ⎡ 0 ⎤
B Braking constant of the motor 1.69×10-4 N･m･s/rad
⎢ −ωs2 r ⎥ ⎢ 0 ⎥
fr 0 ωs2 0
Coulomb friction force of the driving system 131.4 N ⎢ α ⎥ ⎢ ⎥
Ap 3 = ⎢ ⎥ b = ⎢ 0 ⎥
0 0 0 1 p 3
⎢ ⎥
parameters of the controlled object are summarized ⎢ ⎥, α K
⎢ α KT K E ⎥ ⎢ T ⎥
in Table 1. In experiments, the angular ⎢ 0 0 0 − ⎢⎣ rI l R ⎥⎦
displacement of the motor is detected by an ⎢⎣ rI l R ⎥⎥⎦
encoder. The vibration status of the flexible
structure is detected by both an acceleration pickup ks
installed on the structure and a gap sensor installed where ωs and Il are ωs = and
ms
on the table. Control computations in experiments r α 1
are carried out by a digital signal processor (DSP, I l ≡ mb + 2Jm + Jb , respectively..
α r αr
TMS320C31). Its sampling period is 1.0 ms.
IV. NUMERICAL CALCULATIONS AND
If the flexible structure can be modeled by a single- EXPERIMENTS
degree-of-freedom vibration system, we can obtain
the controlled object model consisting of the A. The Control Problem in This Section and the
Feature of the ANSC Method the residual vibration reduction (settling) using the
In this section, we discuss the positioning and estimated information. From such a point of view,
residual vibration reduction of the flexible the ANSC method is a nonstationary controller
structure in Fig. 1. The vibration during access design method possessing some adaptability.
motion is permitted in this section. Then, we can
estimate parameters on the vibration characteristic B. Numerical Calculations
during access motion and use the estimated data This subsection mentions numerical calculation
to reduce the residual vibration. In this section, a examples in [8]. Let the access distance set 0.1 m
very simple method for the natural frequency under the access period is about 2 s. The models
estimation is applied. As mentioned in the previous (14) and (12) are applied to the controller design
section, the influence of damping of the flexible and the response calculations, respectively. We
structure can be neglected in this section. Then, prepare two kinds of weights – the time-varying
the natural frequency is estimated by the “peak- weights on the angular displacement of the motor
(for positioning) ⎡⎣ q pos ( t ) ⎤⎦ and on the relative
2
to-peak” period measurement of the tip
acceleration as in Fig. 3. If the ANSC method is velocity of the tip of the structure (for residual
vibration reduction) ⎡⎣ qvib ( t ) ⎤⎦ . These weights are
2
applied to the model (14) without the weights on
vibration control, the controller does not reduce tuned by taking account of the above problem
vibration. When the weights increase, the vibration setting and the constraints of actual experimental
reduction is begun as in Fig. 4. Therefore, we can system. As a result, Qmd ′ ( t ) , q pos ( t ) , and qvib ( t )
estimate the natural frequency exactly when the are set as follows:
weights are zeros. Different from the ANSC
′ ( t ) = Cobj
Qmd T
( t ) Cobj ( t )
method, the conventional NOR produces some (15)
nonzero feedback gains and reduces vibration even ⎡0 0 q pos ( t ) 0 ⎤
C obj ( t ) = ⎢ ⎥
if the weights become zeros as in Fig. 4 [2]. ⎣0 qvib ( t ) 0 − ( r / α ) qvib ( t ) ⎦
Therefore, NOR requires the exact controlled 1
q pos ( t ) = ( r / α ) ⎡⎣1.0e5 + 2.0e7/ {1.0 + exp ( − 10.0( t − 2.0) )}⎤⎦ 2
object model a priori. However, the ANSC method 1
offers single nonstationary controller design qvib ( t ) = ⎡⎣5.0e3 / {1.0 + exp ( −10.0(t − 1.8) )}⎤⎦ 2 .
realizing the positioning motion, the natural
frequency estimation during access motion, and The weight on the control input rmd′ ( t ) is set to 1.0
′ ( t ) = 1.0 ⎤⎦ in this section. q pos ( t ) and qvib ( t )
⎡⎣ rmd
are shown in Fig. 5 (a) and (b), respectively.

The time-varying feedback gains obtained by these

weights are shown in Fig. 5 (c)–(f). Here, we
calculate (10) under Pmd′ ( 0 ) = 0 . The gains for the
motor variables (Fig. 5 (e) and (f)) increases after
0 s by following the weight q pos ( t ) (Fig. 5 (a)).
The access motion is realized by these gains. On
Figure 3. Estimation of the natural frequency. the other hand, the gains for vibration reduction
(Fig. 5 (c) and (d)) are almost zeros until about
ethod

1.3-1.5 s. Then, the vibration reduction is not

lm

carried out during access motion. The estimation

method
Feedback gain

ntiona

stops when the table passes 1.0 cm before the

Conve

objective settling point. We have to leave the

Proposed

Weight q(t)
Gain is calculation responses out for want of space. The
not 0.
readers can see the results in [8]. The responses
are similar to the experimental results in the
Gain is 0. tf
0 following subsection.
Time t [s]
: Initial point : Terminal point

Figure 4. Comparison of feedback gains (proposed ANSC C. Experiments

method and conventional nonstationary optimal regulator). Experimental result of the ANSC method
 8 80 40

the angular displacement of the motor 7 70 35

the relative velocity of the tip

Square root of the weight on

Square root of the weight on

6 60 30

Natural frequency [Hz]

5 50 25
4 40 20
3 30 15
2 20 10
1 10 5
0 0 0

-1 -10 -5
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s] Time [s]
(a) Weight on the ang. disp. of the motor (b) Weight on the relative velocity of the tip
Figure 7. Natural frequency estimation (experiment).
Feedback gain on the displacement of the tip

500 5

Feedback gain on the velocity of the tip

0 0

-500 -5
includes the result of the case that the nonstationary
-1000 -10
controller based on the time-invariant model whose
-1500 -15
natural frequency is 2.0 Hz is applied (thin lines).
-2000 -20 In the experiments, taking the influence of noises
-2500
0 0.5 1 1.5 2 2.5 3 3.5 4
-25
0 0.5 1 1.5 2 2.5 3 3.5 4
on the acceleration pickup into account, the natural
(c) Feedback gain on the disp. of the tip
Time [s]
(d) Feedback gain on the velocity of the tip
Time [s]
frequency is estimated by the acceleration data
12 0.3 sampled every 15.0 ms. The estimated result is
10 0.25
indicated in Fig. 7. Discrete-time matrices Amd ( n )
Feedback gain on the angular
Feedback gain on the angular
displacement of the motor

and bmd ( n ) are obtained by using the zero-order

velocity of the motor

8 0.2

6 0.15

4 0.1 holder and Taylor approximation of matrix

2 0.05 exponential function (terms whose orders are
0 0
higher than the square of sampling period are
-2 -0.05
0 0.5 1 1.5 2
Time [s]
2.5 3 3.5 4 0 0.5 1 1.5 2
Time [s]
2.5 3 3.5 4
truncated.).
(e) Feedback gain on the ang. disp. of the motor (f) Feedback gain on the ang. velo. of the motor

Figure 5. Time-varying weights and feedback gains. From the comparison of both the results in Fig. 6,
0.12 0.12
we can understand that the ANSC method reduces
0.1 0.1 the residual vibration effectively. Therefore, the
Displacement of the base [m]
Displacement of the tip [m]

0.08 0.08 effectiveness of ANSC is verified experimentally.

0.06 0.06

V. APPLICATION TO MODE SWITCHING

0.04 0.04

0.02 0.02

0 0 CONTROL OF POWER ASSIST CART

-0.02
0 0.5 1 1.5 2 2.5 3 3.5 4
-0.02
0 0.5 1 1.5 2 2.5 3 3.5 4 (ANSC-SERVO CONTROL CASE)
Time [s] Time [s]
(a) Displacement of the tip (b) Displacement of the base In spite of popularization of industrial robots, there
1
0.8
0.1
are some processes which require operator’s
actions instead of the uses of autonomous robots.
Acceleration of the tip [m/s ]
2

0.6
Velocity of the base [m/s]

0.4
0.2
0
For example, many operators handle and position
heavy objects in automotive assembly processes
0
-0.2
-0.1
-0.4
-0.6
[10]. For that reason, various power assist systems
-0.8
-1 -0.2
based on the impedance control have been
developed.
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
Time [s] Time [s]
(c) Acceleration of the tip (d) Velocity of the base
3 10

2 If we consider many applications of power assist

Current of the motor [A]

systems, the positioning objects have been traveled

1 0
Input voltage [V]

-1 -10 to around the power assist system by a conveyance

-2

-3 -20
system as shown in the operation type - I in Fig. 8.
-4 However, in order to improve the efficiency of
-5
0 0.5 1 1.5 2
Time [s]
2.5 3 3.5 4
-30
0 0.5 1 1.5 2
Time [s]
2.5 3 3.5 4 operation and reduce the physical burden of
(e) Current of the motor (f) Input voltage operators, we should pay attention to the
Figure 6. Experimental results cooperation of both the conveyance and the power
(bold: ANSC method; thin: the controller based assisting functions. It is ideal that both the
on a time-invariant natural frequency is used.). automatic conveyance and the power assist
corresponding to the numerical calculation is processes are realized by single actuator and the
shown in Fig. 6 (bold lines). As a reference, Fig. 6 novel controller for the smooth switching of both
 Operation type - I
Operator(s) Taking that situation into account, the author
Conveyance system
studied on the positioning of a cart (Fig. 9) by
(Servo control) Object(s)
means of a smooth switching from the servo control
Object(s) mode (SCM) to the impedance control mode (ICM)
in [9]. The smooth switching in [9] means the
Floor Power-assist system
continuous switching of the control inputs of plural
(Impedance control) control modes. For the purpose of the smooth
Operation type - II switching, the braking mode (BM) was inserted
Operator between SCM and ICM. Both SCM and BM were
Object(s) Actuator

Weight on the servo variable (x 1.0e5)

Weight on the velocity of the cart

12 1400
10 1200
8 1000
Servo control Braking mode Impedance 800
mode control mode 6
600
4
Figure 8. Conveyance and power assist processes. 2
400
200
0 0
-2 -200
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time [s] Time [s]
(a) Weight on the servo variable qs(t) (b) Weight on the velocity qb(t)

Feedback gain on the velocity of the cart

Feedback gain on the servo variable
200 250
0
200
-200
-400 150

-600 100
-800
50
-1000
-1200 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time [s] Time [s]
(c) Feedback gain for the servo variable (d) Feedback gain for the velocity of the cart

Figure 10. Time-varying weights and feedback gains

for SCM and BM.
2.5 0.4
Displacement of the cart [m]

Velocity of the cart [m/s] 0.3

2
Handle 0.2
Weight
Force sensor 1.5 0.1
0
1
-0.1
363

0.5 -0.2
Acceleration pickup -0.3
Auxiliary wheel 0
-0.4
-0.5 -0.5
0 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
60 Time [s] Time [s]
Gears (a) Displacement of the cart (b) Velocity of the cart (dashed line: reference)
300
1 30
0.8
Current of the motor [A]

Encoder 20
Servomotor Driving wheel 0.6
Input voltage [V]

0.4 10
0.2 0
mn Mass of the model 20.0 kg (Nominal) - 50.0 kg (Max) 0
-10
Jm Moment of inertia of the motor 2.13 X 10-5 kg・m2 -0.2
-0.4 -20
Jg Moment of inertia of the motor side gear 8.47 X 10-5 kg・m2 -0.6
-30
Jw -0.8
Moment of inertia of the wheels 4.56 X 10-4 kg・m2
-1 -40
α = rr/rm Gear ratio 15.0 0 1 2 3 4 5 6
Time [s]
7 8 9 10 0 1 2 3 4 5 6 7
Time [s]
8 9 10

rw Radius of the wheel 0.045 m (c) Current of the motor (d) Control input voltage

KT Torque constant of the motor 2 15

Acceleration of the cart [m/s2]

0.18 N・m/A
KE Back electromotive force constant of the motor 0.18 V・s/rad 1.5
10
1
R Resistance of the motor 12.1 Ω
0.5
5
Force [N]

0 0
Figure 9. A power assist cart. -0.5
-5
-1
the processes. It enables us to save control energy -1.5
-10

and improve operators’ comfortable. The operation -2

0 1 2 3 4 5 6
Time [s]
7 8 9 10
-15
0 1 2 3 4 5 6 7
Time [s]
8 9 10

type - II in Fig. 8 shows its scheme. Hereafter, each (e) Acceleration signal of the cart (f) Force of the operator

process is called “mode” in this section. Figure 11. Experimental result.

designed simultaneously by means of a “servo “Reduction of Residual Vibrations in
type” ANSC. The servo type means that we obtain Positioning Control Mechanism,” Trans
the ANSC time-varying feedback gains for the ASME, J Vib, Acoust, Stress, and Reliab in
augmented system consisting of the original Design, Vol 107, No 1, pp 47-52.
controlled object system and an integrator for the [2]. Hara, S and Yoshida, K (1994). “Simultaneous
error between a reference and the corresponding Optimization of Positioning and Vibration
actual value. This technique is similar to so called Controls Using Time-Varying Criterion
LQI control. In such a control problem, controlled Function,” J Robot and Mechatro, Vol 6, No
object parameters such as mass on the cart often 4, pp 278-284.
varies in wide range. But the mass can be measured [3]. Hara, S (2000). “Reference Trajectory
by using a loadcell. Then, we assume that all the Generation Method for Servo Positioning
parameters of the controlled object are time- Control of Flexible Structures,” Fifth Int Conf
invariant and can be determined at the starting time Motion and Vib Contr, Sydney, NSW,
of control. The time-varying Riccati equation for Australia, Vol 1, pp 91-96.
the augmented system is calculated online during [4]. Hara, S (2005). “Unified Design of Time-
SCM and BM. For further details of this study, the Varying Gain Type Access Control and Integral
readers may refer to [9]. Type Servo Control by Means of Nonstationary
Optimal Control Method,” Microsyst
Examples of the time-varying weights and Technolo, Vol 11, No 8-10, pp 676-687.
feedback gains are shown in Fig. 10. “Servo [5]. Hara, S (2002). “Nonstationary Optimal
variable” in Fig. 10 corresponds to the state Positioning Controller Design Using GA Meta-
variable of the integrator. An experimental results Optimization,” 2002 American Contr Conf,
are shown in Fig. 11. In this result, the period Anchorage, AK, USA, pp 3165-3167.
between 0 and about 5 s is SCM. In the vicinity of [6]. Hara, S (2004). “Efficient Design of Time-
6 s, the operator catches the cart and the control Varying Gain Type Controller and Positioning
mode is switched from BM to ICM smoothly. In Control of Vibration Systems,” IEEJ Trans
ICM, the operator can settle the cart easily by Indust Applications, Vol 124, No 8, pp 843-
power assisting. 844.
[7]. Hara, S (2005). “Efficient Design of Time-
VI. CONCLUSIONS Varying Gain Type Controller and Positioning
This plenary talk discussed an adaptive Control of Vibration Systems –Taking Account
nonstationary control (ANSC) method based on of Parameter Variations–,” IEEJ Trans Indust
the progressive time calculations of Riccati Applications, Vol 125, No 9, pp 871-878.
equations and showed its application to positioning [8]. Hara, S (2006). “Adaptive Nonstationary
control problems. Concretely, this talk selected a Control Using Solutions of Time-Varying
positioning problem of a flexible structure installed Riccati Equations–Its Application to
on X-Y table and the mode switching control Positioning Control of Vibration Systems,” 9th
problem of a power assist cart. The effectiveness Int Conf Control, Automat, Robotics and
of the ANSC method was verified via the two Vision, Singapore, pp 403-408.
applications. [9]. Hara, S, Yamada, Y, and Lee, S (2006).
“Power-Assist Control Switching from
ACKNOWLEDGMENTS Adaptive Nonstationary Servo Control to
The author would like thank Dr. Leo Wiryanto and Force Sensorless Nonstationary Impedance
Dr. Roberd Saragih from Institut Teknologi Control,” 2009 IEEE Int Conf Indust Technolo,
Bandung for inviting the author to a plenary talk Churchhill, VIC, Australia, pp 289-294.
in CIAM2010, Bandung, Indonesia. The author’s [10].Yamada, Y, Konosu, H, Morizono, T, and
travel for CIAM2010 is supported by the Daiko Umetani, Y. (1999). “Proposal of Skill-Assist
foundation, Nagoya, Japan. : A System of Assisting Human Workers by
Reflecting Their Skills in Positioning Tasks,”
REFERENCES 1999 IEEE Int. Conf. Systems, Man,
[1]. Yamada, I and Nakagawa, M (1985). Cybernetics, Tokyo, Japan, Vol IV, pp 11-16.
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Some aspects of modelling pollution transport in groundwater aquifers

Robert McKibbin
Centre for Mathematics in Industry, Institute of Information and Mathematical Sciences
Massey University at Albany, Auckland, New Zealand
Email: R.McKibbin@massey.ac.nz

ABSTRACT In the more general case, the advection-

Dissolved chemical species disperse as they are dispersion equations that model the fluid and
advected by a fluid flowing within a permeable species transport then have coefficients that
matrix. Here, some mathematical models that may depend mainly on depth, but with a layer
be used for quantifying the spread of a pollutant composition that changes slowly with
species or tracer in groundwater systems are
horizontal distance.
developed, with a view to estimating passage times
to, and concentrations at, fluid withdrawal points.
The parameters in the pde's that describe the fluid Here, the coefficients in the pde's that describe
flow and pollutant flux may be constant within the fluid flow and pollutant flux are generally
each layer, but can be different in each of the assumed constant with depth in each layer, but
various layers. This allows aquifer systems with can be different in each of the various layers.
bedded structures to be modelled without having to This allows aquifer systems with bedded
resort to full-scale numerical simulations. structures to be modelled without having to
resort to full numerical simulations. The layer
KEYWORDS thicknesses are assumed small compared to the
Porous media; pollutants; tracers; groundwater lateral extent of the aquifer, and their interface
aquifers; mathematical modelling; dispersion. slopes are also small.
I. INTRODUCTION
Within each layer, changes in concentration
As dissolved chemical species are advected by
may occur by advection and dispersion, and
a fluid flowing within a permeable matrix,
also by transfer of the pollutant across the
they are dispersed by mixing as the fluid
layer interfaces when the concentrations
makes its way through complicated pathways
within adjoining layers are not equal. A
in the porous material. This means that
layered system may also be used as an
pollutants that find their way into groundwater
approximate realization of an aquifer with
aquifers not only move "down-stream", but
properties that vary smoothly with depth,
also spread in all directions. In general, the
allowing semi-analytic solution of realistic
rate of dispersion of a chemical species
physical problems where dispersion takes
dissolved in groundwater depends on the
place in all directions (see [4], for example).
porous structure and the fluid speed (see [1],
[2], [3] for various discussions).
Some illustrative results will be presented.
For clarity, the examples shown are where the
In this paper, some mathematical models that
species injection is via horizontal line sources,
may be used for quantifying the spread of a
thereby enabling 2-D solutions to be depicted.
pollutant species in groundwater systems are
The signature of the species at withdrawal
developed; it is assumed that the fluid flow
points is shown to depend strongly on the
remains steady. Usually, groundwater systems
layering, as well as on the injection and
have layered structures determined by
sampling depths.
different events in the geological processes
that formed them. The sub-layers in such a
The models incorporate most of the important
system have different matrix properties; in
structure of groundwater aquifers and the
turn, a typical sub-layer has a thickness and
transport of dissolved species therein, without
matrix properties that may vary with map
having to resort to large-scale simulation,
coordinates (c.f. [4], where layer thicknesses
thereby allowing efficient examination of the
and properties were both assumed constant).
effects of parameter variation.

9
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

II. FLUID FLOW EQUATIONS after use of (1). Here,

Uniform aquifer u  u(x, y, zb (x, y)) , ut  u(x, y, zt (x, y))

The equations of conservation of mass and %b % % %
momentum for the isothermal flow of an are the fluid velocity vectors at the lower and
incompressible fluid (water) in a rigid, upper boundaries where the latter are defined
uniform, permeable (rock) matrix, are [2]: by, respectively:
 u  0
% b (x, y, z)  z  zb (x, y)  0
K K (1)
u  (p   g)  (P) t (x, y, z)  z  zt (x, y)  0
 %  % 
The fluid velocities at these impermeable
Here, u  (u,v, w) is the specific discharge or
% boundaries are parallel to the surfaces, and (3)
Darcy velocity in a Cartesian coordinate becomes, exactly:
system (x, y, z) where the gravitational
acceleration is g  (0,0, g) . The fluid qx q y  

–3 %   (hu )  (hv )  0 (4)
density  [kg m ], fluid dynamic viscosity  x y x y
[kg m–1 s–1] and the matrix permeability K
[m2] are all constants; p [Pa] is the fluid The volume flux components may be written
pressure, P(x, y, z)  p(x, y, z)  p0   gz is in terms of a (stream-) function  (x, y) :
the dynamic pressure [fluid pressure adjusted
for hydrostatic pressure (fluid weight)] and p0  
hu  ,ŹŹŹhv   (5)
is a datum pressure. y x

The bottom and top boundaries at z  zb (x, y) , which satisfy (4) exactly. The mean dynamic
pressure through the layer thickness is:
z  zt (x, y) are impervious and smoothly-
varying, as is the aquifer thickness, which is z ( x,y)
1 t

h(x, y) z (x , y )
given by h(x, y)  zt (x, y)  zb (x, y) . The P(x, y)  P(x, y, z)dz (6)
two-dimensional (horizontal) volume flux b

vector is q(x, y)  (qx (x, y),q y (x, y)) , whose

% Then:
components are defined by:
z ( x, y )
 t
P z z
(hP)  
zt ( x, y )
dz  Pt t  Pb b
qx (x, y)  
zb ( x,y )
u(x, y, z)dz  h(x, y)u (x, y) x z ( x, y )b
x x x
(2) zt ( x, y )
 zt zb

zt ( x,y )
  udz  Pt  Pb (7)
q y (x, y)  
zb ( x, y )
v(x, y, z)dz  h(x, y)v (x, y) zb ( x, y )
K x x
 zt zb
 hu  Pt  Pb
where u (x, y) and v (x, y) are the components K x x
of the mean horizontal Darcy velocity. Then,
after using (1), and where
qx q y zt ( x, y )
 u v 
Pb  P(x, y, zb (x, y)),ŹPt  P(x, y, zt (x, y))
x

y
 
zb ( x, y )
 x  y  dz

zt zb zt zb are the dynamic pressures on the bottom and
ut  ub  vt  vb (3) top boundaries. A similar equation is found
x x y y
for the y-derivative. Then:
  ub  b  ut  t
 % %

10
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

P  z z with:
h   hu  ( Pt  P) t  (Pb  P) b
x K x x
(8)   1     1  
P  z z  0
h
y
  hv  (Pt  P) t  (Pb  P) b
K y y x  h x  y  h y 
(13)
  P    P 
h  h 0
Integration of the z-component of Darcy's Law x  x  y  y 
in (1) gives:

zt ( x, y )
The streamlines and isobars are orthogonal:
qz  hw  
zb ( x, y )
w(x, y, z)dz
  P  0 (14)
zt ( x, y )
(9)
K  P  K
 
zb ( x, y )
   dz  (Pb  Pt )
  z  
Also note that for small interface slopes (i.e.
small ), the dynamic pressure has little
variation with depth, and therefore, to good
It is proposed that this model would apply to approximation,
systems where the horizontal gradients of the
bottom and top confining surfaces are small. P(x, y, z)  P(x, y)  P(x, y) (15)
Consider the mass conservation equation in
(1), rearranged: (Note that, for a uniformly-thick aquifer
( h(x, y)  h , a constant), both  and P are
w  u v  harmonic functions, as used in the classical
    (10)
z  x y  theory.) The mathematical problem here
reduces to solving one or both of (13), subject
If u and v (and hence u and v ) are of order to suitable boundary conditions.
U, the aquifer thickness is of order H and the
horizontal length over which significant Aquifer with uneven sub-layers
variations occur is L, then (10) shows that w is In the more general case, water is supposed to
be flowing steadily though an aquifer system
O(U ) , where   H L is small. Equation
that varies only slightly from the horizontal,
(9) indicates that vertical variations in P are of and which is composed of N layers of
order (  K )Hw  (  K )H U . Therefore the permeable materials that may differ in
last two terms in each equation of (8) are of thickness and matrix properties, all of which
order (  K ) Hw( H L)  (  K ) 2 HU , while may also vary with horizontal position (x, y) .
the first term on the RHS of each is of order The system of total thickness h(x, y) is
(  K )HU . For small , the last two terms on confined between impervious surfaces at:
the RHS of each equation in (8) are O( 2 )
z  z0 (x, y) , z  z N (x, y)  z0 (x, y)  h(x, y)
smaller than the first and may be neglected.
Then, to good accuracy, we may write:
The aquifer sub-layer interface levels are at
z  zi (x, y),Źi  1,ŹK , N  1, with the thickness
P  P  
  u ,ŹŹŹ   v (11) of Layer i given by:
x K y K

So, the mean velocity components may be h (x, y)  zi (x, y)  zi1 (x, y),Źi  1,ŹK ,ŹN ,
i
written in terms of derivatives of  and P as:
N
The system thickness is h(x, y)   hi (x, y) .
i 1
1  K P The normal fluid volume flow at each
u 
h y  x interface is continuous.
(12)
1  K P
v   As discussed above for the uniform aquifer,
h x  y the flow within Layer i is supposed to be

11
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

approximately uniform through the layer's Since the volumetric discharges within the
thickness; the mean specific volume flux separate layers change with horizontal
(Darcy speed) in that layer is denoted ui (x, y) distance, interlayer flows must occur to
[m s–1]. The total volume flow per unit width maintain the total fluid volume. Denoting the
flow per unit interface area from Layer i into
is q  (qx (x, y),q y (x, y)) [m2 s–1], given by the
% Layer i  1 by wi (x, y) , the required mass
sum of the layer fluxes qi (x, y) : balance for Layer i gives, to current accuracy:
%
N N
q(x, y)   qi (x, y)   hi (x, y)ui (x, y) . (16)  q
% q yi 
% i 1 % i 1
wi (x, y)  wi1 (x, y)    xi   (21)
 x y 
The same approximations as for the single-
layer case are made. From (1), the specific
volume flux in each layer depends on the where w0 (x)  wN (x)  0 (the top and bottom
horizontal gradient of the mean dynamic boundaries are impermeable).
pressure Pi (x, y) in that layer; these pressures
III. POLLUTANT FLOW EQUATIONS
and their gradients are, with error O( ) and
O( 2 ) respectively, the same in all layers. Uniform aquifer
With Pi (x, y)  P(x, y) for i  1,Ź,ŹN : The model is of a fluid containing low
concentrations of a soluble pollutant flowing
steadily along a near-horizontal permeable
K i (x, y)  P P 
u ,v  
i i
,
  x y 
(17) layer that lies between impermeable upper and
lower boundaries and has a slowly-varying
thickness, as described in Section II above. In
addition to parameters previously defined, the
where K i (x, y),Źi  1,ŹK ,ŹN [m2] are the layer
 permeable matrix has porosity  (x, y) [–]. In
permeabilities. Note that this means that the
a small x-y planform region of area xy over
flow direction in every layer at any planform
a small time interval t, the change in the
point (x, y) is the same. Substitution of (17)
vertically-averaged mass of pollutant per unit
into (16) gives volume of fluid c (x, y,t) [kg m–3] is due to
advection, dispersion with coefficient D(x, y)
   
N
q(x, y)  qx ,q y   qxi ,q yi [m2 s–1] (that may depend on the local fluid
% i1 speed), and a distributed pollutant source with
(18)
N  1  P P  mass concentration per unit fluid volume
    hi (x, y)K i (x, y)   ,
    x y  f (x, y,t) [kg s–1 m–3], and is described by:
 i1

The layer specific discharges are then:  hxy[c (x, y,t  t)  c (x, y,t)]
    
 y   [qx c ]x  
u ,v 
i i N
K i (x, y)
q , q 
x y
(19)   x

 

  
K k
(x, y)hk (x, y)  x   [q y c ]y  
k1
  y  
 
The layer fluxes are separately given by     c   
ŹŹ y   hD  x   t (22)
 x
  x   
qi (x, y)  ui (x, y)hi (x, y)     
 c  
% % x   hD  y  
K (x, y)hi (x, y)   y  y   
 N i q(x, y) (20)
 
 K k (x, y)hk (x, y) %  hxyf (x, y,t)t 
 k1  
ŹŹŹO( 2 )
As in (4), qx x  q y y  0 .

12
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

This is a vertically-averaged form of the Suppose a mass of pollutant per unit area of
balancing of changes in pollutant mass in a the aquifer, Q [kg m–2], is released as a plane
small REV with advective and dispersive source at a certain time (t  t0 ) and place
fluxes across the boundaries. Divide (22) by
xyt and take the limit as x, y, t  0 : (x  x0 ) into the aquifer. From (23), the
equation for the species concentration c (x,t)
resulting from such a source, represented by
h
c
t


x
qx c 

 
qc
y y
  f (x,t)  Q (x  x0 ) (t  t0 ) , is:

  c    c 
  hD    hD  c  u (x) d[ h(x)D(x)] dx  c
x  x  y  y   
t    h(x)  x
 hf (x, y,t) (26)
2 c
ŹŹŹŹ=D(x) 2  Q (x  x0 ) (t  t0 )
Expand, divide by  (x, y)h(x, y) , use (2) and x
(4) and rearrange, to give:
with c (x,0 )  0,Ź  x   . The solution
c  u ( hD) x  c of (26) clearly depends on the assumed form
   of the dispersion coefficient D(x) . For
t   h  x
example, one possibility is that the dispersion
 v ( hD) y  c is proportional to the mean interstitial flow
   (23)
 h  y speed U  u  :
 2 c 2 c 
ŹŹŹŹŹ D(x, y)  2  2   f (x, y,t) D(x)  U (x) (27)
 x y 

The initial condition is: where  [m] is a dispersion length

(dispersivity) that depends on the matrix
structure (see [1], [2], [3] for discussions).
c (x, y,0 )  0,Ź  x, y   (24) This length is a constant for a uniform aquifer.
Then, since  h(x)D(x)   h(x)U (x)   qx
Example
The general equation (23) for the vertically- is constant, (26) becomes:
averaged species concentration may be
simplified in certain cases. Consider, for c c 2 c
 U (x)  U (x) 2
example, a homogeneous aquifer whose t x x (28)
thickness varies in only in one direction which Q (x  x0 ) (t  t0 )
is also the same as that of the steady fluid
flow; this is the 1-D flow case. Equation (2)
gives the total fluid flow per unit width: Attempts to find an analytic solution for this
equation have been unsuccessful so far; efforts
zt ( x ) are continuing. So standard numerical
methods were used to solve (28).
qx  
zb ( x )
u(x, z)dz  h(x)u (x) (25)
Space allows only a couple of illustrative
examples to be shown here. The thickness
while (4) asserts that qx is a constant. From profile is selected to be periodic, of the form
(11), the gradient of the mean dynamic h(x)  1 a sin( x) .
pressure P(x) is given by
In comparison with aquifers of varying
dP   qx thickness, the case of uniform thickness
 u  . ( a  0 ) may be solved analytically. The mean
dx K K h(x)
interstitial flow speed U is constant and the
equation to be solved is:

13
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

c c 2 c (a)
U  U 2
t x x (29)
Q (x  x0 ) (t  t0 )

(b)
with the well-known analytic solution:

( xx0 Ut )2
Q 
c (x,t)  e 4Ut
(30)
4Ut

The concentration profiles after two particular (c)

times (t = 2, 4) for a system with a  0 and
otherwise the parameters listed in Table 1, is
shown in Figure 1. In Figure 2, the profiles
after the same times are shown for a system
with uneven thickness. (Note that the chosen
times, as well as the parameter values listed in
Figure 2. (a) System shape, as described in the text.
Table 1 are arbitrary.) Figure 2(a) shows the Source is marked at x = 0.
aquifer profile. The pollutant concentration in (b): Predicted pollutant concentration after t = 2.
the aquifer at time t = 2 is shown in Fig. 2(b) (c): Predicted pollutant concentration after t = 4.
and for t = 4 in Fig 2(c). Other parameters are listed in Table 1.

Table 1. Parameter values used for example of pollutant The average interstitial flow speed in both
transport in a porous layer with uneven thickness and cases is about 2.5 m s–1. After times of t = 2
uniform matrix properties. and 4, the maximum concentrations are at
about x = 5 and 10 respectively. The
a 0.8 variations in Fig. 2(b), (c) are due to that of the
 2 fluid speed and hence also the dispersion rate.
 0.1
qx 0.25 An instrument set at a particular position
downstream would detect a wavy profile
 0.2
different from the smooth one of Fig. 1. Sets
Q 1
of field data are now being sought to find if
t0 0 any have this characteristic, and if the inverse
x0 0 problem could then be solved to infer aquifer
thicknesses.

Aquifer with uneven sub-layers

(a) While algebraically more complicated, the
multi-layered case is treated in a similar way,
but with allowance being made for advection
and dispersion of the pollutant across the inter-
layer boundaries.

(b)
Within Layer i, where the vertically-averaged
species concentration is ci (x, y,t) , the
pollutant mass change may be due to:
advection and longitudinal dispersion of the
species along the layer, advective gains or
losses from neighbouring layers according to
Figure 1. Uniform layer the interlayer flow direction and strength,
(a) Predicted pollutant concentration after t = 2. cross-interface dispersion, and sources; this
(b) Predicted pollutant concentration after t = 4.
Other parameters are listed in Table 1. results in Equation (31):

14
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

ci   ci  with  0 (x, y)   N (x, y)  0 (zero dispersive

i hi  qxi ci  i hi DLi 
t x  x  flux across the bottom and top boundaries).
Note that the form of (34) is analogous to that
  c  for resistances in series.
  q yi ci  i hi DLi i 
y  y 
In general, the set of simultaneous pde's (31) is
 1 sgn(wi1 ) 
  i1  wi1  ci1 unable to be solved analytically (c.f. [4] for the
 2  uniform sub-layer thickness case). The system
(31) may be solved numerically over an x-y domain
 1 sgn(wi1 ) 
  i1  wi1  ci ( [xu , xd ]  [ yu , yd ] , for example), with layer
 2 
thicknesses, permeabilities, porosities,
 1 sgn(wi )  longitudinal and transverse dispersion
  i  wi  ci
2 coefficients, and boundary conditions being
 
specified. Solution of the fluid flow equations
 1 sgn(wi )  for the total fluid volume flux q(x, y) then
  i  wi  ci1  i hi f i (x, y,t) %
 2  allows calculation of the pressure gradient, the
longitudinal layer fluid speeds and fluxes, the
where  i (x, y) [m s–1] is the interlayer transfer interlayer fluid flows, the interstitial speeds,
and the interlayer dispersion coefficients.
coefficient between Layers i and i  1 . These
coefficients may be estimated by considering
In the examples shown here, the equations
the dispersive species flux between the
(31) were solved using a standard forward
concentrations ci (x, y,t) , ci1 (x, y,t) in the difference in time with error O(t) , and
two neighbouring layers across the interface at central differences in space with error
z  zi (x, y) where a nominal concentration O(x 2 , xy, y 2 ) (suitable variations of the
 i (x, y,t) is assumed. The dispersive fluxes space formulae are applied at the ends of the x-
below and above the interface must match: and y-ranges to ensure the error order is the
same). Spatial bc's for the pollutant
 c  concentrations are applied only at the
 i ci  ci1 ; i DTi (x, y)  upstream end of the solution region, which is
 z  zzi chosen to be far enough away to allow zero
(32)
 c  concentrations to be specified.
; i1 DTi1 (x, y)  
 z  zzi Example

Consider a multi-layered system where the
where the dispersion coefficients transverse to layer thicknesses vary in the x-direction only,
the main fluid flow are denoted by and where the fluid flows in that direction
D (x, y),Źi  1,ŹK ,ŹN . Vertical derivatives of under the influence of a pressure P(x) .
 Ti
the concentration are approximated by
difference formulae: For a source injected instantaneously as a
plane strip parallel to the y-axis at x  x K in
 i  ci ci1   i Layer K at the initial time t = 0, into a flow
i DTi ; i1 DTi1 (33)
hi 2 hi1 2 where qx  0 , q y  0 , the following boundary

(35) and initial (36) conditions are applied:
After solution for  i (x, y,t) and substitution
into (32), the transfer coefficients are given ci
ci (xu , y,t)  (x , y,t)  0,
by: x u (35)
1 hi 2 h 2  i  1,ŹK ,ŹN
  i1 ,ŹŹi  1,ŹK ,ŹN  1 (34)
 i DTi i1 DTi1
 i

15
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

cK (x, y,0)  f K  (x  x K ), (in Layer 2). One of two assumptions is made:

either there is vertical dispersion [Figs 3(b),
xu  x  xd ,ŹŹŹxu  x K  xd (c)] or not [Figs 3(d), (e)]. The effects of the
(36)
Źci (x, y,0)  0,ŹŹŹxu  x  xd , lens and the dispersion model choice are seen
i  1,ŹK ,ŹN ,Źi  K to be significant. As may be noted, after being
 borne by advection and longitudinal
dispersion, the pollutant's highest
Lack of space precludes a comprehensive set
concentration remains at the level in the
of examples; only one illustration is shown
aquifer where the release was made, but some
here. A confined aquifer with a non-uniform
has spread vertically as well as along. In Figs
total thickness that varies (periodically, in this
3(b), (c) there is a residual amount that has
case) in one horizontal direction only, given
dispersed into, and which is slow to leave, the
by h(x)  H[0.8  0.2cos(5 x 7)] , is low-permeability region. Even neglecting
supposed to be composed of 11 layers (of vertical dispersion in the model, the effect of
equal thickness in this case) – see Fig. 3(a). the low-permeability lens on the fluid
The layers are numbered from bottom to top. movement, and hence that of the vertically-
advected pollutant, is evident in Figs 3(d), (e).
(a)
Because there is little fluid flow within the
lens, the water has to move around it, and
advects the pollutant vertically in the layers
near the ends of the lens. In both Figs 3(d),
(b) (e), this has moved the pollutant towards the
middle of the aquifer, effectively providing
dispersion by advective processes.

IV. SUMMARY & CONCLUSIONS

(c) Some mathematical models that may be used
for quantifying the spread of a pollutant
species or tracer in geologically-layered
groundwater systems have been developed,
with a view to estimating passage times to, and
(d) concentrations at, fluid withdrawal points.
Examples show that the concentration
signature of the species depends strongly on
the layering, as well as on the injection and
sampling depths. Use of geometric stratum
(e) thickness data enables solutions to be found
with modest computational resources.

REFERENCES
Figure 3. (a) System shape, as described in the text, with
[1]. J. Bear (1972). "Dynamics of fluids in
a low-permeability lens shaded. porous media," New York: Dover.
(b – e): Predicted pollutant concentrations after a certain [2]. J. Bear and A. Verruijt (1987). "Modeling
time, are shown under assumptions of: groundwater flow and pollution,"
(b, c) vertical dispersion; (d, e) negligible vertical
Holland: Reidel.
dispersion. Release point is marked *.
[3]. J. Bear and Y. Bachmat (1991).
"Introduction to modeling of transport
In this example, the layer porosities, phenomena," The Netherlands: Kluwer.
dispersion lengths and permeabilities are taken [4]. R. McKibbin (2009). "Groundwater
as uniformly equal, except for a region in pollutant transport: transforming layered
Layers 5 and 6 that represents a clay lens with models to dynamical systems," An. St.
permeability 1/100 of that elsewhere; it is Univ. Ovidius Constanta, Serie
marked in Fig. 3(a). A pollutant is released Matematica Vol 17(3), pp 183-196.
near the top (in Layer 11) or near the bottom

16
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Jets and Bubbles in Fluids – Fluid Flows with Unstable

Interfaces

Larry K. Forbes(1)
(1)
School of Mathematics and Physics, University of Tasmania, Hobart, Tasmania, Australia
(Email: Larry.Forbes@utas.edu.au)

solved numerically. This can represent a

significant challenge both for mathematicians
ABSTRACT and for computers, and until relatively recently
This presentation discusses two types of problems in only steady-state inviscid free-surface flows
interfacial fluid mechanics. The first involves a jet or have yielded to accurate numerical solution. A
plume of light fluid moving upwards through a canonical problem of this type was introduced
surrounding heavy fluid. When viscous effects are
by Tuck & Vanden-Broeck [8]. It asked the
ignored, a symmetric jet rises vertically and
eventually settles down to a shape that is independent deceptively simple question: what is the steady-
of time. However an asymmetric jet evidently state shape adopted by the free surface of a
becomes unstable and never reaches a steady state. stationary inviscid fluid, into which a line source
When viscous effects are introduced, this qualitative is immersed? The authors [8] showed that there
difference between the symmetric and non-symmetric are two different steady solution types, due to
jets disappears, as now neither of them can achieve a the non-linearity of the problem, and a
steady state, since viscosity causes the outer fluid to considerable literature on this problem has since
be entrained in both cases. The second problem is developed.
related to the first, and it involves a light fluid More recently, it has become possible to obtain
injected either cylindrically or spherically into a
numerical solutions to some time-dependent
surrounding heavy fluid. Without viscous effects, the
expanding interface develops points of infinite large-amplitude free-surface flows. This then
curvature within finite time, after which the solution raises the question of whether the non-linear
fails to exist. However, when viscosity is re- steady-state solutions obtained earlier are stable,
introduced, the points of infinite curvature at the in the sense that they would be observed as the
interface are replaced with points of high large-time limits of an unsteady solution.
concentrations of vorticity. This results in over- Furthermore, it is now known that unsteady
turning mushroom-shaped plumes moving radially solutions can become singular within finite time,
outwards. when viscosity is ignored, due to the formation
of points of infinite curvature at the free surface.
KEYWORDS This was first demonstrated by Moore [6].
Fluid mechanics; interfaces; instability: inviscid
Forbes et al. [4] developed a spectral solution
fluid; spectral methods; viscous fluid.
method for solving certain unsteady inviscid
I.INTRODUCTION problems, and confirmed the development of
Free-surface fluid mechanics is an important surface curvature singularities. Later, Forbes [2]
subject, and has practical uses in understanding used that technique to solve for Rayleigh-Taylor
jets, ocean waves, flow around ships, and so on. instability in inviscid fluids, and observed
A classical reference is the article by Wehausen surface curvature singularities close to the
& Laitone [9] that summarizes many of the asymptotically predicted time. He then used a
linearized (small amplitude) results. Large spectral solution to solve the corresponding
amplitude free-surface flows are more difficult problem in viscous fluids, and demonstrated that
to model, since the shape of the fluid region is the inviscid Moore curvature singularity is
unknown in advance. Mathematical models of replaced with a small region of high vorticity.
large-amplitude free-surface flows are therefore This then causes the interface to roll up into
inherently non-linear, and can usually only be mushroom-shaped plumes.

17
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Here, these techniques are extended and applied  2 1  2 1

to the study of jets and bubbles in fluids. Each   0, XL  x  XR (1)
of these situations is characterized by a light x 2 y 2
fluid displacing a heavy fluid, and so can be On the bottom, there is the requirement
subject to Rayleigh-Taylor type instability.  0, x 1
Spectral methods are introduced, to solve these v1 ( x, 0, t )   on y  0 (2)
problems both for inviscid and viscous fluids. sin  , x 1
These flows are realizable in the laboratory, and and at the top of the plume, the condition
may also have relevance to “black smokers” in u1 ( x, h, t )  cos  (3)
the ocean [5] and to modelling star growth [7]. is assumed. On the left-hand interface
x  X L ( y, t ) , there is a kinematic condition
II. INVISCID FLUID JETS
Consider a stationary outer fluid 2, with an inner u1  X L / t  v1X L / y (4)
fluid 1 forming a rising jet, as illustrated in which states that fluid 1 is not free to cross this
Figure 1. interface, and a dynamic condition
1 1 2 ( D  1)
  u1  v12  
1
y (5)
t 2 F 2
2
y that expresses the continuity of pressure at this
interface. The identical conditions to (4) and (5)
also apply on the right-hand interface
y=h
x  X R ( y, t ) in Figure 1.
2 1 2 A spectral solution method has been developed
to solve the non-linear time-dependent problem
x = XL (y,t) x = XR (y,t) given by equations (1) – (5), based on the
technique introduced by Forbes et al. [4]. The

velocity potential is expressed as
x
-1 1
1 ( x, y, t )  x cos   y sin   A0 (t )
Figure 1. Illustration of dimensionless inviscid jet flow.
N
The jet makes an angle  with the bottom vent.
   An (t ) cosh( n x)
The problem is non-dimensionalized so that the n 1

 Bn (t ) sinh( n x)  cos( n y ) (6)

bottom vent has width 2. In addition to the
angle  at which the jet enters from the bottom
vent, there are three other dimensionless and the two interfaces are written
N
parameters that characterize the flow. These are
X L ( y, t )  1  y cot    Cn (t ) sin( n y )
the height h of an upper wall through which n 1
the jet passes, a density ratio D of fluid 2 to N
fluid 1 (so that D  1 ) and a Froude number F X R ( y, t )  1  y cot    Dn (t ) sin( n y ) . (7)
that measures the inlet speed of the jet at the n 1

bottom. The constants in (6) and (7) are defined as

Since fluid 1 is irrotational and incompressible,  n  (2 n  1) /(2 h) . (8)
a velocity potential 1 can be defined in it, so The velocity potential (6) satisfies Laplace’s
that the velocity vector is grad1 . Then the equation (1) and conditions (2), (3) identically.
The kinematic condition (4) and Bernoulli
horizontal and vertical components of velocity equation (5), and their equivalents on the right-
can be calculated as u1  1 / x and hand interface x  X R ( y, t ) , are Fourier
v1  1 / y . The velocity potential satisfies decomposed to yield a system of 4 N  1
Laplace’s equation ordinary differential equations for the 4 N  1
coefficients An (t ) , Bn (t ) , Cn (t ) and Dn (t )

18
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

and these are integrated foward in time using a above light fluid) cause the jet to break up, at
fourth-order Runge-Kutta scheme [1, p.371]. which time the numerical solution technique
It is possible to carry out an asymptotic solution fails.
of the governing equations (1) – (5), on the Inviscid ; h = 15 ; D = 1.05 ; F = 2 ;  = 0.99/2
assumption that the jet is thin. The details are 15 15 15 15
not given here, but the analysis is similar to
classical shallow-water theory [9]. According to
this approximation, there is a steady-state 10 10 10 10
solution in which the vertical speed in the jet is
y
v1 ( y )  sin 2   2( D  1) y / F 2 (9)
5 5 5 5
and the two interfaces have the shapes
X L ( y )   y cos   sin   / v1 ( y )
X R ( y )   y cos   sin   / v1 ( y ) . (10) 0 0 0 0
-2 0 2 -2 0 2 -2 0 2 -2 0 2
When the plume is vertical (    / 2 ) the two t=8 t = 12 x t = 16 t = 20

interfaces in (10) are anti-symmetrical. Figure 3. Inviscid plume at angle   0.99 / 2 at four
h = 15 ; D = 1.05 ; F = 1 different times. Here, h  15 , D  1.05 , F  2 .
15

III. VISCOUS FLUID JETS

Viscosity is now taken into account
10 approximately, using a Boussinesq approach [2].
t=8 Rather than treat the system as two separate
y t = 48
fluids with a sharp interface, the system is
instead modelled as a single fluid with a finite
5
interface thickness. The fluid density is written
Dashed = Asymptotic
as  ( x, y , t )  D   ( x, y , t ) , in which the
perturbation  is regarded as “small”. The
0
-1.5 -1 -0.5 0 0.5 1 1.5 mass conservation equation then splits into an
x
incompressible component
Figure 2. Inviscid vertical plume (    / 2 ) at two
different times. The steady asymptotic solution (10) is also
u / x  v / y  0 (11)
shown (dashed line). Here, h  15 , D  1.05 , F  1 . and a weakly compressible transport equation
Figure 2 shows numerical solutions at the two     2  2  
different times t  8 , 48. The solution was u v   2  2  (12)
t x y  x y 
started with initial conditions X L ( y ,0)  1 ,
for  . Equation (11) is satisfied identically
X R ( y,0)  1 . As time progresses, a contraction using a streamfunction  , in terms of which the
wave moves up the jet at a speed given two velocity components may be written
approximately by (9), and eventually, the jet u   / y and v   / x . (13)
adopts a steady shape close to that predicted by In addition, the vorticity vector (which is the
the asymptotic solution (10). curl of the velocity vector, and represents twice
The evolution of an asymmetric inviscid jet is the angular velocity) has just the one component
shown in Figure 3, for the four different times
t  8 , 12 , 16, 20. This example highlights the    v / x  u / y    2  . (14)
fact that an asymmetric jet does not approach the The viscous Navier-Stokes equations can be re-
steady-state solution (10) as time increases, written in terms of vorticity component (14), and
unlike the symmetric solution in Figure 2. under the Boussinesq approximation this gives
Instead, the jet becomes unstable and develops rise to the vorticity equation
small oscillations; eventually, Rayleigh-Taylor
type instabilities (in which heavy fluid lies

19
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

   integrated forward in time, using the classical

u v fourth-order Runge-Kutta scheme [1, p.371].
t x y
(15) Streamfunction ; h = 15 ; D = 1.05 ; F = 1 ;  = 5
1  1   2  2  15 15 15
  
DF 2 x DRe  x 2 y 2 
The constants  in equation (12) and Re in 10 10 10
(15) are a diffusion coefficient and Reynolds
number, respectively. In this approximate
y
viscous model, boundary layers are ignored on 5 5 5
the top and bottom walls; in addition, artificial
walls are placed at x    to bound the
computational region, and likewise have no 0 0 0
-5 0 5 -5 0 5 -5 0 5
boundary layers. Therefore t=6 t = 12 x t = 18
u  0 on x    (16) Figure 4. Streamlines for viscous vertical plume
and (    / 2 ) at three different times. Here, h  15 ,
D  1.05 , F  1 , and   5 .
 0, x 1
 ( x,0, t )   on y  0 . (17) A viscous plume is shown in Figure 4, for the
1  D, x 1 same case as in Figure 2. Here, the
For the general plume at an angle  , the streamfunction  in (18) has been contoured to
solution for density perturbation and give streamlines for the flow, and is shown at
streamfunction is expressed in the form three different times. Unlike the inviscid case,
there is clearly no approach to a steady state
 ( x, y , t )   S ( x, y )   U ( x, y , t )
flow, since now the outer fluid is entrained into
 ( x , y , t )   S ( x , y )  U ( x , y , t ) , (18) the plume, as a result of viscosity.
in which the “steady” density is Streamfunction ; h = 15 ; D = 1.05 ; F = 2 ;  = 5 ;  = 0.99 /2

1  D, x  y cot   1
15 15 15

 S ( x, y )   (19)
 0, 1  x  y cot   
with a similar form for streamfunction. It is then 10 10 10
appropriate to express the “unsteady” y
components in equation (18) as
 U ( x, y , t )  5 5 5

M N

 B
m 1 n 1
mn (t )cos  m ( x   )  sin( n y )
0 0 0
-5 0 5 -5 0 5 -5 0 5
U ( x , y , t )  t=8 t = 16 x t = 24
Figure 5. Streamlines for viscous plume at angle
M N (20)
 A
m1 n 1
mn (t )sin  m ( x   )  sin( n y )   0.99 / 2 at three different times. Here, h  15 ,
D  1.05 , F  2 , and   5 .
with  n as defined in (8) and  m  m /(2  ) . Figure 5 shows a solution for a viscous plume,
for the same case as illustrated in Figure 3.
These expressions (20) are chosen to satisfy
Again, there is no approach to a steady-state
boundary conditions (16) and (17) identically.
solution, since viscosity causes entrainment of
Equations (20) are substituted into the governing
the outer fluid; this is a feature in common with
system (12) and (15), and after considerable
the symmetric solution of Figure 4. Small
algebra, a (large) system of 2MN ordinary
vortices can be seen either side of the jet, and for
differential equations is obtained for the Fourier the asymmetric case in Figure 5, they eventually
coefficients Amn (t ) and Bmn (t ) . This system is

20
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

cause the jet to develop sideways oscillations as and by conservation of mass

it rises. u2  0 , v 2  0 as r   . (23)
There are now two kinematic conditions at the
IV. RADIAL OUTFLOW IN A interface, replacing equation (4), and these take
CYLINDRICAL BUBBLE the forms
These spectral solution techniques can be
R v j R
applied equally successfully to more complex uj   , j  1,2 on r  R( , t ) . (24)
situations than the cartesian planar geometry t R 
illustrated in sections II. And III. Following the Similarly, there is a dynamic condition that
work of Tuck & Vanden-Broeck [8], a problem expresses the continuity of pressure at the
is now discussed in this section which is interface, and it becomes
 2 1 1
 D  u22  v 22    u12  v12 
intended to represent possibly the simplest case
of a non-planar Rayleigh-Taylor flow. An inner D 
t t 2
light fluid is injected through a line source at the
origin, into a surrounding heavier fluid. The ( D  1)  1 1  t 
 R  ( D  1)   
F2  8 (  t ) F  
interface between the flow fluids is therefore 2

subject to instabilities, and small disturbances at

the initial time t  0 can therefore be expected on r  R( , t ) . (25)
to grow as time progresses. Here, D is the ratio of densities of the outer
Consider a cylinder of light fluid and initial fluid to the inner fluid, and so D  1 . The other
radius a lying along the z -axis of a coordinate parameter F is a type of Froude number based
system. A line source is present on the axis, and on the strength of a constant acceleration of
ejects the light fluid at some constant volume gravity directed radially inwards; in the
rate (per length along the z -axis). There is a laboratory situation this term would be absent,
heavier fluid surrounding this cylinder. Once and in that case F   .
again, the problem is now non-dimensionalized, The two velocity potentials can be written
using the initial cylinder radius as the unit of log r M
length and the source strength divided by that 1 ( r , , t )    Pm (t )r m cos( m )
2 m 1
radius as the unit of speed.
log r M
 2 ( r , , t )    Qm (t )r  m cos( m ) (26)
Inviscid cylindrical outflow model. 2 m 1
When viscosity is ignored in both the light inner and the interface is represented in the form
fluid 1 and the heavy outer fluid 2, velocity
 t M
potentials 1 and  2 can be defined in each R( , t )   H 0 (t )   H m (t )cos( m ) .
fluid, as before. It is appropriate to make use of
 m 1

cylindrical polar coordinates ( r , , z ) , and to (27)

As in Section II., these expressions (26), (27) are
consider the interface between the two fluids to
substituted into the three interfacial conditions
have the form r  R( , t ) . Then Laplace’s (24) and (25), and eventually yield a system of
equations in each fluid, equivalent to equation 3 M  1 ordinary differential equations for the
(1) above, become coefficients Pm (t ) , Qm (t ) and H m (t ) , and
 2 j 1  j   j
2

   0, j  1,2 . (21) these are integrated forward in time using a

r 2 r r  2 fourth-order Runge-Kutta method, as before.
In each fluid, the radial and tangential The initial conditions are taken to be
components of fluid velocity are u j   j / r H K (0)  0 , PK (0)  QK (0)   (28)
and v j  (1/ r ) j /  . Near the line source, representing an initially circular interface with a
small velocity disturbance at the K th Fourier
the potential for the light inner fluid has the
mode.
singular behaviour
1  log r /(2 ) as r  0 (22)

21
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Viscous cylindrical outflow model. small amplitude   0.03 . The solution fails to
Once again, viscous effects can be accounted for continue beyond about t  12 , and to
using a Boussinesq approximation, using understand why, it is necessary to consider the
equations similar to (12) and (15). As in section curvature.
III., a vorticity-streamfunction formulation is D = 1.05 ; F = 1 ;  = 0.03 ; P2(0) = -Q2(0) = 
used, and now the streamfunction is represented 2
in the form
 ( r, , t )   /(2 ) 0
M N

(29)
 Amn (t ) J m  jmn r /   sin( m )
-2

m 1 n 1 -4
replacing equation (20). The density 2
perturbation is written
y
 ( r, , t )  ( D  1)r 2 /  2 0
t = 1 , 2 , ,,, , 11
N
 r  Cn (t )sin( n r /  ) 2 4
-2 -4 -2 0 x
n 1
M N Figure 7. Curvature at the interface, for the same case as
 r  Bmn (t ) J m ( jmn r /  )cos( m ) . (30) shown in Figure 6.
m 1 n 1 Figure 7 shows the interfacial curvature for each
In these expressions (29), (30), the functions J m of the eleven solutions in Figure 6. Initially, the
interface is a circle of radius 1, and so its
are first-kind Bessel functions of order m , and
curvature is   1 , as expected. However, as
jmn are their zeros. These are substituted into the interface evolves, it develops regions of very
the density transport equation and vorticity high curvature at its poles. It appears that the
equation to derive ordinary differential equations curvature becomes infinite there, within about
for the Fourier coefficients, and these are the finite time t  12 , and this then causes the
integrated forward in time. failure of the solution beyond this time. This is
D = 1.05 ; F = 1 ;  = 0.03 ; P2(0) = - Q2(0) = 
2 consistent with ideas in [6], and a Moore
curvature singularity is expected to form.
Density D = 1.05 ; F = 1 ;  = 0.03 ;  = 5 ; K = 2 ; t = 12
5
1
y 3
0 t = 1 ,2 , ... , 11
1
y
-1
-1

-2 -3
-2 -1 0 1 2
x
Figure 6. The evolution of the inviscid interface as time
-5
increases. The solution is shown at eleven different times, -5 -3 -1 1 x 3 5
for K  2 . The scales on the axes are equal.
Figure 8. Viscous solution for the same case as in Figure 6,
In Figure 6, the development of the cylindrical at the time t  12 . The dashed line is the inviscid result at
interface R( , t ) is shown for times the same time. The scales on the axes are equal.
t  1,,11 , for a case F  1 in which a The corresponding viscous solution is shown in
constant inwardly-directed acceleration of Figure 8, for the same case as in Figure 6, at the
gravity is present. The interface initially has a time t  12 . These are contours of the
circular shape, but its velocity was initially perturbed density in equation (30), and they
perturbed at the second mode, K  2 , with the show clearly the location of the effective
interface. The inviscid solution at this same

22
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

time is overlaid on this diagram, and is shown may be given here, but the techniques are similar
with a dashed line. It agrees well with the in principle to those outlined in the previous
viscous result, but it represents the last time at sections.
which it was possible to compute an inviscid Spherical polar coordinates are now used, and to
solution, before the curvature singularity caused simplify the problem as much as possible, only
the failure of that solution. axi-symmetric modes are considered. The flows
Density D = 1.05 ; F = 1 ;  = 0.03 ;  = 5 ; K = 2 ; t = 30 are therefore rotationally symmetric about the
5
z -axis, so that only the radial coordinate r and
polar angle  from the z -axis are involved.
In the inviscid case, the two velocity potentials
y in the inner fluid 1 and outer fluid 2 are now
0 represented in the forms
N
1
1 ( r , , t )     Bn (t )r n Pn (cos  )
4 r n 1
N
1
 2 ( r , , t )     Cn (t )r  n 1 Pn (cos  )
-5 4 r n1
-5 0 x 5
(31)
Figure 9. Viscous solution for the same case as in Figure And a similar expansion is used for the interface
6, at the time t  30 . r  R( , t ) . In these expressions, the functions
A careful study of the vorticity shows that the
Pn are Legendre polynomials. Once again, the
regions in which the inviscid solution developed
curvature singularities are now replaced by velocity potentials (31) and the representation of
regions of high vorticity in the viscous case, at the interface are substituted into the two
the same locations and time. It seems clear that kinematic conditions and the dynamic condition
high curvature in the inviscid model is the at the moving interface, to derive a system of
precise trigger for vorticity to concentrate in the ordinary differential equations for the Fourier –
viscous case. This, in turn, causes the interface Legendre coefficients Bn (t ) , and so on. These
to roll up and develop over-hanging plumes. are then integrated forward in time using a
This is illustrated at the much later time t  30 , Runge-Kutta method.
in Figure 9. The small velocity disturbance at The viscous problem is again modelled using the
the K  2 mode in the initial conditions (28) Boussinesq approximation, as in section III. In
has developed into a large bi-polar plume with this approach, the density of the outer fluid is
two mushroom-shaped outflows. These bi-polar assumed to be only slightly larger than that of
shapes are believed to be of importance in the the inner fluid, and the unstable density ratio
formation of stars in astrophysics [7]. D  1 is of interest here. Once again, a density
Further details of these types of flows are transport equation and vorticity equation may be
available in a forthcoming paper by Forbes [3]. derived, and these are similar to equations (12),
It is, in fact, possible to obtain different non- (15).
linear outflow modes for different choices of the In the following Figures 10 – 12, an axi-
integer K in the initial conditions (28), and symmetric viscous solution is shown, at three
different examples of those flows are available different times. In each solution, the same
from that paper [3]. density contour is illustrated, and has been
chosen to correspond most closely to the
V. AXI-SYMMETRIC OUTFLOW IN A location of the centre of the interface (which is
SPHERICAL BUBBLE of finite thickness in the viscous case). There is
The techniques of the previous section IV. can found to be close agreement between this
readily be extended to study outflow from a density contour and the interface predicted by
point source in three-dimensional geometry. In the inviscid equations (31) for early times;
the interests of space, only a very brief overview however, as in section IV., the inviscid solution

23
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

fails at finite time, due to the formation of locations at which the inviscid solution forms a
Moore-type curvature singularities at the curvature singularity, and eventually the two
interface. plumes form over-hanging regions, as in Figure
12.

VI. CONCLUSIONS
Spectral methods have been used here, to obtain
highly accurate solutions for jets and for
problems involving injected bubbles in fluids.
The inviscid theory may predict the formation of
Moore-type curvature singularities on the
interface within finite time, and these are
replaced with regions of high vorticity in the
viscous theory. This causes the formation of
over-hanging plumes as time progresses.
Figure 10. Viscous axi-symmetric expansion from a point
ACKNOWLEDGEMENTS
sink at time t  8 , for a solution with D  1.05 , F  0.5 ,
This work was supported in part by Australian
  0.03 and the bi-polar mode K  2 .
Research Council grant DP1093658.

REFERENCES
[1]. Atkinson, KE (1978). An Introduction to
Numerical Analysis, Wiley, New York.
[2]. Forbes LK (2009). "The Rayleigh-Taylor
instability for inviscid and viscous fluids", J.
Engin. Math., Vol 65, pp 273-290.
[3]. Forbes, LK (2011). "A cylindrical
Rayleigh-Taylor instability – radial outflow
from pipes or stars", J. Engin. Math., special
Tuck Memorial Issue, to appear.
[4]. Forbes LK, Chen MJ and Trenham CE
Figure 11. Same case as Figure 10, for time t  16 . (2007). "Computing unstable periodic waves
at the interface of two inviscid fluids in
uniform vertical flow", J. Comput. Phys.,
Vol 221, pp 269-287.
[5]. Jupp, T and Schultz, A (2000). "A
thermodynamic explanation for black smoker
temperatures", Nature, Vol 403, pp 880-883.
[6]. Moore, DW (1979). "The spontaneous
appearance of a singularity in the shape of
an evolving vortex sheet", Proc. Roy. Soc.
London, Vol A365, pp105-119.
[7]. Stahler, SW and Palla, F (2004). The
formation of stars, Wiley, Berlin.
[8]. Tuck, EO and Vanden-Broeck, J-M (1984).
Figure 12. Same case as Figure 10, for time t  24 . "A cusp-like free-surface flow due to a
The bubble surrounding the point source in submerged source or sink", J. Austral. Math.
Figure 10 is almost spherical, but as time Soc. Ser. B, Vol 25, pp 443-450.
progresses, the small bi-polar disturbance to the [9]. Wehausen, JV and Laitone, J (1960)
initial velocity grows unstably. Regions of high "Surface Waves", in: Handuch der Physik,
vorticity are formed at precisely the times and Springer-Verlag, Berlin, Vol 9 pp 446-778.

24
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

25
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Boundary Control of Hyperbolic Processes with

Applications in Water Flow

M. Herty(1) and S. Veelken(2)

(1)
RWTH Aachen University, Templergraben 55, D-52064 Aachen, GERMANY
(Email: herty@mathc.rwth-aachen.de)
(2)
RWTH Aachen University, Templergraben 55, D-52064 Aachen, GERMANY
(Email: veelken@mathc.rwth-aachen.de)

ABSTRACT and we refer to [3] for an overview of existing

This paper discusses a novel approach to boundary results. The approach followed in [3, 15] is based
control of hyperbolic conservation laws. As on linearization of the equations (1) and (2) around
working example the St. Venant equations for water the steady state. The arising quasilinear system
flow in open canals are considered. The control , 0, (5)
problem is formulated and its solution is discussed.
, (6)
KEYWORDS
is then analyzed using suitable Lyapunov
Control laws; conservation laws; relaxation
functions [6, 3]. Null–controllability and
methods.
exponential convergence towards the steady
state can be proven [6]. The results extend to
I.INTRODUCTION the non homogenous case, networks of canals
In this paper we are concerned with systems of and also to the case of gas networks including
conservation laws describing for example water friction terms [9]. However, the disadvantage
flow in horizontal open canals. A similar discussion of the previous approach is due to the fact that
holds also true in the case of gas networks the linearization requires a sufficient regularity
described by the isothermal Euler equations. In the
0,1 of the solutions , , , . In
case of horizontal water canals with a constant
rectangular cross–section and without friction the general, this regularity constraint is not met.
flow dynamics are described by a nonlinear system Therefore, in [13] a different approach based
of two conservation laws (St. Venant equations) on relaxation has been proposed. This will be
discussed in the following paragraph.
0 (1)
1 II. RELAXATION OF BOUNDARY
0 (2)
2 CONTROL PROBLEMS
Here, is the water height and its velocity. In [14] a relaxtion approach to conservation
denotes the gravitational constant. The equation is laws has been introduced and numericaly
supposed to hold in a canal parameterized by analyzed. It can be shown [14, 1, 17] that for a
0,1 and for 0. The equation is small parameter and up to order the
supplemented with suitable boundary conditions at following system is an approximation to (1)
0 and 0,1 , respectively. For more details and (2).
on the modeling and arising control questions we 0, (7)
refer to [15, 11, 5, 6, 9, 3, 10, 8, 7, 4, 12] and the 0, (8)
references therein. , (9)
A typical question in the context of water flow [3, . (10)
15] is as follows: how can we control the flow at
the boundaries of the canal in order to stabilize it? The approximation is dissipative provided that
More mathematically, we reformulate this question: satisfies the sub characteristic condition
Design boundary functions ,  , such that √ max , . (11)
,
any given initial data , converges over time to The advantage of the relaxation formulation is due
a steady
1 state. In the considered case the steady to the fact that the hyperbolic structure of the
state is , 0. Boundary conditions are in equations is preserved in this approximation. No
general imposed as linearization of the nonlinear terms is necessary.
,0 , ,0 , 0, (3) The disadvantage is the doubling of variables. The
,1 , ,1 , 0 (4) variables and are not needed to describe the
for some given, smooth functions and . water flow. For small it can be shown that these
This questions has received attention in past years

26
Proceedin
ng of Conf. on IIndustrial and
d Appl. Math., IIndonesia 201
10

variables converge to o and , previouus theorem guuarantees exponnential decay of

solutionns towards a stteady-state for arbirtrary initiial
respectivelly. Further results on relaxattion systems
and their numerical
n perfoormance we refer to [17, 2, data off small —noorm. Obviouslyy, for
16].
For the previous systeem we can noow apply an we obtaain
extension of Theorem 13.12 in [3] in order to
obtain a boundary
b contrrol result. To this end we for marks are in ordder.
. Some rem
reformulate (7)-(10) in n characteristiic variables
Remarrk 1. The ressult strongly depends
d on thhe
  and
value of . This ccontrols to soome extend thhe
. I.e., the previouus system is
norm ofo the initial data as welll as the decaay
equivalent to
rate . For a more detailed
d discusssion we refer to
,  (12) [13].
(13) For a completelyy different proof
p using a
nt Lyapunov ffunction we refer to [133].
differen
where andd .
Numerical experimennts can also be found
f in [13].

In order
o to comppare the bounndary conditionns
impliedd by (15) and (16) with w previoussly
For small values
v of we have introduuced conditionns (Chap. 13.44[3]) we assum me
, (14) from nown on subcrritical flow, i.e., the solutioon
satisfiees
Obviously,, we can reccover and from the
solutions and off the diagonal system (12) . (18)
and (13). In order to simplify thee discussion Furtherrmore, we consider
c so called
c spillwaay
following we prescribe linear
l boundarry conditions conditions, i.e., (3)–(44) are given byy
for some constants an
nd , (199)
, (15)) (200)
, (16)) where is the givenn water level above the cannal
We estimaate the derivativ ves of and with respect an a characteristtic constant ofo the spillwaay.
to the indeependent variab bles at the dataa From the t applicationn it is desiredd to express thhe
in orrder to apply Thheorem 13.12[[3]: controll only in terms
t of the local water heigght
or , rrespectively. To
T this end we w
(17)
apply thet following derivation. We replace in thhe
previouus formulationn the terms of annd
Theorem 1. We assum me that and by the boundary conditionns (15) and thhe
, then therre exists a posiitive number relaxatiion approximaation (14). We W obtain up to
, such thaat if
order and annd
and hence the followinng
there existss , , and such that for equatioon for the contrrol and , respectively,
r
any initiial conditionn in
satisfying the
t boundary conditions
c (15)-(16), , (21)
,
(222)
,
, Under the assumptionn and using thhe
previouus results we expect these controls
c to steeer
, the syystem to rest. Note, that using
u boundaary
and additioonally conditions (16) a ddifferent contrroller could bbe
, d.
derived
the closedd–loop system (12)-(13) witth boundary
III. COONCLUSIONS
conditions (15)-(16) haas a unique solution in
We preesented a noveel idea for boun ndary control of
. This solution satisfies for
hyperbolic systems. Contrary
C to exxisting results no
n
any
linearizzation is needeed in order to obtain a contrrol
result. Future investtigations will be carried out o
consideering generaal conservatioon laws annd
The prooof relies on a particular choice of a numeriical studies.
Lyapunov function, seee Chapter 13.3 [3]. The

27
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

ACKNOWLEDGMENTS [9]. M. Gugat and M. Herty, Existence of

This work has been supported by DAAD classical solutions and feedback stabilization
Partnership Program and the German Research for the flow in gas networks, ESAIM, Control
Foundation HE5386/8-1. Optim. Calc. Var., (2010).
[10]. M. Gugat and G. Leugering, Global boundary
REFERENCES controllability of the de st. venant equations
[1]. D. Areba-Driollet and R. Natalini, between steady states, Annales de l’Institut
Convergence of relaxation schemes for Henri Poincar´e, Non- linear Analysis, 20
conservation laws, Applicable Analysis, 61 (2003), pp. 1–11.
(1996), pp. 163–193. [11]. M. Gugat, G. Leugering, K. Schittkowski,
[2]. M. K. Banda and M. Seaid, Relaxation weno and E. Schmidt, Modelling, stabilization, and
schemes for multidimensional hyper- bolic control of flow in networks of open channels,
systems of conservation laws, Nu- mer. in Online optimization of large scale
Methods Partial Differ. Equations, 23 (2007), systems, M. e. a. Groetschel, ed., Springer,
pp. 1211–1234. Berlin, 2001, pp. 251–270.
[3]. J.-M. Coron, Control and nonlinearity, [12]. M. Herty, M. Gugat, A. Klar, and G.
American Mathematical Society, 2007. Leugering, Conservation law constrained
[4]. J.-M. Coron, G. Bastin, and B. optimization based upon front tracking,
d’Andrea Novel, Dissipative boundary M2AN Math. Model. Numer. Anal., 40
conditions for one-dimensional nonlinear (2006), pp. 939–960.
hyperbolic systems, SIAM J. Control. Optim., [13]. M. Herty and S. Veelken, Boundary control
47 (2008), pp. 1460–1498. of nonlinear hyperbolic equations, preprint,
[5]. J. M. Coron, B. d’Andrea Novel, and G. (2010).
Bastin, A lyapunov approach to control [14]. S. Jin and Z. Xin, The relaxation schemes for
irrigation canals modeled by saint venant systems of conservation laws in arbitrary
equations, in ECC Karlsruhe, 1999. space dimensions, Comm. Pure Appl. Math.,
[6]. J.-M. Coron, B. d’Andr´ea Novel, and G. 48 (1995), pp. 235–276.
Bastin, A strict Lyapunov function for [15]. G. Leugering and E. J. P. G. Schmidt, On the
boundary control of hyperbolic systems of modelling and stabilization of flows in
conservation laws, IEEE Trans. Automat. networks of open canals, SIAM J. Control
Control, 52 (2007), pp. 2–11. Optim., 41 (2002), pp. 164–180.
[7]. J. de Halleux, C. Prieur, J.-M. Coron, B. [16]. H. L. Liu and G. Warnecke, Convergence
d’Andr´ea Novel, and G. Bastin, Boundary rates for relaxation schemes approximating
feedback control in networks of open conservation laws, SIAM J. Numer. Anal.,
channels, Automatica J. IFAC, 39 (2003), pp. 37 (2000), pp. 1316–1337.
1365–1376. [17]. R. Natalini, Convergence to equilibrium for
[8]. M. Gugat, Boundary controllability between the relaxation approximations of
sub and supercritical flow, SIAM Journal on conservation laws, Comm. Pure Appl. Math.,
Control and Optimization, 42 (2003), pp. 49 (1996), pp. 795–823.
1056–1070.

28
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Isogeometric methods for shape modeling and numerical simulation

Bernard Mourrain(1)
(1)
GALAAD, INRIA Méditerranée, Sophia Antipolis, France
Email: mourrain@sophia.inria.fr
Gang Xu(2)
(2)
GALAAD, INRIA Méditerranée, Sophia Antipolis, France
Email: Gang.Xu@sophia.inria.fr

ABSTRACT differential equations that describe the

The topic of this paper is to give an overview of a physical phenom- ena we want to analyse.
recent approach, called isogeometric analysis, that
aims at a seamless integration of geometric This has two important consequences. Firstly,
modeling and numerical computational. It is a a conversion step is needed to go from one
uniform frame- work to describe both the geometry
representation to another, which might deviate
representation and approximate solutions of a
simulation problem on this geometry. It rises corresponding performance analysis.
interesting geometric problems, that we review. We Secondly, this transformation needs to be
describe the general framework of this approach, tightly connected with design parameters when
the interesting properties of B-spline bases, that it one want to optimize the geometry with
exploits. After showing on an example of 3D heat respect to performance analysis.
conduction problem hwo it works, we discuss
geometric issues including the parametrization of The topic of this paper is to give a brief
computational domains and its impact on quality of overview of recent developments, which
approximation, refinement techniques which allow tackle these problems. The approach uses the
to extend the function basis and to develop
same type of mathematical (piecewise non-
adaptive methods to improve efficiently the
accurancy of approximation and geometric issues linear) representation, both for the geometry
related to complex topologies. and for the physical solutions, and thus avoid
this costly forth and back transformations.
KEYWORDS Moreover its reduces the number of
Isogeometric analysis; geometric modeling; parameters needed to describe the geometry,
physical simulation; B-spline; T-spline; simplex which is of particular interest for shape
spline. optimisation.

I. INTRODUCTION This approach was introduced by T. Hughes

In engineer design and simulation of physical and his collaborators under the name of
phenomena, geometry plays an important role. isogeometry in the context of PDE problems
The shape of an object directly influences the [1]. This uniform framework provides more
functionalities that we expect from it. Consider accurate and efficient ways to deal with
just as examples the properler of a ship, the complex shapes and to approximate the
wing of a plane or the structure of a solutions of physical simulation problems. But
mechanical piece in a car engine. Its it also rises interesting geometric problems for
performances (force, drag, resistance) are the representation of shapes and functions on
directly related to its shape. To analyze and shapes that we will describe.
optimize these performances, numerical
simulations are usually performed. In the To present the general idea of this approach,
design process, these objects are usually we simplify the context and consider a surface
described by CAGD tools, which involve patch of
3
on which we want to solve a
parametric non-linear models using bspline differential equations of the form
functions. But in the simulation process, D f x = 0 for x
usually surface or volume discrete meshes are
with boundary conditions:
used to approximate the solutions of partial
N f x = f 0 x for x ∂ OMEGA.

29
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Instead of directly approximating the function

f x on the domain ,
N j ,0 t = { 1 if t j ≤ t t j 1
0 otherwise
 we first parametrize the physical
domain by a computational t− t j
domain D (here a rectangle) by a N j ,n t = N t
t j n− t j j , n− 1
map : D
t j n 1− t
 And then we compute the solution N t
= f ° induced by partial
t j n 1 − t j 1 j 1,n− 1
differential equations with boundary
B-spline cuves can be defined as follows,
conditions on D through the map .
m
P t = ∑ P i N i ,n t
i= 0

where P i are called control points. B-spline

surfaces and volumes can be defined in a
tensor product way,
m l
P u ,v = ∑ ∑ Pi , j N i ,n u N j ,n v
i= 0 j= 0

D m , l ,q

Figure 1. Parametrization of the physical P u ,v ,w = ∑ P i , j , k N i ,n u N j ,n v N k ,n w

i , j ,k = 0
domain.

Therefore the solution f is defined implicitly B-spline representation has many interesting
−1
on by f x = ° x . This method properties, such as local support, partition of
n− 1
naturally extends to cases where the physical unity, C continunity and refinement prop
domain is a volume parametrized by a erties [3], which are desired for numeric
3
cube D in . analysis.

The isogeometric approach consists in III. EXAMPLE

choosing the same type of representation for In this section, we show how to solve a 3D
the parametrization map and the actual heat conduction problem with boundary
. conditions, by using an isogeometric method.
solution function Because we are
interested by representing geometry objects Given a domain closed by the boundary
∂ , this physical problem can be described
which are coming from CAGD, a natural
choice is to use B-spline basis functions. In the as the following PDE,
next section, we will recall their definition and T x =f x
in
give their main properties. In the next section, N T x = T 0 x on ∂
we will show on an example, how this is done
in practice. In section IV, we will discuss where x are the Cartesian coordinates, T
some of the geometric issues related to this represents the temperature field. These
approach before the concluding section. boundary conditions that could be of Dirichlet
or Neumann type are applied on the boundary
II.SPLINE REPRESENTATION ∂ of , T 0 being the imposed
Given a nondecreasing sequence of knots temperature. f is a user-defined function that
= {t 0 , t 1, , t h }t i ≤ t i 1 allows to generate problems with an analytical
solution, by adding a source term to the
the B-spline basis of degree n can be defined classical heat conduction equation.
using the Cox-de Boor recursion formula:

30
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

According to a classical variational approach, 3D control points. The solution value is

1
we seek for a solution T H , such that represented by the color information.
T x = T 0 x on ∂ and:
−∫ T x x d
=∫ f x x d 1

According to the isogemetric paradigm, the

temperature field is represented using B-spline
basis functions. For a 3D problem, we have:
m , l ,q
T u ,v ,w = ∑ T i , j ,k N i ,n u N j , n v N k , n w
i , j ,k = 0

where N i , n functions are B-spline basis

functions and p= u , v , w D are domain
parameters. Then, we define the test functions
x Figure 2. Physical domain and its boundary surfaces.
in the physical domain bys:
−1
M ijk x = N ijk ° x
where IV. GEOMETRIC ISSUES
N ijk p = N i ,m u N j ,l v N k , q w The isogeometric approach provides a uniform
The weak formulation Eq. 1 reads: framework to represent the geometry and the
nr ns nt
physical solutions. This simplifies
∑ ∑ ∑ T rst∫ M rst x M ijk x d significantly the computation process
r= 0 s= 0 t= 0
involved in a simulation or optimisation
= −∫ f x M ijk x d problem. But it also rises new questions,
which are of geometric nature.
Finally, we obtain a linear system similar to
that resulting from the classical finite-element
methods, with a matrix and a right-hand side
defined as:
E ijk ,rst = ∫ M rst x M ijk x d =

∫ p N rst p B
T
p B p p N ijk p J p d P

Sijk = − ∫ f x M rst x d
= −∫ f T p N rst p J p dP Figure 3. Color map of the solution and control points.
J p
where is Jacobian of the
T Injectivity Condition. The first geometric
transformation, B is the transposed of the issue is to guarantee the injectivity of the
inverse of the Jacobian matrix. The above paramaterisation map :D . In a usual
integrations are performed in parametric space context, this physical domain is described by
using classical Gauss quadrature rules. boundary curves or surfaces (see Figure 2);
The computational domain needs to be
An example is given in Figure 2 and Figure 3 parametrised such that the parametrisation
with
2 coincides on the boundary of D , with the
− x y z
f x= sin sin sin parametrisation of the boundary surfaces.
3 3 3 3 . Using B-spline representations, the control
Figure 2 shows the physical domain and its coefficients are known on the boundary. In
boundary surfaces. Figure 3 presents the color Figure 3, we find that the outer control points
map of the solution field and corresponding of the volume parameterisation, wich are

31
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

deduced from the control points of the approximate the solution of our problem. An
boundary surfaces. The problem reduces to interesting characteristic of these
find the interior control points such that the approximation schemes is that their order of
map is a bijection between the approximation is directly related to the degree
computation domain D and the physical of regularity of the bspline space. Figure 4 is
an example of a planar heat conduction
domain . The injectivity of is verified problem analysis, where the error is given in
if its Jacobian does not vanish on D . A terms of the log of the square root of the
sufficient condition for injectivity can be number of control points: The two curves
deduced from the relative position of the represent the L2 error of the approximated
control points of the parametrisation. For solution, when we use a natural position of the
simple shapes, such a parameterisation can be control points (plain blue curve) and we when
constructed from so-called Coons pacthes [2]. we optimize their position (dash red curve).
For more complex shapes, a solution can be We observe that in both cases, the slopes of
found using standard linear programming the curves tend to -4, which is the speed of
techniques on the position of the controlled approximation we expect using bicubic bspline
points. Such an approach is described in [4]. functions.
As a matter of fact, the choice of the free inner
control points has an influence on the quality Using tensor product B-spline functions has
of approximation of the physical solution. As however some drawback in this context. When
shown in [4], the optimal position of the a knot is inserted in one direction of the
control points is not necessarily the natural (or parameter domain, we add not one but the
regular) one. New types of strategies number control points involved in the other
combining the optimisation of the position of parameter direction.
the inner control points with approximation of
the solution and the estimation of the error can This can lead to too many knot insertions if h-
be considered. refinement operations are required in all
parametric directions. Instead, we would like
to have local parameter space refinements
possibilities. To handle this problem, new
types of B-spline basis are considered. So-
called T-splines, introduced by T.W.
Sederberg et al. [5], are a generalisation of
rational bspline functions, which are
associated to rectangular subdivisions of the
parametric domain. These subdivisions allow
to have T-junctions and to perform local
refinements. They have interesting properties
for the isogeometric approach [6]. Other types
of T-splines which are piecewise polynomial
[7] are also considered in isogeometric
problems [8]. These spaces of T-splines are
Figure 4. Error history during refinement for different
parametrization of computational domain
not completely under- stood. In particular,
open questions remain on their dimension and
Function space refinement. A standard and the construction of explicit bases. Another
tranditional technique to improve the quality family of B-spline functions is related to the
of approximation is to refine the triangular control meshes. These splines
computational domain. This process, also extend the concept of simplex splines to a
called h-refinement, consists to insert knots in triangular mesh, attaching a sequence of nodes
the parametric domain, that is to add new to each vertex of the triangulation [9]. They
control points. Note that it does not change the allow to deal with 2D or 3D domains with
parametrisation map but increase the space of arbitrary topology, and with arbitrary degree
(B-spline) functions used to represent it. Thus of regularity. Having the possibility to perform
it provides more freedom to better local refinement with these types of bspline
functions is important in the isogeometric

32
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

approach. To fully exploit these capacities, This recent approach rises interesting
efficient local error estimators are however geometric modeling and representation
needed, which remains a difficult issue in challenges, that need to be addressed for
numerical analysis, whatever the approach further impact of isogeometry in scientific
chosen to approximate the physical solution. computing. This also implies some deep
changes in the numerical tools and techniques
Multipatches. The geometry on which we involved in numerical computation, which is
want to perform the simulation may not be another challenge to address.
composed of one part that can be parametrized
by a simple domain D . It may have holes or ACKNOWLEDGEMENTS
different pieces assembled in a non-manifold We thank Dr. Régis Duvigneau for interesting
discussions on finite element techniques and
on shape optimisation.

REFERENCES
[1]. A.Cottrell, T.J.R.Hughes, Y.Bazilevs. Iso-
geometric Analysis: Toward Integration
of CAD and FEA. Wiley Press, 2009.
[2]. Farin, D. Hansford. Discrete coons patches
Computer Aided Geometric Design,
1999,16(7):691-700
[3] Farin. Curves and surfaces for computer
way: aided geometric design : a practical guide.
This type of geometry requires a special treat- Comp. Science and sci. computing. Acad.
ment for the description of the Press, 1990.
parametrisations of the different parts of the [4]. G.Xu, B.Mourrain, R.Duvigneau, A.
object and the constraints that should be Galligo. Optimal analysis-aware
satisfied along the boundary of these different parameterization of computational domain
components. The topological structure and the in isogeometric anal- ysis. Geometric
geometric and functional basis description Modeling and Processing, LNCS 6130,
should be tightly linked in order to provide an 2010: 236-254.
efficient solution to the simulation problem. In [5]. W. Sederberg, J.M Zheng, A.Bakenov, A.
particular, assembling the (stiffness) matrix E H. Nasri. T-splines and T-NURCCs. ACM
should be optimized according to the support Trans. Graph, 2003, 22(3): 477-484
of the basis functions. [6]. Bazilevs, V.M. Calo, J.A. Cottrell, J.A.
Evans, T.J.R Hughes, S. Lipton, M.A.
V.CONCLUSION Scott, T.W. Sederberg. Isogeometric
The isogeometric approach is a promosing Analysis using T-Splines, Computer
technique which represents in the same Methods in Applied Mechanics and
framework the geometry and for the physical Engineering, 2010,Volume 199, Issues 5-
functions on the geometry. By representing 8: 229-263.
exactly the geometry, it avoids some [7]. X. Li, J.S Deng, F.L Chen. Surface
numerical artefacts that can appear in finite modeling with polynomial splines over
element method with mesh approximation. It hierarchical T-meshes. The Visual
also leads to high order numerical Computer ,2007, 23(12): 1027-1033
approximation scheme, using basis functions
such as splines. These piecewise polynomials [8]. Dörfel, B. Jüttler, and B. Simeon. Adap-
functions which are heavily used in CAGD tive isogeometric analysis by local h-
provide a uniform framework to describe the refinement with T-splines. Computer
geometry and the solutions. Tranditional finite Methods in Applied Mechanics and
element techniques extend naturally to this Engineering,2010, 199(5-8): 264-275
new framework. Shape optimisation methods [9]. G. Xu, G.Z Wang, X.D Chen. Free-form
can be applied more efficiently by moving def-ormation with rational DMS-spline
control points instead of nodes on the finite volumes. Journal of Computer Science
element mesh. and Technology, 2008,23(5): 862-873.

33
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

FOURTH-ORDER QSMSOR ITERATIVE METHOD FOR THE

SOLUTION OF ONE-DIMENSIONAL PARABOLIC PDE’S
 
1
J. Sulaiman, 2M.K. Hasan, 3M. Othman, and 4S. A. Abdul Karim,
1
Mathematics with Economics Programme
Universiti Malaysia Sabah, Locked Bag 2073,
88999 Kota Kinabalu, Sabah, Malaysia
e-mail: 1jumat@ums.edu.my
2
Dept. of Industrial Computing, Universiti Kebangsaan Malaysia,
43600 UKM Bangi, Selangor, Malaysia
e-mail: 2khatim@ftsm.ukm.my
5
Dept. of Communication Technology and Network, Universiti Putra Malaysia,
43400 UPM Serdang, Selangor, Malaysia
e-mail: 3mothman@fsktm.upm.edu.my
4
Fundamental and Applied Sciences Department,
Universiti Teknologi Petronas, Bandar Seri Iskandar,
31750 Tronoh, Perak Darul Ridzuan, Malaysia.
e-mail: 4 samsul_ariffin@petronas.com.my

ABSTRACT equations mainly in partial differential

Presently, many science and engineering problems equations. Recently many researchers have
have been described in mathematical models via proposed low- and high-order approximation
partial differential equations. The solutions of these equations obtained by discretization of existent
problems can be obtained analytically or differential equations in mathematical models.
numerically. In previous studies, the concept of
Both second- and fourth-order difference
quarter-sweep scheme has been pointed out to
accelerate the convergence rate in solving any schemes will lead to a linear system with
system of linear equations generated by using different properties of their coefficient matrix.
approximation equations of boundary value For case of high-order, formulations on the
problems. Based on the same concept, The primary fourth-order approximation equations have
objective of this paper is to investigate the been proposed and discussed by Spotz [10]
application the fourth-order finite difference solver and Gupta et al. [4,5] in order to get more
using the Quarter-Sweep Modified Successive accurate approximate solutions. Each high-
Over Relaxation (QSMSOR) iterative method to order approximation equation will also lead a
solve one-dimensional parabolic partial differential system of linear equations, where the
equations. Consequently, the fourth-order finite-
characteristic and complexity of its coefficient
difference approximation equation of proposed will
be derived by using the Crank-Nicolson scheme. In matrix are different. Instead of standard and
fact, the formulation and implementation of Full- compact fourth-order schemes, a new fourth-
Sweep, and Half-Sweep Successive Over order difference method so called, arithmetic
Relaxation iterative methods, namely FSSOR and average discretization has been presented
HSSOR respectively are also presented. recently for the solution of two dimensional
Throughout implementations of some numerical non-linear singularly perturbed elliptic partial
experiments conducted, it has shown that the differential equations, see in Mohanty [7, 8]
fourth-order solver using the QSMSOR method is and Mohanty and Singh [9].
superior as compared with other SOR methods. The main purpose of this paper is to
examine the efficiency of the fourth-order
quarter-sweep finite difference method for
KEYWORDS
Quarter-Sweep Iterative, SOR Scheme, Fourth- solving one-dimensional diffusion equations.
Order Finite Difference, 1D Parabolic PDE’s Besides the quarter-sweep approach,
formulation of the full- and half-sweep second
order schemes also need to be considered for
I. INTRODUCTION implementing the quarter-sweep iterations.
Science and engineering problems can This is because of implementations of the
always being modelled by differential

34
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

quarter-sweep iterative algorithms will be quarter-sweep iterative algorithms applied

performed together with the full- and half- onto the solid node points of type only until
sweep approaches. In this paper, we also iterative convergence is achieved. Then other
examine the Quarter-Sweep Modified approximate solutions at remaining points
Successive Over Relaxation (QSMSOR) (points of the different type) will be obatined
method for solving one-dimensional diffusion by using the direct method, see Abdullah [1],
equations by using the quarter-sweep fourth- Ibrahim and Abdullah [6], Sulaiman et al. [11,
order finite difference approximation 13].
equation.. This QSMSOR method will be Because of implementations of the full-,
compared to the FSSOR, HSSOR and QSSOR half-, and quarter-sweep iterations involve
methods. the solid node points of type only, it can
To investigate the effectiveness of the be obviously seen that the implementation of
QSMSOR method based on the fourth-order the half- and quarter-sweep iterative methods
discretization scheme, let us consider the one- just involves approximately 50% and 25% of
dimensional diffusion equation as given by the whole inner points as shown in Figures 1b
U 2U and 1c compared to the full-sweep iterative
 , a 0  x  b 0 , 0  t  T, method in Figure 1a.
t x 2
(1)
subject to the initial condition
Ux, t   g1 x , a 0  x  b0 ,
and the boundary conditions a).
Ua 0 , t   g 2 t 
 0tT
Ub 0 , t   g 2 t 
where α is a diffusion parameter.
Before constructing the formulation of the b).
finite difference approximation equation in
case of the full-, half-, and quarter-sweep
iterations over the problem (1), we assume the
solution domain (1) can be uniformly c).
partitioned into (n+1) and M subintervals in Figure 1. a)., b).and c). show the distribution of uniform
node points for the full-, half-, and quarter-sweep cases
the x and t directions. The subintervals in the x respectively.
and t directions are denoted x and t
respectively, which are uniform and defined as By using the θ iterated scheme, the general
b  a0  second-order approximation equation for
x  0  h , m  n  1
m (2) problem (1) can be easily given as
T0   U i1, j1  1  2U i, j1
t   (3)
M   U i1, j1  f i, j1
where,
f i, j1  1  U i1, j
II. Formulation of Fourth-Order ,
 1  21  U i, j  1  U i1, j
Quarter-Sweep Finite Difference
Approximation t
 .
Referring in Figure 1 and to facilitate us h2
for implementing full-, half-, and quarter- The values of θ in Eq. (3), correspond to
sweep iterations, we firstly need to construct 0, 1, and 1 2 , indicates the explicit
three finite grid networks in order to derive the
(classical), fully implicit and the Crank-
full-, half-, and quarter-sweep finite difference
Nicolson (CN) schemes, respectively. In fact,
approximation equations by discretizing the
the truncation errors for each scheme can be
problem (1). Actually, these networks show us
on implementation of the full-, half-, and   
shown as O x   t  , O x   t  ,
2 2

35
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

 2 2

and O x   t  , respectively. Due to
The value of p, which corresponds to 1, 2 and
high accuracy, the second-order CN scheme is 4, represents the full-sweep, half-sweep and
mainly considered to derive an approximation half-sweep cases respectively. Based on
equation of problem (1). Similarly using the Figure 2, it can be seen that we can not
same way to derive Eq. (3) and by substituting calculate approximate values for two node
  1 2 , we can construct the Full-Sweep, points, i  p and i  m  p by using Eq.(5),
Half-Sweep and Quarter-Sweep Crank- because some node points are out of the
Nicolson finite difference approximation solution domain (1). To overcome this
equations, which are indicated as the FSCN, problem, we need to consider the correspoding
HSCN, and QSCN respectively. The second-order approximation equation in Eq.
formulation of all CN finite difference (4). It means that both node points will be
approximation equations at the (j+1) time calculated by Eq. (4).
level, can generally be expressed as Therefore, combination between Eq. (4)
 and (5) can easily lead to a linear system in
 U ip, j1  1   U i, j1
2 matrix form as
 (4) AU  b (6)
 U i p, j1  f i, j1 ~ ~
2 where,
Where
1   0 
 t  2 
 ,
ph 2  b a b c 
 c b a b c 
   
f i, j1  U i  p, j  1   U i, j  U i  p, j
2 2 A     
 c b a b c 
By following the same steps to construct 
second-order CN approximation equation in  c b a b 
(4), we can show on how to derive that the   
0  1 
fourth-order full-sweep, half-sweep and  2 
 m   m 
quarter-sweep finite difference approximation  1x 1
 p    p  
equations and then these approximation   
equations can generally be stated as
cU ip, j1  bU i p, j1  aU i, j1 III. Derivation of the Family of SOR
Methods
 bU i p, j1  cU i p, j1  f i, j1 As can be seen the linear system in Eq.
(5) (6), the coeiffient matrix, A is sparse and large
Where scale. Therefore iterative methods are the
f i, j1  cU i  p, j  bU i  p, j  dU i, j natural options for efficient solutions of
sparse linear system. In this section, we will
 bU i  p, j  cU i  p, j discuss on how to construct various types of
 t 16t the Successive over-relaxation (SOR) method
c , b , introduced by Young [16]. Actually, this
24ph 2 24ph 2 method can be also called the Gauss-Seidel
30t 30t method with a weighted parameter, w is
a 1 , d 1 . widely used to accelerate the convergence rate
24 ph 2 24 ph 2 of the standard Gauss-Seidel method for
solving a linear system of equations.
Now, we construct and formulate FSSOR,
HSSOR, QSSOR and QSMSOR iterative
methods based on linear system in Eq. (6).
Since the linear system can be represented in a
general matrix form for the case of full-sweep,
Figure 2. Distribution of node points for the
computational molecule of the fourth-order full-sweep half-sweep and quarter-sweep. Let us consider
and half-sweep schemes at i  p .
linear system in Eq. (6) as

36
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

AUb (7) approximation equation in (4). There are three

~ ~ items will be considered in comparison such as
Then we need to decompose the coefficient the number of iterations, execution time and
matrix, A in Eq. (7) as follow maximum absolute error. Some numerical
A  LDT experiments were demonstrated in solving the
where L, D and T are lower triangular, following one-dimensional diffusion equation
diagonal and upper triangular matrices as follows
respectively. Therefore, we can write the
general scheme for all SOR methods as follow U  2 U
 , 0  x  1, 0  t  1.0 (10)
(Young [16, 17, 18, 19]) t x 2
U k 1  D  wL1 wT  1  w D  U k   w b  (8) The initial and boundary conditions and exact
~  ~ ~ solution of the problem (10) are given by
where w and Uk  represent as a relaxation
~  2t   42t 
factor and an unknown vector at the k th U( x , t )  e   sinx   3e   sin 2x 
iteration respectively. The choice of
All results of numerical experiments, which
relaxation factor depends upon the properties
were gained from implementations of the
of the coefficient matrix, A. In addition, a
FSSOR, HSSOR, QSSOR and QSMSOR
good choice of parameter can improve the
convergence rate of iteration process. In methods have been tabulated in Table 1. In
practice, the optimal value of w in range implementations mentioned above, the
convergence test considered the tolerance error
1  w  2 will be obtained by implementing
  1010 . Figures 3 and 4 show number of
several computer programs and then the best
iterations and execution time against mesh size
approximate value of w is chosen in which its
number of iterations is the smallest. However, respectively.
a theoretical optimal value of w can be Through numerical solutions obtained in
calculated by the formula: Table 1, it clearly shows that applications of
2 the half- and quarter-sweep iterative methods
w opt  reduces number of iterations and
1  1  2
computational time as as compared to FSSOR
where  is the spectral radius of the iteration iterative method (refer Table 2 and Figures 3
matrix. and 4).
Actually the formulation of the Modified
Successive Over-Relaxation (MSOR) is the Table 1. Comparison of a number of iterations, the
same with Successive Over-Relaxation (SOR) execution time (seconds) and maximum errors for the
iterative methods.
method by imposing the concept of red black No. of Iterations
ordering with the two different relaxation Methods Mesh size
parameter factors such as " w" for the “red” 1024 2048 4096 8192 16384
FSSOR 663 1224 2245 4185 7927
'
equation and “ w ” relaxation parameter HSSOR 356 663 1224 2245 4185
QSSOR 192 356 663 1224 2245
factor for “black equation”. As a result, the QSMSOR 120 225 417 772 1414
iterative scheme for QSMSOR method is Execution time (Seconds)
given by Methods Mesh size
1024 2048 4096 8192 16384
U k 1  D  wL 1  wT  1  w D  U k   w b 
  FSSOR 5.19 20.03 77.85 288.91 1064.61
~  ~ ~ (9) HSSOR 1.19 4.42 17.32 66.82 247.70
QSSOR 0.56 1.89 7.32 28.10 108.08
U k 1  D  w  L 1  w  T  1  w D  U k   w b 
 
QSMSOR 0.26 0.77 2.83 10.62 40.48
~  ~ ~ Maximum Absolute Errors
If choosing w  1 , we get simple Gauss-Seidel Methods Mesh size
1024 2048 4096 8192 16384
(GS) method. FSSOR 4.12e-7 4.13e-7 4.23e-7 5.54e-7 9.58e-7
HSSOR 4.13e-7 4.13e-7 4.13e-7 4.23e-7 5.54e-7
QSSOR 4.13e-7 4.13e-7 4.13e-7 4.13e-7 4.23e-7
QSMSOR 4.01e-7 3.89e-7 3.65e-7 3.17e-7 2.26e-7
IV. Computational Experiments
In this section, we have conducted some
numerical simulations in order to examine the
effectiveness of the QSMSOR method by
using the fourth-order QSCN finite difference

37
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

execution time than the FSSOR, HSSOR, or

QSSOR method. This is as a result of the
computational complexity of the QSSOR
method is approximately 75% less than of the
FSSOR method. For future works, the
capability of the proposed technique will be
examined for the further investigation in
solving any multi-dimensional partial
differential equation (Evans & Sahimi [2];
Abdullah [1]; Ibrahim & Abdullah [6];
Sulaiman et. al [12] and being used as a
smoother in multigrid solvers, see in Evans
Figure 3. Number of iterations versus mesh size of the
FSSOR, HSSOR, QSSOR and QSMSOR methods. and Yousif [3]; Othman and Abdullah [15];
Sulaiman et. al [14].

REFERENCES
[1]. Abdullah, A.R. (1991). ”The Four Point
Explicit Decoupled Group (EDG)
Method: A Fast Poisson Solver”.
International Journal Computer
Mathematics, Vol 38, pp 61-70.
[2]. Evans, D.J & Sahimi, M.S. (1988). ”The
Alternating Group Explicit iterative
method (AGE) to solve parabolic and
hyperbolic partial differential equations”.
Ann. Rev. Num. Fluid Mechanic and Heat
Figure 4. The execution time (seconds) versus mesh size Trans, Vol 2, pp 283-389
of the FSSOR, HSSOR, QSSOR and QSMSOR methods. [3]. Evans, D. J & Yousif, W. S. (1990). ”The
Explicit Block Relaxation method as a
Table 2. Reduction percentage of the number of grid smoother in the Multigrid V-cycle
iterations and execution time for the HSSOR, QSSOR
scheme”. Int. J. Computer Maths. , Vol
and QSMSOR methods compared with FSSOR method
34, pp 71-78.
Methods [4]. Gupta, M. M., Kouatchou, J. & Zhang, J.
Problem (1)
(1997a). ”A Compact Multigrid solver for
Number of Execution time convection-diffusion equations”. Journal
iterations of Computational Physics, Vol 132, No 1,
HSSOR 45.47 - 47.21% 76.73 - 77.93%
70.47 - 70.92% 89.21 - 90.60%
pp 123-129.
QSSOR
QSMSOR 81.43 - 82.16% 94.99 - 96.36% [5]. Gupta, M. M., Kouatchou, J. & Zhang, J.
(1997b). ”Comparison of second and
fourth order discretizations for multigrid
V. CONCLUSIONS Poisson solvers”. Journal of
In this paper, we present the formulation Computational Physics, Vol 132, No 2,
of the the FSSOR, HSSOR, QSSOR and pp 226-232.
QSMSOR methods by using the corresponding [6]. Ibrahim, A. & Abdullah, A.R. (1995).
fourth-order Crank-Nicolson finite difference ”Solving the two-dimensional diffusion
schemes in Eq. 5. Based on fourth-order equation by the four point explicit
discretization schemes for all sweep cases, we decoupled group (EDG) iterative
method”. International Journal Computer
can see that the corresponding second-order
Mathematics, Vol 58, pp 253-256.
approximation equation needs to be applied
[7]. Mohanty, R.K. (1997). ”Order h4
together at two node points, i  p and
difference methods for a class of singular
i  m  p . From the observation of numerical
two space elliptic boundary value
collected in Tables 1 and 3, we can conclude
problems”. Journal of Computational
that the QSSOR method is more superior in
terms of number of iterations and the

38
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

and Applied Mathematics, Vol 81, pp of Elliptic Type”, Trans. Amer. Math.
229-247 Soc., Vol 76, 92-111.
[8]. Mohanty, R.K. (2007). ”The smart- [17]. Young, D.M. (1971). ”Iterative solution
BLAGE algorithm for singularly of large linear systems”. London:
perturbed 2D elliptic partial differential Academic Press.
equations”. Applied Mathematics and [18]. Young, D.M. (1972). ”Second-degree
Computation, Vol 190, pp 321–331. iterative methods for the solution of large
[9]. Mohanty, R.K. & Singh, S. (2006). ”A linear systems”. Journal of
new fourth order discretization for Approximation Theory, Vol 5, pp 37-148.
singularly perturbed two dimensional [19]. Young, D.M. (1976). ”Iterative
non-linear elliptic boundary value solution of linear systems arising from
problems”. Applied Mathematics and finite element techniques”. In:
Computation, Vol 175, pp 1400–1414. Whiteman, J.R. (Eds.). The
[10]. Spotz, W. F. (1995). ”High-Order Mathematics of Finite Elements and
Compact Finite Difference Schemes for
Computational Mechanics”. Ph.D
Applications II, pp 439-464.
Dissertation. The University of Texas at London: Academic Press.
Austin.
[11]. Sulaiman, J., Hasan, M.K. & Othman, M.
Othman. (2007). ”Red-Black EDGSOR
Iterative Method Using Triangle Element
Approximation for 2D Poisson
Equations”. In. O. Gervasi & M.
Gavrilova (Eds). Computational Science
and Its Application 2007. (LNCS 4707),
pp 298-308. Berlin: Springer-Verlag.
[12]. Sulaiman, J., Hasan, M.K. & Othman,
M.(2007). ”Red-Black Half-Sweep
Iterative Method Using Triangle Finite
Element Approximation for 2D Poisson
Equations”. In. Y. Shi et al. (Eds).
Computational Science 2007. (LNCS
4487), pp 326-333. Berlin: Springer-
Verlag.
[13].Sulaiman, J., Othman, M. & Hasan, M.K.
(2004). ”Quarter-Sweep Iterative
Alternating Decomposition Explicit
algorithm applied to diffusion
equations”. International Journal of
Computer Mathematics, Vol 81, No 12,
pp 1559-1565.
[14]. Sulaiman, J., Othman, M. & Hasan, M.K.
(2008). ”Half-Sweep Algebraic Multigrid
(HSAMG) method applied to diffusion
equations”. 2008. In. H.G. Bock et al.
(Eds). Modeling, Simulation and
Optimization of Complex Processes, pp
547-556. Berlin: Springer-Verlag.
[15]. Othman, M. & Abdullah, A.R. (1999).
”An Effcient Multigrid Poisson Solver”.
Int. J. Computer Maths., Vol 71, pp 541-
553.
[16]. Young, D. M. (1954). ”Iterative Methods
for solving Partial Difference Equations

39
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

A Parallel Accelerated Over-Relaxation Quarter-Sweep Point Iterative

Algorithm for Solving the Poisson Equation

Mohamed Othman,(1,2) Shukhrat I. Rakhimov,(2) Mohamed Suleiman(2) and Jumat Sulaiman(3)

(1)
Dept of Communication Tech and Network,
Universiti Putra Malaysia, 43400 UPM Serdang, Selangor D.E., Malaysia
(Email: mothman@fsktm.upm.edu.my)
(2)
Institute for Mathematical Research,
Universiti Putra Malaysia, 43400 UPM Serdang, Selangor D.E., Malaysia
(Email: sh.rakhimov@gmail.com)
(3)
School of Science and Technology,
University Malaysia Sabah, 88999 Kota Kinabalu, Sabah, Malaysia
(Email: jumat@ums.edu.my)

ABSTRACT A number of numerical methods have been

Over-relaxation methods are proved to be formulated to solve the equation. Most of the
efficiently solve a large system of linear equations, methods are based on the concept of
arising from discretizations of scientific and Successive Over-Relaxation (SOR),
engineering problems. Recently, a new accelerated introduced by Young [27], and Accelerated
over-relaxation (AOR) quarter-sweep point-
Over-Relaxation (AOR), introduced by
iterative method was shown to be effective in
solving the Poisson equation, as compared to full- Hadjidimos [23]. Based on SOR, the parallel
and half-sweep approach. In this paper, we present full-sweep point-iterative algorithm for
the parallel implementation of the AOR quarter- solvign large sparse linear systems was
sweep method on a distributed memory invented by Evans [22]. The half-sweep
architecture machine. To confirm its effectness, the approach for the derivation of Explicit De-
experimental results of a test problem which are coupled Group method was introduced by
recoreded and analyzed, show that the parallel Abdullah, see [20]. Subsequently, parallel
AOR quater sweep algorithm is superior as implementation of the EDG method was
compared to the previous parallel AOR full- and developed by Yousif et al. [28]. The SOR
half-sweep algorithms.
quarter-sweep point/group-iterative methods,
which were shown to be superior over the full-
KEYWORDS
Accelerated over-relaxation, point iterative sweep and half-sweep point iterative methods,
methods, Poisson equation, distributed memory was introduced by Othman et al. [24]. At the
architecture, parallel algorithm. same time, the concept of AOR was applied to
full- and half-sweep approaches by Ali et al.
I. INTRODUCTION [21].
Many scientific and engineering problems can
be modeled mathematically as differential In a recent paper [26], authors had developed
equations. The numerical techniques are used the AOR quarter-sweep point iterative method,
to solve a large size problem since the rapid which was shown to be more efficient than the
growth of computer technology. In the case of AOR methods with full- and half-sweep
stationary type of problem, two dimensional approaches. In this paper we introduce the
Poisson's equation is the most popular type of parallel implementation of the AOR quarter-
equation and represented as follows: sweep point iterative method, based on
 2u  2u distributed memory architecture machine.
  f  x, y  (1)
x 2 y 2 II. THE AOR QUARTER-SWEEP POINT
with Dirichlet boundary conditions: ITERATIVE METHOD
u ( x, y )  g ( x, y ) and ( x, y )   . For In the AOR quarter-sweep point iterative
simplicity, we assume the solution domain method, the domain points are divided into
 to be a square unit [0  x, y  1] . three types: , , and , as shown in
Figure 1.

40
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

III. THE PARALLEL AOR QUARTER-

SWEEP POINT METHOD
In [26], it was shown that the chessboard
ordering strategy is the most optimal strategy
for the new method. Therefore, the strategy is
used in the parallel implementation.

A. Domain Decomposition
Since the points in the solution domain are
interdependent, the domain decomposition
mechanism is used to solve the problem of
data dependent at the interboundary. The
number of points at the interboundary must be
very minimal. Let’s denote the number of
processes p and the solution domain is
decomposed evenly into p vertical sub-
Figure 1. The solution domain for the AOR quarter- domains, as shown in Figure 2.
sweep point-iterative method.

The method is based on the five points

quarter-sweep finite difference scheme,
vi2, j  vi2, j  vi, j 2  vi, j 2  4vi, j  4h2 fi, j (2)

The algorithm of the method is as follows.

1. Divide the grid points into the three types

as in Figure 1. Compute the values of h 2 ,
2h 2 , and 4h 2 beforehand and assign
them to variables H , I , and J .
2. For all the points of type , implement
the accelerated over-relaxation:
 vi(k2,1)j  vi(k2,) j  vi(,kj1)2  vi(,kj) 2 
vi , j  r  
 4
 


( k 1)

  v i , j  vi(,kj)  vi(,kj) , Figure 2. Domain decomposition for the parallel AOR
quarter-sweep method with n = 18 and two processors
where (p=2).
1
  vi(k2,1)j  vi(,kj1)2  vi(k2,) j  vi(,kj) 2  J  f i , j 
( k 1)
v i, j
4
3. Check for the convergence. If
the convergence is not achieved, go to B. Interprocess Communication
the step 2. Otherwise, evaluate solutions In chessboard ordering strategy, there are two
for the other points by the following stages, red and black. The stages depend on
formulas: each other, update the data for each stage are
1
  vi 1, j 1  vi 1, j 1  vi 1, j 1  vi 1, j 1  I  fi , j 
required, i.e. twice for for each iteration. Only
vi , j
4 the points at the interboundaries are required
for points of type and i.e. each sub-domains are required to update
1 their data (communications among the
vi , j 
4
 vi1, j  vi, j 1  vi1, j  vi, j 1  H  fi, j  processors). The interprocess communication
procedure is called twice for each iteration,
for points of type .
red iteration then followed by black iteration.
4. Stop.
C. Parallel AOR quarter-sweep Algorithm

41
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

The implementation of the parallel AOR 150, 250, 350 and 450. The convergence test
quater sweep points algorithm is describe as, was the maximum absolute error with the error
1. Divide the grid points into the four types tolerance   1010 . The parallel algorithm
, , and as in Figure 2. Compute was implemented in C language on Sun Fire
the values of h 2 , 2h 2 , and 4h 2 V1280 parallel machine with eight numbers of
beforehand and assign them to variables processor. The results for the parallel AOR
H , I , and J . Decompose the domain quarter-sweep point iterative method are
into sub-domains. reported and shown in Table 1.
2. For all the points of type , implement Table 1. Experimental results for the parallel AOR
the accelerated over-relaxation: quarter-sweep point iterative method.
 vi(k2,1)j  vi(k2,) j  vi(,kj1)2  vi(,kj) 2  n p Time (ms) Speedup Effic.
vi , j  r  
 4 1 511.864 1.00 1.00
  2 308.179 1.66 0.83
( k 1)

  v i , j  vi(,kj)  vi(,kj) ,
150
3
4
235.346
194.690
2.17
2.63
0.72
0.66
where 5 179.336 2.85 0.57
1 6 165.988 3.08 0.51
v i , j   vi(k2,1)j  vi(,kj1)2  vi(k2,) j  vi(,kj) 2  J  f i , j 
( k 1)
7 157.103 3.26 0.47
4 8 152.881 3.35 0.42
3. Call the interprocess communication 1 2365.096 1.00 1.00
procedure, sending the undated points at 2 1358.951 1.74 0.87
the interboundaries to the adjacent 3 972.199 2.43 0.81
processors. 4 770.047 3.07 0.77
250
4. Repeat Steps 2−3 for the points of type . 5 682.750 3.46 0.69
5. Check for the convergence. If 6 603.724 3.92 0.65
the convergence is not achieved, go to 7 552.556 4.28 0.61
8 506.630 4.67 0.58
the step 2. Otherwise, evaluate solutions
1 6120.029 1.00 1.00
for the other points by the following 2 3448.256 1.77 0.89
formulas: 3 2426.846 2.52 0.84
1
vi , j   vi 1, j 1  vi 1, j 1  vi 1, j 1  vi 1, j 1  I  fi , j  350
4 1904.208 3.21 0.80
4 5 1655.956 3.70 0.74
for points of type and 6 1429.563 4.28 0.71
1 7 1292.324 4.74 0.68
4
 vi1, j  vi, j 1  vi1, j  vi, j 1  H  fi, j 
vi , j  8 1147.156 5.33 0.67
1 13 289.598 1.00 1.00
for points of type . 2 7 456.084 1.89 0.94
6. Stop. 3 5 212.513 2.70 0.90
4 3 990.711 3.59 0.90
450
IV. NUMERICAL RESULTS AND 5 3 287.753 4.48 0.90
DISCUSSIONS 6 2 832.822 5.28 0.88
To benchmark the parallel AOR quarter-sweep 7 2 511.695 6.25 0.89
point iterative algorithm, we have 8 2 427.652 7.15 0.89
implemented parallel version of the AOR full- The results of execution time for all the
and half-sweep point iterative methods. The parallel AOR full-, half- and quarter-sweep
numerical experiments were conducted for point iterative algorithms for are shown in
solving the following problem, Figure 3.
 2u  2u
   x 2  y 2  e xy ,
x 2 y 2
( x, y )     0,1   0,1
with the Dirichlet boundary conditions. The
exact solution of the problem is u ( x, y )  e
xy

for ( x, y )   . The experiments were

carried out with several mesh size such as

42
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

[21]. Ali, NHM, and Chong, LS (2007).

Group Accelerated OverRelaxation
Methods on Rotated Grid. Applied
Mathematics and Computation,
191(2):533−542.
[22]. Evans, D.J., (1984). Parallel S.O.R.
iterative methods. Parallel Computing 1: 3-
18.
[23]. Hadjidimos, A (1978). Accelerated
OverRelaxation Method. Mathematics of
Computation, 32(141):149−157.
Figure 3. Execution times for the parallel AOR full-, [24]. Othman, M, and Abdullah, A.R.,
half-, and quarter-sweep point iterative methods (n=450). (2000). An Efficient Parallel Quarter-sweep
Point Iterative Algorithm for Solving
It is shown that the parallel AOR quarter- Poisson Equation on SMP Parallel
sweep method is the fastest among the three Computer, Pertanika Journal of Science and
algorithms at any number of processes run to Technology, 8(2):161−174.
solve the problem. [25]. Othman, M, Abdullah, A.R., and
Evans, D.J., (2004). A Parallel Four Points
V. CONCLUSIONS Modified Explicit Group Algorithm on
In this paper, we discussed the implementation Shared Memory Multiprocessors.
of the parallel AOR quarter-sweep point International Journal of Parallel, Emergent
iterative algorithm. The algorithm has shown and Distributed Systems, 19(1):1−9, 2004.
good speedup and efficiency, especially for the [26]. Rakhimov, ShI, and Othman, M.
large size of grid. Again, the algorithm has (2009). An Accelerated Over-Relaxation
shown their superiority over the previous Quarter Sweep Point Iterative Method For
parallel AOR full- and half-sweep algorithms. Two Dimensional Poisson Equation, Sains
Malaysiana, 38(5), pp. 729−733.
ACKNOWLEDGEMENT [27]. Young, D.M. (1954). Iterative Methods
The research was supported by Fundamental for Solving Partial Difference Equations of
Research Grant Scheme (FRGS), Grant No. 02-01- Elliptic Type. Transactions of American
07321FR under the Ministry of High Education of
Mathematical Society, 76:92−111
Malaysia.
[28]. Yousif, W.S., and Evans, D.J. (1995).
Explicit De-coupled Group Iterative
REFERENCES
Methods and Their Parallel
[20]. Abdullah, AR (1991). "The Four Point
Implementations. International Journal of
Explicit Decoupled Group (EDG) Method:
Parallel, Emergent and Distributed
A Fast Poisson Solver." International
Systems, 7:53−71, 1995.
Journal of Computer Mathematics,
38:61−70.

43
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Value-at-Risk (VaR) using ARMA(1,1)-GARCH(1,1)

Sufianti(1) and Ukur A. Sembiring(2)

(1)
Department of Mathematics, Universitas Pelita Harapan, Tangerang, Banten, Indonesia
Email: sufianti88@gmail.com
(2)
Department of Mathematics, Universitas Pelita Harapan, Tangerang, Banten, Indonesia
Email: ukur_sembiring@uph.edu

ABSTRACT help risk managers in managing portfolio’s

With the economic and social world so uncertain market risk.
today, risk management becomes an essential suit
of armor for many corporate entities. One of the
widely used tools in managing risk is Value-at-Risk The aim of this paper is to introduce daily
(VaR). The main concern in this paper is VaR calculation method using
ARMA(1,1)-GARCH(1,1) model, developed for ARMA(1,1)-GARCH(1,1) model for a
VaR calculation. From the derivation and portfolio, which consists of two risk
simulation, ARMA(1,1)-GARCH(1,1) model with
factors, with related assumptions through
related assumptions is shown to be able to describe
several financial data's characteristics adequately model derivation and simulation using real
and is applicable to real data. Additionally, from data. Real data, in form of share’s closing
backtesting results, VaR calculation using price from arbitrary companies listed in
ARMA(1,1)-GARCH(1,1) model exemplifies a New York Stock Exchange (NYSE), will
better performance than the conventional one, in
represent the major risk factors of a
terms of capital reservation for market risk
coverage. portfolio in the simulation. In addition, a
comparison between VaR value calculated
KEYWORDS with this model and the conventional one
Value-at-Risk; ARMA(1,1)-GARCH(1,1); risk using backtesting method will be
management; risk factor; backtesting conducted to show model’s advantage over
the other.
I. INTRODUCTION
In financial market, value of an instrument
II. VAR CALCULATION FOR
always changes over time, due to risk
PORTFOLIO WITH TWO RISK
factors, such as inflation, natural disasters, FACTORS
or government policies. Therefore, 1. Model Derivation
portfolio holders will be exposed to a risk One-period, discretely sampled return, ,
or the uncertainty of portfolio’s return. can be written as a total of conditional
Several models have been developed to expected return, , and innovation or
make risk measurable, so that it can be
managed according to portfolio manager’s arbitrary component, , with zero mean
risk appetite. One of the methods for and conditional variance :
managing risk is by risk valuation using
Value-at-Risk (VaR).
(1)
Until today, VaR has been widely used as
a reliable risk valuation method and where is independent white noise
possesses an advantage in its ability to process, and is the log
consolidate all risk factors exposing a return.
portfolio. VaR defines the potential loss of
a portfolio for a certain time horizon with Two assumptions will be used :
a specific confidence level. Knowing the 1. is normally distributed (or is a
VaR value for a portfolio will considerably Gaussian white noise);
2. is not constant.

44
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Using the first assumption results in a By Eqs. 5~6, the log-likelihood function is
conditionally normally distributed return:

(2)

The second assumption is used as the basis

in ARMA(1,1) model selection for (7)
modeling the return :
with and are functions of .
To obtain using mle method, the following
(3)
equation must be solved :
and the volatility in the return equation,
(8)
,will be modeled by GARCH(1,1) :

Because the log-likelihood function is

nonlinear in its parameters, Eq. 8 can only be
(4) solved using numerical computation.

GARCH(1,1) is chosen for modeling After getting the parameters, the next step is to
volatility because of its ability to describe calculate 1-step ahead forecast of conditional
several financial data’s characteristics, mean and volatility of using
such as the leptokurtic distribution of
return, volatility clustering, stationarity,
and volatility mean reversion. (9)

After setting up the return and volatility

and by Eq. 2, .
equations, the parameters in Eqs. 3~4 will
be estimated simultaneously using
maximum likelihood estimation method The previous steps of setting equations and
calculating forecasts should be done for each
(mle). Let be the
of the two risk factors. Afterwards, the
vector containing parameters from conditional mean and volatility of the portfolio
ARMA(1,1)-GARCH(1,1) model and can be calculated using
be the set of prior information. Given ,
the likelihood function
can be written as a
product of conditional densities :
(10)
(5)
Since VaR is the α-th quantile of the
with . Because returns are cumulative distribution function of return and
, VaR of the portfolio is
conditionally normally distributed, the
conditional density of the vector is defined as

(11)

(6)

45
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

where is the confidence level estimation. Parameter estimation is conducted

and the value of can be acquired from the using one of the tools in GARCH toolbox,
garchfit, provided by MATLAB. The
standard normal distribution table.
equations acquired:
The VaR calculated from the quantile of the
Arch Coal, Inc. (ACI)
distribution of given information available
at time t is in percentage. The dollar amount of
VaR is then the cash value of the financial
position times the VaR of the log return series. (13)
That is, using a time horizon of 1 day, the
daily VaR of the portfolio for a position of Goldman Sachs Group, Inc. (GS)
is

(12)
(14)
Here, VaR symbolizes the worst loss over 1
day horizon for the position held in portfolio. The following step is to check whether the
2. Simulation models acquired are the suitable ones using the
In this subsection, simulation of VaR same inspections and tests as before. Once the
calculation for portfolio with two risk factors models are approved, 1-step ahead forecast of
will be done using real data. The closing price conditional mean and volatility for ACI and
of Arch Coal, Inc. (ACI) and Goldman Sachs GS are calculated using
Group, Inc. (GS) in 2002-2005 will be used.
Arch Coal, Inc. (ACI)
Before using the data, several tests and
inspections must be conducted to ensure that
they satisfy all assumptions and conditions for
ARMA(1,1)-GARCH(1,1) model usage. (15)
Inspections using ACF, PACF, and Normal
QQ plot, and quantification tests, such as Goldman Sachs Group, Inc. (GS)
Ljung-Box Q-test, Engle’s ARCH test, and
Kolmogorov-Smirnov test will be applied.

After making sure that the data can be used,

the closing price for each risk factor is (16)
converted into log return.
Next, the covariance matrix of ACI and GS is
computed and shown below

(17)

Using Eqs. 15~16, the conditional mean and

volatility for both risk factors can be
calculated. By investing $1 in the portfolio
with 30% in ACI and 70% in GS, the
conditional mean and volatility of the portfolio
is valuated. The following table presents the
results :

Figure 1. Log return of ACI and GS (2002-2005) Table 1. Conditional mean and volatility
ACI GS Portfolio
The next stage is to construct the return and
volatility equations by means of parameter

46
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

0.0015 0.00043953 0.00076012

0.0209 0.0145 0.0134

Thus, with a 99% confidence level, the daily

VaR of the portfolio is

(18)
Figure 2. Backtesting for both VaR calculation models
3. Backtesting VaR
Value-at-Risk or VaR calculation using Ideally, for a time horizon of 250 days and
ARMA(1,1)-GARCH(1,1) model derived confidence level of 99%, the expected
in the previous subsection is useful only if number of exceptions are 2.5 exceptions.
it can predicts loss accurately. Therefore, From the backtesting result, it can be seen
its ability in predicting loss needs to be that both models produce only one
compared with the conventional one. The exception. According to Basel Committee
conventional model can be defined as for Bank Supervision (BCBS) (1996), an
accurate model will fall in the green zone.
Yellow zone indicates that there is a
(19) problem in the model. If a model falls in
the red zone, it will automatically lead into
where is the standard deviation of an assumption of incorrect or inaccurate
the historical data. Both models model.
performances will be compared using
Table 2. Basel Traffic Light (confidence level = 99%)
backtesting method.
Backtesting is a formal statistical Number of Increase in
Zone
technique used for observing whether the exception k
VaR forecast is able to describe the real Green 0-4 0.00
loss over the given time horizon. Historical Yellow 5 0.40
data of the log return of ACI and GS for 6 0.50
500 days will be used and daily VaR 7 0.65
forecast will be done using both models 8 0.75
for the following 250 days, with a position 9 0.85
of $1 and confidence level of 99%. Red 10+ 1.00

Both VaR forecasts will be compared to From Fig. 2, both models fall in the green
the hypothetical P&L from historical data. zone, thus both models are accurate for the
If there is a case where VaR is portfolio in question.
underestimated or the loss is higher than
VaR, it is said to be an exception or VaR using ARMA(1,1)-GARCH(1,1)is
violation towards VaR. If there is too better than the conventional one
many exceptions, then the model applied is
not suitable enough for the portfolio in To show that the first model is better, the
question. formula of Capital Adequacy Ratio (CAR)
will be used.

47
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

risk management plays a very important

role for many corporate entities. One of
(20)
the popular methods in managing risk is by
Value-at-Risk (VaR) calculation.
where
CR = Credit Risk
The advantage of using VaR is its ability
OR = Operational Risk
to quantify a portfolio’s loss potential by
MR = Market Risk = mean of a single number, easily
k = multiplicative factor (See Table 2) understood and used by many people. The
shortcoming of VaR is the difficulties in
CAR is used by Basel Committee to choosing the right model. Using the wrong
regulate the minimum amount of capital model will result in inaccurate results.
reserve for a bank as stated in Basel II
Accord. This rule ensures that a bank will In this paper, the use of ARMA(1,1)-
have enough capital reserve to cover risk GARCH(1,1) model for valuating VaR of
exposed to the bank. For higher risk, more a portfolio with two risk factors is based
capital will be needed as backup and vice its ability to describe financial return’s
versa. leptokurtic distribution, volatility
clustering and volatility mean reversion
From Eq. 20, it can be seen that with the effects, stationarity, and also to
same amount of capital, higher VaR value accommodate the assumptions that its
will result in higher MR which ends up in conditional mean is not constant over time.
lower CAR value. In relation with Basel
Traffic Light, if the model falls in yellow The only drawback of this model is the
or red zone, along with the increase of the assumption where the white noise is
multiplicative factor k, CAR value will normally distributed, which is hardly true
also be lower. If , bank with for most financial data. In reality, there is a
such CAR value will be asked to increase non-normality phenomenon in the return
its capital reserve. When the bank is of financial data.
unable to do it, it can be closed down.
Through the simulation, it is shown that
For the portfolio considered in this paper, the model is applicable to a portfolio
it can be seen from Fig. 2 that VaR constructed using real data. Besides, it is
calculated from the conventional model is quite simple to calculate VaR with this
higher. Therefore, with the same model, only slightly complicated by the
multiplicative factor ( ), to yield an use of numerical computation in
equal CAR value, the capital reserve estimating the parameters. But, it can be
needed by the ARMA(1,1)-GARCH(1,1) solved by using MATLAB.
model is less than the conventional one.
The difference in the capital reserve can be Furthermore, from the backtesting result
used for expanding their business, for for the portfolio (See Figure 2), VaR from
example engaging in other investments. ARMA(1,1)-GARCH(1,1) model is more
This shows that for VaR calculation, with sensitive to the hypothetical P&L
the same accuracy in risk prediction, movement. This proves that with the same
ARMA(1,1)-GARCH(1,1) model is better power and accuracy in predicting risk, the
than the conventional one. ARMA(1,1)-GARCH(1,1) model needs
less capital reserve than the conventional
III. CONCLUSION one, thus making it a better model to be
Financial market is a dynamic used in VaR calculation.
environment and full of risks. Therefore,

48
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

ACKNOWLEDGEMENTS
I would like to acknowledge and extend
my heartfelt gratitude to Mr. Yosef
Oktavianus Senobua, M. Sc. who has
given valuable comments and made the
completion of this paper possible.

REFERENCES
[1]. Bagasheva, BS, Fabozzi, FJ, Hsu,
JSJ, and Rachev, ST (2008).
Bayesian Methods in Finance, John
Wiley & Sons, Inc., New Jersey.
[2]. Basel Committee on Banking
Supervision (2005). Part 2: The First
Pillar – Minimum Capital
Requirements. Retrieved May 31st,
2010,from
http://www.bis.org/publ/bcbs118.htm
[3]. Berry, R (2009). ”Backtesting Value-
At-Risk”, J.P. Morgan Investment
Analytics & Consulting, pp 5-6.
[4]. Bodie, Z, Kane, A, and Marcus, AJ
(2008). Investments, 7th ed.,
McGraw-Hill/Irwin, New York.
[5]. Brockwell, PJ, and Davis, RA
(2002). Introduction to Time Series
and Forecasting, 2nd ed., Springer-
Science+Business Media, Inc., New
York.
[6]. Craig, AT, Hogg, RV, and McKean,
JW (2005). Introduction to
Mathematical Statistics, 6th ed.,
Pearson Education, Inc., New Jersey.
[7]. Franke, J, Hafner, CM, and Hardle,
WK (2008). Statistics of Financial
Markets, An Introduction, 2nd ed.,
Springer-Verlag Berlin Heidelberg.
[8]. Jorion, P (2001). Value at Risk : The
New Benchmark for Managing
Financial Risk, 2nd ed., McGraw-
Hill, New York.
[9]. Nieppola, O (2009). Backtesting
Value-at-Risk Models, Department of
Economics, Helsinki School of
Economics.
[10]. Tsay, RS (2005). Analysis of
Financial Time Series, 2nd ed., John
Wiley & Sons, Inc., New Jersey.

49
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Decline Curve Analysis in a Multiwell Reservoir System

using State-Space Model

S. Wahyuningsih(1), S. Darwis(2), A.Y. Gunawan(3 ), A.K. Permadi(4)

(1)
Statistics Study Program, Faculty of Mathematics & Natural Sciences, Mulawarman University, Samarinda,
East Kalimantan, Indonesia
(Email: swahyuningsih@gmail.com)
(2)
Statistics Research Division, Faculty of Mathematics & Natural Sciences, Institut Teknologi Bandung,
Indonesia
(Email: sdarwis@math.itb.ac.id)
(3)
Industrial and Financial Research Division, Faculty of Mathematics & Natural Sciences, Institut Teknologi
Bandung, Indonesia
(Email: aygunawan@math.itb.ac.id)
(4)
Reservoir Engineering Research Division, Faculty of Petroleum Engineering, Institut Teknologi Bandung,
Indonesia
(Email: akp@tm.itb.ac.id)

ABSTRACT Here, Yt is the production rate at time t , D is

This paper presents a multi-well decline curves
the decline rate, and b is the decline exponent
analysis using state-space model. A predator-prey-
like model was proposed exclusively for such a 0  b  1 . Arps suggested a tentative
system, and the parameter of which was estimated classification of decline curve, based on their
using Kalman filter. The multi-well predator-prey- Yt
like system is represented as a state-space model loss ratios . If the loss ratio is a
dYt / dt
where the state is the model parameters and the
observations are the well production. State-space constant c where c < 0 or the first derivative of
model for decline curves analysis for both single the loss ratio is near b = 0, the curve is the
and multi-well systems, has been tested on real exponential type,
observation as well as simulated. When the well
interactions are quite significant, the use of the
Yt  Y0 .e Dt
proposed model is more benefit.
(2)
KEYWORDS
Decline curve analysis, state-space model, Kalman Eq. (2) represents the relationship between
filter, predator-prey-like model. production Yt and t for constant loss ratio,
where Y0 is the initial production rate. Non
I.INTRODUCTION
Predicting oil and gas production have constant loss ratio corresponds to hyperbolic
attracted many researchers. Fetkovich [2] DCA for 0  b  1 , and harmonic DCA for b
developed type = 1.
curve, Marhaendrajana and Blasingame [4]
extended type curve method to a multiwell Wahyuningsih et.al [5] showed that the
system, and Li and Horn [3] verified Decline univariate Kalman filter is a good approach to
Curve Analysis (DCA) method. One predict next observation based on the Arps
frequently-used technique to predict decline curve analysis in single well case. In
production in oil, gas and geothermal many cases, production of one well is
reservoirs is the DCA which is based on the influenced by other wells in the same
empirical Arps equation [1]. The Arps DCA is reservoir. So, there can be some interaction
described by the differential equation among wells. Sometimes one well can produce
dYt higher than the others.
  DYt b 1
dt
This paper aims propose a state space model
(1)
as DCA model for multiwell system. We
construct state-space model and derive the

50
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

recursive formula of multiwell system in always decrease or constant, however there is

Section 2. Modeling the interaction among not interaction among the wells. One
wells using predator-prey approach is describe production well will take another production
in Section 3. We also give an example to wells. This well as predator, of course another
illustrate the application of the model. The well as prey. The mathematical model of the
final section of the paper consists of some interaction among wells is proposed as :
concluding remarks.
dYt
II. STATE SPACE MODEL   DYt
dt
Consider a state space form for multivariate .(5)
state space model (1):
 D1 s1,2 1,2  s1, n 1,n 
Measurement: Z t 1  Yt 1  bt 1  
s2,1 2,1 D2  s2, n  2, n 
State: Yt 1   Yt  at 1 where D  
     
(3)  
 sn ,1 n,1 sn ,2  n ,2  Dn 
a   0 
where  t   N   . Unlike other time
 bt   0   i , j  0, i , j  0,1,2,, n , and
series model such as autoregressive process,  si , j  1, if Yi ,t as a predator
state space model does not require the random  , i  j,
variables to be stationary. Parameter  can be  si , j  1, if Yi ,t as a prey
estimated by ordinary least square.
Di is decline rate in multiwell system, and
Using Bayesian theorem, the recursive  i , j is interaction coefficient between well
equation from equation system (3) is i and j .
The parameters Di and  i , j can be estimated
Yt 1  Yt  Rt 1 (  Rt 1 )1 et 1  YYt  K t 1et 1
 , simultaneously using Gauss Newton method.
Vt 1  Rt 1  K t 1 Rt 1
For example Figure 1 shows the simulation of
(4) the model for two production wells. The
simulation using the propose model is
where et 1  Z t 1  Zˆ t ()  Yt 1  Yˆt  bt 1 , generated by D1 = 0; 06 and mean 62,099, and
Rt 1  Vt    , and the production well 2 is generated by D1 = 0;
K t 1  Rt 1 (  Rt 1 )1 01 and mean 50,337. It is assumed that there
is called Kalman gain. Kalman gain are some interactions between the wells. Each
determines the weight given to the forecast well have 20 data which is divide into training
error vector. Eqs. (4) are used to update the data set and validation data set. Base on the
mean and covariance and hence the  44.944 5.081 
data we have V0    . Using
distribution of the state, Yt 1 , after the new  8.081 0.907 
observation, Z t 1 has become available.  0.854 0.139 
VAR process we obtain    .
 -0.018 1.013 
III. INTERACTION MODEL The coefficient of interactions for this data are
Assumed that there are some interactions
1,2 = 0.151 and 2,1 = 0; 028. Given different
among production wells. Interaction among
production error, we have different pattern forecasting
wells is analogized as predator-prey-like production (Figure 2,3, and 4).
model. The analogy is not pure predator-prey
model because the oil and gas production

51
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 4. Simulation 4
Figure 1. Simulation 1

Figure 5. Kalman gain of production well 1 and 2

Figure 2. Simulation 2
Figure 1, 2, 3, and 4 show the prediction and
forecast of each well production using vector
and scalar Kalman filter. Prediction by vector
Kalman filter is more realistic than its scalar.
The prediction of scalar Kalman filter tends
underestimate for well 1, and overestimate for
well 2. The results are similar for its forecast.
The vector Kalman filter tends toward a steady
 0.591 0.014 
state with K t 1    for t  6
 0.014 0.620 
(see Figure 5), and the scalar Kalman filter
tends a steady state with Kt 1,1 = 0; 606 for
t  6 , and Kt 1,2 = 0; 616 for t  6 . The
implication of the steady state of Kalman gain
is the forecast model can be simplified as
Figure 2. Simulation 3 multivariate regression model.

CONCLUSIONS
Decline curve analysis using the state space
model can be used to represent prediction
model of production well. A Predator-prey-
like model is a potential approach for

52
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

modeling inter well interaction. The proposed [2] Fetkovich, MJ (1980). “Decline Curve
model is more realistic when the interaction Analysis Using Type Curve”, SPE 4629,
among wells can be negligible i.e. when the pp 1065-1077.
wells are located relatively close to each other. [3] Li, K and Horne, RN (2005). “Verification
of Decline Curve Analysis Model for
ACKNOWLEDGEMENT Production Prediction”, SPE 93878.
This research was supported by Competitive [4] Marhaendrajana, T and Blasingame, TA
Grand 2010, Directorate of Higher Education, (2001). “Decline Curve Analysis Using
Department of National Education, Republic Type Curve Evaluation of Well
of Indonesia. Performance Behavior in a Multiwell
Reservoir System”, SPE 71517.
REFERENCES [5] Wahyuningsih, S, Darwis, S, Gunawan,
AY, Permadi, AK (2008). “Predicting
[1] Arps, JJ (1945). “Analysis of Decline Steam Production Using Kalman Filter”,
Curve Analysis”, Trans. AIME, pp 160- FJTS, Vol.26 Issue 2, pp 219-225.
247.

53
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Study of Role of Interferon-Alpha in Immunotherapy

through Mathematical Modelling

Mustafa Mamat1, Edwin Setiawan Nugraha1, Agus Kartono 2, W M Amir W Ahmad1

1
Department of Mathematics, Faculty of Sciences and Technology, Universiti Malaysia Terengganu, 21030 K.
Terengganu, Malaysia.
Email: mus@umt.edu.my
2
Laboratory for Theoretical and Computational Physics, Department of Physics, Faculty of Mathematical and
Natural Sciences, Institut Pertanian Bogor, Kampus IPB Darmaga, Bogor 16680, Indonesia.
Email: akartono@ipb.ac.id

ABSTRACT Interaction between tumor cells and an

A mathematical model in the form of ordinary immune system has been studied through
differential equations that describing tumor- mathematical model by numerous authors. In
immune interaction in the presence of 1998, a mathematical modeling that describes
immunotherapy is presented. The model is used to dynamics between tumor cells, effector cells
investigate on how the effect of immunotherapy
(the activated immune system cells) and IL-2
treatment when adding interferon alpha (INF-),
not only interleukin-2 (IL-2) and infiltrating
had been proposed by Kirschner and Panetta
lymphocytes (TIL) on the tumor growth. Numerical [10]. The important parameter in their model
simulation shows combination TIL, IL-2 and INF- is antigenicity of tumor (c), measure of ability
 more effective to enhance of strength of immune the immune system to recognize the tumor
system than only combination TIL and IL-2. cells. Their works are able to explain both
short tumor oscillations in tumor size as well
KEYWORDS as long term tumor relapse. They also
Mathematical modeling; ordinary differential investigate the effect of immunotherapy both
equations; immune response; interferon alpha active and adoptive on tumor growth. Then,
(INF-); tumor infiltrating lymphocytes (TIL); they indicate that a treatment with adoptive
interleukin-2 (IL-2). immunotherapy may be a better option to
either as a monotherapy or in conjuction with
1. INTRODUCTION IL-2. Kuznetsov et al.[11] presented a
One types of promising treatment for cancer mathematical model of the cytotoxic T-
disease currently is immunotherapy. The lymphocyte response to the growth of an
principle of this treatment is to increase the immunogenic tumor. The model exhibits a
strength of the immune system to attack cancer number of phenomena that are seen in vivo,
cells. The evidence for the potential of including the immunostimulation of the tumor
immune system control the cancer hats been growth, “sneaking through” the tumor, and the
verified in the laboratory and clinical formation of a tumor “dormant state”. In 2003,
experiment [12, 13]. It has motivated new Szymanska [14] proposed a basic
research into development of immunotherapy mathematical model in the form six ordinary
[1, 3]. There are three main category of this differential equations. The model describes
treatment: immune response modifier, immune response when cancer ells recognized
monoclonal antibodies, and vaccines [4]. The and extend to include two types of
first category contains substances that affect immunotherapy: active and adoptive
immune response, such as interleukin-2 (IL-2), immunotherapy. Her study shows that the
interferons, tumor necrosis factor (TNF), method of adoptive immunotherapy is seem
colony-stimulating factors (CSF), and B-cell better for the patient at least in some cases.
growth factor. In the next category, In 2006, Pillis et al. [4] proposed the
monoclonal antibodies are currently being mathematical model in the form of ordinary
developed to target specific cancer antigens. In differential equations. The model describes
the third category are vaccines, which dynamics of tumor cells and immune cells in
generally used therapeutically, and created
from tumor cells.

54
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

the presence of mmunotherapy and paper is as follow. In section 2, we present a

chemotherapy. There are two immunotherapy system of ordinary differential equation which
drug in their model such as interleukin-2 (IL- describes tumor-immune interaction under the
2) and tumor infiltrating lymphocytes (TIL). influence of immunotherapy. In section 3, we
One of their results show that mixed present set parameters values for this model
immunotherapy and chemotherapy is more from previous works to run simulation.
effective to eliminate tumor cells than both Section 4 present numerical simulations based
immunotherapy alone and chemotherapy on parameters in section 3. In section 5, we
alone. Mathematical model of immunotherapy summarize and discuss our conclusion.
and chemotherapy has been also proposed by
Isaeva and Osivop [9]. There are two 2. MATHEMATICAL MODEL
immunotherapy drug in their model such as Pillis’s model [4] describes the effect of
interleukin-2 (IL-2) and interferon-alpha (INF- interlukin2 (IL-2), tumor infiltrating
). In their study, they compared several of lymphocytes (TIL), and chemotherapy on
treatment strategies such as chemo/immune, tumor growth. Isaeva and Osiopov’s model
immune/chemo and concurrent describes the effect of IL-2, INF- and
chemo/immune therapy. Numerical chemotherapy on tumor growth. In our study,
simulations show that chemo/immuno we modified Pillis’s model [4] by including
sequence is more effective while concurrent INF-. We define six populations in our
chemoimmunotherapy is more sparing. model: tumor cells T(t), natural killer cells
In this paper, we proposed a mathematical N(t), CD8+ T cells L(t), circulating
model for immunotherapy treatment by lymphocytes cell C(t), concentration of IL-2
extending Pillis’s model [4] to include the I(t), INF- concentration I(t). The
presence of interferon-alpha (INF-) and the mathematical model follows used as follows:
absence of chemotherapy. The outline of this

dT dI 
 aT ( 1  bT )  cNT  DT  c' TL  V (t )  gI 
dt dt

(1) (6)
dN T2 ( L / T )l
 eC  fN  g N  pNT where D d.
dT h T 2 s  (L / T )l

(2)

dL D 2T 2 2 piLI
 mL  j L  qLT  ( r1 N  r2 C )T  uNL   vL ( t )
dT k  D 2T 2 gi
(3)
dC
   C
dt

(4)
dI
  iI  jLI  kTI  v I (t )
dt

(5)

55
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Equation (1) describes the growth of tumor term RL  N , L  r1NT [4]. The immune
population. Term 1 express tumor growth, in system is also stimulated by the tumor cells to
the form logistic growth [5]. The interaction produce more CD8+ T-cells. Recognition of
between NK cells with tumor are represented the presence of the tumor is proportional to the
by term of  cNT . Whereas tumor lysis by average number of encounters between
CD8+ T cells has form  DT. Both that terms circulating lymphocytes and tumor. It is
represents negative interaction between two represented by RL C, T   r2CT [4]. The fifth
populations which describes completion for
space and nutrients as well as regulatory action term, I CL  uNL2 , describes the NK cell
and direct cell population. We suppose that the regulation of CD8+ T cells, which occurs
parameter c’ depend on INF- concentration when there are very high levels of activated
CD8+ T cells without responsiveness to
 
I 
as c'  cCTL  2  e I 0  where cCTL is rate cytokines present in the system [4]. The term
 
  of vL ( t ) is a function of time and represents
of tumor inactivation by CTL [7]. This agrees drug intervention term of tumor infiltrating
with the fact that INF- enhances immune lymphocytes to boost the immune system [4].
immune-mediated antitumor responses by Equation (4) describes the rate of change for
increasing expression of MHC molecules on the circulating lymphocyte population. We
tumor cells. assume that circulating lymphocytes assumed
The second equation describes the rate of generate a constant rate and that each cell has
change for the NK cell population. In this a natural lifespan. Thus, the terms are   C
model, the growth term of NK cell population [4].
is tied to the overall immune health levels as The fifth equation describes the concentration
measured by the population of circulating of IL-2 in the bloodstream. The term 1,  iI ,
lymphocytes. Hence, the growth term of NK
cell is represented by term 1 and term 2, express the natural death rate of IL-2 [4]. The
eC  fN . The term of NK cell recruitment next term,  jLI , represents the consumption
describe by Pillis and Radunskaya [5]. Hence, rate of IL-2 [7]. It was found that inhibition of
IL-2 results from an accumulation of immune-
T2 suppressing substances, prostaglandins. Their
the term is g N . Inactivation term of
h T 2 number is proportional to the tumor
cytolytic potential occur when a NK cell has population. Prostaglandins suppress of the
interacted with tumor cells several times and production of IL-2 and can directly destroy it
ceases to be effective. The inactivation term of molecules [7]. The term 3,  kTI , express the
NK cells has form  pNT [5]. IL-2 destruction rate by prostaglandins [11].
The rate of change for the CD8+ T cell The last term, vI ( t ) , is function of time and
population is described by equation (3). Cell represent the amount of IL-2 injected to the
growth term for CD8+ T cells consist only of patient [4].
natural death rate, since no CD8+ T cells are The sixth equation shows the concentration of
assumed to be present in the absence of tumor INF- in the bloodstream. The term 1, V ( t ) ,
cells. Thus, the term is  mL [4]. The term of is function of time and represents the amount
NK cell recruitment describe by Pillis and of INF- injected to the patient [7].
D 2T 2 The term 2,  gI , express the natural death
Radunskaya [5], and has form j L.
k  D 2T 2 rate of INF- [7].
For term 3, we use inactivation term
developed by also Pillis and Radunskaya [5]
which written in the form  qLT . CD8+ T-
cells can also be recruited by the debris from
tumor cells lysed by NK cells. This
recruitment term is proportional to the number
of cells killed. From this, we procure the

56
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Table1. Estimated human parameter values for numerical analysis

Patient 9 Patient10 Units Description Source
a = 4.31 x 10-1 a = 4.31 x 10-1 day-1 Tumor growth rate [6]
b =1.02 x 10-9 b =1.02 x 10-9 cell-1 1/b is tumor carrying capacity [6]
-11 -11 -1 -1
c = 6.41 x 10 c =6.41 x 10 day · cell Fractional (non) ligand transduced tumor cell [6], [7]
kill by NK cells
d =2.34 d =1.88 day-1 Saturation level of fractional tumor cell kill by [7]
CD8+ T Cells. Primed with ligand-transduced
cells, challenged with ligand-transduced
e = 2.08 x 10-7 e = 2.08 x 10-7 day-1 Fraction of circulating lymphocytes that [11]
became NK cells
l = 2.09 l = 1.81 Dimensionless Exponent of fractional tumor cell kill by CD8+ [7]
T cells. Fractional tumor cell kill by
chemotherapy
f = 4.12 x 10-2 f = 4.12 x 10-2 day-1 Date rate of NK cells [6]
-2 -2 -1
g = 1.25 x 10 g = 1.25 x 10 day Maximum NK cells recruitment by ligand- [7]
transduced tumor cells
h = 2.02 x 107 h = 2.02 x 107 cell2 Steepness coefficient of the NK cell [11]
recruitment curve
j = 2.49 x 107 j = 2.49 x 107 day-1 Maximum CD8+ T cell recruitment rate. [6], [7]
Primed with ligand-transduced cells
k = 3.66 x 107 k = 5.66 x 107 cell2 Steepness coefficient of the CD8+ T cell [6], [7]
recruitment curve
m = 2.04 x 10-1 m = 9.12 day-1 Death rate of CD8+ T cells [15]
-6 -6 -1 -1
q = 1.24 x 10 q = 1.24 x 10 day · cell CD8+ T cell inactivation rate by tumor cells [11]
p = 3.42 x 10-6 p = 3.59 x 10-6 day-1 · cell-1 NK cell inactivation rate by tumor cells [7]
-2 -1
s = 8.39 x 10 s = 5.12 x 10 Dimensionless Steepness coefficient of tumor –(CD8+ T cell) [6]
lysis term D. Primed with ligand-transduced
cells, challenged with ligand-transduced.
r1 = 1.10 x 10-7 r1 = 1.10 x 10-7 day-1 · cell-1 Rate of which CD8+ T cells are stimulated to [15]
be produced as a result a tumor cells killed by
NK cells
r2 = 6.50 x 10-11 r2 = 6.50 x 10-11 cell−1 · day−1 Rate of which CD8+ T cells are stimulated to -
be produced as a result a tumor cells
interaction with circulating lymphocytes
u = 3.00 x 10-10 u = 3.00 x 10-10 cell−2 · day−1 Regulatory function by NK cells of CD8+ T -
cells
 = 7.50 x 108  = 5.00 x 108 cell · day−1 Constant source of circulating lymphocytes [8]
−1
 = 1.20 x 10 -2
 = 8.00 x 10 -3
day Natural death and differentiation of [8]
circulating lymphocytes
 = 9.00 x 10-1  = 9.00 x 10-1 day−1 Rate of chemotherapy drug decay [2]
-1 -1
pi = 1.25 x 10 pi = 1.25 x 10 day−1 Maximum CD8+ T cell recruitment curve by [10]
IL-2
gi = 2.00 x 102 gi = 2.00 x 102 cells2 Steepness of CD8+ T cell recruitment curve by -
IL-2
 i = 1.00 x 101  i = 1.00 x 101 day−1 Rate of IL-2 drug decay [10]
-9 −1 −1
cCTL =4.4 x 10 cell day Rate of tumor cells inactivation by CD8+ T [9]
cells
j =3.3 x 10-9 cell−1day−1 Rate consumption IL-2 by CD8+ T cells [9]
-8 −1 −1
k = 1.8 x 10 cell day Inactivation of IL-2 molecules by [9]
prostaglandins

57
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

3. PARAMETER VALUES presence of the INF- in immunotherapy

To complete the development of the treatment.
mathematical model and simulation, it is Then we note that immunotherapy by using
necessary to obtain accurate parameters. The TIL, IL-2 and INF- may limited only
equations (1)-(6) are very sensitive to the effective for tumor sizes around order 106.
choice of parameters and can vary greatly Figure 3A and 3B shows that this treatment is
from one individual to another, so multiple not effective in treating the tumor size 1 × 107.
data sets can be used in order to obtain suitable These figure show that the addition of INF-
parameters ranges. In our study, we used value has no significant effect in controlling the
of parameters from previous works. These tumor growth. This result is consistently with
parameters values are provided at Table 1. previous work [4].

4. NUMERICAL SIMULATION 4.2. ADDITION OF INF- on PATIENT 10

Firstly, we examine the model with the set of In order to examine whether these treatment
parameters representing patient 9 provided at simulations vary from patient to patient, we
Table 1. We present a case which the immune change patient specific parameter extracted
system response to tumor growth in the from Rosenberg’s study and run simulation
presence of INF-. It is to investigate the with the parameters for patient 10 [6]. These
strength of the immune system in controlling parameters are provided at Table 1. We begin
tumor growth after patient got immunotherapy examine the initial tumor cells 1 x 106. The
treatment by addition of INF-. To study result of numerical experiments show that the
behavior model with variation of patient, we immune system can reduction of the tumor
simulate this model using the parameters of cells before day 20 but cannot full of control
the patient 10 provided at Table 1. This patient the tumor cell population as we see in Figure
comparison provides insight into patient- 5A, so that the tumor growth to dangerous
specific parameter sensitivity. level, although patient 10 get 109 TIL from
day 7 through day 8 and IL-2 in 6 pulses at
4.1. ADDITION OF INF- on PATIENT 9 strength 5 × 105 from day 8 to day 11. If we
In this sub section, we examine compare this result with previous numerical
immunotherapy treatment with injection of IL- experiment in patient 9 in Figure 1A, where
2, TIL and INF- to patient. We are interested tumor cells can be killed, thus the immune
to investigate whether the INF- can cause system in patient 10 is weaker than patient 9.
immune system to be stronger. We begin with Then, we study the effect of the addition of
examine the immunotherapy treatment without INF- on dynamics of tumor cells and
INF- to handle 1 x 106 tumor cells, while the immune cells. Our result shows that this
initial of the immune system are 1 x 103 NK treatment cause the immune system is more
cells, 10 CD8+ T cells, and 6 x 108 circulating strength than before and the tumor cells can be
lymphocytes. 109 TIL administered on day 7 to killed completely at day 17 as we seen in
day 8 and IL-2 is administered in 6 pulses at Figure 5B. However it begin relapse again at
strength 5 x 106 on day 8 to day 11. The results day 23.
are shown in Figure 1A and 1B. These Figure
show that this treatment successfully increase
the strength of the immune system and
consequently the immune system is able to kill
the tumor cells completely at day 16. By
addition of INF- in immunotherapy
treatment, I (t) = 5 MU administered for four
days in a ten day cycle, the result show that
this treatment allow the immune system
become stronger to against the tumor cells as
seen Figure 2A and 2B, consequently the
tumor cells is able to killed completely at day
11. All these results demonstrate that the
tumor cells will die more quickly when in

58
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

A B

Figure 1. Dynamics of tumor cells and immune cells are under influence of immunotherapy without addition of INF- when
initial tumor size 1 × 106 cells. Figure B is more detail than Figure A.

A B

Figure 2. Dynamics of tumor cells and immune cells are under influence of immunotherapy with addition of INF- when
initial tumor size 1 × 106 cells. Figure A is more detail than Figure B.

59
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

A B

Figure 3. Dynamics of tumor cells and immune cells are under influence of immunotherapy when initial tumor size 1 × 107
tumor cells. A: immunotherapy without addition of INF-. B: immunotherapy with addition of INF-.

A B

Figure 4. Immunotherapy drugs concentration. A: IL-2 Concentration. B: INF- Concentration

60
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

A B

Figure 5. Dynamics of tumor cells and immune cells are under influence of immunotherapy when initial tumor size 1 × 106
cells. A: immunotherapy without addition of INF-. B: immunotherapy with addition of INF-.

5.DISCUSSION and CONCLUSIONS In the future, this study is still necessary to

In this paper, we have extended Pillis’s model further investigation how the influence of
[4] to include the presence of INF-. The other cytokines such as IL-10, IL-12 on the
model represented by system of ordinary strength of the immune system to find better
differential equation which describes treatment.
dynamics of tumor cells and immune cells
under the immunotherapy treatment. This ACKNOWLEDGEMENTS
system consists vL(t), vI (t) and v(t) that The authors acknowledge the financial support
represents the injections of TIL, IL-2 and INF- from Department of Higher Education,
 into system, respectively. However, the Ministry of Higher Education, Malaysia
effects of INF-  treatment on tumor-immune through Fundamental Research Grant Scheme
interaction are the main focus. (FRGS) Vot 59146.
Our numerical experiments show that the
addition of INF- in immunotherapy treatment REFERENCES
plays a significant role in dynamics of tumor [1]. Blattman, J.N. and Greenberg, P.D.,
cells and immune cells. Based on Figure 1 and (2004). Cancer immunotherapy: A
2, INF- treatments can enhance the immune treatment for the masses. Science,
system and consequently the tumor cells dies 305:200–205.
more quickly. There fore, the immunotherapy [2]. Calabresi, P. and Schein, P.S., (1993).
treatment with combination of TIL, IL-2 and editors. Medical Oncology: Basic
INF- is more effective than only combination Principles and Clinical Management of
TIL and IL-2 without INF- . For parameters Cancer. McGraw-Hill, New York,
from patient 10 and initial tumor size around second edition,.
106, our study show effect of injection of TIL, [3]. Couzin, J. (2002). Select T cells, given
IL-2 and INF- cause reduction of the tumor space, shrink tumors. Science,
cells temporary, then tumor recurrence in the 297:1973.
future. If initial tumor cells reach 107, this
treatment have no an important role on the [4]. de Pillis, L. G., Gu, W. and
immune system since the tumor cells is unable Radunskaya, A. E., (2006). Mixed
killed. immunotherapy and chemotherapy of
tumors: modeling, applications and

61
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

biological interpretations, J. Theoret treatment: Mathematical modelling,

Biol 238(4): 841-862. Computational and Mathematical
[5]. de Pillis, L.G. and Radunskaya, A.E. Methods in Medicine 10(4): 253-272.
(2003). Immune response to tumor [10]. Kirschner, D. and Panetta, J. C., (1998).
invasion. In K.J. Bathe, editor, Modeling immunotherapy of the tumor-
Computational Fluid and Solid immune interaction, J. Math. Biol.
Mechanics, 2: 1661–1668, M.I.T. 37(3): 235-252.
[6]. Diefenbach, A., Jensen, E. R., Jamieson, [11]. Kuznetsov, V., Makalkin, I., Taylor, M.,
A. M. and Raulet, D., (2001). Rael and and Perelson, A., (1994). Nonlinear
H60 ligands of the NKG2D receptor dynamics of immunogenic tumors:
stimulate tumor immunity, Nature 413: Parameter estimation and global
165-171. bifurcation analysis. Bulletin of
[7]. Dudley, M. E. , Wunderlich, J. R., Mathematical Biology, 56(2):295–321.
Robbins, P. F., Yang, J. C., Hwu, P., [12]. O’Byrne, K. J., Dalgleish, A. G.,
Schwartzentruber, D. J., Topalian, S. L., Browning, M. J., Steward, W. P., and
Sherry, R., Restifo, N. P., Hubicki, A. Harris. A. L., (2000). The relationship
M., Robinson, M. R., Raffeld, M., between angiogenesis and the immune
Duray, P. Seipp, C. A., Rogers-Freezer, response in carcinogenesis and the
L., Morton, K. E., Mavroukakis, S. A., progression of malignant disease. Eur J
White, D. E., and Rosenberg, S. A., Cancer., 36:151–169.
(2002). Cancer regression and [13]. Stewart, T.H., (1996). Immune
autoimmunity in patients after clonally mechanisms and tumor dormancy.
repopulation with anti tumor Medicina – Buenos Aire, 56(1):74–82.
lymphocytes, Science 298 (5594):850-
854. [14]. Szyma´nska, Z.(2003). Analysis of
immunotherapy models in the context of
[8]. Hauser, B. (2001). Blood tests. cancer dynamics, International Journal
Technical report, International of Applied Mathematics and Computer
Waldenstrom’s Macroglobulinemia Science 13(3): 407–418.
Foundation. Available at
http://www.iwmf.com/Blood Tests.pdf [15]. Yates, A., and Callard. R., (2002). Cell
Accessed May 2005. death and the maintenance of
immunological memory. Discret Contin
[9]. Isaeva, O. G. and Osiopov, V. A., Dyn S., 1(1):43–59.
(2009). Different strategies for cancer

62
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Improving the performance of the Helmbold universal portfolio with an

unbounded learning parameter

Choon Peng Tan(1) and Wei Xiang Lim(2)

Department of Mathematical and Actuarial Sciences, Universiti Tunku Abdul Rahman, Jalan Genting Kelang,
53300 Setapak, Kuala Lumpur, Malaysia
(1)
Email: tancp@utar.edu.my
(2)
Email: limwx@utar.edu.my

ABSTRACT
Universal portfolios were studied in [1] and [2].
We consider investing in an -stock market using (1)
a Helmbold universal portfolio [3]. It was
demonstrated in [3] that the Helmbold universal
portfolio can perform better than the Cover uniform where is the number of stocks in the market,
universal portfolio [1] on some stock data sets with is the price relative of stock on the nth
much lesser computer memory requirements. It is
the objective of this paper to show that the upper
trading day; ; the learning
bound on the learning parameter recommended parameter is defined by
in [3] is unnecessarily restrictive. By allowing to
take on larger positive or negative values, it may be
where is the total number of trading days
possible to achieve higher investment returns. and . From the
definition, it is clear that is positive and
KEYWORDS
bounded. We assume all logarithms in this
Helmbold universal portfolio; investment wealth;
learning parameter. paper are to base . It is our objective to show
that the upper bound on recommended in [3]
I. INTRODUCTION is unnecessarily restrictive. By allowing to
An investment portfolio is universal if it does take on larger values, it may be possible to
not depend on the underlying distribution of achieve higher investment returns. By
the stock prices. Universal portfolios were maximizing a certain objective function of the
studied by Cover [1], Cover and Ordentlich doubling rate of the wealth or capital function,
[2]. We consider investing in an -stock we show that the resulting universal portfolio
market using a Helmbold universal portfolio obtained is given by (1) with negative. We
[3]. It was demonstrated in [3] that the demonstrate further that by running the
Helmbold universal portfolio can perform Helmbold universal portfolio on some selected
better than the Cover uniform universal stock data from the Kuala Lumpur Stock
portfolio [1] on some stock data sets with Exchange, it is possible to obtain higher
much lesser computer memory requirements. investment returns using positive and negative
Subsequently, Tan and Tang [4] have shown learning rates which are unrestricted.
that the initial starting portfolio of the
Helmbold universal portfolio is an important II. ETA-PARAMETRIC FAMILY OF
parameter influencing its performance. HELMBOLD UNIVERSAL PORTFOLIOS
First, we introduce the eta-parametric family
A portfolio vector is a of Helmbold universal portfolios which is
vector satisfying for and defined by (1) for any real number .
. The Helmbold universal portfolio
Preposition.
is a sequence of portfolio vectors
Consider the objective functions
generated by the following update of , the ith
component of the portfolio vector on the nth
trading day: (2
and

63
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

respectively. In the first case, we want

(3
  the current portfolio to be close to the
where previous portfolio in terms of Kullback-
Leibler distance measure. In the second case,
we want the opposite, i.e. the maximum
(4) separation between the current and the
previous portfolios. In Tan and Tang [4], we
  have shown that the initial starting portfolio is
is the Kullback-Leibler distance measure or a parameter that can affect the final wealth
relative entropy and is positive. By achievable by the Helmbold universal
approximating by portfolio. If the initial starting portfolio is a
, the maximum of the good one, we require that the subsequent
portfolios are close to each other. On the other
objective function is achieved at
hand, if the initial starting portfolio is not a
given by (1) and the maximum of is
good one, we hope to move away from the
also achieved at given by (1) where is
current portfolio to the right one with the
replaced by . highest investment return. In the latter case,
we require maximum separation between the
Proof. current and previous portfolios.
Since is a portfolio vector for ,
we need to intoduce the Lagrange multiplier III. EXPERIMENTAL RESULTS
in maximizing the objective functions We have run the eta-parametric family of
Helmbold universal portfolio on 3 stock data
sets chosen from the Kuala Lumpur Stock
( Exchange. The period of trading of the stocks
selected is from January 1, 2003 until
 
November 30, 2004, consisting of 500 trading
and
days. Each data set consists of 3 company
stocks. Set A consists of the stocks of Malayan
Banking, Genting and Amway (M) Holdings.
( Set B consists of the stocks of Public Bank,
  Sunrise and YTL Corporation. Finally, set C
Helmbold et al. [3] have shown that the consists of the stocks of Hong Leong Bank,
maximum of is achieved at RHB Capital, and YTL Corporation.
given by (1). The maximum of is
The universal wealth achieved by the universal
achieved when the following partial
portfolio after trading days, is given by:
derivatives are zero,

for . We obtain (7)

for
. Summing up the components where is the price-relative vector on day
over , we have and the initial wealth is assumed to be 1 unit.
leading We begin with the initial portfolio
for all the 3 data sets.
to for . For each data set, the portfolios after 500
trading days and the universal wealths
achieved are calculated for selected values of
Remark. and are listed in Tables 1, 2 and 3.
We have shown that the eta-parametric family
of Helmbold universal portfolios is generated Let and
by maximizing two different objective denote the norm and norm of the vector
functions and for and respectively. It is clear from Tables 1, 2 and
3 that and as

64
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

functions of are growing with as gets -0.10 (0.327, 0.342, 0.331) 1.5658
larger. If is restricted between 0 and 0 (0.333, 0.333, 0.333) 1.5650
0.10 (0.340, 0.325, 0.336) 1.5641
(or 0.1326 for , ) as 0.20 (0.346, 0.316, 0.338) 1.5633
recommended in [3], and 0.30 (0.352, 0.307, 0.340) 1.5624
are close to 0. If a larger variation 0.50 (0.364, 0.291, 0.345) 1.5607
of from is required, it can be achieved 0.75 (0.380, 0.271, 0.349) 1.5585
by using a larger universal portfolio.
1.00 (0.395, 0.253, 0.353) 1.5563
3 (0.504, 0.137, 0.359) 1.5398
In Figures 1, 2 and 3, the graphs of the
5 (0.594, 0.070, 0.336) 1.5266
universal wealth against are plotted for
10 (0.748, 0.011, 0.240) 1.4996
the data sets A, B and C respectively, where
the local maxima are shown. We strongly
Table 2. The portfolios after 500 trading days and
believe that the local maxima are also the the universal wealths achieved for selected values of
global maxima over all . For data set A, the and data set B
maximum universal wealth achievable is
at . Here is an example -10 (0.851, 0.149, 0.000) 1.8141
of a Helmbold universal portfolio with a
-5 (0.687, 0.311, 0.002) 1.8109
negative-valued parameter achieving the
-3 (0.601, 0.381, 0.018) 1.8399
maximum wealth. For data sets B and C, the
-1.00 (0.460, 0.397, 0.144) 1.9865
maximum universal wealth achievable are
-0.75 (0.432, 0.387, 0.181) 2.0221
and at
and respectively. Again, this -0.50 (0.402, 0.373, 0.225) 2.0629
demonstrates that if is restricted between 0 -0.30 (0.375, 0.359, 0.265) 2.0993
and 0.1326 as recommended in [3], it is not -0.20 (0.362, 0.351, 0.287) 2.1187
possible to achieve the maximum wealth. -0.10 (0.348, 0.343, 0.310) 2.1389
0 (0.333, 0.333, 0.333) 2.1599
IV. CONCLUSIONS 0.10 (0.319, 0.324, 0.358) 2.1816
This paper demonstrates that it is necessary to 0.20 (0.304, 0.313, 0.383) 2.2041
remove the unnecessary restriction 0.30 (0.289, 0.303, 0.408) 2.2272
imposed on in order to 0.50 (0.260, 0.280, 0.461) 2.2752
0.75 (0.223, 0.250, 0.527) 2.3381
achieve a higher investment wealth. Empirical 1.00 (0.189, 0.220, 0.591) 2.4030
evidence is provided that the maximum 3 (0.032, 0.052, 0.917) 2.8850
investment wealth can be achieved at a 5 (0.004, 0.009, 0.987) 3.1897
negative learning parameter and at large 10 (0.000, 0.000, 1.000) 3.5139
positive learning parameters. The best
achieving the maximum wealth can be Table 3. The portfolios after 500 trading days and
determined from hindsight given the past stock the universal wealths achieved for selected values of
data. It remains unsolved how to choose the and data set C
best at the beginning of the investment
period. -10 (0.023, 0.977, 0.000) 1.3126
-5 (0.138, 0.862, 0.000) 1.3349
Table 1. The portfolios after 500 trading days and
-3 (0.252, 0.741, 0.008) 1.3898
the universal wealths achieved for selected values of
and data set A -1.00 (0.366, 0.519, 0.115) 1.5867
-0.75 (0.368, 0.478, 0.155) 1.6358
-10 (0.005, 0.979, 0.016) 1.4309 -0.50 (0.363, 0.432, 0.204) 1.6931
-5 (0.068, 0.814, 0.118) 1.5449 -0.30 (0.355, 0.394, 0.251) 1.7453
-3 (0.151, 0.638, 0.211) 1.5724 -0.20 (0.349, 0.374, 0.277) 1.7735
-1.00 (0.270, 0.428, 0.302) 1.5722 -0.10 (0.342, 0.354, 0.305) 1.8031
-0.75 (0.286, 0.403, 0.311) 1.5706 0 (0.333, 0.333, 0.333) 1.8341
-0.50 (0.302, 0.379, 0.319) 1.5689 0.10 (0.324, 0.313, 0.363) 1.8664
-0.30 (0.314, 0.361, 0.325) 1.5674 0.20 (0.313, 0.293, 0.394) 1.9000
-0.20 (0.321, 0.351, 0.328) 1.5666 0.30 (0.302, 0.272, 0.426) 1.9348

65
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

0.50 (0.277, 0.233, 0.490) 2.0075

0.75 (0.243, 0.188, 0.569) 2.1030
1.00 (0.208, 0.147, 0.645) 2.2015
3 (0.034, 0.012, 0.954) 2.8734
5 (0.004, 0.001, 0.995) 3.2020
10 (0.000, 0.000, 1.000) 3.4415

Figure 3. Graph of against displaying the local

maximum at for data set C

REFERENCES
[1]. Cover, TM (1991). "Universal portfolios,"
Math. Finance, Vol 1, pp 1-29.
Figure 1. Graph of against displaying the local
maximum at for data set A
[2]. Cover, TM, and Ordentlich, E (1996).
"Universal portfolios with side
information," IEEE Trans. Inform. Theory,
Vol 42, pp 348-363.
[3]. Helmbold, DP, Schapire, RE, Singer, Y,
and Warmuth, MK (1998). "On-line
portfolio selection using multiplicative
updates" Math. Finance, Vol 8, pp
325347.
[4]. Tan, CP, and Tang, SF (2003). "A
comparison of two types of universal
portfolios based on some stock-price data,"
Malaysian J. of Science, Vol 22, pp
Figure 2. Graph of against displaying the local 127-133.
maximum at for data set B

66
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Optimal Design The Interval Type-2 Fuzzy PI+PD Controller And

Superconducting Energy Magnetic Storage (SMES) For Load Frequency
Control Optimization On Two Area Power System

Muh Budi R Widodo(1), M Agus Pangestu H.W(2)

(1)
Electrical engineering of ITS graduate, Surabaya, East java, Indonesia
Email: Boedy@elect-eng.its.ac.id
(2)
Merine Engineering of ITS graduate,Surabaya, East java, Indonesia
Email: Agoezt@ne.its.ac.id

ABSTRACT requires more than one division of power

This paper represents the application of supercon- supplied to the load, so there is no waste of
ducting magnetic energy storage (SMES) and energy [3,4,5]. The problem that arises is the
interval Type-2 interval fuzzy PI+PD (IT2FPI+PD) burden that is always growing all the time, the
controller to improve the performance of power ability of plants to respond in terms of increase
system frequency are two areas that were changed
or decrease output rapidly and accurately from
due to changes in the load of the system. The
results of this study is the result of simulation zero to full load and becomes zero again is
studies using MATLAB. Through this simulation is very important to be considered [1,3,7]. If the
performed to observe the performance frequency frequency changes and hold left until you
response of the system in two areas electricity in reach 10% then the generator will be out of
which use the application of fuzzy PI+PD sync, so it will disrupt the stability of electric
controller, IT2FPI controller, and combination power system, so we need a reliable frequency
IT2FPI+PD with SMES. From the simulation settings [5.12]. Frekeuensi settings and load in
results obtained by the fact that the system uses a electric power system is known as Load
combination IT2FPI and SMES have the best Frequency Control (LFC) [1,3,5,6].
performance. It can be seen from the overshoot
The main objective of the LFC in power
and settling time of the frequency of a system in
their respective areas of -0.000395 p.u and 2.667 systems: maintaining the system frequency
seconds, in two areas each -0.0001091 p.u and 6.72 changes in the allowable limit and minimize
seconds. While the power transfer between areas the power transfer power flow on tie-line [5.8].
for each mising has overshoot and settling time - To make the adjustment burden and frequency,
0.0001925 p.u and 2.139 seconds. the two variables on the frequency and tie-line
power exchange and weighed together by a
KEYWORDS linear combination to form a single variable
Two Area Power Sistem; Load Frequency Control; called Area Control Error (ACE) [1,2,3]. ACE
Interval type-2 Fuzzy PI+PD controller; Supercon- is used as the control signal input in setting the
ducting Magnetic Storage (SMES). load and frequency [1,3,5,6]. At ACE settings
conventional PI control and PD control are the
I. INTRODUCTION main choice among industry [1.9]. usage an
In the power system frequency is a very easy and uncomplicated to make this
important parameter and is representative of controller has a fascination [1.10]. Combined
active power system [1.5]. Good frequency these two kotroler possible to make the ACE
settings can ensure constant rotation for a load to zero, which means the system has stabilized
of synchronous machines and induction [1,5,10]. But konrtoler PI and PD has
machines [1]. This constant rotation is very weakness in determining the gain is amplified,
important to obtain the desired system perfor- so often this control method
mance [1,2]. In large-scale interconnection digabungkandengan other methods such as
systems both large and small generators fuzzy logic. Fuzzy Logic, or fuzzy logic has
connected to one (interconnection), all been known as a reliable and proven method
synchronous machines work so generetor must to improve the performance of fuzzy systems
operate at the same frequency [1,3,5]. The that use simple and requires no mathematical
number of generators that supply the load equation to solve the problem of fuzzy logic

67
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

makes it very at favorite among the f1 + - f2

1  2
researchers. Along with many improvements
perkembangnya fuzzy experience. In 1965 A 1 1
sM1  D1 T/s sM2  D2
lotfi Zadeh Interval Type-2 Fuzzy Logic, an
Ptie
upgrade of Fuzzy Logic. Interval Type-2 - - + -
PL1 PL2
Fuzzy improve the definition of antacedent + +
Pm1 Pm2
and consequent weakness of the fuzzy type-1
with dual functions namely Lower mebership 1
1 sTCH1
1
1 sTCH 2
membership function (LMF) and upper 1 1
Y1 Y2
mebership function (UMF) [1.11] RG1 RG2
1 1
Use of fuzzy controllers in electric power 1 sTg1 1 sTg 2
systems are very popular among researchers,
but to improve the performance of this - -
controller, the electric power system was also + +
installed superconducting Magnetic Storage
(SMEs). These SMES are the direct current Pc1 Pc2

equipment is capable of storing energy in the figure 1. Two area power system [2]
form of the magnetic field. SMES have the
advantage that is able to provide active power III. INTERVAL TYPE-2 FUZZY PI+PD
and reactive power at the same time on power CONTROLLER, SUPER
system with 98% efficiency. This is very CONDUCTING MAGNETIC
helpful in supplying energy when interference STORAGE (SMES)
occurs in the form of changes in load on the In the control frequency, the controller has
system, so the frequency of the system can be an important role is to maintain the system
kept constant [12.13]. The paper used the frequency. To assist in maintaining the
fuzzy PI + PD controller type-2 at two area stability of the controller system installed
power system and superconducting Magnetic SMES units which hace function as batrai. At
Storage (SMES) to improve the system 3.1 and 3.2 points will be discussed on
performance decreases due to changes in load modeling both the equipment..
through simulation using MATLAB. With the
application controller and the SMES in the two 3.1 Design of PIPD Controller [5,6]
area power system is expected to increase the Transfer function of the PI controller in the
performance of the system. form of the z-transform equation is given by,
KP Ki
u (n T ) = (e(n T ) - e(n T - T )) + e(n T )
PI T T
II. LOAD FREQUENCY CONTROL
(LFC) (1
In frequency control (LFC) have two physi- )
u (n T ) = u PI (n T - T ) + K uPi u PI (n T )
cal quantity measured were frequency and tie PI
line load. Two magnitude in comparison with (2
the reference scale and the difference is used )
to determine the corrective measures carried therefore Transfer function of the PD
out by adding or reducing power generated controller in the form of the Z-transform
[2,3,4]. Linear model of power system shown equation is given by,
by figure 2. u
K K
(n T ) = P  e(n T )-e(n T -T )  + D  e(n T )-e(n T -T ) 
PD T T
(3
)
u PD (n T ) = -u (n T - T ) + K uPd u (n T )
PD PD
(4
)
Where, Kp, Ki and Kd is gain of
proportional, integral and differential, Upi and
Upd is output of Fuzzy PIPD controller, and nT
is time variable. Consider of equation 1 to 4

68
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

obtained block diagram of the fuzzy PIPD energy storage system (BESS), SMES
controller that shown in Figure 2. have few advantages that are able to
provide active power and reactive power
generating system simultaneously with the
efficiency reaches 98% [12.13]. Fast res-
ponse to load changes to make SMES are
very interested. Basically the SMES units
consist of superconducting coil, cryogenic
cooling systems, and power Conversion
System (PCS).
Figure 2. IT2FPI+PD controller SMES unit consists of the DC
superconducting coil and the converter is
3.2 Membership function [6] connected to the transformer Y-Δ/YY.
Input membership functions of fuzzy PI + Initial charging coil using a first flow Ido,
PD is indicated by Figure 3, namely in the ignoring losses in the converter, the
form of error (ep) and the delta error (ev).
inductor DC voltage equation can be
expressed,
Ed  2Vd0 c os  - 2Id Rc
(5)
when there are changes in the system load
current and voltage changes, which are stated
Figure 3, Membership Function Input inductor,
1
the membership functions od output shown in E d i( s )  K oi
1  sTdci
Bi  fi ( s )  Pi ( s )
figure 4, which formed as a single output.
1
 K Idi Idi ( s )
1  sTdci
(6)
1
 I di ( s )  E di ( s )
sL i

(7)
figure 4. membership fungtion of output SMES active power changes from the inductor
can be expressed by,
Rule base from input from the membership
function of input and output above is shown in PSMi ( s )  Edi Ido ( s )  Idi Edi ( s )
Table 1. (8)
Table 1. Rule Base Fuzzy Type-2
from equations 6, 7 and 8, it can be made
Delta error (ev) block diagram of a unit of SMEs as shown in
Rule evn evp Figure 5,
Error
epn on zo
(ep)
epp zo op

3.3 Superconducting Magnetic Storage

(SMES) [12,13]
SMES is unidirectional flow devices
which store energy in a magnetic field
[12]. Of the several kinds of energy
Figure 5. Block diagram of SMES
storage technologies such as Compressed IV. SIMULATION RESULT
Air Energy Storage (CAES), Battery
69
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

The simulation using MATLAB 7.3, by second area and tie line power transfer
observing the response frequency stability and between areas also show similar things, it can
power transfer between areas (Ptie). The be seen in Figure 7 and 8. For more details
system was given the disruption in one area about the frequency response of area 1, 2 and
such as load change of 0.1 pu. simulation Ptie can be seen in Table 2 and Table 3.
results shown in Figure 6, 7 and 8.
x 10
-3 respon frekuensi area 1(pu) Table 2. Frequency Response of 1-th and 2-nd area
2
X: 1.788
Y: -0.0001298
Overshot (p.u) Time Settling (second)
0 Kontroler
1-th area 2-nd area 1-th area 2-nd area
-2 Fuzzy PI -0.1167 -0.00688 16.78 17.19
IT1FPI+PD -0.01064 -0.00583 13.96 15.34
amplitude (pu)

4
3
-4 IT2FPI+PD -0.008102 -0.00397 11.96 13.55
2 IT2FPI+PD
-6 -0.000395 -0.0001091 2.669 6.72
1 SMES
-8 1. Fuzzy PI Controller
2. IT1FPI+PD Controller
3. IT2FPI+PD Controller Table 3. P-tie Response on Two Area Power System
-10 4. IT2FPI+PD SMES
Ptie
-12 Kontroler
0 2 4 6 8 10 12 14 16 18 20 Overshoot ts (second)
Waktu (detik) Fuzzy PI -0.0138 >20
Figure 6. frequency response on 1-th area IT1FPI+PD -0.01119 18.88
IT2FPI+PD -0.0071 16.12
IT2FPI+PD
x 10
-3
respon frekuensi area 2(pu) -0.0001925 2.139
1 SMES
X: 3.202
Y: -9.953e-005
0
V. CONCLUSIONS
-1
Applications of combination IT2FPI + PD
-2 and SMES can increase the frequency of the
amplitude (pu)

-3 3
system performance significantly. This can be
2
seen from the overshoot and settling time of
-4
system frequency response and tie line power
-5 1 1.
2.
Fuzzy PI Controller
IT1FPI+PD Controller
transfer as shown by Table 2 and Table 3.
3. IT2FPI+PD Controller
-6 4. IT2FPI+PD SMES

-7
ACKNOWLEDGEMENTS
0 2 4 6 8 10 12
Waktu (detik)
14 16 18 20
Thaks you very much we appreciate to
Figure 7, frequency renspon 2-nd area PSOC laboratory members and ITS, which
0
respon P-tie two area power system (pu) has provided facilities so that research
papers can be resolved in time.
-0.002

-0.004 REFERENCES
[1] Imam Robandi,” Desain Sistem Tenaga
-0.006
4
Modern, Penerbit ANDI, Yogyakarta,
-0.008 3
2006.
2 [2] Hadi Saadat, Power System Analysis 2nd
-0.01
1 1. Fuzzy PI Controller Edition, McGrowHill. 2004.
2. IT1FPI+PD Controller
-0.012 3.
4.
IT2FPI+PD Controller
IT2FPI+PD SMES
[3] Kundur,.P, Power System Stability and
Control, McGraw-Hill,Inc.,1994
-0.014
0 2 4 6 8 10 12
Waktu (detik)
14 16 18 20 [4] Anderson P.M, Fouad A.A, Power
Control and Stability, The lowa State
Figure 8. respon of tie line power between 1-th
and 2-nd area University, Press.1982.
[5] Muh Budi R Widodo, Imam Robandi,
Figure 6 represents a frequency response “Optimization of Fuzzy PIPD Controller
in first area. That figure whoen that the for Excitation System Stability Analysis
frequency response of two area power systems on Single Machine Infinite Bus (SMIB)
by using IT2FPI + PD combined with SMAS using Genetic Algorithm (GA)” ICAST.
has the best response, settling time and 2009
overshoot frequency of each system is - [6] Muh Budi R Widodo, Muhammad abidil-
0.000395 p.u and 2.667 seconds. Response in lah, Imam Robandi,” Optimal Design

70
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Load Frequency Control On Multi Area : the rating change from the error signal
Power System Using Interval Type-2 Id : the current trough into the coil
Fuzzy PI Controller“. APTECS-2009. I0 : initial SMES current
Paper number: 034. Ki : integral gain
[7] Muhammad Abdillah, “Desain Optimal : integral gain discrete domain
KId : constant gain of Id feedback
Fuzzy Logic Load Frequency Control
Kp : proportional gain
pada Sistem Tenaga Listrik Menggunakan
: proportional gain in discrete domain
Artificial Immune System Via Clonal
KSMES : constant gain of SMES
Selection. Tugas Akhir, Jurusan Teknik
Ku : constant gain of Pi controllr
Elektro ITS, 2009. L : inductance of SMES
[8] Jawad Talaq and Fadel Al-basri, M1 : inertia equivalent on area 1
Adapteve gain sceduling foe load M2 : inertia equivalent on area 2
frequency control, IEEE Transc. on nT : continues-time frequency domain
Power Systems, Vol 14. No 1, February, P12 : load exchange between area 1 and area
pp. 145-150, 1999. 2
[9] Charles E. Fosha, Jr., and Olle I.Elgerd, PL1 : non frequency load change on area 1
The Megawatt-Frequency Control PL2 : non frequency load change on area 2
Problem: New approach via Optimal Pm1 : mechanical power on area 1
Pm2 : mechanical power on area 2
Control Theory, IEEE Trans. Vol. PAS.
Pref1 : load reference from prime over on area
No.4, April 1970, pp.563-577. 1
[10] Ceyhun YILDIZ, A. Serdar YILMAZ, Pref2 : load reference from prime over on area
Mahmet BAYRAK, “Genetic Algorithm 2
based PI Controller for Load Frequency PSMES : SMES power
Control in Power system”. Proceding of R1 : speed drop of governor
5th International symposium on Intelegent Rc : commutating resistance
Manufacturing System, may 29- T : time sampling period
31,2006:1202-1210 T12 : synchronous torque coefficient
[11] Jerry M.Mandel, Robert I. Bob John,” TCH1 : time constant of thermal turbine on 1-
th
Type-2 Fuzzy Sets Made Simple”, IEEE,
TCH2 : time constant of thermal turbine on 2-
April, 2002. nd
[12] Satrio haninditho, design Automatic TDC : time constant of converter
generation control (AGC) dengan unit TG1 : time constant of governor on area 1
superconducting energi magnetic storage TG2 : time constant of governor on area 2
(SMES) pada sistem tenaga menggunakan u : output of PI controller
Fuzzy Proportional integral (Fuzzy PI), Y1 : governor output on area 1
unpublised. Y2 : governor output on area 2
[13] Detailed modeling of superconducting z : discrete-time frequency domain
Magnetic Energy Storage(SMES) system. α : firing angle
δ1 : rotor angle on area 1
Transaction om power dilivery, vol 21, no
δ2 : rotor angle on area 2
2, november 1992. ω1 : speed of rotor on area 1
ω2 : speed of rotor on area 2

NOMENCLATURE

Δ : the deviation
A12 : rate area capacity on area 1 and area 2
B1 : bias frequency factor on area 1
B2 : bias frequency factor on area 2
D : load damping constant
D1 : load damping constant on area 1
D2 : load damping constant on area 2
f1 : frequency output on area 1
f2 : frequency output on area 2
E : error signal from the system
Ed : DC voltage applied to conductor
: the error signal rate

71
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Dependence of biodegradability of xenobiotic polymers on population of

microorganism

Masaji Watanabe (1) and Fusako Kawai (2)

(1)
Graduate School of Environmental Science, Okayama University, Okayama, Okayama Prefecture, Japan
(Email: watanabe@ems.okayama-u.ac.jp)
(2)
R & D Center of Bio-based materials, Kyoto Institute of Technology, Kyoto, Kyoto Prefecture, Japan
(Email: fkawai@kit.ac.jp)

ABSTRACT II. MODELING OF EXOGENOUS

Biodegradation of polyethylene glycol is studied. DEPOLYMERIZATION PROCESSES
Analysis is based on exogenous type Let t and M be the time and the molecular
depolymerization model. Techniques developed in weight respectively. Suppose that an M -
previous studies were used to determine the molecules is a molecule with molecular weight
molecular factor of degradation rate. The time
M. Let w(t,M) represent the total weight of M
factor of degradation rate is determined by
assuming a logistic-type growth of the microbial -molecules present at time t, and L be the
population which utilized degraded monomer units amount of weight loss due to one cycle of
as the sole carbon source. exogenous depolymerization. In an exogenous
depolymerization process of PEG, a molecule
KEYWORDS is first oxidized at its terminal, and then an
Biodegradation; polyethylene glycol; ether bond is liberated. It follows that L = 44
mothematiocal model; numerical simulation; (CH2CH2O). Figure 1 shows the weight
differential equation. distribution of PEG before and after
cultivation of the microbial consortium E1.
I.INTRODUCTION Let  t, M  be the degradation rate, which
Microbial depolymerization processes are is the rate of the weight conversion from the
generally classified into two types. Those two class of M -molecules to the class of (M – L)-
type are exogenous type and endogenous type. molecules at time t. Then x  wt , M  and
In an exogenous type depolymerization y  wt , M  L  satisfy
process, monomer units are truncated from
dx M
terminals of molecules stepwise. In an    t , M x   t , M  L  y . (1)
endogenous type depolymerization process, dt M L
molecules are split at arbitrary positions. In This equation is associated with the initial
this study, a mathematical model for condition
exogenous type depolymerization processes is
analyzed to investigate biodegradation of w0, M   f M  , (2)
polyethylene glycol (PEG).
In previous studies, mathematical
techniques are developed for analysis of
exogenous depolymerization processes [1, 2, 3,
4, 5, 6, 7, 8, 9, 10, 11]. In this study, analysis
of exogenous type depolymerization processes
of PEG is continued. Techniques developed in
previous studies are used to determine the
molecular factor of degradation rate. The time
factor of degradation rate corresponds to the
microbial population which utilizes degraded
monomer units as the sole carbon source.

72
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 1. Weight distribution of PEG before and after cultivation of the microbial consortium E1 [3, 6, 8, 11]

where f M  is the initial weight distribution.

dx
Given an additional weight distribution g M    t  M x
at timet  T , dt
M (3)
wT , M   g M  , (3)   t  M  L  y.
M L
equation (1), the conditions (2) and (3) form Let W  , M   wt , M  , where
an inverse problem to find the degradation rate
  0  s  ds ,
t
 t, M  for which the solution of the initial
value problem (1), (2) also satisfies the
condition (3). and let X  W  , M  and Y  W  , M  L  .
Then the equation (3) becomes
III. MOLECULAR FACTOR AND TIME dX M
FACTOR OF DEGRADATION RATE    M  X    M  L  Y. (4)
d M L
A time factor of the degradation rate such as Given the initial weight distribution f M ,
microbial population affects the molecules equation (4) is associated with the initial
regardless of their size. Then the degradation condition
rate is the product of time factor  t  and
molecular factor  M, and the equation (1) W 0, M   f M  . (5)
becomes

73
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 2. Molecular factor of the degradation rate based on the weight distribution before and after cultivation of the
microbial consortium E1 for three and five days [3, 8, 11].
amount of the monmer units consumed by the
Given an additional weight distribution microorganism at time is
g M  at    ,
A  0  t  M wt , M  dM .


W , M   g M  , (6)
The growth rate of the microbial population is
equation (4) and the conditions (5) and (6) proportional to
form an inverse problem to find the
degradation rate for which the solution of the  t  1
1 h  1 h ,
  M wt , M  dM
initial value problem (4) and (5) also satisfies 
A
the condition (6). The inverse problem was 0
solved numerically for the initial weight and satisfies the equation
distribution and the weight distribution after
three days of cultivation shown in Figure 1. A d  1 
numerical result is shown in the figure 2.  k 1  h    , (7)
In the biodegradation process of PEG dt  0  M wt , M  dM 
 
shown in the figure 1, the microbial population
is the only time factor, and the monomer tunits where h and k are positive constants.
truncated from in the exogenous Equation (7) is associated with the initial
depolymerization process is the sole carbon condition
source. Then the time factor  t  is regarded
as the microbial population, and the total

74
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 3. Transition of the weight distribution of PEG after cultivation for three days.

 0    0 . (8) corresponding time evolution of the microbial

population.
IV. SIMULATION OF EXOGENOUS
DEPOLYMERIZATION PROCESS OF V. CONCLUSIONS
PEG BASED ON COUPLED MODEL Change of microbial population was taken into
Once the time factor and the molecular factor consideration in modeling of exogenous
of the degradation rate are found, the transition depolymerization process. Under constant
of the weight distribution can be simulated by supply of carbon source, growth of the
solving the initial value problem (3), (7), (2), microbial population would be logistic with a
(8). In previous studies, the initial value constant carrying capacity [12]. However
problem was solved numerically for a certain under limeted resource, the carrying capacity
set of prameter values of h , k , and  0 . In decreases as microbial population consumes
this study, exogenous depolymerization the carbon source. This scenario was
process of PEG is also simulated for the incorporated into modeling of the equation (7).
following values of parmeters [3, 8]. Comparison between the numerical results and
the experimental results shown in the figure 3
 0  0.029827, k  2.0 , h  700. and 4 indicates that the model is appropriate
for experimental setting.
Numerical results together with experimental In practice, biodegradability should also be
results are shown in the figures 3 and 4. Those dependent on other factors such as temperature,
figures show the weight distribution of PEG dissolved oxygen, etc. Once these essentials
after cultivation for three and five days. The are incorporated into the time factor of the
degradation rate, the exogenous

75
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 4. Transition of the weight distribution of PEG after cultivation for five days.
and Stability, 86, 105 - 114, 2004.
depolymerization model should be applicable doi:
to assess the biodegradability of xenobiotic 10.1016/j.polymdegradstab.2004.03.0
polymers in the environment. 15
[3] M. Watanabe and F. Kawai, Study on
ACKNOWLEDGEMENTS
effects of microbial population in
The authors thank Ms Y. Shimizu for her
degradation process of xenobiotic
technical support. This work was
polymers, Submitted.
supported by JSPS KAKENHI 20540118.
[4] M. Watanabe and F. Kawai, Numerical
REFERENCES simulation of microbial
[1] F. Kawai and M. Watanabe and M. depolymerization process of
Shibata and S. Yokoyama and Y. exogenous type, Proc. of 12th
Sudate, Experimental analysis and Computational Techniques and
numerical simulation for Applications Conference, CTAC-
biodegradability of polyethylene, 2004, Melbourne, Australia in
Polymer Degradation and Stability, September 2004, Editors: Rob May
76, 129 - 135, 2002. and A. J. Robert, ANZIAM J., 46, E,
doi: 10.1016/S0141-3910(02)00006-X C1188 - C1204, 2005.
[2] F. Kawai and M. Watanabe and M. http://anziamj.austms.org.au/V46/CTA
Shibata and S. Yokoyama and Y. C2004/Wata
Sudate and S. Hayashi, Comparative [5] M. Watanabe and F. Kawai,
study on biodegradability of Mathematical study of the
polyethylene wax by bacteria and biodegradation of xenobiotic
fungi, Polymer Degradation polymers with experimental data
introduced into

76
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 5. Transition of the weight distribution of PEG after cultivation for seven days.

analysis, Proceedings of the 7th and Applications Conference,

Biennial Engineering Mathematics EMAC-2005,
Melbourne, Editors: Andrew Stacey xenobiotic polymers based on
and Bill Blyth and John Shepherd and experimental results, BIOSIGNALS
A. J. Roberts, ANZIAM J., 47, C665 2009, Second International Conference
-C681, 2007. on Bio-inspired Systems and Signal
http://anziamj.austms.org.au/V47EM Processing, Proceedings, Porto -
AC2005/Watanabe Portugal, 14-17 January, 2009, 25 - 34,
[6] M. Watanabe and F. Kawai, INSTICC Press
Mathematical analysis of microbial [8] M. Watanabe and F. Kawai, STUDY
depolymerization processes of ON EFFECTS OF
xenobiotic polymers, Proceedings of MICROORGANISM IN
the 14th Biennial Computational DEPOLYMERIZATION PROCESS
Techniques and Application OF XENOBIOTIC POLYMERS BY
Conference, CTAC2008, ANZIAM J., MODELING AND SIMULATION,
50, C930 - C946, 2009. Proceedings of the First International
http://anziamj.austms.org.au/ojs/index. Conference on Bioinformatics,
php/ANZIAMJ/article/view/1465 Valencia, Spain, January 20 - 23, 2010,
[7] M. Watanabe and F. Kawai, Modeling Editors: Ana Fred and Joaquim Filipe
and simulation of biodegradation of and Hugo

77
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 5. Evolution of microbial population

Gamboa, 2010 INSTICC - Institute for exogenous depolymerization of

Systems and Technologies of Information, xenobiotic polymers, Mathematical
Control and Communication, 181 – 186, Biosciences, 192, 19 – 37, 2004.
2010. doi: 10.1016/j.mbs.2004.06.006
[9] M. Watanabe and F. Kawai and M.
Shibata and S. Yokoyama and Y. [11] M. Watanabe, F. Kawai, STUDY ON
Sudate, Computational method for EFFECTS OF MICROORGANISM IN
analysis of polyethylene BIODEGRADATION OF
biodegradation, Journal of XENOBIOTIC POLYMERS BASED
ON MODELING AND SIMULATION,
Computational and Applied Proceedings, International Conference on
Mathematics, 161, 1, 133 – 144, Environmental Research and Technology
December, 2003. doi: (ICERT 2010), 2-4 JUNE 2010,
10.1016/S0377-0427(03)0051-X PARKROYAL PENANG, MALAYSIA,
442-448.
[10] M. Watanabe and F. Kawai and M. [12] J.D. Murray, Mathematical Biology,
Shibata and S. Yokoyama and Y. Second Corrected Edition, Springer –
Sudate and S. Hayashi, Analytical Verlag, Berlin, 1989
and computational techniques for

78
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

PROTOTYPE OF VISITOR DISTRIBUTION DETECTOR

FOR COMMERCIAL BUILDING

Sukarman, Suharyanto, Samiadji Herdjunanto

(1)
Sekolah Tinggi Teknologi Nukir-BATAN (Polytechnic institute of Nuclear Technology-BATAN)
Jl. Babarsari Po.Box 6101 YKBB Post-code 55281
Telp. 0274. 484085 Fax 0274.0274 489715
Email:sukarmanst@batan.go.id, sukarman@hotmail.com
(2), (3)
Departemen of Electrical Engineering, Gajah Mada University Yogyakarta
Jl. Grafika Bulak Sumur Yogyakarta

ABSTRACT distribution pattern daily that is shaped in the

The prototype of visitors distribution detector for form of user interface graphics,and getting of
commercial building, aiming to be able to know the air conditioning temperature control baseds on
distribution pattern of people on commercial the distribution pattern of visitors.
buildings as a function of time, so building According to regulation, Number 30rd,
manager can make a visit report distribution
2007 on Energy, chapter 25, "Energy
patterns, which make electric energy saving
optimization to reduce cost. Energy savings made conservation is the responsibility of national
by regulating the temperature of the air- governments, local governments, employers,
conditioner, base on density of the visitors. To and society." In this case is an energy-
monitor the distribution of visitor use sensors efficient, with regard to rational, sensible and
Passive Infrared (PIR), it was placed 2x3 sensor efficient without compromising comfort and
array with an area of 50 m2 sized room 5.77 m x productivity.
8.66 m. These sensors placed on the roof / ceiling Accordingly, one of the government
pointing downwards with a maximum range of policy to overcome the energy crisis was the
detection distances as high / far as five meters at an issuance of Presidential Instruction No. 2 /
angle of 300. The output of this sensor array is
2008 on Saving Energy and Water is expected
connected to Microcontroller ATMEGA 8535 as a
controller to communicate with a remote control to further accelerate the implementation of
and network modules (modules network) using energy saving and water because the current
NM7010A, LF. Output Module-LF NM7010A energy crisis is not only experience but also
network switch connected to the switch Hub so that the water crisis.
data can be displayed to the network and can be Energy and water savings are not only
done with computer data acquisition. To acquire required for government agencies, but also the
the data used LabVIEW software by the National industrial sector, business / commercial and
Intrument (NI). The result showed that this household. Included in the category of the
prototype sensor with a size of 2x3 array can business sector / commercial such as shopping
effectively detect the spread of visitors with a room
centers, malls, private offices, hotels, and
size area of 50 m2 with a maximum distance of 5
meters. Control action to change the temperature entertainment venues.
setting on the remote control air conditioning work According to the criteria of ASEAN, a
in accordance with the pattern of distribution of building energy classified lowcost if the
people / visitors. maximum energy consumption 150 kWh per
year per square meter. While the energy use of
KEYWORDS 150-200 kWh per square meter per year are
microcontroler, network module, sensor array. classified saving, 200-250 kWH per year per
square meter belonging to ordinary, and more
I. INTRODUCTION than that including energy wasteful buildings.
The purpose of research are getting a Based on a survey conducted by the
prototype for detection of commercial Indonesian Association of Building Physics
distribution of visitors in the building, getting (IAFBI), the average building in Jakarta to
data acquisition of distribution visitors on spend 310 kWh per square meter per year and
commercial buildings, getting a data no more than 10% alone, buildings in Jakarta

79
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

are using the default number is close to the relations. Cool weather conditions will provide
energy [12]. a comfortable atmosphere for visitors who
Office buildings, industrial and come, because it conditions the air to be
commercial buildings, is a great place to adjusted by the number of visitors. If the
absorb a lot of electricity, whether it is for visitor a bit, then the temperature is not too
lighting, fans, water pumps, material cold. Logically, if the crowds, temperatures
processing, material handling, refrigerator, air- would increase. To achieve a comfortable
conditioning system (Air conditioning), spot, ie a state of thermal comfort. At this
elevators and motors other drivers. point the air temperature, air circulation and
Energy consumption in buildings cleaning does not bother about human
consist of lighting, air conditioning, elevators, performance. Thermal comfort standards for
and so forth [7]. Recorded almost a quarter of tropical countries ranging from 260C with 50-
annual world energy supply in the late 80s and 60% humidity [4]..
nearly two-thirds of this energy in the supply Based on a survey of existing
of fuel oil and gas reserves are concerned commercial buildings in Yogyakarta include
about the age of no more than 100 years (30 AMPLAZ, some Batik Mirota Bookstore and
years for Indonesia) (World Energy, 1991). showed that the air conditioner is turned on is
Evaluated from the equipment used, not scheduled. Therefore the energy used for
the energy consumption for air conditioning air conditioning are used continuously at low
(AC) reached 72% of the total electricity temperatures between 16-20 oC since the
usage[16], plus in the early planning is always building opened at 9:00 pm until 20:30 o'clock
exceeds capacity by 10-15% based on pm. if the burden of a large room / number of
consideration for future development and at visitors quite a lot of the air conditioning load
the request of the owner of the building[24]. becomes heavy, especially the unequal
That is, there is wasteful investments because distribution of visitors as a result the air-
the air conditioner is designed from the needs conditioner working at full load. This means
exceed the capacity of the building. the electrical energy used to supply engine
Electric energy waste occurs when the cooling fresh air to be great.
AC is not being to handle peak load, or when Efforts to save energy use electricity
they're low load (partial load), air conditioners for air conditioning, particularly commercial
have to work at full load, so that power buildings of which have been made based on
consumption by the AC remains relatively the distribution of visitors a function of time,
high [1,14, 20]. although other factors also influence such as
Efforts for energy saving in air building materials, infiltration, humidity and
conditioning systems based on research that so forth [23]. If the distribution can be known
never existed [21,13], such as: adjustable fan that it will be able to obtain the benefits that
speed, refrigerant flow rate, indoor many of them can know the place or location
temperature control (thermostat), and water where the most visitors, can then be digelarnya
flow control is carried out experimentally and the activities that the time is right (effective),
simulation. and automatic data requests will be stored as a
In conventional air conditioning systems, data logger that provides efficient traffic on
motors only know of two conditions based on the part of the building management. The
temperature settings. If room temperature is important thing is to make use of data
greater than the setting temperature of the distribution management, management of
motor will be operated (On) and vice versa visitors for the benefit of electricity, especially
will not operate (Off) if temparatur space the engine cooling air-conditioning settings.
smaller than the setting temperature. Saving All that will have an impact on increasing
the system will be obtained at an interval of revenue / sales for the supermarket / mall.
the motor does not operate. The more frequent Although the AC there are sensors that
occurrence of fluctuations due to the smaller are used to measure and regulate the
cooling load will be the ability to save energy conditions at room temperature. However, this
[20]. An air conditioning (AC) is necessary sensor is sensitive and accurate measure of
because it will provide the comfort factor, temperature is only around in / AC machines,
occupational health, productivity, aesthetics but not accurate to measure the room
related to limitations of space and community temperature varies with the number of visitors.

80
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Existing sensors on the AC machine is not source Ip(t), induced by heat flux absorbed by
sensitive and can not be used to determine the the sensor [2, 3]. The equivalent circuit in
distribution or the number of visitors. To be Fig.1 illustrates this interpretation.
able to know the distribution of visitors in a
room then used the PIR (passive infra red)
sensor. PIR sensors are infrared sensors that
can detect the presence of people (due to
temperature changes people) in a room /
building.
Energy conservation should be
implemented because if the energy price
increases continuously, the threat of rising
rates and soaring world oil prices will affect
the basic needs of the institution / manager. In
this condition then the manager should try to Figure 1. Equivalent circuit of pyroelectric
identify and explore the possibility of energy sensor and preamplifier
management that can be done.
Based on the above description as an Assuming the uniform structure of the
electric energy saving measures in commercial pyroelectric material and uniform heating, the
buildings especially for electrical energy current Ip(t) may be determined from
consumption in refrigeration (AC) which it relationship:
needs to be based on the optimal distribution
patterns of visitors in the space / building, dT
IP  P S (1)
therefore the following will be described in dt
further detail. Where,

P  pyroelectric coefficient
II. THEORY BASIS
Pulse radiation sensors may be divided in two dT
 speed of temperature changes of the
basic groups: photon detectors and thermal dt
detectors, in which the radiation absorbed is pyroelectric material
first converted into heat that, in turn, causes a S  surface area of the sensor electrode.
measurable effect. The main difference Let us assume that there are no additional heat
between photon and thermal detector lies in losses, the duration of measured radiation
the fact that the first measures the radiation pulse tI is very short and it meets inequality:
power, and the latter – the radiation energy.
tI  E  C (2)
Among thermal detectors the following
examples may be mentioned: pyroelectrics,
thermocouples, thermistors, and bolometers. E  electric time-constant of the whole
Particularly important group of elements are electric circuit:
pyroelectrics.
Pyroelectric sensor as such is a Rd  R E
capacitor formed by depositing metal E  (C d  C E ) (3)
R d .R E
electrodes on a both surfaces of the thin slice
of pyroelectric material. The absorption of the C  thermal time-constant of the sensor.
radiation pulse of power P(t) by the
pyroelectric material results in a change in its Then it is possible to assume that the speed of
temperature with the value of ΔT, which temperature changes of the pyroelectric
causes a polarisation change, and, material is a linear dependence of the power of
consequently, results in displacement of radiation P(t) falling on the sensor –
electric charges in the pyroelectric material, relationship:
hence the displacement current Ip(t) occurs
[1]. It is possible therefore to accept the dT 
 P (4)
interpretation, that the capacitor of dt c d . .S .d
pyroelectric sensor is loaded from current

81
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

  absorption coefficient of radiation falling

on the sensor
cd  specific heat of the pyroelectric material
density of the pyroelectric material
  surface of the sensor electrode
d  thickness of the pyroelectric material slice
relationship (4) into relationship (1), we obtain
formula (5) describing the intensity of
displacement current Ip(t): Figure 2. a digital signal output of PIR sensor.

P. P.
IP  P (t ), r1  (5)
c d . .d c d . .d

We obtain relation,

I P  r1.P (6)

The value of output voltage signal is given by

1
C
U I P (t )dt (7)

Figure 3. Directions reach PIR sensors with and without

Passive Infra Red Sensor (PIR) sensor caps protective.
,which is the smallest type of KC7783, is
already equipped with fresnel lens and have Figure 3, shown range sensor without caps
the output (output) digital. its use does not protection. Figure 4, shown internal circuit
require the signal conditioner circuit (signal PIR sensor, which is packaged into module.
conditioners) are complicated[11].
PIR sensor has two pieces of the
sensor element connected in
series[2,3,5,6,8,9,10]. This arrangement would
eliminate the signals caused by vibration,
temperature changes and sunlight. PIR sensor
is applied more widely [15,17,18,22].
A passing through the sensor will
activate first and second sensor elements as
shown in the figure 2.

Figure 4. block diagram of the workings of the PIR

sensors

Fresnel lens serves to limit the energy

of the incoming light and temperature as well
as a filter. IR filter works to filter out certain
wavelengths are allowed to pass. Pyroelectric
sensor works based on changes in
environmental heat. When changes in heat will
be generated output voltage, the stress is

82
 Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

strengthened by the amplifier. Finally, the Every sensor on figure 5, PIR sensor with
output voltage of 5 volts was obtained after protective, need the distance-one with other
cross on a comparator. sensors as wide as 2.89 meters.

RESEARCH METHOD For the space with an area 50 is required as

Materials required in this research include are many as six pieces meter2 sensors arranged in
1. UTP cable, this cable used to connect a matrix (3x2) PIR sensor.
between the PC Hubs and Switches.
2. LabVIEW software, used to acquire data
from the module TCP / IP Starter kit
3. PonyProg2000 software - Serial Device 60o
Programmer Version 2.07c Beta, used to
download programs to the microcontroller 5 mtr
module ATMEGA8535.
4. Basic Compiler software, used to program
1.4 mtr 7.2 mtr
the microcontroller ATMEGA 8535.
5. Fiber glass (mica), 30x20 cm to place the
PIR sensor 7783R KC.
6. PC (personal computer) with the
specifications: minimum Pentium II
Processor with 128 MB mimimum RAM,
10 GB hard disk minimum. This computer
is used for functions to acquire data, store Figure 6. PIR sensor detection area
data and control the means and displaying
the historical data. n the figure 6, indicated the scope of the
7. Module TCP / IP starter kit with an sensor is facing downwards direction with
Ethernet port facilities.
angle 300 with the guards, so that for a
8. ATMEGA8535 Microcontroller module,
which is used to process and build a data
sensor can cover a distance of 2.89
communication module TCP / IP starter meters..
kit.
9. Remote control used for transmit data to Modul 2
Array 6
Mikroko Netw Switch
the system air conditioner (AC). Sensor
ntroler ork hub
PIR (3x2)
ATMega Mod
SYSTEM DESIGN
Actuator
Layout design of distributed detection system (remote Control)
of visitor commercial buildings. Tombol
8.66 m Mode low
suhu Up

Tombol
suhu Dn
computer
Mode normal
Port A.0 portA.1 portA.2
Figure 7. diagram block of detection system.
North
In the figure 7, is shown a block diagram of
A A A data acquisition from the PIR sensor array, the
5.77 m
microcontroller module, network modules,
portA.3 portA.4 portA.5 South actuators and personal computers.
Mikrokontroler used to collect data and
mengontorl data to actuator. Microcontroller is
A A A used to retrieve data and mengontorl data to
the actuator. Network module is used to
transmit data distribution of visitor
information on a computer display. Remote
control is used as an actuator that will send
Figure 5. PIR sensor layout indoor commercial building

83
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

data to the engine coolant based on the

distribution of visitors in a room.
The personal computer was used to record the
distribution of the data distribution of visitors
every time.

RESULTS AND DISCUSSION

In table 1, it appears that the distance sensor
with an object (visitors) that varies from 30 cm
to 600 cm. Shows that at a distance of 30 cm -
500 cm, the sensor can still detect the object.

Tabel.1. PIR individual sensor output as a function of

distance at 30o angle coverage
Proximity Output comment
detection (cm) sensor
30 5.09 High Figure 9. data recording of visitor.
60 5.09 High
90 5.09 High
120 5.09 High In the figure 9, shown graphic user
150 5.09 High interface display data recording visitors to the
200 5.09 High entire room (the area until the area-0-5). These
250 5.09 High data will be stored at any moment into a
300 5.09 High memory to be processed further.
350 5.09 High
400 5.09 High
450 5.09 High III. CONCLUSION
500 5.09 High Distribution with Sensor Detector Passive
550 0.0 Low Infrared (PIR) has a maximum range as far as
600 0.0 Low five meters at an angle 0o with no additional
That is shown by the output voltage of 5 volts, protection. On detection of a large angle
while the output of 0 volts means that sensors without additional protection, the PIR sensor
can not detect the object. can detect objects larger than a smaller angle
of detection.
PIR Sensor 2x3 array with no
additional protection can detect an area of
112.5 m2 with a size of 12.09 x 08.06 meters
and at an angle of detection 600. Meanwhile,
if the additional protection can be used to
detect the room with an area of 50 m2, with a
length of 8.66 meters and 5.77 meters wide at
the angle of detection 300.
This prototype detector distribution,
can be used to determine the distribution
pattern of visitors on commercial buildings
everytime, so it could get benefits for the
manager of a commercial building.
The air conditioning temperature
settings can be done automatically (not
Figure 8. space plan manually) based on the pattern of distribution
of visitors, thereby saving can be obtained
In the figure 8, shown by the graph display to
from the use of electric energy consumption
monitor the status of the space already visited.
due to air conditioner.
Light green color shows that there are visitors
and the color blue of the air conditioner which
indicates the engine coolant is at a lower
temperature than the previous set.

84
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

REFERENCES Bangunan Sebagai Upaya Penghematan

[1]. Arismunandar, Wiranto, 1995, “Penyegar Energi.Padang.
Udara”, penerbit Pradnya Paramita [14]. Nasution, H. dan Mat Nawi Wan Hassan,
Jakarta . 2005. Energy saving for air conditioning
[2]. Brox, E Steven, 1991, Passive Infrared / by proportional control, variable and
Acoustic Pool Security System, United constant speed motor compressor. Proc. of
States Patent the 2nd International Conference on
[3]. Carr, JJ, 1993, Sensor and Circuits: Mechatronics 2005, pp. 492-498, May.
Sensors, transducers, and supporting Kuala Lumpur.
circuits for electronic instrumentation, [15]. Odon, Andrzej, 2001, processing of Signal
measurement, and control, PTR Prentice of pyroelectric sensor in laser Energy
Hall, New Jersey. meter, Measurement Science Review,
[4]. Elyza, Rizki, dkk, 2005, Buku Panduan Volume 1, number 1, Polandia.
Efisiensi Energi di Hotel, ISBN 979- [16]. Qureshi, T. Q. dan Tassou, S. A., 1996.
98399-2-0, Jakarta Variable speed capacity control in
[5]. Fraden, S. J. , 1997, Handbook of Modern refrigeration systems, J. Applied Thermal
Sensors, 2nd ed., NY: AIP Press. Engineering,Vol.16, No.2, pp.103-113.
[6]. Gran et al, 1995, Infra-Red Sensor System [17]. Schwarz, Frank, 1996, InfraRed Detector
For Intelligent Vehicle Highway System, For Detecting Motion And Fire An Alarm
United State Patent, USA System Including The Same, United State
[7]. Karyono,Tri Harso, 1999, Arsitektur Patent
Kemapanan Pendidikan kenyamanan dan [18].Sheffer, Eliezer A, 1992,Pattern
Penghematan Energi, PT Catur Libra Recognizing Passive Infrared Radiation
Optima, Jakarta, Jakarta, h..126. Detection System, United States Patent
[8]. Kumada, Akira, 1993, InfraRed Detector [19]. Sumanto, , 2000 “ Dasar-dasar Mesin
With Phyroelectric Detector Element And Pendingin”, Penerbit Andi Yogyakarta.
Chopper Control Circuit, United State [20]. Sumeru dan Sutandi, T., 2007.
Patent Penghematan energi pada mesin pendingin
[9]. Lang, S. B ,1974, Sourcebook of dengan variasi putaran kompresor, Jurnal
Pyroelectrocity (Gordon and Breach, Teknik Mesin, Vol.7, No.2, pp.80-86.
London) Institut Teknologi Sepuluh November.
[10]. Lee, Wade, 1994, Wide Angle Passive [21]. Tojo, K., Ikegawa, M., Shiibayashi, M.,
Infrared Radiation Detector, United States Arai, N. dan Uchikawa, N., 1984. A scroll
Patent compressor for air conditioners. Proc. of
[11]. Mohd Syaryadi, Agus Adria, dan the 1984 International Compressor
Syukurullah, 2007, sistem kendali keran Engineering Conference at Purdue, pp.
wudhuk menggunakan sensor pir Berbasis 496-503, July, Purdue.
mikrokontroler at89c2051,jurnal rekaya [22]. Wijaya, Handoyo 2008, Aplication Note
elektrika, Volume 6 nomor , .hal 14-20 DT51 team Innovative electronic,
[12]. Nasution, H. dan Mat Nawi Wan Hassan, Surabaya
2004. Hemat energi pada sistem “air [23]. Wilbert F.Stoecker, Jerold W. Jones,
conditioning” sebagai upaya mengatasi 1989, ” Refrigerasi dan Pengkondisian
krisis energi di Indonesia. Proc. of the Udara”, Cetakan kedua, Airlangga,
International Scientific Meeting Jakarta, 1989.
Indonesian Student 2004, September, [24]. Yu., P. C. H. , 2001, A study of energy
United Kingdom. use for ventilation and air-conditioning
[13]. Nasution, H, 2005. Aplikasi Kendali systems in Hong Kong, PhD Thesis, The
Logika Fuzzy Pada Sistem Pendingin Hong Kong Polytechnic University.

85
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

APPLICATION ANFIS FOR NOISE CANCELLATION

Sukarman
(1)
Sekolah Tinggi Teknologi Nuklir-Badan Tenaga Nuklir Nasional , Yogyakarta, DIY
(Email: sukarmast@batan.go.id)

ABSTRACT unknown nonlinear dynamics to generate a

APPLICATION ANFIS FOR NOISE distorted noise d(k), which is then added to
CANCELLATION. The objective of Adaptive x(k) to form the measureable y(k). our task is
noise cancellation is to filter out an interference to retrieve the information signal x(k) from
component by identifying a linier model between a overall otput signal y(k), which consist of
measureable noise and the corresponding
information signal x(k) plus d(k), a distorted
unmeasureable interference. Adaptive noise
cancelation using linier filter has been successfully and delay n(k).The result show that ANFIS did
in real-world application such as interference a good performance to remove unknown
canceling in electrocardiograms (ECGs), echo distorted noise signal d(k) from the measured
elimination on long-distance telephone signal y(k).
transmission lines, and antenna side-lobe
interference canceling. Therefore, it can expand the II. ANFIS SYSTEM
concept of linier adaptive noise cancellation into ANFIS is the Adaptive Neuro-Fuzzy training
nonlinear realm using nonlinear adaptive system. In of Sugeno-type Fuzzy Inference Systems.
this experiment, it will show how ANFIS can be
ANFIS uses a hybrid learning algorithm to
used to identify an unknown nonlinear passage
dynamic that transforms a noise source into an identify the membership function parameters
interference component in a detected signal. of single-output, Sugeno type fuzzy inference
Information signal x(k) and a measureable noise systems (FIS). A combination of leastsquares
source signal n(k), noise source goes through and back propagation gradient descent
unknown nonlinear dynamics to generate a methods are used for training FIS membership
distorted noise d(k), which is then added to x(k) to function parameters to model a given set of
form the measureable y(k). our task is to retrieve input/output data [14,15].
the information signal x(k) from overall otput In 1993 Roger Jang, suggested
signal y(k), which consist of information signal Adaptive Neuro Fuzzy Inference system [7,11]
x(k) plus d(k), a distorted and delay n(k).The result
(ANFIS). ANFIS can serve as a basis for
show that ANFIS did a good performance to
remove unknown distorted noise signal d(k) from constructing a set of fuzzy ‘if-then’ rules with
the measured signal y(k). appropriate membership functions to generate
the stipulated input-output pairs. Here, the
KEYWORDS membership functions are tuned to the input-
Anfis; adaptive system; nonlinear system; output data and excellent results are possible.
. Fundamentally, ANFIS is about taking an
I.INTRODUCTION initial fuzzy inference (FIS) system and tuning
The objective of Adaptive noise cancellation is it with a back propagation algorithm based on
to filter out an interference component by the collection of
identifying a linier model between a input-output data. The basic structure of a
measureable noise and the corresponding fuzzy
unmeasureable interference. It can expand the inference system consists of three conceptual
concept of linier adaptive noise cancellation components [4,6,16]: A rule base, which
into nonlinear realm using nonlinear adaptive contains a selection of fuzzy rules; a database,
system[7,10]. In this experiment, it will show which defines the membership functions used
how ANFIS can be used to identify an in the fuzzy rules; and a reasoning mechanism,
unknown nonlinear passage dynamic that which performs the inference procedure upon
transforms a noise source into an interference the rules and the given facts to derive a
component in a detected signal. Information reasonable output or conclusion.
signal x(k) and a measureable noise source Adaptive noise cancellation was first
signal n(k), noise source goes through proposed by Widrow and glover in 1975[1].

86
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Filter out an interference component was

identified by a linier model between a X(k) y(k)=x(k)+d(k)
measureable noise and the corresponding
detected signal
unmeasureable interference. Adaptive noise Information signal
(measurable)
cancelation using linier filter has been
successfully in real-world application such as d(k)
n(k) f
interference canceling in electrocardiograms Distorted Noise
Noise Source
(ECGs), echo elimination on long-distance (measureable)
(not measureable)
telephone transmission lines, and antenna side-
lobe interference canceling[2]. Noise Figure 1. Schematic diagram of noise cancellation
cancellation is a variation of optimal filtering without ANFIS Filter
that is higly advantageos in many application.
The usual method of estimating a in symbols, the detected output signal is
signal corrupted by addition noise is to pass expressed as
the composite signal through a filter that tends
to suppress the noise while leaving the signal
relatively unchanged. The design of such y (k )  x(k )  d (k )
(1)
filters is the domain of optimal filtering, which  x(k )  f (n(k ), n(k  1), n(k  2),...)
originated the pioneering work of wiener and
was extended and enhanced by the work of The function f(.) represents the passage
Kalman, Bucy, and others. Filter used for the dynamics that the noise signal n(k) goes
foregoing purpose can be fixed or through. If f(.) were known exactly, it would
adaptive[8,9]. The design of fixed must be be easy to recover the origin information
based on prior knowledge of both yhe signal signal by subtracting d(k) from y(k) directly.
and the noise, but adaptive filters have the However, f(.) is usually unknown in advance
ability to adjust their own parameters and could be time varying due to changes in
automatically, and their design regquires little the environment. Moreover, the spectrum of
or no prior knowledge of signal or noise d(k) may overlap that of x(k) substantially,
characteristics [3,12]. invalidating the use of common frequency-
Its obviously that it can expand the domain filtering techniques.
concept of linier adaptive noise cancellation
into nonlinear realm using nonlinear adaptive
+ y(k)=x(k)+d(k)
system. In this section, it will show how X(k) +
ANFIS can be used to identify an unknown
nonlinear passage dynamic that transforms a
noise source into an interference component in
d(k)
a detected signal. Under certain conditions, the n(k) f
proposed approach is sometimes more suitable
than noise elimination techniques based on
+
frequency selective filtering. ANFIS
(k x(k)=x(k)+d(k)- (k))
Figure 1, show the schematic diagram -
of ideal situation to which adaptive noise
cancellation can be applied. Here it has an
unmeasureable information signal x(k) and a Figure 2. Schematic diagram of noise cancellation with
measureable noise source signals n(k). the ANFIS Filter.
noise source goes through unknown nonlinear
dynamics to generate a distorted noise d(k), To estimate the distorted noise signal d(k), it
which is then added to x(k) to form the need to pick up a clean version of noise signal
measureable output signal y(k). Our task is to n(k) that is independent of information signal.
retrieve the information signal x(k) from However, it cannot access the distorted noise
overall output signal y(k). which consist of signal d(k) directly since it is an additive
information signal x(k) plus d(k), a distorted component of the overall measureable signal
and delayed version of n(k). y(k). fortunately, as long as the information
signal x(k) is zero mean and not correlated
with noise signal n(k), it can use the detected

87
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

signal y(k) as desire output for ANFIS

training, as shown in figure (2). , such that the ANFIS function
in a least square sense.
The output of ANFIS be denoted by .
The learning rule of ANFIS tries to minimum That x(k) is the information which want to
the error, shows as recover, but it also serves as additive “noise”
in ANFIS training. To simplify the problem,
e( k )
2
 y (k )  dˆ (k )
2 assumed that
1. X(k) is a zero signal for all k
2 (
 x (k )  d ( k )  dˆ ( k ) 2. It has fixed premise parameters and update
2 consequent parameters of ANFIS using
 x (k )  d ( k )  fˆ (n( k ), n(k  1), n(k  1),..
Least Square method.
2)
Assume 1, implies that it can obtain perfect
ˆ training data that are subject only to measure
Where f is function implemented by ANFIS. noise. Assume 2, states that it is using ANFIS
Since x(k) is not correlated with n(k) or its with linear parameter only. Even with perfect
history, ANFIS has no clue how to minimize training data, ANFIS (with modifiable linear
the error component attribute to x. in the other parameter only) would produce a fitting error
word, the information signal x serves as an e(k) equal to difference between a desire
uncorrelated “noise” component in the data output and the ANFIS output. This error term
fitting processing, so ANFIS can do nothing is attributable to measure noise and or
about it except pick up its steady state trend. modeling errors.
Instead, the best that ANFIS can do is to
minimize the error component attributable to III. SIMULATION RESULTS
d(k). In the experiment, it applied ANFIS to two
nonlinear passage dynamics of order 2 and
That is, order 3, respectively. Unknown nonlinear
and passage dynamics were assumed to be defined
this happens to be the desired error measure, it as
is as if it coulds measure d(k) directly.
Equation (2) can expand to
2 2 2
e(k )  x(k )  d (k )  dˆ(k )  2x(k )d (k )  2x(k )dˆ(k ) (3) (7)

ˆ Where n(k) is a noise source and d(k) is

Because x(k ) is not correlated with d (k ) , it resultant from the nonlinear passages
become dyhnamis f(.) attribute to n(k) and n(k-1).
    
E e 2  E x 2  E (d  dˆ ) 2  2Exd   (4)
Figure (11) displays function f(.) as 3-
dimensional surface [13]. Since f(.) is
If x(k) is random signal with zero mean, then unknown, it use ANFIS to approximation this
ANFIS has no way to model it and function under assumption that it does know
approaches zero as n goes that f(.) is of order 2.

to infinity. This implies E[x )] = 0, and it has It assume that information signal x(k) is
expressed as
(5)
(7)
Where, is not affected when ANFIS is
adjusted to minimize . Therefore, Where k is a step count, and the sampling
training ANFIS to minimize the total error period is equal to 5 us. Figure 3 demonstrate
x(k) when k runs from 0 to 1000 (or when
is equivalent to minimize

88
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

time runs from 0 to 5 s). it assume that the

measureable noise is Gaussian with zero mean
and unity variance, as shown in figure 4. The
resulting distorted noise d(k) produced by
nonlinear dynamics in equation (7) is shown
figure 5. The measureable signal at receiving
end, denote as y(k) is equal to sum of x(k) and
d(k) and is demonstrated in figure 6. due to the
nonlinear passage dynamics of f(.) and the
large amplitude of d(k), it is hard to correlate
y(k) and x(k) in the time domain.

Figure 6. measured output signal y(k).

Figure 7 until figure 10 demonstrated the

spectral density distribution of x(k), n(k), d(k)
and y(k), respectively, for first 256 point.
Obviously, spectra of x(k) and d(k) overlap
each other considerably, making it impossible
to employ frequency domain filtering
techniques to remove d(k) and y(k).
To use ANFIS in this situation, it
collected 500 training data pairs of the
Figure 3. information signal x(k). following form:

(8)

With k runs from 1 to 500. It used a four rule

ANFIS to fit the training data, in which each
of the two input was assigned two generalized
bell membership function.

Figure 4. noise signal n(k).

Figure 7. power spectral density of x(k).

Figure 5. distorted noise signal n(k).

89
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 8. power spectral density of y(k). Figure 11.actual nonlinier passage dynamics f(.)

Figure 9. power spectral density of n(k). Figure 12. ANFIS function

Figure 10. power spectral density of d(k).

Figure 13. Training data dsitribution

Figure 12 is ANFIS surface (.) after 20 epoch
of batch learning. Figure 13 is scatter plot of
the training data. Figure 14 is RMSE curve
through 20 epoch. The starting point of RMSE
curve show the error when only the linear
parameter have been indentified by LSE. By
also updating the nonlinear parameter, it was
able to decrease the error further.

90
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 14. RMSE Curve Figure 17. MFs after training for n(k)

Figure 15-18 show MFs before and after

training, reflecting changes in premise
(nonlinear) parameter. Note that the error
cannot be minimize to zero, the minimum
error is regulated by information signal x(k),
which appears as fitting noise.

Figure 18. MFs after training for n(k-1).

By using ANFIS, the estimated resultant

from the nonlinear passage is expressed as
, as shown in
Figure 15. MFs before training for n(k) figure 19. thus, the estimated information
signal , derived as y(k)- , shown in
figure 20. the difference between x(k) and
,is shown in figure 21. Note that, is
already fairly close to x(k). estimated error in
figure 21 is expected to decrease if more
training data used over more training epochs.

Figure 16. MFs before training n(k-1).

91
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 19. ANFIS Output dˆ (k )

Figure 22. Original signal x(k ) .

IV. CONCLUSIONS
ANFIS has applied for Noise Cancellation,
although information signal added by signal
Noise. Information signal x(k) and a
measureable noise source signal n(k), noise
source goes through unknown nonlinear
dynamics to generate a distorted noise d(k),
which is then added to x(k) to form the
measureable y(k).
Information signal x(k) from overall
otput signal y(k), which consist of information
signal x(k) plus d(k), a distorted and delay
Figure 20. Estimated Signal xˆ(k ) n(k). The error cannot be minimize to zero, the
minimum error is regulated by information
signal x(k), which appears as fitting noise. The
result show that ANFIS did a good
performance to remove unknown distorted
noise signal d(k) from the measured signal
y(k). In the future works, ANFIS will be
applied to the actual external source signal.

REFERENCES
[1]. Anonim. 1995. Fuzzy Logic Toolbox User
Guide. The Mathworks, Inc. Apple Hill
Drive, Natick.
[2]. B.Widrow and J. R. Glover, Adaptive
Figure 21. Estimated error Signal xˆ(k ) - x(k ) Noise Cancelling: Principles and
Aplications. IEEE Proceeding, 63:1692-
1716, 1975.
[3]. B. Widrow and D. Strears, Adaptive signal
processing, Prentice hall, upper saddla
river NJ, 1985.
[4]. Fauset, L. 1994. Fudamental of Neural
Network,
[5]. Hanselman, D. Littlefield, B. 1996.
Mastering Matlab, A Comprehensive
Tutorial and 6 Reference. Prentice Hall,
Upper Saddle River, New Jersey.

92
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

[6]. George J. Klir and Bou Yuan, 1995, Fuzzy applications,” Proc. IEEE, vol. 63, no. 12,
set and Fuzy logic, Theory and pp. 1692–1716, 1975.
Applications, Prentice Hall PTR, upper [12]. Widrow, B., Stearns, S.D. 1985. Adaptive
saddle River, new Jersey 07458, USA. Signal Processing. Prentice Hall Inc,
[7]. J.S,R.Jang; C.T Sun;E mizutani,, 1997, Englewood Cliffs, New Jersey.
Neuro_fuzzy and Soft Computing, A [13].Heikki Koivo, 2000, ANFIS (Adaptive
computational Approach to Learning and Neuro-Fuzzy Inference System) By Heikki
Machine Intelligence,Page 523- Koivo, 2000
533.Prentice Hall, USA [14].L.R.D.Suresh, S.Sundaravadivelu, 2006,
[8]. S.S. Haykin, adaptive filter theory, prentice Engineering Letters, 13:3, EL_13_3_8
hall, upper saddle river, NJ, 2nd edition, (Advance online publication: 4 November
1991 2006), Real Time Adaptive Nonlinear
[9]. Ronaldy dkk, 2003, noise Reduction Noise cancellation using Fuzzy Logic for
Menggunakan Filter Adaptif Dengan Optical Wireless Communication System
Algoritma Recursive Least Square (RLS). with Multi-scattering ChannelMember.
Skripsi Jurusan Sistem Komputer, [15].Howard Demuth, Mark Beale , Martin
Universitas Bina Nusantara. Hagan , 1992, Neural Network Toolbox
[10].C P Kurian, S Kuriachan, J Bhat and R S User’s Guide, The MathWorks, Inc.,
Aithal , “ An Adaptive neuro-fuzzy model USAs
for the prediction and control of light in [16]. Robert Fuller, 1995, Neural Fuzzy
integrated lighting schemes”, Lighting Systems, ISBN 951-650-624-0, ISSN
Research & Technology, Volume 37, 0358-5654, Turku Center for Computer
Number 4, 2005, pp 343-352. Science, 1995
[11].Meng Joo Er, and Zhengrong Li,
“Adaptive noise canceling: Principles and

93
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

THE STABILITY OF THE MECD SCHEME FOR LARGE SYSTEM OF

ORDINARY DIFFERENTIAL EQUATIONS
Supriyono
Polytechnic Institute of Nuclear Technology
National Nuclear Energy Agency
Jl. Babarsari Kotak Pos 6101/YKBB Yogyakarta - Indonesia
Email : masprie_sttn@yahoo.com

Abstract difference (CD) scheme (the so-called

The Stability of The MECD Scheme for Large  scheme [4], choosing   1 ), which is
System of Ordinary Differential Equations. The
paper introduce a Modification of extrapolation for called the extrapolated central difference
solving large systems of second order ordinary (ECD) scheme in this paper. As an efficient
differential equations. Some basic problems on this method which can be used in such a case we
method including stability are analyzed from the presented in [6] a method, which is derived by
theoritical points of view. We considered the applying the extrapolation technique to central
application MECD (Modified Extrapolated Central difference solutions. We call the Modified
Difference) scheme mainly to linear systems extrapolated central difference (MECD).
derived from a finite element approximation. The The application of the technique of
MECD scheme is explicity expressed as a second
extrapolation to ODE’s has along history and a
order difference equations. The MECD scheme has
almost the more accuracy as the ECD (extrapolated lot of research papers which indicated its
central difference scheme), but The speed of efficiency has been published. We rever, for
computation in ECD is about twice of MECD. The example, to D.C.Joice survey paper [5] for
upper limit of the time mesh t to ensure that the details and its references. Nevertheless, as far
stability of the MECD scheme in the sense of as the authors know, it is not so widely utilized
in the actual computation of very large
energy is about 2 times that of CD (Central systems in science and engineering. This is
Difference). Therefore   2 is the best possible probably due to that these research do not
estimate for the ECD and   2 is the best necessarily take the size or special character of
possible estimate for the MECD. the systems into account. As a consequence,
the interest is limited mainly to the accuracy
Key Words and the discussion has been directed to the
Ordinary Differential Equations, Extrapolation, behavior of the extrapolation as the
Modification, Finite Element, Method, Stability extrapolation step size tends to zero.
The idea of the extrapolation is
Introduction simple. Consider a one step method to solve an
Solving initial value problems for initial value problems. Let t be the basic step
large systems of second order ordinary
differential equations is one of the most size and y be the approximate value obtained
important problems in scientific computation. by using the step size   t , t 2 , t 4 ,
In the stepwise integration of these problems, Assume that the error permit an expansion of
the discretization of the time derivatives has to
be carefully treated, since in the practical the form y  y1  a12  a24  
computation several hundreds or thousands We seek an approximate value y1 for y1 from
steps are necessary to obtain useful the equations
information and in such process, the
accumulation of truncation error may have yt  y1  a1t 2 ;
serious influence on the results. y t  y1  a1 ( 1 2 t ) 2

The purpose of the present paper in to 2

propose the use of an extrapolation technique Then since
to improve the stability of ECD and MECD
scheme analyzed from the theoritical points of
y1  1 4 y t 2  y t   y1  O(t 4 )
view. In particular, it is shown that an 3
extrapolation applied to the one step central

94
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

the error reduces to O(t ) . The value y1 is

4
introduce z  (i  1,2,3,) by z
i 0
 y' 0
the approximate value obtained in one
extrapolation. z i 1
 z i
 1 tA
2
y i 1
 y , i  1,2,3,
i

If we use the approximate value (5)

yt 4 and add the O() term, then we can get Theorem 1 : The sequence y determined by
i
a higher extrapolation with higher accuracy. (4) satisfies
After introduction, in section 2 we introduce a
one-step integration scheme equivalent to the y i 1
 y   t z i  1 t 2 A y
i 2 i
(6)
standart CD scheme, which is known i  1,2,3,
 scheme. In section 3 the simplest
extrapolation formula for the CD scheme is
y is given by
provided
1
derivied. Section 4 is devoted to give a
difference expression for the solution of the
y  y  t z  1 2 t A y
1 0 0
2
0
(7)
ECD scheme. In section 5 we give a sufficient Substitution of (6) into (5) leads to the
equation
 y
condition to ensure the stability of the MECD

2 tA 2I  2 t A
scheme in the sense of energy. 1 1 2
z i 1 i
Derivation of one-step scheme equivalent to
the central difference scheme.

 I1
2 t
2
A z i

Consider an initial value problem for Therefore the central difference equation (4) is
N ordinary differential equations written in matrix form as
My"Ky  f (t ) y 
 i 1   Q  y i  ;
 
(1)   t  
derived from a finite element approximation of  z i 1   zi 
the equation of motion to describe a linear  I  1 2 t A
2
tI 

Q t  1 tA 2I  1 2 A 
elastic vibration. Here M and K are the mass
and stiffness matrices, respectively. We

 2 2 t   I  1
2 t
2
A 
assume that M is diagonal and K is symetric
and positive definite. The right-hand of (1) is (8)
assumed to be zero in this paper, since this
term has no essential influence on the
provided y 1
is determined by (7). The
following discussion. computation proceeds as follows.
We also assume the inverse inequality
z 0
 y' (0)
[2]
y 
 I  1 t 2 A y  t z i 
 K y , y 
C  2
 2
y K
 0
h
y M


y K 
i 1
i  1,2,3,
i

(2)
Where h is a parameter to denote the size of
z i 1  z i  1 2 tA( y i 1  y i)
the finite element. In order to apply the Note that the computation of the type Ay , i
extrapolation technique we rewrite the central
difference scheme to a one step scheme known which includes the computation of the
1 stiffness matrik K, appear only one time at
as “  scheme”. Introduce A  M K to
write (1) as
each step if the vector Ay is stored..
i
y"  Ay Extrapolation of the central difference
(3) scheme.
The central difference approximation of this In order to derive an extrapolation
system is formula we have to know the asymptotic
expansion of the error. In our case, however,
y 2y  y the extrapolation formula is easily determined.
i 1 i i 1
 Ay i  1,2,3,
t
2 i

(4)

95
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

y  3  4 
   t "  1
y0 2 t y0  1 4 t y0
2 3
Assume that U 0   0  is given as an  z1 z0
 z0     t "  1 3  4 
y0 2 t y0  316 t y0
2 3
 z 1 z0
initial value of the CD scheme and let  5  6 
 1 t y  1  y
4 5
y  y 
  1  be the values  32 128 t
U 1   1  and
0 0

 z1 
U 1  z 1  Therefore hold
4  5 
  y ' (t ) 
12 t y 0
1  1 t y
3 4
after t with step-size t and t
2
,  z1 24 0

respectively, that is,

2

  O t

 
5

U Q V ;U  Q t V
4  5 
48 t y 96 t y
1 t 0 0  z 1  y ' (t )  1 3
1
4
1 2 0 0

(9)
 I1 
 O t  
5

2 t
2
 A 
1

3
 tI 
8 t A  (11)
  32 t A
1 4 2

 
2
Qt 2   t  3 I  1 t A 
y 
  1  1 4
16 t A U
3 2 2

3 U1 U1
if we set
 2  
 z1 
1

 1 2 
 1
 t and
4
 t
5
A
5
 t A  3
 128 32  then by (10) and (11) the terms of O
O t  4
vanish, and we have
Taking the relations z 0  y' (0) and  y t  
y"  Ay into account we have, from equation U 1
 
y ' t 
  O t
5
 
 
(9),
y  y  1 2 t A y  t z
2
0
The vector U 1
is the next approximate value
1 0 0 in our extrapolation. The integration scheme
 y  ty ' (0)  1 t y" (0)
2
0 2
consisting of the process 0 U
y  y  2 t A y  132 t A y
1  U Q U
2 4 2
1 0
;
1 0 0 0 t
2
 t z  1 t A z Q 
3
8 0 0 U 1 t U 0 U 1
 
' " 3 is called the extrapolated central difference
 y  t y  1 t y  1 t y
2 3
0 2 8 0 0 0 (ECD) scheme in this paper.
 
,4
 1 t y
4
32 0 Finite difference expression of the
extrapolated central difference scheme.
y  determined
k

where y  d
  y ( 0)
k
. k The vector by the
dt 0

above process do not satisfy the central

i

Therefore hold difference equation (4). In this section we seek

3  4 
  y ( t )   1  t y  1  t y
y
3 4
a difference equation which governs the
 1

  O ( t )
5
6 0 24 0
extrapolated solution. Let U i 1
 y i 1
, zi 1 
 be the extrapolated value determined by the
U  y , z  :
3  4 
 y1  y (t )   1 24 t y 0  1 96 t y 0
3 4
starting value
 i i i
 O (t )
5

U Q U
i 1
;
t i
2
(10)
Similarly we have
U Q U
i 1
; t
2
i

U  13 4U  U .
i 1 i 1 i 1

96
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Each component is determined as follows.

 p 0   y 0  1 2 t Ay 0  tz 0 
2

 I  
V0     
2 t
1 2
A tI
   q 0   z 0  1 2 tAy 0  1 2 tAp 0 
 
6 t
 1 3
 y    1 t A A  
4 2
 y i   y0  1 Ay 0  1 tz 0 

2
 i 1   24
   tA  1  
 p   8 t 2
6 t A I  2 t A  z i  1  q1   z 0  1 tAy 0  1 tAp1 
3 2
1 2
  1
 
 z i 1   V
 1 2  4 4 
 1

5 4
 t A
3
 t A
 96 24  p   p1  18 t Ap1  1 2 tq1 
2

V2   2   
(12)  q 2   q1  1 4 tAp1  1 4 tAp 2 
The extrapolated central difference scheme is
expressed in the following form.
Then U 1
is determined by

Theorem 2 : The vector y determined by

i
 y   1 4 p 2  p 0 
U 1   1    3
(12) satisfy the following difference equations.  z1   13 4q 2  q 0  
y i 1
2y 
i
y i 1 (15)
The vectors to be calculated in this step are
t
2

Ay0 , Ap0 , Ap1 , Ap2 . Here we note that the


 A  1 t A
12
2 2
 y  12 1 24 t
i
4
Ay
3
i 1
vectors p 0 , p1 and p 2

(13) Stability of the extrapolated central

difference scheme.
1 It is well known that the central
Since A  
M K we have difference scheme (4) is stable only under a
yi 12 yi  yi 1 condition
t
h C1

M
t
2
for a certain constan C which depends on
 K I  1 t M y
1
2 1
12
K
i
the constant C 0
in the inverse inequality and


1
12  24 t
4
K M 1K  y 2

i 1
0
on elastic constants [3]. Is this condition
sufficient or relaxed for the extrapolated
scheme ? To discuss this problem we
(14) introduce a matrix D defined by
This is the difference expression of the
D  1 t M K and write (14) as
2 1
extrapolated central difference scheme. Note 12
that a little decreasing of stiffness and a very follows
little dumping are caused by the extrapolation. :
See also (15). Equation (14) is called the
yi 12 yi  yi 1 
 K I  D D y
extrapolated central difference equation. 2
M
t
2 i
Modified Extrapolation of The Central

 KD  y  y   0
Difference Scheme 1 2
The developing of the ECD method i 1
process as follows. Let U 0  y , z 0 and   2 i

y , z be the initial vector and the next

0

U 1
 1
(16)
1
We first derive an energy inequality for this
vector to be obtained in one step of the
equation. The inner product of the both sides
extrapolation, respectively. In this
extrapolation process three intermediate of (16) and y i 1
 y i 1
yields
vectors 0
, V
1
and 2V V
are introduced as
follows :

97
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

2 2
y i 1 y i y n 1 y n

 K I D 1
2D
2
 y i, y  
 K I D 1
2D
2
 y n, y 
t M
i 1
t M
n 1

1 KD
2
2
 y  y , y i i 1 i 1
 y i 1
  1 K D  y
4
2
n 1
 y , y
n n 1
 y n

K  , y 
2 n
y y 4 D y
  1 y y
2
 
 i i 1  K I D 1 
2 D y i 1, y
2
i 1
i 1 i 1 i 1 l 1

t i

y y
2

  y 0, y 
M
Summing this equation for i  1,2,3,  , n we  1 0  K I D 1
2D
2

have t M
1

 y  y , y  y 
2
y n 1 y n
  1 KD
2

 K I D 1 
2 D y n, y
2 4 1 0 1 0

t M
n 1 (18)
Using the identity

1 K   , y
n

2 D y y
2
i i 1 i 1
 y i 1
 ab  1    1 2 a b , the above
2a b
2 2 2

i 1
equation can be written as follows.
2

y1 y 0
2
y n 1 y n
 
 K I D 1
2D
2
 y 0, y  t
t 1
M
M

1

K I D 1
 2 D
2
y n, y n   

(17)
To estimate the third term in the left-hand side
2

 K I  D  2 D y n 1 , y n 1 
1
2   
we set ei  y  y and write 
 1  K I  D  y n 1 y n , y  y n  
 
i i 1
2 n 1
e i ,e j  K D 2 e i ,e j ;
K  y , y y
n

4 D y
1
2
 
2 i 1 i i 1 l

ei  e i ,e j i 1
2
y y
Since  1 0
n n 2 t
 e i , e i  1 e i   ei
K I  D  12 D  y 0, y  K I  D  12 D  y , y
M

e 
i 1 i 1 1
2 2
2 0 1
n 2 2 2
 1  i 1 e i  e i 1  e i
2
i 1  1  K I  D 
2 y1 y 0, y1 y  0

n 2 (19)
  12 e i 1
2
1 
2 ei 1 ei  12 e i
2
Therefore, the extrapolated central difference
i 1 scheme is stable if the following conditions are
satisfied
Substituting this into (17), we have Condition (1) : 
K I D 1
2D
2
 is
positive definite.

98
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Condition (2) : Stability of the Modified Extrapolated

2 Central Difference Scheme
z As is seen (22) the stability limit of MECD
1
2
K I  D z, z  is positive for scheme from (15) we rewrite as follows.
t
1 12
2
h
4 4
M
 0 ;
any N-dimensional vector z.
5 1
Theorem 3 : Let C
be the constant appearing  
C 0C1
0
4
in the inverse inequality (2) and set 144

  C0
t Be solving this inequality we have
then the following holds :
h 4

2

1. The central difference scheme is stable in 4
1  1  16 h
the sense of energy if  2
(24)
(20)
2. The conditions (20) is sufficient for Therefore, roughly speaking   2
(corresponding to 1  16h  0 ) is the
4
Condition (1) and Condition (2).
The conditions maximum value of  for the MECD scheme,
2 althought the above analysis gives only a
 C 0 t    sufficient condition of the stability. Comparing
  
2
 12 3  1 this result with (20) we can expert that the
 h  time mesh t can be taken by about 2 times
(21) larger than that of CD.
is sufficient for Condition (1) and the
condition References
4
 t  1. Adam.A.R., “Sobolev Spaces”, Academic
1 12 24  
2
Press, 1975.
1 0
 2. Ciarlet.P.G., “Finite Element Methods for
 C1  Elliptic Problems”, North-Holland
(22) Publishing Company, 1978.
is sufficient for Condition (2). 3. Fujii., “Finite Element Galerkin Method for
As is seen in (22) the stability limit is Mixed Initial-Boundary Value Problems in
relaxed by the extrapolation. To explain this Elasticity Theory”, Center for Numerical
fact, we rewrite (21) as follows. Analysis The University of Texas at
Austin, 1971.
1 12
2
h
4 4
 0 ; 4. Hsu.T.R, “The Finite Element Method in
1 1 Thermomechanics”, Allen and Unwin, Inc,
 
C 0C1 4 1986.
24 5. Joice.D.S, “Survey of Extrapolation
Be solving this inequality we have Process in Numerical Analysis”, SIAM
4 Review. Vol.13, No.4 (1971), 435-490.

2
 6. Supriyono and Miyoshi.T, “A Modified
4
1  1  16 h Extrapolation Method for Large Systems of
Second Order Central Difference
(23)
Equations”, Japan Journal of Industrial and
Applied Mathematics, Vol.12, No.3,
October 1995, Kinokuniya Tokyo Japan,
439-455.

99
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Real Time Performance of Fuzzy PI+Fuzzy PD Self Tuning Regulator In

Cascade Control
Mahardhika Pratama (1) , Syamsul Rajab ( 2 ) ,Imam Arifin (3) , Moch.Rameli ( 4 )
(1, 2,3, 4)
Electrical Enginering Department,. Faculty of Engineering,. Sepuluh Nopember Institute of Technology,.
Surabaya,. Indonesia
Email : ta2ngzzhh@gmail.com

Abstract complexness to determine rule and

This paper discussed real time performance membership function. In order to deal with
evaluation of fuzzy PI+fuzzy PD self tuning that problem fuzzy PI + fuzzy PD can be used,
regulator for controlling flow and pressure of because we can decrease the complexness,
pressure rig 38-714 in the cascade structure. The however we can get the characteristic of Fuzzy
results were compared with conventional PI
PID controller.
controller. We used fuzzy PI+fuzzy PD self tuning
regulator for master controller and P controller for Reference [2] introduces a kind of fuzzy
slave controller. Experimental evaluations were controller with on-line self-regulation
conducted with several load configuration function. The controller is capable of
including minimal load, nominal load, and automatically adjusting parameters according
maximal load. From overall those results could be to the controlled system error, and has the
concluded that improving controller was much features of strong adaptability. While main
faster, and robust responces than conventional PI controller solve control task, other controller
controller. give self regulation online.
This paper is organized as a follows.
Keyword Section 2 discussed about cascade control and
cascade control, fuzzy PI+PD, self tuning regulator. fuzzy PI + fuzzy PD self tuning regulator.
Section 3 discussed about system design
I. Introduction including controller design, and experimental
Cascade control is one of the most popular set up. In the Section 4, we spoke about
complex control structures that can be found in experimental result and discussed the results
the process industries, implemented in order to taken from experiment. In the end of this
improve the disturbance rejection properties of paper, we gave some conclusions used for next
the controlled system [1]. On the other hand, research.
Controllability of outside loop is improved
because inside controller speeds up response. II.A Cascade control
Therefore, Non linearity of process are Cascade control is one of complex control
handled by that loop and removed from the system structure that are often used. Using
more important outer loop. cascade structure will get many advantages.
Using PID controllers are very difficult in One of those is disturbance effecting inner loop
the complex system especially for non linear can be detected more fastly than normal
systems, PID controller can’t be directly structure, because inner controller compensate
applied. Fuzzy set theory invented with Zadeh it, before it can influence outer loop. Figure 1 is
eliminates those problems. With very simple showing cascade contol architecture.
approached, it can overcome the problems in
the complex process. Even without knowing
dynamical model of the system, it can be used.
Heuristic approach, making suitable with
human feeling, improves boolean on off
principle. The problem on fuzzy logic is
determining fuzzy parameter that just based on
feeling of designer. Therefore, Combination Figure 1. Block diagram of cascade control.
between PID and fuzzy can be proposed to
make pattern on resulted response. C1, C2 are outer, and inner controler. P2,P1
Fuzzy PID appears into use on recently are inner, and outer process. Disturbance can
research. The weakness of this controller is come from inner or outer loop. The most
needing more input (3 input), that can cause important, remembered in cascade structure, is

100
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

process inner loop must be much faster than Table 1. Rule base of Fuzzy PI + Fuzzy PD
outer loop. E NB NM NS Z PS PM PB
de
NB NB NB NB NB NM NS Z
II.B Fuzzy PI+ Fuzzy PD
NM NB NB NB NM NS Z PS
Fuzzy PI+ Fuzzy PD is as same as Fuzzy
NS NB NB NM NS Z PS PM
PID, however the main reason for avoiding
Z NB NM NS Z PS PM PB
fuzzy PID is more input needed than fuzzy PI+
fuzzy PD, therefore we have more complexity PS NM NS Z PS PM PB PB
for determining rule base and membership PM NS Z PS PM PB PB PB
function than fuzzy PI+ fuzzy PD[3]. PB Z PS PM PB PB PB PB
A classical PI controller is described as :
1 II.C Self Tuning Regulator
u pi  K c (e(t ) 
i  e(t )dt (1)
The major weaknest of fuzzy controller is
With differential operator, we can get from in the steady state error. Mainly, it can not
Equation 1 : eliminate static error and the static error
du pi (t ) variation rate. Because of that, we overcome
de(t ) e(t )
 Kc (  ) that problem based on actual output of plant.
dt dt i We determine rule of self tuning regulator
u pi (k )  u pi (k  1) e(k )  e(k  1) e(k ) from plant output as can describe on graphic in
 Kc (  )
Ts Ts i Figure 2.
Kc
u pi  K c e  e(k )
i
u pi (k )  u pi (k  1)
u pi 
Ts
Ts u pi  u(k ) pi  u pi (k 1)
Ts u pi  u pi (k )  z 1u pi (k )
K upi
u pi (k )  u pi (2)
1  z 1 Figure 2. Graphic of error.
In the equation 2 is mathematical model for a
fuzzy PI. A classical PD controller is given as In the point a, error is negative big and
follows : de
de(t )
raise steady state if >0. therefore we must
u pd  K c (e(t )   d ) (3) dt
dt increase a number of u . In the point b, error is
In the discrete form, equation 3 can be changed de
as follows : negative smaller than in the pont a, and >0,
dt
e( k )  e( k  1) so we must decrease a number of u to avoid
u pd ( k )  K c (e( k )   d )
Ts excessive overshoot. In the point c, error is
u pd (k )  K c e(k )   d K c e(k ) (4) de
possitive big and < 0, in order to get steady
Equation 4 is showing a fuzzy PD, fuzzy PI+ dt
fuzzy PD is the combination of fuzzy PI and state, u must be increased. In the point d, error
fuzzy PD. de
is positive small and < 0 to approach
dt
u(k )  u pi (k )  u pd (k ) (5) steady state value u should be decreased. All
rule base of self tuning regulators are given in
Table 2.
Table 1 is showing the rule of fuzzy PI+fuzzy
We use five membership functions and
PD controller. We used triangular shape for 7
triangular shapes. Figure 3 is exhibiting fuzzy
membership function in each input or output.
PI+ fuzzy PD self tuning regulator structure.

Table 2. Rule base of self tuning regulator.

101
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

e NB NS Z PS PB
de
NB PB PB PB Z Z
NS PB PB PS Z NS
Z PB PS Z NS NB
PS PS Z NS NB NB
PB Z NS NB NB NB

We use five membership functions and

triangular shapes. Figure 3 is exhibiting fuzzy
PI+ fuzzy PD self tuning regulator structure.

III. System Design Figure 4. Pressure rig 38-714.

In the experiment, we built cascade control
configuration. In the inner loop, we controlled IV. Experimental Results
flow. In the outer loop, we controlled pressure. Experiments are conducted to evaluate
Figure 4 exhibits picture of pressure rig 38-714. fuzzy PI+ fuzzy PD self tuning regulator. For
comparison, we used pure PI controller. PCI
1711 from advantech was used for converting
analog to digital data and digital to analog data.
Software used in these experiments was
Labview 7.1. Figure 5 is showing experimental
set up for experiments.
In the inner loop, P signal from P controller
changes from digital to analog to control flow,
flow output detected with flow sensor converts
from analog to digital data. Process in the outer
loop are that control signal generated from
fuzzy PI+PD self tuning regulator changes
Figure 3. Fuzzy PI+Fuzzy PD self tuning regulator from analog to digital to control pressure,
structure. pressure sensor was detected pressure output
converted from analog to digital data.
PI parameters was obtained with analytical
Table 3. Controller parameters. method (Kp=0.56, Ti=32,5). Input from these
Kp Ki Kd experiments are 1,1 volt. Figure 6, and 7 are
Inner Loop 4,11 0,42 10,15
Outer Loop 0,25 - -
showing plant inner and outer loop responces
with fuzzy PI+PD self tuning regulator as outer
loop controller.
For designing fuzzy PI + fuzzy PD self
tuning regulator in the outer loop and
constructing P controller in the inner loop, we
needed to identificate plant parameter with
open loop configuration. After that, we used
rule 1 of ziegler nichols law. Table 3 is
showing controller parameters in the inner and
outer loop.

Figure 6. Experimental set up.

From Figure 7 and 8, we could conclude

that proposed controller improved plant
responces in the outer and inner loop.
Responces with proposed controller was much
faster than pure PI controller to raise steady
state output. Even, Responce with PI controller
didn’t attain steady state output on the 1000

102
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

iteration Table 4. given output parameters of

outer loop.
Table 5. Output parameters of inner loop
Fuzzy PI+PD PI
self tuning
regulator
Steady state 1,548 -
output
Settling time 57 iteration -
(5%)
Overshoot 2% -

Conclussions
From these experiments, fuzzy PI+ fuzzy
PD self tuning regulator could improve
performance of pressure rig 38-714, because it
can make responce become much faster than
Figure 7. Outer loop responce. using pure PI controller. Using fuzzy PI+ fuzzy
PD reduced complexity of fuzzy PID in the
obtaining rule and membership function. In
addition to control structures, cascade control
proved faster than normal structure in the
rejecting disturbance.

References
[1] Arrieta,O, Vilanova.R, Belaguer,P.(2008). “
Procedure for cascade control Systems design :
Choise of Suitable PID Tunings”. Int. J. of
Computers, Communications & Control. ISSN.
Vol. III (2008), No. 3, pp. 235-248.
[2] Qian,Z, Xi-jin, G, Zhen,W, Xi-lan, T,
Figure 8. Inner loop responce. (2006) “. Application of Improved Fuzzy
Controller in Networked Control Systems ”. J.
China Univ. of Mining &Tech. (English
Edition). Vol.16, No.4.
Table 4. Output Parameters of outer loop
Fuzzy PI+PD PI
[3] Kumar,V, Rana,R,N,S, Gupta, V. (2008). “
self tuning Real Time Performance Evaluation of a Fuzzy
regulator PI + Fuzzy PD Controller for Liquid Level
Steady state 1,1 - Process. INTERNATIONAL JOURNAL OF
output INTELLIGENT CONTROL AND SYSTEMS.
Settling time 20 iteration -
(5%)
VOL. 13. NO. 2. 89-96.
Overshoot 10% -

103
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

APPROXIMATION OF RUIN PROBABILITY FOR INVESTMENT

WITH INDEPENDENT AND IDENTICALLY DISTRIBUTED RANDOM
NET RETURNS AND MULTIVARIATE NORMAL MEAN VARIANCE
MIXTURE DISTRIBUTED FORCES OF INTEREST IN FIXED PERIOD
Ryan Kurniawan(1) and Ukur Arianto Sembiring(2)
(1)
Department of Mathematics, Universitas Pelita Harapan, Karawaci, Banten, Indonesia (Email:
canis_boyz@yahoo.co.id)
(2)
Department of Mathematics, Universitas Pelita Harapan, Karawaci, Banten, Indonesia (Email:
ukur_sembiring@uph.edu)

Abstract discount formula, can be used under certain

An investor can gain negative investment net conditions, and its use has been discussed for
returns that are obtained from several periods. The Pareto distributed and lognormal-invers
negative returns will force the investor to save Gaussian (LNIG) distributed . In this paper,
some reserve fund for liability payment. We can
the authors will discuss the use of the formula
determine the reserve fund needed to anticipate it
with tail probability distribution from the sum of for Burr and inverse Burr distributed , and
discounts of random investment net returns under for the , which have the distribution family
random forces of interest. The problem is of log multivariate normal mean-variance
referenced from journal [1]. In [1], the forces of mixture (LMNMVM). The authors chose the
interest are Normal Inverse Gaussian (NIG) stated distributions because of their large roles
random variables, in which NIG is a member of in financial studies.
multivariate normal mean-variance mixture
(MNMVM) distribution class, while investment Certain violations of the conditions
yields are independent Pareto random variables
required for the discount formulas will also be
which have regularly varying property for their
distribution tails. In this research, instead of Pareto
discussed. In this case, the authors chose the
distribution, we will use other distributions which logmultivariate t distribution for . It will be
also have regularly varying distribution tail (such shown that under such distribution, the
as Burr distribution) for modeling investment net discount formula cannot be used.
returns, and MNMVM distribution class for
modeling forces of interest. II.THEORY AND DEFINITIONS
II.1. Discount Formula
As was explained, the main focus in
Key Words this paper is the approximation of the
Investment yield; force of interest; normal mean- probability in Eq. 1. The discount formula was
variance mixture distribution class; slowly varying
found in [1] under certain conditions for the
function; ruin probability.
random variables and . For the random
variables , i.i.d. property has to be satisfied.
I.INTRODUCTION
This paper is motivated by [1] which Moreover, their distribution tail, , has to be
focuses on the approximation of the tail regularly varying.
probability of the sum of weighted random
variables of the form Definition 2.1. A function is
regularly varying iff there exists some
Pr (1) such that

where the weights, , are also random (2)

variables. The probability in Eq. 1 is the where is a function such that
ruin probability of a company during for every
periods from now, given the initial capital ,
where the random variables represent the
investment net returns, and represent the (3)
discount factors. In [1], the approximation
formula, which will now be called, the

104
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

If is regularly varying for some Definition 2.3. The random vector in

, denote . Definition 2.1. is multivariate t-distributed,
denoted by iff , where
Below is the theorem from [1] which is a parameter vector in , and
contains the discount formula.
, is an inverse Gamma
Theorem 2.1. Let be i.i.d. random distributed random variable with density
variables with distribution for some
and be random variables, not . (6)
necessarily i.i.d. Then,

Pr (4) Definition 2.4. The random vector in

Definition 2.1. is a generalized hyperbolic
if there exists a such that random variable, denoted by
for every . “~” denotes an asymptotic relation iff ,
for , meaning that the approximation will be where and are parameter vectors in ,
more accurate as tends to infinity. and , is a generalized inverse
Gaussian(GIG) distributed random variable
II.2. MNMVM Distribution Class with density

Definition 2.2. A random vector is

MNMVM-distributed, denoted by
iff

(5) , (7)

where where is a modified Bessel function of

1. , where is a the third kind with index
multivariate normal random variable,
and is an identity covariance III. MAIN RESULTS
matrix, both of which are elements of III.1. Theoritical Results
; For investment problem, in Eq. 1
2. is a scalar nonnegative random is of the form
variable and is independent from
3. is a matrix, and ; (8)
4. is a measurable
function. for every , where is a random force of
interest in period .
From the above definition, the Now, let for
MNMVM distribution class has the following every , where ,
properties. , and is a
nonnegative scalar random variable. Let
Fact 2.1. .
be i.i.d. with distribution for
Fact 2.2. If , then some . Then, by Theorem 2.1 and
, where Definition 2.2,

and are parameter matrix and

vector, respectively. Pr (9)

Below are the definition of the two under the assumption that the conditions in
members of MNMVM distribution class which Theorem 2.1 are satisfied. Moreover, by Fact
will be discussed. 2.2, , where

105
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

and . case, is t -distributed. Note

Hence, by Fact 2.1, that for every ,
 
(10)

(11) , (16)

Through Eq. 9-11, the following where , doesn’t exist. Hence,

proposition is obtained. the discount formula cannot be used in this
case.
Proposition 3.1. Let be i.i.d. random
variables with distribution for some III.2. Numerical Results
and be MNMVM To look at how well the the
random variables, with is of the form given approximation using the discount formula
by Eq. 8 for every . If there exists a works, the authors simulate the probability in
Eq. 1 by generating 5,000,000 data, each of
such that exists which is the sum in Eq. 1 where is of the
for every , then
form given by Eq. 8 in the case where
is GH -distributed with
Pr (12
, , ,
)
, , and , and
In the case where is
GH -distributed with
and
,

(13)
(17)

where and is a moment The numerical results are obtained in

generating function of , with two cases. In the first case, are Burr-
and . distributed with cdf
Because exists for every , the
condition of Proposition 3.1 is satisfied so (18)
that
is a member of . In the second case,
Pr
are inverse Burr-distributed with cdf

(19)
(14)

where is a member of .
The numerical results are absolute
(15) relative errors of the discount formula
approximations to the simulation results,
Now, a violation will of the condition .
of Proposition 3.1 will be discussed. In this

106
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 1. Absolute relative error plot for Burr-distributed Figure 2. Absolute relative error plot for inverse Burr-
with distributed with

It can be shown from Fig. 1-2 that

the error decreases as tends to infinity.
This is consistent with the asymptotic
relation used in the discount formula. Also, (22)
the error is relatively small. From Fig. 1,
the maximum error is only 0.18, while from satisfies Eq. 3. Again, let be a positive
Fig. 2, it is 0.035. These show how well the real number, then,
formula works.

III.3. Regularly Varying Property of (23)

Burr and Inverse Burr Distribution Tail
Note that
It will be shown that Burr
distribution with cdf (18) is a member of
, or equivalently, the function
(24)
(
and
20)

satisfies Eq. 3. Let be a positive real

number, then,
.
(25)

Hence, through Eq. 23-25, satisfies Eq. 3.

. Thus, inverse Burr distribution is a member
(21) of .

Hence, Burr distribution is a member of IV. CONCLUSION

. The discount formula
approximation has been tested for GH-
Now, it will be shown that distributed forces of interest and Burr and
inverse Burr distribution with cdf (19) is a inverse Burr distributed net returns. The
member of , or equivalently, the resulting approximation as has been
function observed in the simulation result is quite
good. It is consistent with the asymptotic
property of the approximation and the
errors are small. However, not all of the
member of MNMVM distribution class can

107
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

be used as the distribution of the forces of [2] Kalemanova, Anna and Werner, Ralf
interest to apply the formula. As has been (2006). “A short note on the efficient
seen, the multivariate t distribution violates implementation of the Normal Inverse
the condition used in Proposition 3.1. Gaussian distribution,” Scandinavian
Other distributions beside Burr and inverse Actuarial Journal.
Burr can also be used as the distribution of
the net returns as long as they have [3] Klugman, Stuart A., Panjer, Harry H.,
regularly varying property for their and Wilmot, Gordon E. (2008). Loss
distribution tail. However, the i.i.d. Models, New Jersey, John Wiley &
property of the net returns must also be Sons, Inc.
taken into account as it is also one of the
conditions required to use the formula. [4] Luciano, Elisa and Semeraro, Patrizia.
(2009). A Generalized Normal Mean-
REFERENCES Variance Mixture for Return Processes
[1] Goovaerts, Marc J., et. al.(2005). “The in Finance, Collegio Carlo Alberto.
tail probability of discounted sums of
Pareto-like losses in insurance,” [5] McNeil, Alexander J., Frey, R diger,
Scandinavian Actuarial Journal, Taylor and Embrechts, Paul (2005).
and Francis, pp 446-461. Quantitative Risk Management, New
Jersey, Princeton University Press.

108
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Stochastic History Matching for Composite Reservoir

Sutawanir Darwis1, Agus Yodi Gunawan2, Sri Wahyuningsih1,3, Nurtiti Sunusi1,4, Aceng Komarudin
Mutaqin1,5, Nina Fitriyani1
1
Statistics Research Division, Faculty of Mathematics and Natural Sciences, Institut Teknologi Bandung,
Indonesia, 2Industrial and Financial Mathematics Research Division, Faculty of Mathematics and Natural
Sciences, Institut Teknologi Bandung, Indonesia , 3Statistics Study Program, Faculty of Mathematics and
Natural Sciences, Universitas Mulawarman, Indonesia, 4Mathematics Department, Faculty of Mathematics and
Natural Sciences, Universitas Haluoleo, Indonesia, 5Statistics Department, Faculty of Mathematics and Natural
Sciences, Universitas Islam Bandung, Indonesia
sdarwis@math.itb.ac.id,

ABSTRACT I. INTRODUCTION
The act of adjusting a model of a reservoir until it During history match, reservoir models are
closely reproduces the past behavior of historical calibrated to improve forecast reliability. In
production is termed history matching. The most history matching applications, the
accuracy of the matching depends on the quality of parameters to be estimated consist only the
the reservoir model and the quality of production
static parameters. In EnKF, the static and
data. Once a model has been history matched, it
can be used to simulate the future reservoir dynamic parameters are updated
behavior with a high degree of confidence. simultaneously which introduces
Stochastic history matching is design to estimate inconsistencies between the updating of static
reservoir properties such as permeability which are and dynamic (Gu, and Oliver, 2004). The
needed for forecasting reservoir production. The outcomes of history matching motivates the
history matching can be formulate as a state development of stochastic framework. The
estimation problem where the parameters of EnKF has gained popularity as stochastic
interest are added to the state and estimated with based methodology for history matching.
other states. A composite reservoir consist of two Naevdal et al (2002) and Naevdal et al. (2007)
regions with different properties (permeability,
discussed the application of EnKF on reservoir
porosity, viscosity, compressibility) and used as a
model for fluid flow. A thin discontinuity is models. Modeling reservoir is a complicated
assumed to separate the two regions. Oil fluid is task due to the fact that the information of
flowing through a porous medium. The diffusivity reservoir properties is limited; parameters such
equation is considered as an important expression as permeability and porosity is very uncertain.
in petroleum engineering. The fluid flow Therefore, the model must be improved from
modeling for composite reservoir is a nonlinear the available measurements by updating the
estimation problem. The ensemble Kalman filter model parameters. A composite reservoir is
(EnKF) was introduced as a way to extend the a system two concentric regions with a single
Kalman filter (KF) to nonlinear problem. This well at the center (Demski, 1987). The two
paper combines a sequential EnKF history
regions have different properties. Fluid flow
matching with composite reservoir to provide an
estimation of reservoir properties. The diffusion through regions is represented by diffusivity
flow model, analytical solution, nonlinear state equations. The solution pressure as a function
space model, estimation and prediction will be of permeability yields a nonlinear model The
discussed. Two case studies are discussed to linearity of state model is an important
explain the methodology: EnKF infinite reservoir assumption of KF. It is frequently that the
history matching and composite reservoir history models are nonlinear such as the pressure as
matching. The case studies showed that the function of permeability in a composite
proposed method can be considered as an reservoir. The EnKF was developed by
alternative of the previous known history matching Evensen (2007) for nonlinear models. The
methods
method uses Monte Carlo sampling to estimate
the posterior distribution. EnKF can be used
Keywords
composite reservoir; EnKF history matching to integrate observations by updating
sequentially a sample of reservoir modes
during the simulation; each reservoir model is
kept up to date as observations are assimilated.

109
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

The mean and covariance of the prior are assume t noisy observations
estimated using Monte Carlo samples, then Y   Y1 , , Yt  ; Yi | X  x ~ N  x,  2 
t

used KF update formula to obtain the posterior

distribution. . The history matching can be   n 1 1  t  y    n 1  
1

X | y ~ N   2  2     2i  2   ,  2  2  
formulate as a state estimation problem where       i 1          
 
the parameters of interest are added to the state The posterior mean can be written as
and estimated with other states. Although the  n2 
2 2  ny  
static parameters (permeability, porosity) are E X | Y  y  2  2
 2    2 2 
y  
  n    
2
   n 
not varying in time, the estimate of them are
   K  y  
changing with the observations available
(Chen, Oliver, and Zhang, 2009). This paper The prior is adjusted according to gain
 n 
2
combines a sequential EnKF history
matching with composite reservoir to provide K . The prior variance is
an estimation of reservoir properties. The  2
 n 2 
diffusion flow model, analytical solution, updated according to
nonlinear state space model, estimation and Var  X | Y  y   1  K  2 . Vector.
prediction will be discussed. Two case
studies are discussed to explain the Assume that the state X  p 1 has normal
methodology: EnKF infinite reservoir history
prior X ~ N  μ, P  , and the observations
matching and composite reservoir history
matching. The case studies showed that the follows the model Y | X = x ~ N  Hx, R  ,
proposed method can be considered as an
alternative of the previous known history where H  t  p  is the observation matrix.
matching methods The posterior distribution is given by

II. METHODOLOGY

X | Y = y ~ N  H t R -1 H + P -1 
1
H R
t -1
y + P -1μ  ,  H t R -1 H + P -1 
1

The methodology starts with the sequential  N  μ + K  y - Hμ  ,  I - KH  P 
estimation based on Bayes inference. The
composite reservoir model is developed using Posterior distribution known as analysis step,
the basic homogeneous reservoir model, μ is the forecast based on previously collected
followed by the EnKF formulation for history data, and P is the forecast error covariance.
matching problems The prior μ is updated based on forecast error
Let X denote the unobservable (state)
weighted by the gain K
quantities and Y is the observations. The
Sequential estimation. Define the following
probability model can be factored into prior,
notations:
data distribution, marginal and posterior
Y1:t   Y1 , , Yt  , X 0:t   X 0 , , X t  .
f  x | y   f  x  f  y | x  / f  y  , where
The analysis is
f  x  is the prior quantifies a prior
then f  x 0:t | y1:t   f  x 0:t   f  y1:t | x 0:t 
understanding of the unobservable based on t
historical information or from a forecast where f  x0:t   f  x0   f  x j | x j1 
model, f  y | x  is the sampling distribution j1

of the observation given the unobservable represent the prior of the state, f  x 0  is the
quantifies the measurement errors, f  y  is initial state. The observations are
t
the marginal distribution known as prior
independent f  y1:t | x0:t    f  y j | x j  .
predictive distribution, f  x | y  is the j1
posterior known as the update of prior given The analysis becomes sequential updating
the observation Y. (estimate)
Scalar. Let the prior of the univariate X be t

 
X ~ N ,  2 . Conditioned on X = x, f  x0:t | y1:t   f  x0   f  x j | x j1   f  y j | x j 
j1

. The result shows that as data becomes

available, the previous estimate can be updated

110
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

without having to start from the beginning. X it|t  X it|t 1  K t  Yt   it  HX it|t 1  ,

Kalman filter (KF). Consider the model
 
1
X t  FX t 1   mt ,  mt ~ N  0, Q  , and the K t  Pˆ t|t 1H t HPˆ t|t 1H t  R ,
measurement Yt  HX t   ,  ~ N  0, R  .
o
t
o
t it ~ N  0, R  , i  1, , m
The error covariance for analysis and forecast Radial flow (reservoir infinite acting) into a
Pt|t  E  X  E  X | y1:t    X  E  X | y1:t   
t
well, the pressure P depends on the radius r
  and time t
 2 P 1 P c P
 
Pt|t 1  E  X  E  X | y1:t 1    X  E  X | y1:t 1   
t

  r 2
r r .0002637 k t
2rhk P
. Theforecast distribution P  r,0   Pi , P  , t   Pi , .001127 |r r  q
X t | y1:t 1  
~ N X t|t 1 , Pt|t 1 has mean and  r w
covariance X t|t 1  E  X t | y1:t 1   FX t 1|t 1 , The solution is given by the expression
Pt|t 1  Var  X t | y1:t   Q  FPt 1|t 1Ft . The  70.61Q0B0   948cr 
2
P  r, t   Pi    Ei  
analysis is given by  kh   kt 
  H t R 1H  P 1 1  H t R 1 y  P 1 X B , is the oil formation volume (well fields
X t | y1:t ~ N  
t|t 1 t t|t 1 t|t 1 0

 t 1 
  barrels/stock tank barrels), Q0 is the oil flow,
1  1

 H R H t  P t|t 1 
B0Q0  q is the flow rate, Ei is the
with mean and variance

X t|t  X t|t 1  K  y t  HX t|t 1  exponential integral Ei  x  
e t
,
x t dt
Pt|t   I  KH  Pt|t 1 ,
Fluid flow through composite reservoir
K  Pt|t 1H  H PH  R 
1
t t
regions represented by the diffusivity
equations:
Ensemble Kalman filter (EnKF). EnKF is a
Monte Carlo simulation in the context of KF. P1 1 P1  c  P1
   rw  r  R
Monte Carlo samples are used to estimate the r 2 r r  k 1 t
forecast using the nonlinear model. The mean
and variance of forecast were estimated using P2 1 P2  c  P2
   rR
Monte Carlo samples (ensemble), and then are r 2 r r  k 2 t
used in KF update for analysis step. The
P1  r, 0   Pi , P2  r, 0   Pi , P1  rw , t   Pw
forecast is given by
1 m P2 P k k
f  x t | y1:t 1    f  x t | x it 1|t 1  . The P1  R, t   P2  R, t  ,
r
|r R   1 |r R ,      
r
m i 1   1   2
update at time at time t is given by
P2  , t   Pi
 
m
1
f  x t | y1:t   f  y t | x t   N x it|t 1 , Pˆ t|t 1 .
m i 1 Consider a semi infinite composite regions
Assuming independent samples  , 0   0,   , the diffusivity equations are

x it 1|t 1 , i  1, , m , the basic steps are: (1)
 P1 1 P1
2

x it|t 1  f  x it 1|t 1    m,i

t ,  m,i ~ N  0, Q  ,  -<x<0
t x 2 k1 t
1 m
  
t
Pˆ t|t 1  
m  1 i 1
ˆ
X it|t 1  X t|t 1
ˆ
X it|t 1  X t|t 1
 2 P2 1 P2
 , x0
x 2 k 2 t
1 m
, X̂ t|t 1   X it|t1 (2) Update each P P
m i 1 k1 1 |x 0  k 2 2 |x 0 , P1  0, t   P2  0, t 
x x
sample using KF
P1  x, 0   0, P2  x, 0   0, P1  , 0   P

111
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

The solutions are given by (Carslaw, Jaeger, Pi  4000 psi , h = 15 ft, c  12 106 psi ,
1980: 319)
t  .5 hour . Figure 1 shows the

erfc
 2i  1   x   observations (represented by circles) and the
  evolution of transmissivity and storativity.

i 
2 k1t 
P1  x, t   P    The observations update the static parameters.
i 0  erfc  2i  1   x  The results shows that the EnKF converge to
 2 k1t  the true values after a number of iterations.

2P  i  2i+1   kx
P2  x, t   
1   i 0
 erfc
2 k1 t
k1 k  1
k ,  2 k, 
k2 k1  1
erfc  x   1  erf  x  ,
x
2
erf  x    erf     1 ,
2
e t dt ,
0
erf   x   erf  x 

III. RESULTS AND DISCUSSION

Almendral-Vazquez and Syversveen (2006)
gave a review of the application of EnKF in
the field of history matching of oil reservoir.
The interest is in estimating permeability
based on observations of the bottom hole
pressure (BHP). However, in this paper the
interest is in estimating the transmissivity Figure 1. EnKF for transmissivity T  kh /  and
T  kh /  and storativity S  ch based on storativity S  ch homogeneous reservoir. The
observations of bottom hole pressure. The results shows that the filter recover the true sorativity and
state consist of the static (S, T) and the transmissivity after a certain number of iteration.
dynamic bottom hole pressure P(r, t);

X  S T P  r,t  . 
The simulated bottom Composite reservoir plays important role in
reserve evaluation.. The interest is in
hole pressure are obtained from P(r,t) at well estimating the permeability based on bottom
radius r  rw . The observations is the hole pressure. The proposed composite model
can be considered as reinjection process. The
observed pressure; Y  Pobs  r, t  . The static
principal modes of transport is diffusion which
parameters are modeled as causes diffuse process from high pressure to
St  St 1 , Tt  Tt 1 ; t  t k  kt ; S0 , T0  regions with lower pressure. Considered as
are drawn from a normal distribution. Define one dimensional flow the flow from the
an interval [0, T], a set of observations was injector to the well can represented by one
dimensional flow channel. The basic solution
generated using  Strue , Ttrue  . The for interpretation is
observations at t obs was perturbed P  x, t   P2  x, t  
P  rw , t obs ,Strue , Ttrue    . The simulation 2P  i  2i+1   kx . The dynamic
was set up for T = 5000, rw  0.1 ft ,

1   i 0
 erfc
2 k1 t
  1.5 cp , Q0  300 STB/day , of the reservoir is represented by the analytic
solution or the simulator. The concept of fit is
B0  1.25 bbl/STB ,   0.15 ,
formalize as the difference between the

112
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

simulated pressure Psim  x, t  generated by

the analytical solution/reservoir simulator and
the pressure from observations Pobs  x, t 
from P2  x, t  using
 k1 , k 2 true ; Psim  x, t   Pobs  x, t  . The
interpretation can characterized by three
factors: (1) state space modeling and initial
reservoir parameters distribution (2)
characterization of prior distribution and (3)
updating (posterior) process. The experiment
was set up for T = 2000, ensemble size m =
100, number of observations n = 40, initial
pressure 4000, time step t  .5 , true
(b) Permeability region 2
permeabilities k1,true  43, k 2,true  54 .
Figures 2 – 5 show the results for composite
reservoir varying the well position x = 100 and
75 m. For all of three cases there is a good
match between the simulated and the true
pressure, more ever the initial parameter
distributions converge to the true values. For
the cases studied, the EnKF estimate
converges to the true parameter values, as
expected. The experiment shows that EnKF is
a promising method for updating reservoir
properties such as permeability and porosity.

Figure 2. Permeability progress and pressure

(simulated and observed)   200, x  100

(a) Permeability region 1

(a) Permeability region 1

113
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Figure 4 Permeability progress and pressure (simulated

and observed)   200, x  100
(b) Permeability region 2

Figure 3 Permeability progress and pressure (simulated

and observed)   200, x  100 (a) Permeability region 1

(a) Permeability region 1

(b) Permeability region 2

Figure 5 Permeability progress and pressure (simulated

(b) Permeability region 2
and observed)   200, x  75

114
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

IV. SUMMARY AND CONCLUSIONS Wikle C K, and Berliner L M, 2006, A

The ensemble Kalman filter is an iterative Bayesian Totorial for Data Assimilation,
procedure suitable for nonlinear models such Physica D, doi:10.1016/j.physd.2006.09.017
as composit reservoir models. The use of
EnKF for updating reservoir models,
analytically or numerically, using information
from observations reamains a problem of
intense interest. Currently, the equation for a
three phase system should be solved by the
simulator. An analytical approach is needed as
a part of validation process. In this paper,
investigate a special case, the analytical
solution a single phase system was used in
place of simulator. The applicability of the
method has been studied through a synthetic
examples. The experiment shows that EnKF
is a promising method for updating reservoir
properties such as permeability and porosity.
A 2D radial composite reservoir models is a
topic for further research.

REFERENCES
Almendral-Vazquez A, Syversveen A R, 2006,
The Ensemble Kalman Filter – Theory and
Application in Oil Industry, Norsk
Regnesentral, Norwegian Computing Centre
Carslaw H S and Jaeger J C, 1980, Conduction
of Heat in Solids, Oxford Clarendon Press
Chen Y, Oliver D S, and Zhang D, 2009, Data
Assimilation for Nonlinear Problems by
Ensemble Kalman Filter with
Reparameterization, Journal of Petroleoum
Science and Engineering, 66, 1-14
Demski J A, 1987, Decline Curve Derivative
Analysis for Homogeneous and Composite
Reservoir, Stanford University, California
Evensen G, 2007, Data Assimilation: The
Ensemble Kalman Filter, Springer
Gu Y, and Oliver D S, 2004, History Matching
of the PUNQ-S3 Reservoir Model Using the
Ensemble Kalman Filter, SPE 89942, SPE
Annual Technical Conference and Exhibition,
Houston, 26-29 Sept 2004
Naevdal G, Mannseth T, and Vefring E H,
2002, Near-Well Reservoir Monitoring
Through Ensemble Kalman Filter, SPE75235
SPE Improved Oil Recovery Symposium,
Tulsa California 13-17 April 2002
Naevdal G, Bianco A, Cominelli A, Dovera
L, Lorentzen R J, and Valles B, 2007, State
Estimation of a Large Scale System in the
Petroleoum Industry: the Ensemble Kalman
Filter for Updating Reservoir models, 8th Int
Synposium on Dynamics and Control of
Process System, June 6-8 2007, Mexico

115
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

PROBABILITY ANALYSIS OF RAINY EVENT WITH THE WEIBULL

DISTRIBUTION AS A BASIC MANAGEMENT
IN OIL PALM PLANTATION
Divo D. Silalahi
SMART Research Institute, PT. SMART Tbk, Riau, Indonesia
E-mail: divosilalahi@yahoo.co.id

Abstract distribution which has a representative and

The three parameter Weibull Distribution can be symmetric nature of data.
used to determine the probability of rainy event Weibull Distribution with three
number, both at the minimum, maximum, and parameters is proposed to use for the
average level. The three parameter Weibull probability of rainy event determination. This
Distribution has a symmetric pattern on density
distribution is one of the probability
function of the data. High Potential of rainfall
occurs during October to December, and the low distribution with the symmetric and
potential of rainfall occurs on February and June. representative parameters of an event
Therefore, this rainfall information can be used to changing. It has three parameters namely; A as
estimate the early season of effectiveness planting a scale, B as a threshold and C as a shape
and fertilizing in the field. At the minimum level of parameter. This distribution can be used to
rainy event, the highest probability level larger than determine the probability of rainy event at
99,06% occurs on April, at the average level the various levels of characteristic.
highest probability level larger than 53,77% occurs Therefore, in order to determine the
on September and the lowest 34,14% occurs on rainfall potential in a region, it is necessary to
July. It means that the lowest error of weather
apply a systematic study of the probability
predict will occurs on September and largest on
July. Then at the maximum level of monthly rainy value by determination of rainy event number
event the highest probability level larger than in a mathematical model.
4,97% occurs on November. Comparison results
for P75 and Pmean, on March, April, and September B. MATERIAL AND METHOD
has similar characteristics to the potentiallity of As a method to determine the probability
rainy event. This similar phenomenon also occurs
of rainy event, was obtained data from the
on May and July. The Strength correlation between
of P75 and Pmean is 0,914.
climate station in Division VII, Libo estate,
Riau. The data was collected since 1991 to
2009. Figure 1. shows the frequency of rainy
Keywords event at the climate station.
Three parameter Weibull distribution, correlation, 60

goodness of fit test

50

A. INTRODUCTION 40

Probability value is very useful in the 30

plantation sector, especially in oil palm

20
plantation. Probability value can be used to
Frequency

estimate an event changing, such as rainy 10 Std. Dev = 4.18

event which is used to predict the rain 0

Mean = 8.9
N = 228.00

potential in a region estate. Probability of 2.0 6.0 10.0 14.0 18.0 22.0
4.0 8.0 12.0 16.0 20.0
rainy event can also be used to determine the
Rainy Event
characteristics of rainfall both at the minimum,
maximum, and average level as one of key Figure 1. Frequency of rainy event since 1991 to 2009
factors in fertilizer management, the amount of As systematic estimation is used
evapo-transpiration, planting patterns theoretical method in the three parameter
adjustment, etc. Considering the rainy event Weibull distribution. Theoretical methods
always change, so it is necessary to use using maximum likelihood estimation method
statistical analysis using a specify probability as a determinant estimated parameters of the
distribution function. Then based on these

116
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

distribution, normality test was made using By using the natural logarithm of the
Kolmogorov-Smirnov test. likelihood function equation can be expressed
in form:
B.1. Three Parameter Weibull Distribution K = L (A ,B ,C )
 C -1

This distribution was introduced by 
 N   N 
K = L n  C N A -N   (X i -B )  A -N C   (X i -B )  A N 

  i= 1   i= 1  
Waloddi Weibull, a Swedish physicist, in
 N  X -B  C 
1939. As the name implies this distribution has exp  -  i
 i= 1  A  

 
three parameters, namely A as the scale
parameter, B as a location parameter and C as
B.3. Goodness of Fit Test
a shape parameter.
Hypothesis testing is a procedure to
Probability density function of random
make a decision about the rightness or
variable X as a Weibull Distribution density
wrongness of hypothesis that aims to achieve a
function (Bain, 1991) is given in the form:
decision to accept the hypothesis based on
C  X  B  C 1
  X  B C 
exp   
analysis of certain samples. One of the sample

f X (X)   A  A    ,B  X  
  A   analysis to test the validity of theoretical

0 , others distribution that have been assumed to be
(1) justified or refuted by statistical test for
Based on the probability density goodness of fit is Kolmogorov-Smirnov test.
function, the cumulative distribution function
of the three parameter Weibull Distribution is C. RESULT AND DISCUSSION
expressed in forms: C.1. Probability Analysis of Rainy Event

Using Theoretical Methods
 X  B   C

1  exp     ,X  B, A  0, C  0
Fx (X)  

 A   C.1.1. Description of Data Observation
0 ,X  B

(2) After calculating the primary data of

rainy event on January to December since
To determine the variant of the three 1999 to 2009, can be seen that there are some
parameter as follows: differences rainy event in every level of
Var(X)  E(X2 )  (E(X))2 distribution.
 2  1
E(X 2 )  A 2  1    2AB 1    B2
, a variant that is: Frequency
22
 C  C
20
  2   1  
2
18 Min
Var (X )  A 2  1      1     16 Mean
  C    C   
14 Max
(3) 12
10
B.2. Parameter Estimation with Theoretical 8

Method 6
4

Theoretical method were used to estimate 2

0
the parameters A, B, and C is the maximum Jan Feb M ar Apr M ei Jun Jul Aug Sept Oct Nov Dec

likelihood method (Cohen, et al., 1984). Month

Maximum likelihood estimates can be Figure 2. Distribution of monthly rainy event on January
obtained by solving the first derivative to December.
equation of likelihood function in Weibull
Distribution. The probability density function If be seen from the rainy event number,
is expressed in form: the lowest minimum spread number of
N
C  X -B 
C-1
  X -B C 
L(A,B,C)=  i  exp -  i  
monthly rainy event occurrs on January,
  A  
i=1 A  A    February, May, and July. The highest average
 C    Xi -B    N  Xi -B C  number of rainy event occurs in November.
N N C-1

=      exp - A  
 A   i=1  A    i=1   
Then the maximum of rainy event number
 N 
C
 N 
-1
 N  X -B C 
=CNA-N  (Xi -B) A-NC  (Xi -B) AN.exp - i  
 
occurs in October and November. From this
 i=1   i=1   i=1  A  
assumption, it can be known that the High
(4) (2.12) Potential of rainfall occurs on October,

117
 Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

November and Desember. While the low Min 1 41.7% 5.58 2.67 1.60 98.9%
potential of rainfall occurs in February and May Mean 8 41.7% 5.58 2.67 1.60 39.5%
June. Therefore base on this assumption. It can Max 16 41.7% 5.58 2.67 1.60 2.8%
be identified the early season of effectiveness Min 2 33.1% 46.95 -39.34 26.18 96.5%
for planting and fertilizing in the field. June Mean 7 33.1% 46.95 -39.34 26.18 49.1%
Max 10 33.1% 46.95 -39.34 26.18 2.5%
Frequency
100% Min 1 52.7% 5.91 1.78 1.45 99.2%
90% July Mean 8 52.7% 5.91 1.78 1.45 34.1%
80% Min
Max 19 52.7% 5.91 1.78 1.45 0.9%
70% Mean
60% Min 3 50.2% 5.13 2.83 1.26 98.6%
Max
50% August Mean 8 50.2% 5.13 2.83 1.26 36.4%
40% Max 18 50.2% 5.13 2.83 1.26 2.0%
30%
Min 3 34.1% 9.17 1.30 2.73 99.0%
20%
10% September Mean 10 34.1% 9.17 1.30 2.73 53.8%
0%
Jan Feb M ar A pr M ei Jun Jul A ug Sept Oct No v Dec
Max 17 34.1% 9.17 1.30 2.73 1.3%
Month Min 2 49.3% 10.69 0.46 2.02 98.0%

Figure 3. Probability distribution level of rainy event in October Mean 11 49.3% 10.69 0.46 2.02 37.9%
minimum, maximum, and average. Max 21 49.3% 10.69 0.46 2.02 2.4%
Min 4 36.0% 14.77 -0.14 3.07 98.0%
At the minimum level of rainy event, the November Mean 14 36.0% 14.77 -0.14 3.07 41.7%
highest probability level larger than 99,06% Max 21 36.0% 14.77 -0.14 3.07 5.0%
occur on April, at the average level the highest Min 5 31.3% 9.59 3.54 2.41 98.9%
probability level larger than 53,77% occurs on December Mean 13 31.3% 9.59 3.54 2.41 38.0%
September and the lowest 34.14% occurs on Max 20 31.3% 9.59 3.54 2.41 2.6%
July. It means that the lowest error of weather i. rate iv. scale parameter vii. probabilty
predict will occurs on September and largest ii. rainy event v. location parameter
on July. Then at the maximum level of iii. coefficient of variant vii. shape parameter
monthly rainy event the highest probability
level larger than 4.97% occurs on November. C.1.2. Goodness Test of Fit
Differences probability level of Pmean on rainy
event in each month may caused by different Based on the result of probability
coefficients of variant. If coefficients of obtained, be carried out Kolmogorov Smirnov
variants is ever larger and the average of rainy test using the ratio between of rainy events
event number ever lower. So the probability number by the total number that occur, then
level of Pmean will be going to ever lower. determined the maximum difference (DN)
between observed cumulative distribution
Description the probability range of rainy function (SN(X)) and cumulative distribution
event number since 1991 to 2009 as follows. function theoretical (F *(X)).
Table 1. Description the probability range of rainy event Table 2. Kolmogorov Smirnov test
number.
Month | DN | D α = 0 ,0 5 P-value α (alpha)
N =19
Month i ii iii iv v vi vii
January 0.132 0,301 0.404 0,05
Min 1 51.8% 9.65 -0.36 2.12 98.4% February 0.129 0,301 0.906 0,05
January Mean 9 51.8% 9.65 -0.36 2.12 39.2% March 0.153 0,301 0.612 0,05
Max 20 51.8% 9.65 -0.36 2.12 0.8% April 0.127 0,301 0.917 0,05
Min 1 41.2% 9.04 -2.48 3.90 97.6% May 0.119 0,301 0.948 0,05
February Mean 6 41.2% 9.04 -2.48 3.90 45.9% June 0.182 0,301 0.486 0,05

Max 10 41.2% 9.04 -2.48 3.90 3.0% July 0.095 0,301 0.976 0,05

Min 5 36.6% 9.95 13.24 3.64 99.2% August 0.112 0,301 0.907 0,05
March 10 36.6% 9.95 13.24 3.64 41.3% September 0.112 0,301 0.969 0,05
Mean
20 36.6% 9.95 13.24 3.64 1.6% October 0.109 0,301 0.802 0,05
Max
6 27.9% 4.46 5.80 1.49 99.1% November 0.077 0,301 0.991 0,05
Min
April 10 27.9% 4.46 5.80 1.49 40.1% December 0.161 0,301 0.702 0,05
Mean
Max 16 27.9% 4.46 5.80 1.49 3.2%

118
 Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Fitted Weibull Distribution Weibull Plot

C.1.4. Correlation between P75 and Pmean
40 99.9
99
Table 4. Correlation between P75 and Pmean
90
70
30

cumulative percent
50
Coefficients of
30 Regression Equation Sig-F Sig-t
percentage

20
10
correlation
20
5
P75 = 0,7213* Pmean - 0,8525 0,000 0,000 0,914
10 1
0.5

0.1
The Strength correlation between of P75
0
0.1 1 10 100 0.1 1 10 100 and Pmean is 0,914. So the regression equation
above can be used to facilitate using the three
parameter Weibull Distribution with
Figure 4. Fitted The three parameter of Weibull
Distribution and Weibull Distribution Plot.
probability level larger than 75%.

From the Kolmogorov-Smirnov test in D. CONCLUSION

table (2) and figure (4) above, can known that |
DN | of the Weibull Distribution for the According to the simulation results
parameter distribution show the maximum probability of rainy event number with the
difference is lower than D N=19α=0,05
= 0,301. It three parameter Weibull Distribution, can be
conclude that:
means the data distribution of rainy events
number on each month since 1991 to 2009 was 1. The three parameter Weibull Distribution
carried out representative result. P-value of can be used to determine the probability
Kolmogorov-Smirnov which has been of rainy event number
generated from each distribution larger than α 2. The three parameter Weibull Distribution
= 0,05. So, can be assumed that the probability has a symmetric pattern on density
distribution of each month has a representative function of the data.
and symmetric nature of the data. 3. High Potential of rainfall occurs during
October to December, and the low
C.1.3. Difference Decrease of P75 and Pmean potential of rainfall occurs on February
Table 3. Differences decrease of P75 and Pmean
and June.
4. At the minimum level of rainy event, the
Standard Coefficients
Month P Mean P75
Deviation of Variant
highest probability level larger than
Januari 9 5 4,24 51,83%
99,06% occurs on April, at the average
level the highest probability level larger
Februari 6 4 2,35 41,16%
than 53,77% occurs on September and
Maret 10 7 3,64 36,59%
the lowest 34.14% on July. It means that
April 10 7 2,75 27,95%
the lowest error of weather predict will
Mei 8 5 3,20 41,66%
occurs on September and largest on July.
Juni 7 5 2,19 33,06%
Then at the maximum level of monthly
Juli 8 4 3,77 52,73%
rainy event the highest probability level
Agustus 8 4 3,82 50,24%
larger than 4.97% occurs on November.
September 10 7 4,23 34,10% 5. Comparison results for P75 and Pmean, on
Oktober 11 6 4,90 49,31% March, April, and September has similar
November 14 10 4,71 36,03% characteristics to the potentiallity of rainy
Desember 13 8 3,77 31,27% event. This similar phenomenon also
Based on this comparison result, on occurs on May and July. The Strength
March, April, and September has similar correlation between of P75 and Pmean is
characteristics to the potentiallity of rainy 0,914.
event. This similar phenomenon also occurs on
May and July. E. REFFERENCES

Bain, L.J and M. Engelhardt. 1991.

Introduction to Probability and
Mathematical Statistics. California:
Duxbury Press.
Casella, G. and Berger, R.L. 1990. Statistical
Inference. California. Brooks/Cle

119
Proceeding of Conf. on Industrial and Appl. Math., Indonesia 2010

Publishing Company. in Enginering Kao, J. H. K. (1959). A Graphical Estimation

and management Science. John Willey of Mixed Weibull parameters in Life
& Sons. Inc. testing of Electron Tubes.
Cohen, A. C., Whitten, B. and Ding, Y. Technometrics 1, pp. 389-407.
(1984). Modified Moment Estimation Mann, N. and Fertig, K. (1975). A Goodness-
for the Three-Parameter Weibull of-Fit-Test for the two-Parameters vs.
Distribution. Journal of Quality Three-Parameter weibull; confidence
Technology 16, pp. 159-167. bounds for threshold. Technometrics
Haan,C.T. 1977. Statisticsl Methodes in 17, pp. 237-245.
Hydrology. The Lowa State University Sellers, W.D. 1965. Physical Climatology.
Press. Ames. Lowa. 378P. University of Toronto Press, 272 pp.
Hines, W.W and Montgomery, D.C. 1996. Weibull, W. (1951). A Statistical Distribution
Probability and Statistics in Function of Wide Applicability.
Enginering and management Science. Journal of Applied Mechanics 18, pp.
John Willey & Sons. Inc. 294-298

120
Proceedingg of Conf. on In
ndustrial and Appl.
A Math., Bandung-Indoneesia 2010

Calculating Areaa of Earth

h’s Surfacee
Baased on Disscrete GPSS Data

wan (1), Aripinn Iskandar (2)

Alexandder A S Gunaw
(11)
Mathematics Department, BBinus Universitty, Jakarta, Inddonesia
(Email: aagunng@binus.eduu (1))
(2)
Computer Enggineering Lab, Binus Universsity, Jakarta, Inndonesia
(E
Email: aripin_8888@yahoo.coo.id (2))

ABSTRA
ACT will allow
a the dettection of thhe latitude annd
In this paper, it is presentted the method d to calculate longituude on earthh’s surface as a in [1]. By B
area of Earth’s
E surfacee based on discrete
d GPS considdering the eartth’s shape, thhe discrete GPPS
(Global Positioning Sysstem) data. The T obtained data of earth’s surfface must be projected onnto
discrete GP PS data take thet format of latitude, and a flat plane or 2 dimensional views. In thhis
longitude. To utilize the GPS data, thee coordinates
paper, the Lam mbert Confformal Connic
must be projected on nto a flat plane
p or 2
dimensionaal view. Oncee the projectedd coordinates Projecction is used to determinee the projecteed
are obtainned, points, lin nes and polyggons can be coordiinates, and based on the obtaineed
quantified.. The objectiive of the paper p is to projected coordinnates, pointss; lines annd
manipulatee discrete GPS S data to definne a polygon polygoons can be quuantified. Thee next step is to
by using thhe Lambert Coonformal Conic Projection, calculaate the projeected area of the polygoon
and then to t determine the projected area of the based on the disccrete GPS data d of earthh’s
polygon byy using Green n’s Theorem. The
T resulted surfacee by using Grreen’s Theoreem.
algorithm is
i implemented d in software.
REA CALCU
II. AR ULATION
KEYWO
ORDS Lamb bert Conform mal Conic Proojection
Discrete GPS
G data; Eartth surface; Arrea; Lambert
A Lam mbert conform mal conic prrojection as in
Conformall Conic Projecttion; Green’s Theorem.
T
[4] is a conic map projection. In I essence, thhe
I.INTRO ODUCTION projection superim mposes a coone over thhe
Area is a quantity expressing the two- spheree of the E Earth, with two standarrd
dimensionnal size of a defined
d part of
o a surface, paralleels to the gloobe and interssecting it. Thhis
typically a region bounded by a closed curve. minimmizes distortioon from projjecting a threee
For certaain closed cuurves, such as Square, dimensional surface to a tw wo-dimensionnal
Paralleloggram, and Regular
R polyygon, it is surfacee. There is no distortiion along thhe
simple to calculate areea by definitee formulae. standaard parallels,, but distorrtion increasees
Unfortunaately, the arrea calculation on the furtherr from the choosen parallelss.
Earth's suurface is commplicated by thhe fact that
the latitudde and longiitude line segments are
actually curves
c and thhat the distannce between
points separated by the t amount degrees of
longitude varies accordding to the co osine of the
latitude. Furthermore,
F Earth is noot a sphere.
Earth is a little flat att the poles (tthe distance
from the center to a pole p is shortter than the Figgure 1. Lambert Conformal Con
nic Projection
distance from centerr to anywheere on the As thhe name inndicates, mapps using thhis
equator) and
a is a little rough. Thus to calculate projection are coonformal, th hat means a
area of Eaarth’s surfacee, first we mu ust consider mapping which preserves angless.
the earth’s shape andd then derivve the area
calculation algorithm. To connsider the earth’s shape, it
i is used GR
RS
80, or Geodetic Refference Systeem 1980, that is
The propo osed approacch here is to exploit the a geoddetic referennce system consisting of a
availabilitty of microwwave signalss from the global reference ellipsoid. A referencce
thirty sattellites of the
t Global Positioning ellipso
oid, customarrily chosen too be the sam
me
System (G GPS). The monitoring
m off GPS data size (vvolume) as thhe figure off the Earth, is

134
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

described by its equatorial radius a and

eccentricity e. For GRS 80, these are: Below are the integrals representing Green’s
a = 6,378,137 m; theorem as in [2]. On the left there is area
e = 0.081819191 integration (double integral) and on the right
Let λ be the longitude, λ0 the reference there is line
 M L 
longitude, φ the latitude, φ0 the reference
latitude, and φ1 and φ2 the standard parallels.
    dA   L dx  M dy
x y 
By using this reference system, several integration (single integral).
additional parameters need to compute before
transformation can be undertaken, that is: (8)
ln m1  ln m 2 A   dA where M and L are two
n (1)
ln t1  ln t 2 functions M(x,y) and N(x,y).
m1 The area of any real region is defined as:
F (2)
n t1 
n
(9)
where
If two functions M and L is chosen such that
cos
m 1 1
(3)
1  e 2 sin 2  M  x and L y
2 2
t

tan 4    

2

1e sin 
1 e sin 
 (4)

  aF tn (5) 1
A  dA  2  x dy  y dx
Note:
 m1 and m2 are obtained by evaluating m
(Eq. 3) using φ1 and φ2 the standard M L
parallels.   1 ; thus the Green’s Theorem
 t0, t1 and t2 are obtained by evaluating t
x y
(Eq. 4) using φ0, φ1 and φ2 could be used to calculate area. In this
 0 are obtained by evaluating  using t0 paper, we choose:

Then the transformation of spherical (10)

coordinates to the plane via the Lambert
conformal conic projection is given by: Therefore, the formulae for area
calculation is:
x   sin n   0  (6)
y   0   cosn  0  (7) (11)
In the application, we can choose the reference
coordinate (φ0, λ0) as the average of all data The line integral can be evaluated using the
and the standard parallels φ1 and φ2 as the following discrete approximation as in [5],
1 n
minimum and the maximum latitude in the
data.
A  x i y i  y i x i
2 i 0
(12)

In the application, area is typically a region

bounded by a closed loop polygon and thus for
Green’s Theorem convenience, the initial point (x0, y0) of (Eq.
Green’s theorem is a very interesting theorem 12) is set equal to the end point (xn, yn) in this
which shows a relationship between a closed closed loop polygon. The sign of area
area and the path surrounding this closed area. calculation is depend on the orientation of
closed loop polygon, that is: (+) for

135
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

counterclockwise orientation and (-) for 4. The resulting is the decimal value of
clockwise orientation. degrees (0.9874) and sum the degrees
GPS (Global Positioning System) to the decimal using the symbols for
The Global Positioning System (GPS) as in [3] degrees (°) (-73.9874°)
is a space-based global navigation satellite
system that provides reliable location and time In the software implementation, the above
information in all weather and at all times and conversion procedure must be done to convert
anywhere on or near the Earth where there is all discrete GPS data to decimal degrees
an unobstructed line of sight to four or more (dd.dddddd) format.
GPS satellites. It is maintained by the United
States government and is freely accessible by Software Implementation
anyone with a GPS receiver. A GPS receiver Software Implementation is the final, and most
calculates its position by precisely timing the involved, step in the research. We choose
signals sent by the GPS satellites high above Visual Basic as programming language, and
the Earth. two data format that is MMC (Microsoft
Media Catalog) file and Excel file.
The research as basis of this paper is used a
special GPS receiver that is EM-406 from The software which called as “arealator” is
GlobalSat. GPS Data Format of this receiver developed based on the Lambert conformal
for both latitude and longitude is "degrees, conic projection and the Green’s Theorem for
minutes, and decimal minutes" format— earth’s area calculation. The discrete GPS data
ddmm.mmmm. However, the Lambert will be first converted to decimal degrees
conformal conic projection requires longitude (dd.dddddd) format before calculating earth’s
and latitude to be expressed in decimal degrees area. The below figure is the GUI (Graphics
(dd.dddddd) with a corresponding sign User Interface) of the software:
(negative for south latitude and west
longitude). Hence, we need to perform some
conversions.

Figure 3. Arealator GUI

Furthermore, the software is linked to Google

Figure 2. GPS Receiver
map via www.gpsvisualizer.com, hence we
could see the actual location of the GPS data
The following procedure converts the where the area is calculated.
ddmm.mmmm format to decimal degrees
format:
1. For example, we have "degrees,
minutes, and decimal minutes" format —
ddmm.mmmm that is (-73° 59’ 14.64")
2. Divide the decimal minutes by 60
(14.64 / 60 = 0.244)
3. Sum the resulting to the minute and
divide by 60 (59.244 / 60 = 0.9874)

136
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

mountainous regions. This limitation will be

the first issue in our next step research.

REFERENCES
[1] C.G. Carlson, “What do latitude and
longitude readings from a DGPS receiver
mean?”, South Dakota State University,
Brookings, 1999.

[2] D. Varberg, E.J. Purcell, “Calculus”, 9th

ed., Prentice Hall, 2006.
Figure 4. Google Map

[3] J-M. Zogg, “GPS Basics: Introduction to

III. CONCLUSIONS
the System”, U-blox ag, Thalwil,
There are advantage and disadvantage of our
Switzerland, 2002.
approach in solving the earth’s area
calculation problem by projecting the latitude
[4] O.S. Adams, “General Theory of The
and longitude data (ignoring elevation) onto 2
Lambert Conformal Conic Projection”,
dimensional views. The advantage is, of
Washington Goverment Printing Office,
course, that calculus can be used to derive the
1918.
algorithm of area calculation. Nevertheless,
the main disadvantage of the approach is that
[5] S.C. Chapra, R.P. Canale, “Numerical
it does not count the impact of elevation data
Methods for Engineers”, 5th ed., McGraw
to the area calculation. Therefore, the
Hill, 2006.
approach that proposed in the paper is not
suitable applying to measure area in

137
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Study on Application of Machine Vision using Least-Mean-Square (LMS)

Hendro Nurhadi(1) and Irhamah(2)

(1)
Department of Mechanical Eng., Institut Teknologi Sepuluh Nopember (ITS), Surabaya, East Java,
INDONESIA
(Email: hdnurhadi@me.its.ac.id)
(2)
Department of Statistics, Institut Teknologi Sepuluh Nopember (ITS), Surabaya, East Java, INDONESIA
(Email: irhamah@statistika.its.ac.id)

ABSTRACT on using LMS in Machine Vision’s

In recent years, implementations of image applications.
processing for Machine Vision using LMS (Least-
Mean-Square), PCA (Principal-Component- II. Video Enhancement Using Content-
Analysis) and SVD (Singular-Value- adaptive Least Mean Square Filters
Decomposition) in industrial applications are
The current television system is evolving
widely performed. This study discusses the
significant differentiation of linear function due to towards an entirely digital, high resolution and
an irregularity in one of the data. This study high picture rate broadcasting system. As
focuses in three cases: video enhancement using compatibility with the past weights heavily for
content-adaptive LMS filters; machine vision a popular consumer service, this evolution
system for the automatic identification of robot progresses rather slowly, [4]–[5]. The progress
kinematics parameters; and visual servo auto- is particularly slow compared to the revolution
focusing using recursive weighted least-squares for we observe in modern display technology,
machine vision inspection. In conclusion, LMS is a which in the last decade eliminated the
robust tool that can be used to be an appropriate traditional bottleneck in television picture
approach for some applications in machine vision.
quality. Methods, that can bridge the resulting
gap between broadcast and display picture
KEYWORDS
Machine vision; least square; component analysis; quality, are therefore increasingly important
value decomposition; image processing; video and may profit from the rapid developments in
enhancement; identification. semiconductor technology.
This last trend enables a steadily increasing
I.INTRODUCTION processing power at a given price level, and
LMS, PCA and SVD are common tools which makes the implementation of complex video
used to solve many problems for applications enhancement algorithms feasible in consumer
in Machine Vision. It introduces that the linear equipment, [6]–[7].
function differs significantly due to an We started by reviewing image and video
irregularity in one of the data. LMS approach resolution enhancement techniques proposed
also couldn’t obtain proper results if there is a in scientific papers and patent literature. This
miss in the experiment. In this study, we’ll review revealed two interesting approaches,
learn and see how reliable is this LMS method both derived from image restoration theory,
for such Machine Vision’s applications. We’ll i.e. the Least Mean Square (LMS) or Wiener
discuss them in according to three cases: filtering approach, and the Bayesian image
Video Enhancement Using Content-adaptive restoration approach, [8]–[9].
Least Mean Square Filters [1], Machine Vision We focused on the LMS-filtering methods as
System for the Automatic Identification of particularly Kondo’s classification based LMS
Robot Kinematics Parameters [2], and Visual filtering approach for video resolution up-
Servo Auto-focusing using Recursive conversion method attracted our special
Weighted Least-Squares for Machine Vision attention due to its high performance and
Inspection [3]. relatively simple hardware implementation.
Using one of user friendly method such as Furthermore we showed that although the
LMS (Least-Mean-Square), PCA (Principal- LMS filtering approaches are based on
Component-Analysis) and SVD (Singular-Value- objective metrics, particularly the MSE
Decomposition) in application of Machine criterion, it is still possible, for video up-
Vision is the main goal of writing this study of scaling, to design subjectively more pleasing
papers. In this study report, it will be focused filters using the LMS metric.

138
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

We also concluded that a direct mapping of When encoding each pixel into 1-bit, , with
the content-adaptive LMS filtering to the ADRC:
problem of intra-field de-interlacing does not
outperform all existing intra-field methods.  F  FMIN 
However, the classification based LMS Q   SD  0.5 (1)
filtering approach was shown to successfully  FMAX  FMIN 
replace the heuristics in two other de-
interlacing designs, including the vertical Here is the luminance value of the SD
temporal filtering and the data mixing for pixel. and are the maximum and
hybrid deinterlacing combining motion- minimum luminance values of the pixels in the
compensated and edge-adaptive techniques. classification aperture respectively. . is the
floor operator. Let be the luminance value
of the original (not the up-converted) HD
pixels and be the value of the interpolated
ones, which is a weighted sum of the nine SD
pixels in the interpolation window. The
equation to interpolate pixels on position A is:
Figure 1. The training process according to Kondo [1]
The SD (standard definition) video signal is a down- FHI  2  i  2  , 2  j  2   
scaled version of the HD (high definition) signal.
2 2

 w FSD  2  i  2k   1, 2  j  2l   1
Sample pairs are extracted from the training material
and classification of samples is done by ADRC kl , c
(Adaptive Dynamic Range Coding). Using LMS k 0 l 0
(Least Mean Square) algorithm within every class, (2)
optimal coefficients are figured out and stored in a
Look-Up Table (LUT). where wkl,c are weights for class c. Suppose
one class contains in total a number of t
Finally, we investigated the classification samples in the training process, then the error
based LMS filtering approach for chrominance of the pth interpolation sample is:
resolution enhancement and coding-artifact
reduction. In the chrominance up-conversion
e p ,c  FHD , p  FHI , p
problem, we combined luminance and
chrominance information in an innovative e p ,c  FHD , p 
classification of the local image patterns 2 2
leading to an improved performance compared
 w kl , c FSD , p  2  i  2k   1, 2  j  2l   1
to earlier designs. k 0 l 0

In coding-artifact reduction, the relative

 p  1, 2, , t 
position information of the pixel inside the (3)
coding block along with the local image Consequently, the total (squared) error of this
structure for classification was shown to give class can be expressed as:
an interesting performance and elegant
t
optimization. This reveals a more general
usage of LMS filtering for video enhancement, ec2   e2p,c (4)
p 1
which can avoid many heuristic optimizations
that are commonly seen in this area.
To find the minimum, we calculate the first
The classification-based LMS filtering derivative of 2 with respect to each
algorithm is described in a stack of about 50
patents invented by Kondo and owned by
Sony. We explored the options for
classification-based LMS filtering applied in
de-blocking of severely compressed digital
video.

139
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

ec2 t  e p ,c  the subjective impression. However, the LMS-

  2  e p ,c based filter design for image up-conversion
wkl ,c p 1  wkl ,c  based on the objective MSE (mean-square-
t error) metric can effectively avoid heuristic
  2 FSD , p  2  i  2k   1, 2  j  2l   1 e p ,c approaches in the optimal filter design.
p 1

 k  0,1, 2; l  0,1, 2 
(5)

The minimum occurs when the first derivative

is zero, which leads to the following equation
for each class:

 X 00,00 X 00,01  X 00,22   w00,c  Y0 

X X 10,01  X 10,22   w01,c   Y1 
 10,00  
          
    
 X 22,00 X 22,01  X 22,22   w22,c  Y8 
(6)

The coefficients wkl,c can be obtained by

solving above equation for each class. Here,

t
X kl ,qr   FSD , p  2  i  2k   1, 2  j  2l   1
p 1

FSD , p  2  i  2q   1, 2  j  2r   1
 k , q  0,1, 2; l , r  0,1, 2 
(7)
and
t
Y3k l   FSD , p  2  i  2k   1, 2  j  2l   1
p 1

FHD , p  2  i  2   1, 2  j  2  
 k  0,1, 2; l  0,1, 2  Figure 2. An image portion from up-converted
(8) Bicycle using: Original image, (A) Standard TK
(Kondo) method, (B) JT method, (C) Standard TK
method with classification-based sharpness
The classification-based global LMS enhancement, (D) Classification-based subjective
optimization with 3 × 3 aperture gives the optimal up-conversion using 3 × 3 aperture, (E)
better performance/price ratio compared to Classification-based subjective optimal up-conversion
other LMS filtering based up-conversion using the diamond shape aperture, (F) Subjective
optimal up-conversion based on 3 × 3 aperture using
methods. Off-line LMS optimization improves localized approach.
the computational efficiency and provides
additional flexibility.

The classification-based global LMS

optimization can be used to build
classification-based sharpening filters, i.e.
classification-based subjectively optimal LMS
filter. Image quality remains a subjective issue
and no objective measure can reliably reflect

140
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

objective metrics, particularly the MSE

criterion, it is still possible, for video up-
scaling, to design subjectively more pleasing
filters using the LMS metric.
As a conclusion, a direct mapping of the
content-adaptive LMS filtering to the problem
of intra-field de-interlacing does not
outperform all existing intra-field methods.
However, the classification-based LMS
filtering approach was shown to successfully
replace the heuristics in two other de-
interlacing designs, including vertical temporal
filtering and data mixing for hybrid de-
interlacing combining motion-compensated
and edge-adaptive techniques.

III. Machine Vision System for the

Automatic Identification of Robot
Kinematics Parameters
This paper presents an efficient, noncontact
measurement technique for the automatic
identification of the real kinematics parameters
of an industrial robot. The technique is based
on least-squares analysis and on the Hayati
and Mirmirani kinematics modeling
convention for closed kinematics chains. The
measurement system consists of a single
camera mounted on the robot’s wrist. The
Figure 3. An image portion from up-converted Siena
using: Original image, (A) Standard TK method, (B) camera measures position and orientation of a
JT method, (C) Standard TK method with passive target in six degrees of freedom.
classification-based sharpness enhancement, (D)
Classification-based subjective optimal up-conversion
using 3 × 3 aperture, (E) Classification-based
subjective optimal up-conversion using diamond
shape aperture, (F) Subjective optimal up-conversion
based on 3 × 3 aperture using localized approach.

From this overview, Kondo’s classification

based LMS filtering approach for video
resolution up-conversion attracted our
attention due to its high performance and
Figure 4. Kinematic configuration of the system [2]
relatively simple hardware implementation.
It is also recognized that the content-adaptive
LMS filtering, especially Kondo’s Target position is evaluated by applying least-
squares analysis on an over determined system
classification-based LMS filtering, which
of equations based on the quaternion
outperforms Li’s localized LMS-filtering, can
representation of the finite rotation formula.
be used for a broad area of video enhancement
To enhance the accuracy of the measurement,
applications. Examples of applications
a variety of image processing functions
elaborated in this thesis include resolution up-
including sub-pixel interpolation are applied.
conversion, de-interlacing, chrominance
The unknown kinematics parameters are
resolution up-conversion and coding artifact
identified by solving
reduction.

Furthermore, those were showed that although M1M 2M 3M 4  I (9)

the LMS filtering approaches are based on or

141
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Fi  M 1  M 41 M 31 M 21  0 (10) characteristics including area and centroid are
simultaneously evaluated.
The machine vision system is used to
The form characteristics are used to resolve
determine the position and orientation of the
the correspondence problem, i.e., relating the
target with respect to the camera. The camera
2-D objects in the image to the three-
used is a high-resolution charge-coupled-
dimensional spheres in the target. Following
device (CCD)-type camera.
this, the maximum gradient pixels P(i,j) are
identified for each contour in the image. For
each pixel P(i,j), a sub-pixel interpolation
technique is applied to determine the exact
position of the maximum gradient with sub-
pixel accuracy. A detailed description of this
technique is presented in. To summarize the
approach, the gradient values G(P(i+m,j+n)) of a
5x5 kernel around pixel P(i,j) are used for the
sub-pixel interpolation. The Gauss function
S(x,y) represented by
Figure 5. Hayati and Mirmirani kinematic model
  G  P e
2 2
1 2  x  m    y  n  
2 2

S  x, y   im, j n
 

The target is passive in nature implying that m 2 m 2

the only light emanating from the target is (11)
ambient reflected light. Ideally, the target is
composed of a 3-D array of precision spheres must then be maximized with respect to the x
accurately positioned in three-dimensional and y coordinates measured from the center of
space. To facilitate the identification of each the P(i,j) pixel.
sphere within the image, one of the spheres is The sub-pixel interpolation technique
intentionally oversized. In order to derive generates a set of n points with coordinates
three-dimensional information from a single (xi,yi) which accurately describe the contour of
(2-D) perspective view, the coordinates of the each object. The objects being circular, the
centers of the spheres must be known, a priori, equation of a circle
relative to the target reference frame. Although
it is not mathematically required, it is Fi  ac , bc , rc    xi  ac    yi  bc   rc2
2 2

preferable from an accuracy point of view if

the centers of the spheres do not all lie on the (12)
same plane. This introduces a “perspective
effect” which accentuates the visual combined with a least-squares technique allow
consequences of small rotations about axes the identification of the center coordinates
parallel to the camera plane. (ac,bc) as well as the radius rc of each circle
The position of the target with respect to the corresponding to one sphere of the target.
camera is determined by analyzing a single
clear image of the target. The target is IV. Visual Servo Auto-focusing using
composed of an array of spheres, which appear Recursive Weighted Least-Squares for
as circles when projected onto the 2-D image Machine Vision Inspection
sensor. Various image-processing algorithms Auto-focusing of an electronic camera is an
are applied in order to accurately determine important task in automated machine vision
the center of each of the circles in the image. applications. It helps to obtain clear images for
A Sobel filter is applied to isolate the edges of accurate inspection. For example, in electronic
the image, i.e., those areas with an elevated assembly industry, the camera focus needs to
spatial gradient. Further image segmentation be constantly adjusted according to the size of
(i.e., thresholding) yields a binary image different components to obtain sharp images
containing only edge information. A contour so that accurate data could be extracted from a
tracking algorithm is then applied to the binary pool of information available without much
image in order to identify the contours of the processing power.
objects within the image. A number of form

142
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

In this paper, we propose a visual servo auto- Hence, the RWLS estimation of quadratic
focusing mechanism using a moments based curve parameters is obtained to be
focus measure. A recursive weighted least-
squares focusing algorithm is developed to Pi 1 H i 1
drive a camera to the focused position. It is xi  xi 1   i  H i xi  (14)
shown that the proposed auto-focusing scheme H i Pi 1 H i  R
can guide the camera to the focused position in and
a fast speed regardless of the initial position Pi 1 H i H iT Pi 1
and the object to be focused. Pi  Pi 1  T (15)
H i Pi 1 H i  R

where R is a constant weight assigned to the

RWLS and Pi is covariance matrix of ni.

A visual servo mechanism for auto-focusing is

presented. The approach involves moments
feature extraction, quadratic curve parameters
estimation using recursive weighted least
squares, profile construction, searching for
minimum point of the estimated quadratic
curve and moving the camera toward the
focused position. The proposed scheme does
not require any prior knowledge about the
assembly process and the object to he focused
on, it is a robust auto-focusing scheme suitable
for wide range of application.

III. CONCLUSIONS
Content-adaptive Least Mean Square (LMS)
Figure 6. Flowchart of proposed visual servo auto-
filters, i.e. content-adaptive filters designed via
focusing [3] linear regression have been investigated in this
thesis for applications in video and image
Now we develop the recursive algorithm to enhancement.
estimate the quadratic curve and the camera Our interest in these filters was triggered when
focused positions zi. Considering the noise creating an overview of state of the art video
pollution in practice, the obtained focus up-scaling and resolution up-conversion
measure is added with a Gaussian zero-mean techniques using both objective and subjective
random noise n. Thus, we have evaluation. From this overview, Kondo’s
classification based LMS filtering approach
 i  H iT xi  ni for video resolution up-conversion attracted
(13)
our attention due to its high performance and
relatively simple hardware implementation.
T
 1 Li’s localized LMS filtering approach for
 
2
 1
where H i      1 and resolution up-conversion was investigated in a
 z i   zi   comparative study as an alternative.
We recognized that the content-adaptive LMS
x i  a i ci  .
T
bi filtering, especially Kondo’s classification-
based LMS filtering, which outperforms Li’s
It is desired to obtain an estimate of xi localized LMS-filtering, can be used for a
recursively when additional information is broad area of video enhancement applications.
available in the form of new moments feature. Examples of applications elaborated in this
To do this, both moments feature extracted thesis include resolution up-conversion, de-
from two consecutive camera locations, i-1 interlacing, chrominance resolution up-
and i, are partitioned and its moments conversion and coding artefact reduction.
estimation error cost function is minimized.

143
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Finally, we investigated the classification- 5th Conference (EPTC 2003) 10-12 Dec.
based LMS filtering approach for chrominance 2003, pp.777-780.
resolution enhancement and coding-artefact [4] J. Whitaker, Standard Handbook of Video
reduction. In the chrominance up-conversion and Television Engineering, 4th Edition,
problem, we combined luminance and ISBN: 0071411801, McGraw-Hill
chrominance information in an innovative companies, 2003.
classification of the local image patterns [5] H. Kaneko and T. Ishiguro, “Digital
leading to an improved performance compared television transmission using bandwidth
to earlier designs. In coding-artefact reduction, compression techniques”, IEEE
the relative position information of the pixel Communications Magazine, Vol. 18, No. 4,
inside the coding block, along with the local pp. 14-22, Jul. 1980.
image structure for classification, was shown [6] U. Reimers, Digital video broadcasting –
to give interesting performance and elegant the international standard for digital
optimisation. television, Springer-Verlag Berlin
Heidelberg New York, ISBN: 3-540-
REFERENCES 60946-6, 2001.
[1] Meng Zhao, “Video Enhancement Using [7] U. Reimers, “Digital video broadcasting”,
Content-adaptive Least Mean Square IEEE Communications Magazine, Vol. 36,
Filters”, PhD Dissertation, Technische No. 6, pp. 104-110, Jun. 1998.
Universiteit Eindhoven 2006. ISBN 90- [8] T. Sikora, “MPEG Digital Video-Coding
386-0774-1 - ISBN 978-90-386-0774-0. Standards”, IEEE Signal Processing
[2] Patrick Rousseau, Alain Desrochers, and Magazine, Vol. 14, No. 5, pp. 82-100, Sep.
Nicholas Krouglicof, “Machine Vision 1997.
System for the Automatic Identification of [9] P. Gastaldo, S. Rovetta, R. Zunino,
Robot Kinematic Parameters”, IEEE “Objective quality assessment of MPEG-2
Transactions on Robotics and Automation, video streams by using CBP neural
Vol.17, No.6, Dec.2001, pp.972-978. networks”, IEEE Transactions on Neural
[3] Chee Hwa Chin, Ying Zhang, Changyun Networks, Vol. 13, No. 4, pp. 939-947, Jul.
Wen, “Visual Servo Auto-focusing using 2002.
Recursive Weighted Least-Squares for
Machine Vision Inspection”, IEEE
Electronics Packaging Technology, 2003

144
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Cooperative Linear Quadratic Game for Descriptor System

Salmah
Department of Mathematics, Gadjah Mada University, Yogyakarta, DI Yogyakarta, Indonesia
(Email: syalmah@yahoo.com)

T
 
ABSTRACT 1
2 0
In this paper the cooperative linear quadratic game J i (u1 , u 2 ,..., u n )  x T (t )Qi x (t )  u iT (t ) Ri u i (t ) dt ,
problem will be considered. In the game, either
there is one individual who has multiple objectives
or there are a number of players all facing linear (1)
quadratic control problem, will cooperate together
with all matrices are constant, Qi  Qi , and
T
to obtain optimal result. For ordinary system case
the set of Pareto efficient equilibria can be Ri positive definite. The inclusion of player
determined for these games. In this paper the result
will be generalized for descriptor system. j’s control effort into player I’s cost function is
dropped, due to the open-loop information
KEYWORDS structure, this term drops out in the analysis.
Dynamic, game, cooperative, descriptor, system. The players give control vector to the system

I. INTRODUCTION Ex(t )  Ax(t )  B1u1 (t )  B2u2 (t ) ,

x(0)  x0
Dynamic game theory brings three keys to
many situations in economy, ecology, and (2)
elsewhere: optimizing behavior, multiple
( nr ) x ( nr ) ( n  r ) xm
agents presence, and decisions consequences. with E , A   , Bi   1
, x(t)
Therefore this theory has been used to study descriptor vector n+r dimension. While ui(t),
various policy problems especially in macro- i=1,2 are control vector mi dimension which
economic. In applications one often
encounters systems described by differential is done by i-th player, i=1,2. Matrix E
equations system subject to algebraic generally singular with rank E  n. The initial
constraints. The descriptor systems, gives a vector x 0 is assumed to be a consistent initial
realistic model for this systems. state.
In policy coordination problems, questions
arise, are policies coordinated and which We recall the following result from theory of
information do the parties have. One scenario descriptor (see [7]) system for the differential
is cooperative open-loop game. According algebraic equation
this, the parties can not react to each other’s Ex  Ax(t )  f (t ), x(0)  x0
policies, and the only information that the
(DAE)
players know is the model structure and initial
state. And the associated pencil
E  A
II. PRELIMINARIES (3)
In this paper we will consider a linear open- System (DAE) is called regular if the
loop dynamic game in which the player satisfy characteristic polynomial det(E  A) is not
a linear descriptor system and minimize
identically zero. If the system (DAE) is not
quadratic objective function. For finite horizon
regular then any consistent initial conditions
problem, solution of generalized Riccati
do not uniquely determine solutions (see [13]).
differential equation is studied. If the planning
If the system (DAE) is regular, the roots of
horizon is extended to infinity the differential
characteristic polynomial are the finite
Riccati equation will become an algebraic
eigenvalues of the pencil. If E is singular the
Riccati equation. Formally, the players are
pencil (3) is said to have infinite eigenvalues
assumed to minimize the performance criteria
which is the zero eigenvalues of the inverse

145
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

pencil E  A . In the next discussion we Next, let [VW ] be an orthogonal matrix

recall the Weierstrass canonical form. such that image V equals the null space (
N ( E T ) ) of E T and image W equals the
Theorem 1 If (3) is regular, then there
exist nonsingular matrices X and Y such null space of E. Then
that E  [ E1 0][VW ]  E1V , where E1 is full
T T

I 0 
J 0  column rank. The next lemma
Y T EX   n  andY T
AX   
characterizes pencils which have an index
0 I r 
 0 N 
of at most one.
(4)
where J is a matrix in Jordan form whose Lemma 1 The following statements are
elements are the finite eigenvalues, equivalent:
I k  R k k is the identity matrix and N is a (i) pencil (4) is regular and has at most
nilpotent matrix also in Jordan form. J index one.
and N are unique up to permutation of  E 
(ii)rank   T    n  r
Jordan blocks.
 V A 
If (3) is regular the solutions of (DAE) ( rank ( E  VV T A)) .
take the form (iii) rank ([EAW])  n  r
x(t )  X1 z1 (t )  X 2 z2 (t ) ( rank(E  AWW )) .
T

where with X  [ X1 X 2 ] Y  [Y1T Y2T ] ,

Since we do not want to consider
X1 Y1T  R( nr )n , X 2  Y2T  R( n r )r and
derivatives of the input function in this
paper, we restrict the analysis to regular
t
z1(t)  eJt z1(0)   eJ (t s)Y1 f (s)ds z1(0)  [In 0]X 1x0 index one systems here. The above
0
discussion motivates then the next
k 1
di assumptions.
z2 (t) NiY2 f (t)
i 0 dti
Assumption 1 Throughout this section
under the consistency condition: the next assumptions are made w.r.t.
k 1
di system (1):
[0 I r ] X 1 x0   N iY2 i f (0) 1. matrix E is singular;
i 0 dt 2. det ( E  A)  0 ;
Here k is the degree of nilpotency of N . 3. rank ([ EAW ]  n  r .
That is the integer k for which N k  0
and N k 1  0 The index of the pencil (3) In the cooperative game, each player have
and of the descriptor system (DAE) is the a set of possible outcomes then one
degree k of nilpotency of N . If E is outcome (which in general does not
nonsingular, we define the index to be coincide with a player’s overall lowest
zero. cost) is cooperatively selected. It is
From the above formulae then the solution reasonable that the cooperative desicion is
x(t ) will not contain derivatives of the strategy that have the property that if
function f if and only if k  1 . In that different strategy is chosen, then at least
case the solution x(t ) is called impulse one of the players has higher costs. Or,
stated diferently, is solutions that the
free. In general, the solution x(t ) involves
players outcome can not be improved by
derivatives of order k 1 of the forcing all players simultaneously. This strategy is
function f if 3 has index k . called Pareto efficiency.

146
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Definition 1 Let U is the set of admissible x2 (t )  0 I r Y T ( B1u1 (t )  B2 u 2 (t ))

strategies. A set of strategies ˆ is called  Y2 ( B1u1 (t )  B 2 u 2 (t ))
Pereto efficient if the set of inequalities (7)
J i ( )  J i (ˆ ), i=1,2,...,N, where at least one
of the inequalities is strict, does not allow for Substitution of (7) into the cost functions
any solution   U . The corresponding point (6) shows that (u1 () u2 ()) are open-loop
( J 1 (ˆ),..., J N (ˆ ))   N Nash equilibrium actions for the game (1),
is called a Pareto solution. The set of all (2) if and only if (u1 () u2 ()) are open-
Pareto solutions is called Pareto frontier. loop Nash equilibrium actions for the
game
Usually there is more then one Pareto x1 (t )  Jx1 (t )  Y1 B1u1 (t )  Y2 B 2 u 2 (t ) ,
solution. It is noted that here we are not
x1 (0)  I n 0X 1 x 0
make definiteness assumptions for matrix
Qi . Here we follow the line in [5] which is (8)
derive Pareto solutions for linear quadratic
game with ordinary systems without with cost functionals J i (u1 u2 ) for the
definiteness assumptions for matrix Qi . players given by
While in [4] there is more simple discussion
about Pareto solutions of linear quadratic
game with definiteness assumptions for T I n 0 
matrix Qi .  1 1

0 { x (t ) u (t ) u (t )  0
T T T
2   B Y 
T
1 2
T

 0  B Y 
T
2 2
T

III. THE FINITE PLANNING HORIZON  x1 (t ) 

I 0 0  
Let we consider the game (1), (2) under X Qi X  n
T
  u1 (t ) 
0  Y2 B1  Y2 B2 
the assumption that T is finite. The game u (t )
 
(1), (2) has a set of open-loop Nash
equilibrium actions (u1 () u2 ()) if and  uiT (t ) Ri ui (t )dt
only if (u1 () u2 ()) are open-loop Nash
 x1 (t ) 
equilibrium actions for the game
 
T
  { x (t ) u (t ) u (t ) M i  u1 (t ) }dt
T
1
T
1
T
2

I n 0  x1 (t )   J 0   x1 (t ) 
0
u 2 (t )
  Y T B1u1 (t )
0
 0  x 2 (t )  0 I r   x 2 (t ) (9)
where

 Qi Vi Wi 
 x (0)   
 Y T B2 u 2 (t ),  1   X 1 x0 M i  ViT 

R1i Ni  
 x 2 (0)  
WiT


NiT R2i 
(5)
where player i has the quadratic cost (10)
functional:
with
  x (t)  
 
T
J i (u1 , u2 ,...,un )    x1T (t) x2T (t) Qi x(t) 1  
0 x2 (t) 
~
 u (t ) R u (t ) dt
(6)

From (5) it follows that

147
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Qi  X1T Qi X1Vi   X1T Qi X 2Y2 B1 Next we will consider conditions for the cost
functions J i in (1) are convex. From [5] we
Wi   X1T Qi X 2Y2 B2  Ni  B1T Y2T X 2T Qi X 2Y2 B2 
recall some preliminary result.
R11  B1T Y2T X 2T Q1X 2Y2 B1  R1
Lemma 4 Assume that U is convex. We will
R21  B2T Y2T X 2T Q1X 2Y2 B2  consider the linear quadratic cost function
R12  B1T Y2T X 2T Q 2X 2Y2 B1 T

J ( s, T , u , x0 )   { x T (t ) u T (t ) M  
 x(t )

R22  B Y X Q 2X 2Y2 B2  R 2
T T
2 2
T
2
0
u (t )


2 p1T (t )   x(t )
p 2T (t )    c 2 (t )}dt
IV. SOLUTION OF COOPERATIVE u (t )
GAME
(11)
In the following discussion, the set of Subject to the state dynamics
parameters, A, plays a crucial role. x (t )  Ax(t )  Bu(t )  c1 (t ), x(s)  x0 ,
(12)
N
A  {  ( 1 ,...,  N )  i  0 and  i  1} Where u, pi and ci aresuch that (11) and
i 1
Q V
. (12) have solution, M   . Let
The next two lemmas give a caracterization of V
T
R 
Pareto efficient solutions. The first lemma give x0   n . Then J (0, T , u, x0 ) is convex as a
identification for Pareto efficient whithout
convexity assumptions, while in second function of u if and only if J (0, T , v, x0 )  0
lemma states how to find Pareto efficient with for all v where
convexity assumptions. The proof of those
lemmas can be found in [4].
T

J (0, T , v, x0 )   z T (t ) v T (t ) M  
 z (t )
 dt
N 0
v(t ) 
Lemma 2 Let  i  (0,1) , with i 1
i  1. with z (t )  Az (t )  Bv (t ), z ( s )  0 .
(13)
Assume ˆ U such that
N
Note that in the second equivalence does not
ˆ  arg min{  i J i ( )} .
i 1
depend on x 0 . In particular it follows from
Then ˆ is Pareto efficient. that if J is convex for one x 0 then is convex
for all x 0 .
Lemma 3 Assume that the strategy space U is
convex. Moreover, assume that the payoffs J i Lemma 5 Assume that U is convex. Consider
are convex, i=1,...,N. Then, if ˆ is Pareto the linear quadratic cost function (11) and
(12). Then if J (0, T , u, x0 ) is for some x 0 ,
efficient, there exist   A , such that for all
 U , J (0, T , u, x0 ) is convex foar all s  0 and for
N N
all x 0 .
 J
i 1
i i ( )    i J i (ˆ ) .
i 1
Corollary 1 Consider the linear quadratic
N cost function (11) and (12). Then
Note that if J i are convex, then  J
i 1
i i ( ) J ( s, T , u, x0 ) is convex for all s  0 and x 0
is convex too for an arbitrary   A . In [5], if and only if J (0, T , u, x0 ) is convex for all
characterization for the space of admissible (an) x 0 . Which holds if and only if
control U is convex is given.

148
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

J (0, T , v,0)  0 for all v, where J is given by u  ( )  arg min  1 J 1   2 J 2 , subject to

uU
(13). (12).
From [5] we get the following theorem. The following theorem give procedure to
calculate the Pareto efficient solution of the
Theorem 1 Assume that either T is finite or game with descriptor system.
(A,B) is stabilizable. Then J (0, T , u, x0 ) is
convex for all (an) x 0 if and only if Theorem 2 Consider the cooperative game
(9), (2) with T   , U  L2 ,e , s and (A,B)
inf J (0, T , v,0) exists.
stabilizable. For   A let
Note 1 In [5] we can find theorem that
~ ~
consider the existence of the following cost Q V
function. M ( )   1 M 1   2 M 2   ~ T ~ ,
V R
T

J   { x T (t ) u T (t ) M 
 x(t )
 dt Where
0
u (t ) ~
Q   1Q1   2 Q2 ,
with x (t )  Ax(t )  Bu(t ), x(0)  x0 , ~
V  [1V1   2V2 , 1W1   2W2 ] , and
(14)
~ R N1  R N2 
The following Riccati equations plays R   1  11T    2  12T .
 N 1 R21   N 2 R22 
important role in the next discussion. ~ ~ 1
Furthermore let S  BR B . Assume that
T
K (t )  AT K (t )  K (t ) A
(15) below has a stabilizing solution X i for
 ( K (t ) B  V T ) R 1 ( B T K (t )  V )  Q, ;
 i  1 , i=1,2 respectively.
K (T  0 ~ ~ ~ ~
A T X  XA  ( XB  V ) R 1 ( B T X  V T )  Q  0
(DRE) .
0  AT K  KA  (KB  V T )R 1 (BT K  V )  Q (15)
Then the set of all cooperative Pareto
.
solutions is given by
(ARE)
{( J 1 (u  ( )), J 2 (u  ( )))   A} .
Let B  B1 B2 . Combining the results of Here
~ ~
Lemmas 2 and 3 and Theorem 1 and Note 1, u  (t )   R 1 ( B T X s  V T ) x (t )
we can derive existence and computational 
~
 R 1 B T  e  Acl (t  s ) X s c( s)ds
T
algorithms for both the finite and infinite
horizon game problems. We just state for t
infinite horizon problem. Where X s is the stabilizing solution of (15)
~ ~ ~
Corollary 2 Consider the coocperative game and, with Acl  A  BR 1V T  S X s ,
(1,2) with T   , U  L2 ,e , s and (A,B) the closed-loop system is
stabilizable. Assume (ARE), with x (t )  Acl x(t )  Sm(t )  c(t ), x(0)  x0 .
R Ni  In case c (.)  0 with these actions the
V  Vi Wi  and R   1Ti  has a
 N i R2 i  corresponding cost are
~ ~
stabilizing solution K i , for i=1,2 respectively. J i ( x 0 , u  )  x 0T M i x 0 , where M i is the
Then all Pareto efficient solutions are unique solution of the Lyapunov equation.
obtained by determining for all   A ,

149
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

~ ~

AclT M i  M i Acl   I
~ ~
 ( X s B  V }R 1  [6] Engwerda, J.C., “Necessary and Sufficient
Conditions for Solving Cooperative
 I  . Differential Games”, CentER Discussion
M i  ~ 1 T ~T 
 R ( B X s  V ) Paper, Tilburg University, The
Netherlands, No 2007-42, 2007.
[7] Engwerda and Salmah, “The Open-Loop
V. CONCLUSIONS Linear Quadratic Differential Game for
In this paper we consider the solution of index one Descriptor Systems”,
cooperative linear quadratic game for Automatica, Vol. 45, Number 2, 2009.
descriptor system. Using the theory of convex [8] Katayama, T., and Minamino, K., “Linear
analysis we can get the result for finite and Quadratic Regular and Spectral
infinite horizon game. Basically, the result Factorization for Continuous Time
were obtained by reformulating the game as an Descriptor Systems”, Proceedings of the
ordinary linear quadratic differential game. 31st Conference on decision and Control,
Following the lines and combining the results Tucson, Arizona, 1992, pp 967-972.
of [5] and [7], similar conclusions can be
[9] Lewis, F.L., “A survey of Linear Singular
derived. Systems”, Circuits System Signal Process,
The above result cn be generalized vol.5, no.1, 1986, pp 3-36.
straightforwardly for N-player game.
[10] Minamino, K., “The Linear Quadratic
REFFERENCES Optimal Regulator and Spectral
Factorization for Descriptor Systems”,
[1] Basar, T., and Olsder, G.J., Dynamic Master Thesis, Department of Applied
Noncooperative Game Theory, second Mathematics and Physics, Faculty of
Edition, Academic Press, London, San Engineering, Kyoto University, 1992.
Diego, 1995.
[11] Salmah, Bambang, S., Nababan, S.M., and
[2] Dai, L., Singular Control Systems, Wahyuni, S., “Non-Zero-Sum Linear
Springer Verlag, Berlin, 1989. Quadratic Dynamic Game with Descriptor
[3] Engwerda J., “On the Open-loop Nash Systems”, Proceeding Asian Control
Equilibrium in LQ-games”, Journal of Conference, Singapore, 2002, pp 1602-
Economic Dynamics and Control, Vol. 22, 1607,.
1998, pp 729-762. [12] Salmah, “N-Player Linear Quadratic
[4] Engwerda, J.C., LQ Dynamic Dynamic Game for Descriptor System”,
Optimization and Differential Games Proceeding of International Conference
Chichester: John Wiley & Sons, 2005, pp Mathematics and its Applications SEAMS-
229-260. GMU, Gadjah Mada University,
[5] Engwerda, J.C., “A Note on Cooperative Yogyakarta, Indonesia, 2007.
Linear Quadratic Control”, CentER
Discussion Paper, Tilburg University, The
Netherlands, No 2007-15, 2007.

150
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

ARMA Model Identification using Genetic Algorithm

(An Application to Arc Tube Low Power Demand Data)

Irhamah(1), Dedy Dwi Prastyo(2)and M. Nasrul Rohman(3)

(1), (2)
Department of Statistics, Institut Teknologi Sepuluh Nopember, Surabaya, East Java, Indonesia
Email: irhamah@statistika.its.ac.id, dedy-dp@statistika.its.ac.id

ABSTRACT usually involve subjective judgement, which

The most crucial steps in building ARIMA time results many models in identification step [3] .
series models are to identify and build model based In the last decade, artificial
on available data. One of the common intelligence concepts (fuzzy system, neural
identification methods is correlogram, but when the network, genetic algorithm) have been
data sets have mixed ARMA effect, correlogram
proposed as forecasting tools. As known, GAs
cannot provide clear lags to identify. Therefore, in
this paper we study the use of Genetic Algorithm are an effective search and optimisation
(GA), an effective search and Optimisation method that simulates the process of natural
method, as identification method of ARMA selection or survival of the fittest. The GAs are
models. The best model from GA is compared to population to population approach, can escape
the best model from correlogram. Their application from local optima and are very effective in
in Arc Tube low power demand data show that GA global search. GA can also handle any kind of
is better in terms of smaller MSE. objective function and constraints. Reference
[8] studied about how to build ARMA model
KEYWORDS using GA. Reference [9] applied GA as
Time series; ARIMA; Identification; Correlogram; alternative method to identify ARIMA models.
Genetic Algorithm; MSE.
In this paper, we study the use of
Genetic Algorithm (GA) as identification
I. INTRODUCTION
method of ARMA models. The best ARMA
A time series is a time ordered
model from GA is compared to the best
sequence of observations. Time series occurs
ARMA model from Correlogram
in a variety of fields: agriculture, business and
identification method. The comparative study
economics, engineering, geophysics, medical
is applied on the bi-weekly Arc Tube low
studies, meteorology, etc. One of the most
power demand data [11] from January to
popular method in time series is
November 2005 based on RMSE-in sample
Autoregressive Integrated Moving Average
and MAPE-out of sample. Softwares used in
(ARIMA) developed by Box and Jenkins [1].
this study are Minitab and Matlab. GA toolbox
The stages of building ARIMA models consist
developed by [2] is available in Matlab.
of Model Identification, Model Estimation,
Diagnostic Checking and Forecasting. The
II. MIXED AUTOREGRESSIVE
most crucial steps are to identify and build a
MOVING AVERAGE (ARMA)
model based on the available data.
MODELS
One of the commonly used
A stationary and invertible process can
identification methods is correlogram which
be represented either in a MA form or in an
uses sample autocorrelation function (ACF)
AR form. A problem with either representation
and sample partial autocorrelation function
is that it may contain too many parameters,
(PACF). The goal is to match patterns in the
even for a finite-order MA and a finite-order
sample ACF and sample PACF with the
AR model because a high-order model is often
known patterns of the theoretical ACF and
needed for good approximation. A large
PACF of ARIMA models. However, when the
number of parameters reduces efficiency in
time series data sets have mixed ARMA
estimation [12]. Thus, in model building, it
effect, the plots cannot provide clear lags to
may be necessary to include both AR and MA
identify. The orders of a mixed ARMA model
simultaneously as ARMA model:

151
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

 p ( B ) Z t   q ( B ) at ... (5)
(1)
The equation is clearly nonlinear in
where the parameter. Hence, for a general ARMA
 p ( B )  1  1 B  ...   p B p (2) model, a nonlinear least squares estimation
and procedure must be used to obtain estimates.
We can use conditional least squares (CLS)
q ( B )  1  1 B  ...   q B q (3) procedure to find the least squares estimates
that minimize the error sum of squares:
A mixed process of considerable practical n

importance is ARMA(1,1) (when p = q =1). S (  ,  ,  )   at2 (  ,  ,  , Z , a , Z ) (6)

 ~ ~ ~ ~ init ~ init ~
t 1

III. ARMA MODEL IDENTIFICATION where Z init and ainit are initial values. One of
AND MODEL ESTIMATION the algorithms commonly used to achieve a
Model identification refers to the proper and faster convergence in nonlinear
methodology in identifying the required estimation is Levenberg-Marquardt.
transformations, such as variance stabilising
transformations and differencing V. GENETIC ALGORITHM FOR ARMA
tranformation. The decision is to find proper IDENTIFICATION
orders of p and q . The procedure in Genetic Algorithm (GA) is a
correlogram identification method is to match population-based search method. GAs
patterns in the sample ACF and sample PACF categorized as a global search heuristics,
with the known patterns of the theoretical ACF classed under Evolutionary Algorithm that use
and PACF of ARIMA models. Table 1 technique inspired by evolutionary biology:
summarizes the important results for selecting principles of natural selection and genetics.
p and q [12]. Reference [6] developed the idea and concepts
behind the GA and many authors have refined
Table 1. Characteristics of theoritical ACF and PACF for his initial approach [10]. The researchers who
stationary processes used GAs to solve complex real-world
Process ACF PACF problems report good results from these
AR( p ) Tails off as Cuts off after techniques, for example the study of [7].
exponential decay lag p
or damped sine
Solutions to GA are represented by
wave data structures, often a fixed length vector,
MA( q ) Cuts off after lag Tails off as called chromosomes. Chromosomes are
q exponential composed of genes which have value called
decay or alleles, and the position of the gene in the
damped sine
wave chromosome is called locus [5].
ARMA( p , q Tails off after lag Tails off after Chromosomes have been represented by
) (q- p) lag (p- q) vectors, strings, arrays, and tables. Alleles
have been represented by a simple binary
After identifying a tentative model, system of zeroes and ones, integers, real
the next step is to estimate the parameters in number, and letter of alphabet, or other
the model. Consider a simple ARMA(1,1) symbols [4]. Each chromosome has a fitness
model value which is indicative of the suitability of
  the chromosome as a solution to the problem
Z t  1 Z t 1  at  1at 1 [5].
(4)
In this research, the steps of GA for
To calculate at , we note that
ARMA model identification are:
 

at  Z t  1 Z t 1  1at 1 Step 0. [Define] Define operator settings of

GA suitable with the problem. This
   

 Z t  1 Z t 1  1 ( Z t 1  1 Z t  2  1at  2 ) GA uses binary encodings that

represent orders of p and q . For
  

 Z t  (1  1 ) Z t 1  11 Z t  2  12 at  2 )

152
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

example, chromosome (10010,11010)

represents ARMA([1,4];2,[4] ).
Step 1. [Initialization] Create an initial
population P of PopSize Box-Cox Plot of data
Low er C L Upper C L

chromosomes. 160000
Lambda
(using 95,0% confidence)
Estimate 0,79974
140000
Step 2. [Fitness] Evaluate the fitness f(Ci) of 120000
Low er C L
Upper C L
0,56870
1,06275

each chromosome Ci in the 100000

Best Value 1,00000

StDev
population. The fitness is 1/MSE. 80000
High fitness denoted by low MSE. 60000

Step 3. [Selection] Apply Roulette Wheel 40000

Selection to produce Mating 20000

Limit

Population M with size PopSize. -2 -1 0 1 2

Lambda
3 4 5

Step 4. [Crossover] Pair all the chromosomes

Figure 2. Box-Cox transformation analysis of data
in M at random forming PopSize /2
pairs. Apply crossover with
The indication of nonstationary in the men,
probability pc to each pair and form
suggesting differencing is needed. The sample
PopSize chromosomes of offspring.
ACF and PACF of differenced data are shown
Step 5. [Mutation] With a probability of
in Figures 3-4. These plots indicate some
mutation pm, mutate the offspring.
tentative models: ARIMA (1,1,0), ARIMA
Step 6. [Replace] Evaluate fitness of new
(2,1,0), ARIMA (0,1,1), dan ARIMA (0,1,2).
offspring. Replace the old population
with newly generated population.
Step 7. [Test] If the stopping criterion is met Autocorrelation Function for diff
(with 5% significance limits for the autocorrelations)

(no further improvement in fitness) 1.0

0.8
then stop and return to the best 0.6

solution in current population, else go 0.4

Autocorrelation

0.2
to Step 2. 0.0
-0.2
-0.4

VI. RESULTS -0.6

-0.8
The data analysis is divided into two -1.0

major parts: building ARMA model using 1 5 10 15

Lag
20 25 30

correlogram and GA as identification methods.

Figure 3. Sample ACF of differenced data
6.1. MODEL FROM CORRELOGRAM Partial Autocorrelation Function for diff
(with 5% significance limits for the partial autocorrelations)
The data plot in Figure 1 suggests that 1.0
the series is possibly nonstationary in mean 0.8

and variance. 0.6

Partial Autocorrelation

0.4
0.2
0.0
Time Series Plot of Data Asli
-0.2
-0.4
200000
-0.6
-0.8
150000 -1.0
Data Asli

1 5 10 15 20 25 30
100000 Lag

50000 Figure 4. Sample PACF of differenced data

After identifying tentative models, the

0
1 14 28 42 56 70 84 98 112 126
Index
next step is to estimate and test the parameters.
Figure 1. Time series plot of data
Parameter estimation and test that summarized
The Box-Cox analysis as shown in Figure 2 in Table 2, show that both second-order of AR
indicates that no transformation is needed and and MA are not significant. This step suggests
the series is stationary in variance, since the two simple models: ARIMA(1,1,0) and
optimal lambda is 1. ARIMA(0,1,1). We check the adequacy of the
model: white noise and normal residual. From

153
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Table 3, it is only ARIMA(1,1,0) that satisfy proceed with modeling the residual using
white noise assumption. From normality test, ARMA. The steps of building ARMA model
this model also has normal residual (p-value for residual follow Box-Jenkins’s ARIMA
from Kolmogorov-Smirnov normality test: modeling as mentioned before: identification,
0.081). parameter estimation and testing, and
adequacy checking. These suggest
Tabel 2. Parameter Estimation Testing ARMA(5,4) model to represents the residual.
Parameter Coeff p-value The final model that satisfies best
ARIMA(1,1,0) 1 -0.5501 0 model selection criteria is:

ARIMA(2,1,0)
1 -0.5948 0 zt  zt-1  0.548 zt -1  0.439 zt-3  0.007 zt-5 
2 -0.0915 0.262 at  0.236 at -2  0.793 at -3 
ARIMA(0,1,1) 1 0.594 0 0.533 at-4  0.362 at-5  X t
1 0.6242 0 (9
ARIMA(0,1,2) )
2 -0.1198 0.164
where
Tabel 3. White-Noise and normality test X t  1.93 et -1  1.612 et-2  0.965 et-3  0.492 et-4 
white (10)
Ljung - Box noise/ - 0.201et-5  rt  1.99 rt 1  1.64 rt  2  0.356 rt 3
normal?  0.192 rt  4
ARIM lag 12 24 36
Yes/
e represents AR process while r represents
A P_Va MA process of residual. The final model
0.441 0.348 0.558 Yes
(1,1,0) l satisfies white noise and normal residual
ARIM lag 12 24 36 assumption.
No/
A P_Va
0.026 0.001 0.001 -
(0,1,1) l Tabel 4. GA’s Model Identification Results
N 10 20
G 15 43
Thus the final model from identification using
correlogram is ARIMA(1,1,0)    
   
-0.527 0 0 0
Parameter

zt  zt-1  0.5501zt-1  at (7) -0.205 0 0 -0.065

0.202 0.624 -0.049 0.237
-0.135 -0.514 -0.631 -0.948
6.2. MODEL FROM GA 0.083 0.565 0 0.306
The second part of data analysis is the MSE 7.90E+08 9.64E+08
application of GA as identification method. N 40 40
GA is applied to differenced data. The analysis G 83 9
is conducted on different number of    
chromosomes: 10, 20, 40 and 100. The    
-0.544 0 -0.536 0
Parameter

number of generation when the process is

0 0.227 0 0.225
stopped, parameter estimates and MSE for 0.434 0.778 0.427 0.773
each N are reported in Table 4. The best model -0.012 -0.523 -0.023 -0.522
based on minimum MSE is ARMA([1,3,5] , [ 0.003 0.374 0 0.378
2,3,4,5]): MSE 7.87E+08 7.78E+08
N 100 100
zt  zt-1  0.548 zt -1  0.439 zt-3  0.007 zt-5  G 79 40
at  0.236 at -2  0.793 at -3  (8)    
   
0.533 at-4  0.362 at-5 -0.556 0 -0.544 0
Parameter

0 0.239 0 0.227
0.44 0.792 0.434 0.778
from combination of N = 100 and G = 67. 0 -0.536 -0.012 -0.523
Similar with the steps in 6.1, the best 0 0.362 0.003 0.374
model from GA is tested against white noise MSE 7.76E+08 7.87E+08
and normality assumption. The model satisfies N 100
normal residual assumption (p-value > 0.15) G 67
but it is not a white noise process. Then we

154
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

  Optimization. Canada: John Wiley & Son

  Inc.
-0.548 0

Parameter
0 0.236 [5] Goldberg, D. E. 1989. Genetic
0.439 0.793 Algorithms in Search, Optimization and
0 -0.533 Machine Learning. New York, USA:
0.007 0.362 Addison-Wesley.
MSE 7.73E+08 [6] Holland J.M. 1975. Adaptation in
N = number of chromosome Natural and Artificial Systems. Ann
G = number of generation when convergence is Arbor: The University of Michigan
reached
Press.
[7] Irhamah and Ismail, Z. 2009. A Breeder
VII. THE COMPARISON Genetic Algorithm for Vehicle Routing
The best model from correlogram is Problem with Stochastic Demands.
compared to the best model from GA based on Journal of Applied Sciences Research,
RMSE in-sample and MAPE out of sample. 5(11), 1998-2005.
The results are summarized in Table 5. [8] Minerva, T. and Poli, I. 2001. Building
ARMA model with Genetic Algorithms.
Lecture Notes in Computer Science,
Table 5. RMSE and MAPE from Correlogram and GA 2037, 335-342
Correlogram GA [9] Ong, C.S., Huang, J.J. and Tzeng G.H.
RMSE 34.286,086 25.243,370 2005. Model identification of ARIMA
family using genetic algorithms. Journal
MAPE 12,47%. 10,657%. Applied Mathematics and Computation,
VIII. CONCLUSION 164, 885-912
ARMA identification method using [10] Ortiz, F., Simpson, J. R., Pignatiello, J.
GA produces smaller RMSE in-sample and J., and Langner, A. H. 2004. A Genetic
MAPE out-of-sample, compared to Algorithm approach to Multiple-
Correlogram identification method. This study Response Optimization. Journal of
has limitation on the use of MSE as single Quality Technology. 36 (4).
criteria in fitness function. Further researches [11] Sudarsono, I. 2008. Peramalan Arc Tube
can be done by including other criterias on the Dalam Pembuatan Lampu Light Capsule
fitness function. Super (Lcs) Dengan Metode Arima Box-
Jenkins Di Pt. Panasonic Lighting
ACKNOWLEDGEMENTS Indonesia. Final project: Institut
The authors are very grateful to Penelitian Teknologi Sepuluh Nopember
Produktif ITS Tahun Anggaran 2010 and [12] Wei, W.W.S. 2006. Time Series
Institut Teknologi Sepuluh Nopember who Univariate and Multivariate Methods,
kindly supported this research funding. Second Edition. Canada: Addison
Wesley Publishing Company, Inc.
REFERENCES
[1] Box, G.E.P., Jenkins, G.M. and Reissel,
G.C. 1994. Time Series Analysis
Forecasting and Control, 3rd Edition.
Englewood Cliffs : Prentice Hall.
[2] Chipperfield, A.J. and Fleming, P.J.
1995. The MATLAB Genetic Algorithm
Toolbox. Applied Control Techniques
Using MATLAB, IEE Colloquium, 101-
104.
[3] Cryer, J.D. and Chan, Kung-Sik. 2008.
Time Series Analysis, with Applicationa
in R, Second Edition. Springer.
[4] Gen, M. and Cheng, R. 2000. Genetic
Algorithms and Engineering

155
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

A Particle Swarm Optimization for Employee Placement Problems in the

Competency Based Human Resource Management System

Joko Siswanto(1) and The Jin Ai(2)

(1)
Industrial Management Research Group, Industrial Technology Faculty,
Institut Teknologi Bandung, INDONESIA
(Email: j.siswanto@ti.itb.ac.id)
(2)
Department of Industrial Engineering, Universitas Atma Jaya Yogyakarta, INDONESIA
(Email: jinai@mail.uajy.ac.id)

ABSTRACT Optimization (PSO) to find best methods for

This paper discussed on how particle swarm these employee placement problems.
optimization could be applied for solving the
employee placement problems in the competency PSO is a population based search method
based human resource management. The employee proposed by Kennedy and Eberhart [2], which
placement problems are the problems to were motivated by the group organism
simultaneously place many people to many jobs in
behavior such as bee swarm, fish school, and
an organization. After the particle swarm
mechanism to solve the problem is defined and bird flock. PSO imitated the physical
explained, simple case study is presented to movements of the individuals in the swarm as
illustrate the capability of the proposed method. a searching method, altogether with its
cognitive and social behavior as local and
KEYWORDS global exploration abilities. One of PSO
Employee placement problem; Human resource advantage is its simplicity of its iteration step
management; Competency based human resource which only consists of updating two set of
management system; Particle swarm optimization; equations. PSO is widely used as a solution
Evolutionary computing. methodology for solving numerous
combinatorial optimization problem such as
I. INTRODUCTION job shop scheduling [3], vehicle routing [4],
Competency Based Human Resource and project scheduling [5].
Management (CBHRM) System is a
standardized process to manage people Due to its simplicity and unexplored potential
optimally in organization from recruitment, in the HRM area, this paper will discuss on
selection, placement up to termination process how particle swarm optimization could be
based on job competency profiles and applied for solving the employee placement
individual competencies in order to achieve problems in the CBHRM system. Specifically,
organization goals, missions and vision [1]. it will describe on how the solution of the
One of CBHRM system function is placement. problem, which is the placement of the
It is a process to put the right persons at the employees, could be represented as a multi-
right places at the right time which is very dimensional particle. Also, the decoding
critical for the success of any modern method for translating particle into employee
organizations. placement is also explained.
Usually, a placement problem involves a multi Simple case study will be presented at the end
criteria decision making process. At a simplest of this paper to illustrate the capability of the
case, an employee can be rotated or promoted proposed particle swarm optimization
to a certain job within an organization one by algorithm for solving the employee placement
one sequentially based on a set of criteria of problem.
past performance, current competencies and The advantages and disadvantages of this
future expectations. But sometimes, in more algorithm will also be discussed further,
complex problem, organization needs to place altogether with its opportunity for
many people to many jobs, even for the whole improvement and extension.
organization, simultaneously. This paper will
demonstrate the application of Particle Swarm

156
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

II. EMPLOYEE PLACEMENT B. Mathematical Formulation

PROBLEM The EPP can be formulated as the following
A. Problem Definition integer programming problem:
The employee placement problem (EPP) in an
organization can be defined as the problem to

Maximize Z    i   ij   ij  xij 
m n

place many employees to many jobs (1)

i 1 j 1
simultaneously based on a set of criteria of
past performance, current competencies and
future expectations. Subject to
m

Regarding to the competencies criterion, the  xij  1 , for j (2)

i 1
employees’ competencies should be aligned n
with the job competency profile. The job  xij  1 , for i (3)
competency profiles provide a list of j 1

competencies and the minimal scores on those xij  0 , for employee i that is not qualified
competencies required to hold the jobs, while
to hold job j (4)
the employees’ competencies are the
quantitative score of each employee on those xij 0,1 , for i , j (5)
competencies. The minimal score on a
competency is the quantification of capability where:
required on the competency. Therefore, m : number of potential employees
employee with lower scores than minimal n : number of available jobs
required scores of a certain job position is not xij : binary assignment variable, xij = 1 if
qualified to hold that job. [6] For the employee i assigned to job j, xij = 0
placement criteria, the closeness among otherwise
employee’s competencies and job competency i : index of employee, i = 1 … m
profiles is the measurement basis of the j : index of job, j = 1 … n
competency performance score of an αi : past performance score of employee i
employee on a particular job. βij : current competency performance score
of employee i on job j
The generic EPP could be defined as the γij : future expectation score of employee i
problem to place a set of employee consisting on job j
of m potential people into a set of n available
jobs in order to maximize the total weighted The objective function in Eq. 1 is showing that
score of the criteria, subject to the required the higher the past performance of an
competencies. In this generic definition, it is employee, the bigger chance the employee
assumed that a job can be filled at most by a being placed in any jobs. Also, it is implied
single employee. that the employee placement tends to place an
employee in a job that is maximizing the
Each criterion may comprised of many sub- current competency performance and the
criteria, that is could be defined in hierarchical future expectation scores in all jobs.
form. For example the competency criterion
can be divided into major competency, Eq. 2 states that all job to be fulfilled by at
supporting competency, and field competency; most one employee. Whenever no employee is
whereas the major competency is comprised of qualified to hold a job, it is not necessary to
five sub-competencies. Using a proper place any employee to that job. Eq. 3 shows
methodology, such as the analytic hierarchy that one employee is placed at most to one job.
process (AHP), the weights of each In the Eq. 4, the binary assignment variables
competency and sub-competency can be are limited by employees’ qualification on the
determined. In the mathematical formulation available jobs.
defined below, these weights are utilized for The variables domain is defined in the Eq. 5.
obtaining the total weighted score of the In nature, the mathematical formulation of
criteria as the objective function of the EPP has m·n binary variables. Therefore, if the
decision problem. EPP is being solved using total enumeration

157
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

technique, it has 2mn alternative solutions that representing a solution of the problem. The
should be evaluated. ability of a particle to search for solution is
represented by its velocity vector (  l ) which
III. DATA PREPROCESSING drives particle movement. In the PSO iteration
There are three steps of preprocessing the step, every particle moves from one position to
human resource data into parameters required the next based on its velocity. Moving from
in the EPP: job and employee sets definition, one position to another, a particle is evaluating
CBHRM data extraction, and criteria different prospective solution of the problem.
evaluation. The basic particle movement equation is
presented below:
A. Job and Employee Sets Definition
In the first step, the number of available jobs is lh   1  lh    lh   1 (6)
identified. For the job with many available
where:
positions, every position is defined as different
lh   1 : Position of the l th particle at the
job so that each job can be fulfill only by
hth dimension in the   1
th

single employee. The job ID is assigned based

iteration
on its importance or rank, i.e. the first job
lh   : Position of the l th particle at the
(j=1) is the most important job or the highest-
hth dimension in the  th
rank job and the last job (j=n) is the least
iteration
important job or the lowest-rank job.
lh   1 : Velocity of the l th particle at the
hth dimension in the   1
th

After the job set is defined, the employee set is

iteration
defined, i.e. by listing the candidates that are
possible to be placed at least one job in the job
PSO also imitated swarm’s cognitive and
set.
social behavior as local and global search
abilities. In the basic version of PSO, the
B. CBHRM Data Extraction
particle’s personal best position (  l ) and the
In this step, the CBHRM data is extracted to
global best position (  g ) are always updated
find the job competency profiles for each job
and maintained. The personal best position of
in the job set, past performances of each
a particle, which expresses the cognitive
employee in the employee set, current
behavior, is defined as the position that gives
competency performances of employees, and
the best objective function among the
future expectations of employee placed into
positions that have been visited by the particle.
particular job.
Once a particle reaches a position that has a
better objective function than the previous best
C. Criteria Evaluation
objective function for this particle, i.e.
Using the AHP method, the particular
Z  l   Z   l  , the personal best position is
CBHRM data is being processed into the score
updated. The global best position, which
criteria: the past performance score of
expresses the social behavior, is the position
employee i (αi), the current competency
that gives the best objective function among
performance score of employee i on job j (βij),
the positions that have been visited by all
and the future expectation score of employee i
particles in the swarm. Once a particle reaches
on job j (γij). At the end of this step, all
a position that has a better objective function
parameters required in the EPP are available
than the previous best objective function for
so that the EPP is ready to be solved.
 
whole swarm, i.e. Z   l   Z  g , the global
best position is also updated.
IV. PSO METHOD FOR SOLVING EPP
A. PSO Algorithm [7]
The personal best and global best position are
As mentioned before, PSO is a population
used for updating particle velocity. In each
based search method that imitated the physical
iteration step, the velocity  is updated based
movements of the individuals in the swarm as
on three terms: inertia, cognitive learning and
a searching method. In PSO, a swarm of L
social learning terms. The inertia term forces
particles served as searching agent for a
particle to move in the same direction as
specific problem solution. A particle’s position
previous iteration. This term is calculated as a
(  l ), which consists of H dimensions, is
product of current velocity with an inertia

158
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

weight ( w ). The cognitive term forces particle  Move each particle based on Eq. 6, after
to go back to its personal best position. This updating particle velocity based on Eq. 7.
term is calculated as a product of a random Step 3: Termination
number ( u ), personal best acceleration  If the terminating criterion is reached, i.e.
constant ( c p ), and the difference between   T , the stop the iteration. The solution
personal best position  l and current position corresponding with the last global best
 l . The social term forces particle to move to position is the best solution found by this
the global best position. This term is algorithm.
calculated as a product of a random number (  Otherwise, set the iteration counter
u ), global best acceleration constant ( c g ),    1 , and back to Step 2.
and the difference between global best
position  g and current position  l . To be B. Solution Representation
more specific, the velocity updating equation In the PSO, a problem specific solution is
is expressed as follow: represented by position of particle in multi-
dimensional space. The proposed solution
lh   1  wlh    c p u  lh  lh    representation of EPP with m employees and n
(7) jobs is a m dimensional particle. Each particle
cg u  gh  lh    dimension is encoded as a real number. These
where: m dimensions are related to employees, in
lh   : Velocity of the l th particle at the which each employee is represented by one
hth dimension in the  th dimension. The position value in each particle
iteration dimension will be represented the priority
 lh : Personal best position of the l th weight of its corresponding employee to be
particle at the hth dimension in placed into jobs in the decoding steps.
the  th iteration
 gh : Global Personal best position at
the hth dimension in the  th
iteration

In the velocity-updating formula, random

numbers is incorporated in order to randomize
particle movement. Hence, two different
particles may move to different position in the
subsequent iteration even though they have
similar position, personal best, and global best.

Algorithm 1: Basic PSO Algorithm Figure 1. A Solution Representation of EPP (m=5)

Step 1: Initialization
 Set the PSO parameters: T , L , w , c p , C. Decoding Method
cg . The decoding method is required to transform
 Set the iteration counter,   1. a particle (represented by its position) into a
 Generate L particles with random initial problem specific solution, which is the
position (  l ) and zero velocity ( l  0 ). placement of employees into jobs in the EPP.
 Set the initial personal best position the
same as its position (  l  l ). As mentioned before, the first step in the
Step 2: Iteration – Particles Movement decoding method is the extraction of employee
 Decode each particle into a problem priority weight from the position value. Each
specific solution and evaluate the employee will be given a priority weight from
objective function of the solution. Set the its corresponding particle dimension. For
objective function value as the fitness example, the particle depicted on Fig. 1 can be
value of the particle Z  l  . transformed into following priority weight for
 Update the personal best position of each five employees: 1.075, 0.344, 3.150, 4.593,
particle, set  l  l if Z  l   Z   l  . 2.728.
 Update the global best position, set It is defined that the priority of an employee to
 
 g   l if Z   l   Z  g . be placed into jobs is correspond to its priority

159
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

weight. Therefore, employee with higher V. CASE STUDY

priority weight will be given more priority A simple case study is conducted to illustrate
than employee with lower priority weight. So, the capability of the proposed particle swarm
continuing the example, the fourth employee optimization algorithm for solving the
will be given the first priority and finally the employee placement problem. The case
second employee will be given the last comprises of a problem to place five
priority. The complete information related to employees into three available jobs.
employee priority could be kept into a list Hypothetical CBHRM data is used here, which
illustrated in Table 1. consists of job competency profiles, past
performances of each employee, current
Table 1. An Employee Priority List competency performances of employees and
Employee ID Priority Weight Priority Rank future expectations of employee placed into
4 4.593 1 particular job.
3 3.150 2
5 2.728 3 To test the performance of the proposed PSO,
1 1.075 4 the algorithm is coded into computer program
2 0.344 5 using C# language. PSO parameters used to
solve this case are: number of particle L = 30,
After the employee priority list is created, the number of iteration T = 200, decreasing inertia
placement of employee into job is performed. weight w from 0.9 to 0.4, personal best
One by one each employee in the employee acceleration constant cp = 2, and global best
priority list, starting from the first rank, is acceleration constant cg = 2. Since the PSO
placed into a job considering the rank of job, has random property, five replication of the
availability of job, and employee’s algorithm is run.
qualification. An employee will be placed at
the highest rank job that is matched with For comparison purpose, total enumeration of
employee qualification and has not assigned to possible solutions is performed. All possible
other employee yet. It is possible to have a solutions are evaluated, so that the best
situation where is no more available job for an employee placement can be determined.
employee. Finally, the total employee Among the five PSO replications performed,
placement could be conducted and the result three replications provide the same result as
could be displayed as illustrated in Table 2. the best employee placement and the other
replications provide a solution which its
Table 2. An Employee Placement objective function is very close with the
Employee ID Job ID objective function of the best employee
4 2 placement.
3 –
VI. CONCLUSIONS AND FURTHER
5 1
WORKS
1 3
2 –
The simple case study above shows that the
proposed solution representation and decoding
It is implied from the example illustrated in method are effective for solving the EPP using
Table 2, that the fourth employee is not basic PSO. The effectiveness of this method is
qualified for the first job so that this employee still need to be confirmed using larger sized
is assigned to the second job. Also, the third and real-world problem.
employee is qualified only for the second job.
Since the second job is already assigned to the It is noted that the result on this paper is
fourth employee, this employee could not be gained by pure PSO algorithm. Hence, it is
placed at any job. The fifth employee is possible to improve the result by more
qualified for the first job and the first sophisticated PSO variants and features. Also,
employee is met the qualification of the third it is possible to hybridize this PSO with other
job. Therefore, no more job available for the technique, i.e. local search method. It is also
second employee. possible to improve the performance of the

160
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

proposed algorithm by parameter optimization

and programming implementation.

Integrating this EPP solving module inside the

CBHRM system, including automated data
extraction, is the ultimate direction of this
research.

REFERENCES
[1]. Siswanto, J (2007). "Integrated
Competency Based Human Resource
Management System: Implemented Model
in Indonesian Crown Corporations,"
Proceedings of the 8th Asia Pacific
Industrial Engineering and Management
Systems (APIEMS 2007) Conference,
Kaoshiung, Taiwan.
[2]. Kennedy, J and Eberhart, R (1995).
"Particle Swarm Optimization,"
Proceedings of IEEE International
Conference on Neural Networks, Vol 4, pp
1942-1948.
[3]. Pongchairerks, P and Kachitvichyanukul,
V (2009). "A two-level Particle Swarm
Optimisation algorithm on Job-Shop
Scheduling Problems," International
Journal of Operational Research, Vol 4,
No 2, pp 390-411.
[4]. Ai, TJ and Kachitvichyanukul, V (2009).
"Particle swarm optimization and two
solution representations for solving the
capacitated vehicle routing problem,"
Computers and Industrial Engineering,
Vol 56, No 1, pp 380-387.
[5]. Zhang, H, Li, H and Tama, CM (2006).
"Particle swarm optimization for resource-
constrained project scheduling,"
International Journal of Project
Management, Vol 24, No 1, pp 83-92.
[6]. Siswanto, J, Cahyono, E and Wangi, S
(2009). "An optimal method of job
employee matching in corporate
organization," Proceedings of the 4th
International Conference on Mathematics
and Statistics (ICoMS 2009), Bandar
Lampung, Indonesia.
[7]. Ai, TJ (2008). "Particle swarm
optimization for generalized vehicle
routing problem," Dissertation, Asian
Institute of Technology, Thailand

161
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Measuring Similarity between Wavelet Function and Transient in a Signal

with Symmetric Distance Coefficient

Nemuel Daniel Pah(1)

(1)
Department Electrical Engineering, Universitas Surabaya, Surabaya, Indonesia
(Email: nemuelpah@ubaya.ac.id)

ABSTRACT flexibility comes with a need of methods to

Wavelet transform has been developed and applied select wavelet function.
in various areas of science and engineering. It Common wavelet selection methods were
offers better signal processing capabilities due to
based on the matching between the properties
the existence of a large number of wavelet
functions. The superiority comes with a of wavelet function and the properties of the
requirement of methods to select a proper wavelet analyzed signal [1-8]. The methods require
function for each signal or application. Selection both, the wavelet function and the analyzed
methods based on signal properties are not always signal, to be defined mathematically. All
applicable due to the lack of mathematical wavelet functions are defined mathematically,
definition of most analyzed signals. Wavelet but, most analyzed signals are natural signals
function can be selected based on its shape
similarity to interested transient in input signal. The
with lack of mathematical expression. Other
selection method is subjective since there is no wavelet function selection method was based
quantified parameter to measure this similarity. on visual shape similarity between interested
This paper introduces a new parameter, symmetric transient in input signal and wavelet function
distance coefficient (SDC) to measure similarity [9-14]. The method, was not accurate since the
between wavelet function and transient in a signal. similarity was only judged by visual
It is based on a fact that wavelet coefficients of a
inspection.
transient that has similar shape and similar time
support to a wavelet function are always This paper introduces a new parameter to
symmetric. The parameter measures similarity by measure similarity between wavelet function
measuring the degree of symmetry in wavelet and transient in a signal by measuring the
coefficients. The paper also reports an experiment degree of symmetry in wavelet coefficients.
to demonstrate procedures to measure this The new parameter is called ‘symmetric
similarity using SDC. distance coefficient’ (SDC) [15]. This paper is
organized into five sections. Following this
KEYWORDS
Wavelet; symmetric; transient; transformation. introduction, Section II elaborates the issue of
measuring shape similarity. Section III
I.INTRODUCTION definition of SDC, while Section IV presents
Wavelet Transform (WT) is a recent the procedure and experimets to measure
development in signal processing [1, 2]. shape similarity between a wavelet function
Wavelet transform of a signal f  L2(R) is and transients in a signal. Finally, Section V
defined as a correlation between the signal and gives the concluding remark.
a dilated-shifted wavelet function ψs,σ.
II. MEASURING SHAPE SIMILARITY

1 t  Shape similarity between a wavelet function
Wf ( s ,  )  f ,  s*,   f (t )
 s
 *(
s
) dt and a transient in a signal can be observed on
the symmetric pattern in wavelet coefficients
(1)
in the neighborhood of the center of the
transient, t = tc [16]. It is based on the fact that
The magnitude of wavelet coefficient, |Wf(s,
correlation function of two similar shape
σ)|, indicates how well they correlate around t
functions is always symmetric (even-
= σ, and is determined by the matching
symmetric). Consequently, if a wavelet
between the properties of wavelet function and
function has similar shape to a transient at t =
the properties of the analyzed signal. WT
tc, the wavelet coefficients in the neighborhood
offers a large option of wavelet functions. This
of σ = tc are even-symmetric. This symmetry

162
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

feature is not altered by signal amplification.

There is a difficulty in using the symmetric
nature to measure shape similarity because in
the real world, it is nearly impossible to find a
wavelet function that has exactly a similar
shape to any transients in input signal. One can
only find a wavelet function that is almost
similar. This condition will produce nearly
even-symmetric wavelet coefficients. To
measure this ‘nearly similar’ condition, it is
Figure 1. A nearly even-symmetric function at t = tc with
necessary to have a parameter to measure
time support [tc-ts , tc+ts]. The SDC(tc,ts) is the ratio of
symmetry in gradual levels. Unfortunately, the energy of (A-B) to the total energy of A and B.
symmetry is only defined as a discrete
parameter (i.e. symmetry or asymmetry). The The range of SDC in (3) is [0, 2] which
new parameter, introduced in this paper, was does not meet common sense. The equation is
developed to measure symmetry in gradual modified to shift its range to [-1,1] where -1
levels. The parameter is applied to the case of
represents odd-symmetry, 0 represents
wavelet function selection.
asymmetry, and 1 represents even-symmetry
as in (4).
III. SYMMETRIC DISTANCE ts

  f (t   ) f (t c   )  d
COEFFICIENT 2
c
A function, f(t), is considered symmetric SDC (t c , t s )  1  0
ts
around t = tc if
f (t c   ) d 
2

t s
 f (tc   )   f (tc   )[ t ,t ] s s (4)
(2)
SDC is independent to input signal
where  is any value in a time support of [-ts, amplification. Suppose a signal, f(t), is even-
ts]. Equation (2) strictly distinguishes between symmetry around t = 0 i.e.
symmetric and asymmetric function. SDC  f (t )  f (t )t[ts ,ts ] . Another signal, g(t), is
measures the degree of symmetry in gradual
levels. It is inspired by the concept of equal to f(t) but asymmetric due to
e(t )t[0,t ] s

symmetric distance in image analysis that :

measures the minimum effort required to
modify pattern in an image into its symmetric  f (t ) t  [t s ,0]
shape [17-19]. g (t )  
 f (t )  e(t ) t  (0, t s ]
SDC is the ratio between the energy of the
difference between left and right side of a (5)
function around its center t = tc, and the
energy of the function within the same time then SDC at t = 0 with support [-ts ,ts] is:
ts
support of t = [tc-ts , tc+ts ]. It is illustrated in
e ( ) d
2

Fig. 1. SDC (0, t s )  1  t 0

ts s

  f (t   ) f (t c   )  d 2f ( )  2 f ( )e( )  e 2 ( ) d
2 2
c
0
SDC (t c , t s )  0
ts
(6)
f (t c   ) d 
2

t s

(3) If the signal is then amplified by a factor of A:

 A f (t ) t  [  t s ,0 ]
g (t )  
 A f (t )  A e(t ) t  (0, t s ]
(7)

163
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

the SDC becomes

ts

A 2  e 2 ( ) d
SDC (0, t s )  1  t s 0

 2A f ( )  2 Af ( ) Ae ( )  A 2 e 2 ( ) d
2 2

(8)

By canceling out the amplification factors, A,

equation (8) proves that SDC can measure the
level of symmetry, independent to the strength
of input signal or transient.
The ability to measure symmetry
independent to signal strength may create
problem if the input signal has too many Figure 2. The SDC of various types and levels of
symmetry.
unexpected small ripples (see Fig. 3). To avoid
measuring SDC of unwanted small transients,
a threshold level, th, is added to the equation
to ignore transients with energy less than the
threshold level.
ts

f (tc   )d  th
2
SDC(tc , t s ) if
 ts
SDCth (tc , t s , th)  
 0 else
(9)

Equation (8) and (9) are the mathematic

definition of SDC.
Fig. 2 shows SDCth(t,0.25,0.1) of even-
symmetric, nearly even-symmetric, odd-
symmetric, nearly odd-symmetric and
asymmetric transients located at t = 0.75 s,
1.75 s, 2.75 s, 3.75 s, and 4.75 s, respectively.
The SDC of the first transient,
SDCth(0.75,0.25,0.1), is 0.99. It means that the
transient is 99% even-symmetry. The second Figure 3. a) A signal with an even-symmetric transient
added with noises. b) The SDCth(t,0.125,0), c) The
transient is only 90% even-symmetry, while SDCth(t,0.125,5).
the third to fifth transients are 99% odd-
symmetry, 92% odd-symmetry, and 0% IV. MEASURING PROCEDURES AND
symmetry, respectively. The figure shows how EXPERIMENT
SDC can measure various types and levels of The procedure to measure similarity between a
symmetry. wavelet function, ψs,σ, and a transient located
Fig. 3 shows the necessity of using at t = tc in an input signal, f(t), is as follows:
threshold. It shows SDC of a signal with a 1. Apply WT to f(t) using the selected wavelet
symmetric transient at t = 0.475 s and small function, ψs,σ, to the whole time span of f(t).
noises. When threshold was set to zero (no 2. Calculate SDCth(tc,ts,th) of the wavelet
threshold) the SDC of the investigated coefficients at time location  = tc. The
transient was obscured by a large number of time support parameter, ts, should be half of
SDCs of the noises (Fig. 3b). When a the selected wavelet function’s time
threshold value of 5 was used, the SDC of the support or half of the interested transient’s
noises were removed (Fig. 3c). time support. Use an appropriate threshold,
th, to ignore small ripples in wavelet

164
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

coefficients. ability to measure shape similarity between a

3. The value of SDCth(tc,ts,th) indicates the wavelet function and transients in signal. The
degree of shape similarity between the experiment demonstrates how the parameter
selected wavelet function and the transient. can be used to find a wavelet function that has
The closest the value to 1, the more similar the most similar shape to T-waves in ECG
the two functions. signal.
The parameter and procedure was applied to
electrocardiograph (ECG) signal [7, 20] to find
a wavelet function that has the most similar
shape to T-waves (marked with ‘T’ in Fig. 4a).
An ECG signal, downloaded from MIT-BIH
PhysioBank database record aami3a.dat
(AAMI-EC13) [21] was used in this
experiment (Fig. 4a). The signal was re-
sampled with a sampling rate of 360 Hz.
Wavelet coefficients were calculated using
various wavelet functions. However, the
results of only three wavelet functions are
presented in this paper (i.e. db2 at scale 28,
Gaus2 at scale 8, and Mexh at scale 12) due to
their significant results. The SDC calculation
was then applied to all wavelet coefficients
with a half time support of 0.158 s and a
threshold value of 0.1. The results (Fig. 4b-d)
suggested that db2 at scale 28 is the wavelet
function with the most similar shape to T-
waves. The average SDCth(t,0.158,0.1) of the
four T-waves when calculated using db2
wavelet at scale 28 was 0.915. The value was Figure 4. a) ECG signal (MIT-BIH aami3a.dat – AAMI
higher than that of Gaus2 wavelet at scale 8 EC13) with sampling rate 360 Hz, b) – d) The
and Mexh wavelet at scale 12 (0.88, and 0.82, SDCth(t,0.158,0.1) calculated on wavelet coefficients
respectively). The experiment suggests that the using Db2 at scale 28, Gaus2 at scale 8, and Mexh at
scale 12, respectively.
parameter and the procedure can be used to
find wavelet function that has the most similar REFERENCES
shape to an interested transient. [1] I. Daubechies, Ten Lectures on Wavelets.
Pennsylvania: SIAM, 1992.
V. CONCLUSION [2] S. Mallat, A Wavelet Tour of Signal
This paper introduces a new parameter called Processing. London: Academic Press,
‘symmetric distance coefficient’ (SDC) and the 1999.
procedure to measure shape similarity between [3] S. Mallat and W. L. Hwang, "Singularity
a wavelet function and a transient in a signal. Detection and Processing with
The parameter measures the similarity by Wavelets," IEEE Transaction on
measuring the symmetric pattern in wavelet Information Theory, vol. 38, pp. 617-
coefficient. It is based on the fact that wavelet 643, 1992.
transform of a transient that has the same [4] N. D. Pah and D. K. Kumar,
shape as a wavelet function is always "Thresholding Wavelet Networks for
symmetric. The parameter is useful as a Signal Classification," International
guidance to select suitable wavelet function Journal of Wavelets, Multiresolution,
for a specific transient in a signal or and Information Technology, vol. 1, pp.
application. 243-261, 2003.
The paper also presents an experiment to [5] D. K. Kumar and N. D. Pah, "Neural
demonstrate the procedure and the parameter’s Network and Wavelet Decomposition for

165
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Classification of Surface [14] P. Zhou and W. Z. Raymer, "Motor Unit

Electromyography," Electromyography Action Potential number Estimation in
and Clinical Neurophysiology, vol. 40, the Surface Electromyogram: Wavelet
2000. matching Method and Its Performance
[6] J.-K. Zhang, T. N. Davidson, and K. M. Boundary," presented at The 1st
Wong, "Efficient Design of Orthonormal International IEEE EMBS Conference on
Wavelet Bases for Signal Neural Engineering, Capri Island, Italy,
Representation," IEEE Transaction on 2003.
Signal Processing, vol. 52, pp. 1983- [15] N. D. Pah, "Measuring Wavelet
1996, 2004. Suitability with Symmetric Distance
[7] C. Li, C. Zheng, and C. Tai, "Detection Coefficient," presented at Seminar
of ECG Characteristic Points Using Nasional Soft Computing, Intelligent
Wavelet Transform," IEEE Transaction System and Information Technology,
on Biomedical Engineering, vol. 42, pp. Surabaya, Indonesia, 2005.
21-28, 1995. [16] P. Abry and P. Flandrin, "Multiresolution
[8] S. Karlsson, J. Yu, and M. Akay, "Time- Transient Detection," presented at
Frequency Analysis of Myoelectric Proceedings of the IEEE-SP International
Signals During Dynamic Contractions: A Symposium on Time-Frequency and
Comparative Study," IEEE Transaction Time-Scale Analysis, 1994.
on Biomedical Engineering, vol. 47, pp. [17] Y. Keller and Y. Shkolnisky, "A Signal
228-238, 2000. Processing Approach to Symmetry
[9] S. Kadambe and G. F. Boudreaux- Detection," IEEE Transaction on Image
Bartels, "Application of the Wavelet Processing, vol. 15, pp. 2198-2206,
Transform for Pitch Detection of Speech 2006.
Signals," IEEE Transaction on [18] I. Choi and S.-I. Chien, "A Generalized
Information Theory, vol. 38, pp. 917- Symmetry Transform with Selective
924, 1992. Attention Capability for Specific Corner
[10] A. J. Hoffman and C. J. A. Tollig, "The Angles," IEEE Signal Processing Letters,
Application of Classification Wavelet vol. 11, pp. 255-257, 2004.
Networks to the Recognition of Transient [19] A. Imiya, T. Ueno, and I. Fermin,
Signals," IEEE, pp. 407-410, 1999. "Symmetry Detection by Random
[11] V. J. Samar, "Wavelet Analysis of Sampling and Voting Process for Motion
Neuroelectric Waveforms: A Conceptual Analysis," International Journal of
Tutorial," Brain and Language, pp. 7-60, Pattern Recognition, vol. 17, pp. 83-125,
1999. 2003.
[12] R. Q. Quiroga, O. W. Sakowitz, E. Basar, [20] J. S. Sahambi, S. N. Tandon, and R. K. P.
and M. Schurmann, "Wavelet Transform Bhatt, "Using Wavelet Transform for
in the Analysis of the Frequency ECG Characterization," in IEEE
Composition of Evoked Potentials," Engineering in Medicine and Biology,
Brain Research Protocols, vol. 8, pp. 16- 1997, pp. 77-83.
24, 2001. [21] A. L. Goldberger, L. A. N. Amaral, L.
[13] R. Zhang, G. McAllister, B. Scotney, S. Glass, J. M. Hausdorff, P. C. Ivanov, R.
McClean, and G. Houston, "Combining G. Mark, J. E. Mietus, G. B. Moody, C.
Wavelet Analysis and Bayesian K. Peng, and H. E. Stanley,
Networks for Classification of Auditory "PhysioBank, PhysioToolkit, and
Brainstem Response," IEEE Transaction PhysioNet: Components of a New
on Information Technology in Research Recource for Complex
Biomedicine, vol. 10, pp. 458-466, 2006. Physiologic Signals," Circulation, vol.
101, pp. e215-e220, 2000.

166
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

An Implementation of Investment Analysis using Fuzzy Mathematics

Novriana Sumarti and Qino Danny

Industrial and Financial Mathematics Research Group, Institut Teknologi Bandung, Indonesia
Email: novriana@math.itb.ac.id

ABSTRACT Scholes formula [4], real options valuation [2],

Buckley [1] develops fuzzy analogues of problems insurance [6], and determination of factors in
in financial mathematics, such as the fuzzy present macroeconomics [3].
value and fuzzy future value of fuzzy cash
amounts, using fuzzy interest rates. Using fuzzy In this paper, the fuzzy analysis is used in
approach in analyzing problems in financial
valuing a new investment of photocopy
mathematics, it tolerates facts happened in real
world that future cash amounts and interest rates service, which is a promising business in
are quite often in estimation values. Indonesia, especially in a perimeter of schools
and universities. The valuation of this business
In this paper we implement the fuzzy analysis in uses data from the existing business in
valuing a new investment on a photocopy service at Bandung, Indonesia.
Bandung. In valuing this investment, we analyze
some scenarios related to the choices of the II. FINANCE IN FUZZY
machines’ types, the appropriate time to buy goods,
and need of additional sale activity on top of the Denote A  ( a1 , a2 , a3 ) and B  (b1 , b2 , b3 )
main business. Using data from an existing and two fuzzy numbers. Another way to write
similar business, we can determine the best fuzzy numbers [5] is in the difference between
investment scenario which can be choose. the middle number with the left and right
numbers, such as
KEYWORDS
Financial Mathematics; cash flow analysis;
A  (a2, a2 a1, a3 a2 ) or simply A  (a,1,2 ) ,
investment valuation; Fuzzy logic. and B  (b2 , b2  b1, b3  b2 ) or B  (b, 1, 2 ) . The
arithmetic operations between them become
I. INTRODUCTION
Uncertainty and vagueness in investment A B  (a b,(a 1) (b  1),(a 2 ) (b  2 )) ,
valuation are sometimes difficult to deal with. A B  (a b,(a 1) (b  2 ),(a 2 ) (b  1)) ,
To represent them mathematically and to
provide formalized tools for dealing with the A B  (ab, a (b1)  b (a1), a (b 2)  b (a 2))
imprecision, the fuzzy sets theory was
specifically design to those purposes. The 1 1 a 2 a 1 
 , , .
main characteristic of fuzzy system is its A  a a(a (a 2 )) a(a (a 1)) 
suitableness for uncertain or approximate
reasoning, especially for the system with a Now we discuss future values after n period:
mathematical model that is difficult to derive.
Furthermore, fuzzy logic allows decision S n  A  (1  r )n , where A is fuzzy amount
making with estimated values under invested today at rate r per period. Notes that
incomplete or uncertain information.  and  are fuzzy multiplication and
addition operations. In present value analysis,
Fuzzy systems have been used in decision we define two formulae:
making process in engineering since it was
1. PV1 ( S , n )  A if and only if A is a
proposed by Lotfi Zadeh in his paper [7]. In
fuzzy number and A  (1  r )  S .
financial mathematics, Buckley [1] had done n

the pioneer works since 1987, explaining the

fuzzy point of view of some important 2. PV2 ( S , n )  A if and only if A is a
definitions and equations of mathematics fuzzy number and A  S  (1  r ) .
n

finance. Nowadays there are many researches

on the implementation of fuzzy logic in this
area, such as in option pricing using Black

167
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Theorem: Let 1  ( x | S , n ) and investment. On other way, it can be used to

calculate the time when the initial capital for
 2  ( x | S , n) denotes the membership investment is paid back by the cash flows.
functions for PV1 ( S , n ) and PV2 ( S , n)
In the next section we discuss two main plans
respectively. If S is negative, PV1 ( S , n ) is designed in this business. Plan A considers the
exist and the membership function type of the photocopy machines. On the other
hand, plan B considers the time and the way of
1  ( x | S , n ) is determined by purchasing goods for the business. Using data
f i ( y | A)  f i ( y | S )(1  f i ( y | r ))  n , i  1, 2 of cash flow and cash out from existing
. photocopy business in Bandung, Indonesia, all
business plans can be valuated.
On the other hand, if S is positive,
PV2 ( S , n) is exist and the membership III. VALUATION OF PLAN A
Plan A consist of three choices of the
function  2  ( x | S , n ) is determined by photocopy machines will be used. There are
f i ( y | A)  f i ( y | S )(1  f 3i ( y | r ))  n , i  1, 2 two types of photocopy machines based on
their prices. The expensive type is IRR 500
. The proof of the above theorem can be found
with price (23, 0.5, 1) million IDR. The cheap
in [1].
one is 6050 with price (10, 0.5, 1) million
IDR. However, there is a significant difference
In valuation of an investment, we need to
in the operating cost where the cheap type
compute Net Present Value (NPV), the needs higher cost than the expensive one.
difference between the present value of the Three choices in Plan A are explained in
future cash flows and the amount to be figure 1 below.
paid initially for investment. If its value is
positive, the investment should be made, Figure 1. Plan A
otherwise it traditionally should not be
Type of Machines
made. A real option analysis can be
performed in addition to the NPV analysis
if its value is negative. The fuzzy formula 1. #2 IRR 5000 2. #2 6050
of NPV is defined as in [1].
3. #1 IRR 5000, #1 6050
Let A  A0 , A1 ,, An denote cash flow,
The cash out is the cost from purchasing of
where A0 is positive and A j , j  1, 2,, n two machines, the venue rental, electricity,
can be either positive of negative. Let r worker’s salary, equipments for book binding.
denotes a positive interest rate. The fuzzy In addition to the photocopy service, there is a
NPV is small shop selling stationary, snacks, and
n beverages, including ice cream. Heuristically
NPV ( A, n)  A0   PVk ( i ) ( Ai , i ) , this small shop can significantly increase the
i 1 income because the snacks and ice cream
where  is fuzzy addition and attract customers who wait for the work to be
finished. The cost for the shop is also
considered as cash out.
1, Ai negative
k (i )   . From analyzing choices 1 to 3, it gives some
2, Ai positive conclusions. The capital needed in beginning
Another analysis to value an investment is for the choice of two IRR 5000 machines is
Internal Rate of Return (IRR). This value the largest amount among other choices. The
determines the rate of return when NPV equals second highest is the choice of mixed IRR
zero for a given number n. The higher IRR, 5000 and 6050. In term of the time when the
the more desirable it is to undertake the capital could be paid back by the cash flow,
both choices need 8 years. On the other hand,

168
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

the choice of two 6050 machines could give 3 gives 107-542 million IDR, scenario 4 gives
back in year 6. From these results, the latter is 75-610 million IDR.
chosen in designing other scenarios in the next
section. To conclude, the highest profit gains from
scenario 1 where all goods are purchased and
IV. VALUATION OF PLAN B paid cash from the first year. However, the
In Plan B, we make the following different debt from the cash shortage of this scenario is
scenarios and it is described in Table 1. The also the highest. It means that the risk in this
capital owned by the investor is only 50 scenario is also the highest.
million IDR. The shortage of cash in each plan
is filled by credit scheme from a bank. V. CONCLUSIONS
Fuzzy analysis on valuation of an investment
Table 1. Scenario B on photocopy service gives more realistic
results due to the characteristic of fuzzy logic
Descrip- Scenarios
tion 1 2 3 4
which allows uncertainty and imprecision of
#2 6050 Cash Cash 3y the data. The results show that the best choice
Cash of the machine types is both the cheap one,
yI & yIII yI & yIII credit
#2
yI yI & yIII yI & yIII yI
and the best scenario with the highest profit
workers comes from the highest risky investment too.
#1 2y
2y credit 2y credit
Motor- Cash credit
yI yI REFERENCES
cycle yI
#1 Car 5y
5y credit 5y credit
Cash
yI yI
credit [8]. J.J. Buckley, The fuzzy mathematics of
yI finance, Fuzzy Sets and Systems 21 (3)
Small 2y (1987) 257–273.
Cash
shop Cash Cash credit
yIII
yI [9]. Carlsson, C., A fuzzy approach to real
option valuation, Fuzzy Sets and Systems,
Scenario 1: All are paid in cash at the first 139 (2003) 297–312.
year. [10]. C. Kahraman (editor), Fuzzy
Scenario 2: Cash at the first year are paid for Engineering Economic with Applications,
one machine, one worker and the shop Spinger-Verlaag Berlin, 2008.
equipments. In the third year, cash is paid for [11]. Lee, CF, Tzeng GH and Wang SY, A
another machine and worker. A motorcycle new application of fuzzy set theory to the
and a car are bought on credit schemes for 2 Black–Scholes option pricing model,
and 5 years respectively. Expert Systems with Applications, 29
Scenario 3: This is the same with scenario 2 (2005) 330–342.
unless the shop opens in the third year. [12]. G.J. Klir and G. Yuan, Fuzzy Sets and
Scenario 4: All goods are bought on credit Fuzzy Logic, Theory and Applications,
schemes for different periods of time in the Prentice-Hall, New Jersey, 1995.
first year. [13]. Shapiro, A.F., Fuzzy logic in
insurance, Insurance: Mathematics and
Calculating cash flow and cash out during 8 Economics, 35 (2004) 399–424.
years, the results show that scenario 1 gives [14]. Zadeh, L.A. (1965). "Fuzzy sets",
NPV from around 88 -616 million IDR, Information and Control 8 (3): 338–353.
scenario 2 gives 60-576 million IDR, scenario

169
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Simulation of Susceptible Areas to the Impact of Storm Tide Flooding

along Northern Coasts of Java

Nining Sari Ningsih(1), Safwan Hadi(2), Dwi F. Saputri (3), Farrah Hanifah(4), and Amanda P.Rudiawan
(5)
(1)
Research Group of Oceanography, Faculty of Earth Sciences and Technology, ITB, Bandung, West Java,
Indonesia
Email: nining@fitb.itb.ac.id
(2)
Research Group of Oceanography, Faculty of Earth Sciences and Technology, ITB, Bandung, West Java,
Indonesia
Email: safwan@ppk.itb.ac.id
(3)
Research Group of Oceanography, Faculty of Earth Sciences and Technology, ITB, Bandung, West Java,
Indonesia
Email: dfajar.saputri@gmail.com
(4)
Study Program of Oceanography, Faculty of Earth Sciences and Technology, ITB, Bandung, West Java,
Indonesia
Email: hanifah_farrah@yahoo.com
(5)
Master Program in Geodetic and Geomatic Engineering, Faculty of Earth Sciences and Technology, ITB,
Bandung, West Java, Indonesia
Email: oceanrocksz@yahoo.com

ABSTRACT KEYWORDS
Storm tide event as the sum of the astronomical Storm surges; storm tide; inundation; run-up.
tides and storm surges generated by the Cyclones
Hagibis and Mitag in November 2007 has been I. INTRODUCTION
simulated to quantify wave height, run-up, and Severe storm surges generated by tropical
inundation along the northern coasts of Java by cyclone in South China Sea, Pacific and Indian
using a two-dimensional ocean model. The applied
Oceans have often hit the northern coasts of
model was run in the period from 12 November
2007 until 13 December 2007. Java. For example, (1) severe flooding caused
by high astronomical tides and cyclones
The model domain covers South China Sea, Hagibis and Mitag, have inundated coastal
Philippines Waters, and Pacific Ocean as regions areas of the northern part of Java during 24 –
where both cyclones being generated, Java Sea, and 27 November 2007, such as Jakarta,
coastal areas of northern part of Java. The storm Semarang, Tegal, and Brebes, damaging
tide event was simulated by imposing tidal houses and infrastructures, disturbing access to
elevations at the open boundaries, 6-hourly wind the Soekarno-Hatta International Air Port in
and air pressure data. Tidal elevation data derived Jakarta, and forcing people to leave their
from the tide model driver (TMD) of Padman and
homes; and (2) in March 2008, Karimunjawa
Erofeeva (2005), while wind and air pressure data
were obtained from NCEP (National Centers for and Jepara waters in Java Sea were battered by
Environmental Prediction). cyclone Nicholas generated in Australian
waters, damaging fishing ships, causing
The simulation results show that the storm tides several ship accidents, and disturbing tourism
had attacked and inundated coastal areas of the activities and delivery of foods to the
northern part of Java. Along the coastal areas, Karimunjawa Island due to the lack of sea
highest surge existed at Losari, which is located at transportation.
Province of Central Java. Maximum distances of
storm tide flooding at each provinces of Java In the case of flooding of 24 – 27 November
occured at Banten Bay (Banten Province), 2007, the flooding along the coastal area was
Penjaringan (DKI Jakarta Province), Cilamaya
worse than previous flooding events because
(West Java Province), Semarang (Central Java
Province), and Surabaya and Sampang (East Java surges coincided with high tides caused by
Province). spring tide and perigee conditions. In that case,
the advancing surge generated by the cyclones

170
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Hagibis and Mitag combined with the was set from 17o 5’ S – 20o 30’ U and 99o 10’
astronomical tide to create a storm tide. – 140o 00’ E, which covers South China Sea
and Philippines Waters, Pacific Ocean as
Storm tides are the sum of the astronomical regions where both cyclones being generated,
tides (the daily changes in water level due to Java Sea, and coastal areas of northern part of
gravitational interactions between the earth, Java (as shown in Figure 1). The grid sizes
moon, and sun), the storm surges (the transient vary from ± 200 km in the Pacific Ocean,
changes due to the effects of a storm), and South China Sea, and Philippines Waters to
long-term changes (sea level rise, seasonal and fine resolution of about 5 km in the coastal
decadal changes). The storm tide can cause area. Since the present study focuses on the
severe flooding in coastal regions, especially northern coasts of Java, the higher resolution
in low-lying areas, frequently causing losses of of grid sizes in those areas is performed of
lives and substantial economic damages. The about 150 m in order to include inundation
degree of flooding depends on storm intensity, processes in the model simulation.
fluctuations in astronomically generated tides,
and the slope of the continental shelf.

Since coastal zones are at risk of storm tides, it

is necessary to study crucial coastal zones
affected by storm tide flooding, especially
along northern coasts of Java. Moreover,
accurate estimates of maximum storm surge
(MSS) and inundated (flooded) areas caused
by storm tides are necessary, primarily from
the view of flood defense.

As far as we know, storm tide investigation

(Sourceof map: http://www.wikimapia.org)
along the northern coasts of Java is still
limited, especially study of vulnerable coastal
zones to the impact of storm tide flooding.
Therefore, we are interested in studying
influences of storm surges combined with the
astronomical tide along the northern coasts of
Java. In this study, we only considered storm
tides as the sum of the astronomical tides and
storm surges. Hence, we neglected the
influences of the long-term changes.

Estimations of surge height, water run-up

height, and inundation areas due to the impact
of tides and storm surges were carried out in
this study by considering the Cyclone Hagibis
during 19 – 27 November 2007 and Cyclone
Figure 1. Computational domain, bathymetry (in meters),
Mitag during 20 – 27 November 2007 as a and model meshes.
generating force of storm surges and by using
a barotropic hydrodynamic numerical model
(2D version of Mike 21 developed by [3]). The storm tide event, which was generated by
tides, the Cyclone Hagibis (Figure 2a) with
II. MATERIAL AND METHODS wind speed range about 17.99 – 35.98 m/s and
A finite volume method is used in the applied Cyclone Mitag (Figure 2b) with wind speed
model. The method allows for a substantial range about 17.99 – 41.12 m/s, was simulated
flexible discretization of the model domain. by imposing tidal elevations at open
An unstructured grid comprising of triangle or boundaries, winds, and air pressures. The tidal
quadrilateral element was used in the elevation data was derived from the tide model
horizontal plane [3]. The computational area driver (TMD) of [5]. The TMD is a Matlab

171
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

package for accessing the harmonic and a radiation boundary condition was used
constituents, and for making predictions of for currents. When wind forcing was included,
tide height and currents. It has 1/4° x 1/4° the tilt facility of the applied model has been
resolution and eight tidal constituents (M2, S2, used to improve the boundary conditions.
N2, K2, K1, O1, P1, Q1) and was used for Tilting provides a correction of water level at
predicting the tidal elevations at the open each point along the boundary based on the
boundaries. We used 6-hourly wind and air steady state Navier Stokes equations. A
pressure data with 2.5° x 2.5° resolution detailed description of this tilt facility can be
obtained from NCEP (National Centers for found in [3]. Further, flooding and drying
Environmental Prediction). (FAD) capabilities of the model were enabled
to simulate water run-up and the inundation
processes of storm tides caused by tides and
(a) both cyclones.

Bathymetry was generated using the 1 minute

resolution GEBCO data [4] and topography
was resolved using the 90 m resolution Digital
Elevation Model (DEM) of the Shuttle Radar
Topographic Mission (SRTM), [7]. However,
a changed coastline (taken from navigational
maps) was constructed for northern coasts of
Java to take into account specific topographic
features there, which are not described
accurately enough in the standard topography.
The parameterization of bottom friction is
based on Manning’s approach, with the ranges
of friction coefficient n from 0.01 for smooth
concrete to 0.06 for poor natural channels [1].
Since no other information of bottom friction
(Sourceof map: http://www.wikimapia.org)
is available along the computational domain,
we used a constant manning number of
(b) 0.03125 m-1/3s as suggested by [3]. Horizontal
diffusion is needed for numerical stability. It is
a Smagorinsky type with the Smagorinsky’s
constant = 0.28.

III. SIMULATION RESULTS

3.1. Model Verification
Following [6], we compared the simulated
storm surges, obtained by removing the tidal
part from the storm tide simulation, with
Figure 2. Depression tracks of: (a). the Cyclone Hagibis weekly data of TOPEX Poseidon Sea Level
during 19 – 27 November 2007 and (b). the Cyclone Anomaly (SLA),
Mitag during 20 – 27 November 2007. (Source: (http://apdrc.soest.hawaii.edu). A sea-level
http://agora.ex.nii.ac.jp/digital anomaly is the difference between the total
typhoon/year/wnp/2007.html.en).
sea-level and the average sea-level for this
time of year [2]. Figure 3 shows validation of
The numerical simulation was carried out for the simulated storm surge height with the
31 days (12 November – 13 December 2007), TOPEX Poseidon SLA at the whole
which covered the occurrence of both computational domain, whereas examples of
cyclones. Zero normal flow was applied to the surge height validation at a location, such
solid boundaries, while along open boundaries as at Penjaringan (marked by 5 in Figure 5)
tidal elevation varying both in time and space can be seen in Figure 4.
were specified to accommodate tidal forcing

172
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

SLA-TOPEX [m]
Model result [m]

(a)

(b) Figure 4. Validation of the computed storm surge height

at 00:00 UTC on 21 and 28 November 2007; and on 5
and 12 December 2007. Blue line (the simulation
results); red dots (the TOPEX Poseidon SLA); red square
(period of the cyclone Hagibis); green square (period of
the cyclone Mitag), at Penjaringan (marked by Point 5 in
Figure 5).

3.2. Surge Heights and Inundation Areas

Figure 5 shows that there were 36 areas most
affected by the storm surge generated by the
Cyclones Hagibis and Mitag, namely: a).
along the northern coast of Banten Province:
(1). Merak, (2). Banten Bay, (3). Tirtayasa,
and (4). Mauk; (b). along the northern coast of
Jakarta Province: (5). Penjaringan and (6).
Cilincing; (c). along the northern coast of
West Java Province: (7). Karawang Cape, (8).
Tambak sumur, (9). Sungaibuntu, (10).
Figure 3. Validation of the computed storm surge height Cilamaya, (11). Pamanukan, (12). Indramayu,
at 00:00 UTC on December 5, 2007. (a). Simulated
results (in cm). Arrows indicate speed and direction of
(13). Balongan, and (14). Cirebon; (d). along
winds (m/s); (b). The TOPEX Poseidon SLA (in cm), the northern coast of Central Java Province:
(Source: www.apdrc.soest.hawaii.edu). (15). Losari, (16). Tegal, (17). Pekalongan,
(18). Gringsing, (19). Semarang, (20). Jepara,
In general, storm surge heights obtained from (21). Bumimulyo, and (22). Rembang; (e).
the simulation show a good agreement with along the northern coast of East Java
those of the TOPEX Poseidon SLA. However, Province-1 (Main Land): (23). Tambakboyo,
the simulated storm surge was smaller than (24). Labuhan, (25). Pangkah Cape, (26).
those of the TOPEX Poseidon SLA, probably Surabaya, (27). Sidoarjo, (28). Bangil, (29).
due to the estimation of the parameters used in Probolinggo, (30). Gending, (31). Gerinting
the model (e.g., the bottom- and wind-drag Cape, and (32). Panarukan; (e). along the
coefficients) is not adequate. northern coast of East Java Province-2
(Madura Island): (33). Pamekasan, (34).
Sampang, (35). Modung Cape, and (36).
Sapulu.

173
Proceedingg of Conf. on In
ndustrial and Appl.
A Math., Bandung-Indoneesia 2010

Mitag.. It can be seeen from thee Figure 5 thhat

In additioon, the Figuure 5 also shows the Losarii (marked 15 in the Figuree 5), located at
maximum m storm surgee height (MS SSH) at the Centraal Java Proovince), exp perienced thhe
36 locatio
ons (marked by
b Points 1 – 36), which highesst surge amonng the other stations, about
are most prone to thhe impact off the storm 17.4 cmm, at 5:45 UT
TC on 5 Deceember 2007.
surge gennerated by thee Cyclones Hagibis
H and

12.0
12.3
11.6 13.3 13.5
12.2 13.5
7 8 13.0 13.6
1
5
6 9 10 12 1
11.5 13.1 14.3
11
13
3 17.4 36
6
13.5 13.0 35 14.4 13.2
3 4 20
2 21 22 34 33
12.7
7 13.6 23
24 25
26
12
2.2
12.5 12.2 12.2 12.4
14.2 1 15
14 16
17 18 19 12.6 13.1
27
13.9 13.1 14.5
14.8
8 32
14.9 28
8 29 30 31
13.4
4
14.6 14.0
1
14.1

1
15 cm

(Sourceeof map: Google E

Earth)

Figure 5. The maximum storms surge heigghts (in cm) at thhe 36 locations, which
w are most ssusceptible to thhe impact of the
storm surgge generated by the
t Cyclones Haagibis and Mitagg in November 20 007. Name of Loocations: (1). Merak, (2). Bantenn
Bay, (3). Tirtayasa,
T (4). Mauk,
M (5). Penjaringan, (6). Cilincing, (7). Karawwang Cape, (8). T
Tambaksumur, (9). Sungaibuntuu,
(10). Cilamaya,
C (11). Pamanukan, (122). Indramayu, (113). Balongan, (14). Cirebon, (15). Losari, (16). Tegal, (17).
Pekalonggan, (18). Gringssing, (19). Semaarang, (20). Jeparra, (21). Bumim mulyo, (22). Rem
mbang, (23). Tam mbakboyo, (24).
Labuhan, (225). Pangkah Caape, (26). Surabaaya, (27). Sidoarjjo, (28). Bangil, (29). Probolingggo, (30). Gendinng, (31). Gerintinng
Cape, (32).. Panarukan, (333). Pamekasan, (3 34). Sampang, (335). Modung Caape, and (36). Saapulu. The higheest surge occurreed
at the Losari (m
marked 15 in Figuure 5).

Figure 6. Astronomic
A tide, storm tide, and storm
s surge heigghts at the Penjarringan (marked 5 in Figure 5); brown
b dash squarre
(periood of the cyclonne Hagibis); bluee dash square (peeriod of the cycloone Mitag).

174
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Figure 6 shows an example of the calculated Banten Province, marked 2 in Figure 5),
heights of astronomic tide, storm tide, and Penjaringan (at the Jakarta Province, marked 5
storm surge at the Penjaringan (marked 5 in in Figure 5), Cilamaya (at the West Java
Figure 5). Based on the simulated results in Province, marked 10 in Figure 5), Semarang
this present study, it is found that the derived (at the Central Java Province, marked 19 in
storm surge heights are of order of few Figure 5), and Surabaya and Sampang (at East
centimeters (< 20 cm), so it seems that the Java Province, marked 26 and 34 in Figure 5,
Cyclones Hagibis and Mitag in November respectively). Among the 36 locations in
2007 did not generate high storm surges along Figure 5, Rembang (marked 22 in Figure 5)
the northern coasts of Java. However, as it will experienced minimum distance of storm tide
be shown later, the coasts of northern Java flooding (table and figure not shown). An
were still severely flooded. analysis of topography slope shows that the
slope of topography may contribute to the
Table 1 shows detailed maximum horizontal distance of storm tide flooding. For instance,
distance of storm tide flooding, vertical run-up topography of the Sampang (marked 34 in
heights, and total water depth around the Figure 5) is gentle of about 0.012o compared
location of maximum inundation distance at 6 to that of the Rembang of about 0.545o. This
locations, which represent most susceptible difference may contribute to the existence of
areas along the northern coast of Java to the distant storm tide flooding at the Sampang;
impact of the storm tide flooding at each whereas the steep slope of the Rembang
province of Java, namely Banten Bay (at the decreases the amount of flooding.

Table 1. Maximum horizontal distance of storm tide flooding, vertical run-up heights, and total water depth
around the location of maximum distance of inundation
Run – up and Inundation (Flooding)
Total Water Depth around the
Vertical Run-up at the
Maximum Horizontal Location of Maximum
Location of Maximum
No. Locations Distance of Inundation (m) Horizontal Distance of
Distance of Inundation (m)
Inundation (m)
Storm+ Storm+ Storm+
Tides Storm Tides Storm Tides Storm
Tides Tides Tides
Banten Province:
2 Banten Bay 558.5 558.5 0.0 0.483 0.428 0.055 0.15 0.13 0.02
Jakarta Province:
5 Penjaringan 6275.3 5958.7 316.6 0.537 0.483 0.054 0.33 0.28 0.05
West Java Province:
10 Cilamaya 2967.6 2967.6 0.0 0.535 0.489 0.046 0.12 0.08 0.04
Central Java Province:
19 Semarang 1278.3 972.4 305.9 0.570 0.472 0.098 0.17 0.12 0.05
East Java Province-1
(Main Land):
26 Surabaya 5305.4 5305.4 0.0 2.099 2.037 0.062 0.27 0.19 0.08
East Java Province-2
(Madura Island):
34 Sampang 6552.3 6552.3 0.0 1.559 1.494 0.065 0.07 0.05 0.02

In this present study, the model results of extensively in both electronic and print media.
surge height have been quantitatively and They reported that the prone areas to the
qualitatively verified with the TOPEX impact of the storm tide flooding in November
Poseidon SLA. In addition, the simulated 2007 were the northern coasts of Jakarta,
results have shown the coastal areas of the Brebes, Tegal, and Semarang where those
northern coasts of Java which are most prone areas are similar to the several inundation
to the impact of storm tide flooding. The areas of storm tide obtained from this study,
present simulated results of the susceptible namely Penjaringan and Cilincing at Jakarta
areas to the impact of storm surges are (marked by 5 and 6 in Figure 5, respectively),
qualitatively similar to those reported Losari (near the Brebes, marked 15 in Figure

175
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

5), Tegal (marked 16 in Figure 5), and existed at the Semarang (marked 19 in Figure
Semarang (marked 19 in Figure 5). 5), about 1278.3 m with run-up height at the
location of maximum inundation of about
The waves generated by the tides and cyclones 0.570 m. In East Java Province, it occurred at
attacked and inundated coastal areas of the Surabaya and Sampang (marked 26 and 34 in
northern coasts of Java. In the Banten Figure 5) with the distance of flooding and
Province, maximum distance of storm tide water run-up height (at Surabaya) about
flooding existed at the Banten Bay (marked 2 5305.4 m and 2.099 m, respectively.
in Figure 5), about 558.5 m with run-up height Meanwhile, at Sampang they are about 6552.3
at the location of maximum inundation of m and 1.559 m, respectively (Table 1).
about 0.483 m. In the Jakarta Province, the
maximum distance of storm tide flooding In Figures 7 and 8, we present examples of
reached about 6275.3 m at the Penjaringan simulated inundation area and the total water
(marked by 5 in Figure 5) with run-up height depth along the storm tide inundated area at
of about 0.537 m. In the West Java Province, it the Penjaringan (marked 5 in Figure 5) and at
occurred at the Cilamaya (marked by 10 in the Semarang (marked 19 in Figure 5),
Figure 5) with the distance of flooding and respectively. In the Figures, the boundaries of
water run-up height about 2967.6 m and 0.535 the simulated inundation area after the storm
m, respectively. In Central Java Province, tide event are shown as well.
maximum distance of storm tide flooding

5958.7 (a) 6275.3 (b)

H1= 0.483
H1= 0.537
5463.1 5606.9
3057.4
3057.4

H3= 0.473 H3= 0.532

H2= 0.491 H2= 0.543

(c)

Figure 7. The simulated inundation area (distance and height of flooding, in meters) at the Penjaringan (marked 5 in Figure 5), at
7:15 UTC on 26 November 2007. The figure shows topography contours of 0 and 3 m measured from the mean sea level (MSL);
(a). Inundation area due to tides; (b) Inundation area due to storms and tides; and (c) Total water depth of the simulated storm tide
inundation area (in meters)

176
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

(a) 1278.3
(b)
972.4

H1= 0.472
H1= 0.570
534.4 534.4

H2= 0.472 H2= 0.569

(c)

Figure 8. The simulated inundation area (distance and height of flooding, in meters) at the Semarang (marked 19 in Figure 5), at
19:30 UTC on 30 November 2007. The figure shows topography contours of 0 and 3 m measured from the MSL; (a). Inundation
area due to tides; (b) Inundation area due to storms and tides; and (c) Total water depth of the simulated storm tide inundation area
(in meters)

IV. CONCLUSIONS ACKNOWLEDGEMENTS

This paper investigated wave height, run-up, Parts of this research were funded by KK
and inundation of storm tide along the Research of ITB in 2010. We gratefully
northern coasts of Java, generated by tides and acknowledge the ITB supports. We also
the Cyclones Hagibis and Mitag in November greatly appreciate assistance from D. A.
2007 by using 2D hydrodynamic model. The Utami, N. Rakhmaputeri, and H. Timotius
present simulated results showed that although whose contribution to literature about storm
the Cyclones Hagibis and Mitag did not surges aspects have been invaluable
generate high storm surges along the coasts of
Java (surge heights < 20 cm), the coasts were REFERENCES
still severely flooded. The highest surge [1] Arcement, Jr GJ, and Schneider, VR
existed at the Losari (norther coast of Central (1984). “Guide for selection Manning’s
Java). Meanwhile, the maximum distance of roughness coefficients for natural channels
storm tide flooding occurred at the Banten Bay and flood plains,” Technical Report, US
(Banten Province), the Penjaringan (Jakarta Department of Transportation, Federal
Province), the Cilamaya (West Java Province), Highways Administration Reports no.
the Semarang (Central Java Province), and the FHWA-TS-84-204, 1984, 62 p.
Surabaya and Sampang (East Java Province). [2] Chambers D. (1988). “Monitoring El Niño
with satellite altimetry”, The University of
Storm tide investigation performed in this Texas at Austin, Center for Space
study is important for simulation of areas Research. Webpage:
prone to the impact of storm tide flooding, http://www.tsgc.utexas.edu/info
such as the northern coasts of Java. The results /teamweb.html
of this present study could be significantly [3] DHI Water and Environment (2005).
valuable for designing both proper “MIKE 21 & MIKE 3 flow model FM,”
management plans and investment policies in Hydrodynamics and Transport Module,
the northern coasts of Java. Scientific Documentation, Denmark.

177
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

[4] IOC, IHO, and BODC (2003). “General wind-induced anomalies in the South-
bathymetric chart of the oceans,” China Sea”. Paper presented at the Asia
Centenary Edition of the GEBCO Digital Oceania Geosciences Society (AOGS)
Atlas, BODC, Liverpool, U.K. 2009 Annual General Meeting, Singapore.
[5] Padman, L, and Erofeeva, S (2005). “Tide [7] SRT (2007). “Shuttle Radar Topography
model driver (TMD) manual,” Earth & Mission X SAR/SRTM,” Webpage:
Space Research. http://www.dlr.de/srtm.
[6] Tkalich, P, Kolomiets, P, and
Zheleznyakm M (2009). “Simulation of

178
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Fuzzy Finite Difference on Calculation of an Individual’ Bank Deposits

Novriana Sumarti and Siti Mardiah

Industrial and Financial Mathematics Research Group, Institut Teknologi Bandung, Indonesia
Email: novriana@math.itb.ac.id

ABSTRACT determination of factors in macroeconomics

Fuzzy analogues of problems in financial [7].
mathematics had been developed initially more
than two decades ago by Buckley [1], such as the In this paper, the fuzzy set theory is firstly
fuzzy present and future values of fuzzy cash explained in section II, and also some
amounts, using fuzzy compound interest rates. Due
definition of macroeconomics’ factors which
to the fact that future cash amounts and interest
rates are quite often in estimation values, the fuzzy potentially give effect on the value of money.
approach results in solutions whose values in
intervals where they might be in. II. FUZZY APPROACH ON FINANCE
Now we discuss fuzzy membership function
In this paper we implement fuzzy finite difference and fuzzy arithmetic operations used in this
formulae in calculating the balance of an research. A fuzzy triangular number
individual’s bank deposits [2]. It involves not only
compound interest, but also some factors of
A  (a1 , a2 , a3 ) can described by its
macroeconomics which cause the change in the membership function
value of money, in different time periods. For  x  a1
instance the inflation, the taxation, the purchasing  a  a , a1  x  a2
power of the capital, etc. Real data taken from an  2 1
employee of a company has been used to estimate  a x
the time when some significant plans can be  A ( x)   3 , a2  x  a3 (2)
conducted in the future.  a3  a2
0, others.
KEYWORDS 
Financial Mathematics; finite difference; 
compound interest rate; macroeconomics; Fuzzy It can be also described by its α-cuts method,
set theory. A ( )  [ AL ( ), AR ( )] , (3)

I. INTRODUCTION where AL ( ) and AR ( ) are functions of

One of main characteristic of fuzzy system is α‫[ א‬0, 1],
its suitableness for uncertain or approximate AL ( )  a1   (a2  a1 ),
reasoning, especially for the system with a (4)
mathematical model that is difficult to derive. AR ( )  a3   (a2  a3 ).
Furthermore, fuzzy logic allows decision A
making with estimated values under
incomplete or uncertain information. Since it
was proposed by Lotfi Zadeh in his paper α
[3], there were many application of fuzzy
set theory made on decision making process
in engineering, economics and financial a1 a2 a3 x
A L ( ) AR ( )
mathematics. Pioneer works in financial
Mathematics had been done by Buckley [1], 
Figure 1: Description of a fuzzy number A
explaining the fuzzy point of view of some
important definitions and equations of Denote B  [ BL ( ), BR ( )] another fuzzy
mathematics finance. Nowadays there are
many researches on the implementation of number. The membership function of numbers
fuzzy logic in this area, such as in option resulted from arithmetic operations between
pricing using Black Scholes formula [4], real those fuzzy numbers used in this paper are
options valuation [5], insurance [6], and determined by a method [2, 9, 10] equivalent

179
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

to the extension principle method, which is III. FINITE DIFFERENCE OF THE

written below. BALANCE OF BANK ACCOUNT
A()  B()  (AL()  BL(), AR()  BR()) ,
The balance of a bank deposit increases due to
A() B()  (AL() BL(), AR() BR()) , interest paid the bank. If the bank applies
A() B() [min{A ()B (), A ()B (), compound interest, the balance calculated by
L L L R
finite difference (recursive) formula [2] is
AR()BL(), AR()BR()},max{AL ()BL(), ~ ~ ~
F n 1  (1  I ) Fn , n  N (5)
AL ()BR(), AR()BL(), AR()BR()}], ~
where F n is the fuzzy amount of the account
If 0  b ( ) for all α, then ~
A()  1 1 1 balance at time n, I is the fuzzy compound
 A()  [min{AL () , AL () ,
B() B() BL () BR () interest rate. Using (3) and write
1 1 1 .
Fn ( )  Fn  [ Fn, L , Fn, R ] , we get
AR () , AR () },max{AL () ,
BL () BR () BL () [ Fn, L , Fn, R ]  [ F0, L , F0, R ][1  IL ,1  IR ]n ,
1 1 1 with   [0,1] . Rewrite all fuzzy numbers
AL () , AR () , AR () }],
BR () BL () BR ()
in the form of (4), such as F  (F, F1, F2)
One way to write fuzzy numbers [8] is in the
difference between the middle number with and I  (I, I , I ) , the above formula becomes
1 2
the left and right numbers. For example, 
Fn  [ Fn, L , Fn, R ] 
 
A  (a2 , a2 a1, a3 a2 ) or simply
[(F1 (  1)  F )(1  ( I1 (  1)  I )) n ,
A  (a,1,2 ) . (4)
( F  F2 (  1)  F )(1  (I 2 (  1)  I )) n ].
In this paper we will calculate the balance of
an individual bank account during a period of Now we consider that the balance can decrease
time where some macroeconomics factors give due to effects of macroeconomics factors, the
influence on the value of money. Those factors inflation J , the tax portion p on the interest
being considered here are the inflation, the
rate and the CPI H . To maintain the balance,
taxation and the purchasing power of the
capital. Inflation is an increase in the price there are deposits bn added during the life of
of a basket of goods and services that is the account. If money income stays the same,
representative of the economy as a whole. but the price of most goods goes up, the
We use data of Indonesia’s inflation effective purchasing power of that income
during 2008 – 2010. Taxation is a charge falls. Falling purchasing power can thus be
taken on the subjects of a state by the part of inflation. Let us denote real purchasing
power with
government, or on the members of a
corporation or company, by the proper F
Yn  n
authority. Taxation also applies on the H
interest of bank deposits if the balance is so the finite difference will be
~ ~ ~
more than IDR 7.5 million. Finally, the ~ (1  (1  p ) I ) Fn  b(1  p)
purchasing power is the value of money, as Y n 1  ~ ~
measured by the quantity and quality of H (1  J )
products and services it can buy. Data used ~ ~
(1  (1  p) I ) ~ b(1  p)
to show the purchasing power is the  ~
Yn ~ ~
, n  N.
consumer price index (CPI) from Indonesia (1  J ) H (1  J )
during 2008-2010. Using the same procedure explained
previously, we get

180
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

[  Yn ,l ,  Yn , r ]   b1 =(2.336,3, 4.672) if n=jFitr

b = 50,54,56 if n=7 after 3 years
1  (1  p )  I l 1  (1  p )  I r n    
[ , ] [ Y0,l ,  Y0,r ] bn   2
1  Jr 1   Jl  b1  b2 if n=jFitr =7 after 3 years
 0 if others
(1  p )  bl (1  p )  br
[  , ]
H r (1   J r )  H l (1   J l ) where jFithr is the month of his religious
n 1
1  (1  p )  I l 1  (1  p )  I r n  s 1 celebration day. The month of jFithr is

s 0
[
1  Jr
,
1   Jl
] .
started from Sep 2010, so jFitr =9. From
year 2 and so on, the month of jFitr will be
(6)
8, 8, 8, 9, 9, 9, 10, 10,10,... This
Detailed steps in the above process can be
found in [2]. In the next section we implement
arrangement agrees with Islamic calendar
the above formula in the design of important where the Eid Fitr’ will be coming earlier
plans of an individual life in the future. 11 days from the day in the previous day.

IV. IMPLEMENTING SCENARIO Fuzzification of macroeconomics data as in

A scenario is designed to see the importance the Appendix is given below. The interest rate
of the finite difference formula (6) in the of the mortgage is In =(0.0525,0.060722,
previous section. The chosen value of α is 0.8.
Data is taken from the main salary and 0.0725), the CPI is H =(0,1008 ; 0,230846429
additional income of a fresh-graduate ; 0,6401) and the inflation J =(0.0214,
employee of a big company owned by 0.10994, 0,1214).
government. His take-home salary is IDR The balances of the bank account are shown in
4,672,000 per month. In the month of his tables below. In table 1, the tax applies on 4th-
religious celebration day, Eid Fitr’ for his case, month of year 1 and first type bonus is added
there is one month salary added to his account. on 9th-month.
After 3 (three) years since his first day at the
office, there will be a one-year bonus paid by
the employer on every June. Yi,j Yn_Left Yn_Right

He commits to himself to save around IDR 1,1 2,202,449.15 2,396,750.57

2,700,000 per month from the main salary. If 1,2 4,512,910.71 4,756,459.52
the bonuses are received, he will save the 1,3 6,854,644.18 7,125,149.31
amount of 50% to 100% of the total bonuses. 1,4 9,217,341.37 9,517,090.09
With this financial condition, he plans to
1,5 9,748,395.30 13,768,894.82
spend his money to buy a house with price
IDR 350,000,000 and the expense of his 1,6 13,520,184.90 14,630,674.51
wedding party worth around IDR 100,000,000 1,7 16,143,712.89 16,712,726.06
to 200,000,000. 1,8 18,543,693.72 19,051,488.84
1,9 19,251,736.06 25,328,515.42
The fuzzy finite difference formula (6) is used
to set time when he could take a mortgage for 1,10 21,570,017.26 31,040,733.39
Table 1: Year 1
buying a house and the time when he will hold
a quite big wedding party without needs of Yi,j Yn_Left Yn_Right
financial help from his parents and relatives.
Assume that he starts to save from January 4,4 113,081,653.29 120,230,577.44
2010. 4,5 116,592,656.94 124,114,841.36
4,6 128,503,202.85 159,670,382.92
Now we define parameters: (in million IDR)
4,7 57,764,208.52 79,167,635.88
Fn =(2, 2.7,3) , n = 0,1,2,...
4,8 55,291,707.38 84,054,477.35
4,9 54,541,915.90 85,561,648.29

181
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

4,10 64,213,442.80 73,464,253.45

4,11 66,849,938.85 71,820,082.13 REFERENCES
4,12 68,131,492.52 72,359,956.12 [15]. J.J. Buckley, The fuzzy mathematics
Table 2: year 4
of finance, Fuzzy Sets and Systems 21 (3)
The house mortgage with rates as shown in the
(1987) 257–273.
Appendix needs IDR 100 million as a down-
[16]. K.A. Chrysafis, B.K. Papadopoulos,
payment and monthly payment as much as
G. Papaschinopoulos, On the fuzzy
[2,183,165.914 , 2,264,552.576] for 10 years.
difference equations of finance, Fuzzy Sets
Notice that he needs to maintain his balance in
and Systems 159 (2008), 3259–3270.
decent amount after he takes the mortgage and
[17]. Zadeh, L.A. (1965). "Fuzzy sets",
pay the monthly installment. Taking it as a
Information and Control 8 (3): 338–353.
consideration, he takes a mortgage on 7th-
[18]. Lee, CF, Tzeng GH and Wang SY, A
month of year 4 as shown in table 2.
new application of fuzzy set theory to the
Black–Scholes option pricing model,
Yi,j Yn_Left Yn_Right Expert Systems with Applications, 29
8,5 242,817,812.55 271,025,964.98 (2005) 330–342.
8,6 209,654,062.28 289,218,014.15 [19]. Carlsson, C., A fuzzy approach to real
8,7 78,403,973.47 123,056,558.28 option valuation, Fuzzy Sets and Systems,
139 (2003) 297–312.
8,8 77,465,352.27 126,858,691.37 [20]. Shapiro, A.F., Fuzzy logic in
8,9 92,800,394.98 111,628,988.83 insurance, Insurance: Mathematics and
8,10 99,131,867.71 112,174,729.18 Economics, 35 (2004) 399–424.
8,11 103,805,191.59 116,137,939.78 [21]. C. Kahraman (editor), Fuzzy
Engineering Economic with Applications,
8,12 108,243,959.32 120,921,526.88
Spinger-Verlaag Berlin, 2008.
9,1 112,726,452.11 125,982,250.51 [22]. W.H. Steeb, Y. Hardy, R. Stoop, The
9,2 117,309,594.02 131,209,342.39 Nonlinear Workbook, Singapore (2005)
9,3 122,007,626.59 136,583,243.21 554-555.
9,4 126,826,653.75 142,103,744.28 [23]. G.J. Klir and G. Yuan, Fuzzy Sets and
Fuzzy Logic, Theory and Applications,
9,5 131,771,322.29 147,774,853.57 Prentice-Hall, New Jersey, 1995.
9,6 151,538,633.97 198,691,011.26
9,7 152,570,638.09 243,496,494.54
Table 3: Year 8 and 9
In 6th-month of year 8, the money is enough
to pay back the mortgage, so it occurs in the
7th-month. Finally, he can hold a wedding
party starting on 6th-month of year 6, which is
shown in table 3.

V. CONCLUSION

Fuzzy finite difference equation can be used to

determine the time when the balance of the
bank account is suitable to take a mortgage
and to spend it in a wedding party. In the case
observed in this research, a fresh-graduate
employee could buy a house of IDR 350
million in year 4, and holding a wedding party
of IDR 100 – 200 million in year 9.

182
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Appendix

Mortgage rates Month-Year Inflation CPI

Bank Des Jan Feb April 2010 3.91 % 0.1378
2009 2010 2010
BTN (Kredit March 2010 3.43 % 0.1402
Ringan February 2010 3.81 % 0.1427
Batara) 9.75% 10.50% 9.75% January 2010 3.72 % 0.1392
BTN
(platinum) 11.25% 11.75% 11.25% December 2009 2.78 % 0.1397
BTN (Kredit November 2009 2.41 % 0.141
Griya Utama) 12% 12% 12%
October 2009 2.57 % 0.1461
BRI 12.25% 12.25% 12.25%
September 2009 2.83 % 0.1525
BCA 10.05% 10.50% 10.05%
August 2009 2.75 % 0.1646
BII 11.49% 11.49% 10.49%
July 2009 2.71 % 0.1668
CIMB
NIAGA 10.05% 10.50% 10.50% June 2009 3.65 % 0.1665
OCBC NISP 12.25% 12.25% 12% May 2009 6.04 % 0.1703
PERMATA 9.90% 12.25% 9.90% April 2009 7.31 % 0.1801
MEGA 13.99% 13.99% 13.99% March 2009 7.92 % 0.1836
MANDIRI 11.05% 11.75% 11.05% February 2009 8.60 % 0.1819
BNI 11.50% 11.50% 11.50% January 2009 9.17 % 0.1837
DANAMON 11.99% 11.99% 11.99%
Bank account
PANIN 10.80% 10.80% 10.80% Year Interest rates
5.25%
5.75%
2009
5.50%
6%
6.25%
6.50%
2010 6.75%
7%
7.25%

183
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

An Implementation of Fuzzy Linear System in Economics

Novriana Sumarti and Cucu Sukaenah

Industrial and Financial Mathematics Research Group, Institut Teknologi Bandung, Indonesia
Email: novriana@math.itb.ac.id

ABSTRACT economists by [5], where it measures

Many problems in mathematical models can be some concept of inflation,
written in linear systems. Especially in unemployment, recession, market-
numerical analysis, all problems in differential dominating companies, money, the gross
equations are usually rewritten in linear
national product (GNP), the national
systems so they can be solved numerically. In
modelling real world problems involving income, investment, households, and so
vagueness and imprecision, fuzzy approach on. In [6], the efficiency of the fuzzy
could tolerate facts which are quite often approach in economics put into an
captured in estimation values. observation and the results was positive
that the approach can make a significant
In this paper we implement the methods of contribution to the foundations of
solving system equations Ax  b where econmics. Some other works in
values of every element of matrix A and vector implementing fuzzy analysis in finance
b are in intervals. In finding suply and demand and economics can be found in [2,7,8].
curves based on data on Indonesia’s rice
production during 1987 - 1997, we build fuzzy In this paper, we explain concepts of fuzzy
linear systems using a method of fuzzy linear linear system and linear regression in
regression on observed data. The clasic
section II. An implementation of concepts
solution will not found from these linear
systems. Based on a method in [1,3], a vector in the preceeding section is available in
solution can be found using α-cuts method and section III where the demand and supply
Cramer’s rule so the equilibrium point between curves of data on Indonesia’s rice
supply and demand data can be determined. production during 1987 – 1997 are built.
In the process of finding their equilibrium
KEYWORDS point, it occurs that the classic solution
Linear systems; Fuzzy arithmetics; α-cuts cannot be found, so it needs the method of
method; Cramer’s rule; Economics; vector solution which is expained in
section IV. Finally, the chosen equilibrium
I. INTRODUCTION point and its interpretation in economics
are explained in section V.
The fuzzy sets theory was specifically
design to represent mathematically II. FUZZY LINEAR SYSTEM AND
uncertainty and vagueness on REGRESSION METHOD
quantification of real world problems and
to provide formalized tools for dealing Consider a fuzzy linear system of
with the imprecision. Due to the main equations of the form:
characteristics of the fuzzy system, which Ax  b (1)
are its suitableness for uncertain or
where the elements A ij and B j , i,j = 1, . .
approximate reasoning, and its
acknowledgment of decision making . , n, of the matrix A and vector b
with estimated values under incomplete or respectively are triangular fuzzy numbers.
uncertain information, it has been used in A fuzzy triangular number
decision making process in engineering A  (a , a , a ) can described by its
1 2 3
since it was proposed by Lotfi Zadeh in membership function
his paper [4].
In economics, the fuzzy sets theory was
firtsly used to bridge the gap between
mathematical- and language-oriented

184
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

 x  a1 Y  A0  A1 X (5)

 a  a , a1  x  a2
 2 1 where parameters Ai , i  0,1 are
 a x estimated from the data. Here we need
 A ( x)   3 , a2  x  a3
 a3  a2
two linear equations of the above form. In
0, others. the first equation, Y and X are
 respectively the quantities of the supply
 and prices of a good. In the second
(2 equation, Y and X are respectively the
) quantities of the demand and prices of a
It can be also described by its α-cuts good. At the end, we will find the
method, equilibrium point between the quantities
A ( )  [ AL ( ), AR ( )] , (3) of suply and demand.
where AL ( ) and AR ( ) are functions
According to [10], parameters in (5) are
of α‫[ א‬0, 1], found by minimizing the fuzzy least
AL ( )  a1   (a2  a1 ), square, that is the total square error of the
(4) output. The objective function to be
AR ( )  a3   (a2  a3 ).
minimized is
A
J   (YˆR  YˆL )dM X dRX (6)

where YˆL , YˆR are the estimator of YL , YR .

α
Quantities M X and R X are midpoint and
radius from the coressponding fuzzy
a1 A L ( )
a2 AR ( )
a3 x number X , where
Figure 1: Description of a fuzzy number A X L  X R X  X L
MX  , RX  R .
2 2
Denote B  [ BL ( ), BR ( )] another The constraint of the optimization
fuzzy number. The membership function problem (6) is below
of numbers resulted from arithmetic
operations between those fuzzy numbers
M A0  M A1 M X j  RA1 RX j  M Yj 
used in this paper are determined by a
method [2,9,10 ] equivalent to the RA0  M A1 RX j  RA1 M X j  RYj
extention principle method, which is where ( X j , Yj ), j  1, , N are
written below.
A() B() (AL() BL(), AR()  BR()) , observed data samples. Detailed
explaination about the above method
A() B()  (AL() BL(), AR() BR()) , can be found in [10]. To solve the above
A() B() [min{A ()B (), A ()B (), problem, there are many optimization
L L L R
methods available. However, in this paper
AR()BL(), AR()BR()},max{AL()BL(), we use the simulated annealing method
AL()BR(), AR()BL(), AR()BR()}], which is explained in detail and
implemented on another optimization
problem in [11].
If 0  b ( ) for all α, then A()  A()  1 .
 B()  B() III. SOLVING FUZZY LINEAR
SYSTEM CLASICALLY
Now we discuss fuzzy linear regression in
order to analyze the linear relationship Data in the Appendix taken from Statistik
between variables, in this case, Y and X , Bulog 1998 are fuzzificated. An example
in equation of the resulted fuzzy numbers of the rice

185
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

production can be found in the Appendix. where k  [0,1] , k=1,2,3,4 for A and
The numbers are also scaled to simplify
k=5,6 for B . If dimension of A  is n x n,
the computation. The results are put into
the regression process and then we then k = 1,2,..., n  n . The system (7)
2

determine the equilibrium point of the becomes

demand and supply curves. The two  A

   
A   x    B  
12 
resulted curves of the form (5) are 11

1
  1

rewritten into linear equation system  A

  21    
A   x2    B 2 
22   
below
(8)
Use Cramer’s rule to solve system (8) .
X 1  (0.6125, 0.5200, 0.4275) X 2  The solutions of (8) are functions in
(0.3817, 0.4559, 0.5301)   [0,1] . The plot of x1 and x 2 for all
(7)
X 1  (0.3157, 0.1615, 0.0073) X 2  values of α can be found in figure 1 and 2
respectively.
(0.6833, 0.7357, 0.7881)
where X 1  ( x11 , x12 , x13 ) is the quantity
1

0.9
of the rice production in the situation in 0.8
which the suply and demand are the same,
and X 2  ( x21 , x22 , x23 ) is the price in that
0.7

0.6
situation. Here we need the condition
alpha-cuts
0.5
xi1  xi 2  xi 3 , i  1, 2 . Writing in its α- 0.4

cuts representation (3) and using 0.3

arithmetic operations in section II, we find 0.2

that x11  1.0026, x13  0.7926, 0.1

x21  0.6140, x23  1.0138 . Here 0

2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5

x11  x13 which is not acceptable. So the

Quantity of Product 10
x 10

Figure 1: Quantity of product on equilibrium point

clasic solution of fuzzy linear system is
not available. 1

0.9

IV. VECTOR SOLUTION METHOD 0.8

Buckley [1] proposed a new method called 0.7

the vector solution method. First we define 0.6

alpha-cuts

new linear equation system of the same 0.5

dimension coming from restasting all 0.4
fuzzy numbers in the system (7) into the
0.3
corresponding forms of (4). Notice that
system (7) involves a fuzzy matrix A  of 0.2

2x2 dimension and a fuzzy vector B of 2

0.1

0
dimension. Using (4), denote 0 200 400 600 800 1000
Price
1200 1400 1600 1800

Aij ( )  [aijL ( ), aijR ( )] Figure 2: Price on equilibrium point

 . Here
an ij-th element of matrix A After scaling back to the original
aijL ( )  aij1   (aij 2  aij1 ), magnitude and choose α = 1, it conclude
that the quilibrium point occurs at the
aijR ( )  aij 3   (aij 2  aij 3 ). production of 3.2083 1010 kg and the
Define new elements for i,j = 1,2 price of IDR 378.1.
 A   aijL ( )  [aijR ( )  aijL ( )]k and
ij
V. CONCLUSIONS
 B   bijL ( )  [bijR ( )  bijL ( )]k ,
ij
The classic solution of fuzzy linear system
can be non exist. The alternative method is
calculating a vector solution by using α-

186
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

cuts method and Cramer’s rule. Finding

the equilibrium point of the suply and APPENDIX
demand curves on the data of rice
production during 1987 – 1997 is a good Production (kg) Price (IDR)
example of the implementation of the (2.259e+010,2.404e+010, (140.6, 175,
alternative method. 2.552e+010) 209.4)
(2.404e+010,2.552e+010, (175, 209.4,
REFERENCES 2.698e+010) 243.8)
(2.552e+010,2.698e+010, (175, 209.4,
[24]. J.J. Buckley, Y. Qu, Solving 2.843e+010) 243.8)
systems of C. Kahraman (editor), (2.552e+010,2.698e+010, (209.4, 243.8,
Fuzzy Engineering Economic with 2.843e+010) 278.1)
(2.552e+010,2.698e+010, (243.8, 278.1,
Applications, Spinger-Verlaag Berlin,
2.843e+010) 312.5)
2008. (2.698e+010,2.843e+010, (243.8, 278.1,
[25]. C. Ponsard, Fuzzy mathematical 2.991e+010) 312.5)
models in economics, Fuzzy Sets and (2.698e+010,2.843e+010, (312.5, 346.9,
Systems 28 (3), 1988, 273-283. 2.991e+010) 381.3)
[26]. G.J. Klir and G. Yuan, Fuzzy Sets (2.698e+010,2.843e+010, (312.5, 346.9,
and Fuzzy Logic, Theory and 2.991e+010) 381.3)
Applications, Prentice-Hall, New (3.284e+010,3.43e+010, (312.5, 346.9,
Jersey, 1995. 3.577e+010) 381.3)
[27].A. Bisserier, R. Boukezzoula, S. (3.43e+010,3.577e+010, (381.3, 415.6,
Galichet, An interval approach for 3.723e+010) 450)
fuzzy linear regression with imprecise (3.43e+010,3.577e+010, (415.6, 450,
data,IFSA-EUSFLAT 2009, 1305- 3.723e+010) 484.4)
1310.
Indonesia’s rice production and consumption
[28].P. Luic, D. Teodorovic, Simulated 1987 - 1997
annealing for the multi-objective
aircrew rostering problem, Production Consumption
Transportation Research Part A 33(1) Year (Kg) (Kg) Price
1999, 19-45.
1987 24047398200 23932084680 175
[29]. linear fuzzy equations, Fuzzy Sets
and Systems 43 (1991) 33–43. 1988 25006390800 24650879870 210
[30]. J.J. Buckley, The fuzzy 1989 26834862600 27670245620 210
mathematics of finance, Fuzzy Sets 1990 27107762400 25938534960 250
and Systems 21 (3) (1987) 257–273.
1991 26811343200 25821981000 270
[31]. Muzzioli S. and Reynaerts H.,
Fuzzy Linear System of the form , 1992 28945521000 29962225150 295
Fuzzy set and systems 157 (2006) 939- 1993 28909125000 27245040800 330
951. 1994 27983538000 26833493200 340
[32]. Zadeh, L.A. (1965). "Fuzzy sets", 1995 34823747700 27566148280 360
Information and Control 8 (3): 338–
1996 35767191600 31328276850 400
353.
[33]. A. Pfeilsticker, The systems 1997 35438970000 27721333500 450
approach and fuzzy set theory
bridging the gap between Fuzzification on the production data
mathematical and language-oriented
economists, Fuzzy Sets and Systems 6
(3) 1981, 209-23.
[34]. M. Fedrizzi, M. Fedrizzi, W.
Ostasiewicz, Towards fuzzy modelling
in economics, Fuzzy Sets and Systems
54(3) 1993, 259-268

187
 Compact Finite Difference Method for Solving Discrete Boltzmann
Equation

PRANOWO1 & A. GATOT BINTORO2

1
Senior Lecturer of Informatics Engineering, Atma Jaya University, Yogyakarta, Indonesia.
(Email: pran@mail.uajy.ac.id)
2
Lecturer of Industrial Engineering Atma Jaya University, Yogyakarta, Indonesia

ABSTRACT
Fourth compact finite difference (FD) method for
solving two dimensional Discrete Boltzmann
f
 e  f  
f  f eq   (1)
Equation (DBE) for simulation of fluid flows is t 
proposed in this paper. The solution procedure is
carried out in Eulerian framework. BGK eq
where f is the equilibrium distribution and
(Bhatnagar–Gross–Krook) scheme is adopted to
approximate the collison term. The convective
 is relaxation time. After discretizing the
terms are discretized using 4th compact finite velocity space e into various directions, the 2-
difference method to improve the accuracy and D Boltzmann equation for the velocity
stability. Te semidiscrete equations are updated distributon function f i may be written as
using 4th order explicit Runge-Kutta method.
Preliminary results of the method applied on the discrete Boltzmann equation.
Taylor-Green vortex flows benchmark are
presented. We compared the numerical results f i
 e i  f i   i

f  f i eq
(2)

with other numerical results, i.e. explicit 2nd and 4th
FD, and exact solutions. The comparisons showed t 
excellent agreement. The discrete velocity e i is expressed as:

KEYWORDS
Compact finite difference; Boltzmann; BGK; ;

Taylor vortex
 0,0 , i 1
 
e i   cos  i , sin  i  ,  i  i  1 , i  2,3,4,5
I.INTRODUCTION  4
In the last decade the lattice-Boltzmann  2 cos  i , sin  i  ,  i  i  1  , i  6,7,8,9
method (LBM) has attracted much attention in  4
the simulation of fluid dynamics problems. 8
Unlike conventional computational fluid    fi (3a)
dynamics methods, which discretize the i 0

macroscopic governing equations directly, the 8

LBM method solves the gas kinetic equation at u j   f i eij ; (3b)
the mesoscopic scale, i.e. the discrete i 0
Boltzmann equation with the Bhatnagar–
Gross–Krook (BGK) relaxation for the  e .u e .u 2 u 2 
collision operator. The BGK relaxation fi eq
 i  i 2  i 4  
 cs 2c s 2 
process allows the recovery of Navier Stokes  
equations through Chapman Enskog expansion
for low Knudsen number. with 1  4 9 , 2  3  4  5  1 9 ,
and 6  7  8  9  1 36 . The
In the gas kinetic theory, the evolution of the
pressure can be calculated from p  cs  with
3
single-particle density distribution function
f t, x,e which represents the probability of sound velocity c s  1 3 in lattice unit
density of a particle with unit mass moving
with velocity e at point x at time t, is
governed by the Boltzmann equation:

188
  II. DISCRETIZATION
and the kinematic viscosity of fluid is   . The linear convective terms of equations (2)
3 are discretized using 4th compact finite
.
difference method:

1  f i  2  f  1  f 
    i   i 
6  x  k 1,l 3  x  k ,l 6  x  k 1,l
(5a)
f k 1,l  f ik ,l
2x

1  f i  2  f  1  f 
    i   i 
6  y  k ,l 1 3  y  k ,l 6  y  k ,l 1
(5b)
Figure 1. Velocities in 2-D Lattice  f i k ,l 1   f i k ,l 1
Boltzmann model (D2Q9)
2 y

In Lattice Boltzmann method eq. (2) is solved

in the form of

f i x  e i , t  1  f i x, t  
(4)

1

 f x, t   f x, t 
i i
eq

using x  y  t  1 . The use of unit

square mesh elements is restrictive. Several
extension to the LBM have been developed to
overcome this restriction. Reference [1] used
finite difference method (FDM) with 2nd Figure 2. Finite Difference Stencil
upwind discretization for convective terms. In
ref. [1] the FDM is extended to curvilinear After discretizing the convective terms using
coordinates with non-uniform grids. 4th compact FD, we obtain semi discrete
Unfortunately the 2nd upwind makes the stencil equation of (2).
longer, so it is not easy to handle the boundary
condition. Reference [2] used FDM on non- f i
 e i  f i 
 f i  f i eq 
 L f i  (6)
uniform grids. They used implicit temporal
discretization to improve the stability. Many
t 
modified FDM were proposed to improve the
stability and numerical accuracy. Upwind Then the time update is performed using
FDM suffers from large dissipation error and classical 4th explicit Runge Kutta method.
standard 2nd suffers from large dispersion
error. Spectral method [5] offers exact III. ANALYSIS OF DISCRETIZATION
differentiation but suffers from low flexibility The analysis of spatial and temporal
in treatment of boundary condition. discretization are given in this section. For
simplicity, linear advective equation is takes as
In this paper, the 4th order compact FDM is example.
proposed to discretize the convective terms of
u u
eq. (2). The method is preffered due to high a  0, x  0,2 
accuracy and flexibility [4]. For improving the t x (7)
stability, the 2nd explicit Runge Kutta method u x,0  e ikx
is used to integrate the semi-discrete equation.

189
where a is velocity constant. f i
 L f i  (10)
The exact solution of eq. (7) is easily
t
computed, and we have
where L is the residual terms which contains
the spatial terms of the governing equations.
ux, t   ei kxt  . (8)
By subtituting f i  f i t e into eq. (6), we
kx

where k is the wave number and  is angular obtain

frequency. From the exact solution, we
obtained the exact dispersion relation: f i
 f (11)
  ak . By subtituting the local solution t
u i x, t   U i ei k  to eq. (6), we obtain
* i
x  t

the numerical dispersion relation [5]: Where f is the Fourier coefficient, i  1

, and  is complex number. The left terms of
  eq. (11) is expanded by using 4th explicit
 1.5 sin kx  
Runge-Kutta scheme to obtain the
ik x  i
*  (9)
 2  amplification factor:
 1  coskx  
 4 
f i n 1 1
 1  t   t 
2
G (12)
From the eq. (9), it can be seen that numerical fi n
2
dispersion relation of 4th compact FD has no
imaginair components, so it can be concluded
The stability condition requires that the
that 4th compact FD is conservative and has no
amplification factor must be bounded,
dissipation error. For acceptable dispersion
error k * x  kx  10 2 , we found that
G 1 (13)
k x  1.0893. Therefore, the spatial grid
*

points per wavelength

(PPW) is
PPW  2 1.0893  5.7683. From the
right term of eq. (9) the eigenvalues of 4th
compact FD can be calculated and they are
purely imaginary, covering
 1.732a x,1.732a x

Figure 4. The stability region in complex plane

The stability of 2nd explicit Runge-Kutta The stability region in complex plane can be
scheme can be analyzed by considering eq. seen in figure 4. The eigenvalues of 4th RK in
(6), Figure 3. Dispersion error imaginary axis are covering

190
 2.828,2.828t , the stability condition of We compare the results with explicit 2nd and
the fully discrete equation is 4th order FD. We take constant t  5e  3

at From figure 5, we can see that 4th compact FD

CFL   2.828 1.732 has the lowest error among the r methods. The
x (14)
accuracy orders are in a good agreement with
CFL  1.6328 theoritical results.
CFL is Courant Friedrichs Lewy number. The second problem to be solved is 2-D
decaying Taylor –Green vortex flow problem
IV. NUMERICAL RESULTS [5]. The Taylor-Green vortex flow has the
The performance of numerical is tested by following analytic solutions to incompressible
application to 2 benchmark problems. The first Navier-Stokes equation in 2-D:
problem to be solved is 1-D linear convection
problem [3]:
u x  x, y, t   U 0 cosk x x sin k y y e

 kx k y 2 t

u u
 2  0, x  0,2  u y  x, y , t  
kx
U 0 cosk y y sin k x x e x y
 k  k 2 t

t x ky
ux,0  e sin x   2

U2
u x  x, y , t    0  cos2k x    k x 
 sin 2k y y 
with periodic boundary condition 4  x k  
  y  
The exact solution to above equation is a right-
moving wave of the form: 
 kx k y 2 t
e 1

u x, t   e sin  x 2t  where U 0 is initial velocity amplitude, k x and

and x  0.0982,x  0.0491,x  0.0245. k y are the wave number in x dan y direction.

Figure 5. Convergence History for linear convection

problem

Figure 7.Velocity fields at t=2

We use 2-D system of size 32  32 with

periodic boundary condition in both directions.
The simulation parameters are U 0  0.01,
k x  k y  2 , t  0.005, and   0.0018.
The initial condition of velocity distributon
Figure 6. Accuracy Order

191
function f i actually are unknown, it is not accurate than explicit 4th FD. Figure 9a and 9b
show the comparisons.
easy how to generate consistent initial
condition of f i . Research for generating
consistent initial condition of f i is still in
progress [6]. In this paper, we use a simple
approach, we use the equilibrium distribution
function f i eq to intialize f i .

Figure 9a.Vertical velocity error at t=2 and y=3.043

Figure 8. Density distribution at t=2.

Figure 7 and 8 show the computed results for

velocity field and density at t  2 .
Numerical and exact solutions of vertical
velocity for t = 2 and t = 150 are compared in
figure 8, showing excellent agreement. Figure 9b.Vertical velocity error at t=150 and y= 3.043

Figure 10 shows the evolution of averaged

error for 4th compact FD, 2nd and explicit 4th
FD schemes. It can be seen that the averaged
error of 4th compact FD and explicit 4th FD are
almost equal and the the averaged error of 2nd
FD scheme is higher than others.

Figure 8. Comparison of vertical velocity at t= 2 and

t=150

We compared the vertical velocity error of 4th

compact FD with explicit 2nd and 4th FD, the
comparisons show that 4th compact FD much Figure 10. Convergence history of Taylor-Green vortex
problem
more accurate than 2nd and slightly more

192
V. CONCLUSIONS
In this paper, we have presented a 4th compact [3]. Hesthaven, JS, Gottlieb, S, and Gottlieb D
finite difference method for solving two (2004). “Spectral Methods for Time
dimensional Discrete Boltzmann Equation. Dependent Problems,” Cambrige
The proposed method has been verified for the University Press, UK.
1-D convective equations and Taylor-Green [4]. Lele SK (1992). "Compact Finite
vortex flows benchmark. The excellent Difference Schemes with Spectral-like
agreement with exact solution and results of Resolutions," Journal of Computational
2nd and explicit 4th FD shows the excellent Physiscs, 103, pp 16-42.
accuracy and stability of the proposed method. [5]. Wilson, RV, and Demuren, A O (2004).
“Higher-Order Compact Schemes for
ACKNOWLEDGEMENTS Numerical Simulation of Incompressible
The work described in this paper was fully Flows,” NASA/CR-1998-206922, ICASE
supported by a grant from LPPM of Atma Jaya Report No. 98-13, Hampton, USA.
Yogyakarta University.. [6]. Mei, R et al. (2006). "Consistent initial
conditions for Lattice Boltzmann
REFERENCES simulations," Computers & Fluids, 35, pp
[1]. Mei, R, and Shyy, W. (1998). "On the 855-682.
Finite Difference-Based Lattice Boltzmann
Method in Curvilinear Coordinates,"
Journal of Computational Physiscs, 143,
pp 426-448.
[2]. Toelke, J. et al.(1998). "Implicit
discretization and non-uniform mesh
refinement approaches for FD
discretizations of LBGK models," Int.
Journal of Modern Physics C, 9(8), pp
1143-1157.

193
 Natural convection heat transfer with an Al2O3 nanofluids at low Rayleigh
number

(3 )
Zailan Siri(1), Ishak Hashim(2) and Rozaini Roslan
(1)
Institute of Mathematical Sciences, University of Malaya, 50603 Kuala Lumpur, Malaysia
(Email: zailansiri@um.edu.my)
(2)
Centre for Modelling & Data Analysis, School of Mathematical Sciences, Universiti Kebangsaan Malaysia,
43600 Bangi, Selangor, Malaysia
(Email: ishak_h@ukm.my)
(3)
Centre for Research in Applied Mathematics, Faculty of Science, Arts & Heritage
Universiti Tun Hussein Onn Malaysia, 86400 Parit Raja, Johor, Malaysia

ABSTRACT nanoparticles size ranging from 11-150nm. In

Natural convection flow for variable sidewalls addition, they was derived a correlation of the
temperatures in a square enclosure heated from the thermal conductivity ratio as a function of
side is investigated. The left wall has a variable hot nanoparticle size and temperature. They
temperature and the right wall has a constant cold conclude that a higher temperature and smaller
temperature. The top and bottom horizontal straight
nanoparticle size increase a thermal
walls are kept adiabatic. The fluid in the enclosure
is a water-based nanofluids containing Al2O3 conductivity nanofluids.
nanoparticles. To solve the nonlinear governing
differential equations a finite difference method is The viscosity properties of nanofluids in the
applied. It is shown that the addition of Brinkman model was shown to underestimate
nanoparticles has the effect of decreasing the the effective viscosity of nanofluids [15,19].
maximum absolute value of the stream function and The Brinkman model also does not consider
therefore natural convection decreases. the nanofluids temperature and nanoparticle
size. Beside, the Brinkman model was derived
KEYWORDS for larger particles sizes compared to the size
Natural convection; variable wall temperatures; of nanoparticles. Therefore, Nguyen et. al [15]
nanofluids; Al2O3 nanoparticles; finite difference
have responded with extensive experiment
method.
under a wide range temperatures and several
I.INTRODUCTION nanoparticles size to determine the effective
Natural convection flow in a square enclosure viscosity of nanofluids. They demonstrated
is an important phenomenon in engineering that viscosity drop sharply with temperature,
systems due to its wide applications in especially for high concentration of
electronic cooling and solar energy collections nanoparticles. The smaller nanoparticles size
[17]. Choi [3] reported that to enhance heat decrease the viscosity nanofluids. They also
transfer is by using solid particles in the base derived a viscosity ratio correlation.
fluid (i.e nanofluids) in the range of sizes 10-
50nm. Nanofluids poses as promising Abu-Nada et. al [1] have applied the
alternative for the heat transfer enhancement. correlation of thermal conductivity of Chon [5]
However, the viscosity of nanofluids was and the viscosity data given by Nguyen et. al
found increase abnormally that will suppress [15] to predict the heat transfer behaviour in a
the buoyancy current [22]. differentially heated rectangular enclosure.
They observed that heat transfer was reduced
The classical model of Maxwell-Garnett is the by increasing the volume fraction of
first model to explain the thermal conductivity nanaparticle above 5% at a high Rayleigh
behavior of the nano-scale solid particles in the number. This new finding corrects a finding of
base fluid but the model does not consider an the effect nanoparticles concentration on heat
important mechanisms for heat transfer in transfer enhancement reported by
nanofluids such as Brownian motion and effect [11,20,18,2,14,16,6, 12]. However, Abu-Nada
of temperature as well as nanoparticle size. et. al [1] tested for a fixed nanoparticles size
Chon et. al [5] was the first conducted and in the rectangular enclosure only.
experimental of Al2O3 nanofluids over a wide
range of temperature from 21-71oC and
194
The purpose of this paper is to analyze  nf  (1   )  f   s
numerically the heat transfer behaviour in a
knf
square enclosure filled nanofluids with  nf 
variable sidewalls temperatures. Al2O3 (  c p )nf
nanoparticle will be used in this study. (  c p )nf  (1   )(  c p ) f   (  c p ) s
II. MATHEMATICAL FORMULATION (  )nf  (1   )(  ) f   (  ) s .
Fig. 1 shows a schematic diagram of the
differentially heated square enclosure. The left The dynamic viscosity ratio of water-Al2O3
enclosure has a variable hot temperature (Th) nanofluids in the ambient condition was
and the right enclosure has a constant cold derived by Nguyen et. al [15] as follow:
temperature (Tc). The top and bottom  nf  nf
horizontal straight walls are kept adiabatic.  0.904e0.148 and  1  0.025  0.015 2
f f
The fluid in the enclosure is a water-base
nanofluids containing Al2O3 nanoparticles. The for 47 and 36 nm particle-size respectively.
continuity, momentum, and energy equations While the thermal conductivity ratio of water-
for the laminar and steady state natural Al2O3 nanofluids calculated by Chon et. al [5]
convection in the two-dimensional enclosure is
0.369 0.7476
can be written in the dimensional form as knf  df   kf 
follows Ghasemi and Aminossadati [8,9]  1  64.7 0.764
    Pr 0.9955 Re1.2321
kf  dp   kp 

f  f k bT
with Pr  and Re  , and the
 f f 3 2f l f
symbol kb  1.3807  1023 J / K , l f  0.17nm,
d f  0.384nm and u f  2.414  10510247.8 T 140

is the Boltzmann constant, mean path of fluid

particles, molecular diameter of water and the
temperature dependence of water viscosity
Figure 1. Physical model respectively. Thermo-physical properties of
water and Al2O3 are as Abu-Nada et. al [1] can
u v be found in Table 1.
 0 (1)
x y
Table 1. Thermo-physical properties of water and Al2O3
u u 1 p  nf   2 u  2 u  Physical properties Water Al2O3
u v      (2)
x y  nf x  nf  x 2 y 2  Cp (J/kgK) 4179 765
ρ (kg/m3) 997.1 3970
u u 1 p  nf   2 v  2 v 
u v      k (Wm-1K-1) 0.613 25
x y  nf y  nf  x 2 y 2 
β × 10-5 (1/K) 21 0.85
1
   nf g T  Tc  (3)
 nf
The appropriate initial and boundary
conditions are:
T T   2T  2T 
u v   nf  2  2  (4) t  0 : u  v  0,    c , 0  x  L, 0  y  L, (5)
x y  x y  
t  0 : u  v  0,  0, x  0 & L, (6)
where, x
x
u  v  0,   ( h   0 )( )   0 , y  0, (7)
L
x
u  v  0,   ( 0   c )( )   c , y  L. (8)
L

Introducing the stream function and vorticity

that defined as:
195
   v u g  l 3 Pr
u
y
, v
x
,  
x y
(9) Ra  Th  Tc  is the Rayleigh number,
 f0
The stream function (9) satisfies the continuity f
Eq. (1). To find the vorticity equation is by and Pr  0 is the Prandtl number.
 f0
eliminating the pressure between the two
momentum equations, i.e by taking y-
derivative of Eq. (2) and substracting the x- The boundary conditions for variable sidewalls
derivative of Eq. (3). This gives: temperature [10] are:

  0 for all solid boundaries (13)

    nf   2  2 
      2
1
y x x y  nf  x 2 y 2   and   1  X  on the left hot wall (14)
X 2
2
T
  nf g  
1
 (10) 2 1
 nf  x     2 and   X on the right cold wall (15)
X 2
 T  T   2T  2T  2 
   nf  2  2  (11)    2 and  0 on the adiabatic walls (16)
y x x y  x y  Y Y
 
2 2
   (12) Once we know the temperature we can obtain
x 2 y 2 the rate of heat transfer from the hot wall,
which are given in terms of the average
Nusselt number as:
knf 
Nu   
l4
Now we introduce the following non- dY . (17)
l1 k f X
dimensional variables
x y l 2  vl
X  , Y  ,  ,  , V , III. SOLUTION APPROACH
l l  f0  f0  f0 We employed finite difference method to solve
ul knf  nf  nf Eq. (10)-(12) subject to the boundary
U , k ,  ,  , conditions (13)-(16). The central difference
 f0 kf 0  f0  f0
method is applied for discretizing the
T  Tc equations. Next, the solution of the algebraic
 (13)
Th  Tc equations was performed using Gauss-Seidel
iteration. The unknowns Θ, Ψ and Ω are
where the subscript "0" stands for the reference calculated until the following convergence
temperature which is considered at the ambient criterium is fulfilled:
condition.   in, j 1   in, j
i, j
, (18)
By using the dimensionless parameters in Eq.  n 1
i, j
(13) together with Eq. (9)-(12) then, we can i, j

rewrite as: where ξ is either Θ, Ψ and Ω, n represents the

    Pr    2  2   iteration number and  is the convergence
  
Y X X Y  p  X 2 Y 2  criterium. In this study, the convergence
1      criterion is set at  106 . Different mesh sizes
f from 11×11 to 101×101 are used to carry out
Where,  p     this study. It is clear from the grid
Ra Pr   1       (10) independence test as shown in Fig. 2 that a
 f   X 
  41×41 uniform grid is enough to investigate
    k   
2 2
 the current problem.
    2  (11)
Y X X Y ( c ) X 2
Y 
1      p p 
( cp ) f
2  2 
   (12)
X 2 Y 2

196
 paper is hoped to be a useful guide for the
experimentalists to study the natural
convection heat transfer in square enclosure
filled with nanofluids.

(a)

Figure 2. Nu for different grid system

IV. RESULTS AND DISCUSSIONS

A numerical analysis has been conducted to
investigate the effect of Al2O3-water nanofluid (b)
filling a two-dimensional square enclosure. We
present numerical results for streamlines and
isotherms for a value of the Ra  103 and
0%    5% . We limited our investigation on
volume fraction of 0%    5% as explained
by [7,22,4,23,13,24]. From their investigations,
nanofluids were found to exhibit substantially
higher thermal properties particularly thermal
(c)
conductivity even when the concentrations of
suspended nanoparticles are very low i.e
  5%

The flow and energy transport in the enclosure

for the water-based Al2O3 nanofluid in the case
of variable sidewalls temperature with
Ra  103 is shown in Fig. 3 for various solid
volume fractions. The flow is single-cellular
and the shape of the main cell is circular. The
addition of nanoparticle affects the diameter of Figure 3. Contour plots of the stream function (left) and
the cell. In all cases, the flow rotates in isotherm (right) for (a) pure water (b)   0.04
clockwise direction. As shown by Fig. 3, the and (c)   0.05
maximum absolute value of the stream
function,  max decreases as volume fractions ACKNOWLEDGEMENTS
increases, therefore natural convection The authors would like to acknowledge the
decreases. Furthermore the stream function financial support received from the Grant
decreases which means natural convection is UKM-GUP-BTT-07-25-173 and Univeristy of
weaken. Malaya.

V. CONCLUSIONS REFERENCES
The problem of natural convection heat [7]. Abu-Nada, E, Masoud, Z, Oztop, HF, and
transfer in a square enclosure filled with Al2O3 Campo, A (2009). "Effect of nanofluid
nanofluids has been studied numerically. variable properties on natural convection
Various solid volume fractions at low Rayleigh in enclosures," Int J Thermal Sci in press.
numbers have been considered for a variable [8]. Aminossadati, SM, and Ghasemi, B
sidewalls temperature. The flow and (2009). "Natural convection cooling of a
temperature fields as well as the heat transfer localised heat source at the bottom of a
rate has been analyzed. We can conclude that nanofluid-filled enclosure," Eur J Mech.
heat transfer within the enclosure were found B/Fluids Vol 28 pp 630-640.
to depend upon, volume fraction of
nanoparticles. The theoretical prediction in this
197
[9]. Choi, SUS (1995). "Enhancing thermal nanoparticle suspensions (nanofluids)," J
conductivity of fluids with nanoparticles," Appl Phys Vol 99 pp 084314.
ASME Fluids Eng Div, Vol 231, pp 99-105 [20]. Muthtamilselvan, M, Kandaswamy, P,
[10]. Choi, SUS, Zhang, ZG, Yu, W, and Lee, J (2009). "Heat transfer
Lockwood, FE, and Grulke, EA (2001). enhancement of copper-water nanofluids
"Anomalous thermal conductivity in a lid-driven enclosure," Commun
enhancement in nanotube suspensions," Nonlinear Sci Numer Simulat in press.
Appl Phys Lett Vol 79 pp 2252-2254. [21]. Nguyen, CT, Desgranges, F, Roy, G,
[11]. Chon, CH, Kihm, KD, Lee, SP, and Galanis, N, Mare, T, Boucher, S, and
Choi, SUS (2005). "Empirical correlation Angue Minsta, H (2007). "Temperature
finding the role of temperature and particle and particle-size dependent viscosity data
size for nanofluid (Al2O3) thermal for water-based nanofluids hysteresis
conductivity enhancement," Applied phenomenon," Int J Heat Fluid Flow Vol
Physics Letters, Vol 87, pp 1-3. 28 pp 1492-1506.
[12]. Das, MK, and Ohal, PS (2009). [22]. Ogut, EB (2009). "Natural convection
"Natural convection heat transfer of water-based nanofluids in an inclined
augmentation in a partially heated and enclosure with a heat source," Int J
partially cooled square cavity utilizing Thermal Sci in press.
nanofluids," Int J Numer Meth Heat Fluid [23]. Ostrach, S (1988). "Natural convection
Flow Vol 19 pp 411-431. in enclosures," J Heat Transfer, Vol 110, pp
[13]. Eastman, JA, Choi, SUS, Li, S, and 1175-1190.
Thompson, LJ (1997). "Enhanced thermal [24]. Oztop, HF, and Abu-Nada, E (2008).
conductivity through the development of "Numerical study of natural convection in
nanofluids," Proceedings of the partially heated rectangular enclosures
Symposium on Nanophase and filled with nanofluids," Int J Heat Fluid
Nanocomposite Materials II Vol 457 pp 3- Flow Vol 29 pp 1326-1336.
11. [25]. Polidori, G, Fohanno, S, and Nguyen,
[14]. Ghasemi, B, and Aminossadati, SM CT (2007). "A note on heat transfer
(2009). "Natural Convection Heat Transfer modelling of newtonian nanofluids in
in an Inclined Enclosure Filled with a laminar free convection," Int J Thermal
Water-Cuo Nanofluid, " Numerical Heat Sci Vol 46 pp 739-744.
Transfer, Part A Vol 59 pp 807-823. [26]. Tiwari, RK, and Das, MK (2007).
[15]. Ghasemi, B, and Aminossadati, SM "Heat transfer augmentation in a two-sided
(2010). "Periodic natural convection in a lid-driven differentially heated square
nanofluid-filled enclosure with oscillating cavity utilizing nanofluids," Int J Heat
heat flux," Int J Thermal Sci Vol 49 pp 1- Mass Transfer Vol 50 pp 2002-2018.
9. [27]. Wang, X, Xu, X, and Choi, SUS
[16]. Kandaswamy, P, Eswaramurthi, M (1994) "Thermal conductivity of
(2008). "Density maximum effect on nanoparticle-fluid mixture, " J
bouyancy-driven convection of water in a Thermophysics Heat Transfer, Vol 13, pp
porous cavity with variable side wall 474480.
temperatures," Int J Heat Mass Transfer [28]. Wang, X, Xu, X, and Choi, SUS
Vol 51 pp 1955-1961. (1999). "Thermal conductivity of
[17]. Khanafer, K, Vafai, K, and Lightstone, nanoparticle--fluids mixture," J
M (2003). "Buoyancy-driven heat transfer Thermophys Heat Transfer Vol 13 pp 474-
enhancement in a two-dimensional 480.
enclosure utilizing nanofluids," Int J Heat [29]. Xuan, Y, Li, Q, and Yu, W (2003).
Mass Transfer Vol 46 pp 3639-3653. "Aggregation structure and thermal
[18]. Kumar, S, Prasad, SK, and Banerjee, J conductivity of nanofluids," AIChE J Vol
(2009). "Analysis of flow and thermal field 49 pp 1038-1043.
in nanofluid using a single phase thermal [30]. Zhu, H, Zhang, C, Liu, S, Tang, Y,
dispersion model," Appl Math Model in and Yin, Y (2006). "Effects of
press. nanoparticles clustering and alignment on
[19]. Li, CH, and Peterson, GP (2006). thermal conductivities of Fe3O4 aqueous
"Experimental investigation of temperature nanofluids," Appl Phys Lett Vol 89 pp
and volume fraction variations on the 023123-1-0231123-3.
effective thermal conductivity of
198
NOMENCLATURE Subscript
Cp spesific heat capacity c cold
g gravitational acceleration f fluid
k thermal conductivity h hot
Nu Nusselt number nf nanofluid
Pr Prandtl number 0 reference value
Ra Rayleigh number p particle
T fluid temperature
U, V dimensionless velocity components Superscript
in the X- and Y-direction — mean
X, Y dimensionless space coordinates

Greek symbols
α thermal diffusivity
β thermal expansion coefficient
ν kinematic viscosity
ø solid volume fraction
Ψ dimensionless stream function
Θ dimensionless temperature
Ω dimensionless temperature
ρ density of fluid
μ dynamic viscosity

199
Optimization model for estimating productivity growth in Malaysian food
manufacturing industry

Nordin Hj. Mohamad(1), and Fatimah Said(2)

(1)
Institute of Mathematical Sciences, University of Malaya, 56300 Kuala Lumpur, Malaysia
(Email: nordinhm@um.edu.my)
(2)
Faculty of Economics and Administration, University of Malaya, 56300 Kuala Lumpur, Malaysia
(Email: fatimahs@um.edu.my).

ABSTRACT technology and other non-physical factors do

In this study, a mathematical programming-based contribute to the dynamics of productivity
optimization technique known as data envelopment growth. This non-physical contributor is
analysis, DEA is used to compute and analyze the known as total factor productivity, TFP. In
decomposition of Malmquist index of total factor short, TFP addresses any effect in total output
productivity, TFP into technological change,
not caused by inputs or economies of scale and
technical efficiency change and scale efficiency
change by utilizing an output-oriented DEA model is often found to be a significant contributor to
under the assumptions of constant and variable output growth. Improvement in TFP will
returns to scale. The methodology is applied to enable the industry to generate a larger output
selected 5-digit Malaysian food manufacturing from the same resources, and hence shifting it
industries using annual time-series data for the to a higher frontier. The technological change
period 2002 – 2007. The results suggest that the component of productivity growth captures
TFP growth is largely due to positive technological shift in the frontier technology and can be
change rather than technical efficiency change. interpreted as providing a measure of
innovation. Technical efficiency improvement
KEYWORDS or catching up effect, on the other hand, is
Data envelopment analysis, Malmquist productivity measured by the difference between the
index, technological change, technical efficiency frontier output and the realized output. Thus,
change.
the decomposition of TFP into technological
change and technical efficiency change is
I. INTRODUCTION therefore useful in distinguishing innovation or
Decomposing productivity growth is important adoption of new technology by best practice
in identifying and understanding the sources of firms from the diffusion of technology.
growth to aid and provide directions to The rest of the paper is organized as follows.
decision makers in their policy making. The next section reviews selected literature in
Traditionally, productivity can be productivity performance analysis in the
mathematically defined as the relationship manufacturing sector, This is followed by
between a set of input values and the output definition of DEA output distance function for
values. This gives rise to the concept of partial two time periods, the formulation and
productivity which represents the change of decomposition of Malmquist TFP growth
output produced corresponding to each input index. The methodolody is applied to a set of
used such as labour productivity and capital 32 selected Malaysian food manufacturing
productivity. However, as technology sub-industries for the period 2002–2007.
progesses, it is observed that it is possible to Results and findings are presented, followed
produce more from less inputs by adopting by concluding remarks in the final section.
better means and methods of production. It is
therefore essential to conduct and analyze the
II. LITERATURE REVIEW
productivity trends as well as the technological
Various techniques have been used in the study
changes in order to understand the industrial
of productivity growth. This includes the
situation and the productivity dynamics.
Divisia index model, growth accounting
Thus, output or productivity growth is not
approach, aggregate or frontier production
attributed to growth in inputs only.
function estimates, stochastic varying
Improvement in input qualities, efficient use of
coefficient approach and the non-parametric
production processes, adoption of new
DEA.
200
 An empirical analysis which focuses on the which are defined by a distance function
convergence hypothesis and scale effects in relative to two different time periods.
explaining different productivity growth rates
within the manufacturing sectors of Canada Definition 1. Consider K decision making
and the United States is addressed by Mullen units (DMUs), each utilizing inputs
and Williams [7]. Fare et al. [2] analyzes X k(t )   N to produce outputs Yk(t )   M , k
productivity growth in 16 of Taiwan’s
manufacturing industries during the period = 1,2,…,K at time t. The output distance
1978-1992 by utilizing the DEA approach to function for DMU-k with respect to two
compute the Malmquist TFP index.
different time periods, k k k D (t ) ( X (t 1) , Y (t 1) )
Mahadevan and Kim [4] examines the sources
under the assumptions of constant returns to
of output growth of four selected South Korean
scale, CRS, is defined by an output oriented
manufacturing industries from 1980 to 1994
DEA linear programming problem such that
using firm level data within each industry.
On the Malaysian scenario, Zulaifah and
Maisom [8] utilized the Dollar and Sokoloff {Dk(t ) ( X k(t 1) , Yk(t 1) )}1 = Max  k (1)
model and an econometric approach to s.t
determine the intensity of use of factors of K
production in mamufacturing industries. They  j X (j t()n )  X k( t(n1)) ,
n  1,2,3,..., N ,
obtained an estimated growth rate of 4.32 j 1 (2)
percent per annum for all industries for the K

period 1985-1998 with light and medium  Y

j 1
j
(t )
j (m)   k Yk((tm1)) ,
m  1,2,3,..., M , (3)
industries exhibiting the highest TFP growth
rate of 17.3 percent per annum, followed by  j  0, j  1,2,..., K .
heavy industries at 11.7 percent per annum and (4)
resource based industries at 6.4 percent per
annum. However, no analysis was conducted A similar definition applies to
(t ) (t ) (t ) ( t 1) (t ) (t )
on the components or sources of TFP growth. D ( X ,Y ) , D
k k k k ( X ,Y )
k k and
A recent study by Ismail [3] found that the
technical efficiency change in food-based Dk(t 1) ( X k(t 1) , Yk(t 1) ) .
industry for the period 1985-2003 is low with a
negative overall mean of -0.976 percent per Definition 2. The Malmquist index of TFP for
annum. Only twenty percent of the twenty-five
sub-industries investigated experience positive DMU-k, k k k M ( X (t 1) , Y (t 1) , X (t ) , Y (t ) )
k k can be
technical efficiency change. However all the specified as the geometric mean of the
sub-industries experience positive technical or productivity changes between two time periods
technological change with a mean value of such that
1.036 or 3.6 percent change. The TFP index M k ( X k(t 1) , Yk(t 1) , X k(t ) , Yk(t ) )
exhibits positive growth with an overall mean
of 1.011 (ranging from 0.949 to 1.073).
In this study we utilize the concept of DEA 1/ 2
Malmquist productivity index to decompose
 D (t ) ( X (t 1) , Y (t 1) )  D (t 1) ( X (t 1) , Y (t 1) ) 
  k (t ) k (t ) k(t )  k (t 1) k (t ) k(t )  .
and analyze the TFP change into technical  Dk ( X k , Yk )  Dk ( X k , Yk ) 
efficiency change and technology or frontier
shift. The technical efficiency change will
(5)
further be decomposed into pure technical
efficiency change and scale efficiency change.
A value of Mk > 1 indicates positive TFP
The model is applied to thirty-two selected
growth or gain, Mk < 1 indicates TFP decline
Malaysian food manufacturing sub-industries
or loss, and Mk = 1 implies stagnation or no
for the period 2002-2007.
change in TFP for DMU-k from time t to t+1.
Equation (5) can also be equivalently written
III. DEA MALMQUIST INDEX as
Fare et al. [1] construct the DEA-based
Malmquist productivity index as the geometric
mean of two Malmquist productivity indexes

201
  D (t 1) ( X ( t 1) , Y (t 1) )  The complete decomposition for DMU-k thus
M k (...)   k (t ) k ( t ) (kt ) . becomes
 Dk ( X k , Yk ) 
Mk(.) = (TFP growth)k
= (TEC)k . (FS)k
1/ 2
 Dk(t ) ( X k(t 1) , Yk(t 1) )  Dk(t ) ( X k(t ) , Yk( t ) )  = (PTEC)k.(SEC)k.(FS)k , k=1,..,K.
 ( t 1) ( t 1) (t 1)  ( t 1) ( t ) ( t )  . (11)
 Dk ( X k , Yk )  Dk ( X k , Yk ) 
For evaluation under the assumption of VRS,
(6) an additional convexity constraint
K

The first component, 

j 1
j  1,
is imposed when solving the linear
 D (X )
( t 1)
,Y ( t 1) ( t 1) programming problem (1)-(4).
TECk   k
,
(t )
k
(t )
k
(t )
 D ( X ,Y )  k k k (7) IV. EMPIRICAL IMPLEMENTATION
Data source
measures the change in technical efficiency The data used in the study are annual time-
over the two periods, whether or not DMU-k is series data for 32 selected 5-digit
getting closer to its efficiency frontier over Malaysian food manufacturing sub-
time. This term is normally referred to as the industries for the period 2002-2007,
catching up effect since it measures the degree
compiled from the Annual Survey of
of catching up to the best-practice frontier over
time. Malaysian Manufacturing Industries,
published by the Department of Statistics,
The second component, Malaysia [5]. Table 1 lists these sub-
industries and their 5-digit Malaysian
1/ 2 Standard Industrial Code (MISC). A single
 D(t ) ( X (t 1) , Y (t 1) )  D(t ) ( X (t ) , Y (t ) ) 
FSk   (kt 1) k (t 1) k (t 1)  (kt 1) k (t ) k (t )  measure of output, value added deflated by
 Dk ( X k , Yk )  Dk ( X k , Yk )  the consumer price index for food, is used.
(8) Cost of inputs, total number of workers
measures the technology change over the two and total fixed assests constitute three
periods, whether or not the frontier is shifting measures of input. The cost of inputs and
out over time. It can be viewed as a geometric total fixed assets were deflated by producer
mean of the change or shift in the frontier of price index for goods in the domestic
technology or innovation experienced by economy for manufactured goods. Both
DMU-k from time t to t +1. Hence the deflators with 2000 as the base year were
Malmquist TFP index is simply the product of obtained from the Economic Report
technical efficiency change and technological published by the Ministry of Finance,
change. That is,
Malaysia [6].
TFP growth = (TEC) . (FS) (9)
Table 1. The Malaysian food manufacturing sub-
industries
Fare et al. [1] further propose an enhanced 5-Digit Sub-industry
decomposition of TEC (relative to CRS
MSIC
frontier) into a pure technical efficiency 15111 Production of poultry and poultry products
change, PTEC component (relative to a 15119 Production of meat and meat products
variable returns-to-scale, VRS frontier), and a 15120 Production of fish and fiah products
residual scale efficiency change, SEC 15131 Pineapple canning
15139 Canning of other fruits and vegetables
component which captures changes in the
15141 Manufacture of coconut oil
deviation between the VRS and CRS 15142 Manufacture of crude palm oil
technology. That is, 15143 Manufacture of refined palm oil
15144 Manufacture of palm kernel oil
TEC = (PTEC) . (SEC) (10) 15149 Manufacture of other oils and fats
15201 Manufacture of ice cream
15202 Manufacture of milk
15311 Rice milling
202
15312 Flour milling DMU is operating under irs or drs can be
15319 Manufacture of other mill products determined by observing its TE and PTE
15322 Manufacture of glucose and maltose
15323 Manufacture of sago and tapioca products scores, such that
15330 Manufacture of animal feeds
15411 Manufacture of biscuits and cookies
15412 Manufacture of bakery products Table 2. Mean efficiency scores, 2002-2007
15420 Manufacture of sugar 5-Digit Technical Pure technical Scale Returns
15431 Manufacture of cocoa products MSIC efficiency efficiency efficiency to scale
15432 Manufacture of chocolate and sugar products
15440 Manufacture of macaroni and similar products 15111 0.5259 0.5823 0.9093 drs
15491 Manufacture of ice (excluding dry ice) 15119 0.5299 0.5868 0.9242 drs
15492 Manufacture of coffee 15120 0.6235 0.7749 0.8071 drs
15493 Manufacture of tea 15131 0.5327 0.6088 0.8853 irs
15494 Manufacture of spices and curry powder 15139 0.7509 0.7706 0.9723 irs
15495 Manufacture of nut and nut products 15141 0.5172 0.9947 0.5196 irs
15496 Manufacture of sauces 15142 0.5520 1.0000 0.5520 drs
15497 Manufacture of snack (cracker/chips) 15143 0.8482 0.9750 0.8671 drs
15499 Manufacture of other food products 15144 0.7438 0.7615 0.9764 irs
15149 0.5668 0.6029 0.9296 drs
15201 0.7973 0.8200 0.9734 irs
Next, we solve the DEA output-oriented model 15202 0.8931 1.0000 0.8931 drs
under the assumptions of CRS and VRS for 15311 0.3780 0.3978 0.9595 drs
each year. Results for the mean efficiency 15312 0.8299 0.8588 0.9607 drs
15319 0.7638 0.8517 0.9035 irs
scores and returns to scale are summarized in 15322 0.6287 1.0000 0.6287 irs
Table 2. 15323 0.6081 0.7035 0.8841 irs
15330 0.6026 0.7194 0.8360 drs
Technical and scale efficiency 15411 0.6213 0.7376 0.8454 drs
Out of the 32 sub-industries, only one sub- 15412 0.7194 0.9836 0.7326 drs
15420 0.9957 0.9986 0.9970 mpss
industry (15491 Manufacture of ice) obtains a 15431 0.6182 0.6525 0.9482 irs
scale efficiency score of 100 percent, implying 15432 0.8817 0.9260 0.9503 drs
that it is technically efficient in all years under 15440 0.6864 0.7458 0.9176 drs
evaluation and is operating on the frontier at 15491 1.0000 1.0000 1.0000 mpss
15492 0.8174 0.8364 0.9797 drs
the most productive scale size, mpss. Two 15493 0.5854 0.7295 0.7855 irs
other sub-industries (15420 Manufacture of 15494 0.6568 0.6701 0.9800 drs
sugar and 15496 Manufacture of sauces) are 15495 0.8205 0.8930 0.9142 irs
close to mpss. In fact, both sub-industries 15496 0.9415 0.9545 0.9837 mpss
achieved scale efficiency score of 100 percent 15497 0.7156 0.7609 0.9379 drs
15499 0.9463 1.0000 0.9463 drs
in five of the six years under consideration. Average 0.7093 0.8093 0.8844
The result also implies that, in general, more Std.dev 0.1577 0.1602 0.1223
than 90 percent of the sub-industries were Maximum 1.0000 1.0000 1.0000
Minimum
operating inefficiently and they need to 0.3780 0.3978 0.5196
increase their output (or reduce their inputs) to
become efficient. The average PTE score was
80.93 percent during 2002-2007. This finding
Note: drs and irs refer to decreasing and
suggests that if these sub-industries were
increasing returns to scale respectively.
operating efficiently, they could have produced
19.07 percent more output. Nevertheless, more
than sixty percent of the sub-indutries were
• if TE = PTE , CRS prevails
more than 90.0% scale efficient.
• if TE ≠ PTE , then
Returns to scale K
Apart from the inefficiency that could arise in
the conversion process, another reason for the

j 1
j 1
→ irs,
inefficiency of the inefficient units could be K
attributed to the scale of operations. DMUs
that do not operate at the most efficient (or

j 1
j 1
→ drs.
productive) scale size cannot be fully efficient.
The inefficiency may arise because it is The last column in Table 2 records the returns
operating under decreasing returns to scale, drs to scale based on the most frequent occurance
or increasing returns to scale, irs. Whether a observed during the years under consideration.
203
As mentioned earlier, three (9.375%) sub- 15420 1.046 1.041 1.005 1.002 1.003
industries appeared to be operating at their 15431 1.069 1.079 0.991 0.985 1.006
15432 0.976 1.054 0.926 0.946 0.979
mpss. Eighteen (56.250%) of sub-industries 15440 0.940 1.001 0.939 0.975 0.963
exhibited drs. These sub-industries should 15491 1.041 1.041 1.000 1.000 1.000
scale down their scale of operation if they were 15492 1.040 1.007 1.033 1.009 1.024
to operate on the frontier. The remainder 15493 1.093 1.037 1.054 1.003 1.051
15494 0.992 1.008 0.984 0.997 0.987
eleven (34.375%) exhibited irs. These sub- 15495 0.932 1.022 0.912 0.935 0.975
industries should expand their scale of 15496 0.946 1.013 0.934 0.952 0.981
operation in order to become scale efficient. 15497 0.923 1.016 0.909 0.934 0.973
The average scale efficiency score in the 15499 0.976 1.009 0.967 1.000 0.967
sample for the period 2002-2007 was 88.48 Average 1.003 1.034 0.970 0.975 0.996
Std.dev 0.051 0.027 0.043 0.048 0.036
percent, ranging from a minimum of 51.96 Maximum 1.093 1.103 1.054 1.033 1.076
percent to a maximum of 100 percent. Minimum 0.892 1.001 0.847 0.788 0.907
Note: All Malmquist index averages are
Malmquist productivity change geometric means.
Table 3 presents a summary of the annual
geometric means of the Malmquist Fifteen (46.87%) of the sub-industries showed
productivity index and its components. As can positive TFP growth while another seventeen
be observed, on average, the TFP for food (53.3%) recorded negative growth. The highest
manufacturing sub-industries showed a small TFP growth comes from sub-industry 15493
increase of 0.30 percent per annum, ranging Manufacture of tea (9.3 percent per annum)
from -10.8 percent to 9.3 percent. The while the lowest is from sub-industry 15131
technological or frontier shift recorded a Pineapple canning (-10.8 percent per annum).
positive change for all sub-industries, implying The remainder twenty-six (81.25%) indicated a
that growth was largely attributable to decline in TFP performance.
innovation. The technological change (frontier
shift or innovation) improved by 3.4 percent Technological change (frontier shift)
(ranging from 0.1% to 10.3%) per annum All sub-industries, on average, experienced
while technical efficiency regressed by 3.0 technological progress since the FSk index
percent (ranging from -15.3% to 5.4%) per attains a value greater than one for all
annum. Hence catching up is a problem facing k=1,...,32. The average score was 1.034,
most sub-indutries. Only six (18.75%) of the indicating a 3.4 percent technological progress
sub-industries showed improvement in all per annum. The highest technological progress
components. This includes two which of 10.3 percent per annum was achieved by
exhibited mpss, three drs and one irs. sub-industry 15323 Manufacture of sago and
tapioca products while the lowest innovative
Table 3. Mean Malmquist productivity index change,
2002-2007 improvement of 0.1 percent per annum was
recorded by sub-industry 15440 Manufacture
5-Digit Mk(.) FSk TECk PTECk SECk of macaroni and similar products.
MSIC (TFP growth)
15111 1.015 1.009 1.006 0.986 1.021 Technical efficiency change (catching up
15119 0.982 1.044 0.941 0.887 1.061 effect)
15120 0.980 1.024 0.957 0.944 1.014
15131 0.892 1.053 0.847 0.788 1.075 Only seven (21.9%) showed improvement in
15139 1.029 1.043 0.987 0.973 1.014 technical efficiency with sub-industry 15493
15141 1.049 1.071 0.979 1.000 0.979 Manufacture of tea attaining the highest score
15142 1.027 1.016 1.011 1.000 1.011 of 1.054 (catching up rate of 5.4% per annum).
15143 1.019 1.033 0.988 1.033 0.956
15144 0.984 1.041 0.946 0.959 0.986
Twenty-five (78.1%) sub-industries appeared
15149 0.939 1.015 0.925 0.966 0.958 to be lagging behind with sub-industry 15131
15201 1.091 1.097 0.995 1.004 0.991 Pineapple canning recording the lowest score
15202 0.991 1.038 0.955 1.000 0.955 of 0.847 (a decline of -15.3% per annum). On
15311 1.013 1.036 0.977 0.963 1.015 average, the group was found to be staggering
15312 0.974 1.005 0.969 0.972 0.997
15319 0.999 1.010 0.989 0.997 0.992 behind at -3.0% per annum. This indicates that
15322 0.975 1.075 0.907 1.000 0.907 technical efficiency is not improving in line
15323 1.086 1.103 0.985 0.915 1.076 with technological progress. In other words,
15330 1.075 1.038 1.036 1.031 1.005 the gap to the efficient frontier is widening.
15411 1.009 1.012 0.996 1.024 0.973
15412 0.989 1.011 0.979 1.011 0.968

204
Pure technical efficiency change 0.3 per cent per annum despite an encouraging
As mentioned earlier, TEC is the product of frontier shift or innovative improvement of 3.4
PTEC and SEC. Eight (25%) of the sub- percent per annum. This is due to a decline in
industries indicated increase in pure technical the catching up effect or TEC of -3.0 percent
efficiency with sub-industry 15143 per annum which is further attributable to
Manufacture of palm oil taking the lead with decrease in both PTEC and TEC. Only three
improvement of 3.3 percent per annum. sub-industries were found to be operating
Eighteen (56.25%) indicated a decrease with efficiently (exhibiting mpss) while twenty-nine
sub-industry 15131 Pineapple canning exhibit variable returns to scale, indicating the
retaining the lowest score of negative growth needs for operation adjustments. The findings
at -21.2 percent per annum. The remainder six suggest that eighteen and eleven of these sub-
(18.75%) sub-industries showed no change industries should scale down and expand their
during the period under consideration as scale of operations respectively if they were to
indicated by their PTEC score of unity. be operating on the efficient frontier.
Scale efficiency change The study is not without limitations. DEA is
SEC of fourteen (43.75%) of the sub-industries non-stochastic and does not capture random
contibuted positively to the productivity noise; thereby may have over estimated the
change since their scores exceed one. Sub- magnitude of inefficiencies. The data utilized
industry 15323 Manufacture of sago and in the study are aggregated sub-industries data
tapioca products recorded the highest score of and not firm level data. This is because firm
1.076 (a change of 7.6 percent per annum), level data is not easily accessible. The study
while sub-industry 15322 Manufacture of also assumes that all sub-industries under
glucose and maltose recorded the lowest score evaluation are fairly homogenous, utilizing
of 0.907 (a change of -9.3 percent per annum). similar set of inputs to produce identical
The average score for the group is 0.996 (a outputs. This can only be achieved if we are
small decrease of 0.4 percent per annum). evaluating a group of firms operating similar
business activities such as banking or financial
Observations institutions, hospitals and others.
From the above discussions, we can highlight a Lastly, the study provides avenues for further
few observations. exploration. The methodology can be revised,
• The Malmquist TFP index for Malaysian expanded and applied to other public and
food manufacturing industry indicated only a private organizations.
small increase of 0.3 percent per annum.
• The TFP growth is largely due to innovation REFERENCES
(a positive shift in the frontier) rather than [1]. Fare, R,; Grosskopf, S,; Norris, M, and
technical efficiency change (catching up Zhang, Z (1994). ‘‘Productivity growth,
effect). technical progress and efficiency change in
• A decrease in TEC is attributable to both industrialized countries,’’ American Economic
decrease in PTEC and SEC. Review, Vol 84, No 1, pp 66-83.
• Sub-industry 15493 Manufacture of tea [2]. Fare, R,; Grosskopf, S, and Lee, W
achieved the highest TFP growth with all (2001). ‘‘Productivity and technical change:
components indicating positive changes. the case of Taiwan,’’ Applied Economics, Vol
• Sub-industry 15131 Pineapple canning 33, pp1911-1925.
recorded the lowest TFP growth with the [3]. Ismail, R (2009). ‘‘Technical efficiency,
lowest TEC score despite encouranging technical change and demand for skills in
improvement in FS score (above group- Malaysian food-based industry,’’ European
average). The low TEC was due to the lowest Journal of Social Sciences, Vol 9, No 3,
PTEC despite a relatively high SEC. pp504-515.
[4]. Mahadevan, R, and Kim, S (2003). ‘‘Is
V. CONCLUSIONS output growth of Korean manufacturing firms
In this study, we have estimated the Malmquist productivity driven,’’ Journal of Asian
TFP index and its decompositions using the Economics, Vol 14, pp669-678.
output-oriented DEA distance functions for 32 [5]. Malaysia (various years). Annual survey
Malaysian food manufacturing sub-industries of manufacturing industries, Department of
for the period 2002-2007. The findings Statistics, Malaysia.
indicate that TFP only grew at a slow rate of

205
[6]. Malaysia (various years). Economic
Report, Ministry of Finance, Malaysia.
[7]. Mullen, JK, and Williams, M (1994).
‘‘Convergence scale and relative production
performance of Canadian-US manufacturing
industries,” Applied Economics, Vol 26,
pp739-750.
[8]. Zulaifah, O, and Maisom, A (2001).
‘‘Pattern of total factor productivitry (TFP)
growth in Malaysian manufacturing industries,
1985-1995,’’ in Yew, TS and Alias, R (eds.),
Selected Readings on Economic Analysis of
Industries and Natural Resources, Universiti
Putra Malaysia Press, 2001, pp72-105.

206
 Numerical study of natural convection in a porous cavity with transverse
magnetic field and non-uniform internal heating

Habibis Saleh(1), Ishak Hashim(2 ) and Rozaini Roslan(3 )

(1,2)
Modelling & Data Analysis Research Centre, School of Mathematical Sciences
Universiti Kebangsaan Malaysia, 43600 Bangi, Selangor, Malaysia
(Email: twokbi@yahoo.com (1))
(3)
Centre for Research in Applied Mathematics, Faculty of Science, Arts & Heritage
Universiti Tun Hussein Onn Malaysia, 86400 Parit Raja, Johor, Malaysia

ABSTRACT local heat generation within the porous

The effect of a transverse magnetic field on the material. Such a situation can arise through
steady convective flow within a square region filled radioactive decay or through, in the present
with a fluid-saturated porous medium having non- context, a relatively weak exothermic reaction
uniform internal heat generation at a rate within the porous material. This can happen,
proportional to a power of the temperature
for example in the self-induced heating of coal
difference is studied numerically. The side vertical
walls are maintained isothermally at different stockpiles and bagasse-piles. Convective
temperatures. The top and bottom horizontal flows are governed by nonlinear partial
straight walls are kept adiabatic. The porous differential equations, which represent
medium modeled according to Darcy’s law. The conservation laws for the mass, momentum
formulation of differential equations is and energy. Numerical treatment usually used
nondimensionlized and solved by finite difference to solve the equations or if possible by an
method. It is found that with application of an analytic method. Numerical investigation of
external magnetic field, the flow and temperature natural convection in porous enclosure having
field and the heat transfer rate are significantly uniform internal heating firstly conducted by
modified. Ref. [1]. Then solved analytically by Ref. [2].
Recently, Ref. [3] moved away from the study
KEYWORDS
of uniform internal heating to that of non-
Natural convection; Darcy's law; Magnetic field;
Non-uniform heat generation; Finite difference uniform internal heating and solved the
method. problem numerically and analytically.

I.INTRODUCTION Previously abundant studies of natural

Convective flows in porous media have convection in the presence of magnetic field in
occupied the central stage in many an enclosure filled with a viscous and
fundamental heat transfer analysis and has incompressible fluid has been studied for
received considerable attention over the last example by Ref. [4] and Ref. [5]. However,
few decades. This interest is due to its wide there are very few studies on natural
range of applications, for example, high convection of a conducting fluid saturating a
performance insulation for buildings, chemical porous medium in the presence of magnetic
catalytic reactors, packed sphere beds, grain field in an enclosure. To the best of our
storage and such geophysical problems as frost knowledge, the first investigation of this
heave. Porous media are also of interest in problem is due to Ref. [6] who studied the
relation to the underground spread of effect of an electromagnetic field in the
pollutants, solar power collectors, and to horizontal direction in a rectangular porous
geothermal energy systems. Convective flows cavity when the side walls are asymmetrically
can develop within these materials if they are heated. Then Ref. [7] extended this to include
subjected to some form of temperature an inclined magnetic filed when the side walls
gradient. are symmetrically heated. In reality,
convective flows in porous media as
In some situations of considerable practical consequence of asymmetrically heated having
importance, porous material provides its own internal heat source is a more appropriate
source of heat, giving an alternative way in model for practical problem. Therefore, the
which a convective flow can be set up through present paper investigates the effect of an

207
 
inclined magnetic field on steady natural · I  0,
convection in a square cavity filled with a    (5)
porous medium saturated with an electrically I   (  V  B),
conducting fluid having internal heat   0 [1   (T  Tc )] (6)
generation. The rate of internal heat
generation is assumed proportional to a power
of the temperature difference.

II. MATHEMATICAL FORMULATION

We consider the steady, two-dimensional
natural convection flow in a square region
filled with an electrically conducting fluid-
saturated porous medium, see Figure 1.
The co-ordinate system employed is also
depicted in this figure. The top and bottom
surfaces of the convective region are
assumed to be thermally insulated and the
sloping surfaces to be heated and cooled at
constant temperatures Th and Tc ,
respectively. Heat is assumed to be
generated internally within the porous
medium at a rate proportional to
(T  Tc ) p , ( p  1) . This relation, as Figure 1. Schematic representation of the model
explained by Ref. [3], is an approximation 
of the state of some exothermic process. where V  (u, v) is the fluid velocity
 vector, T is the fluid temperature, P is
A uniform and constant magnetic field B 
the pressure, B is the external magnetic
is applied normal to gravity direction. The 
viscous, radiation and Joule heating effects field, I is the electric current,  is the

are neglected. The resulting convective electric potential, g gravitational
flow is governed by the combined acceleration vector, K is the permeability
mechanism of the driven buoyancy force of the porous medium,  m is the effective
and the retarding effect of the magnetic thermal diffusivity,  is the density,  is
field. The magnetic Reynolds number is
the dynamic viscosity,  is the coefficient
assumed to be small so that the induced
magnetic field can be neglected in favor of of thermal expansion, c p is the specific
the applied magnetic field. heat at constant pressure,  is the
electrical conductivity, 0 is the reference
Under the above assumptions, the density and  is the associated electric
conservation equations for mass,
momentum under the Darcy field. Eqs. 4 and 5 reduce to  2  0 , the
approximation, energy and electric transfer unique solution is   0 since there is
are given by: always an electrically insulating boundary

· V  0, (1) around the enclosure. Thus, it follows that
 K    the electric field vanishes everywhere [4].
V  (P   g  I  B), (2) Furthermore, eliminating the pressure term
 in Eq. 2 in the usual way then the
 q ''' (3) governing Eqs. 1~6 can be written as
(V · )T   m  2T  0 (T  Tc ) p ,
cp (4)

208
u v Ha  q0''' 2 (T ) p 1 / ( m  c p ) is the
  0, (7)
x y internal heat generation parameter and
u v gK  T  KB02  u  Ha   KB02 /  is the Hartmann number
   , (8)
y x  x   y  for the porous medium. Once we know the
T T   2T  2T  temperature we can obtain the rate of heat
u v  m  2  2  transfer from each of the vertical walls,
x y  x y  which are given in terms of the mean
q ''' (9)
Nusselt number as
 0 (T  Tc ) p .
cp 1
  
 Nuh      dY (15)
0
X  X 0
where B0 is the magnitude of B and  is
the kinematic viscosity of the fluid. The
at the hot wall and
following are the boundary conditions:
1
on x   : u  0, T  Tc ,   
Nuc      dY (16)
0
X  X 1
on x  0 : u  0, T  Th ,
T
on y  , y  0 : v  0,  0. (10) at the cold wall.
y
III. FINITE DIFFERENCE METHOD
Now we introduce the following non- We employed the finite difference method
dimensional variables to solve Eqs. 12 and 13 subject to 14. The
central difference method was applied for
x y   T  Tc discretizing the equations. The solution of
X  , Y  ,U  u,V  v,   (11)
  m m T the algebraic equations was performed
using the Gauss-Seidel iteration with
Introducing the stream function  defined relaxation method. The unknowns  and
as U   / Y and U   / X , and  are calculated until the following
using expressions 11 in Eqs. 7~9, we convergence criterion is fulfilled:
obtain the following partial differential
equations in non-dimensional form:  n 1 n 
  i, j   i, j 
 2  2 
max    , (17)
(1  Ha 2 )    Ra , (12)   n 1 
X 2
Y 2
X  i, j 
 2  2    
 2  G p   . (13)
X 2
Y Y X X Y where  is either  or  , n represents
the iteration number and  is the
Subject to the boundary conditions convergence criterion. In this study, the
convergence criterion is set at   106 .
on X  L :   0,   0, The mean Nusselt number for each vertical
on X  0 :   0,   1, walls was calculated using the integration
 trapezoidal rule.
on Y  L, Y  0 :   0,  0. (14)
Y
IV. RESULT AND DISCUSSION
Figure 2 shows the streamlines and isotherms
where Ra  gK  LT / ( m ) is the for the strong heating case (G=5) with
Rayleigh number, Ra=300, p=1 and a range of values of Ha. In

209
the absence of a magnetic field, the fluid
motion as shown in the figure 2(a) is described
as follow. Since the temperature of the left
wall is higher than that of the fluid inside the
enclosure, the wall transmits heat to the fluid
and raises the temperature of fluid particles
adjoining the left wall. As the temperature
rises, the fluid moves from the left wall (hot)
to the right wall (cold), falling along the cold
wall, then rising again at the hot wall. This
movement creates a clockwise vortex cell
inside the enclosure and the isotherms start
either from the hot wall or from the bottom
wall and end at the top wall or cold wall.
When the magnetic field is relatively
strengthened (figure 2(b)), the central
streamlines cell are becoming upright and the
maximum temperature drifts towards the
center of the top wall. The internal heat
generation enhances the flow near the cold
wall and forces the streamlines in a denser
distribution. On the other hand, the negative
buoyancy caused by G produces a vertical
downward flow motion in the vicinity of the
top corner hot wall. The maximum
temperature increase above that on the heated
wall. This is due to the fact that the hot fluid
reaching the top left corner of the enclosure is
unable to reject energy since the velocities are
small. Finally, for large Ha (figure 2(c)), the
core vortex is elongated vertically and the
isotherms are almost parallel to the vertical
wall, this implies that conduction is dominant.
Figure 2. Contour plots of the stream function and
temperature for Ra=300, G=5, p=1 and (a) Ha=0, (b)
The variations of the mean Nusselt number Ha=5, (c) Ha=20.
along the hot wall and the cold wall with Ha
for several values of G are shown in Figure 3.
Naturally, the heat transfer is maximum in the
absence of a magnetic field, because the
convection is maximum for this situation. In
general, the mean Nusselt number along the
hot wall and the cold wall initially decreases
steeply with Ha. As the value of Ha is made
larger, the strength of the convective motion is
progressively suppressed and the mean Nusselt
number along the hot wall and the cold wall
towards unity for G=0. For a fixed moderate
value of Ha, increasing G has the effect of
decreasing the mean Nusselt number along the
hot wall and increasing the mean Nusselt
number along the cold wall.

(a)

210
 REFERENCES
[31]. Haajizadeh, M, Ozguc, AF, and
Tien, CL (1984). “Natural convection
in a vertical porous enclosure with
internal heat generation,” Int J Heat
Mass Transfer, Vol 27, pp 1893-1902.
[32]. Joshi, MV, Gaitonde, UN, Mitra, SK
(2006). “Analytical study of natural
convection in a cavity with volumetric heat
generation,”J Heat Transfer, Vol 128,
pp176-182.
[33]. Mealey, LR, and Merkin, JH (2009).
“Steady finite rayleigh number convective
flows in a porous medium with internal
(b) heat generation,” Int J Thermal Sci, Vol
48, pp 1068-1080.
Figure 3. Plots of the mean Nusselt number along the [34]. Alchaar, S, Vasseur, P, and Bilgen, E
hot wall (a) and along the cold wall (b) against Ha for the (1995). “Natural convection heat transfer
values of G labeled on the figure with Ra=100, p=1.
in a rectangular enclosure with a
transverse magnetic field,” J Heat
V. CONCLUSIONS Transfer, Vol 117, pp 668-673.
The present numerical study exhibits many [35]. Al-Najem, NM, Khanafer, KM, El-
interesting features concerning the effect of the Refaee, MM, (1998). “Numerical study of
inclined magnetic fields on natural convection laminar natural convection in tilted
in square enclosure filled with a porous enclosure with transverse magnetic field,”
medium having non-uniform internal heat Int J Numer Meth Heat Fluid Flow, Vol 8,
source. The results of the numerical analysis pp 651-672.
lead to the following conclusions. In general, [36]. Bian, W, Vasseur, P, Bilgen, E, and
effect of magnetic field is to retard the flow Meng, F, (1996). “Effect of an
circulation and normalize the temperature electromagnetic field on natural
distribution. Combination of the large of convection in an inclined porous layer,”
applied magnetic field having a strong internal Int J Heat Fluid Flow, Vol 17, pp 36-44.
heat source was found to be most effective in
[37]. Grosan, T, Revnic, C, Pop, I,
suppressing the rate of heat transfer along the
hot wall. However, combination of the large of Ingham, DB, (2009). “Magnetic field
applied magnetic field having a low internal and internal heat generation effects on
heat source was found to be most effective in on the free convection in a rectangular
suppressing the rate of heat transfer along the cavity filled with a porous medium,”
cold wall. Int J Heat Mass Transfer, Vol 52, pp
1525-1533.

211
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

THE DISTRIBUTION PATTERN AND ABUNDANCE

OF ASTEROID AND ECHINOID
AT RINGGUNG WATERS SOUTH LAMPUNG
Arwinsyah Arka*, Agus Purwoko*, Oktavia

* Department of Biology, Faculty of Mathematics and Natural Sciences, Sriwijaya University. 0818561648

ABSTRACT Station II (sandy beach): S 05º 33’23.18”,

Te research of the distribution pattern and E 105º 15’ 8.24”
abundance of Asteroid and Echinoid was carried out Station III (muddy beach): S 05º 33’17.24”,
in 28 June-11 July 2008 at Ringgung waters South E 105º 15’11.19”
Lampung. The aim of the research is to collect Station IV (muddy beach): S 05º 33’10.30”,
information about distribution pattern and
E 105º 15’14.15”
abundance of Asteroid and Echinoid at Ringgung
waters. The research was made by applying square Station V (muddy beach): S 05º 33’ 4.36”,
transect method at five stations. The result was E 105º 15’17.10”
analyzed descriptively. The research showed that
there were 2 species of Asteroid which consist of Asteroidea and Echinoidea were observed and
Archaster typicus and Culcita novaeguineae and 2 counted. Species name is macthed by Clark &
species of echinoid which consist of Diadema Rowe book . Diversity indeces is applied for
setosum and Aganum sp. with distribution patterns data analysing.
is clumped and the abundance of Ateroid and
echinoid is about 0,2-5.7 ind/m2. The factors that RESULTS AND DISCUSSION
influenced distribution pattern were differences
At least, we found 4 species ta Ringgung area.
habitat and living habits of each species.
They are classified as well as:
Key words
distribution pattern, abundance, asteroidea and 1. Fylum : Echinodermata
echinoidea, ringgung waters Class : Asteroidea
Ordo : Valvatida
INTRODUCTION Family : Archasteridae
Echinodermata consist of 5 classes such as: Genus : Archaster
Asteroidea , Echinoidea , Holothuroidea , Spesies : Archaster typicus,
Ophiuroidea and Crinoidea . Muller & Troschel
Sea urchin gonad can be consummed. Cina,
Hongkong, Korea, Jepang and American
people farm sea urchin, hoever, sea stars are
decoretted value.
Sea urchin abundand at Ringgung area
(Lampung Selatan) but no scientific
information is available. This research will find
out sea urchin population.
METHOD
Research carry out on 28 Juni - 11 Juli 2008 at
Ringgung. Kabupaten Lampung Selatan
Lampung Province.
Ringgung is recreasion site and fish farm
Sampling method was Purposive Sampling.
There are 5 stations which laid a 100 meter
transect which is consisted 10 sampling plots
(1 x 1 m). The cordinate of 5 stations are:

Station I (sandy beach) : S 05º 33’29.12”,

E 105º 15’13.72”

212
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Figure 1. Archaster typicus,

Muller & Troschel

2. Fylum : Echinodermata
Class : Asteroidea
Ordo : Valvatida
Family : Oreasteridae
Genus : Culcita
Spesies : Culcita novaeguineae

Figure 2. Culcita novaeguineae, Muller & Troschel

3. Fylum : Echinodermata
Class : Echinoidea
Ordo : Diadematoida
Family : Diadematidae
Genus : Diadema
Spesies : Diadema setosum

Figure 3. Diadema setosum, Leske

213
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

4. Fylum : Echinodermata
Class : Echinoidea
Ordo : Clypeasteridae
Family : Laganidae
Genus : Laganum
Spesies : Laganum sp.

Figure 4. Laganum sp.

Table 1. Morisita index at each station

Distribution pattern is cluster or clumped.

Density (ind/m²)
No. Jenis
Station I Station II Station III Station IV Station V

1 Archaster typicus 3.6 3.5 4.4 4.9 4.3

2 Culcita novaeguineae - - 0.2 0.3 -
3 Diadema setosum 3.3 3.7 4.6 5.7 3.5
4 Laganum sp. - - 1.0 1.4 0.6

Table 2. Density of Sea urchin at Ringgung

The highest density found at station IV and the lowest density is station I.
Stasiun Shenon index (H') Sameness index (e) Dominan Index (C)

I 0.99 Low 0.99 0.5

II 0.99 0.99 0.5
III 1.48 Average 0.74 0.39
IV 1.53 0.76 0.39
Low
V 1.29 0.82 0.44

Table 3. Diversity indeces

CONCLUSIONS
1. Asteroidea and Echinoidea distribution is REFFERENCES
cluster. Clark, A. M. and F.W.E. Rowe. 1971.
2. Archaser typicus density is 3,5-4,9 ind/m² Monograph of Shallow-Water Indo-
and Culcita novaeguineae is 0,2-0,3 ind/m² West pacific Echinoderms. Trustees
3. Sea urchin Diadema setosum density is 3,3- of The British Museum (Nat. His.).
5,7 ind/m² and Laganum sp is 0,6-1 London. 238 page
ind/m².
4. Archaster typicus dan Diadema setosum
found at all staions. They well adapted at
sandy and muddy habitats.

214
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Low biomass of macrobenthic fauna at a tropical mudflat: an effect of

latitude?
Agus Purwoko a, b & Wim J. Wolff a, *
a
Dept. of Marine Benthic Ecology and Evolution, University of Groningen, P.O. Box 14, 9750 AA Haren, The
Netherlands
E-mail addresses: w.j.wolff@rug.nl
b
Dept. of Biology, Sriwijaya University, Palembang, Indonesia
E-mail addresses: agus_purwoko2002@yahoo.com

ABSTRACT population has built up sufficiently to graze it.

The macrobenthic animal biomass of the intertidal Hence, they postulated a relationship between
area of the Sembilang peninsula of South Sumatra, intertidal benthic biomass and latitude
Indonesia, has been studied in 2004. Monthly suggesting this to result in lower biomasses in
(March – August) 21 core samples were taken at tropical areas compared to temperate areas.
each of six sampling stations. Macrobenthic fauna
was identified at the lowest taxonomical level
possible and counted. Biomass was measured as We studied the benthic fauna of the intertidal
ash free dry weight (afdw). The average biomass flats at Sembilang peninsula, South Sumatra,
over all stations and months was 3.62 g afdw m-2, Indonesia, in 2004 and found low biomasses
the highest biomass (14.09 g afdw m-2) found at a as well. Hence, in this paper we report on our
station in one month was due to abundant results and investigate the possible existence
occurrence of the bivalve Anadara granosa. Low of a positive relationship between latitude and
biomass of macrobenthic fauna at Sembilang biomass of macrobenthic fauna.
peninsula cannot easily be explained but is in line
with low biomasses found elsewhere in the tropics. For temperate and subtropical estuaries
For that reason we analyzed a data set of 268 soft-
Kalejta and Hockey (Kalejta & Hockey 1991)
bottom intertidal biomasses collected world-wide
to look for a relationship with latitude. It was found a negative relationship between
shown that average biomass of intertidal estuarine invertebrate production and distance
macrobenthic fauna in the tropics was significantly from the equator and a positive relationship
lower (p<0.05) than that at non-tropical sites. A between production and mean annual ambient
significant second-order relationship between temperature. However, production is not a
biomass of macrobenthic fauna and latitude was good predictor of biomass, so this paper does
established. not allow generalizations about a relationship
between benthic biomass and latitude. Riese et
Keywords al (1994) compared two temperate mudflats
biomass, macrobenthos, intertidal fauna, tropics, with a tropical one and concluded that biomass
Indonesia, Sembilang was lowest at the tropical site, although no
biomass values were given. Piersma et al.
1. Introduction (1993) compared benthic biomass in 20
estuaries between 57 o N and 34 o S and
(Warwick & Ruswahyuni 1987; Rehm 2006) concluded that there was no relationship
studied the benthic fauna at the N coast of between biomass and latitude although their
Central Java, Indonesia, and found low Fig. 2B suggests otherwise. Ricciardi and
intertidal biomasses. As an explanation they Bourget (1999) analyzed a data set from 245
suggested that in tropical waters with more or sedimentary shores and found that macro
less continuous phytoplankton production invertebrate biomass for sedimentary shores
zooplankton grazers are able to remain in did not vary linearly with latitude although
phase with phytoplankton and prevent that a peak values were found for northern and
large part of the primary production reaches southern temperate areas. All analyses
the bottom as food for benthic fauna. This suffered from the fact that data from the
should be in contrast to temperate latitudes tropics were less abundant than those from
where phytoplankton production is highly higher latitudes (Alongi 1990). We decided to
seasonal and much of the spring bloom settles reanalyze the data sets of Piersma et al. (1993)
at the bottom before the zooplankton

215
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

and of Ricciardi and Bourget (Ricciardi & is also influenced by the monsoon. During the
Bourget 1999) to investigate if the low rainy season, the height of the high tide will be
biomass at Sembilang tidal flats could be up to 1-2 m higher. Also low tide levels are
attributed to a biomass – latitude relationship. higher in that period, meaning that the tidal
flats emerge slowly (Tide table from Daftar
pasang surut Sungsang, Computer program
2. Methods TideWizard, and field observation).

2.1. Area description Salinity measured at low tide was 12-13 in the
wet season and 17-18 in the dry season
(Figure 1 about here) (Purwoko, unpublished observations).

The Sembilang peninsula (Fig. 1: 10 59’ to 20 2.2. Description of sampling stations

15’ S and 1040 45’ to 1040 53’ E) is located at
the eastern coast of South Sumatra, Indonesia, We selected 6 sampling stations (Fig. 1) with
and it is part of the Sembilang Nasional Park. varied characteristics:
It is influenced by the Musi River, the
Musibanyuasin River and some smaller 1. Station 1: Tj. (Tanjung) Carat (S: 20 16.324'
tributaries. Originally, the land was covered by & E: 1040 55.117' by Gekko 202 Garmin
mangrove and swampy forest. The most GPS). Located in the estuary of the Musi river;
seaward belt of the mangrove vegetation is the sediment is sandy, and the station is
still formed by Sonneratia and Avicennia strongly influenced by fresh water.
where the sediment is sandy; however, where
the sediment is muddy, Rhizophora occurs. 2. Station 2: S. (Sungai) Bungin (S: 20 14.955'
Nypa palms appear where the sediment is & E: 1040 50.714'). Located in the estuary of
highly influenced by fresh water. In the the Musibanyuassin river. The surface layer of
mangrove area at the Sembilang peninsula, the sediment is soft and high in organic matter
approximately 200 to 500 meter landward of and has a thickness of more than 1 meter. On
the beach, there are 4 000 ha of shrimp ponds. the sediment young Avicennia occur.
The coastline of the area faces directly Bangka
Strait and the East China Sea. 3. Station 3: Solok Buntu (S: 20 11.063' & E:
1040 54.764'). Located near the estuary of the
The area is characterized by a monsoon Musibanyuasin river. The surface layer of the
climate. The annual rainfall is 2 000 – 2 500 sediment is muddy, and reaches 40 cm depth.
mm. The maximum rainfall occurs in Mangrove vegetation consists mainly of
November and December (260 – 275 mm per Avicennia.
month) and the minimum in July and August
(140 – 200 mm per month). We have 4. Station 4: S. Barong Kecil (S: 20 9.872' &
distinguished a wet season from October to E: 1040 54.587'). The site is near the minor
April and a dry season from May to estuary of S. Barong Kecil. The sediment is
September. Daily temperature ranges between muddy and the depth of this layer reaches 50
20 – 32 o C and the humidity varies from 70 to cm.
90 percent (Source: Stasiun Klimatologi Klas I
Kenten Palembang). 5. Station 5: S. Siput (S: 20 5.824' & E: 1040
54.102'). The sediment is muddy and the soft
The sediment of the tidal flats ranges from fine layer reaches 40 cm; the adjacent mangrove
to medium sands to very soft mud with high vegetation is mostly Avicennia; close to the
organic content. Mostly, the sediment is acidic site young trees grow.
and the C:N ratio is high (Soeroyo & Suryaso
1999). 6. Station 6: S. Dinding (S: 20 1.924' & E: 1040
50.573'). The sediment is muddy and contains
There is one high tide and one low tide daily. many dead shells. The depth of the soft layer
The tide reaches a maximum height of 4 reaches 40 cm. The site is facing directly the
meters above low-tide level. Further, the tide

216
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

South China Sea, so it is exposed to high

waves. We used the data set . collected by Ricciardi
and Bourget (Ricciardi & Bourget 1999) and
2.3. Sampling procedure kindly made available by Dr. A. Ricciardi as a
starting point for our analysis. This data set
Each station consisted of 3 parallel line comprises 245 soft-bottom localities. To these
transects each consisting of 7 sampling points. soft-bottom data we added our own data, in
The distance between the line transects varied which we treated each of our sampling stations
from 5 to 10 m; the distance between the as one locality, resulting in 6 more locations.
sampling points was randomly chosen and Finally, we added the data from Piersma et al.
varied between 3 and 20 m. The first core (1993: Table 1) excluding those also reported
sample of the first transect was taken at the by Ricciardi & Bourget (Ricciardi & Bourget
lowest water level and the second till the 1999). In total our data set consisted of 268
seventh one were at further distance towards biomass values from soft-bottom tidal flats in
the mangrove. In the field we could not tropical, subtropical, temperate and polar
distinguish beforehand between areas of areas. To obtain normality we converted our
normal density of animals and areas of biomass data by log 10(afdw + 0.1). The
extremely high density (e.g., beds of bivalve Statistics 7 program was used for all data
filter feeders). Hence, we could not stratify our analyses.
samples accordingly. Samples were taken with
a circular corer. Core diameter was 15 cm and 3. Results
the sampling depth 30 cm. At each sampling
point 1 core sample was taken. The core 3.1. Biomass at Sembilang
samples directly were extracted by double
layers of 1 mm sieve in the field and the (about here Tables 1 and 2)
animals were collected from the sieve by hand.
Tables 1 and 2 (see also Figure 1) summarize
Sometimes sampling was difficult. Rough seas our results for the macrobenthic biomass of
caused difficulty approaching the sampling the intertidal area at Sembilang. The animal
sites, whereas big waves disturbed sampling biomass averaged over the 6 stations and the 5
activities, especially during sieving of the core months of sampling at Sembilang amounted to
samples. This may have reduced to some 3.62 g afdw m-2. The highest biomass of
extent the number of animals obtained. To macrobenthic animals occurred at station 4 in
overcome this problem, on some occasions June and July 2004 and was caused mainly by
intact core samples have been broken into a high biomass small individuals of the fast-
pieces before sieving and animals were growing bivalve Anadara granosa (Broom
collected directly from the core when we saw 1982). The biomass at station 4 (afdw: 14.09 g
them, especially worms. m-2) was significantly higher than those at the
other stations (ANOVA: p<0.05). Most of the
Macrobenthic fauna collected was preserved biomass derived from bivalves. Other animals,
in 70 % alcohol mixed with 3 % of formalin. such as gastropods, crabs and polychaetes
Sampling activities took place during low tide reached their highest biomass level in June
at Sungsang determined by a tide table (ANOVA: p<0.05), July and May 2004
(TideWizard) in March, May, June, July and (ANOVA: p<0.05), respectively (Table 2).
August 2004. The animal fauna in the samples
were identified to the lowest taxonomic level Taxa influenced by place (stations) and time
possible (Dharma 1988, 1992); Computer (months) were Tellina remies, T. timorensis,
Program: Poly Key), counted and ash-free dry Clithon oualaniensis, Nereididae, Maldanidae
weight (afdw) (Winberg & Duncan 1971) was and Lumbrineridae. Their biomasses were
measured at the laboratory at Palembang. The significantly different between stations and
statistical calculations were carried out by the between months, as indicated by the different
Statistics 7 Program. small letter (in brackets) in Tables 1 and 2.

2.4. Reanalysis of literature data

217
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

3.2. Biomass – latitude relationship 4. Discussion

4.1. Biomass per station at Sembilang
The analysis of the Piersma et al. (1993) data
set while keeping apart data from northern and The highest biomass occurred at station 4
southern latitudes, did result in positive which differed significantly from all other
correlations for latitude and biomass but none stations with by far the highest value (14.09 g
was significant (regression analysis: North: R2 m2) of ash-free dry weight of macrobenthic
= 0.15, p = 0.12; South: R2 = 0.55, p = 0.26). animals. This is mainly the result of the
The outcome improved when using absolute abundant occurrence of the bivalve Anadara
values for latitude but also was not significant granosa, which was rare or lacking at the
(R2 = 0.16, p = 0.07). other stations. A. granosa is a fast-growing
species occurring in dense beds. We
When we analysed our main data set of encountered especially 1 – 1.5 cm sized
biomasses while keeping apart data from individuals which according to (ref.) are less
northern and southern latitudes, the t-test and than 1 year old. However, their growth is not
regression analysis showed that the mean fast enough to explain their occurrence in our
values of macrobenthic biomass in tropical samples in June without a previous occurrence
and non-tropical areas were not significantly in May. Apparently, we have missed the A.
different for both northern (R2 = 0.002, p = granosa bed when sampling in May. This is
0.60; t-test p = 0.13) and southern (R2 =0.09, p assumption is supported by the strongly
= 0.62; t-test p = 0.32) latitudes. However, clustered distribution we observed in June.
when we classified our data without When we omitted the biomass values of
differentiating between southern and northern Anadara, statistical analysis showed that there
latitudes, Fig. 2 suggests that average were no significant differences among the 6
biomasses differs by latitude. Indeed, the mean stations for the total biomass of the
biomass of tropical (defined as between the macrobenthic fauna.
Tropic of Cancer at 23 o N and the Tropic of Polychaete biomass was significantly different
Capricorn at 23 o S) areas was significantly (t- between stations. The highest biomass value
test, p = 0.01) different from the biomass of was found at station 4. Further analysis
non-tropical areas. showed that the polychaete families
Nereididae, Maldanidae and Lumbrineridae
(About here Figure 2) each had their highest biomass at different
stations. The highest biomass per polychaete
Further regression analysis indicated that the taxon, such as Nereididae, Maldanidae and
hypothesis of a linear relationship was rejected Lumbrineridae, occurred at station 2, station 6
(R2 = 0.00, p = 0.74). The best fitting the and station 5, respectively.
relationship between biomass of macrobenthic
fauna and latitude zone was obtained by 4.2. Biomass per month at Sembilang
polynomial regression (R2 = 0.032, p = 0.003).
The equation of the regression is log 10(afdw There was also an influence of the month of
+ 0.1) = 0.2846 + 0.0241x - 0.0003x2, where x sampling on the biomass of the macrobenthic
is latitude values (Figure 3). fauna at Sembilang peninsula. Biomass of the
total macrobenthic fauna was not significantly
(About here Figure 3) different over the 5 months sampling period.
However, bivalves, gastropods and
As may be expected, in our large data set polychaetes showed significant differences.
water temperature is coupled significantly to The highest bivalve and gastropod biomass
latitude (R2 = 0.78, p = 0.00). Biomass (as was found in June, and the highest polychaete
afdw) shows a weak negative linear biomass in May 2004. Biomass of Tellina
correlation with water temperature (R2 = remies was lowest in August whereas biomass
0.011, p = 0.097) with zero biomass at about of T. timorensis was highest in July and
37 o Celsius. August 2004. Omitting the Anadara biomass
value did not influence the ANOVA result for
bivalves and total macrobenthic fauna

218
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

biomass, and there was no time influence for however, during long drought periods local
the Sembilang peninsula tidal flat. Maldanidae ducks also searched for food at the Sembilang
and Lumbrineridae had their highest biomass peninsula tidal flats. However, shorebird
value in May but Nereididae in August 2004. numbers at Sembilang are not exceptionally
high (Purwoko, in prep.).
4.3. Comparison of Sembilang biomass with
other studies Although we have no records on predation by
fish, there is evidence that some species prey
We found an average biomass over all stations on macrobenthic fauna. During field
and all months of 3.62 g ash-free dry weight observations at low tide, we observed tracks of
m-2. Biomass value ranged from 2.95 to 36.31 fish disturbance left on the substrate.
g afdw at east Java (Erftemeijer & Swennen
1990) and was 2.89 g at central Java (Warwick Humans can be important predators as well.
& Ruswahyuni 1987). Broom (1981) reported However, at Sembilang before 2005 Anadara
that the biomass value at a Malaysian mudflat and other shellfish were not harvested. On the
was 23.45 g afdw, so the Sembilang biomass other hand shrimps were caught with small
value was almost the same as the Java values trawls and standing nets. These activities may
but it was lower than the Malaysian one. have caused disturbance of the bottom leading
Parulekar et al.(1980) reported that the to lower macrobenthic biomass.
biomass of the macrobenthic fauna at a
tropical estuary in India was 54.2 g m-2 wet 4.4. Biomass - latitude study
weight. However, when it was converted to
afdw (Ricciardi & Bourget 1998), it was 3.34 The mean biomass of macrobenthic fauna in
g afdw m-2, being very close to our Sembilang the tropics was shown to be significantly
value. In Mauritania (Africa, 200 latitude), the lower compared to the mean biomass of
biomass was much higher (17.00 g ash-free macrobenthic fauna at non-tropical latitudes.
dry weight m-2) (Wolff et al. 1993). Ricciardi However, we could not demonstrate a
and Bourget (Ricciardi & Bourget 1999) significant linear relationship between benthic
reviewed a large number of intertidal biomass biomass and latitude. Instead we established a
studies and found for tropical sedimentary significant second-order relationship with the
shores average biomasses of about 7.00 g m-2 maximum biomass predicted at 40.17 degree
ash-free dry weight (0 – 20o N) and about 8.50 of latitudes. This is in line with our earlier
g m-2 (0 - 20o S). We conclude that the value conclusion of low biomasses in the tropics, but
for Sembilang peninsula is on the low side. we find it difficult to suggest a biological
explanation for this relationship.
Ricciardi and Bourget (1999) suggest also that
macrobenthic biomass at soft-sediment shores Tropical beaches and tidal flats are subject to
may be lowered by very muddy sediment, a climatic and physical disturbance (review by
high slope of the shore, high exposure to (Alongi 1990). Large rain fall, storms, sun
waves, and small waves. The Sembilang tidal exposure, and high temperatures disturb the
flats indeed are very muddy, but neither show macrobenthic habitat. However, physical
a high slope of the shore, nor, in most cases, disturbance is also found at higher latitudes
they are highly exposed to waves. The same and it is not immediately clear why it should
authors also suggest a relationship with have less effect on benthic biomass. Maybe,
latitude (see below). They do not consider our latitude – biomass relationship is the result
biological factors which of course may be of two stress gradients. One gradient might be
connected to the physical factors mentioned. related to the occurrence of freezing
temperatures and ice scour, leading to a
Exclusion experiments showed that at negative relationship between latitude and
Sembilang shorebirds were important biomass from the poles towards temperate
predators of the macrobenthic community. latitudes. The other one could be related to
Resident birds, mainly storks and herons, prey increasing temperatures leading to increasing
especially on fish, whereas migratory birds, heat stress on intertidal flats going from
mainly waders, feed on macrobenthic fauna; temperate latitudes to the equator.

219
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

(WJW) has seen tidal flats all over the world

Other possible explanations are ecological in and observed that in many developing tropical
nature. Warwick & Ruswahyuni (1987) countries coastal ecosystems are exploited
speculated that in the tropics less of the very intensively. Although we cannot support
primary production by phytoplankton reaches this observation quantitatively, it might mean
the bottom because production is more or less that tropical mudflats on average are exploited
continuous and in phase with zooplankton more heavily than non-tropical flats.
grazers. On the other hand, they argue, in
temperate regions primary production is Acknowledgments
highly seasonal and much of the spring bloom
settles to the bottom before the zooplankton First AP would like to thank to Sriwijaya
population has built up sufficiently to graze it. University, Palembang, Indonesia for
This might be true at Sembilang as well. sponsoring his study. AP thanks Ake and Usup
Phytoplankton occurs in low densities at for their assistance during sampling and Ismail
Sembilang (Sutomo 1999), possibly because for acting as a speedboat driver. He is also
of the high abundance of zooplankton at the glad that Edi P. assisted at measuring the
same time (Romimohtarto 1999). However, afdw. Finally, special thanks are addressed to
data are lacking to test the generality of this Dr. Harry ten Hove who guided us in
explanation. identifying the macrobenthic fauna and Dr.
Anthony Ricciardi who supplied his world-
The same applies to the suggestion that wide data set on macrobenthic biomasses.
tropical flats are subjected to higher
exploitation pressures by humans. One of us

References Estuaries: Part III - Annual Cycle of

Macrofaunal Distribution , Production and
Alongi, M. D. (1990). "The ecology of tropical Trophic." Indian Journal of Marine Science
soft-bottom benthic ecosystem." 9: 189-200.
Oceanographic Marine Biology 28: 381-496. Piersma, T., de Goeij, P. and Tulp, I. (1993).
Broom, M. J. (1981). "Structure and "An evaluation of intertidal feeding habitats
seasonality in a Malaysian mudflat from a shorebird perspective: Towards
community." Estuarine coastal shelf science relevant comparasions between temperate
15: 135-150. and tropical mudflats." Netherlands Journal
--- (1982). "Structure and seasonality in a of Sea Research 31(4): 503-512.
Malaysian mudflat community." Estuarine, Rehm, P., Thatje, S., Siegel M. U., and Brandt
Coastal and Shelf Science 15(2): 135-150. A. (2006). "Comparation and distribution of
Dharma, B. (1988). Siput dan kerang the paracarid crustacean fauna along a
Indonesia (Indonesian Shell). Jakarta., PT. latitudinal transect of Victoria Land (Ross
Sarana Graha.: 111 pages. Sea, Antartica) with special emphasis on the
--- (1992). Siput dan kerang Indonesia Cumacea." Polar Biol: 11 pages.
(Indonesian Shell II). Wiesbaden, Germany., Ricciardi, A. and E. Bourget (1998). "Weight-
Verlag Chista Hemmen: 135 pages. to-weight conversion factors for marine
Erftemeijer, P. and C. Swennen (1990). benthic macroinvertebrates." Marine
"Density and biomass of macrobenthic fauna Ecology Progress Series (MEPS) 163: 245-
of some intertidal areas in Java, Indonesia." 251.
Wallaceana 59-60: 1-6. --- (1999). "Global patterns of
Kalejta, B. and P. A. R. Hockey (1991). macroinvertebrate biomass in marine
"Distribution, abundance and productivity of intertidal communities." Marine Ecology
benthic invertebrates at the Berg River Progress Series (MEPS) 185: 21 -35.
estuary, South Africa." Estuarine Coastal Riese, K., E. Herre and M. Srurm (1994).
and Shelf Science 33(2): 175-191. "Biomass and abundace of macrofauna in
Parulekar, H. A., V. K. Dhargalkar and Y. S. intertidal sediments of Koeningshafen in the
S. Singbal (1980). ""Benthic Studies in Goa

220
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Northern Wadden Sea." Helgoleander

Meeresuntersuchungen 48(2-3): 201 - 215.
Romimohtarto, K. (1999). Komposisi dan
sebaran zooplankton (in Indonesian).
Ekosistem perairan sungai Sembilang,
Musibanyuasin, Sumatera Selatan. K.
Romimohtarto, Djamali, A. and Soeroyo
(editors). Jakarta, Pusat Penelitian dan
Pengembangan Oseanologi-LIPI (Lembaga
Ilmu Pengetahun Indonesia/Indonesian
Scientific Institute): 37-54.
Soeroyo and Suryaso (1999). Sifat-sifat kimia
tanah mangrove (in Indonesian). Ekosistem
perairan sungai Sembilang, Musibanyuasin,
Sumatera Selatan. K. Romimohtarto,
Djamali, A. and Soeroyo (editors). Jakarta,
Pusat Penelitian dan Pengembangan
Oseanologi-LIPI (Lembaga Ilmu
Pengetahuan Indonesia/Indonesian Scientific
Institute): 15-20.
Sutomo (1999). Kondisi korofil dan seston (in
Indonesian). Ekosistem perairan sungai
Sembilang, Musibanyuasin, Sumatera
Selatan. K. Romimohtarto, Djamali, A. and
Soeroyo (editors). Jakarta, Pusat Penelitian
dan Pengembangan Oseanologi-LIPI
(Lembaga Ilmu Pengetahuan
Indonesia/Indonesian Scientific Institute):
29-36.
Warwick, R. M. and Ruswahyuni (1987).
"Comparative study of the structure of some
tropical and temperate marine soft-bottom
macrobenthic communities." Marine
Biology V95(4): 641-649.
Winberg, G. G. and A. Duncan (1971).
Methods for the estimation of production of
aquatic animals. London and New York,
Academic press: 174.
Wolff, W. J., G. A. Duiven, P. Duiven, P.
Esselink, A. Gueye, A. Meijboom, G.
Moerland and J. Zegers (1993). "Biomass of
macrobenthic tidal flat of the Banc d'Arguin,
Mauritania." Hydrobiologia 258: 151-163.

221
 Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Average AFDW (gram/m2 monthly) at Sembilang peninsula Average

Species Station 1 Station 2 Station 3 Station 4 Station 5 Station 6 Station

Bivalves
1 Anadara granosa 0.0431 0.0000 1.1113 12.2685 0.0000 0.2047 2.2713
2 Hecuba scortum 0.0122 0.0098 0.0391 0.0095 0.0289 0.0046 0.0173
3 Solen sp. 0.1533 0.0000 0.0000 0.0000 0.0000 0.0000 0.0256
(a) (a) (b) (c) (bc) (a)
4 Tellina remies 0.0210 0.0892 0.5047 0.8736 0.6869 0.2765 0.4087
5 Tellina timorensis 0.2631(a) 1.8528(b) 0.0302(a) 0.0841(a) 0.1461(a) 0.0815(a) 0.4096
Total Bivalves 0.4927(a) 1.9518(a) 1.6852(a) 13.2357(b) 0.8619(a) 0.5673(a) 3.1324

Gastropods
(a) (b) (a) (a) (a) (a)
1 Clithon oualaniensis 0.0350 0.1648 0.0114 0.0734 0.0177 0.0430 0.0575
2 Littorina melanostoma 0.0000 0.0004 0.0000 0.0002 0.0000 0.0005 0.0002
3 Nassa serta 0.0517 0.0637 0.0333 0.0636 0.0337 0.0135 0.0432
4 Thais buccinea 0.0111 0.0151 0.0306 0.0096 0.0038 0.0298 0.0167
Total Gastropods 0.0978(bc) 0.2440(c) 0.0753(bc) 0.1467(c) 0.0551(a) 0.0868(bc) 0.1176

Decapods (Crabs)
1 Ocypodidae 0.0231 0.0551 0.2012 0.0463 0.2535 0.0350 0.1024
2 Leucociidae 0.0000 0.0116 0.0000 0.0000 0.0493 0.0000 0.0101
Total crabs 0.0231 0.0667 0.2012 0.0463 0.3028 0.0350 0.1125

Polychaetes
(b) (b) (a) (a) (a) (a)
1 Nereididae 0.1111 0.1182 0.0609 0.0654 0.0685 0.0721 0.0827
(a) (bc) (ab) (c) (c) (d)
2 Maldanidae 0.0040 0.0103 0.0060 0.0250 0.0153 0.0277 0.0147
(a) (ab) (ab) (b) (c) (ab)
3 Lumbrineridae 0.0079 0.0230 0.0288 0.0553 0.1133 0.0291 0.0429
4 Capitellidae 0.0027 0.0127 0.0034 0.0071 0.0064 0.0092 0.0069
5 Sternaspidae 0.0124 0.0111 0.0527 0.5013 0.0042 0.0075 0.0982
6 Unidentified worms 0.0085 0.0082 0.0142 0.0089 0.0000 0.0240 0.0106
Total Worms 0.1465(a) 0.1835(a) 0.1660(a) 0.6629(b) 0.2077(a) 0.1698(a) 0.2561

(a) (a) (a) (b) (a) (a)

Total all species 0.7601 2.4460 2.1278 14.0915 1.4275 0.8590 3.62

Table 1. The average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna at different stations (see Fig. 1) at
Sembilang peninsula from March to August 2004. Different letters (in brackets) after biomass values in the same
row (species) indicate significant differences (ANOVA: p<0.05).

Average afdw (gram/m2) at Sembilang peninsula Average

Species March May June July August Month
Bivalves
1 Anadara granosa 0.1706 0.0000 7.7037 3.4820 0.0000 2.2713
2 Hecuba scortum 0.0224 0.0067 0.0269 0.0115 0.0191 0.0173
3 Solen sp. 0.0000 0.0000 0.1278 0.0000 0.0000 0.0256
4 Tellina remies 0.5746 (b) 0.5163 (b) 0.4092 (b) 0.3420 (ba) 0.2011 (a) 0.4087
5 Tellina timorensis 0.0629 (a) 0.1427 (a) 0.1996 (a) 0.8470 (b) 0.7960 (b) 0.4096
Total bivalves 0.8305 (a) 0.6657 (a) 8.4672 (b) 4.6824(b) 1.0163 (a) 3.1324
Gastropods
1 Clithon oualaniensis 0.0224 (a) 0.0673 (ab) 0.0673 (ab) 0.1115 (b) 0.0190 (a) 0.0575
2 Littorina melanostoma 0.0000 0.0000 0.0009 0.0000 0.0000 0.0002
3 Nassa serta 0.0853 0.0269 0.0853 0.0075 0.0112 0.0433
4 Thais buccinea 0.0067 0.0125 0.0375 0.0193 0.0072 0.0167
Total gastropods 0.1145 (a) 0.1068 (a) 0.1911(b) 0.1383 (ab) 0.0375 (a) 0.1176
Decapods (Crabs)
1 Ocypodidae 0.0584 0.0147 0.0640 0.1446 0.0771 0.0718
2 Leucociidae 0.0000 0.0571 0.0213 0.0579 0.0675 0.0408
Total crabs 0.0584 0.0718 0.0853 0.2025 0.1446 0.1126
222
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Polychaetes
1 Nereididae 0.0224 (a) 0.0673 (b) 0.0494 (b) 0.1020 (c) 0.1723 (d) 0.0827
2 Maldanidae 0.0013 (a) 0.0269 (bc) 0.0269 (c) 0.0062 (a) 0.0121 (b) 0.0147
3 Lumbrineridae 0.0180 (a) 0.0629 (b) 0.0269 (a) 0.0594 (b) 0.0474 (b) 0.0429
4 Capitellidae 0.0011 0.0028 0.0180 0.0029 0.0099 0.0069
5 Sternaspidae 0.0016 0.4169 0.0314 0.0253 0.0158 0.0982
6 Unidentified worms 0.0131 0.0196 0.0135 0.0018 0.0053 0.0106
Total worms 0.0575 (a) 0.5964 (b) 0.1661 (a) 0.1976 (a) 0.2628 (a) 0.2561

Total all species 1.0609 1.4408 8.9097 5.2208 1.4612 3.62

Table 2: Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna in different months at Sembilang peninsula
from March to August 2004. Different letters (in brackets) after biomass values in the same row (species) indicate significant
differences (ANOVA: p<0.05).

Figure 1. Sampling stations along the coast of Sembilang peninsula. Each station consists of three transects each with seven
sampling points.

223
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

18
16
14

Average biomass
12
10
8
6
4
2
0
0-9.9 10-19.9 20-29.9 30-39.9 40-49.9 50-59.9 > 60
Latitude class

Figure 2. Average biomass (g afdw m-2) per 10 o latitude, northern en southern latitudes combined.

3.0

2.5

2.0

1.5
Log (afdw+0.1)

1.0

0.5

0.0

-0.5

-1.0

-1.5
-10 0 10 20 30 40 50 60 70 80
Latitude

Figure 3. The relationship between latitude and biomass. The relationship shown

fits the equation y = 0.2846 + 0.0241x - 0.0003x2 (0.95 conf. int.) with y = log10 (afdw + 0.1)

224
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Density and biomass of the macrobenthic fauna of the intertidal area in

Sembilang national park, South Sumatra, Indonesia
Agus Purwoko a, b & Wim J. Wolff a, *
a
Dept. of Marine Benthic Ecology and Evolution, University of Groningen, P.O. Box 14, 9750 AA Haren, The
Netherlands
E-mail addresses: w.j.wolff@rug.nl
b
Dept. of Biology, Sriwijaya University, Palembang, Indonesia
E-mail addresses: agus_purwoko2002@yahoo.com

ABSTRACT Piersma et al. (2005) at Mauritania, Northeast

Intertidal area of the Sembilang peninsula of South Australia, Western Australia, Mozambique,
Sumatra, Indonesia had been studied in 2004. Western Australia respectively.
Three replicated transect lines were located at each
of six sampling stations. One transect line consisted The Sembilang tidal flat area in South-east
of randomly 7 cores sampling monthly.
Sumatera, Indonesia, is an important stop-over
Macrobenthic fauna was identified at the lowest
taxonomical level, and counted. Biomass was site for migrating shorebirds. The area supplies
measured as ash free dry weight. The data food for resident and migrating birds. Starting
compiled by applying statistical analyzing. Stations in 1988, Danish and Netherlands researchers
and months variables showed significant collaborated with the Environmental Study
difference. The most abundance of macrobenthic Center staff of Sriwijaya University in
animals found was bivalve Tellina timorensis. exploring the Sembilang coastal area,
Biomass of the total macrobenthic fauna was not censusing the birds population and measuring
significantly different over the 5 months sampling socio-economical aspects (Danielsen, 1990 &
period. However, Anadara showed the highest Djamali, 1999). Some researchers, such as
biomass value.
Simanjuntak (1999), Soeroyo (1999), and
Suryaso (1999) from the Indonesian National
Introduction
Institute of Oceanography (Lembaga
Intertidal areas on sedimentary shores are
Oceanografi Nasional –Lembaga Ilmu
places of dynamic biological processes. In
Pengetahuan Indonesia) determined the bio-
many areas the intertidal is important for
geophysical and chemical characteristics of
sustaining the economical and ecological
mangrove areas of Sembilang and Bangka
functions of coastal areas for plants, animals,
strait. The first inventory of the macrobenthic
and local people. One of the ecological
fauna of the Sembilang tidal flats was
indicators of the state of the intertidal
conducted by Purwoko (1992). The same
ecosystem is animal macrobenthic biomass.
author (1996) also studied the macrobenthos
Observations on the biomass of the intertidal
biomass at S. (Sungai) Tengkorak, a river
areas on sedimentary shores have been made
mouth at the Sembilang peninsula.
mostly in temperate and subtropical regions
(e.g., Piersma et al., 1993, Ricciardi &
The objective of our quantification of the
Bourget, 1999). Alongi (1990) also reports
macrobenthos biomass is to determine the
that there is little information on intertidal
potential food supply for migrating shorebirds
macrobenthic fauna in the tropics; most of the
and any other predators (e.g., fish and
studies of tropical macrobenthos have been
fishermen) at the shores of Sembilang
carried out in lagoons and estuaries in India.
National Park.
More recent studies on macrobenthos of
tropical tidal flats are among others those by
Material and Methods
Wolff et al. (1993), Dittmann (1995 & 2000),
Area description
Pepping et al. (1999), De Boer (2000), and
The Sembilang peninsula (Fig. 1: 10 59’ to 20 Musibanyuasin River and some smaller
15’ S and 1040 45’ to 1040 53’ E) is located at tributaries. Originally, the land was covered by
the Eastern coast of South Sumatra, Indonesia, mangrove and swampy forest. The most
and it is part of the Sembilang Nasional Park. seaward belt of the mangrove vegetation is
It is influenced by the Musi River, the still formed by Sonneratia and Avicennia
225
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

where the soil is sandy; however, where the 2. Station 2: S. (Sungai) Bungin (S: 20 14.955'
soil is muddy, Rhizophora occurs. Nypa palms & E: 1040 50.714'). Located in the estuary of
appear where the soil is highly influenced by the Musibanyuassin river. The surface layer of
fresh water. In the mangrove area at the the sediment is soft and high in organic matter
Sembilang peninsula, approximately 200 to and has a thickness of more than 1 meter. On
500 meter from the beach, there are 4,000 ha the sediment young Avicennia occur.
of shrimp ponds. The coastline of the area
faces directly Bangka strait and the East China 3. Station 3: Solok Buntu (S: 20 11.063' & E:
Sea. Two villages are nearby. Sungsang 1040 54.764'). Located near the estuary of the
village is at the southern part and Sembilang Musibanyuasin river. The surface layer of the
village at the northern end of the peninsula. To soil is muddy, and reaches 40 cm depth.
reach Sungsang from the capital city Mangrove vegetation consists mainly of prepat
(Palembang) takes 2.5 hours by boats (90 km). ( Avicennia). At sea nearby, there are some
pole houses.
The area is characterized by a monsoon
climate. The annual rainfall is 2,000 -2,500 4. Station 4: S. Barong Kecil (S: 20 9.872' & E:
mm. The maximum rainfall happens in 1040 54.587'). The site is near the minor
November and December (260 – 275 mm per estuary of S. Barong Kecil. The soil is muddy
month) and the minimum in July and August and the depth of this layer reaches 50 cm.
(140 – 200 mm per month). We have
distinguished a wet season from October to 5. Station 5: S. Siput (S: 20 5.824' & E: 1040
April and a dry season from May to 54.102'). The soil is muddy and the soft layer
September. Daily temperature ranges between reaches 40 cm; the adjacent mangrove
20 – 32 o C and the humidity varies from 70 to vegetation is mostly Avicennia; close to the
90 percent (Source: Stasiun Klimatologi Klas I site young trees grow.
Kenten Pelambang).
6. Station 6: S. Dinding (S: 20 1.924' & E: 1040
The sediment of the tidal flats is varied. It 51.838'). The sediment is muddy and contains
ranges from fine to medium sands to very soft many dead shells. The depth of the soft layer
mud with high organic content. Mostly, the reaches 40 cm. The site is facing directly the
soil is acid and the C:N ratio is high (Suryaso, South China Sea.
1999).
Salinity at these stations was measured in 2005
There is one high tide and one low tide daily. and 2006 by Argento nitrite titration of field
The tide reaches a maximum height of 4 samples. The results show that the intertidal
meters above low-tide level. Further, the tide area of Sembilang peninsula is a brackish
is also influenced by the monsoon. During the water environment with major differences
rainy season, the height of the high tide will be between the wet season (June) and the dry
up to 1-2 m higher. Also low tide levels are season (October) (Table 1). The different
higher in that period, meaning that the tidal stations show minor differences of salinity,
flats emerge slowly (Daftar pasang surut both in the water and in the sediment.
Sungsang, and computer program:
TideWizard). Sampling procedure
Each station consisted of 3 parallel line
Description of sampling stations transects each consisting of 7 sampling points.
We selected 6 sampling stations (Fig. 1) with The distance between the line transects was 5
varied characteristics: m; the distance between the sampling points
was randomly chosen and varied between 3
1. Station 1: Tj. (Tanjung) Carat (S: 20 16.324' and 20 m. The first core sample of the first
& E: 1040 55.117' by Gekko 202 Garmin transect was taken at the lowest water level
GPS). Located at the estuary of the Musi river; and the second till the seventh one were at
the soil is sandy, and the station is strongly further distance towards the mangrove.
influenced by fresh water. Samples were taken with a circular corer. Core
diameter was 15 cm and the sampling depth 30
cm. At each sampling point 1 core sample was

226
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

taken. The core samples directly were The highest density of macrobenthic fauna,
extracted by a 1mm sieve in the field and the mainly bivalves, occurred at station 2,
animals were collected from the sieve by hand. however, the highest biomass occurred at
Macrobenthic fauna collected was preserved in station 4 (afdw: 14.1 g m-2) which is
70 % alcohol mixed with 3 % of formalin. significantly higher than at the other stations
Sampling activities took place during low tide (p<0.05).
determined by a tide table at Sungsang mouth
(TideWizard) in March, May, June, July and Comparison over time
August 2004. The animal fauna in the samples The highest standing biomass of macrobenthic
were identified to the lowest taxonomic level fauna occurred in June, but it was not
possible (Dharma 1988; Dharma 1992; significantly different from that in other
Computer Program: Poly Key), counted and months in 2004. Although the highest biomass
Ash Free Dry Weight (AFDW) (Winberg occurred in June, the highest number of
1971) was measured at the laboratory at animals occurred in July 2004. Most of the
Palembang. animal biomass derived from bivalves. Other
animals reached their highest biomass level at
The data on the weather were collected from a different time, such as gastropods in March,
the weather station at Palembang. The water crabs in August and worms in May 2004 (table
quality data were gathered by other 4b).
researchers (Simanjuntak 1999).
Animals which are influenced by place
The statistical calculations were carried out by (stations) and time (months) were Tellina
the Statistics 7 Program. remies, T. timorensis, Clithon oualaniensis,
Nereididae, Maldanidae and Lumbrineridae.
Results They were significantly different, as indicated
Species composition and density by the different small letter (in brackets) in
table 4a and 4b.
We caught 17 taxa of macrobenthic fauna at
the Sembilang peninsula. The composition of Discussion
the fauna and the average density of the Quality of the data
animals are shown in Table 2. The most
abundant macrobenthic species are the bivalve Sometimes sampling was difficult. Rough seas
Tellina timorensis and the gastropod Clinthon sometimes caused difficulty approaching the
oualaniensis. T. timorensis comprises 27.9 sampling sites, whereas big waves disturbed
percent of the total macrobenthic fauna; its sampling activities, especially during sieving
density is 1737.4 individuals per m2. of the core samples. This may have reduced
the number of animals obtained. To overcome
Comparison of stations this problem, on some occasions intact core
The density of animals at the 6 stations is samples have been broken into pieces before
significantly different (p < 0.05). The highest sieving and animals were collected directly
density of macrobenthic animals occurred at from the core when we saw them, especially
station 2 in July and August 2004 (Figure 2, worms. Due to the fine sieve (0.5 mm mesh)
Table 3). The distribution of the macrobenthic used to sieve sediment, worms broke into
fauna is contagious (Morisita index P = 2.250). fragments. This may have had an impact on
the number of worms counted.
Figure 3 show the percentages of the main
taxonomic groups at the stations at the Comparison of densities per station
Sembilang peninsula. Polychaete worms were The average densities of all species together
the most abundant animals at 5 out of 6 over the entire period differed significantly
sampling stations, and the bivalves usually between stations (Table 3). This was also true
were second. for the average densities of all species per
month, except for March (Table 3). Species
Table 4a and 4b illustrates the distribution of densities also differed. For example,
biomass as ash free dry weight (afdw). Tellina timorensis and Clithon oualaniensis
were found especially at station 2. Soft

227
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

sediment and high organic matter content at although the highest biomass of bivalves
station 2 might be favourable for these two occurred at station 2 (p<0.05), there were no
species. significant differences among the 6 stations for
the total biomass of the macrobenthic fauna.
Comparison of densities per month Other bivalves like Tellina remies and T.
Table 3 shows that average total densities per timorensis showed significant differences of
station differed significantly (Duncan biomass with biomass of T. remies highest at
test).Wet (May and June 2004) and dry (July station 4 and T. timorensis at station 2. Like T.
and August 2004) seasons were expected to be timorensis, Clithon had a high biomass at
correlated with the density of macrobenthic station 2.
fauna (sampling stations, ANOVA: p < 0.05);
however, further tests (Duncan test and t-test) Polychaete biomass was significantly different
could not show a significant difference between stations. The highest biomass value
between wet and dry season in 2004. was found at station 4. Further analysis
showed that the polychaete families
When we look further into the macrobenthic Nereididae, Maldanidae and Lumbrineridae
fauna at lower taxonomic level, we find that each had their highest biomass at different
Tellina timorensis and Clithon oualaniensis stations. The highest biomass of Nereididae,
show significantly different densities (Duncan Maldanidae and Lumbrineridae occurred at
test; p<0.05) in dry and wet seasons. They station 2, station 6 and station 5, respectively.
were more abundant in the wet season. It This may imply that the Nereididae preferred
might be hypothesized that they settled more the habitat which is frequently influenced by
in that period because they require some fresh water, but that the others avoided that
resource connected to the lower salinity during habitat.
the wet season.
Comparison of biomass per month
Comparison of densities with other studies There was also an influence of the month of
The average density of macrobenthic species sampling on the biomass of the macrobenthic
in our study site (6,219.6 individuals per m2) fauna at Sembilang peninsula. Biomass of the
was lower than that (22,591 to 52,914 total macrobenthic fauna was not significantly
individuals per m2) at a mangrove area in different over the 5 months sampling period.
Florida, USA (Sheridan 1997). However, the However, bivalves, gastropods and worms
proportion of bivalves and worms (82%) was showed significant differences. The highest
higher at Sembilang peninsula than that in the biomasses of bivalves, gastropods and worms
Zuari estuary in India (70%) (Parulekar et al., were found in July, June and May 2004,
1980). respectively.
Biomass of Tellina remies was lowest in
Comparison of biomass per station August whereas biomass of T. timorensis was
Although the highest density of macrobenthic highest in July and August 2004. Omitting the
fauna occurred at station 2, the highest Anadara biomass value did not influence the
biomass occurred at station 4. This station analysis of biomass of bivalves, because of the
differed significantly from all other stations. same result of ANOVA. Maldanidae and
Station 4 which had a low density (286.06 Lumbrineridae had their highest biomass value
individuals per m2) compared to station 2 in May but Nereididae in July.
(474.61 individuals per m2), contained by far
the highest value (14.1 g m2) of ash-free dry Comparison of biomass with other studies
weight of macrobenthic animals. This is We found an average biomass over all stations
mainly the result of of the abundant and all months of 3.6 g ash-free dry weight m-
2
occurrence of the bivalve Anadara granosa, . Parulekar et al. (1980) reported that the
which was rare or lacking at the other stations. biomass of the macrobenthic fauna at a
The biomass of Anadara did not differ tropical estuary in India was 54.2 g m-2 (wet
between our 6 stations significantly, but it weight). However, when the value of ash-free
influenced the result of of the statistical dry weight was converted to wet weight
analysis. When we omitted the biomass values (Ricciardi and Bourget, 1998), the Sembilang
of Anadara, the statistical analysis showed that average biomass was 58.8 g m-2, being very

228
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

close to the Indian value. Ricciardi and on the measuring the AFDW. Finally, special
Bourget (1999) reviewed a large number of thanks addressed to Harry ten Hove who
intertidal biomass studies and found for guided me to identify the macrobenthic fauna.
tropical sedimentary shores average biomasses
of about 7.0 g m-2 ash-free dry weight (0 – 20o References
N) and about 8.5 g m-2 (0 - 20o S). We Alongi, D. M., 1990. The ecology of tropical
conclude that the value for Sembilang soft-bottom benthic ecosystem.
peninsula is on the low side. Oceanography and Marine Biology Annual
Review 28: 381-496.
Relationship with environmental factors Danielsen, F. & W. Verheugt, 1990.
Intertidal benthic biomass is the result of Integrating conservation and land-use
recruitment, growth and mortality. Among the planning in the coastal region of South
factors influencing these processes are water Sumatra. Asian Wetland Bureau-PHPA,
quality, sediment quality and biotic factors Bogor, Indonesia,.
such as predation. De Boer, W. F., 2000. Between the tides. The
imact of human exploitation on an intertidal
Water quality includes salinity. High river ecosystem, Mozambique. PhD thesis,
discharges may impede settlement and growth University of Groningen.
of marine and many estuarine animals as well Dharma, B., 1988. Siput dan kerang Indonesia
as cause high mortality because of too low (Indonesian Shells)., PT. Sarana Graha,
salinities. Hence, it is hypothesized that Jakarta.
biomass will decrease from the most seaward Dharma, B., 1992. Siput dan kerang Indonesia
location (station 6) to the most riverward (Indonesian Shells II). Verlag Christa
locations (stations 1 and 2). However, there is Hemmen. Wiesbaden, Germany.
no clear trend in the salinity data (Table 1) as Dittmann, S., 1995. Benthos structure on
well as the biomass data (Table 4a); only tropical tidal flats of Australia. Helgoländer
station 4 has a significantly higher biomass Meeresuntersuchungen 49: 539-551.
whereas all other stations do not differ Dittmann, S., 2000. Zonation of benthic
significantly. communities in a tropical tidal flat of
Simanjuntak (1999) reported that the wet northeast Australia. Journal of Sea Research
season caused reduced salinity. This was 43: 33-51.
confirmed by our study. Djamali, A. &. Sutomo, 1999. Kondisi sosial
ekonomi budaya, sosial ekonomi budaya dan
The sediment at station 1 is sandy, whereas all perikanan. In Romimohtarto K., A. Djamali,
other stations are very muddy. This difference & Soeroyo (eds), Ekosistem perairan sungai
does not result in a clear difference between Sembilang, Musibanyuasin, Sumatera
these stations with regard to benthic biomass. Selatan. Pusat Penelitian dan Pengembangan
With the exception of station 4, all stations Oseaneologi-LIPI, Jakarata.
have average biomasses not differing Parulekar, A. H., V. K. Dhargalkar, & Y. S. S.
significantly. Singbal, 1980. Benthic Studies in Goa
Estuaries: Part III - Annual Cycle of
We have insufficient data on predation of Macrofaunal Distribution, Production and
benthic fauna. We have made bird censuses Trophic Relations. Indian Journal of Marine
(Purwoko, in prep.) en we have observed that Science 9: 189-200.
Anadara granosa is fished near station 4, 5 Pepping, M., T. Piersma, G. Pearson & M.
and 6 in 2005. We do not have any Lavaleye, 1999. Intertidal sediments and
information on fish and other nekton feeding benthic animals of Roebuck Bay, Western
in the intertidal at high tide. Australia. NIOZ-Report 1999-3: 212 pp.
Netherlnads Institute for Sea Research,
Acknowledgements Texel.
First I would like to thank to Sriwijaya Piersma, T., P. de Goeij, & I. Tulp, 1993. An
University for sponsoring my study. Secondly, evaluation of intertidal feeding habitats from
AP thanks to Ake for his assistance during a shorebird perspective: Towards relevant
sampling and Ismail for acting as a speedboat comparisons between temperate and tropical
driver. I was also glad that Edi P. assisted me

229
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

mudflats. Netherlands Journal of Sea Musibanyuasin, Sumatera Selatan. Pusat

Research 31: 503-512. Penelitian dan Pengembangan Oseaneologi-
Piersma, T., G. B. Pearson, R. Hickey & M. LIPI, Jakarata.
Lavalaye, 2005. The Long Mud. Benthos Soeroyo, 1999. Struktur, komposisi, zonasi
and shorebirds of the foreshore of Eighty- dan produksi serasah mangrove. In
mile Beach, Western Australia. NIOZ- Romimohtarto K., A. Djamali, & Soeroyo
Report 2005-2: 218 pp. Royal Netherlands (eds), Ekosistem perairan sungai Sembilang,
Institute for Sea Research, Texel. Musibanyuasin, Sumatera Selatan. Pusat
Purwoko, A., 1992. The inventory of Penelitian dan Pengembangan Oseaneologi-
macrobenthos in S. Sembilang - Sungsang. LIPI, Jakarata.
Lembaga Penelitian Univ. Sriwijaya. Suryaso, 1999. Peranan proses fisika dalam
Palembang, Indonesia. evolusi mangrove. In Romimohtarto K., A.
Purwoko, A., 1996. The carrying capacity of Djamali, & Soeroyo (eds), Ekosistem
intertidal area: a feeding habitat of the stop perairan sungai Sembilang, Musibanyuasin,
over of migratory birds in the Sembilang Sumatera Selatan. Pusat Penelitian dan
peninsula, South Sumatra, Indonesia. Pengembangan Oseaneologi-LIPI, Jakarata.
Environmental Science. Wageningen., Winberg, G. G. & A. Duncan, 1971. Methods
Wageningen Agricultural University, for the estimation of production of aquatic
Wageningen. animals. Academic press, London and New
Ricciardi, A. & E. Bourget, 1999. Global York.
patterns of macroinvertebrate biomass in Wolff, J. W., G. A. Duiven, P. Duiven, P.
marine intertidal communities. Marine Esselink, A. Gueye, A. Meijboom, G.
Ecology Progress Series 185: 21-35. Moerland & J. Zegers, 1993. Biomass of
Simanjuntak, M., 1999. Kondisi kimia macrobenthic tidal flat fauna of the Banc
oseanografi perairan hutan mangrove. In d'Arguin, Mauritania. Hydrobiologia 258:
Romimohtarto K., A. Djamali, & Soeroyo 151-163.
(eds), Ekosistem perairan sungai Sembilang,

230
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Figure 1: Location of the 6 sampling stations at Sembilang peinsula. Source: Sutaryo et al. (2001)

Macrobenthic fauna at Sembilang peninsula in

2004
1000.00

900.00
individual/square mtr

800.00

700.00
March
600.00 May
500.00 June
400.00 July
August
300.00

200.00

100.00

0.00
Station 1 (Tj. Station 2 (S. Station 3 Station 4 (S. Station 5 (S. Station 6 (S.
Carat) Bungin) (Solok Buntu) Barong Kecil) Siput) Dinding)

Stations

Figure 2: The density of macrobenthic fauna at 6 different sampling stations at Sembilang peninsula from March to August
2004.

231
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Figure 3: Distribution of main taxonomic groups of macrobenthic animals at Sembilang peninsula in 2004

Average Salinity (psu)

June 2006 Oct. 2006 2005
Stations water water sediment
Station 1: Tj. Carat 13.01 17.86 25.50
Station 2: S. Bungin 12.54 17.34 26.50
Station 3: Solok Buntu 12.33 18.12 26.33
Station 4: S. Barong K 12.43 18.54 26.50
Station 5: S. Siput 12.53 17.66 27.00
Station 6: S. Dinding 12.43 18.34 27.34

Table 1. The salinity of the intertidal area of Sembilang peninsula. Samples were taken at low tide.

232
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

TAXA Animal density (2004)

Bivalves Indivual/m2 Percentage
1 Anadara granosa 21.54 0.3
2 Hecuba scortum 202.02 3.2
3 Solen sp. 48.48 0.8
4 Tellina remies 705.72 11.3
5 Tellina timorensis 1737.37 27.9
Total bivalves 2715.15 43.7
Gastropods
1 Clithon oualaniensis 870.03 14.0
2 Littorina melanostoma 16.16 0.3
3 Nassa serta 145.45 2.3
4 Thais buccinea. 70.03 1.1
Total gastropods 910.44 14.6
Crabs
1 Ocypodidae 148.15 2.4
2 Leucociidae 2.69 0.0
Total crabs 150.84 2.4
Worms
1 Nereididae 630.30 10.1
2 Maldanidae 595.29 9.6
3 Lumbrineridae 692.26 11.1
4 Capitellidae 258.59 4.2
5 Sternaspidae 439.06 7.1
6 Unidentified worms 355.56 5.7
Total worms 2443.10 39.3
Total macrobenthic
fauna 6219.53 100.0

Table 2 Average density of each macrobenthic taxon at Sembilang peninsula over all stations and months in 2004
2
Density macrobenthic fauna in 2004 (individuals/m ) Average
Stations March May June July August (station)
Station 1 (Tj. Carat) 204.71 226.26 (c) 183.16 (b) 145.45 (b) 142.76 (b) 180.47(c)
(c) (a) (a) (a) (a)
Station 2 (S. Bungin) 140.07 180.47 296.30 929.29 826.94 474.61
(bc) (b) (b) (b) (c)
Station 3 (Solok Buntu) 220.88 293.60 175.08 167.00 169.70 205.25
(b) (a) (b) (b) (b)
Station 4 (S. Barong Kecil) 167.00 414.81 366.33 266.67 215.49 286.06
(a) (b) (b) (b) (bc)
Station 5 (S. Siput) 142.76 568.35 96.97 258.59 274.75 268.28
Station 6 (S. Dinding) 156.23 261.28 (c) 339.39 (a) 177.78 (b) 204.71(b) 227.88(bc)
(c) (ab) (ac) (b) (ab)
Average (month) 171.94 324.13 242.87 324.13 305.72

Table 3: Average density of macrobenthic fauna at 6 stations and in 5 months at Sembilang peninsula. Different letters (in
brackets) after density values in the same column (March to August, and Average months) indicate significantly different
values (Duncan test: p<0.05). Last row excluded from the columns. Different letters (in brackets) after density values in the
last row indicate significant differences (p<0.05) between months. March values not significantly different.

Macrob. Species Average AFDW (gram/m2 monthly) at Sembilang peninsula Average

Bivalves Station 1 Station 2 Station 3 Station 4 Station 5 Station 6 Station
1 Anadara granosa 0.0431 0.0000 1.1113 12.2685 0.0000 0.2047 2.2713
2 Hecuba scortum 0.0122 0.0098 0.0391 0.0095 0.0289 0.0046 0.0173
3 Solen sp. 0.1533 0.0000 0.0000 0.0000 0.0000 0.0000 0.0256
4 Tellina remies 0.0210(c) 0.0892(c) 0.5047(b) 0.8736(a) 0.6869(ba) 0.2765(c) 0.4087
5 Tellina timorensis 0.2631(b) 1.8528(a) 0.0302(b) 0.0841(b) 0.1461(b) 0.0815(b) 0.4096
(b) (b) (b) (a) (b) (b)
Total Bivalves 0.4927 1.9518 1.6852 13.2357 0.8619 0.5673 3.1324
Gastropods
1 Clithon oualaniensis 0.0350(b) 0.1648(a) 0.0114(b) 0.0734(b) 0.0177(b) 0.0430(b) 0.0575
2 Littorina melanostoma 0.0000 0.0004 0.0000 0.0002 0.0000 0.0005 0.0002

233
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

3 Nassa serta. 0.0517 0.0637 0.0333 0.0636 0.0337 0.0135 0.0432

4 Thais buccinea. 0.0111 0.0151 0.0306 0.0096 0.0038 0.0298 0.0167
Total Gastropods 0.0978(bc) 0.2440(abc) 0.0753(a) 0.1467(ab) 0.0551(c) 0.0868(bc) 0.1176
Decapods (Crabs)
1 Ocypodidae 0.0231 0.0551 0.2012 0.0463 0.2535 0.0350 0.1024
2 Leucociidae 0.0000 0.0116 0.0000 0.0000 0.0493 0.0000 0.0101
Total crabs 0.0231 0.0667 0.2012 0.0463 0.3028 0.0350 0.1125
Worms
(a) (a) (b) (b) (b) (b) (b)
1 Nereididae 0.1111 0.1182 0.0609 0.0654 0.0685 0.0721 0.0827
(d) (bc) (dc) (b) (b) (a)
2 Maldanidae 0.0040 0.0103 0.0060 0.0250 0.0153 0.0277 0.0147
3 Lumbrineridae 0.0079(c) 0.0230(bc) 0.0288(bc) 0.0553(b) 0.1133(a) 0.0291(bc)
0.0429
4 Capitellidae 0.0027 0.0127 0.0034 0.0071 0.0064 0.0092 0.0069
5 Sternaspidae 0.0124 0.0111 0.0527 0.5013 0.0042 0.0075 0.0982
6 Unidentified worms 0.0085 0.0082 0.0142 0.0089 0.0000 0.0240 0.0106
(b) (b) (b) (a) (b) (b)
Total Worms 0.1465 0.1835 0.1660 0.6629 0.2077 0.1698 0.2561
Total macrobenthic
(b) (b) (b) (a) (b) (b)
Animals 0.7601 2.4460 2.1278 14.0915 1.4275 0.8590 3.62

Table 4a: The average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna at different locations at Sembilang
peninsula from March to August 2004. Different letters (in brackets) after biomass values in the same row (species) indicate
significant differences (p<0.05)

Macrob. Species Average afdw (gram/m2) at Sembilang peninsula Average

Bivalves March May June July August Month
1 Anadara granosa 0.1706 0.0000 7.7037 3.4820 0.0000 2.2713
2 Hecuba scortum 0.0224 0.0067 0.0269 0.0115 0.0191 0.0173
3 Solen sp. 0.0000 0.0000 0.1278 0.0000 0.0000 0.0256
(a) (a) (a) (ba) (b)
4 Tellina remies 0.5746 0.5163 0.4092 0.3420 0.2011 0.4087
(b) (b) (b) (a)
5 Tellina timorensis 0.0629 0.1427 0.1996 0.8470 0.7960 (a) 0.4096
(b) (b) (ab) (a)
Total bivalves 0.8305 0.6657 8.4672 4.6824 1.0163 (b) 3.1324
Gastropods
(b) (ab) (ab) (a) (b)
1 Clithon oualaniensis 0.0224 0.0673 0.0673 0.1115 0.0190 0.0575
2 Littorina melanostoma 0.0000 0.0000 0.0009 0.0000 0.0000 0.0002
3 Nassa serta 0.0853 0.0269 0.0853 0.0075 0.0112 0.0433
4 Thais buccinea 0.0067 0.0125 0.0375 0.0193 0.0072 0.0167
(b) (b) (a) (ab) (b)
Total gastropods 0.1145 0.1068 0.1911 0.1383 0.0375 0.1176
Decapods (Crabs)
1 Ocypodidae 0.0584 0.0147 0.0640 0.1446 0.0771 0.0718
2 Leucociidae 0.0000 0.0571 0.0213 0.0579 0.0675 0.0408
Total crabs 0.0584 0.0718 0.0853 0.2025 0.1446 0.1126
Worms
(d) (c) (c) (b) (a)
1 Nereididae 0.0224 0.0673 0.0494 0.1020 0.1723 0.0827
(c) (ab) (a) (c) (b)
2 Maldanidae 0.0013 0.0269 0.0269 0.0062 0.0121 0.0147
(b)
3 Lumbrineridae 0.0180 0.0629 (a) 0.0269 (b) 0.0594 (a)
0.0474 (a) 0.0429
4 Capitellidae 0.0011 0.0028 0.0180 0.0029 0.0099 0.0069
5 Sternaspidae 0.0016 0.4169 0.0314 0.0253 0.0158 0.0982
6 Unidentified worms 0.0131 0.0196 0.0135 0.0018 0.0053 0.0106
Total worms 0.0575 (b) 0.5964 (a) 0.1661 (b) 0.1976 (b) 0.2628 (b) 0.2561
Total
macrobenthic animals 1.0609 1.4408 8.9097 5.2208 1.4612 3.62

Table 4b: Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna in different months at Sembilang
peninsula from March to August 2004. Different letters (in brackets) after biomass values in the same row (species) indicate
significant differences (p<0.05)

234
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Intelligent traffic light system for AMJ highway

Nur Ilyana Anwar Apandi(1), Puteri Nurul Fareha M. Ahmad Mokhtar(2), Nur Hazahsha Shamsudin
and Anis Niza Ramani(3) and Mohd Safirin Karis
(1)
Mathematics Lecturer, Faculty of Electrical Engineering, Universiti Teknikal Malaysia Melaka (UTeM),
Melaka, Malaysia.
Email: ilyana@utem.edu.my
(2)
Department of Control, Instrumentation & Automation, Faculty of Electrical Engineering
Email: diba_iman@yahoo.com
(3)
Department of Industrial Power, Faculty of Electrical Engineering, UTeM, Malaysia.
Email: nurhazahsha@utem.edu.my,anisniza@utem.edu.my, safirin@utem.edu.my

ABSTRACT numbers of road user lead to increasing of

The fuzzy logic system to reduce the long term traveling times. Traffic in a city is very much
average waiting times is discussed. A fuzzy logic affected by traffic light controllers. Nowadays,
control scheme to regulate the flow of heavy the problem of traffic light system that most
morning traffic approaching a set of three occur is the long waiting time for vehicles
intersections along (Alor Gajah-Melaka-Jasin)
standing at certain traffic light before passing
AMJ highway is presented. The signal timing
parameters, green phase splits and offset are to another intersection especially during peak
adjusted based on the actual traffic approaching hour. Many of the lights seem to be set for
each intersection. An adaptive Fuzzy Logic Traffic longer red phases causing more vehicles to be
Controller is used to adjust the green phase splits at a standstill for unnecessary long periods of
approaches of each traffic signal. This system uses time. This problem caused the higher level
a fuzzy rule based system for its decision making to efficiency usage of time and fuel to the driver.
coordinate all three intersections simultaneously. There are many conventional controllers for
This system provides the controller with traffic traffic light system that has been discussed by
densities in the lanes and allows a better previous research project [2]-[5] to minimize
assessment of changing traffic patterns. As the
the waiting times at traffic light. Based on
traffic distributions fluctuate, the fuzzy controller
can change the signal light accordingly. Marco Wiering et al [2, 3] which utilized the
multi-agent Reinforcement Learning (RL) to
KEYWORDS train traffic signal controller, the goal is to
Intelligent traffic light system; Fuzzy logic control minimize the overall waiting time of cars in a
system; Multiplan system. city.
Fuzzy logic has been introduced and
I.INTRODUCTION successfully applied to a wide range of
The traffic light system at AMJ (Alor Gajah – automatic control tasks. The fuzzy controller
Melaka Tengah – Jasin) highway road consist changes the cycle time of the lights depending
of many intersections and each intersection upon the observed accumulation of cars
have their own traffic light which certainly behind the green and red lights and the current
busy during peak hour. During this hour, there cycle time. It allows the implementation of
is increase number of vehicles which come real-life rules similar to the way humans
from different destination will pass through would think [4]. The implementation of an
the sequence of traffic lights which this intelligent traffic lights control system using
situation may caused the traffic problem when fuzzy logic technology has the capability of
there is slightly delay of traffic green light mimicking human intelligence for controlling
changing. traffic lights.
This paper consider the road of AMJ highway Fuzzy logic traffic lights control is an
from the junction of Jalan Duyung-Ayer alternative to conventional traffic lights
Keroh until Jalan Melaka Sentral where main control which can be used for a wider array of
destination from rural areas to the city traffic patterns at an intersection. A fuzzy
located[1]. logic controlled traffic light uses sensors that
Traffic light is used to regulate and optimize count cars instead of proximity sensors which
the flow of converging traffic. The growing only indicate the presence of cars.

235
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

A multiagent system is proposed [6] to weight of vehicles at queue area and the
overcome the lack of interactions between the extension of green light will depend on their
neighbouring intersections. A set of three weights. The sensors which are located behind
intersections is coordinated using local fuzzy the load detector counts the number of
logic controllers located at each traffic signal. vehicles coming to the intersection.
A multiagent system employing a fuzzy This research is design to overcome
knowledge base is then proposed to coordinate traffic flow problem during peak hour
the three intersections based on the traffic situation, which the ideas is the load detector
conditions at all three intersections. will detects a heavy traffic load and the sensor
The intelligent traffic light system for AMJ at arrival area count the number of vehicles
highway is design to run the traffic smoothly crossing it, the longer extension of green light
during peak hour. At a constant speed, will be adjust to make sure that the vehicles
vehicles can easily smooth travelling through a travelling smoothly through traffic light A
sequence of traffic lights without have to stop continuosly.
and waiting too much time for the next trip.
III. TRAFFIC FUZZY CONTROLLER
II. SYSTEM DESIGN
A. Design Fuzzy Logic Control
The design controller analyzes a vehicle flow There are a few steps in design fuzzy logic
through a sequence of traffic lights standing at controller. The input and output of the system
various distances and then controls these must be identified before designing the fuzzy
traffic lights time duration while trying to keep logic controller. List of input and output for
the vehicle flow at the maximum speed the fuzzy logic controller is shown in Table 1.
allowed for the road. Figure 1 shows the Table 1. Input and output for the controller
situation of traffic flow approaching three No Input Output
1 Numbers of queue vehicles Extension of
intersections. The traffic flow approaching the 2 Numbers of arrival vehicles green light
three intersections is regulated by adjusting the 3 Speed of arrival vehicles
green phase. A fuzzy logic controller is
employed to adjust the green phase of the B. Fuzzy Inference System Editor
individual traffic signals. The three The Fuzzy Inference System Editor or FIS
intersections are coordinated by adjusting their Editor described the trafficlight where The
respective green phase. The restriction used in Mamdani method is used [7]. The Mamdani
this system consists of a maximum detection method or max-min operation is used in the
of 100 vehicles which is calculated based on processing stage or inference engine stage.
the case study that has been made. While, Centroid method is used for the
defuzzification stage.

These traffic lights system is using three fuzzy

input variables and one fuzzy output variable.
The inputs which are the quantities of traffic
on the queue side, the quantity of traffic on the
arrival side and the speed of vehicle needed to
flow approaching the three intersections. The
output fuzzy variable is the extension of time
Figure 1: Traffic flow approaching three intersections needed for the green light.

Based on Figure 1, the system design B. Membership Function Editor

will allow vehicles to flow from A to D at In constructing the membership function, the
three intersection running smoothly. A fuzzy input and output variables have to be
logic traffic controller uses counter sensors quantized into several modules or fuzzy
that count the number of vehicles and load subsets. According to the traffic lights control
detectors that detect the weight of vehicles system, four membership functions for input 1
stand on it for each lane. The load detector is (queue) which are many (MY), medium (MD),
placed underground of the road which is small (SM), few (FW), four membership
behind each traffic light. It will detect the functions for input 2 (arrival) which are many

236
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

(MY), medium (MD), small (SM), few (FW), of numbers are take based on the case study
three membership functions for input 3 (speed) that has been done.
which are fast(FS), constant (CNT), slow(SL)
and three membership functions for output B. Rule Editor
(extension of green light) which are long (LO), This fuzzy rule base serves for insertion of
medium (MD), short (SH) are needed to new fuzzy rule used in this design as in
quantized each of the fuzzy variables. following:
If (queue is MD) or (arrival is MY) or (speed
is CNT) then (extension_of_green_light is
LO)
If (queue is FW) or (arrival is MY) or (speed
is FS) then (extension_of_green_light is
SH)
If (queue is SM) or (arrival is SM) or (speed is
CNT) then (extension_of_green_light is
(a) queue SH)
If (queue is FW) or (arrival is FW) or (speed is
FS) then (extension_of_green_light is SH)
If (queue is MY) or (arrival is MD) or (speed
is CNT) then (extension_of_green_light is
LO)
If (queue is SM) or (arrival is MD) or (speed is
(b)arrival FS) then (extension_of_green_light is MD)
If (queue is MY) or (arrival is MY) or (speed
is SL) then (extension_of_green_light is
LO)
If (queue is FW) or (arrival is FW) or (speed is
CNT) then (extension_of_green_light is
SH)

(c) speed IV. RESULTS & DISCUSSIONS

Based on the case study that has been made,
the maximum vehicles flow smoothly through
an intersection during peak hour are about 100
vehicles. Table 2 below shows the data of
extension of green light for current traffic
situation.
Table 2: Case study data for current traffic situation
(d) extension of green light Number of vehicle Extension of
Figure 2. The Membership function for input fuzzy sets: passing the Green Light
(a) the number of queuing vehicles (b)the number of intersection (seconds)
arrival vehicles (c)the speed of arrival vehicles and 100 60
(d)extension of green light. 80 48
The Membership Function Editor as in Figure 60 39
2 shows that the queue and arrival input 30 30
presenting the numbers of vehicle. All of the
Based on the data, the number of vehicles
variables is using triangular shape membership
passing the intersection did not separate to
function or trimf.
queue and arrival sides. The amounts of
As shown in Figure 2, the range of queue side
vehicles are calculated through the video that
is between 0 to 70 vehicles, while the range of
has been taken at the study area. Real case
arrival side is between 0 to 30 vehicles where
studies as shown in Table 2 found that 60
the speed of vehicle is in the range 0 to
seconds of extension green light for 100
80km/h. The output fuzzy variable shown in
vehicles travel throung the intersection.
Figure 2 is the extension of green light with
The comparison of extension of green light
the range between 0 to 60 seconds. The range
between two situations of speed is also carried

237
 Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

out. Table 3 shows the extension of green light light. The adjustments are made automatically
when the fuzzy logic controller system is in response to the traffic situation.
applied. The data is taken based on the Next, a simulation was carried out to compare
constant and various speed conditions. the performance of the fuzzy logic controller
with Multiplan system. Our extension analysis
Table 3: Extension of green light by using fuzzy logic shows that the flow densities of the fuzzy logic
controller system
Input Output controller system is better than Multiplan
Queue Arrival Constant Various Extension Extension system because of the range waiting time and
(No. of (No. of Speed* Speed of Green of Green moving time are shorter and do help the traffic
vehicles) vehicles) (km/h) (km/h) Light* Light consistent.
(seconds) (seconds) Based on the result, the fuzzy logic controller
70 30 60 20 41.1 53.6
70 0 60 20 36.1 36.1
system proposed will improve AMJ highway
30 30 60 40 34.3 35 flow especially during peak hour. It will make
20 30 60 70 30.5 32.1 the road users to have a smooth drive and also
50 10 60 50 32.1 36.1 can avoid traffic problem. The results
40 20 60 50 36.5 34 observed that the fuzzy logic controller system
0 30 60 80 30.1 19.9 provides better performance in terms of total
70 10 60 50 37.5 36.7
50 20 60 30 38.2 38.4
waiting time as well as total moving time. Less
waiting time not only reduce the fuel
consumption but also reduce air and noise
Result shown as in Table 5.2, by using fuzzy
pollution.
logic controller system, for ‘many’ condition
at constant speed of 60km/h, the green light
ACKNOWLEDGEMENTS
extension let 41.1 seconds only. It can reduces
The authors wish to express their thanks to the
31.5% of waiting time. While, for ‘many’
Universiti Teknikal Malaysia Melaka (UTeM)
condition with speed of 20km/h, the extension
under UTeM PJP Grant (PJP/2010/FKE (11)
of green light give 53.6 seconds which reduce
S33)
10.7% of waiting times. For ‘medium’
condition, with consideration up to 60 vehicles
REFERENCES
travel through the intersection, 32.1 seconds
[1]. http://maps.google.com.my/maps
and 36.1 seconds extension of green light. It
[2]. Marco Wiering (2003), Intelligent Traffic
reduces 46.5% and 39.8% of waiting time for
Light Control, ERCIM News No. 53,
constant and various speeds condition
April
respectively.
[3]. M. Wiering, J. Vreeken, J. V. Veenen and
A. Koopman, (2004) Simulation and
As for current case study, only 39 seconds is
Optimization of Traffic in a City. IEEE
needed for 60 vehicles to travel through the
Intelligent Vehicles Symposium,
intersection. The extension of green light is
University of Parma, pp. 453-458.
depending on the changes of various vehicle
[4]. K. K. Tan, M. Khalid and R. Yusof,
speeds and the number of vehicles coming on
(1996) Intelligent Traffic Light Control by
the arrival side.
Fuzzy Logic. Malaysian Journal of
Computer Science, Vol. 9-2, pp. 29-35.
V. CONCLUSIONS
[5]. L. F. Pedraza, C. A. Hernandez and O.
Fuzzy logic to control the traffic lights is
Salcedo,( 2008) Intelligent Model Traffic
presented to reduce the long term average
Light for the City of Bogota. Electronics,
waiting times.
Robotics and Automotive Mechanics
The fuzzy logic traffic lights controller
Conference, pp. 354-359.
performed better than the Multiplan system
[6]. M. Masoud. (2006), Multi-agents System
due to its flexibility. The flexibility involves
for Intelligent Control of Traffic Signals.
the number of vehicles sensed at the incoming
Int Conference on Computational
junction and the extension of the green light.
Intelligence of Modeling Control and
The traffic flowing through a set of three
Automation.
intersections is optimized by adjusting the
[7]. E.H. Hamdani (1974), Application of
signal timing parameters, extension of green
fuzzy algorithm for control of simple
dynamic plant, Proc. IEEE 121 12

238
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Goodness of Fit Test for Gumbel Distribution Based on Kullback-

Leibler Information using Several Different Estimators
S. A. Al-Subha , K. Ibrahimb , M. T. Alodatc , A. A. Jemaind,
a,b,d
School of Mathematical Sciences, Universiti Kebangsaan Malaysia, Selangor, Malaysia
c
Department of Statistics, Yarmouk University, Irbid, Jordan
Email: salsubh@yahoo.coma , Kamarulz@ukm.myb , alodatmts@yahoo.comc , azizj@ukm.myd

ABSTRACT underlying sample data is of the normal or

In this paper, our objective is to test the statistical exponential type.
hypothesis H o : F (x )  F o (x ) for all x Many studies have been carried out to study
goodness of fit tests by using of Kullback-
against H 1 : F (x )  Fo (x ) for some x , Leibler information. Kinnison (7) tested the
where F o (x ) is a known distribution function. Gumbel distribution using a correlation
In this study, a goodness of fit test statistics for coefficient type statistic. Arizono and Ohata
Gumbel distribution based on Kullback-Leibler (2) proposed a test of normality based on an
information is studied. The Gumbel parameters are estimate of the Kullback-Leibler information.
estimated by using several methods of estimation Song (11) presented a general methodology
such as maximum likelihood, order statistics, for developing asymptotically distribution-free
moments, L-moments and LQ-moments. The goodness of fit tests based on the Kullback-
critical value based on the statistics which involves Leibler information. Also, he has shown that
the Kullback-Leibler information under the
the tests are omnibus within an extremely
assumption that H o is true is computed using large class of nonparametric global
Monte Carlo simulations. alternatives and to have good local power.
The performance of the test under simple Ibrahim et al. (1) found that the goodness of fit
random sampling is investigated. Ten different test based on Kullback-Leibler information
distributions are considered under the alternative supports the results which indicate that the chi-
hypothesis. Based on Monte Carlo simulations, for square test is most powerful under RSS than
all the distributions considered, it is found that the
SRS for some selected order statistics.
test statistics based on estimators found by moment
and L-moment methods have the highest power,
except for Laplace, Student-T (4) and Cauchy In this paper, we introduce a goodness of
distribution. fit test for Gumbel distribution which is based
on the Kullback-Leibler information. We
KEYWORDS estimate the Gumbel parameters by using
Goodness of fit test; Kullback-Leibler Information; several methods of estimation such as
Entropy;Gumbel distribution; maximum likelihood; maximum likelihood, order statistics, moments
order statistics; moments; L-moments. and L-moments. According to Hosking (5), L-
moments have the theoretical advantages over
I. INTRODUCTION the conventional moments of being able to
There are many areas of application of characterize a wider range of distributions and,
Gumbel distribution such as environmental when estimated from a sample, the method is
sciences, system reliability and hydrology. In more robust to the presence of outliers in the
hydrology, for example, the Gumbel data. Also, the parameter estimates obtained
distribution may be used to represent the from L-moments are sometimes more accurate
distribution of the minimum level of a river in in small samples than the maximum likelihood
a particular year based on minimum values for estimates. We compute the percentage values
the past few years. It is useful for predicting and the power based on the statistics which
the occurrence of that an extreme earthquake, involves the Kullback-Leibler information
flood or other natural disaster. The potential using Monte Carlo simulations.
applicability of the Gumbel distribution to
represent the distribution of minima relates to This paper is organized as follows. In
extreme value theory which indicates that it is Section 2, we define the test statistic. We
likely to be useful if the distribution of the define the estimators of Gumbel distribution in

239
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Section 3. In Section 4, we define two 

algorithms to calculate the percentage points I (f , f 0 )  f ( x )log f ( x )dx

and the power function of the test statistic 
under an alternative distribution. In Section 

5, a simulation study is conducted to study the  f ( x )log f 0 ( x ;  ,  )dx

power of the test statistic and we state our 
conclusions in Section 6.
1 n
 n 
II. TEST STATISTICS 
n
 log  2 m  x
i 1
i  m :n  x i  m :n  

Let X 1 , X 2 ,..., X n be a random sample
1 n

from the distribution function F ( x ) with 

n
 logf
i 1
0 ( x ; ˆ , ˆ ) ,
quantile
(4)
function Q (u )  F 1 (u ) and let where m , called window size, is a positive
X 1:n  X 2:n  ...  X n :n denote the integer (m  n / 2), xi:n  x1:n for i  1 and
corresponding order statistics. We are x i :n  x n :n for i  n . For the Gumbel
interested in testing the hypothesis
H o : F (x )  Fo (x ) for all x distribution, I (f , f 0 ) can be estimated,
vs. H 1 : F (x )  Fo (x ) denoted as I mn , and be given by
 n
 X i  m :n  X i m :n  
1 n
for some x , where Fo (x ) is a Gumbel I mn   
n i 1
log 
 2m 
distribution function of the following form
  x   1 n  x  ˆ 
Fo ( x ;  ,  )  exp   exp  
 ,  log ˆ +  exp   i .
ˆ 
   n i 1 
(1) (5)
Many estimators for  and  are considered.
and its density function is The estimators derived are maximum
1  x   x   likelihood, order statistics, moments and L-
f o ( x;  ,  )  exp    exp   ,
       moments. The purpose is to find the estimator
which is able to give better power.
(2)
where  is a location parameter,  is a scale III. ESTIMATORS OF  , 
parameter, x,   (, ), and   0. We will introduce four different types of
estimators for  and  , which are
We employ the Kullback-Leibler
mle, me, os and lm.
information which is given by

f (x )
I (f , f 0 )   f (x )log f 0 (x ;  ,  )
dx . i)Maximum Likelihood Estimator ( mle ) :
 We denote the mle of  ,  as
(3)
The quantity I (f , f 0 ) describes the amount
ˆ , ˆ , respectively. Let X 1 , X 2 ,..., X n be a
random sample from (2). The log-likelihood
of ‘Information’ lost for approximating f (x )
function is given by
using f 0 ( x ). The larger value of I (f , f 0 )
indicates the greater disparity between f (x ) n
 x  
l ( ,  )  n log       i
and f 0 ( x ). It known that I (f , f 0 )  0 if i 1   
and only if f (x )  f 0 (x ) for all x  0. n
 x  
  exp   i .
Hence the test can be designed as follows. i 1   
Reject H o vs H 1 if I (f , f 0 ) is large. (6)
Vasicek (12) and Song (11) find that

240
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

After taking the derivatives with respect to iv)L-moment Estimator (lm ) :

 and  and equating to 0, we obtain the
equations given as L-moment r as explained in the Hosking (5),
n
is
ˆ  x   x iw i and ˆ   ˆ log  z  . 1 r 1  r  1
r    1 r  j : r , r  1, 2,...
j
i 1 
(7)
r j 0  j 
where (13)
where
 x  1 n z


z i  exp   i  , z  z i and w i  i .
 ˆ  n i 1 nz r :n   xf

r :n (x )dx .

(14)
Using r :n can be shown that
ii)Method of Moment Estimator  me  :
The mean and variance for Gumbel 2:2    1.2704 ,
distribution are given by 1:2    0.1159 ,
2 and 1:1 = .
     and  2   2. (8)
6 The L-moment estimators of  ,  are given
by
The moment estimators of the two parameters
1 ˆ  ˆ
are ˆ   ˆ 2:2  ˆ1:2   2:2 1:2 ,
6 2 log 2 1.3863
ˆ  s
 (15)
and respectively. Given that
ˆ  x  ˆ (9) n 1

n  i  x
2
where s, x are the sample standard deviation ˆ1:2  ,
n  n  1
i :n
i 1
and mean respectively, and n 1
  0.57721566 is Euler's constant.
i x
2
ˆ 2:2  ,
n  n  1
i 1:n
i 1
iii)Order Statistics Estimator (os ) :
ˆ1:1 =x,
th (16)
The p quantile for the Gumbel distribution is
Q ( p ;  ,  )  F 1 ( p )     log   log p  , IV. ALGORORITHM FOR POWER
COMPARISON
0  p 1 Consider   0.05 , random samples of size
(10) n  12, 18, 24, 36, and window of size
By taking p  0.25 and 0.75, the estimators
m  1, 2,3, 4. Let X 1 , X 2 ,..., X n
be a
of  ,  are
random sample from Gumbel with
1
ˆ 
1.5725
 0.3266F 1 (0.75)  1.2459F 1 (0.25)    0,   1, Fo (x ;0,1). We estimate the
parameters  and  from the sample by
1 1 1 1
 F (0.75)  F (0.25) maximum likelihood (7), Method of Moment
5 5 Estimator (9), Order Statistics Estimator (11)
(11) and (12), and L-moment Estimator (15) and
and (16). Calculate the test statistics T  I mn
1
ˆ 
1.5725
 F 1 (0.75)  F 1 (0.25)  . using equation (2.5). We repeat the previous
steps 40, 000 times to get T 1 ,..., T 40,000 .
(12)
Determine the percentage point d  of T

241
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

which is given by the (1   )100

th
quantile of
V. SIMULATION RESULTS
the distribution of T 1 ,..., T 40,000 . The power of each test is approximated based
To calculate the power of T at different on a Monte Carlo simulation of 40,000
alternative distributions: iterations according to the algorithm of
Let X 1 , X 2 ,..., X n be a random sample from Section 4. We compare the efficiency of the
tests for different sample sizes:
H , a distribution under H 1. We estimate the n  12, 18, 24, 36, different window sizes:
parameters  and  from the sample by m  1, 2,3, 4 and different alternative
maximum likelihood method (7), Method of distributions: Normal (0, 1),
Moment Estimator (9), Order Statistics
Estimator (11) and L-moment Estimators (15) Logistic (0, .7), Laplace (0, 1),
and (16). Calculate the test statistic T  I mn StudentT (12), StudentT (4),
in formulas (5). We repeat the previous steps Cauchy (0, 1), Exponential (1),
40, 000 times to get T 1 ,..., T 40,000 . To W eibull ((1.5), 2), W eibull (2, .5), and
calculate the Power, determine Lognormal (0.2, .4). The
1 40,000
Simulation results are presented in the Tables
T (H ) 
40, 000

t 1
I (T t  d  ) , where (1) and (2).
I (.) stands for indicator function.

Table 1. Percentage points for the test statistics I mn for different sample sizes n  12, 18, 24, 36, window
sizes m  1, 2,3, 4 and   0.05.

n m mle me os lm n mle me os lm

12 1 .831 .820 1.091 .810 24 .571 .672 .706 .580

2 .614 .629 .889 .615 .389 .421 .531 .400
3 .561 .613 .882 .447 .352 .385 .496 .368
4 .566 .623 .881 .596 .346 .386 .585 .364
18 1 .655 .679 .850 .659 36 .487 .512 .567 .497
2 .460 .495 .692 .475 .318 .339 .403 .327
3 .422 .467 .660 .449 .269 .295 .362 .283
4 .430 .472 .669 .451 .259 .284 .347 .270

Table 2. Power estimates of theI mn statistics under different alternative distributions with different sample
sizes n  12, 18, 24, 36, window sizes m  1, 2,3, 4 and   0.05.

Normal (0, 1) Logistic (0, .7) Laplace (0, 1)

n m mle me os lm mle me os lm mle me os lm

12 1 .098 .113 .114 .114 .083 .057 .174 .146 .122 .218 .282 .197
2 .120 .163 .125 .142 .125 .196 .179 .174 .156 .260 .304 .244
3 .135 .183 .132 .164 .149 .214 .184 .205 .175 .277 .561 .381
4 .164 .218 .133 .201 .190 .245 .196 .230 .212 .304 .313 .283
18 1 .120 .188 .177 .161 .126 .058 .273 .218 .196 .333 .449 .330
2 .169 .237 .179 .208 .180 .293 .279 .271 .250 .389 .472 .365
3 .173 .268 .188 .229 .194 .314 .296 .284 .251 .397 .474 .385
4 .183 .289 .198 .255 .225 .333 .296 .302 .247 .400 .477 .378
24 1 .134 .255 .183 .211 .157 .063 .309 .291 .284 .436 .531 .432
2 .189 .317 .202 .265 .231 .515 .331 .345 .357 .484 .557 .485
3 .232 .354 .211 .308 .248 .407 .343 .371 .361 .500 .567 .490

242
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

4 .237 .364 .224 .325 .260 .416 .350 .388 .346 .492 .565 .490
36 1 .182 .355 .225 .296 .247 .070 .427 .406 .431 .595 .702 .586
2 .254 .446 .288 .391 .341 .533 .453 .482 .525 .657 .723 .651
3 .319 .509 .301 .425 .376 .547 .467 .520 .565 .665 .729 .673
4 .348 .512 .316 .459 .408 .563 .485 .535 .548 .669 .732 .663

Table 2. (Cont.)

StudentT (12) StudentT (4) Cauchy (0, 1)

n m mle me os lm mle me os lm mle me os lm

12 1 .095 .139 .147 .125 .102 .197 .241 .184 .467 .580 .671 .563
2 .123 .181 .157 .161 .128 .241 .249 .226 .498 .584 .687 .575
3 .145 .207 .154 .189 .161 .263 .262 .248 .468 .579 .684 .559
4 .181 .135 .171 .218 .205 .281 .252 .272 .452 .549 .683 .537
18 1 .109 .220 .232 .192 .150 .310 .363 .287 .679 .755 .842 .742
2 .160 .269 .230 .248 .208 .351 .376 .337 .711 .776 .854 .785
3 .197 .295 .245 .265 .251 .367 .381 .343 .690 .751 .850 .733
4 .210 .320 .251 .285 .256 .368 .378 .352 .616 .709 .849 .686
24 1 .146 .291 .260 .260 .201 .389 .429 .368 .808 .854 .924 .849
2 .213 .357 .274 .314 .281 .435 .452 .419 .849 .875 .931 .870
3 .238 .386 .287 .336 .325 .463 .456 .434 .836 .863 .923 .856
4 .254 .401 .300 .385 .325 .464 .463 .436 .793 .841 .925 .686
36 1 .220 .409 .356 .357 .310 .523 .574 .519 .929 .949 .982 .949
2 .311 .494 .392 .434 .398 .591 .601 .576 .955 .965 .984 .961
3 .355 .535 .406 .483 .470 .605 .605 .589 .959 .962 .985 .960
4 .372 .552 .404 .492 .482 .602 .603 .589 .794 .955 .984 .953

Exponential (1) W eibull ((1.5), 2) W eibull (2, .5)

n m mle me os lm mle me os lm mle me os lm

12 1 .133 .177 .073 .171 .223 .278 .120 .263 .120 .071 .042 .058
2 .168 .219 .065 .218 .318 .323 .124 .332 .062 .061 .044 .062
3 .198 .190 .061 .200 .292 .301 .155 .308 .065 .058 .041 .062
4 .141 .137 .039 .151 .186 .223 .064 .243 .065 .061 .042 .066
18 1 .246 .270 .098 .264 .382 .424 .166 .410 .167 .071 .045 .063
2 .322 .327 .103 .359 .500 .518 .189 .530 .069 .063 .047 .066
3 .354 .353 .097 .371 .519 .528 .390 .530 .071 .068 .044 .071
4 .310 .309 .082 .327 .445 .482 .136 .494 .077 .070 .050 .069
24 1 .324 .365 .198 .363 .491 .557 .333 .553 .198 .065 .049 .066
2 .471 .463 .217 .483 .649 .675 .412 .686 .077 .073 .052 .073
3 .485 .481 .218 .503 .675 .697 .475 .719 .087 .075 .053 .080
4 .470 .475 .216 .488 .638 .674 .394 .690 .087 .078 .047 .085
36 1 .468 .538 .364 .520 .700 .756 .584 .733 .312 .070 .051 .068
2 .659 .673 .452 .692 .845 .871 .672 .866 .089 .081 .053 .085
3 .710 .721 .487 .737 .877 .897 .717 .905 .102 .089 .064 .091
4 .698 .736 .489 .735 .885 .901 .701 .903 .104 .103 .059 .101

243
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Table 2. (Cont.)

Lognormal (0.2, .4)

n m mle me os lm

12 1 .085 .091 .051 .089

2 .101 .100 .056 .089
3 .096 .085 .047 .091
4 .084 .067 .039 .067
18 1 .099 .125 .045 .104
2 .128 .146 .045 .137
3 .127 .189 .041 .135
4 .106 .118 .038 .117
1 .108 .157 .070 .125
2 .164 .187 .070 .173
3 .170 .189 .066 .181
4 .162 .179 .068 .162
1 .144 .210 .098 .167
2 .210 .265 .100 .227
3 .226 .269 .106 .258
4 .215 .278 .093 .256

From the above tables, we make the W eibull (2, .5) , highest power for
following remarks: Cauchy.
1. The percentage points for the test
decrease as the sample size n VI. CONCLUSION
increases.
An accurate estimation of parameters of the
2. The power increases as the sample Gumbel distribution in statistical analysis is
size n increases. importance. In this paper, we have introduced
a test statistic of goodness of fit for Gumbel
3. The power increases as the window distribution based on Kullback-Leibler
size m increases. information measure. We considered ten
different distributions under the alternative
4. In the case of moment estimator, the
hypothesis. It is found that the test statistics
test has the highest power for Normal,
based on estimators found by moment and
Logistic, Student T (12) , and order statistic methods have the highest power,
Exponential distributions. except for weibull and Lognormal
distributions. In the case of Cauchy, the test is
5. In the case of L-Moment estimator, found to have the highest power for all
the test has the highest power for estimators but the test has the lowest power for
W eibull ((1.5), 2) and lognormal W eibull (2, .5) . The theory developed could
distributions. be extended easily to other distributions. Also,
6. In the case of Maximum Likelihood we can apply RSS, extreme RSS and median
estimator, the test has the highest RSS on this test statistics.
power for W eibull (2, .5)
ACKNOWLEDGEMENTS
distribution. The authors would like to thank Universiti
Kebangsaan Malaysia for supporting this work
7. When the estimates are considered, the
under the research grant UKM-ST-06-
test has the lowest power for
FRGS0096-2009.

244
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

REFERENCES [7]. Kinnison R. (1989). “Correlation

[1]. Ibrahim, K. M.T. Alodat, A.A. Jemain & coefficient goodness of fit test for the extreme
S.A Al-Subh. (2009). “Chi-square Test for value distribution,” Amer. Statistician. 43, pp
Goodness of Fit Using Ranked Set Sampling,”
98-100.
Tenth Islamic Countries Conference on
Statistical Sciences. December 20-23, 2009, [8]. Kullback, S. (1959). “Information Theory
Cairo, Egypt.
and Statistics,” New York, Wiley.
[2]. Arizono I. and Ohta H. (1989). “A test for
[9]. Kullback, S., Libler R. A. (1951).” On
Normailty based on Kullback-Leibler
information and sufficiency,” Ann. Mathemat.
Information,” Amer. Statistician. 43, pp 20-22.
Statist. 4, pp 49-70.
[3]. D'Agostine, R., Stephens. M. (1986).
[10]. Paulino P. R., Humberto V. H. and Jose
“Goodness of fit Techniques,” Marcel Dekker
V. A. (2009). “A goodness-of-fit test for the
Inc., New York
gumbel distribution based on Kullback-Leibler
[4]. David, H.A. and H.N. Nagaraja (2003). Information,” Comm. In statistics- Theory and
”Order Statistics,” 3rd Edn., Wiley, New Methods, 38, pp 842-855.
Jersey, ISBN: 0-471-38926-9.
[11]. Song K. S. (2002). Goodness-of-fit tests
[5]. Hosking, J. R. M. (1990). “L-moments: based on Kullback-Leibler discrimination
Analysis and estimation of distribution using information. IEEE Trans. Inform. Theor. 48,
linear combinations of order statistics”, J. Roy. pp 1103-1117.
Statist. Se. B. 52, pp 105-124.
[12]. Vasicek O. (1976). “A test for normality
[6]. Gumbel, E.J. (1958). Statistics of based on sample entropy,” J. Roy. Statist. Soc.
Extremes. Columbia University Press, New 38, pp 54-59.
York, ISBN: 0486436047, pp 377.

245
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Impact of shrimp pond development on biomass of intertidal macrobenthic

fauna: a case study at Sembilang, South Sumatra, Indonesia
Agus Purwoko1, 2, Arwinsyah Arka2 and Wim J. Wolff1
1
Dept. of Benthic Marine Ecology and Evolution, University of Groningen, P.O. Box 14, 9750 AA Haren, The
Netherlands
E-mail addresses: w.j.wolff@rug.nl
2
Dept. of Biology, Sriwijaya University, Palembang, Indonesia
E-mail addresses: agus_purwoko2002@yahoo.com

ABSTRACT Annual mangrove litter production at some

The macrobenthic animal biomass of the intertidal nearby Indo-Pacific locations averaged 1107 g
area of the Sembilang peninsula of South Sumatra, dry wt m-2 at Hong Kong (Lee 1989, 1989 a),
Indonesia, has been studied in 1996, 2004 to 2006. 497 g dry wt m-2 at Sri Lanka (Amarasinghe &
Monthly (August – October 1996) 25 core samples Balasubramaniam 1992), and Bunt (1995)
were taken at each of plot sampling stations, (March
found annually on average 1752 g dry wt m-2 at
– August 2004) 21 core samples were taken at each
of six sampling stations, (March – August 2005) 30 Papua New Guinea. Annual litter production at
core samples were taken at each of six sampling Sembilang, Indonesia, was 1376 g dry wt m-2
stations, (June and October 2006) 30 core samples (Soeroyo 1999). It has been demonstrated that
were taken at each of six sampling stations. benthic animals, mostly crustaceans, feed on
Macrobenthic fauna was identified at the lowest this resource (Odum and Heald, 1972, 1975). In
taxonomical level, and counted. Biomass was mangrove forest crabs (Lee 1989), gastropods
measured as ash free dry weight (AFDW). and polychaetes (Alongi & Sesakumar 1992)
Development of shrimp ponds matched the feed on mangrove leaves. Lee (1989) found that
declining biomass of macrozoobenthic significantly crabs consumed 50 % of mangrove leaves.
(Anova: p < 0.05) for long term (1996 to 2006).
Three year of observation (2004-2006) show that
sites close to the shrimp ponds (pond sites) Part of the net canopy production of mangrove
significantly had lower biomass of macrobenthic forest is exported. Odum and Heald (1972,
animals compare to biomass value at sites further 1975) suggested that this may amount to about
from the ponds (non-pond sites). Mangrove organic 50% of the production and Alongi and
matter probably was not the main source of food for Sesakumar (1992), and Alongi at al. (2004)
macrobenthic fauna and pond effluent might not found an export of 40% and 25% respectively.
responsible for declining macrozoobenthic biomass. This exported material may enhance the benthic
It seemed that disturbance of habitat of fauna of adjoining aquatic habitats such as tidal
macrozoobenthos by fisheries coincided with flats.
decreasing biomass.

Key words: impact, shrimp pond, biomass, However, in a review Lee (1995) questioned
macrobenthos, intertidal fauna, Indonesia Sumatra, that benthic biomass of the macrofauna had a
Sembilang, positive relationship with the availability of
detritus originating from mangrove leaves.
INTRODUCTION Indeed, Lee (1999) experimentally found no
Around 1990 Indonesian mangroves occupied relationship between mangrove leaves and
4.25 million ha which represented about 20 % biomass of macrobenthic animals. Also,
of the world’s mangroves. The largest part was Bouillon et al. (2002) and Hsieh et al. (2002)
found at Irian Jaya (2.94 million ha) and 1.31 demonstrated with stable isotopes that many
million ha occurred at more populated areas benthic animals in mangrove forests derived
such as Java, Sumatra, and Kalimantan. South their food from other sources than mangrove
Sumatra counted 195000 ha (Choong et al. leaves and litter.
1990) with 77500 ha (Danielsen & Verheugt
1990) in our study area Sembilang National Conversion mangrove forest to ponds involved
Park. cutting mangrove forest, dredging and
dumping, and discharging water to estuaries.

246
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Those activities may have impacts to benthic impact of pond construction and operation on
biomasses at tidal flat. Dredging and dumping the macrobenthic fauna of the tidal flats on that
are important activities which lead to turbidity site. We did so by comparing the biomass in the
and enhanced sediment deposition, and growth pre-pond construction (1996) and after-pond
of bivalves may be impaired by elevated construction period (2004-2006), and by
concentrations of suspended matter (Karel comparing locations close to the shrimp ponds
1999). A three-year study on the effects of with sites at further distance after pond
reclamation activities on the macrobenthic construction.
fauna at a Singapore coastal area found that
macrobenthos abundance significantly METHODS
decreased over time close to the reclaimed area Area description
but it increased again with distance from this
area (Lu et al. 2002).. Kenny and Rees (1994) The Sembilang peninsula (Fig. 1: 10 59’ to 20
reported that a dredged site had not fully 15’ S and 1040 45’ to 1040 53’ E) is located at
recovered after 7 months but Lu and Wu (2000) the Eastern coast of South Sumatra Province,
found that the benthic community fully Indonesia, and it is part of the Sembilang
recovered in less than 15 months. National Park. It is influenced by the Musi
River, the Musibanyuasin River and some
Pond effluents lead to reduced water quality by smaller tributaries. Originally, all land was
increasing suspended particles (Jackson et al. covered by mangrove and swampy forest. The
2003), total nitrogen and phosphate (Jackson et most seaward belt of the mangrove vegetation
al. 2004; Lemonnier & Faninoz 2006), salinity, is still formed by Sonneratia and Avicennia
acidity (Cowan et al. 1999) and biochemical where the soil is sandy; however, where the soil
oxygen demand (Trott & Alongi 2000; Cowan is muddy, Rhizophora occurs. Nypa palms
et al. 1999), dissolved organic matter appear where the soil is highly influenced by
(Lemonnier & Faninoz 2006), eutrophication fresh water. In the mangrove area at the
(McKinnon et al. 2002; Alongi et al. 2000) and Sembilang peninsula, approximately 200 to 500
adding antibiotics (Le & Munekage 2004; Le et meter from the beach, there are 4,000 ha of
al. 2005). Phytoplankton may benefit of shrimp ponds. The coastline of the area faces
increasing nutrients. However, it will improve directly Bangka strait and the East China Sea.
shoot biomass of mangrove seedling Two villages are nearby. Sungsang village is at
(Rajendran & Kathiresan 1996). the southern part and Sembilang village at the
northern end of the peninsula.
Two families from Lampung (Southern South
Sumatra) Province established 4 ha of shrimp Description of sampling stations
ponds in the Sembilang National Park in 1996.
Local government caught the investors, but For our 2004-2006 study we selected 6
from 1997 to 2002 the government did not pay sampling stations (Fig. 1) with varied
attention to Sembilang developments. About characteristics:
2,000 families established 4,000 ha (traditional)
shrimp ponds from 1998 to 2002. They left 1. Station 1: Tj. (Tanjung) Carat (S: 20 16.324'
some mangrove area as a green belt, and their & E: 1040 55.117' by Gekko 202 Garmin GPS).
ponds are connected to some small rivers (Fig. Located in the estuary of the Musi river; the
1). soil is sandy, and the station is strongly
influenced by fresh water.
Purwoko (1996) described abundance and
biomass of macrobenthic fauna at Sembilang 2. Station 2: S. (Sungai) Bungin (S: 20 14.955'
before this large-scale pond construction; his & E: 1040 50.714'). Located in the estuary of
site of observation was about 10 km from the the Musibanyuassin river. The surface layer of
first established pond. During and after the the sediment is soft and high in organic matter
pond development at Sembilang peninsula, no and has a thickness of more than 1 meter. On
study on the impacts of shrimp pond the sediment young Avicennia occur.
development on the coastal ecosystems was
conducted at Sembilang. Our study objective 3. Station 3: Solok Buntu (S: 20 11.063' & E:
was to determine the positive or negative 1040 54.764'). Located near the estuary of the

247
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Musibanyuasin river. The surface layer of the & Duncan 1971) was measured at the
soil is muddy, and reaches 40 cm depth. laboratory at Palembang.
Mangrove vegetation consists mainly of
Avicennia. At sea nearby, there are some pole In 2004-2006 each station consisted of 3
houses. parallel line transects each consisting of 7
sampling points in 2004 and 10 sampling points
4. Station 4: S. Barong Kecil (S: 20 9.872' & E: for 2005 and 2006. The distance between the
1040 54.587'). The site is near the minor estuary line transects varied from 5 to 10 m; the
of S. Barong Kecil. The soil is muddy and the distance between the sampling points was
depth of this layer reaches 50 cm. randomly chosen and varied between 3 and 20
m. The first core sample of the first transect
5. Station 5: S. Siput (S: 20 5.824' & E: 1040 was taken at the lowest water level and the
54.102'). The soil is muddy and the soft layer second till the seventh one were at further
reaches 40 cm; the adjacent mangrove distance towards the mangrove. Circular core
vegetation is mostly Avicennia; close to the site diameter and sampling depth were the same as
young trees grow. At this station we made also for the 1996 core samples. At each sampling
our 1996 observations. point 1 core sample was taken. The core
samples directly were extracted by double
6. Station 6: S. Dinding (S: 20 1.924' & E: 1040 layers of 1 mm sieve in the field and the
50.573'). The sediment is muddy and contains animals were collected from the sieve by hand.
many dead shells. The depth of the soft layer Macrobenthic fauna collected was preserved in
reaches 40 cm. The site is facing directly the 70 % alcohol mixed with 3 % of formalin.
South China Sea, so it is exposed to high Sampling activities took place during low tide
waves. at Sungsang in March, May, June, July and
August 2004, March, April, May, June, July
Stations 1, 2 and 6 were further removed from and August 2005, and June and October 2006.
the shrimp ponds; and stations 3, 4, and 5 The animals in the samples were identified to
relatively close to ponds or at least close to the lowest taxonomic level possible (Dharma
estuaries of small rivers that served as an outlet 1988, 1992); Computer Program: Poly Key),
of pond discharges (Fig. 1). So, our after- counted and Ash-Free Dry Weight (AFDW)
construction comparison has been made by measured like 1996’s samples.
comparing stations 1, 2 and 6 on the one hand
with stations 3, 4 and 5 on the other hand. In order to compare years, data sets were taken
from the same month of each year. In this case,
Sampling procedure data sets for June in 2004, 2005 and 2006 were
comparable. The statistical calculations such as
In 1996, two plots (100 * 100 meters) were Analysis of Variance (ANOVA), Duncan test
established at station 5. Plot x was situated at and T-test were carried out by the Statistics 7
the upper half of the shore and plot y was set at Program.
the lower shore. Sampling started in August and
took place three times; the periods between two RESULTS
samplings were four weeks. There were 25 core Comparison of pre-construction and post-
samples collected at random guided by random construction period
numbers at each plot. Samples were taken with a
circular corer. Core diameter was 15 cm and Table 1 from Purwoko (1996) gives the
the sampling depth 30 cm. The core samples biomasses found at the two plots at station 5 in
directly were extracted by 5 mm sieve in the 1996, that is in the pre-construction period of
field and the animals were collected from the ponds at Sembilang peninsula. Higher biomass
sieve by hand. Macrobenthic fauna collected of macrobenthic fauna found at plots x and y in
was preserved in 70 % alcohol mixed with 3 % October 1996 differs significantly from the
of formalin. Sampling activities took place biomasses in August and September 1996.
during low tide at Sungsang. The fauna in the However, biomass of macrobenthic animals at
samples was identified to the lowest taxonomic plot x was not significantly different from those
level possible (Dharma 1988, 1992), counted at plot y from August to October 1996 (no
and Ash-Free Dry Weight (AFDW) (Winberg letters shown). In 1996, 17 taxa were identified

248
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

and the average density at the plots was 690 species of macrobenthic fauna which had a
individuals m-2. significant difference had lower biomass at
pond sites (Table 5).
Table 2 present the biomass data per station
averaged over the period 2004-2006. Generally, DISCUSSION
total biomasses of macrobenthic fauna at Pre-construction compared to post-
stations 1, 3 and 5 were significantly lower construction period of ponds at Sembilang
(ANOVA: p<0.05) than those at stations 2, 4 peninsula
and 6. The same differences occurred for total
biomass of bivalves. The highest biomass of In August and October 1996, biomass of
polychaetes was found at station 1, differing macrobenthic animals at station 5 was
significantly from the other stations. However, significantly higher than in August and October
each family of polychaetes had a different 2004-2006. This biomass of macrobenthic
pattern of distribution of biomass compared to fauna gained mainly from Anadara biomass
the total biomass of all polychaetes. Biomass of followed by Tellina remies. Biomasses at plot x
gastropods at station 1, 3, 4 and 5 was and plot y were not significantly different in
significantly lower than biomass values at 1996, meaning that the macrobenthic fauna
stations 2 and 6. biomass was not influenced by the time of
inundation, and beach steepness (Jiang & Li
1995; Ricciardi & Bourget 1999)
Comparison of sites in post-construction period
Due to the larger sieving mesh in 1996 (5 mm)
Table 2 gives the biomass data for the period compared to the sieving mesh applied in 2004
after pond construction. It is, however, based to 2006 (1 mm), the biomass values in 1996 are
on averages for different months and different supposed to have been even higher than the
numbers of months. To better compare the biomasses actually recorded. Small polychaetes
different years Table 3 compares the biomass apparently passed through the sieve since we
values obtained at each station in June of each did not find any Maldanidae, Capitellidae and
year. We found in general low biomass values, Sternaspidae in 1996. These animals
but with significant differences between years. approximately contributed average 2 percent to
Also the stations differ significantly from each the total biomass value in 2004-2006 (Table
other (ANOVA: p<0.05; not shown in Table) 4).
per year.
Table 4 gives biomasses per species averaged In 1996 macrobenthic animals were larger in
over all stations in June of each year. At the numbers of taxa and density than in 2004, 2005
Sembilang peninsula, total biomass of and 2006. Species found in 1996 and absent in
macrobenthic fauna was significantly lower in 2004, 2005 and 2006 were Cerithium sp.,
2005 compared to 2004 and 2006. There were Clypeomorus sp., Vexilles sp., and Natica sp.
also significant differences among the stations
from 2004 to 2006 (Table 3). For example, at We conclude that macrobenthic animal biomass
station 5, biomass of macrobenthic animals in at Sembilang decreased by nearly an order of
2006 was significantly higher than in 2004 and magnitude between 1996 before the
2005. construction of shrimp ponds and 2004-2006
Bivalves, gastropods, crabs and some well after the pond construction.
polychaetes species were significantly different
between 2004 and 2005. Biomasses of bivalves Comparison of stations after-pond construction
and gastropods reached significantly higher period
values in 2004, but crabs (Ocypodidae) in
2006. Most polychaetes (Nereididae, A three-year observation (2004 to 2006)
Lumbrineridae, Sternaspidae and unidentified showed a significantly lower value of
worms) showed higher values of biomass in macrobenthic animal biomass at the stations 1,
2006 (Table 4). 3 and 5 compared to the stations 2, 4 and 6
Generally, in 2004 to 2006 the pond sites had (Table 2). This detailed evaluation does not
significantly lower biomass of macrobenthic support our hypothesis that the established
animals compared to non-pond sites. All ponds caused the decrease of the biomass of

249
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

macrobenthic fauna. However, based on a Hence, it is unlikely that at Sembilang waste

comparison of non-pond and pond sites, we water from shrimp ponds has been responsible
could show that biomass of macrobenthic fauna for a difference between pond sites and non-
at pond sites on average was significantly lower pond area.
compared to the biomass value at non-pond
sites (Table 5). Also all macrobenthic species Hence, we consider the possibility that fishing
which showed significant differences had lower may have influenced the biomass of the
biomass at the pond sites (Table 5). This intertidal macrobenthic fauna. Fishing at
supports our hypothesis that biomass of Sembilang aims at shrimp, fish and bivalves.
macrobenthic animals declined significantly The number of active fishing vessels amounts
due to pond establishment. to about 1970 of which daily 40-60 are active
in our tidal flat area (Djamali and Sutomo,
Mangrove litter fall were believed to be a 1999). Before 1999, fishermen used gill nets to
source of organic matter for macrobenthic catch shrimps for export abroad. They hardly
animals both in the mangrove area tidal flat harvested shrimps near the intertidal areas
nearby (Odum & Heald 1975; Lui et al. 2002; (Djamali & Sutomo 1999). After 1999, almost
Bosire et al. 2004). In a review, however, Lee all fishing boats have been equipped with
(1995) questioned that benthic biomass had a modified mini trawls to catch shrimps
positive relationship with the availability of (Purwoko, unpubl. obs.). These boats operate at
detritus originating from mangrove leaves. This the intertidal areas during high tide. The trawls
is supported by the results of Bouillon et al. move over and through the muddy top layer of
(2002) and Hsieh et al. (2002) who the sediment, supposedly disturbing the
demonstrated by means of stable isotopes that macrozoobenthic habitats. Since shrimp vessels
most benthic organisms do not feed on operate at all stations they may have
mangrove-derived organic matter. This is in contributed to the general decrease of biomass
line with the conclusion that macrozoobenthic over time.
biomass does not depend only on food
availability (mangrove litter fall) but also food Comparing June 2004, 2005 and 2006 (Table
quality. Mangrove litter contains tannin which 3), a significantly lower average biomass was
is not palatable for most macrozoobenthic found at Sembilang peninsula in June 2005.
species (Alongi & Sesakumar 1992) and Lee Further, stations 1, 4 and 5 had lower biomass
(1999) found negative relationship between of macrobenthic fauna, too. This was probably
solube tannin and biomass of macrozoobenthic due to disturbance of the habitat of
in sediment. These conclusions affect the macrobenthic animals by a fishery for bivalves.
credibility of our hypothesis and force us to Starting in March 2005, fishermen harvested
look into other explanations. Anadara at the intertidal flats of Sembilang
peninsula. At station 2, they collected T.
Discharge water contains nutrients, organic timorensis instead of Anadara. In May 2005,
carbon and suspended solids which might have fishermen harvested Anadara at station 4 and in
antibiotics. Options to prevent and reduce those June 2005 they collected Anadara at station 5,
pollutants entering the environment are to after June, they moved to intertidal areas
provide settling ponds (Jackson et al. 2003; further North (station 6). It is assumed that a
Gautier et al. 2001; Lee 1993; Halide et al. month after massive disturbance, biomass value
2003), ponds with vegetation (Sansanayuth et of macrobenthic animal could not have
al. 1996; Halide et al. 2003) and mangrove recovered completely, but it could be recovered
forest as a filter (Nielsen et al. 2003). The latter after 2 months by showing significantly higher
of these options has been put into practice at biomass of macrobenthic animals at station 3
Sembilang. However, whenever pond effluent which was harvested in March and April 2005.
was discharged directly into creeks, by 1 km At least three fishing boats operated by
down-stream and within 1-2 months after shoveling the sediment and collecting Anadara
discharge ceased, water quality was recovered every day at high tide. However, no fishing
(Trott & Alongi 2000). Trott et al. (2004) found boats operated on station 1. Kenny and Rees
that discharge of pond waste carbon (C) and (1994) reported that the benthic community at a
nitrogen (N) during shrimp harvest periods did dredged site had not fully recovered after 7
not cause eutrophication further downstream. months and Lu and Wu (2000) supported by

250
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

demonstrating that the benthic community fully REFERENCES

recovered in less than 15 months. Alongi, D. M., D. J. Johnston and T. T. Xuan
(2000). "Carbon and nitrogen budgets in
Mesh width of the net of fishermen boat was 2 shrimp ponds of extensive mixed shrimp-
cm, and they collected Anadara bigger than 2 mangrove forestry farms in the Mekong delta,
cm (Djamali 1999). In addition, a group of on Vietnam." Aquaculture Research 31(4): 387-
average 20 people manually collected Anadara 399.
at intertidal areas at low tide. In 1999/2000 Alongi, D. M., A. Sasekumar, V. C. Chong, J.
(Djamali & Sutomo 1999) fishermen caught Pfitzner, L. A. Trott, F. Tirendi, P. Dixon and
130 ton wet weight of Anadara which is G. J. Brunskill (2004). "Sediment
equivalent to about 3 ton afdw at intertidal and accumulation and organic material flux in a
subtidal areas; in 2005 we estimate a catch of managed mangrove ecosystem: estimates of
32 ton wet weight (0.8 ton afdw) per month land-ocean-atmosphere exchange in
from the stations 3 to 6 in the period March to peninsular Malaysia." Marine Geology
June (Purwoko, unpubl. obs.). This can be Material Exchange Between the Upper
compared to the average biomass of Anadara at Continental Shelf and Mangrove Fringed
our stations (0.898 g afdw m-2; Table 2) which Coasts with Special Reference to the N.
implies a standing stock for the Sembilang tidal Amazon-Guianas Coast 208(2-4): 383-402.
flats of about 24 tons afdw. Alongi, D. M. and A. Sesakumar (1992).
Benthic communities. Coastal and estuarine
Further, Tellina remies also showed no studies: Tropical mangrove ecosystem. A. I.
significant difference of biomass value from Roberstson and D. M. Alongi. Wachington,
2004 to 2006. Field observation showed that T. DC, American Geophysical Union: 137-171.
remies migrated in August 2005. We observed Amarasinghe, M. D. and S. Balasubramaniam
about 15 to 20 individual animals m-2 floating (1992). "Net primary productivity of two
at the water surface during high tide moving mangrove forest stands on the northwestern
North. Massive harvesting of Anadara probably coast of Sri Lanka." Hydrobiologia 247(1):
was the triggering of T. remies migration. This 37-47.
behaviour might explain why T. remies showed Bosire, J. O., F. Dahdouh-Guebas, J. G. Kairo,
no significant differences over time. S. Cannicci and N. Koedam (2004). "Spatial
We concluded that converting mangrove to variations in macrobenthic fauna
ponds coincided with decreasing biomass of recolonisation in a tropical mangrove bay."
macrobenthic animals from 1996 to 2006, and Biodiversity and Conservation 13(6): 1059-
from 2004 to 2006 the nearby presence of 1074.
shrimp ponds coincided with low biomass of Bouillon, S., N. Koedam, A. Raman and F.
the macrobenthic fauna. However, loss of Dehairs (2002). "Primary producers
mangrove organic matter and discharge of sustaining macro-invertebrate communities in
shrimp pond water might not be responsible for intertidal mangrove forests." Oecologia
decreasing biomass of macrozoobenthic. 130(3): 441-448.
Habitat disturbance by fisheries was probably Bunt, J. S. (1995). "Continental scale patterns
the most reasonable explanation for declining in mangrove litter fall." Hydrobiologia
macrozoobenthic biomass.. 295(1): 135-140.
Choong, E. T., R. S. Wirakusumah and S. S.
Acknowledgements Achmadi (1990). "Mangrove forest resources
First AP would like to thank to Sriwijaya in Indonesia." Forest Ecology and
University, Palembang, Indonesia for Management 33-34: 45-57.
sponsoring my study. AP thanked to Ake and Cowan, V. J., K. Lorenzen and S. J. Funge-
Usup for their assistance during sampling and Smith (1999). "Impact of culture intensity
Ismail for acting as a speedboat driver. AP was and monsoon season on water quality in Thai
also glad that Edi P. assisted me on the commercial shrimp ponds." Aquaculture
measuring the AFDW. Finally, special thanks Research 30(2): 123-133.
are addressed to Dr. Harry ten Hove who Danielsen, F. and W. Verheugt (1990).
guided AP to identify the macrobenthic fauna. Integrating conservation and land-use
planning in the coastal region of South

251
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Sumatra. Bogor, Indonesia, Asian Wetland management." Journal of Coastal

Bureau-PHPA: 172 pp. Conservation 5: 69-80.
Dharma, B. (1988). Siput dan kerang Indonesia Kenny, A. J. and H. L. Rees (1994). "The
(Indonesian Shell). Jakarta., PT. Sarana effects of marine gravel extraction on the
Graha.: 111 pp. macrobenthos: Early post-dredging
--- (1992). Siput dan kerang Indonesia recolonization." Marine Pollution Bulletin
(Indonesian Shell II). Wiesbaden, Germany., 28(7): 442-447.
Verlag Chista Hemmen: 135 pp. Le, T. X. and Y. Munekage (2004). "Residues
Djamali, A. (1999). Sebaran dan Komposisi of selected antibiotics in water and mud from
Fauna Bentik. Ekosistem perairan sungai shrimp ponds in mangrove areas in Viet
Sembilang, Musibanyuasin, Sumatera Nam." Marine Pollution Bulletin 49(11-12):
Selatan. K. Romimohtarto, A. Djamali and 922-929.
Soeroyo. Jakarata, Pusat Penelitian dan Le, T. X., Y. Munekage and S.-i. Kato (2005).
Pengembangan Oseaneologi-LIPI: 49-54. "Antibiotic resistance in bacteria from shrimp
Djamali, A. and Sutomo (1999). Sosial farming in mangrove areas." Science of The
Ekonomi Budaya dan Perikanan. Ekosistem Total Environment 349(1-3): 95-105.
Perairan Sungai Sembilang, Musibanyuasin, Lee, S. Y. (1989). "The importance of
Sumatera Selatan. K. Romimohtarto, A. sesarminae crabs Chiromanthes spp. and
Djamali and Soeroyo. Jakarta, Pusat inundation frequency on mangrove (Kandelia
Penelitian dan Pengembangan Oseanologi- candel (L.) Druce) leaf litter turnover in a
LIPI: 67-76. Hong Kong tidal shrimp pond." Journal of
Gautier, D., J. Amador and N. Federico (2001). Experimental Marine Biology and Ecology
"The use of mangrove wetland as a biofilter 131(1): 23-43.
to treat shrimp pond effluents: preliminary --- (1989 a). "Litter production and turnover of
results of an experiment on the Caribbean the mangrove Kandelia candel (L.) druce in a
coast of Colombia." Aquaculture Research Hong Kong tidal shrimp pond." Estuarine,
32(10): 787-799. Coastal and Shelf Science 29(1): 75-87.
Halide, H., P. V. Ridd, E. L. Peterson and D. --- (1993). "The management of traditional tidal
Foster (2003). "Assessing sediment removal ponds for aquaculture and wildlife
capacity of vegetated and non-vegetated conservation in Southeast Asia: Problems and
settling ponds in prawn farms." Aquacultural prospects." Biological Conservation 63(2):
Engineering 27(4): 295-314. 113-118.
Hsieh, H. L., C. P. Chen, Y. G. Chen and H. H. --- (1995). "Mangrove outwelling: a review."
Yang (2002). "Diversity of benthic organic Hydrobiologia 295(1): 203-212.
matter flows through polychaetes and crabs --- (1999). "The Effect of Mangrove Leaf Litter
in a mangrove estuary: d13C and d34S Enrichment on Macrobenthic Colonization of
signals." Marine Ecology Progress Series Defaunated Sandy Substrates." Estuarine,
227: 145-155. Coastal and Shelf Science 49(5): 703-712.
Jackson, C. J., N. Preston, M. A. Burford and P. Lemonnier, H. and S. Faninoz (2006). "Effect
J. Thompson (2003). "Managing the of water exchange on effluent and sediment
development of sustainable shrimp farming characteristics and on partial nitrogen budget
in Australia: the role of sedimentation ponds in semi-intensive shrimp ponds in New
in treatment of farm discharge water." Caledonia." Aquaculture Reseach 37: 938-
Aquaculture Management of Aquaculture 948.
Effluents 226(1-4): 23-34. Lu, L., P. L. B.Goh and M. L. Chou (2002).
Jackson, C. J., N. Preston and P. J. Thompson "Effects of coastal reclamaation on riverine
(2004). "Intake and discharge nutrient loads macrobenthic infauna (Sungei Punggol) in
at three intensive shrimp farms." Aquaculture Singapore." Journa of Aquatic Ecosystem
Reseach 35: 1053-1061. Stress and Recovery 9: 127-135.
Jiang, J. X. and R. G. Li (1995). "An ecological Lu, L. and R. S. S. Wu (2000). "An
study on the Mollusca in mangrove areas in experimental study on recolonization and
the estuary of the Jiulong River." succession of marine macrobenthos in
Hydrobiologia 295(1): 213-220. defaunated sediment." Marine Biology
Karel, E. (1999). "Ecologycal effects o 136(2): 291-302.
dumping of dredged semdiments; options for

252
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Lui, T. H., S. Y. Lee and Y. Sadovy (2002). receiving shrimp farm effluent." Estuarine,
"Macrobenthos of a tidal impoundment at the Coastal and Shelf Science 59(2): 197-207.
Mai Po Marshes Nature Reserve, Hong Winberg, G. G. and A. Duncan (1971).
Kong." Hydrobiologia V468(1): 193-211. Methods for the estimation of production of
McKinnon, A. D., L. A. Trott, D. M. Alongi aquatic animals. London and New York,
and A. Davidspm (2002). "Water column Academic press: 174.
production and nutrient characteristics in
mangrove creeks receiving shrimp farm
effluent." Aquaculture Research 33(1): 55-73.
Nielsen, O. I., E. Kristensen and D. J.
Macintosh (2003). "Impact of fiddler crabs
(Uca spp.) on rates and pathways of benthic
mineralization in deposited mangrove shrimp
pond waste." Journal of Experimental Marine
Biology and Ecology 289(1): 59-81.
Odum, W. E. and E. J. Heald (1975). Mangrove
forests and aquatic productivity. In: A.D.
Hasler (ed.) Coupling of Land and Water
Systems. A. D. Hasler. Berlin, Springer: 129-
136.
Purwoko, A. (1996). The carrying capacity of
intertidal area: a feeding habitat of the stop
over of migratory birds in the sembilang
peninsula, South Sumatra, Indonesia.
Environmental Science. Wageningan.,
Wageningen Agricultural University: 45.
Rajendran, N. and K. Kathiresan (1996).
"Effect of effluent from a shrimp pond on
shoot biomass of mangrove seedling."
Aquaculture Reseach 27: 745-747.
Ricciardi, A. and E. Bourget (1999). "Global
patterns of macroinvertebrate biomass in
marine intertidal communities." Marine
Ecology Progress Series (MEPS) 185: 21 -35.
Sansanayuth, P., A. Phadungchep, S.
Ngammontha, S. Ngdngam, P. Sukasem, H.
Hoshino and M. S. Ttabucanon (1996).
"Shrimp ponf effluent: Pollution problem and
treatment by constructed wetlands." Water
Science and Technology 34(11): 93-98.
Soeroyo (1999). Struktur, komposisi, zonasi
dan produksi serasah mangrove. Ekosistem
perairan sungai Sembilang, Musibanyuasin,
Sumatera Selatan. K. Romimohtarto,
Djamali, A. and Soeroyo. Jakarta, Pusat
Penelitian dan Pengembangan Oseanologi-
LIPI: 55-66.
Trott, L. A. and D. M. Alongi (2000). "The
Impact of Shrimp Pond Effluent on Water
Quality and Phytoplankton Biomass in a
Tropical Mangrove Estuary." Marine
Pollution Bulletin 40(11): 947-951.
Trott, L. A., A. D. McKinnon, D. M. Alongi, A.
Davidson and M. A. Burford (2004). "Carbon
and nitrogen processes in a mangrove creek

253
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

AFDW macrobenthic fauna at

station 5 (S.Siput) in 1996
Treatments August Sept Oct
plot x 16.33 (a) 14.71 (a) 37.74 (b)
(a) (a) (b)
plot y 20.44 15.32 33.38

Table 1 Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna at plots x and y of station 5 at Sembilang
peninsula in different months. Different letters (in brackets) after biomass values in the same row indicate significant
differences (ANOVA: p<0.05).

2
WAFDW (gram/m ) Location: Sembilang peninsula in 2004-2006
Bivalves Station 1 Station 2 Station 3 Station 4 Station 5 Station 6
1 Anadara granosa 0.0144 0.0000 0.3704 4.1691 0.0000 0.8413
2 Hecuba scortum 0.0041 0.0358 0.1490 0.0085 0.2251 0.1136
(b) (b) (b) (b) (a) (b)
3 Solen sp. 0.0588 0.0774 0.0425 0.0377 0.0039 0.0410
(a) (a) (a) (a) (a) (b)
4 Tellina remies 0.0184 0.0801 0.2533 0.5155 0.7711 8.5615
(a) (b) (a) (a) (a) (a)
5 Tellina timorensis 0.1112 4.3512 0.2700 0.1933 0.1908 0.3260
(a) (b) (a) (b) (a) (c)
Total Bivalves 0.2069 4.5445 1.0852 4.9241 1.1908 9.8833
Gastropods
1 Clinthon oualaniensis 0.0117 0.0576 0.0039 0.0250 0.0059 0.0153
2 Littorina melanostoma 0.0000 0.0001 0.0000 0.0001 0.0000 0.0002
3 Nassa serta. 0.0176 0.0956 0.0128 0.0221 0.0164 0.0907
4 Thais buccinea. 0.0037 0.0066 0.0118 0.0032 0.0016 0.0115
Total Gastropods 0.0329 (a) 0.1599 (b) 0.0286 (a) 0.0504 (a) 0.0240 (a) 0.1176 (c)
Decapods (Crabs)
(a) (a) (a) (a) (b) (a)
1 Ocypodidae 0.0373 0.0184 0.1885 0.0412 1.5910 0.0271
2 Leucociidae 0.0799 0.0174 0.0346 0.0193 0.0189 0.0076
(a) (a) (a) (a) (b) (a)
Total decapods 0.1171 0.0358 0.2231 0.0605 1.6099 0.0347
Polychaetes (Worms)
1 Nereididae 1.6800 (b) 0.3009 (a) 0.1455 (a) 0.3151 (a) 0.1716 (a) 0.2186 (a)
(a) (b) (a) (a) (a) (a)
2 Maldanidae 0.0039 0.0445 0.0069 0.0158 0.0176 0.0301
(a) (b) (b) (b) (c) (c)
3 Lumbrineridae 0.0065 0.0154 0.0196 0.0288 0.0456 0.0416
4 Capitellidae 0.0009 0.0143 0.0114 0.0026 0.0022 0.0036
5 Sternaspidae 0.0041 (a) 0.0394 (a) 0.0839 (b) 0.2346 (b) 0.0344 (a) 0.0469 (a)
6 Unidentified worms 0.0189 (a) 0.0370 (a) 0.0051 (a) 0.0041 (a) 0 (a) 0.2609 (b)
(c) (a) (a) (b) (a) (b)
Total Polychaetes 1.7145 0.4034 0.2724 0.6009 0.2715 0.6017
(a) (b) (a) (b) (a) (c)
Total macrob. Animals 2.0715 5.1436 1.6093 5.6359 3.0962 10.6374

Table 2. Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna taxa at six sampling stations at Sembilang
peninsula in 2004 to 2006. Different letters (in brackets) after biomass values in the same row (species of macrobenthic
animals) indicate significant differences (ANOVA: p<0.05).

AFDW of macrobenthos fauna at Sembilang peninsula in June from

2004 to 2006 over 6 stations
Locations 2004 2005 2006
(a) (a) (b)
Station 1 2.23 3.45 5.34
(a) (b)
Station 2 0.62 1.48 2.24 (b)
(a) (b) (a)
Station 3 1.35 1.90 1.42
(b) (a) (a)
Station 4 47.45 1.05 1.08
Station 5 0.87(a) 0.49 (a) 9.95 (b)
(a) (b) (b)
Station 6 0.94 2.82 2.83

Table 3 Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna in different stations at Sembilang peninsula
in June 2004 to 2006. Different letters (in brackets) after biomass values in the same row indicate significant differences
(ANOVA: p<0.05) between years.

WAFDW (gram/m2) Sembilang peninsula in June …

254
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Bivalves 2004 2005 2006

1 Anadara granosa 7.7037 0.1163 0.0000
2 Hecuba scortum 0.0269 0.1131 0.0007
3 Solen sp. 0.1278 (a) 0.1854 (b) 0.0725 (a)
(b) (a) (a)
4 Tellina remies 0.4092 0.3645 0.0283
5 Tellina timorensis 0.1996 0.1688 0.1012
(b) (a) (a)
Total Bivalves 8.4672 0.9481 0.2027
Gastropods
1 Clinthon oualaniensis 0.0673 0.0000 0.0000
2 Littorina melanostoma 0.0009 0.0000 0.0000
(b) (a) (ab)
3 Nassa serta. 0.0853 0.0019 0.0134
4 Thais buccinea. 0.0375 0.0000 0.0000
Total Gastropods 0.1911 (b) 0.0019 (a) 0.0134 (a)
Decapods (Crabs)
1 Ocypodidae 0.0640 (a) 0.2985 (a) 1.6395 (b)
2 Leucociidae 0.0000 0.0000 0.0000
(a) (a) (b)
Total Decapods 0.0640 0.2985 1.6395
Worms
1 Nereididae 0.0494 (a) 0.5594 (a) 1.4588 (b)
2 Maldanidae 0.0269 0.0094 0.0126
(a) (a) (b)
3 Lumbrineridae 0.0269 0.0251 0.0430
4 Capitellidae 0.0180 0.0003 0.0000
(a) (a) (b)
5 Sternaspidae 0.0314 0.0220 0.1571
6 Unidentified worms 0.0135 (ab) 0.0011 (a) 0.2828 (b)
Total Worms 0.1661 0.6174 1.9543
Total macrob. Animals 8.91 (b) 1.87 (a) 3.81 (b)

Table 4 Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna taxa at Sembilang peninsula in June 2004 to
2006. Different letters (in brackets) after biomass values in the same row indicate significant differences (ANOVA: p<0.05).

Location: Sembilang peninsula in 2004-2006

WAFDW (gram/m2)
Non-pond
Bivalves sites Pond sites
1 Anadara granosa 1.8032 1.5132
2 Hecuba scortum 0.1920 0.1275
3 Solen sp. 0.1325 ** 0.0280
**
4 Tellina remies 3.4328 0.5133
**
5 Tellina timorensis 3.3016 0.2180
**
Total Bivalves 8.8621 2.4001
Gastropods
1 Clinthon oualaniensis 0.0628 0.0116
2 Littorina melanostoma 0.0002 0.0000
**
3 Nassa serta. 0.1228 0.0171
4 Thais buccinea 0.0163 0.0055
Total Gastropods 0.2020 ** 0.0343
Decapods (Crabs)
**
1 Ocypodidae 0.6530 0.6069
**
2 Leucociidae 0.0917 0.0242
**
Total Decapods 0.7447 0.6312
Worms
1 Nereididae 1.6042 ** 0.2107
**
2 Maldanidae 0.0558 0.0134
3 Lumbrineridae 0.0598 0.0314
4 Capitellidae 0.0168 0.0054
5 Sternaspidae 0.1623 ** 0.1176
6 Unidentified worms 0.1273 ** 0.0031
**
Total Worms 1.9941 0.3816
Total macrob. Animals 11.8030 ** 3.4471

255
Proceeding of Conf. on Industrial and Appl. Math., Bandung-Indonesia 2010

Table 5. Average biomass in g ash-free dry weight (g m-2) of macrobenthic fauna taxa at non-pond (station 1, 2 and 6) and
pond sites (station 3, 4 and 5) at Sembilang peninsula in 2004 to 2006. Different ** after biomass values in the same row
(species of macrobenthic animals) indicate significant differences (ANOVA: p<0.05).

Figure 1 Sampling stations along the coast of Sembilang peninsula. Each station consists of three transect each with seven
(2004), ten (2005 &2006) sampling points, plot 1 & 1 (100*100 m, 1996) at station 5.

256