World Academy of Science, Engineering and Technology 63 2010

Automatic Generation Control of Multi-Area

Electric Energy Systems Using Modified GA
Gayadhar Panda, Sidhartha Panda and C. Ardil

feedback [5]. Recently, AGC based on fuzzy controller has

Abstract—A modified Genetic Algorithm (GA) based optimal also been proposed [11]. Among the aforementioned
selection of parameters for Automatic Generation Control (AGC) of controllers, the most widely employed one is the fixed gain
multi-area electric energy systems is proposed in this paper. controller, like integral controller or PI controller due to its
Simulations on multi-area reheat thermal system with and without low cost and high reliability in operation. Fixed gain
consideration of nonlinearity like governor dead band followed by controllers are designed at nominal operating points and may
1% step load perturbation is performed to exemplify the optimum
no longer be suitable in all operating conditions. For this
parameter search. In this proposed method, a modified Genetic
Algorithm is proposed where one point crossover with modification reason, some authors have applied the variable structure
is employed. Positional dependency in respect of crossing site helps control [14] to make the controller insensitive to the system
to maintain diversity of search point as well as exploitation of parameter changes. However, this method requires the
already known optimum value. This makes a trade-off between information of the system states, which are very difficult to
exploration and exploitation of search space to find global optimum predict and collect completely.
in less number of generations. The proposed GA along with For optimization of any AGC strategy, the designer has to
decomposition technique as developed has been used to obtain the resort to studying the dynamic response of the system. Study
optimum megawatt frequency control of multi-area electric energy of dynamic response of modern large multi area power system
systems. Time-domain simulations are conducted with trapezoidal
costs in terms of computer memory and time. Moreover,
integration along with decomposition technique. The superiority of
the proposed method over existing one is verified from simulations
inclusion of governor dead band nonlinearities in the system
and comparisons. model makes the solution process more complex. Therefore,
there is always an impelling motive to search for a suitable
Keywords—Automatic Generation Control (AGC), Reheat, mathematical model and an easier method of solution.
Proportional Integral (PI) controller, Dead Band, Genetic Algorithm All these motivate a more practical approach for parameter
(GA). optimization in AGC of multi-area electric energy systems
using modified Genetic Algorithm. A digital simulation is
I. INTRODUCTION used in conjunction with the proposed Genetic Algorithm
(GA) to determine the optimum parameters of the AGC for
M EGAWATT frequency control or Automatic Generation
Control (AGC) problems are that of sudden small load
each of the performance indices considered. Genetic
Algorithms are used as parameter search techniques, which
perturbations which continuously disturb the normal operation
utilize the genetic operators to find global optimal solutions
of an electric energy system. In the literature, there has been
[9-12]. For solution of system state equations, decomposition
considerable effort devoted to automatic generation control of
technique [13] along with trapezoidal integration method is
interconnected electric energy system [1-5]. These approaches
used to overcome the difficulty of large multi-area power
may be classified into two categories as follows:
system.
1. Energy storage system: Examples are pumped storage
A number of different methods for optimizing well-behaved
system, superconducting magnetic energy storage system,
continuous functions have been developed which rely on
battery energy storage system etc. [6, 7].
using information about the gradient of the function to guide
2. Control strategy: This category focuses on the design of
the direction of search [13]. If the derivative of the function
an automatic generation controller to achieve better
cannot be computed, because it is discontinuous, for example,
dynamic performance [8-10].
these methods often fail. Such methods are generally referred
In this paper, design of AGC controller is investigated. Many
to as hill climbing. They can perform well on functions with
controllers have been proposed for AGC problem in order to
only one peak (unimodal functions). But on functions with
achieve a better dynamic performance. Examples are the
many peaks, (multimodal functions), they suffer from the
proportional integral (PI) control [3], state feedback control
problem that the first peak found will be climbed, and this
based on linear optimal control theory [4], and output
may not be the highest peak. Having reached the top of a local
maximum, no further progress can be made.
Gayadhar Panda is with the Electrical Engineering, IGIT, Sarang, Orissa, The significant contributions of this work are as follows:
India, Pin: 759146. (e-mail: p_gayadhar@yahoo.com)
• In this work, a modified Genetic Algorithm is proposed.
S. Panda is with the National Institute of Science and Technology (NIST)
Berhampur, OR 760007 India (e-mail: panda_siddhartha@rediffmail.com).
One point crossover with modification is employed.
Positional dependency in respect of crossing site helps to
C. Ardil is with National Academy of Aviation, AZ1045, Baku, maintain diversity of search point as well as exploitation
Azerbaijan, Bina, 25th km, NAA (e-mail: cemalardil@gmail.com)
of already known optimum value. This makes a trade-off

717
 World Academy of Science, Engineering and Technology 63 2010

between exploration and exploitation of search space to • Case-B Four-area interconnected reheat thermal system
find global optimum in less number of generations. with governor dead band nonlinearity.
• The proposed GA along with decomposition technique as The physical connection of four-area interconnected power
developed has been used to obtain the optimum megawatt system has been depicted in Fig. 2. A typical block diagram of
frequency control of multi-area electric energy systems. a single area (kth area) perturbation model is shown in Fig. 3.
Time-domain simulations are conducted with trapezoidal
A. Sub System Model of Multi-Area Reheat Thermal System
integration along with decomposition technique.
• Since decomposition technique is used, there are no A mathematical model is to be developed to study the load
limitations in number of areas interconnected. The frequency dynamics of multi area interconnected reheat
solutions of state equations are found by trapezoidal thermal electric energy system using decomposition technique
integration method, which is more tolerant towards a [12].
large integration time interval. Therefore, there is a
considerable saving in computational time and core
memory.
• In this technique, care has been taken in the model to
include nonlinearity due to governor deadband.
This paper is organized as follows. The system model is
discussed in section two. In section three, trapezoidal
integration with decomposition technique is described. In
section four, an overview of proposed Genetic Algorithm for
function optimization is given. Then the results of simulation
study for four-area reheat thermal power station with and
without consideration nonlinearity due to governor deadband
are in section five. Finally some conclusions are drawn and
future works are suggested in section six. Fig. 3 Block Diagram of Single Area (kth Area) Perturbation
Model with Nonlinearity (Considering Deadband)
II. SYSTEM MODEL
Here the tie-line power deviations ΔPTiek can be assumed as
The system under consideration is exposed to small change
in load during its normal operation; so a linear model is an additional power disturbance to any area k. Accordingly,
sufficient for its dynamic representation. Fig. 1 shows the using decomposition technique the area wise sub system
transfer function model for single area representation of reheat equations are derived considering small load change as
thermal system. follows:
.
X1 = ΔF, X 2 = ΔX E , X 3 = ΔPG , X 4 = X 3 , X 5 = ΔPTie , X 6 = ΔPC .
Referring to Figure 3 of kth area, we can write

KP KP KP
Δ P D (t )
. 1
X 1 = − X 1 + X 3 − X 5 −
TP TP TP TP
. 1 1
X 2 = A 21 X 1 + X 2 + X 6
Tg Tg
Fig. 1 Transfer Function Model of Singe-Reheat Thermal System
.
X 3 = X 4
.
X 4 = A41 X 1 + A42 X 2 + A43 X 3 + A44 X 4 + A46 X 6
. M M
X 5 = 2π ∑ Tki X 1 − 2π ∑ Tki ΔFi
i =1 i =1
i≠K i≠ K
.
X 6 = − K Ik BX 1 − K Ik X 5
Referring to Figure 3 (a) for any area k, the above equations
can be written as:
Fig. 2 Physical Connection of Four-Area System
⎡ . ⎤
⎢⎣ X ⎥⎦ = [ A ]k [ X ]k + [U ]k (1)
k
In this case, the proposed algorithm is applied to the following
two cases: where
• Case-A Four-area interconnected reheat thermal system.

718
 World Academy of Science, Engineering and Technology 63 2010

T
⎡ . ⎤ ⎡ . . . . . .
⎤ which makes the system non-linear. The proposed method is
X
⎢⎣ ⎥⎦ = ⎢⎣ X 1 X 2 X3 X 4 X 5 X 6⎥ applied to investigate the effect of dead band on the dynamic
k ⎦k
performance. The magnitude of dead band is taken as 0.06 %.
For cases presented here the initial position of dead band is
⎡ 1 K p K p ⎤ selected so that the entire dead band of each area has to be
⎢ − 0 0 − 0 ⎥
⎢ T p Tp Tp ⎥ traversed before a response is secured. The system can be
⎢ 1 1 ⎥ anywhere within the dead band. The block diagram of single-
A − 0 0 0
⎢ Tg ⎥
21
Tg area (kth area) perturbation model with governor deadband is
⎢ ⎥
[ A ]k = ⎢ 0 0 0 1 0 0 ⎥ depicted in Figure 3.
⎢ A 41 A 42 A 43 A 44 0 A 46 ⎥ The factor ΔF varies rather slowly with time and so if the
⎢ M ⎥
⎢2π∑ T 0 0 0 0 0 ⎥ integration time interval is chosen sufficiently small, one can
⎢ i =1
ki
⎥ reasonably assume that during any particular time interval
⎢ i≠ k ⎥
each area operates entirely in side the dead band or outside it.
⎣⎢ − K I B 0 0 0 − K I 0 ⎦⎥ k
Therefore the system to be considered becomes essentially a
⎡ ⎤ piecewise linear system when governor dead band is included.
⎢ Kp M
[U] = ⎢− ΔPD (t) U2 (t) 0 U4 (t) − 2π∑TkiΔFi 0⎥⎥
T
k
For each area there will be two [A] Matrices, one for operation
Tp i=1
inside the dead band region and the other for operation outside
⎣⎢ i≠k ⎦⎥k it. They are derived as follows.

For linear model the value for A’s and U’s are as follows: Nonlinear model (Considering dead band DB)
1 , Kr ,
A21 = − A41 = − (i) Operation inside Dead Band:
RT g RT g Tt
There will be no signal proportional to frequency
⎛ 1 Kr ⎞ deviation. Therefore values of A’s and U’s are as
A 42 = ⎜ − ⎟
⎜TT T ⎟ below:
⎝ t r g Tt ⎠
A21 = 0, A41 = 0, U 2 (t ) = 0,
1 (T r + Tt ) , Kr U 4 (t ) = 0
A43 = − , A44 = − A46 = other A’s remain same as the values in linear case.
TrTt T r Tt T g Tt
U 2 (t ) = 0,U 4 (t ) = 0 (ii) Operation outside Dead Band:
When frequency deviation greater than the DB the
System equation for last area can be obtained as follows: signal will be proportional to ( ΔF -

M DB)sign( ΔF ). Therefore the value of A’s and U’s

Since ∑ ΔP Tiei =0 (2) are as follows:
1 , Kr
i =1
A 21 = − A41 = −
M −1 RT g RT g Tt
ΔPTieM = − ∑ ΔPTiei (3) DB
i =1 U 2 (t ) = sign ( Δ F ) ,
Making this substitution, one gets the A matrix for last area RT g
( ) ( )
(i.e. Mth area) to be of the order N − 1 × N − 1 . The state U 4 (t ) =
Kr
DBsign ( Δ F )
equation for last area is given below: RT g T t
The other A’s for both of the above cases are as
below:
⎛ 1 K r ⎞⎟ , 1
A42 = ⎜ − A43 = −
⎜TT ⎟
T g Tt ⎠ T r Tt
⎝ t r
(T + T t ) , A = K r
A 44 = − r 46
T r Tt T g Tt
The other values remain same as in linear model.

B. Subsystem Model of Mutli-Area Reheat Thermal System III. TRAPEZOIDAL INTEGRATION WITH DECOMPOSITION
with Deadband Nonlinerity TECHNIQUE
The speed governor dead band has a great effect on the Applying trapezoidal integration technique to the equation:
dynamic performance of electric energy system. For more
⎡ . ⎤
realistic analysis the governor dead band has to be included ⎢⎣ X ⎥⎦ = [ A ][ X ] + [U ]

719
 World Academy of Science, Engineering and Technology 63 2010

one gets IV. PROPOSED MODIFIED GA ALGORITHM

[X(t +Δt)] −[X(t)] =[A] [X(t +Δt)] +[X(t)] +[U(t +Δt)] +[U(t)] (4)
The following two modifications have been proposed
Δt 2 2 • Modification in parent selection
where • Modification in crossover mechanism
Δt integration time interval A. Proposed Parent Selection
X (t ) values of state variables at the end of time t Depending upon the values of fitness function, pairs of
X (t + Δt ) values of state variables at the end of time strings are selected from the population pool at random for

(t + Δt )
forming a mating pool. In a simple GA approach this is
termed as reproduction. And the strings are selected into the
On simplification, from equation (4), we obtain mating pool by simple Roulette wheel selection. In this
proposed algorithm, the following modifications are applied
⎡ Δt ⎤ ⎡ Δt ⎤ [U(t + Δt)] +[U(t)] Δt
⎢I − A 2 ⎥[X(t + Δt)] = ⎢I + A 2 ⎥[X(t)] +
for the selection of parents so that the strings with large values
(5)
of fitness are copied more into the mating pool.
⎣ ⎦ ⎣ ⎦ 2
• The first parent in each reproduction is the string having
[X (t + Δt )] = [SM ][Y ] (6)
the best fitness value. The second parent is selected from
the ordered population using normal selection technique.
where • At the ith reproduction, first parent is the best string of the
−1
⎡ Δt ⎤ population arranged in the order of fitness values. Second
[SM ] = ⎢I − A 2 ⎥ parent is selected from the ordered population using
⎣ ⎦
normal selection technique.

Δt ⎤
[X (t )] + [U (t + Δt )] + [U (t )] Δt
⎡ B. Proposed Crossover
and [Y ] = I + A
⎢ ⎥ Crossover is an algorithm for artificial mating of two
⎣ 2⎦ 2
individual chromosomes with an expectation that a
combination of genes of individuals of high fitness value may
When equation (6) is used with the decomposition technique,
produce an offspring with even higher fitness. It represents a
the procedure becomes iterative. The equation (6) can be
way of moving in the solution space based on the information
written as:
derived from the existing solutions. This makes exploitation
and exploration of information encoded in genes.
[X ( i +1)
(t + Δt )] = [SM]⎡⎢[X1(t )] + [U (i+1) (t + Δt ) + U (t )] Δt ⎤⎥
2⎦
(7) In this proposed algorithm, the following modifications
⎣ have been proposed with an intuition to have better trade-off
where between exploration of unknown solution space and
⎡ Δt ⎤ exploitation of already known knowledge of solution to find
X 1(t ) = ⎢ I + A [X (t )]
⎣ 2 ⎥⎦ the global optimum in less number of generations. In this
work, one point crossover also called Holland crossover is
and i denotes the iteration count.
Pc ∈ [0.6, 0.95] with
From equation (7) it can be seen that X (t + Δt ) consists
(i +1) adopted with a probability
modifications in exchange of chromosomal materials.
of two parts, one is a fixed part and another is a variable part, In a binary coded chromosome if the value of right most
which varies in every iteration. Equation (7) can be written as:
bits is changed [1 → 0, 0 → 1] , the search point in the

X (i +1) (t + Δt ) = X 2(t + Δt ) + X 3(i +1) (t + Δt )

search space shifts to a nearby point. This helps in refining the
(8) optimum point in the already known search space. As one
proceeds towards the left from the right most bit of the
where X2 is the fixed part and given by chromosome, the shifting of search point in the search space
increases and it depends on the position of the bit in the

[X 2(t + Δt )] = [SM ]⎡⎢[X 1(t )] + [U (t )] Δt ⎤⎥

chromosome whose value is changed. The shifting is highest
with the change in the left most bit. This facilitates to explore
⎣ 2⎦ new region in the search space by shifting the search point
wide apart from the current optimum position in the search
and X3 is the variable part and given by space.

[X 3( i +1 )
(t + Δ t )] = [SM ][U (i +1 ) (t + Δ t )] Δ t .
2

It is observed that it will be enough to iterate on X3 only.

Once convergence is obtained, X (t + Δt ) can be determined
as X2+X3. Fig. 4 (a) Example of Exploration and Exploitation in the Search
Space

720
 World Academy of Science, Engineering and Technology 63 2010

Therefore, it is evident that the exploitation of already V. If the search goal is achieved, or an allowable generation
known region or exploration of unknown region in the search is attained, stop and return, else go to third step (III).
space is relatively depending upon the position of the bit in
the chromosome whose value changes. In a chromosome // Procedure Proposed Crossover
change in the bits towards the right from the middle position // n Æ population size
contribute more towards the exploitation of already known // cs Æ crossover site
region. Similarly, change in the bits towards the left from the // l Æ length of chromosome
middle position contributes more towards exploration of new // midl Æ midpoint of chromosome length
region in the search space. This is shown in the Fig. 4(a).
Thus the positional dependency of crossing site in respect begin
of middle point of the chromosome helps to maintain diversity //Selection of two chromosomes
of the search point as well as improve the value of already Chromo (cn) = best chromo of pop(n)
known optimum value. Chromo (cn-1) = Roulette wheel pop(n-1)
Here the mechanism of crossover is not same as that of one for i=1 to 2 do
point crossover. In this proposed scheme, the exchange of begin
chromosomal material between two parents is made //Selection of crossover site
cs → rand (1,l)
considering the position of crossover site with respect to the
mid point of the chromosome. If the crossover site falls
if (cs towards right midl)
towards the right of the mid point of the chromosome, the
offspringm = chromo (cn) upto cs + chromo (cn-1)
right side chromosomal material from the crossover site of the
after cs
fitter parent is replaced with that of other parent to form one
offspring. If the crossover site falls towards the left of the else
offspring = chromo (cn-1) upto cs +chromo (cn)
middle position of the chromosome, the left side chromosomal
material from the crossover site of the fitter parent is replaced after cs
with the other parent to form one offspring. end
Fig. 4(b) shows an example of crossover procedure. Thus by end
generating one random number, only one offspring is Fig. 4 (c) The Pseudo Code for the Proposed Crossover
produced by crossover. For each pair of parent, two random
numbers are generated to produce two offspring. The pseudo V. SIMULATION RESULTS
code for the proposed crossover is shown in Fig. 4(c). To demonstrate the effectiveness of the proposed algorithm,
some simulations have been carried out for a four-area
interconnected AGC power system structured as in Fig. 2.
Simulations for the proposed method are carried out on a
computer with following specifications: Pentium-4 processor,
1.89 GHz.. The system parameters are tabulated in the
Appendix-A. In this part of the study, a conventional AGC,
which is only integral is considered. The parameters involved
in the feedback are the integral controller (KIk) and the
frequency bias constant (Bk). The optimal values of these
parameters depend upon the cost function used for
optimization. Each individual in the initial population has an
associated performance index (PI) value. The performance
Fig. 4 (b) Example of Crossover Procedure indices [14] used here are of the form:
I. The integral of the square of the error criterion (ISE). It is
C. Algorithm Steps for Optimization of AGC Parameters given be
With the above descriptions, the procedure of a proposed ∞
⎛ M −1 M
⎞
genetic algorithm for power system AGC is given as follows: ISE = ∫ ⎜ ∑ ΔPTie2 k + ∑ Δf k2 ⎟dt
I. Generate randomly a population of parameter strings to 0 ⎝ k =1 k =1 ⎠ (9)
form parameter vector.
II. The integral of time-multiplied absolute value of the error
II. Using digital simulation of system model, to calculate the
criterion (ITAE). The criterion penalizes long-duration
fitness (ISE and ITAE) for each string in the population.
transients and is much more selective than the ISE. A
III. Create offspring strings by proposed GA operators: system designed by use of this criterion exhibits small
reproduction, crossover and mutation. overshoot and well damped oscillations. It is given by
∞
IV. Evaluate the new strings and calculate the fitness for each ⎛ M −1 M
⎞
string. ITAE = ∫ t ⎜ ∑ ΔPTiek + ∑ Δf k ⎟dt
0 ⎝ k =1 k =1 ⎠ .. (10)

721
 World Academy of Science, Engineering and Technology 63 2010

To simplify the analysis, the four interconnected areas were obtaining optimal AGC parameters of interconnected power
considered identical. The optimal parameter value is such that system.
KI1=KI2=KI3=KI4=KI and B1 = B2 = B3 =B4 = B
The optimum values of the parameters KI and B for two 0.03

different test cases corresponding to the performance indices 0.02

considered, as obtained using proposed Genetic Algorithm

frequency deviation, Hz.

(GA), are summarized in Table-1(a) and Table-1(b) 0.01

respectively. 0.00

TABLE I (A) OPTIMUM VALUES OF KI & B (CASE-A) -0.01

Parameter ISE ITAE -0.02 with ITAE

B 0.401 0.392 with ISE

KI 0.706 0.807 -0.03

PI 0.546 2.346 -0.04

0 5 10 15 20 25 30

time, s
TABLE II (B) OPTIMUM VALUES OF KI & B (CASE-B)
Fig. 6 (a) Frequency deviation in Area-1
Parameter ISE ITAE
B 0.340 0.321 0.010
KI 0.861 0.640 0.008

PI 0.554 2.236 0.006

frequency deviation, Hz.

0.004

1.0 0.002

0.000
t ra d it io n a l G A
0.9 p ro p o s e d G A -0.002
Objective Function (ISE)

-0.004
with ITAE
0.8 -0.006
with ISE
-0.008

0.7 -0.010

-0.012
0 5 10 15 20 25 30
0.6
time, s

0.5
Fig. 6 (b) Frequency deviation in Area-3
0 10 20 30 40 50

G e ne ra ti o n num b e r 0.002

Fig. 5(a) Objective Function (ISE) vs. Generation Number 0.000

tie-line power deviation, p.u.

-0.002
1.6

t ra d it io n a l G A
1.5 -0.004
p ro p o s e d G A
Objective Function (ITAE)

with ITAE
with ISE
1.4 -0.006

1.3 -0.008
0 5 10 15 20 25 30

tim e, s
1.2
Fig. 6 (c) Tie-line power deviation in Area-1
1.1
0 10 20 30 40 50
0.0025
G e ne ra ti o n num b e r
0.0020
tie-line power deviation, p.u.

Fig. 5(b) Objective Function (ITAE) vs. Generation Number 0.0015

with ITAE
with ISE

0.0010

Fig. 5(a) shows the profile of objective function (ISE) 0.0005

versus generation number. It is revealed from the figure that
the ISE value is smaller for the same generation number with 0.0000

the proposed GA as compared to traditional GA. Similarly; it -0.0005

is also observed from the Fig. 5(b) that the value of ITAE is
-0.0010
smaller at the same generation number with the proposed GA 0 5 10 15 20 25 30

compared to the traditional GA. As it is evident that the time, s

proposed GA works better in comparison to traditional GA for Fig. 6 (d) Tie-line power deviation in Area-3

722
 World Academy of Science, Engineering and Technology 63 2010

0.012 0 .0 3

0.010 0 .0 2
generated power deviation, p.u.

frequency deviation, Hz.

0.008 0 .0 1

0 .0 0
0.006

-0 .0 1
0.004
with ITAE
with ISE -0 .0 2 w ith IS E
0.002
w ith IT A E
-0 .0 3
0.000

-0 .0 4
0 5 10 15 20 25 30
0 5 10 15 20 25 30
tim e , s
tim e, s

Fig. 6 (e) Generated power deviation in Area-1 Fig. 7 (a) Frequency deviation in area-1

0.003
0.010

w ith ITA E
generated power deviation, p.u.

0.002 w ith IS E
0.005

frequency deviation, Hz.

0.001

0.000

0.000

-0.005

-0.001
with ISE
with ITAE
-0.010

-0.002
0 5 10 15 20 25 30
-0.015
tim e, s
0 5 10 15 20 25 30
Fig. 6 (f) Generated power deviation in Area-3 tim e, s

Fig. 7 (b) Frequency deviation in area-3

Fig. 6 Response of Four-area Reheat Thermal System with Optimal
AGC Parameters (Case-A). Dotted curve corresponding to ISE and
solid curve corresponding to ITAE. 0.002

Fig. 6 shows the dynamic response of four-area 0.000

tie-line power deviation, p.u.

interconnected power system corresponding to Table-1(a).

Each figure contains two curves corresponding to two -0.002

different performance indices. In Fig. 6, the dotted curves

show the responses with the performance index ISE, whereas -0.004

the solid curves show the responses with the performance with IS E
with ITA E
index ITAE. The response obtained when the parameters are -0.006

set according to the ITAE indicate that the damping of

oscillation is much improved and the transient error in -0.008
0 5 10 15 20 25 30
frequency, tie-line power and generated power is also much tim e , s
reduced. Table-2(a) shows transient overshoot with different Fig. 7 (c) Tie-line power deviation in area-1
performance indices for Test Case A.
TABLE II(A) OVERSHOOT WITH DIFFERENT PERFORMANCE INDICES FOR TEST
CASE A
0.0030
Area Nature of Maximum
0.0025
no. response Overshoot/Undershoot
tie-line power deviation, p.u.

with ISE
corresponding to 0.0020
with ITAE

performance index (PI) 0.0015

Area-1 ISE ITAE 0.0010

Δf1 0.0225 Hz. 0.0185 Hz.
ΔPTie1 0.0073 p.u. 0.0073 p.u.
0.0005

ΔPG1 0.0118 p.u. 0.010 p.u.

0.0000

-0.0005

Δf3
-0.0010
Area-3 0.011 Hz 0.010 Hz. 0 5 10 15 20 25 30

ΔPTie3 0.0021 p.u. 0.0021 p.u. time, s

ΔPG3 0.0036 p.u. 0.0025 p.u. Fig. 7 (d) Tie-line power deviation in area-3

723
 World Academy of Science, Engineering and Technology 63 2010

when the control parameters are set based on minimizing the

0.012 ITAE is clearly shown. The numerical results of the transient
0.010
overshoot with respect to different performance indices are
presented in Table-2(b). It is clearly evident from this table
generated power deviation, p.u.

0.008 that the performance of the system with the proposed GA has
0.006
been improved a lot.

0.004
VI. CONCLUSION
0.002 with ISE
with ITAE
The application of the proposed method to the AGC
0.000
problem reveals that the system performance is highly
improved. The results of this proposed algorithm when
0 5 10 15 20 25 30
compared with the results of other investigators derived in
time, s different methods indicate its correctness and effectiveness in
Fig. 7 (e) Generated power deviation in area-1 finding the optimal AGC parameters of multi-area systems.

0.003 APPENDIX
Appendix-A
generated power deviation, p.u.

0.002 with IS E
with ITA E The nominal system parameters are
0.001
f = 60 Hz, Tg = 0.08 Sec, Tr = 10.0 Sec, Hk = 5.0 Sec, Kr=
0.5, Tt = 0.3 Sec, 2πTki = 0.05
0.000

-0.001

REFERENCES
-0.002
0 5 10 15 20 25 30 [1] J. Wood, B. F. Woolenberg, Power Generation Operation and Control,
time, s John Wiley and Sons, 1984.
Fig. 7 (f) Generated power deviation in Area-3 [2] O. l. Elgerd, Electric energy Systems Theory – An Introduction,
Fig. 7 Dynamic response of Four-area Reheat Thermal System with McGraw Hill Co., 2001.
[3] J. Nanda and B. Kaul, “Automatic generation control of an
Optimal AGC Parameters (Case-B). Dotted curve corresponding to
interconnected power system”, IEE Proc., Vol. 125, no. 5, May 1978,
ISE and solid curve corresponding to ITAE. pp. 385-390.
[4] O. I. Elgerd and C. E. Fosha, ., “Optimum megawatt-frequency control
TABLE II(B) OVERSHOOT WITH DIFFERENT PERFORMANCE INDICES FOR TEST of multi-area electric energy systems”, IEEE Trans. On PAS, vol. PAS-
CASE B 89, no. 4, April 1970, pp. 556-563.
Area no. Nature Maximum [5] M. Aldden and H. Trinh, “Load frequency control of interconnected
power systems via constrained feedback control schemes”, Computer
of Overshoot/Undershoot and Electrical Engineering, Vol. 20, no. 1, 1994, pp. 71-88.
respons corresponding to [6] H.J. Kunish, K.G. Kramer and H. D. Dominik, “Battery energy storage –
e performance index (PI) Another option for load-frequency control and instantaneous reserve”,
ISE ITAE IEEE Trans. Energy Convers., vol. EC-1, no. 3, Sept. 1986, pp. 46-51.
[7] S. Banerjee, J.K. Chatterjee and S.C. Tripathy, “Application of magnetic
Δf1 0.0315 Hz. 0.03 Hz. energy storage unit as load frequency stabilizer”, IEEE Trans. Energy
Area-1 ΔPTie1 0.0016 p.u. 0.0014 p.u. Convers., vol. 5, no. 1, March 1990, pp. 46-51.
ΔPG1 0.0118 p.u. 0.0099 p.u. [8] Y.L. Karnavas and D.P. Papadopoulos,”AGC for autonomous power
system using combined intelligent techniques”, Electric Power System
Research, 62 (2002), 225-239.
[9] Y.L. Abdel-Magid and M.M. Dawound, “Optimal AGC tuning with
Δf3 0.008 Hz 0.005 Hz. genetic algorithms”, Electric Power System Research, 38 (1997), 231-
Area-3 ΔPTie3 0.0022 p.u. 0.0023 p.u. 238.
ΔPG3 0.0025 p.u. 0.0025 p.u. [10] S.K. Aditya and D. Das, “Design of load frequency controllers using
genetic algorithm for two-area interconnected hydro power system”,
Elect. Power Compon. Syst., vol. 31, no. 1, Jan. 2003, pp. 81-94.
There are three variables plotted out in the Fig. 7. First one [11] Z. M. Al-Hamouz, and H. N. Al-Dowaish, “A New Load Frequency
is the frequency variation of area-1 and area-3, second one is Variable Structure Controller using Genetic Algorithms”, Elect. Power
AGC tie-line power error of area-1 and area-3, and third one Syst. Res., vol. 55, no. 1, July 2000, pp. 1-6.
is Generated power deviayion in area-1 and area-3. Each [12] D. Rerkpreedapong, A. Hasanovic and A. Feliachi, “Robust Load
Frequency Control using Genetic Algorithms and Linear Matrix
figure contains two curves corresponding to two different Inequalities”, IEEE Trans. Power Syst., vol. 18, no. 2, May 2003, pp.
performance indices. In Fig. 7, the dotted curves show the 855-861.
responses with the performance index ITAE, whereas the [13] A.M. Panda, “Automatic generation control of multi area interconnected
solid curves show the responses with the performance index power system considering nonlinearity due to governor deadband”,
Archives of control sciences, Vol. 7(XLIII), no. 3-4, March 1998, pp.
ISE. It is seen that the deadband tends to produce continuous 285-299.
sinusoidal oscillation whereas without deadband the responses [14] W. Chan and Y. Hsu, “Optimal control of power generation using
are more damped. The superiority of the responses obtained variable structure controllers”, Elect. Power Syst. Res.”, vol. 6, 1983, pp.
269-276.

724
 World Academy of Science, Engineering and Technology 63 2010

Gayadhar Panda is working as an Assistant

Professor in the department of Electrical
Engineering, IGIT, Sarang, Orissa, India. He was
born in Orissa, India, in 1970. He received the
B.E. degree in electrical engineering from the
Institution of engineers (I), the M.E. degree in
electrical engineering from Bengal Engineering
College, Shibpur, Howarh, India, and the Ph.D.
degree in electrical engineering from Utkal
University, Orissa, India, in 1996, 1998, and
2007, respectively.
From 1999 to 2008 he was with the Department of Electrical and Electronics
Engineering, Faculty of Engineering, National Institute of Science and
Technology, Berhampur, Orissa, India where he is currently a Professor. His
research interests are in the areas of stability aspects of power transmission,
power electronics, and development of renewable energy. Dr. G. Panda is a
member of the IE and ISTE of India.

Sidhartha Panda is working as a Professor at

National Institute of Science and Technology
(NIST), Berhampur, Orissa, India. He received the
Ph.D. degree from Indian Institute of Technology,
Roorkee, India in 2008, M.E. degree in Power
Systems Engineering from UCE, Burla in 2001 and
B.E. degree in Electrical Engineering in 1991.
Earlier, he worked as an Associate Professor in
KIIT Deemed University and also in various other
engineering colleges for about 15 years.
He has published above 50 papers in various International Journals and acting
as a reviewer for some IEEE and Elsevier Journals. His biography has been
included in Marquis' “Who's Who in the World” USA, for 2010 edition. His
areas of research include power system transient stability, power system
dynamic stability, FACTS, optimization techniques, model order reduction,
distributed generation, image processing and wind energy.

C. Ardil is with National Academy of Aviation, AZ1045, Baku, Azerbaijan,

Bina, 25th km, NAA

725