Electric Power Systems Research 81 (2011) 458464

Contents lists available at ScienceDirect

Electric Power Systems Research
j our nal homepage: www. el sevi er . com/ l ocat e/ epsr
Differential evolution algorithm for optimal reactive power dispatch
A.A. Abou El Ela
a
, M.A. Abido
b
, S.R. Spea
a,
a
Electrical Engineering Department, Faculty of Engineering, Menouya University, Egypt
b
Electrical Engineering Department, King Fahd University of Petroleum and Minerals, Saudi Arabia
a r t i c l e i n f o
Article history:
Received 21 August 2009
Received in revised form 12 August 2010
Accepted 4 October 2010
Available online 30 October 2010
Keywords:
Reactive power dispatch
Differential Evolution algorithm
Fuel cost minimization
Voltage prole improvement
Voltage stability enhancement
a b s t r a c t
Reactive power dispatch(RPD) is one of the important tasks inthe operationandcontrol of power system.
This paper presents an efcient and reliable evolutionary-based approach to solve the RPD problem. The
proposed approach employs differential evolution (DE) algorithm for optimal settings of RPD control
variables. The proposed approach is examined and tested on the standard IEEE 30-bus test system with
different objectives that reect power losses minimization, voltage prole improvement, and voltage
stabilityenhancement. Thesimulationresults of theproposedapproacharecomparedtothosereportedin
the literature. The results demonstrate the potential of the proposed approach and showits effectiveness
and robustness to solve the RPD problem.
Crown Copyright 2010 Published by Elsevier B.V. All rights reserved.
1. Introduction
The purpose of the reactive power dispatch (RPD) in power sys-
tem is to identify the control variables which minimize the given
objective function while satisfying the unit and systemconstraints.
This goal is achieved by proper adjustment of reactive power vari-
ables like generator voltage magnitudes, switchable VAR sources
and transformer tap setting [1].
Inthe past two decades, the problemof RPDfor improving econ-
omy and security of power system operation has received much
attention. The main objective of optimal reactive power control is
to improve the voltage prole and minimizing system real power
losses via redistribution of reactive power in the system. In addi-
tion, the voltage stability can be enhanced by reallocating reactive
power generations. Therefore, the problem of the RPD can be opti-
mized to enhance the voltage stability, improve voltage prole and
minimize the system losses as well [24].
To solve the RPD problem, a number of conventional opti-
mization techniques [5,6] have been proposed. These include
the Gradient method, Non-linear Programming (NLP), Quadratic
Programming (QP), Linear programming (LP) and Interior point
method. Though these techniques have been successfully applied
for solving the reactive power dispatch problem, still some dif-
culties are associated with them. One of the difculties is the
multimodal characteristic of the problems to be handled. Also, due
to the non-differential, non-linearity and non-convex nature of the

Corresponding author.
E-mail address: shi spea@yahoo.com (S.R. Spea).
RPD problem, majority of the techniques converge to a local opti-
mum. Recently, Evolutionary Computation techniques like Genetic
Algorithm (GA) [7], Evolutionary Programming (EP) [8] and Evolu-
tionary Strategy [9] have beenappliedto solve the optimal dispatch
problem. In this paper, a newevolutionary computation technique,
called Differential Evolution (DE) algorithm is used to solve RPD
problem.
Recently, differential evolution (DE) algorithm has been pro-
posed and introduced [1013]. The algorithm is inspired by
biological and sociological motivations and can take care of opti-
mality on rough, discontinuous and multi-modal surfaces. The DE
has threemainadvantages: it canndnear optimal solutionregard-
less the initial parameter values, its convergence is fast and it uses
fewnumber of control parameters. In addition, DE is simple in cod-
ing and easy to use. It can handle integer and discrete optimization
[1013].
The performance of DE algorithm was compared to that of dif-
ferent heuristic techniques. It is found that, the convergence speed
of DE is signicantly better than that of GAs [12]. In [14], the per-
formance of DE was compared to PSO and evolutionary algorithms
(EAs). The comparison is performed on a suite of 34 widely used
benchmark problems. It is found that, DE is the best perform-
ing algorithm as it nds the lowest tness value for most of the
problems considered in that study. Also, DE is robust; it is able to
reproduce the same results consistently over many trials, whereas
the performance of PSO is far more dependent on the randomized
initialization of the individuals [14]. In addition, the DE algorithm
has been used to solve high-dimensional function optimization (up
to 1000 dimensions) [15]. It is found that, it has superior perfor-
mance on a set of widely used benchmark functions. Therefore, the
0378-7796/$ see front matter. Crown Copyright 2010 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2010.10.005
A.A.A.E. Ela et al. / Electric Power Systems Research 81 (2011) 458464 459
DE algorithm seems to be a promising approach for engineering
optimizationproblems. It has successfullybeenappliedandstudied
to many articial and real optimization problems [1620].
In [2123], the DE algorithm is used as an optimization tool for
the reactive power optimization with the propose of minimizing
the system power losses while maintaining the dependant vari-
ables including voltages of PQ-buses and reactive power outputs of
generators, within limits. In [24], the DE algorithm is used to solve
reactive power dispatch for voltage stability enhancement.
Inthis paper, a novel DE-basedapproachis proposedtosolve the
RPD problem. The problem is formulated as a nonlinear optimiza-
tionproblemwithequality andinequality constraints. Inthis study,
different objectives are considered such as minimizing the power
losses, improving the voltage prole, and enhancing power sys-
tem voltage stability. The proposed approach has been examined
and tested on the standard IEEE 30-bus test system. The poten-
tial and effectiveness of the proposed approach are demonstrated.
Additionally, the results are compared to those reported in the
literature.
2. Problem formulation
The objective of RPD is to identify the reactive power control
variables, which minimizes the objective functions. This is mathe-
matically stated as follows:
2.1. Problem objectives
In this study, the following objectives are considered:
2.1.1. Minimization of system power losses
The minimization of system real power losses Ploss (MW) can
be calculated as follows:
f
1
= Ploss =
nl

k=1
g
k
[V
2
i
+V
2
j
2V
i
V
j
cos(
i

j
)] (1)
where nl is the number of transmission lines; g
k
is the conductance
of the kth line; V
i
and V
j
are the voltage magnitude at the end buses
i and j of the kth line, respectively, and
i
and
j
are the voltage
phase angle at the end buses i and j.
2.1.2. Voltage prole improvement
Bus voltage is one of the most important security and service
quality indices. Improving voltage prole can be obtained by min-
imizing the load bus voltage deviations from 1.0 per unit. The
objective function can be expressed as:
f
2
=

i NL

V
i
1.0

(2)
where NL is the number of load buses.
2.1.3. Voltage stability enhancement
It is very important to maintain constantly acceptable bus volt-
age at each bus under normal operating conditions, after load
increase, following systemconguration changes, or when the sys-
temis being subjected to a disturbance. The non-optimized control
variables may lead to progressive and uncontrollable drop in volt-
age resulting in an eventual widespread voltage collapse.
Enhancing voltage stability can be achieved through minimiz-
ing the voltage stability indicator L-index values at every bus of the
system and consequently the global power system L-index [25].
L-index gives a scalar number to each load bus. This index uses
information on a normal power ow. L-index is in the range of zero
(no load case) and one (voltage collapse). Details of L-index calcula-
tionandderivationare giveninAppendix A. Inorder to enhance the
voltage stability and move the systemfar fromthe voltage collapse
point, the following objective function can be used:
f
3
= L
max
(3)
2.2. System constraints
2.2.1. Equality constraints
These constraints represent load ow equations:
P
Gi
P
Di
V
i
NB

j=1
V
j
[G
ij
cos(
i

j
) +B
ij
sin(
i

j
)] = 0 (4)
Q
Gi
Q
Di
V
i
NB

j=1
V
j
[G
ij
sin(
i

j
) B
ij
cos(
i

j
)] = 0 (5)
where i =1,. . .,NB; NB is the number of buses, P
G
is the active power
generated, Q
G
is the reactive power generated, P
D
is the load active
power, Q
D
is the load reactive power, G
ij
and B
ij
are the transfer
conductance and susceptance betweenbus i and bus j, respectively.
2.2.2. Inequality constraints
These constraints include:
1. Generator constraints: generator voltages, and reactive power
outputs are restricted by their lower and upper limits as follows:
V
min
Gi
V
Gi
V
max
Gi
, i = 1, . . . , NG (6)
Q
min
Gi
Q
Gi
Q
max
Gi
, i = 1, . . . , NG (7)
2. Transformer constraints: transformer tap settings are bounded
as follows:
T
min
i
T
i
T
max
i
, i = 1, . . . , NT (8)
3. Shunt VAR constraints: shunt VAR compensations are restricted
by their limits as follows:
Q
min
ci
Q
ci
Q
max
ci
, i = 1, . . . , N
C
(9)
Security constraints: these include the constraints of voltages at
load buses and transmission line loadings as follows:
V
min
Li
V
Li
V
max
Li
, i = 1, . . . , NL (10)
S
li
S
max
li
, i = 1, . . . , nl (11)
3. Differential evolution algorithm
3.1. Overview
In 1995, Storn and Price proposed a newoating point encoded
evolutionary algorithm for global optimization and named it dif-
ferential evolution (DE) algorithm owing to a special kind of
differential operator, which they invoked to create new off-spring
from parent chromosomes instead of classical crossover or muta-
tion [10].
Similar to GAs, DE algorithm is a population based algorithm
that uses crossover, mutation and selection operators. The main
differences between the genetic algorithm and DE algorithm are
the selection process and the mutation scheme that makes DE self
adaptive. InDE, all solutions have the same chance of being selected
as parents. DE employs a greedy selection process that is the best
newsolution and its parent wins the competition providing signi-
cant advantageof convergingperformanceover genetic algorithms.
460 A.A.A.E. Ela et al. / Electric Power Systems Research 81 (2011) 458464
Initialization of Chromosomes
Mutation Differential Operator
Crossover
Selection
Fig. 1. DE cycle of stages.
3.2. DE computational ow
DE algorithm is a population based algorithm using three oper-
ators; crossover, mutation and selection. Several optimization
parameters must also be tuned. These parameters have joined
together under the common name control parameters. In fact, there
are only three real control parameters in the algorithm, which are
differentiation (or mutation) constant F, crossover constant CR, and
size of population NP. The rest of the parameters are dimension of
problem D that scales the difculty of the optimization task; max-
imumnumber of generations (or iterations) GEN, which may serve
as a stopping condition; and low and high boundary constraints of
variables that limit the feasible area [10,11].
The proper setting of NP is largely dependent on the size of
the problem. Storn and Price [10] remarked that for real-world
engineering problems with D control variables, NP=20D will prob-
ably be more than adequate, NP as small as 5D is often possible,
although optimal solutions using NP<2D should not be expected.
In [13], Storn and Price set the size of population less than the
recommended NP=10D in many of their test tasks. In [14], it is
recommended using of NP4D. In [15], NP=5D is a good choice
for a rst try, and then increase or decrease it by discretion. So, as
a rough principle, several tries before solving the problem may be
sufcient to choose the suitable number of the individuals.
The DE algorithm works through a simple cycle of stages, pre-
sented in Fig. 1.
These stages can be cleared as follow:
3.2.1. Initialization
At the very beginning of a DE run, problem independent vari-
ables are initialized in their feasible numerical range. Therefore,
if the jth variable of the given problem has its lower and upper
bound as x
L
j
and x
u
j
, respectively, then the jth component of the
ith population members may be initialized as,
x
i,j
(0) = x
L
j
+rand(0, 1) (x
u
j
x
L
j
) (12)
where rand(0,1) is a uniformly distributed random number
between 0 and 1.
3.2.2. Mutation
In each generation to change each population member

X
i
(t), a
donor vector v
i
(t) is created. It is the method of creating this donor
vector, which demarcates between the various DE schemes. How-
ever, in this paper, one such specic mutation strategy known as
DE/rand/1 is discussed.
To create a donor vector v
i
(t) for each ith member, three param-
eter vectors x
r1
, x
r2
and x
r3
are chosen randomly from the current
population and not coinciding with the current x
i
. Next, a scalar
number F scales the difference of any two of the three vectors and
the scaled difference is added to the third one whence the donor
vector v
i
(t) is obtained. The usual choice for F is a number between
0.4 and 1.0. So, the process for the jth component of each vector
can be expressed as,
v
i,j
(t +1) = x
r1,j
(t) +F (x
r2,j
(t) x
r3,j
(t)) (13)
3.2.3. Crossover
To increase the diversity of the population, crossover operator
is carried out in which the donor vector exchanges its components
with those of the current member

X
i
(t).
Two types of crossover schemes can be used with DE technique.
These are exponential crossover and binomial crossover. Although
the exponential crossover was proposed in the original work of
Storn and Price [10], the binomial variant was much more used in
recent applications [14].
In exponential type, the crossover is performed on the D vari-
ables in one loop as far as it is within the CR bound. The rst time
a randomly picked number between 0 and 1 goes beyond the CR
value, no crossover is performed and the remaining variables are
left intact. In binomial type, the crossover is performed on all D
variables as far as a randomly picked number between 0 and 1 is
within the CR value. So for high values of CR, the exponential and
binomial crossovers yield similar results.
Moreover, in the case of exponential crossover one has to be
aware of the fact that there is a small range of CR values (typically
[0.9, 1]) to which the DE is sensitive. This could explain the rule of
thumb derived for the original variant of DE. On the other hand, for
the same value of CR, the exponential variant needs a larger value
for thescalingparameter F inorder toavoidprematureconvergence
[26].
In this paper, binomial crossover scheme is used which is per-
formed on all D variables and can be expressed as:
u
i,j
(t) =

v
i,j
(t) if rand(0, 1) < CR
x
i,j
(t) else
(14)
u
i,j
(t) represents the child that will compete with the parent x
i,j
(t).
3.2.4. Selection
To keep the population size constant over subsequent genera-
tions, the selection process is carried out to determine which one of
the child and the parent will survive in the next generation, i.e., at
time t =t +1. DE actually involves the Survival of the ttest princi-
ple in its selection process. The selection process can be expressed
as,

X
i
(t +1) =

U
i
(t) if f (

U
i
(t)) f (

X
i
(t))

X
i
(t) if f (

X
i
(t)) < f (

U(t))
(15)
where, f () is the functionto be minimized. FromEq. (10) we noticed
that:

If u
i
(t) yields a better value of the tness function, it replaces its
target

X
i
(t) in the next generation.

Otherwise,

X
i
(t) is retained in the population.
Hence, the population either gets better in terms of the tness
function or remains constant but never deteriorates.
3.3. DE-based approach implementation
The proposed DE-based approach has been developed and
implemented using the MATLAB software. Several runs have been
done with different values of DE key parameters such as differ-
entiation (or mutation) constant F, crossover constant CR, size of
population NP, and maximumnumber of generations GEN which is
used here as a stopping criteria to nd the optimal DE key param-
eters. In this paper, the following values of DE key parameters are
A.A.A.E. Ela et al. / Electric Power Systems Research 81 (2011) 458464 461
selected for the optimization of power losses and voltage stability
enhancement:
F = 0.2; CR = 0.6; NP = 150; GEN = 500
and DE key parameters for the optimization of voltage deviations
are selected as:
F = 0.2; CR = 0.6; NP = 50; GEN = 500
The rst step in the algorithm is creating an initial population.
All the independent variables which include generator voltages,
transformer tap settings and shunt VAR compensations have to be
generated according to Eq. (12), where each independent parame-
ter of each individual in the population is assigned a value inside
its given feasible region. This creates parent vectors of independent
variables for the rst generation.
After, nding the independent variables, dependent variables
will be found froma load owsolution. These dependent variables
includegenerators reactivepower, voltages at loadbuses andtrans-
mission line loadings. It should be mentioned that, the real power
settings of the generators are taken from [4].
4. Results and discussion
The proposed DE-based algorithm has been tested on the stan-
dard IEEE 6-generator 30-bus test system shown in Fig. 2. The
systemdata is given in Appendix B [27]. This systemhas 19-control
variable as follows: 6-generator voltage magnitude, 4-tap trans-
former setting and 9-switchable VAR.
To demonstrate the effectiveness of the proposed algorithm,
three different cases have been considered as follows:
Case1: Minimization of system power losses.
Case2: Improvement of voltage prole.
Case3: Enhancement of voltage stability.
4.1. Case1 (minimization of system power losses)
In the rst case, the proposed algorithm is run with minimiza-
tion of real power losses as the objective function. As mentioned
above, the real power settings of the generators are taken from[2].
The convergence characteristic of the algorithm is shown in Fig. 3.
The algorithmreaches a minimumloss of 4.5550MW. The optimal
29
30
27
28
25 26
24 23
19 18 15
20 17
21
22 16 14
10
6
9
11
1
2 5
7
8 4
12 13
3
Fig. 2. Single line diagram of IEEE 30-bus test system.
values of the control variables are given in the second column of
Table 1. The minimum loss obtained by the proposed algorithm is
compared with the results reported in [24] using the evolution-
ary computation techniques for the same test system. The results
of the comparison are given in Table 2. From the comparison, the
proposed algorithmgives the minimumlosses which demonstrate
the effectiveness of the proposed algorithm.
4.2. Case2 (improvement of voltage prole)
In the second case, the proposed DE-based approach is applied
for improvement of voltage prole. The convergence characteristic
Table 1
Optimal settings of control variables for different cases.
Case1: minimization of power losses Case2: voltage prole improvement Case3: voltage stability enhancement
V
1
1.1000 1.0100 1.0993
V
2
1.0931 0.9918 1.0967
V5 1.0736 1.0179 1.0990
V
8
1.0756 1.0183 1.0346
V
11
1.1000 1.0114 1.0993
V
13
1.1000 1.0282 0.9517
T
11
1.0465 1.0265 0.9038
T
12
0.9097 0.9038 0.9029
T
15
0.9867 1.0114 0.9002
T
36
0.9689 0.9635 0.9360
Q
c10
5.0000 4.9420 0.6854
Q
c12
5.0000 1.0885 4.7163
Q
c15
5.0000 4.9985 4.4931
Q
c17
5.0000 0.2393 4.5100
Q
c20
4.4060 4.9958 4.4766
Q
c21
5.0000 4.9075 4.6075
Q
c23
2.8004 4.9863 3.8806
Q
c24
5.0000 4.9663 4.2854
Q
c29
2.5979 2.2325 3.2541
Power losses (MW) 4.5550 6.4755 7.0733
Voltage deviations 1.9589 0.0911 1.4191
Lmax 0.5513 0.5734 0.1246
462 A.A.A.E. Ela et al. / Electric Power Systems Research 81 (2011) 458464
0 50 100 150 200 250 300 350 400 450 500
4.5
4.6
4.7
4.8
4.9
5
5.1
5.2
5.3
5.4
5.5
Generations
P
l
o
s
s
(
M
W
)
Fig. 3. Power losses variations of Case1.
Table 2
Comparison of power losses for different methods.
Method Ploss (MW)
Strength pareto evolutionary algorithm [2] 5.1170
Genetic algorithm [3] 4.5800
Genetic algorithm-based approach [4] 4.6501
Proposed algorithm 4.5550
of the algorithm for this case is shown in Fig. 4. The optimal values
of the control variable settings obtained in this case are given in
the third column of Table 1. In this case, the voltage deviations are
reducedfrom1.1606intheinitial stateto0.0911withareductionof
92.15%. The comparison with the results reported in [1] is given in
Table 3 where 79.11% reduction is achieved. From the comparison,
the proposed algorithmgives the best results for voltage deviations
which demonstrate the effectiveness of the proposed algorithm.
0 50 100 150 200 250 300 350 400 450 500
0.05
0.1
0.15
0.2
0.25
0.3
Generations
V
o
l
t
a
g
e

D
e
v
i
a
t
i
o
n
s
Fig. 4. Voltage deviations variations of Case3.
Table 3
Comparison of voltage deviations for different methods.
Method Voltage deviations
Particle swarm optimization [1] 0.2424
Proposed algorithm 0.0911
0 50 100 150 200 250 300 350 400 450 500
0.124
0.125
0.126
0.127
0.128
0.129
0.13
Generations
L
m
a
x
Fig. 5. Lmax variations of Case3.
Table 4
Comparison of Lmax value for different methods.
Method Lmax
Strength pareto evolutionary algorithm [2] 0.1397
Proposed algorithm 0.1246
4.3. Case3 (enhancement of voltage stability)
In the third case, the proposed DE-based approach is applied
for enhancement of voltage stability as the objective. The conver-
gence characteristic of the algorithmfor this case is shown in Fig. 5.
The optimal values of the control variable settings obtained in this
case are given in the fourth column of Table 1. In this case, the
maximum L-index of the system has been reduced from 0.2144
in the initial state to 0.1246. Hence, it is observed that there is
an increase in performance of the system. Thus it results in the
enhancement of voltage stability level of the system. The results of
the comparison with the results reported in the literature are given
in Table 4. From the comparison, the proposed algorithm gives the
best results for L
max
which demonstrates the effectiveness of the
proposed algorithm.
5. Conclusions
In this paper, a differential evolution (DE) optimization algo-
rithm has been proposed, developed, and successfully applied to
solve reactive power dispatch(RPD) problem. The RPDproblemhas
beenformulatedas a constrainedoptimizationproblemwhere sev-
eral objective functions have been considered to minimize power
losses, to improve the voltage prole, and to enhance the voltage
stability. The proposed approach has been tested and examined
on the standard IEEE 30-bus test system. The simulation results
demonstratetheeffectiveness androbustness of theproposedalgo-
rithm to solve RPD problem. Moreover, the results of the proposed
DE algorithm have been compared to those reported in the litera-
ture. The comparisonconrms the effectiveness andthe superiority
of the proposed DE approach over the classical and heuristic tech-
niques in terms of solution quality.
Acknowledgement
Dr. M.A. Abido would like to acknowledge the support of King
Fahd University of Petroleum & Minerals.
A.A.A.E. Ela et al. / Electric Power Systems Research 81 (2011) 458464 463
Appendix A. L-index calculation
For voltage stability evaluation, an indicator L-index is used. The
indicator value varies in the range between 0 (the no load case)
and 1 which corresponds to voltage collapse. The indicator uses
bus voltage and network information provided by the load ow
program.
Table A1
Load data.
Bus no. Load Bus no. Load
P (p.u.) Q (p.u.) P (p.u.) Q (p.u.)
1 0.000 0.000 16 0.035 0.018
2 0.217 0.127 17 0.090 0.058
3 0.024 0.012 18 0.032 0.009
4 0.076 0.016 19 0.095 0.034
5 0.942 0.190 20 0.022 0.007
6 0.000 0.000 21 0.175 0.112
7 0.228 0.109 22 0.000 0.000
8 0.300 0.300 23 0.032 0.016
9 0.000 0.000 24 0.087 0.067
10 0.058 0.020 25 0.000 0.000
11 0.000 0.000 26 0.035 0.023
12 0.112 0.075 27 0.000 0.000
13 0.000 0.000 28 0.000 0.000
14 0.062 0.016 29 0.024 0.009
15 0.082 0.025 30 0.106 0.019
Table A2
Line data.
Line no. From bus To bus Line impedance
R (p.u.) X (p.u.)
1 1 2 0.0192 0.0575
2 1 3 0.0452 0.1852
3 2 4 0.0570 0.1737
4 3 4 0.0132 0.0379
5 2 5 0.0472 0.1983
6 2 6 0.0581 0.1763
7 4 6 0.0119 0.0414
8 5 7 0.0460 0.1160
9 6 7 0.0267 0.0820
10 6 8 0.0120 0.0420
11 6 9 0.0000 0.2080
12 6 10 0.0000 0.5560
13 9 11 0.0000 0.2080
14 9 10 0.0000 0.1100
15 4 12 0.0000 0.2560
16 12 13 0.0000 0.1400
17 12 14 0.1231 0.2559
18 12 15 0.0662 0.1304
19 12 16 0.0945 0.1987
20 14 15 0.2210 0.1997
21 16 17 0.0824 0.1932
22 15 18 0.1070 0.2185
23 18 19 0.0639 0.1292
24 19 20 0.0340 0.0680
25 10 20 0.0936 0.2090
26 10 17 0.0324 0.0845
27 10 21 0.0348 0.0749
28 10 22 0.0727 0.1499
29 21 22 0.0116 0.0236
30 15 23 0.1000 0.2020
31 22 24 0.1150 0.1790
32 23 24 0.1320 0.2700
33 24 25 0.1885 0.3292
34 25 26 0.2544 0.3800
35 25 27 0.1093 0.2087
36 28 27 0.0000 0.3960
37 27 29 0.2198 0.4153
38 27 30 0.3202 0.6027
39 29 30 0.2399 0.4533
40 8 28 0.6360 0.2000
41 6 28 0.0169 0.0599
Table A3
Generator data.
Bus no. Cost coefcients
a b c
1 0.00 2.00 0.00375
2 0.00 1.75 0.01750
5 0.00 1.00 0.06250
8 0.00 3.25 0.00834
11 0.00 3.00 0.02500
13 0.00 3.00 0.02500
For multi-node system:
I
bus
= Y
bus
V
bus
(A.1)
By segregating the load buses (PQ) from generator buses (PV),
Eq. (A.1) can be rewritten as

I
L
I
G

Y
1
Y
2
Y
3
Y
4

V
L
V
G

(A.2)

V
L
I
G

H
1
H
2
H H
4

I
L
V
G

(A.3)
where V
L
, I
L
is the voltages and currents for PQ buses; V
G
, I
G
is
the voltages and currents for PV buses; H
1
, H
2
, H
3
, and H
4
is the
submatrices generated from Y
bus
partial inversion.
Let

V
ok
=

H
2ki

V
i
(A.4)
where NG is the number of generators
H
2
= Y
1
Y
2
(A.5)
L
k
=

1 +
V
ok
V
k

(A.6)
where L
k
is the L-index voltage stability indicator for bus k.
Stability requires that L
k
<1 and must not be violated on a con-
tinuous basis. Hence a global system indicator L describing the
Table A4
The minimum and maximum limits for the control variables along with the initial
settings.
Min. Max. Initial
P
1
50 200 99.24
P
2
20 80 80.0
P5 15 50 50.0
P
8
10 35 20.0
P
11
10 30 20.0
P
13
12 40 20.0
V
1
0.95 1.1 1.05
V
2
0.95 1.1 1.04
V5 0.95 1.1 1.01
V
8
0.95 1.1 1.01
V
11
0.95 1.1 1.05
V
13
0.95 1.1 1.05
T
11
0.9 1.1 1.078
T
12
0.9 1.1 1.069
T
15
0.9 1.1 1.032
T
36
0.9 1.1 1.068
Q
c10
0.0 5.0 0.0
Q
c12
0.0 5.0 0.0
Q
c15
0.0 5.0 0.0
Q
c17
0.0 5.0 0.0
Q
c20
0.0 5.0 0.0
Q
c21
0.0 5.0 0.0
Q
c23
0.0 5.0 0.0
Q
c24
0.0 5.0 0.0
Q
c29
0.0 5.0 0.0
Power losses (MW) 5.842
Voltage deviations 1.1606
Lmax 0.2144
464 A.A.A.E. Ela et al. / Electric Power Systems Research 81 (2011) 458464
stability of the complete systemis L =L
max
{L
k
}, where {L
k
} contains
L indices of all load buses.
In practice L
max
must be lower than a threshold value. The
predetermined threshold value is specied at the planning stage
depending on the system conguration and on the utility policy
regarding the quality of service and the level of system decided
allowable margin.
The objective is to minimize L
max
, that is,
L
max
= max

L
k

, k = 1, . . . , NL (A.7)
Appendix B. System data
Data for IEEE 30-bus test system (100MVA base) are given in
Tables A1A4.
References
[1] S. Durairaj, P.S. Kannan, D. Devaraj, Multi-objective VAR dispatch using particle
swarm optimization, Emerg. Electr. Power Syst. 4 (2005) 1.
[2] M.A. Abido, Multiobjective optimal VAR dispatch using strength pareto evo-
lutionary algorithm, in: 2006 IEEE Congress on Evolutionary Computation,
Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada, July 1621,
2006.
[3] P. Subburaj, N. Sudha, K. Rajeswari, K. Ramar, L. Ganesan, Optimum reactive
power dispatch using genetic algorithm, Acad. Open Internet J. 21 (2007).
[4] S. Durairaj, D. Devaraj, P.S. Kannan, Genetic algorithm applications to opti-
mal reactive power dispatch with voltage stability enhancement, IE (I) J. EL 87
(2006) 4247.
[5] K.Y. Lee, Y.M. Park, J.L. Ortiz, A united approach to optimal real and reactive
power dispatch, IEEE Trans. Power Appar. Syst PAS.104 (5) (1985) 11471153.
[6] S. Granville, Optimal reactive power dispatch through interior point methods,
IEEE Trans. Power Syst. 9 (1) (1994) 98105.
[7] K. Iba, Reactive power optimization by genetic algorithms, IEEE Trans. Power
Syst. 9 (2) (1994) 685692.
[8] Q.H. Wu, J.T. Ma, Power system optimal reactive power dispatch using evolu-
tionary programming, IEEE Trans Power. Syst. 10 (3) (1995) 12431249.
[9] C. Das Bhagwan, Patvardhan, A new hybrid evolutionary strategy for reactive
power dispatch, Electr. Power Res. 65 (2003) 8390.
[10] R. Storn, K. Price, Differential EvolutionA Simple and Efcient Adaptive
Scheme for Global Optimization over Continuous Spaces, Technical Report TR-
95-012, ICSI, 1995.
[11] S. Das, A. Abraham, A. Konar, Particle Swarm Optimization and Differential
Evolution Algorithms: Technical Analysis, Applications and Hybridization Per-
spectives, Available at www.softcomputing.net/aciis.pdf.
[12] D. Karaboga, S. Okdem, A simple and global optimization algorithm for engi-
neering problems: differential evolution algorithm, Turk. J. Electr. Eng. 12 (1)
(2004).
[13] R. Storn, K. Price, Differential evolution, a simple and efcient heuristic strat-
egy for global optimization over continuous spaces, J. Global Optim. 11 (1997)
341359.
[14] J. Vesterstrm, R. Thomsen, Acomparative study of differential evolution, parti-
cle swarmoptimization, andevolutionary algorithms onnumerical benchmark
problems, in: IEEE Congress on Evolutionary Computation, 2004, pp. 980
987.
[15] Zhenyu Yang, Ke Tang, Xin Yao, Differential evolution for high-dimensional
function optimization, in: IEEE Congress on Evolutionary Computation (CEC
2007), 2007, pp. 35233530.
[16] J. Lampinen, A Bibliography of Differential Evolution Algorithm, Available at
http://www.lut./jlampine/debiblio.htm.
[17] D.G. Mayer, B.P. Kingborn, A.A. Archer, Differential evolutionan easy and ef-
cient evolutionary algorithm for model optimization, Agric. Syst. 83 (2005)
315328.
[18] M.D. Kapadi, R.D. Gudi, Optimal control of fed-batch fermentation involving
multiple feeds using differential evolution, Process Biochem. 39 (11) (2004)
7091721.
[19] B.V. Babu, P.G. Chakole, J.H.S. Mubeen, Differential Evolution Strategy for Opti-
mal Design of Gas Transmission Network, Available at www.vsppub.com.
[20] R. Balamurugan, S. Subramanian, Self-adaptive differential evolution based
power economic dispatch of generators with valve-point effects and multiple
fuel options, Comput. Sci. Eng. 1 (1) (2007) 1017.
[21] G.A. Bakare, G. Krost, G.K. Venayagamoorthy, U.O. Aliyu, Differential evolution
approach for reactive power optimization of Nigerian grid system, in: Power
Engineering Society General Meeting, 2007, pp. 16.
[22] L. Yong, S. Tao, Wu Dehua, Improved differential evolution for solving optimal
reactive power ow, in: Power and Energy Engineering Conference, 2009, pp.
14.
[23] X. Zhang, W. Chen, C. Dai, A. Guo, Self-adaptive differential evolution algorithm
for reactivepower optimization, in: FourthInternational ConferenceonNatural
Computation, 2008, pp. 560564.
[24] K. Vaisakh, P. Kanta Rao, Differential evolution based optimal reactive power
dispatch for voltage stability enhancement, J. Theor. Appl. Inform. Technol.
(2008) 638646.
[25] M.A. Abido, Optimal power ow using particle swarm optimization, Electr.
Power Energy Syst. 24 (7) (2002) 563571.
[26] Z. Daniela, A comparative analysis of crossover variants in differential evolu-
tion, Comput. Sci. Inform. Technol. (2007) 171181.
[27] A.A. Abou, E.L. Ela, M.A. Abido, S.R. Spea, Optimal power owusing differential
evolution algorithm, Electr. Eng. 91 (2) (2009) 6978.