1

2013 Nirma University International Conference on Engineering (NUiCONE)

Reactive Power Optimization Using Genetic

Algorithm
Raj K Kapadia, Nilesh K Patel
Electrical Dept., Sankalchand Patel College of Engg., Visnagar

Abstract--Reactive power optimization has an important role II. REACTIVE POWER OPTIMIZATION PROBLEM
to play in operation of power system. Optimum scheduling of
1) Introduction
reactive power reduces system losses. In this paper reactive
power optimization function is taken with the objective to The objectives of reactive power (VAR) optimization are
minimize active power losses in transmission network. This to minimize system active power losses, to improve the
objective function is subjected to power system inequality voltage profile, and to determine optimal VAR compensation
constraints and equality constraints of active and reactive power placement keeping the values of various operating parameters
balance. Genetic Algorithms are very popular in complex within their limits. To achieve these objectives, power system
optimization problem solving. Because of their fitness for
operators utilize control options such as adjusting generator
parallel computing, ease of implementation and universality. In
this paper reactive power optimization is done on IEEE-14 & excitation, transformer tap changing, place different type of
IEEE 30-bus bus test system, for this a MATLAB program is compensation devices. However, the size of power systems
developed based on Binary type of Genetic Algorithm. and prevailing constraints produce strenuous circumstances
Index Terms—Reactive power, Genetic Algorithm, Crossover, for system operators to correct voltage problems at any given
Mutation, MATLAB. time. In such cases, there is certainly a need for decision -
making tools in predominantly fluctuating and uncertain
I. INTRODUCTION computational environments. There has been a growing

I n today’s market due to the deregulation of electric power

system the concept and practice for electricity has been
changed, also there are many political and economic issues
interest in VAR optimization problems due to its advantages.
Most conventional methods used in VAR optimization are
based on linear programming and nonlinear programming.
which restricts private power producers to meet the increase Some simplification is needed in these methods to solve the
demand. Better utilization of the existing power system and problem but it also may induce local minima. Recently, new
its resources to increase capabilities with economic cost of methods based on artificial intelligence have been used in
generation is become essential. The objective of an Optimal VAR optimization and planning such as an artificial neural
Power Flow (OPF) algorithm is to find optimal control network (ANN), expert system, genetic algorithms (GAs),
variables which minimizes generation cost, loss etc. or and evolutionary programming (EP). EP and GAs are good
maximizes loadability, profits etc. while maintaining an methods to obtain global optimal optimizations. However, the
acceptable system performance in terms of limits on excessive time consumption of EP and GAs will limit their
generators’ real and reactive powers, line flow limits, output applications in power systems, especially during real - time
of various compensating devices etc. operation.
When OPF arrived in early 1960, it has marked the end of 2) Reactive Power Optimization Function
the classical period of economic load dispatch. Both are The function [3], [11], [12], [13] is as under,
optimization problem having the same objective of cost
minimization. However, ELD only considers the real power
(
Ploss = f x1 , x2 ) (1)
generation and have only one equality constraint, the real
power balance equation, on the other hand OPF considers the
Ploss = ¦ G (V
K ∈N E
K i
2
+ V − 2ViV j cos θ ij
j
2
) (2)

power output of generators, bus voltages, reactive power

compensators and transformer’s tap settings as a constraints In above equation ( )
f x1 , x2 denotes the active power loss of
with keeping in mind fuel cost minimization objective.
the transmission network, x1 is the control variable vector
OPF is a large-scale, static optimization problem which is
non-convex and nonlinear with both continuous and discrete [VG TK QC ], this control variables simply means that
control variables. Even in the absence of discrete control
variables, the OPF problem is non-convex due to the you have to optimize these variables. On the other hand x2
existence of the nonlinear (AC) power flow equality is the dependent variable vector [VL QG ] , these variables
constraints.
are the one which are going to be affected by the values of
control variables, VG is the generator voltage (continuous),

978-1-4799-0727-4/13/$31.00 ©2013 IEEE

2

TK is the transformer tap setting (integer), Qc is the shunt Selection and Sorting
Crossover and Mutation
capacitor or reactor (integer), VL is the load bus voltage, QG Until stopping criteria
is the generator bus reactive power generation,N E indicates There has been much literature available on this particular
topic i.e. Genetic Algorithm but a step by step guide for
the numbers of transmission lines in the network, GK is the programming it into MATLAB is not handy thing that you
conductance of the line K , θ ij is the voltage angle will get on internet, so in this section of the paper, a try is
made to explain a MATLAB program for Genetic Algorithm.
difference between the buses i and j . The numbers of We will try to understand Genetic Algorithm by following
equality and inequality constraints are as under: case study.
PGi − PDi − Vi ¦ V j (Gij cos θ ij + Bij sin θ ij ) = 0 (3) Case Study- Minimize
j∈N i f ( x ) = 0.001562 P12 + 0.00194 P22 + 0.00482 P32

QGi − QDi − Vi ¦ V j (Gij sin θ ij − Bij cos θ ij ) = 0

+ 7.92 P1 + 7.85P2 + 7.97 P3 + 949
(4)
Subject to,
j∈N i
P1 + P2 + P3 − 850 = 0
Vi min ≤ Vi ≤ Vi max (5)
And
TKmin ≤ TK ≤ TKmax (6) 150 ≤ P1 ≤ 600
min max
Q Gi ≤ QGi ≤ Q Gi (7) 100 ≤ P2 ≤ 400
Q min
C ≤ QC ≤ Q max
C (8) 50 ≤ P3 ≤ 200
Equation number (3) and (4) are the active power balance for To solve this case study we’ll going to have an algorithm
all the buses excluding the slack bus and reactive power which is shown in Figure-1 and all the elements of the figure
balance for all load buses respectively. While equation are explained.
number (5), (6), (7), and (8) are the upper and lower limits for Start

Voltages at all buses, Transformers tap-setting, Reactive Initialization of Population

i.e. Chromosomes

power generation at all generator buses and Reactive power

Function Calculation
compensation if any provided.
Fitness Calculation

III. GENETIC ALGORITHM

Sorting

1) Introduction of Genetic Algorithm

Many human inventions were inspired by nature. Artificial Selection

neural networks are one example. Another example is Crossover and

Mutation
Genetic Algorithms (GA). GA search by simulating
Function Calculation
evolution, starting from an initial set of solutions or
hypotheses, and generating successive "generations" of Fitness Calculation

solutions. This particular branch of AI was inspired by the

way living things evolved into more successful organisms in
Check Gen <= Maxgen
nature. The main idea is survival of the fittest, a.k.a. natural No
And
Check Tolr <= Error
selection.
The genetic based mechanism for problem solving was Yes

first conceived by Professor John Holland of the University Stop

of Michigan, Ann Arbor, in 1965. Genetic Algorithm owes Fig. 1. Algorithm for Case study
their popularities as solutions in various search and Initialization of Population
optimization problems to their global perspective, widespread Population is a matrix having strings of binary numbers i.e.
applicability, and inherent parallelism. 0 and 1 which are nothing except potential solution to the
2) Algorithm problem i.e. well known as chromosome. We generate a
The major steps involved are the generation of a random population to start the solution. There are many ways
population of solutions, finding the objective function and to generate population; some of them are as under.
fitness function and the application of genetic operators. chromosome = round(rand(pop_size,nbit));
These aspects are described briefly below. They are described chromosome = randi([0,1],pop_size,nbit);
in detail in the following subsections. Where,
Randomly initial population pop_size = number of potential solutions
Repeat nbit = it is a multiplication of number of bit and number of
Evaluate objective function variables. For example in our Case we have 3 variables and if
Find fitness function we take 10 bits per variable we have nbit = 30.
These are binary strings and thus they must be converted into
 3

real numbers for further use by using following formula [10]: those whose fitness is high and roulette wheel selection
Pi upper
− Pi lower procedure is very efficient for selection of parent. Roulette
Pi = Pi lower + n
× Pi decoded wheel selection procedure uses fitness of that string to make
2 −1 slot sizes. That is, in the biased roulette wheel, good solutions
Where, have a larger slot size than the less fit solutions. The roulette
Pi is the real value of the variable and subscript i is for 1 to wheel is spun to obtain a reproduction candidate. This
pop_size. process is explained as under.
n is the number of bits of the variable. Step-1. Calculate probability of each chromosome
Pi lower and Pi upper are the lower and upper limits of the fitni
probi = pop _ size
variable respectively.
¦ fitn i
Pi decoded is the binary to decimal converted value of the i

variable. Step-2. Calculate cumulative probability of each

Function Calculation chromosome
i
cumprobi = ¦ prob j
After decoding all the strings of the initial population and
getting values of each variable, we put them into our function
j =1
and find the function value. It should be noted that the size of
Step-3. Generate a random number r between 0 and 1.
this function matrix is pop_size by 1.
Fitness Calculation [8] Step-4. If r ≤ cumprobi then select the i string.
Fitness is the ability of a particular string to solve the Step-5. Repeat Step-3 and Step-4 for the number of
problem. It is also a way for satisfying equality constraints. chromosome is needed for reproduction.
Implementation of a problem in genetic algorithm is realized The whole procedure is explained with the help of an
within the fitness function and in order to emphasize the example, say we have pop_size = 5, other related data are in
‘best’ chromosomes and speed up convergence of the following Table-1.
iteration procedure, fitness is normalized into the range String 1 2 3 4 5
between 0 and 1. There are many ways to calculate fitness of Fitness 28 18 14 9 26
a string a most simple way of Fitness calculation is done in Prob. 0.2947 0.1847 0.1473 0.0947 0.2736
following steps. Cumm. 0.2947 0.4794 0.6267 0.7214 1
Step-1. First calculate the error by converting our equality Prob.
constraint equal to the error. For our case Table-1 Roulette Wheel selection example
Ei = P1 + P2 + P3 − 850 MATLAB code for roulette wheel selection process is as
under:
Step-2. It can be also converted in equality constraint as
pop_size = 5;
P1 + P2 + P3 = 850 and from this the fitness is given by, fitn = [28 18 14 0 26];
§ E · for i = 1:pop_size
fitni = 1 ÷ ¨1 + α i ¸ prob(i) = fitn(i)/sum(fitn);
© 850 ¹ end
Where α is the scaling factor. comprob = cumsum(prob);
Sorting parents = zeros(1,pop_size);
This is one of the important part of the algorithm, as it for i = 1
r = rand;
gives search direction to the program by sorting fitness in
for j = 1:pop_size
descending order and according to that order it sorts initial
if r < comprob(j)
population, error and function values. Here error is sorted parents(i) = j;
only to terminate the program except this there is no use of it, break;
and functions values are sorted to display optimize function end
value. end
Initial population is sorted and from this population we end
keep some for reproduction, the amount we keep is given by Result for this program is like if r = 0.1682 then string 1 is
population multiplied by selection probability, and from this selected.
kept population we select highly fit population for Crossover and Mutation
reproduction using roulette wheel selection procedure. Selected strings, now onwards called parents, are used to
Reproduction means we use crossover and mutation operator do crossover and mutation to produce children which are
on this existing population and produce new population. responsible for the global search property of the GA.
Selection Crossover operator basically combines substructures of
In this section we select some of the population kept after two parent chromosome to produce new structure which is
sorting for reproduction of the population. We want to select child. There are many types of crossover like single point,
4

double point, multipoint, uniform, matrix etc. 1 1 1 0 1 0 3rd string, product

The final operator in the genetic algorithm is mutation; the 1 1 1 0 1 1 of 1st and 2nd string
mutation operator is used to inject new information into the
population. Mutation process changes randomly the newly chromosome =
generated offspring. For binary genetic algorithm mutation is
preferred which switches a few randomly chosen bits from 1 1 1 1 1 0 1
4th string, product of
to 0 and vice-a-versa. There are many types of mutation 1 0 1 0 1 0
1st and 2nd string
processes like uniform, boundary, non-uniform etc. 1 1 1 0 1 0
After mutation, the new generation is completed and the 1 0 1 1 0 1
procedure begins again with the function calculation of the
population and so on. Random Bitwise Mutation
One simple MATLAB code for single point crossover and
uniform mutation is shown as under. mutrow =
clc;
clear all; 3 4
chromosome = round(rand(4,6))
str = ('Single Point mutcol =
CrossOver');
disp (str); 4 2
chromosome(3,:) =
[chromosome(1,1:3)
chromosome =
chromosome(2,4:6)]
chromosome(4,:) = Bit number (3,4)
1 1 1 1 0 1
[chromosome(2,1:3) Mutate from 0 to 1.
chromosome(1,4:6)] 1 0 1 0 1 0
str = ('Random Bitwise 1 1 1 1 1 0
Mutation'); 1 0 1 1 0 1 Bit number (4,2)
disp (str); Mutate from 0 to 1.
mutrow = randi([1,4],1,2) chromosome =
mutcol = randi([1,5],1,2)
for i = 1:2 1 1 1 1 0 1
chromosome(mutrow(i),mutcol(i)) = 1 0 1 0 1 0 Final population
1 1 1 1 1 0
abs(chromosome(mutrow(i),mutcol(i)) 1 1 1 1 0 1
-1)
end The next task is to recalculate function values and fitness
Result of the above program is as under. In this result original
values. After this we will check for the convergence
population which is generated at the start. From this
conditions of the program i.e. maximum number of
population string 3rd and 4th is selected for crossover and
generation and error, if both the values are under limit then
mutation, both these are product of 1st and 2nd strings. 1st and
we repeat the same procedure and if one of them violates the
2nd string is cut from half. 3rd string has first half of 1st string
limit then we terminate the program and best values of
and second half of 2nd string while 4th string has first part of
function and variables are our result.
2nd string and second part of 1st string, also 3rd and 4th string is
operated for mutation at points (3,4) and (4,2) at 3rd and 4th
IV. REACTIVE POWER OPTIMIZATION USING GENETIC
strings respectively. ALGORITHM
chromosome =
1) Introduction
1 1 1 1 0 1 Reactive power flow is needed in an alternating-current
1 0 1 0 1 0 transmission system to support the transfer of real power over
Original population the network.
0 0 0 1 1 1
1 1 1 0 1 1 One of the objectives of reactive power optimization is
minimizing the real power losses in the branches of the
Single Point CrossOver considered network. In this approach, the security is
implicitly guaranteed by the respect of the operational limits.
chromosome = To simplify the model, the losses minimization was obtained
by minimizing the real power produced by the slack
1 1 1 1 0 1 generator.
1 0 1 0 1 0 The objective function for the same problem is well
explained in Section II.
 5

2) Algorithm population size is not giving any further improvement in

The implementation steps of Genetic Algorithm based results an only became burden to the program.
algorithm to solve the problem can be written as follows; The variables consists of,
Step-1. Input the system data. Generator Bus Active Generation = 4, Generator Bus
Step-2. Generate random binary chromosomes within their Voltages = 5, Transformer’s Tap Settings = 3.
respective limits, these chromosomes contains values for Probability of crossover and mutation = 0.5 and 0.005,
generator bus voltages, active power except slack bus, and Crossover probability and Mutation probability has also
transformer tap setting. been tuned, for crossover the range of 0.1 to 0.8 has been
Step-3. For each chromosome, run NR load flow analysis and employed and for mutation the range was 0.001 to 0.03.
find out load bus voltages and angles, reactive power Maximum generation = 10.
generation, and calculate line flows. Start
Step-4. Evaluate function and find fitness of each
1
chromosome using fitni = . Generate random chromosomes
funi
Step-5. Sort fitness of the chromosomes in descending order
and rearrange them based on their fitness. Run Load Flow Analysis

Step-6. Use roulette wheel selection procedure to select

parents to produces children for the next generation.
Step-7. Crossover and Mutation operation is done on parent Evaluate function and fitness

chromosomes and children chromosomes are produced.

Step-8. For each new chromosome run NR load flow analysis
and reevaluate the function and find fitness of each Sorting and Selection
chromosome.
Step-9. If stopping criterion is not violated then repeat steps 5
to 8.
Crossover and Mutation
There are mainly two type of stopping criteria, one is
maximum number of generation is reached and second there
is no significant improvement in function value, in this paper Run Load Flow and Revaluation
first one is taken as a stopping criteria. of function and fitness

As here load flow analysis has been used all the inequality
constraints are automatically satisfied while the control
No
variables are kept within their limits using GA. Following Check for termination
Figure-2 shows flow chart for reactive power optimization conditions

using genetic algorithm.

The different control variables for reactive power Yes

optimization are active power generation, generator bus Stop

voltage, transformer’s tap setting, shunt/series Fig. 2. Algorithm for Reactive Power Optimization Using GA
compensations. 2) IEEE 30-bus test system
The IEEE 30-bus test system is used in this work again
V. RESULTS linedata and busdata are not shown, in IEEE 30-bus system
MATLAB is used for programming of the reactive power there are 30 buses and 41 lines. Out of the 30 buses six buses
optimization problem on IEEE 14-bus system and on IEEE are generator bus and rest of buses are load buses. Out of 41
30-bus system. lines 37 are transmission lines and on 4 lines contain tap
1) IEEE 14-bus test system changing transformers.
The IEEE 14-bus test system is used in this work and The network has active load of 283.4 MW and reactive
linedata and busdata are not shown, in IEEE 14-bus system load of 126.2 MVAR.
there are 14 buses and 20 lines. Out of the 14 buses five buses Genetic Algorithm parameters for Reactive Power
are generator bus and rest of buses are load buses. Out of 20 Optimization.
Population size = 20, Bit per variable = 20,
lines 17 are transmission lines and on 3 lines contain tap
Variable = 17,
changing transformers.
The variables consists of,
The network has active load of 259 MW and reactive load Generator Bus Active Generation = 5, Generator Bus
of 73.5 MVAR. Voltages = 6, Transformer’s Tap Settings = 4, Shunt
Genetic Algorithm parameters for Reactive Power Capacitor = 2.
Optimization. Probability of crossover and mutation = 0.5 and 0.005,
Population size = 20, Bit per variable = 20, Variable = 12, Maximum generation = 10,
The population size has been tuned because large value of Table-2 shows values of generated variables using GA for
6

IEEE 14-bus test system. VI. REFERENCES

Variable Value [1] Anastasios G. Bakirtzis, Pandel N. Biskas, Christoforos E. Zoumas
V1 (p.u.) 1.0175 And Vasilios Petridis “Optimal Power Flow By Enhanced Genetic
Algorithm”, IEEE Transaction On Power Systems, Vol. 17, No. 2,
V2 (p.u.) 1.0182
pp. 229-235, May 2002.
V3 (p.u.) 1.0035 [2] C. M. Wankhade And A. P. Vaidya, “Optimal Power Flow
V6 (p.u.) 0.9968 Evaluation Of Power System Using Genetic Algorithm”,
V8 (p.u.) 1.0038 International Journal Of Power System Operation And Energy
Management, Vol. 1, No. 4, pp. 77-82, 2012.
Pgen1 (MW) 21.065 [3] Chaohua Dai, Weirong Chen, Yunfang Zhu And Xuexia Zhang,
Pgen2 (MW) 70 “Seeker Optimization Algorithm For Optimal Reactive Power
Pgen3 (MW) 50 Dispatch”, IEEE Transactions On Power System, Vol. 24, No. 3,
pp. 1218-1231, Aug. 2009.
Pgen6 (MW) 70 [4] Haupt Randy L. And Haupt Sue E., “Practical Genetic Algorithm”,
Pgen8 (MW) 50 2nd Edition, A John Wiley & Sons, Inc., Publications, 2004, pp.
Tap8 0.998 27-47.
[5] Huang Wei, Xu Chunli, Zhang Jianhua And Hu Shan'ang, “Study
Tap9 0.989 Of Reactive Power Optimization Based On Immune Genetic
Tap10 0.995 Algorithm”, Transmission and Distribution Conference and
Table-2. Result for IEEE 14-bus test system. Exposition, 2003 IEEE PES , Vol. 1, pp. 186-190, Sep. 2003.
Table-3 shows value of generated variable using GA for [6] Iba, K., “Reactive Power Optimization By Genetic Algorithm”,
IEEE 30-bus test system. Power Systems, IEEE Transactions on , Vol. 9, No. 2, pp. 685-692,
May 1994.
Variable Value [7] Jizhong Zhu, “Optimization Of Power System Operation”, A John
V1 (p.u.) 1.011 Wiley & Sons, Inc., Publications, 2009, pp. 409.
V2 (p.u.) 0.998 [8] Kothari D. P. And Dhillon J. S., “Power System Optimization”, 4th
Edition, PHI Learning Private Limited, Jan 2009, pp. 501-532.
V5 (p.u.) 1.01 [9] Kwang Y. Lee And Mohamed A. El-Sharkawi , “Modern Heuristic
V8 (p.u.) 1.001 Optimization Techniques – Theory And Application To Power
V11 (p.u.) 0.991 System”, A John Wiley & Sons, Inc., 2008, pp. 471-500.
[10] Liu, Z.Q. And Wang, H.Q., “Reactive Power Optimization And
V13 (p.u.) 1.008
Voltage Control Using An Improved Genetic Algorithm”, Power
Pgen1 (MW) 167.367 System Technology (POWERCON), International Conference, pp.
Pgen2 (MW) 25.97 1-5, October 2010.
Pgen5 (MW) 15.66 [11] P. Aruna Jeyanthy And D. Devaraj, “Multi-Objective Genetic
Algorithm For Reactive Power Optimization Inclusing Voltage
Pgen8 (MW) 30.61 Stability”, International Journal Of Engineering Science And
Pgen11 (MW) 28.73 Technology, Vol. 2, No. 7, pp. 2715- 2729, 2010.
Pgen13 (MW) 24.73 [12] S. Durairaj, P. S. Kannan And D. Devaraj, “Application Of
Genetic Algorithm To Optimal Reactive Power Dispatch
Tap11 1.0999 Including Voltage Stability Constraint”, Journal Of Energy &
Tap12 1.006 Environment, Vol. 4, pp. 63-73, 2005.
Tap15 0.9419 [13] Sreejaya, P. And Rejitha, R., “Reactive power and Voltage Control
in Kerala Grid and Optimization of Control Variables Using
Tap36 0.9737 Genetic Algorithm”, Power System Technology and IEEE Power
Qsh10 (MVAR) 12.67 India Conference, POWERCON, Joint International Conference,
Qsh24 (MVAR) 7.69 pp. 1-4, Oct. 2008.
Table-3. Result for IEEE 14-bus test system.
3) Conclusion
Following Table-4 shows result for active power losses and
reactive power generation.
System Parameter RPOUGA
IEEE 14-bus test Active power loss 2.065
system Reactive gen. 64.012
IEEE 30-bus test Active power loss 9.667
system Reactive gen. 114.031
Table-4. Active power loss and Reactive power generation table
The results in Table-2 for IEEE 14-bus system and Table-3
for IEEE 30-bus system depicts couple of things, like the
active power loss has reduced substantially and reactive
generation has reduced and the reason behind this is nothing
but the optimization of the control variables. So it can be said
that reactive power optimization can be done by controlling
generator bus voltages, active power generation and
transformer tap settings.