Volume 1, Issue 1

ISSN: - 2306-708X
December 2012
2012-13 International Journal of Information Technology and Electrical Engineering

Enhancement of Harmonic State Estimation Using Harmonic Prediction

1
Hussam M. M. Alhaj, Nursyarizal Mohd Nor, Vijanth S.Asirvadam and M. F. Abdullah
1
Department of Electrical and Electronic Engineering, Universiti Teknologi PETRONAS, Malaysia
E-mail: 1hussamalhaj13@hotmail.com

ABSTRACT
The world wide increasing using nonlinear loads, mostly consisting of power electronics devices in many applications that bring
a great concern about power quality problems caused by these nonlinear loads, harmonics pollution is main concern of power
quality issues due to its harmful effect on power system network, several harmonic state estimation H.S.E methods have been
devolved to estimate the harmonic magnitude level and to identify the harmonic sources ,This paper presents a concept, method
and function of harmonic state estimation and implementation of the singular value decomposition S.V.D algorithm on power
system harmonic state estimation for partially observable systems with harmonic prediction method to compensate the loss of
the observation and maintain the redundancy of the measurements. The result shown that the proposed harmonic prediction
method can compensate the loss of the measurement and therefore; enhance the performance of S.V.D algorithm when the
loss of observation occurs.

Keywords: harmonic state estimation, observablity, singular value decomposition S.V.D algorithm, harmonic prediction
determine solvability of the measurement equation first.
1. INTRODUCTION
The presence of voltage and current waveform distortion Reference [7] classified the bus bars to tow sets which
is generally expressed in terms of harmonic frequencies which Source of harmonic and non source of harmonic and used the
are integer multiples of the fundamental frequency [1].The information of non source to improve the observability. The
main sources of the harmonics are the nonlinear loads such as singular value decomposition algorithm was introduced in
arc furnace, fluorescent lighting and adjustable speed drives [8,9] instead of normal equations to solve the underdetermined
and other power electronic equipment that have been used in system and found this method more robust, however, it need
transmission and generation level such as Flexible AC higher computational cost and by inspection of the null space
Transmission System (F.A.C.T.S) devices and static switch. vectors gives the observability and particular solution can be
When a source of sinusoidal voltage is applied to a nonlinear determined. Harmonic state estimation via sparsity
device or load, the resulting current is not perfectly sinusoidal. maximization to transform the system from unobservable to
In the presence of system impedance this current causes a non- observable system under proper measurement arrangements
sinusoidal voltage drop and, therefore, produces voltage Proposed in [10]. Artificial neural networks to provide initial
distortion at the load terminals [2]. estimates of the harmonic sources (pseudo measurements)
based on the measured harmonics and fundamental load flows
The major adverse effects of the harmonics are the and then improve the estimates by State estimation introduced
heating of the induction motors, transformers , capacitors and in [11,12] .
the over loading of neutrals and so on , Many techniques have
been developed to mitigate the distortion (harmonic) and to Total least (TLS) square method to estimate the
avoid its harmful effect on power system network (such as harmonic measurement errors with parameters errors
active and passive filter). To do any action related to the presented in [13]. A new approach to harmonic state
mitigation of the harmonics the network has to be monitored estimation is proposed by firstly classifying the network nodes
and the harmonics level known in the network. Due to the into four types in [14].Utilization of Evolutionary Strategy
size, complexity and high cost of power quality meters the advantage such as its modeling facilities as well as its
system is not fully monitored, hence there is need for potential to solve fairly complex problems presented in [15].
estimation. The main goal of harmonic state estimation is, Although these efforts in harmonic state estimation, it is still
therefore, to economically generate the best estimation of the challenge to estimate reliably all the network state variables in
harmonic magnitude and the source location at unmonitored the system-wide or even in moderate size network when the
bus bars from limited measurements [3]. measurements fewer than the state variables. Usually the
harmonics measurements installed in the gird and via
This problem of harmonic state estimation first was communication channel the harmonic information sent to the
introduced by Heydt in 1989[4] he used least square control unit where all the calculation and estimation done
estimation to estimate the harmonic state and identifying the using computerize tools but sometimes this information been
harmonic source and the problem of harmonic state estimation lost in random fashion due to communication error and this
handled as an inverse of harmonic power flow. In [5, 6] loss effect the performance of harmonic state estimation due to
adopting the traditional normal equation that use for few numbers of measurements. Most of the works done in
fundamental frequency state estimation which require the harmonic state estimation focusing in harmonic state
system to be fully observable. To solve the problem of the estimation method or algorithm itself. In this paper the
observability the authors separate the observability analysis to problem of harmonics information loss addressed and
 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

x can be determined from the measurement information z

prediction method based on Finite Impulse Response (F.I.R) The system is said to be observable if all state variables
Adaptive Filters with least mean square and Weights Update
Based Error Displacement (W.U.E.D) introduced to predict sometimes the system is not observable hence, Virtual
the future harmonics that can compensate the observation loss measurements and pseudo measurement are used to improve
the observabiltiy, Virtual measurements are type of
The paper is organized as follows: first, the art of information does not need metering and pseudo
harmonic state estimation explained and the S.V.D and measurements is a kind of information we can get from
harmonic prediction methods were described and then Results historical dada. [16].
and conclusion were discussed respectively.
In the traditional state estimation which fundamental
2. BACKGROUND frequency state estimation the system is Over-determined or
completely determined and full or partial observable due to
2.1. State estimation abundant measurements devices such as revenue meters ,

measurement set data vector z to the state variables vector x

The general form of the state estimation which related the therefore, a unique solution can be obtained by using standard
techniques such as L.U decomposition, back-substitution,
and the system topology is Cholesky decomposition or Gauss-Jordan elimination.
However, in harmonic state estimation the technical challenge
z= Hx + 1 is to solve the measurements equations for underdetermined

n > ! due to the limit number of harmonics measurements

system, that is, systems with more unknowns than equations,
Where z is a (m 1) vector of known measurements
and x is a (n 1) vector of state variables (unknown and information. In these systems, a unique solution may not
quantities) for which the equation must be solved. Hx Is a exist [16].
(m n) measurement function relating the known
measurements to state variables and is measurement error 2.2. Singular value decomposition (S.V.D)
with standard deviation .
Equation (1) in harmonic state estimation represent as
involves finding the n vector x that
Equation (1) is widely solved as weighted lest square

minimizes the index jx where VhYh = Ih 6

(WLS) which

jx = 1/2z Hx Wz Hx Where Ih representing the harmonic injection currents
in each bus, Vh harmonic voltage in each bus, Yh the
(2)

system bus admittance matrix and h denote the harmonic

W = R order for instance at fundamental frequency h = 1 .
 = 1/
Where

(3)
Equation (5) in harmonic state estimation is usually
R  Is the measurement covariance matrix where  is underdetermined system due to limited harmonic
the variance of the measurement error. W, can be used to measurements therefore, the matrix of the equation being
represent meter accuracy or reliability. Engineering judgment singular and the solution by standard technique which normal
equation cannot obtained however (S.V.D) algorithm can
error is unknown, W is typically assumed to be a unit matrix,
is normally needed for the selection of W, and if measurement
provides stable and unique solution for partial observable
system.[9] S.V.D is based on a theorem from linear algebra
Differentiating jx yields the first-order optimal
assuming that all measurements are equally accurate [16].
which says that a rectangular matrix A can be decomposed
conditions. into product of three matrices , an orthogonal matrix U ,a
diagonal matrix S and the transpose of an orthogonal matrix V
Gx = H  Wz
. [18]
(4)

Where x is the state estimate and G = H  WH is the state

The theorem is usually presented something like this:

A () = U(( S() V))


estimation gain matrix.
(7)
H WHx = H Wz
 
Where the columns of U are orthonormal eigenvectors of
(5)
AA , the columns of V are orthonormal eigenvectors of A A,
and S is a diagonal matrix containing the square roots of
The equation (5) will be calculated usually using standard
eigenvalues from U or V in descending order which are called
techniques such as L.U decomposition, back-substitution,
singular values [18].The columns of U correspond to the non-
Cholesky decomposition or Gauss-Jordan elimination. [17]
zero singular values and the columns of V correspond to the
State estimation problem can be classified to three situations.
zero singular values are the column space and null space of A
Over-determined system when the number of equations (m)

which m > n, completely determined system when the

greater than the number of unknown state variables (n)
respectively[9].In underdetermined system while most
traditional techniques fail, S.V.D is able to provide a
state variables (n) which m = n and Under-determined system
number of equations (m) equal to the number of unknown
particular solution and a null space vector for each singularity

x = x ,  + )/0)1 2 k  x)  8

unknown state variables (n) which m < n.
when the number of equations (m) less than the number of
3

12
 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

Where x ,  is a particular solution, x )  is null space

vector and k  constant .The another advantage of SVD is that
Table II Harmonic voltage measurement
Bus No. Harmonic volt
5 0.00232+j*0.00073
From the null space vector the system island can be 6 0.00225+j*0.00067
determined observable or not by inspection the element of the
8 0.00253+j*0.00003
null space vector if the element is zero this meaning null space
element will not alter the particular solution and therefore
Table III Harmonic current measurement
observable otherwise the null space element can alter the
Bus No. Harmonic current
solution and in this case unobservable [9]. 1 0.0045+j*0.0021
2 0.0049+j*0.0009
3. RESULT AND DISCUSSION 5 0.0071+j*0.0071
6 0.0063+j*0.0136
3.1. SVD implemented in harmonic state estimation
The S.V.D harmonic state estimation algorithm has All the passive elements are considered to behave
been developed and implemented in the system with data linearly with frequency [20].

5 = 6789:;8:
found in reference [19].

<= > = ?><=

(9)
(10)
<@ > = ? B
A
The system shown in Fig.1 Consist of 8 bus bar, 4
C
harmonic current measurements and 3 harmonic voltage (11)
measurements to estimate the rest of harmonic voltage from
EC IC ,
IC,G LC
this information and the admittance matrix
D H =D H D H 12
ECG ICG, ICG,G LCG
Voltage
measurement

Where the current IMN is the phasor

Current

PI P
frequency h injected at node,N ,i.e IM = M < M , YMN,N is
measurement current at
N N

the equivalent admittance at frequency h at node N, VMN is the

phasor voltage at frequency h injected at node N and N is
1 6

number of the nodes of the electrical network . Using the

5 information above the (S.V.D) harmonic state estimation
3 4
implemented as shown in Fig. 2.[19]

2 7
8

Fig.1 8 Bas bars system

Based on the harmonic voltage in buses number (5, 6,
and 8) and the harmonic current in (1, 2, 5, and 6) and the
admittance data of the system, the harmonic voltage in bus
number (1, 2, 3, 4, and 7) was calculated by (S.V.D). The
harmonic order assumed to be one just to explain how
(S.V.D) method works in harmonic state estimation. The
information of this system given in tables below as fellow:

Table I the system impedance data

Bus Bus R X 1/2B
No. No. P.u P.u P.u
1 2 0.00099 0.00820 0.02600
1 3 0.00020 0.00800 0.00000
2 3 0.00010 0.01200 0.00000
3 4 0.03000 0.18700 0.30000
4 5 0.00080 0.01200 0.05000
5 6 0.00100 0.02000 0.03000
6 7 0.00150 0.05200 0.04000 Fig.2S.V.D implementation
5 7 0.00956 0.05498 0.09092
5 8 0.01000 0.06000 0.10000
 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

The result of estimated harmonics voltage from S.V.D 3.3. The Finite Impulse Response F.I.R Adaptive Filters
as in table IV below
May be understood as a self-modifying filter that adjusts
Table IV S.V.D estimated harmonic voltage
its coefficients in order to minimize an error function. This
Bus Real Image Magnitude Angle error function, also referred to as the cost function, is a
No. (Deg.) distance measurement between the reference or desired signal
1 0.000006 0.000012 0.000013 114.48232 and the output of the adaptive filter [21]. The basic
2 0.000003 0.000019 0.000019 100.38569 configuration of an adaptive filter, operating in the discrete-
3 0.000009 -0.000031 0.000032 -73.78603 time domain k is shown in Fig. 3.
4 0.002283 0.000777 0.00242 18.558

_^ ] ^ ^
7 0.002780 0.000815 0.002897 16.34612

a
c

`^
But when the loss of the observation due to any error

b^
occur the result will diver from the actual result. For instance -
if the current measurement in bus No1 and 6 lost the SVD
result of harmonic voltage will change as shown in tables
below .(all the underlined value changed) Fig. 3 Adaptive filter

The input signal is given by xk the reference signal or

Table V harmonic voltage estimation when, current information lost
desired signal dk or yk represents the desired output
in bus1
y ^ k is the output of the adaptive filter, and hence the error
signal (that usually includes some noise component), and
Bus Real Image Magnitude Angle
No. (Deg.)
1 0.000001 0.000010 0.000010 -86.47639 signal is defined as:

ek = dk y ^ k 13

2 0.000002 0.000016 0.000016 96.13432
3 0.000001 -0.000006 0.000007 -80.06974
4 0.002283 0.000777 0.00242 18.80749 3.4 . Least Mean Square Algorithm (L.M.S)
7 0.002780 0.000815 0.002897 16.34612 It is a celebrated algorithm and is widely used due to
simplicity, low computational complexity and robustness. The

require matrix inversion and the update of kth coefficient

Table VI harmonic voltage estimation when, current information lost
simplicity of (L.M.S) comes from the fact that it does not
in bus6
Bus Real Image Magnitude Angle requires only one multiplication and one addition [22].The
No. (Deg.) L.M.S algorithm for Nth-order (F.I.R) adaptive filters
1 0.000006 0.000012 0.000013 114.48232
shown in steps below [22].
2 0.000003 0.000019 0.000019 100.38569
3 0.000009 -0.000031 0.000032 -73.78603

Initialization: wX =0
Parameters: M= filter order, = step size
4 0.002733 0.000970 0.002900 19.54524
7 0.000608 0.000146 0.000625 13.49526 Computation: for k =0, 1, 2
y ^ k = w  xk 14
ek = dk y ^ k (15)
(a)

Wk = wk 1 + ekxk 16

From the results of these tables random observation loss (b)
due to any error such as communication error, instrument (c)
calibration etc, significantly deteriorate the performance of
2
Where the step size
0 < < 17
S.V.D harmonic state estimation so one step ahead
harmonic prediction may compensate the random observation
loss and maintain the measurements redundancy and therefore; (0[
enhance the performance of S.V.D when the information (0[ is the largest eigenvalue of the input signal due to
lost. the difficulty of determination of the step size that is a trade-
3.2. Harmonic prediction off between the steady-state excess error and the speed of
To predict one step ahead real time harmonics to convergence.[9].
compensate the loss of harmonics observation due to the 3.5. Weights Update Based Error Displacement (W.U.E.D)
communication error in harmonic state estimation, The Finite The output error is one of the most effective factors in the
Impulse Response F.I.R Adaptive Filters with least mean convergence of the weights, since it determines in which
square and Weights Update Based Error Displacement direction the weights will step parallel to the received
W.U.E.D utilized. W.U.E.D used to eliminate the effect of observation, as well as it contributes in the length of the
the output error on the predictor performance when random step[23]. In general when the presented training data to the
losses to the observation (measurement) happen predictor is consistent and continuous the output error
decreases until the weights attain the optimal values, then the
weights will be frozen. But when the observations are lost in
random fashion the output error doesnt direct the weights
optimally to the to the solution region due to the random
losses of the predictor input.W.U.E.D reduces the negative

14
 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

Using the historical information ofAM i cosM i and

AM i sinhi in Eq.26 to predict the future value of harmonics
effects of the output error by controlling the weights update

where coswM t and sinwM t are constant.

based on error displacement (D.e). Shown by the following
steps.

The prediction of A M i cosM i and A M i sinM i

Step 1: Computing the value of the error displacement:

y1 y1^
De1 = e e 18
done in parallel for each harmonic order which Eq. (26) dealt
y1 as tow equation and then the magnitude and phase angle for
Step 2: Checking whether the error displacement is smaller each predicted harmonic order can be calculated as

De1 < f 19

than a threshold ():
|IM | = AM i cosM i  + A M i sinM i  28

A M i sinM i
Step 3: If the error displacement is smaller than the threshold,
M = :;8 29
A M i cosM i
the weights are updated:

wk = wk 1 + ekxk 20

The frame work of L.M.S and (W.U.E.D) harmonic
Otherwise the weights are frozen: predictor as shown in figure below:

wk = wk 1 21

3.6. Model of Instantaneous Current Harmonics.

As mentioned above the harmonics are represented as
integer multiples of the generated frequency.

ii = cA M i cos 2hfX t + M i  + ni 22

M
h 1,3,5,7, 

A Mi is harmonics time- varying amplitude

fX
h is harmonics order
is the fundamental frequency

ni
M i is the harmonics order phase angles
is random noise

t = 0: rt: number of points

rt = 1/fw
fw = sampling frequency
For simple mathematical calculationrt used instead of t Fig .4(L.M.S and W.U.E.D)

Using the Trigonometry in Eq. 23 The historical data collected by fluke 43B analyzer from
cosA + B cosA cosB sinA sinB 23
tow power supplies one supply 15 personal computers desktop
(P.C) and the other supply 33 computers desktop working
Eq. 22 will be in the following form: independently as single loads harmonics source and the
yi = wi i + ni 24
fundamental frequency, sample frequency equal to 50Hz,
51200Hz respectively and the number of points is 1024 while
Where the noise added to data has standard deviation equal to 0.1 and
A  cos i , A i sin i ANi cosN i , AN i sinN i
zero mean the dada number used to train the predictor was
i = z  i { 24 1000, the step size of least mean square taken as 0.05 for 14
PC data and 0.1 for 33 P.C data . The results are shown in
wi =  cosw t , sinw t , , coswNt , sinwN t  (25) Tables and figures below:

To predict any harmonic order magnitude and phase angle Table VII Mean Square Error for harmonic prediction (14P.C)

yi = coswM t AM i cosM i sinwM t AM i sinM i + ni 26)

Harmonic Mean Square Error for 14 PC
order
AN i cosN i AN i sinN i
Model the individual harmonic order for instance if the
harmonic order is third Eq. (26) will be as 3 0.0176 0.0211

yi = cosw} t A} i cos} i sinw} t A} i sin} i + ni 27

5 0.0171 0.0174
7 0.0161 0.0164
 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

Table VII Mean Square Error for harmonic prediction (33P.C)

Harmonic Mean Square Error for 33 PC
order
AN i cosN i AN i sinN i

3 0.0563 0.0488
5 0.0451 0.0310
7 0.0216 0.0216

Five Amperes of the fundamental current added to the

actual and predicted harmonics order to show the shape of
predicted harmonics for 14 P.C data and (25) Amperes Fig.9 Fifth harmonic predicted (33 P.C)
added to 33 P.C data.

Fig.11 Seven harmonic predicted (14 P.C)

Fig.5 Third harmonic predicted (14 P.C)

From the tables and figures the presented methods able to

predict one step ahead future harmonics with small mean
square error this can compensate the loss of harmonics
measurements in harmonic sate estimation.

4. CONCLUSION
Harmonic State Estimation is vital tool for monitoring
and determining the harmonics level and sources of the
system, due to the cost of harmonic measurements the system
is in underdetermined and the traditional algorithm such as
L.U decomposition, back-substitution, Cholesky
Fig.6 Fifth harmonic predicted (14 P.C)
decomposition for over determined system fail to give solution
but S.V.D able to give solution . Harmonic State Estimation
is real time implementation using communication channels
and computerizes tolls usually in the control center unit due to
complexity of acquiring harmonic information from the
network sometimes the information lost in random fashion and
therefore; affect the performance of the estimator. Prediction
of future harmonic may compensate the loss of harmonic
information and enhance the estimator when the loss happens.

5. ACKNOWLEDGMENT
The authors gratefully acknowledge University
Technology PETRONAS (UTP) for providing the facilities
Fig.7 Seven harmonic predicted (14 P.C) and (ERGS) for financial support.
.

REFERENCES
[1] Fourier, J B J (1822). Thhorie Analytiyue de la Chaleur
(book), Paris.
[2] Power System Harmonics, Second Edition J. Arrillaga,
N.R. Watson, 2003 John Wiley & Sons, Ltd ISBN: 0-
470-85129-5

Fig.8 Third harmonic predicted (33 P.C)

 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

[3] Neville R. Watson, Power quality state estimation. [20] Enrique Acha, Manuel Madrigal, Power systems
European Transaction on Electrical power 2010; harmonics: computer modeling and analysis, Wiley,
20:19-3 2001.
[4] G. T. Heydt, Identification of harmonic sources by a [21] J. A. Apolinario Jr, QRD-RLS Adaptive Filtering,
state estimation technique, IEEE Trans., Power Springer 2009.
Delivery, Vol. 4, No. 1, pp. 569576, 1989. [22] Hayes, Statistical Digital Signal Processing and
[5] ZP Du, J Arrillaga, N.R Watson, Continuous Modeling, New York: John Wiley & Son, Inc, 1996.
harmonic state estimation of power systems, IEE [23] Vijanth S. Asirvadam and Musab J.O. Elamin, Wireless
Proceedings Generation, Transmission and system identification for linear network, Signal
Distribution; vol. 143, issue 4, pp. 329336, Jul 1996. Processing & Its Applications, CSPA 2009. 5th
[6] ZP Du, J Arrillaga, N.R Watson, S Chen, International Colloquium on 6-8 March 2009
Identification of harmonic sources of power systems
using state estimation, IEE Proceedings Part-C,
Generation, Transmission and Distribution, vol. 146, Hussam Mubarak M. Alhaj was born in Sudan. He has been
issue 1, pp.712, Jan 1999. working towards PhD Degree in Electrical Engineering
[7] Van Long Pham a, Kit Po Wong a, Neville Watson b, emphasis on Power Systems from Universiti Teknologi
Jos Arrillaga, A method of utilizing non-source PETRONAS Malaysia. His current research interest includes
measurements for harmonic state estimation, Electric power system harmonic state estimation, harmonic prediction,
Power Systems Research 56 (2000) 231241 frequency estimation and adaptive Signal processing
[8] C Madtharad, S Premrudeepreechacharn, N.R Watson, algorithms.
Power system state estimation using singular value
decomposition, Electric Power System Researc, 67, Ir. Dr. Nursyarizal Mohd Nor obtained his PhD in Electrical
pp. 99107, 2003 Engineering from Universiti Teknologi PETRONAS (UTP),
[9] K .C.C Yu, N.R Watson, J Arrillaga, Three phase Malaysia in year 2009. In year 2001 he obtained his MSc in
harmonic state estimation using SVD for partially Electrical Power Engineering from The University of
observable systems, Proceedings of PowerCon 2004, Manchester Institute of Science and Technology (UMIST),
Singapore, Nov. 2004. UK. His areas of specialization are Analysis and optimization
[10] H. Liao Power System Harmonic State Estimation and of large scale power systems and State estimation. He has
Observability Analysis via Sparsity Maximization , several publications at his credit. His research interests are in
Power Systems, IEEE Transactions on Feb. 2007 Power Economics Operation and Control, Power Quality,
[11] R. K. Hartana and G.G. Richards, Harmonic Source Renewable Energy and Power System Analysis
Monitoring and Identification Using Neural Networks,
IEEE Trans. on Power Systems, vol. 5, no. 4, pp. 1098- Dr. Vijanth S. Asirvadam is from an old mining city of
1104, November 1990. Malaysia called Ipoh. He studied at the University of Putra,
[12] M. Negnevitsky, M. Ringrose Monitoring Multiple Malaysia for the Bachelor Science (Hon) majoring in
Harmonic Sources in Power Systems Using Neural Statistics and graduated in April 1997 before leaving to
Networks, Power Tech, 2005 IEEE Russia Belfast Queens University Belfast to do his Masters
[13] Niancheng Zhou, Li lin and jizhong Zhu, An where he received his Masters Science degree in
Approach to Harmonic State Estimation of Power Engineering Computation with a Distinction. He later joined
System, J. Electromagnetic Analysis & Applications, the Intelligent Systems and Control Research Group at
2009, 3: 192-194 Queens University Belfast in November 1999 where he
[14] H.T.Nguyen, JJ.Yang and S.S.Choi, On harmonic completed his Doctorate (Ph.D.) researching Online and
state estimation and the evalution of harmonic power Constructive Neural Learning methods. He took previous
contribution from sources, IEEE 2010 employments as a System Engineer and later as a Lecturer at
[15] Elcio F. de Arruda, Nelson Kagan and Paulo F. the Multimedia University Malaysia and also as a Senior
Ribeiro, Harmonic Distortion State Estimation Using Lecturer at the AIMST University. Since November 2006, he
an Evolutionary Strategy IEEE Transactions on serve as a Senior Lecturer and later as Associate Professor at
Power Delivery, vol. 25, NO. 2, April 2010 the department of Electrical and Electronics Engineering,
[16] Ali Farzanehrafat and Neville R.Watson, review of Universiti Teknologi PETRONAS. His research interest
Power Quality State Estimation,IEEE, Universities includes linear and nonlinear system identification, model
Power Engineering Conference (AUPEC), 2010 20th validation and application of computing system in signal and
Australasian image processing. Dr Vijanth is member of Institute of
[17] T. L. Tan, S. Chen and S.S. Choi An Overview of Electrical and Electronics Engineering (IEEE) and also a
Power Quality State Estimation, Power Engineering member of Institute of Engineering Technology (IET) . He has
Conference, IPEC 2005. The 7th International over 100 publications in proceedings and journals in the field
[18] Kirk Baker, Singular Value Decomposition Tutorial, of computational intelligence, control, signal and image
March 29, 2005 processing.
[19] Arrillaga, N.R. Watson and S. Chen, Power System
Quality Assessment. John Wiley & Sons LTd.,
Chichester (UK), 2000

17
 Alhaj. H. M. M, Nor. N. M, Asirvadam. V. S and Abdullah. M. F: Enhancement of Harmonic State Estimation Using Harmonic Prediction
2012-13 International Journal of Information Technology and Electrical Engineering

M. F. Abdullah completed his Master of Electrical

Engineering from Universiti Tenaga Nasional (UNITEN) in
2006 and now pursuing PhD at Universiti Teknologi
PETRONAS (UTP). He is currently a senior lecturer at UTP
since 2009. Prior to that, he was working with utility company
Tenaga Nasional Berhad (TNB) since 1989 and serving
distribution Division for 12 years and transmission division
for 8 years. His experience in distribution Division includes
planning, construction, maintenance, metering and protection.
In transmission division, he was working in maintenance
department that responsible for Substation, lines, cables,
protection, Telecontrol and technical support. He is a
Professional Engineer (Board of Engineers Malaysia),
Competent Engineer (Energy Commission), member of The
Institute of Engineers, Malaysia (IEM) and member of The
Institute of Electrical and Electronics Engineers (IEEE)

18