2012 Fourth International Conference on Computational Intelligence and Communication Networks

A New Attempt to Optimize Optimal Power Flow

based Transmission Losses using Genetic Algorithm
Jinendra Rahul*1, Yagvalkya Sharma#2, Dinesh Birla*3
*
Electrical Engg. Department, University College of Engineering, RTU, Kota.
1
[email protected]
3
[email protected].
#
Electronics Department, Rajasthan Technical University
2
[email protected]

Abstract — This paper presents a new method using GADS Particle Swarm Optimization, Simulated
Toolbox in MATLAB (A Genetic Algorithm Approach) to find
the optimal solution of optimal power flow based transmission Annealing, adaptive Tabu Search, Harmony Search
losses. Optimal power flow (OPF) is a key area of concern in etc. are being applied in the area.
electric industries. The basic OPF solution is obtained with Classical optimization methods suffer from
objective function as production cost minimization while
satisfying a set of system operating constraints. For reactive drawback of getting trapped in a local minima and
power optimization the OPF problem is formulated as premature convergence. Although some of the
minimization of system active power losses and improvement in Meta-heuristic methods are successful in locating
voltage stability of the system. In this paper GA based optimal
power flow solution is presented for IEEE 30-bus test power the optimal solution yet they also are rather slow in
system with objective as transmission losses minimization and convergence.
optimal results by GA are also compared with solution obtained This paper describes a new approach using
by using Particle Swarm Optimization Technique.
Genetic Algorithm to optimize transmission losses
Keywords— Genetic algorithm (GA), optimal power flow (OPF), using optimal power flow. In doing so, GADS
Particle Swarm Optimization (PSO). Toolbox in MATLAB is used. The proposed
method has been tested on IEEE 30-bus test power
I. INTRODUCTION system.
Controlled voltage magnitude, Generator MW
Carpentier first defined the OPF problem in early
power, reactive power injection from reactive
1960s and OPF soon many researchers were
power sources and transformer tap setting etc. are
attracted towards this artistic area of research.
the controllable parameters in power system. The
Optimal Power Flow problem of power system
objective in OPF problem is to optimize the total
operation and planning is a constrained nonlinear
transmission loss by optimizing the controllable
and occasionally combinatorial optimization
variables within their limits. Hence, during normal
problem. Since its introduction by Carpentier, many
system operating conditions there is no violation on
investigators have attempted to solve this problem
other system parameters viz. MVA flow of
of OPF [1]. With the advancements in computing
transmission lines, load bus voltage magnitude,
technologies, it is becoming possible to consider
generator MVAR etc.
many aspects of the OPF problem more vigorously
The organization of the paper is as follows:
as it progressively became easy to formulate and
Section II explains OPF problem objective function,
solve complicated large-scale problems.
equality and inequality constraints; Section III
Many classical optimization techniques such as
describes the Genetic Algorithm procedure to solve
Linear programming [1], Quadratic programming
non-linear optimization problems; PSO is discussed
[2], Dynamic programming [3], Lagrangian
in Section IV. A comparison of simulation results
Relaxation approaches and Gradient methods as
obtained using GA and PSO for IEEE 30-bus
well as Artificial Intelligence Techniques such as
system is presented in Section V.
Genetic Algorithm [4], Evolutionary algorithms [5],
II. OPTIMAL POWER FLOW PROBLEM
Dinesh Birla is Associate Professor in the Department of EE, University
College of Engg., Rajasthan Technical University, Kota. The general optimal power flow problem can be
Jinendra Rahul and Yagvalkya Sharma are PG students in the University described as a constrained non-linear optimization
College of Engg., Rajasthan Technical University, Kota.
problem as under:

978-0-7695-4850-0/12 $26.00 © 2012 IEEE 566

DOI 10.1109/CICN.2012.212
 4 67
≤ ≤
Minimize    ! ! !

Subject to-  = 0 , equality constraints

Where,
 ≥ 0 , inequality constraints
4 , 67 are upper and lower limits of voltage
A. Objective function magnitude at bus n.
The objective function for total power 84 , 867 are upper and lower limits of tap
transmission loss can be constituted as mentioned position of transformer n.
below [6,7]: *  , *
4 67
are upper and lower limits of reactive
power source n.
4 67
!, , !, are upper and lower limits of power
 

 =   ,  +  − 2  cos −  

 
generated by generator n.
  The penalty function can be constituted as under:
 = 9 + β; + β< + β= + β 9 + β> + β!
Where,
 is the voltage magnitude at bus n. Where,
 is the total number of transmission lines. @$
 is the voltage angle at bus n. β; = ?  ! − "
, is the conductance of line n-m. 
@$
B. System Constraints
− | | | |%&, % cos'(, − 
The system constraints as equality and inequality

constraints are as under [8]. +  
@$
1) Equality constraint: Power flow equations:
β< = ? *! − *"
For Real Power balanced, 
@$
$
−  | | | |%&, % cos'(, − 
! − " −  | | | |%&, % cos'(, −  +  )


+  
=0
@C

β= = ? AB0, * − *67 

For Reactive Power balance,
$

@C
*! − *" + | | | |%&, % sin'(, −  +  ) = 0 
+ + ? DAB'0, *4 − * )E

Where, @F
n is 1, 2, 3,……….. ./01 . β 9 = ? AB0, 8 − 867 
! is the real power generation at bus n. 
*! is the reactive power generation at bus n. @F

" is the real power demand at bus n. + ? DAB'0, 84 − 8 )E
*" is the reactive power demand at bus n. 
@$
3 is the total number of buses.
(, is the angle of bus admittance element n,m. β> = ? AB0,  − 67 
&, is the magnitude of Y bus element n,m. 
@$

2) Inequality constraint: Variable limits as + ? DAB'0, 4 −  )E
inequality constraints are as follows: 

4 ≤  ≤ 67
84 ≤ 8 ≤ 867
*4 ≤ * ≤ *67

567
 @G
the constraints of the problem are selected. This
! 
67 
β! = ? AB0, ! − happens to be the first ‘parents’ chromosome pool.

@G (b) Selection of parents and reproduction: After
4 
+ ? DAB'0, ! − ! )E random generation of initial population, evaluation

of each chromosome is done according to the
Where, goodness of fit in terms of objective function value
H3 is the total number of buses. and the chromosomes satisfying constraints are
H= is the total number of reactive power sources. ranked. Reproduction forms a new generation by
H9 is the total number of transformers. choosing those individuals from the old population
H! is the total number of generators. that have the highest value of the fitness function.
Ranking chromosomes according to respective
III. GENETIC ALGORITHM high objective function value helps to perform the
Perhaps Rechenberg suggested the earliest task of copying chromosomes with a higher
instance of an evolution-based technique related to objective value to the next generation and
modern GAs in the 1960’s [9]. John Holland is discarding the ones with poor values [12].
generally credited with the invention of GAs. Suitable penalty function (PF) for awarding the
Holland’s 1975 monograph [10] describes the basic goodness of a solution on one hand, and to punish
approach to the population-based search the solution not satisfying constraints on the other is
characteristics of today’s GAs. A second major used with objective function [13].
milestone in the history of GAs came up with the
publication of David Goldberg’s 1989 book [11]. Start
The publication of Goldberg’s text accelerated the
application of GAs. Since then, search algorithms
Bus System Data Analysis
from GAs have been applied in a diverse array of
disciplines, and the benefits resulting from these
efforts have been numerous.
Forming of Y-Bus and perform Load Flow
The GA Algorithm works by creating an initial
population of N possible solutions in the form of
candidate chromosomes (or simply, individuals).
Each of these individuals is a representation of the Apply GA to Optimize Objective Function
target system. Initial population could be generated
at random, or as genetic material from some
previous procedure. A population consists of a No
collection of chromosomes. Every chromosome Is Generation=Maximum?
represents a possible solution. Each chromosome
consists of one or more genomes. A particular Yes
feature in the chromosome is represented by a
genome. The nature of a genome depends on the Save
underlying data representation. For bit (or binary)
representations, the genome is a series of bits. For a
real number representation, the genome is an End
integer or floating-point number.
Fig. 1: Flowchart of GA application to transmission loss minimization
GA Procedure:
(c) Creation of ‘offspring’, constraint checking and
The The GA procedure to find the optimal selection of ‘parents’ for next generation: Strings with a
solution for the optimization problems is described higher objective function value have a higher
in the following steps. probability of contributing one or more offspring in
(a) Initialization: A GA begins with a population the next generation. By properly applying the
of potential solutions. These solutions are encoded crossover and mutation, the process of breeding
in the chromosomes and chromosomes satisfying new solutions (creating a new population) by
selecting the chromosomes satisfying the

568
constraints and having the largest objective function TABLE I
CONTROL VARIABLES LIMITS UESED FOR OPTIMAL POWER FLOW
value continued for the specified number of
Variables Lower limits Upper limits
generations. MW 200 260
!
(d) Insertion of new chromosomes and Termination: ! MW 40 80
Some new individuals are also introduced into the V1 – V30 p.u. 0.9 1.1
8 Mvar 0.9 1.1
next generation. These individuals are of randomly 8 Mvar 0.9 1.1
valued genes (created in the same way as the GA’S *! Mvar -30 -20
initial population). New individuals could *! Mvar 50 70
significantly increase genetic diversity but also have *!b Mvar 25 43
*!c Mvar 15 30
an adverse effect on convergence if overused. The *!d Mvar 15 26
criterion of maximum number of generations is *!e Mvar 00 10
used for the GA program to terminate. However,
there is an additional provision also to terminate the To find a set of optimal power flow solution, the
program if one or more individuals achieved a proposed method, Genetic Algorithm was employed.
satisfying objective function value. For benchmarking, Particle Swarm Optimization
was also employed. Table II & III gave the
IV. PARTICLE SWARM OPTIMIZATION (PSO) simulation results of tests.
Kennedy and Eberhart developed a particle Current Point
swarm optimization algorithm based on the 250

behavior of individuals (i.e., particles or agents) of

a swarm. Its roots are in zoologist’s modeling of the 200

movement of individuals (i.e., fish, birds, and

150
insects) within a group [14]. It has been noticed that
Current point

members of the group seem to share information 100

among them to lead to increased efficiency of the
group. The particle swarm optimization algorithm 50
searches in parallel using a group of individuals
similar to other AI-based heuristic optimization 0

techniques.
Each individual corresponds to a candidate -50
0 5 10 15 20 25 30 35
solution to the problem. Individuals in a swarm Number of variables: 38

approach to the optimum through its present Fig. 2: Final variables value using GA.
velocity, previous experience, and the experience of
TABLE II
its neighbors. In a physical n-dimensional search TRANSMISSION LOSS FLOW ALLOCATION RESULTS IN MW FOR
space, the position and velocity of individual i are IEEE 30-BUS SYSTEM USING GA

represented as the velocity vectors. Using these S. No. Lines Transmission Loss (MW)
information individual i and its updated velocity 1 1-2 0.396858
2 1-3 0.494204
can be modified under the following equations in 3 2-4 0.681841
the particle swarm optimization algorithm. 4 3-4 2.219145
5 2-5 0.746898
V. SIMULATION RESULTS 6 2-6 0.031840
In this paper, an IEEE 30-bus test power system 7 4-6 2.216592
8 5-7 0.438401
was chosen as a test system. The Particle Swarm 9 6-7 1.121250
Optimization and Genetic Algorithm methods were 10 6-8 0.462380
employed for comparison. Each method was carried 11 6-9 0.785908
12 6 - 10 0.160106
out by 30 trials. Table 1 shows the limits of
13 9 - 11 0.000000
variables used in Optimal Power Flow problem. 14 9 - 10 0.759651
15 4 - 12 0.063781
16 12 - 13 0.000000
17 12 - 14 0.106695
18 12 - 15 0.246274
19 12 - 16 0.060512
20 14 - 15 0.001038

569
 21 16 - 17 0.018412 accurately found using GA and PSO. It is
22 15 - 18 0.059823
23 18 - 19 0.008044
concluded that GA method provides better results
24 19 - 20 0.021404 amongst the search algorithms to obtain a near
25 10 - 20 0.115865 global solution of the optimal power flow problems.
26 10 - 17 0.012167
27 10 - 21 0.096352 REFERENCES
28 10 - 22 0.044869
[1] B. Stott and J. L. Marinho, “Linear Programming for Power System
29 21 - 22 0.000301 Network Security Applications”,IEEE Transactions on Power
30 15 - 23 0.026510 Apparatus and Systems, vol. PAS-98, pp. 837-848, 1979.
31 22 - 24 0.026856 [2] G. F. Reid and L. Hasdorf, “Economic Dispatch using Quadratic
32 23 - 24 0.003830 Programming”, IEEE Transactions on Power Apparatus and Systems,
33 24 - 25 0.017590 vol. PAS-92, pp. 2015-2023, 1973.
[3] H. W. Dommel and W. F. Tinney, “Optimal Power Flow Solutions”,
34 25 - 26 0.015453 IEEE Transactions on Power Apparatus and Systems, Vol. PAS-87, pp.
35 25 - 27 0.044511 1866-1876, October 1968.
36 28 - 27 0.363467 [4] G. Bakirtzis, N. Biskas, E. Zoumas and Vasilios Petridis, “Optimal
37 27 - 29 0.115114 Power Flow by Enhanced Genetic Algorithm”, IEEE Transactions on
38 27 - 30 0.180955 Power Systems, Vol.17, pp. 229-236, 2002.
[5] Jason Yuryevich and Kit Po Wong, “Evolutionary Programming
39 29 - 30 0.057791
Based Optimal Power Flow Algorithm”, IEEE Transactions on Power
40 8 - 28 0.000654 Systems, Vol.14, pp. 1245-1250, 1999.
41 6 - 28 0.276080 [6] A.J. Wood and B.F. Wollenberg, “Power Generation, Operation and
Control”, Wiley-Interscience, New York, 1996.
Current Function Value: 12.4994 [7] William D. Rosehart, Victor H. Quintana and Claudio A. Cañizares,
350 “Multiobjective Optimal Power Flows to Evaluate Voltage Security
Costs in Power Networks”, IEEE Transactions on Power Systems, Vol.
300 18, No. 2, May 2003.
[8] Kyung II Min, Sang-Hyeon Ha, Su-Won Lee,and Young-Hyun Moon,
“Transmission Loss Allocation Algorithm using Path-Integral Based
250
on Transaction Strategy”, IEEE Transaction on power systems, vol.25,
pp.195-205, Feb.,2010.
Function value

200 [9] I. Rechenberg, “Cybernetic Solution Path of An Experimental Problem,

(Royal Aircraft Establishment Translation No. 1122, B. F. Toms,
150 Trans.), Farnborough hants: Ministry of Aviation, Royal Aircraft
Establishment, August, 1965
[10] J. H. Holland, “Adaptation in Natural and Artificial Systems”,
100
University of Michigan Press, Ann Arbor, 1975.
[11] David E, Goldberg, “Genetic Algorithms in Search, Optimization, and
50 Machine Learning”, ISBN 81-7808-130-X, Pearson Education Asia Pte
Ltd., 2000.
0 [12] Z. Haibo, Z. Lizi and M. Fanling, “Reactive Power Optimization Based
0 2 4 6 8 10 12 14 16 18 20 on Genetic Algorithm”, International Power Conference on Power
Iteration
System Technology, pp.1448 – 1453, 18-21 August 1998.
[13] Mohd Herwan Sulaiman, Mohd Wazir Mustafa and Omar Aliman,
Fig. 3: Function value characteristic using GA. “Transmission Loss and Load Flow Allocations via Genetic Algorithm
TABLE IIII Technique”, IEEE Xplore, TENCON 2009.
SIMULATION RESULTS OF IEEE 30-BUS TEST POWER SYSTEMS [14] J. Hazra1 and A. K. Sinha, “A Study on Real and Reactive Power
Optimization using Particle Swarm Optimization”, International
Total Power Transmission Losses (MW) Conference on Industrial and Information Systems, pp. 323 – 328, 9-11
August 2007.
PSO 13.55
GA 12.4994

For 30-bus test system the average power loss

solutions are 17.72 MW and 12.4994 MW for the
PSO and GA methods, respectively. This reveals
that the GA method is more efficient method for
solving the optimal power flow problem with the
power loss objective.
VI. CONCLUSIONS
This work was conducted by 30 trials on a study
for IEEE 30-Bus system. The comparison of results
obtained by GA and PSO for each of 30 trials is
presented in the paper. It is demonstrated by the
results that a set of optimal solutions with respect to
the power transmission loss objective can be

570