You are on page 1of 8

RESEARCH PAPER

International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

A Fast Computational Genetic Algorithm for


Economic Load Dispatch
M.Sailaja Kumari1, M.Sydulu2
Email: 1Sailaja_matam@Yahoo.com
1, 2
Department of Electrical Engineering
National Institute of Technology, Warangal
Andhra Pradesh, India
il 2 d l h @ h i

Abstract— This paper presents a Fast Genetic Algorithm (FGA) reached. If the condition is not satisfied, the population
approach for solving Economic Load Dispatch (ELD) problem. members are arranged in descending order of their fitness
GAs perform powerful global searches, but their long value and genetic operators are applied to produce new and
computation times limit them when solving large scale better fit population from old population while applying
optimization problems. The present paper describes a method to genetic operators like recombination, elitism, crossover and
overcome this limitation by starting with random solutions
mutation. This completes one iteration and in GA terminology
within the search space and narrowing down the search space by
considering the minimum and maximum errors of the population one generation. The generation count is updated and the
members. Since the search space is restricted to a small region process is repeated.
within the available search space, the algorithm works very fast. SGA is capable of locating the near-optimal solutions, but
This feature of the algorithm is attractive when applied to ELD requires a large number of generations to converge. In
of large systems. The convergence of the algorithm can be addition to SGA, there exists a number of advanced GAs,
expected with in one or two generations independent of size of designed to have advanced features. Suitable combination of
the system. The results have been demonstrated for ELD of GA operators also are used [2] to enhance the performance of
standard 3 generator, 6 generator, 20 generator and 38 generator SGA.
systems with and without consideration of transmission system
Economic Load Dispatch (ELD) is one of the important
losses. In all the cases the Fast GA shows reliable convergence.
The final results obtained using Fast GA are compared with
optimization problems in modern Energy Management
conventional GA and found to be encouraging. Systems (EMS). ELD determines the optimal real power
settings of generating units in order to minimize total fuel cost
Index Terms— Economic dispatch, Genetic algorithms, of thermal plants. Various mathematical programming
Incremental fuel cost of generators, Optimal dispatch. methods and optimization techniques have previously been
applied for solution of ELD. These include Lambda iteration
I. INTRODUCTION method, participation factors method and gradient methods.

G ENETIC Algorithms (GAs) are optimization algorithms ELD problems in practice are usually hard for traditional
based on the mechanics of natural selection and natural mathematical programming methodologies because of the
genetics. GAs have been developed by John Holland in 1960s. equality and inequality constraints.
Some fundamental ideas have been borrowed from genetics GAs have also been applied for solution of ELD. [3, 4]
and are artificially used to construct more robust algorithms present ELD using GA. A unit based encoding scheme is
requiring minimal problem information [1]. Unlike classical used, which restricts the applicability of GA to large-scale
search and optimization methods GA starts its search with a systems. In unit based encoding the chromosome length
random set of solutions, instead of single solution. increase with number of units in the system. [5] presented a
The Holland’s GA that uses simple genetic operators is lambda-based GA approach for solving ED problem. The
known as Simple Genetic Algorithm (SGA). SGA starts with solution time grows approximately linearly with problem size
a random population of solutions (chromosomes in GA rather than geometrically. An Extended compact Genetic
terminology). Each population member is then evaluated for Algorithm (ECGA) and an adaptive discretization technique
the given objective function and is assigned fitness. The named split-on-demand (SOD) were proposed [6] to solve
algorithm then checks for the stopping condition. The ELD problem. A repair operator is defined for making the
condition can be either all population members assume equal infeasible solutions to satisfy the equality constraint. [7]
fitness values or when a maximum number of generations are proposed a combined economic and emission dispatch by
considering both the economy and emission objectives. A
349

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

comparison of GA, Micro GA and Evolutionary Programming by the load. Also, the power output of each unit must be
is brought out for different size of systems. [8] presents greater than or equal to the minimum power permitted and
Economic power Dispatch using a hybrid technique, where less than or equal to the maximum power permitted on that
GA parameters are controlled using Fuzzy logic technique. unit.
In the present paper a fast computation genetic algorithm is
B. Transmission system Losses
proposed based on lambda-based GA approach. The paper
presents a methodology to overcome long computation times To achieve true ELD transmission system losses must be
involved with simple Genetic Algorithms. The fast taken into account, as power generating units are spread over
computation feature of the developed algorithm is large areas. Using B-coefficients method, the network losses
advantageous and can be used in on line power system are expressed as a quadratic function of unit generations as
studies.
NG NG
Ploss = ∑∑ PGi BijPGj (5)
II. ECONOMIC LOAD DISPATCH i=1 j=1

A. Problem Formulation In (5) Bij are called as B-coefficients or loss coefficients.


The objective of Economic Load Dispatch (ELD) for power
system consisting of thermal generating units is to find the III. Simple Genetic Algorithm (SGA)
optimal combination of power generations that minimizes the
A. Representing a solution in GA
total generation cost while satisfying the specified equality
and inequality constraints. The fuel cost function of the To find the optimal decision variables, to optimize an
generator is represented as a quadratic function of generator objective function and to satisfy the constraints, the variables
active powers. are represented in binary strings. Eqn.(6) gives a mapping
function defined to [1] take care of variable bounds. In (6) ‘li’
is the string length used to code ith variable, DV(si) is the
The minimization function J can be obtained as sum of the
decoded value of the string si. Ximin and Ximax are lower and
fuel costs Fi of all the generating units.
upper bounds of variable ‘xi’.
NG
Min J= ∑ Fi (1) ⎛ x max -x min ⎞ (6)
i=1 x i =x imin + ⎜ i li i ⎟ DV(si )
⎝ 2 -1 ⎠
Subjected to
Variable bounds are taken care by the mapping function given
NG
(2) by (6). After choosing a string representation, a random
∑P
i=1
Gi =PD +Ploss population of solutions is generated. Then fitness is assigned
to each population member, which represents the goodness of
PGimin ≤ PGi ≤ PGimax (3)
each solution. The string is evaluated in the context of
objective function and constraints. In the absence of
In (1) fuel cost of generating unit is given by constraints, the objective function is treated as fitness
2
Fi = a i + bi PGi + ciPGi (4) function. In case of a minimization problem, the solution with
smaller fitness is better.
Where ai, bi, ci are cost coefficients of unit i, PGi is real power
generation of unit ‘i’. Genetic Operators: The SGA uses reproduction operator,
elitism operator, crossover operator and mutation operator.
This is a constrained optimization problem, that may be B. Reproduction operator
solved using [9] advanced calculus methods that involve The main goal of reproduction operator is to make
Legrange function. The necessary condition for the existence duplicates of the best fit solutions in the population and
of minimum cost operating condition for thermal power eliminate least fit solutions, while keeping the population size
system is that the incremental cost rates of all the units be constant. Some commonly used methods include tournament
equal to some undetermined value Lambda (λ). Along with selection, proportionate selection or Roulette Wheel Selection
the necessary condition, the equality constraint, the sum of the (RWS), and ranking selection. In tournament selection
power outputs must be equal to the total power demand by the tournaments are played between two solutions and the better
load plus losses, must be satisfied. If transmission system solution is chosen and placed in the mating pool. Two other
losses are neglected, the equality constraint becomes, the sum solutions are selected again randomly and another slot in the
of the power outputs must be equal to the total power demand mating pool is filled with a better solution. Each solution can
350

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

take part in the tournament twice. The process is repeated until E. Mutation Operator
the mating pool is filled with new good quality solutions. The crossover operator is mainly responsible for bringing
diversity in the population; mutation operator is also used for
In proportionate selection, solutions are assigned copies bringing further diversity in the population. The bit wise
proportional to their fitness values. It can be thought of as a mutation operator flips the selected bit from 1 to 0 and vice
roulette-wheel mechanism, where the wheel is divided into N versa, with a mutation probability Pm.
(population size) divisions, where the size of each is marked SGA is very simple and straightforward. Reproduction
in proportion to the fitness of each population member. operator selects good strings; crossover operator recombines
Thereafter the wheel is spun ‘N’ times, each time selecting them to form two better strings. The mutation operator alters
the solution indicated by the pointer. In RWS a string with a the strings locally. If bad strings are created in crossover they
higher fitness has a higher probability of being copied into the
are eliminated by the reproduction operator. GAs with these
mating pool. In RWS the number of copies a population
simple operators, constitute a potential search and
member gets in the next population is equal to fi/favg. This
selection operator is slow compared to tournament selection, optimization algorithms.
as it merely calculates the average fitness all population
members. Fitness scaling techniques are used to enhance the IV. ELD using SGA and FGA
performance of RWS operator. For solution of ELD using SGA, incremental fuel cost of
Ranking selection ranks the population members according the generators i.e. lambda is encoded in the chromosome. The
to their fitness, from worst (rank1) to best (rank N), where N algorithm for implementing ELD without losses using Simple
is the population size. Each member in the sorted list is GA is as follows.
assigned fitness equal to the rank of the solution in the list.
Thereafter the proportional selection operator is applied with 1. Read population size, chromosome length, unit data,
the ranked fitness values and N solutions are chosen from Pdemand, Probability of Elitism, Crossover and mutation
mating pool. 2. Randomly generate population of chromosomes.
C. Crossover operator 3. Decode the chromosomes using (6).
The reproduction operator makes copies of the best fit 4. Lambda_act = Lambda_min + (Lambda_max -
solutions in the next generation, but it cannot create any Lambda_min)* Decoded lambda
diversity in the population members. It only makes more 5. Use the Lambda_act and cost coefficients of the
copies of best fit solutions at the cost of least fit solutions. generators, and calculate real power output of the
Crossover and mutation operators create new solutions in the generators (Pgen).
population. There exist a number of crossover operators in the 6. Calculate the error of each chromosome as (Sum of Pgen)
literature, but in all the operators, two solutions are randomly – Pdemand.
selected from the mating pool. These are called parent 7. Fitness(i) of each chromosome is calculated as
chromosomes. Then a crossover site is randomly selected and 1/(1+error(i)/Pdemand).
a portion of the string is exchanged between the two parent 8. Arrange the chromosomes in the descending order of
chromosomes. These are called child chromosomes. The their fitness.
process is repeated until the population is filled with new 9. Check if error(1) ≤ 0.0001*Pdemand
chromosomes. This is the concept behind single point and 10. If yes STOP and calculate Optimal fuel cost and Pgen of
multi point crossover operators. Another type of crossover units
known as uniform crossover creates a random mask and Else
generates the child chromosomes from parents using the bit
11. Check if Fitness(1)=Fitness(last chromosome)
value in the mask.
12. If yes print ‘All chromosomes have equal values’,
D. Elitism Operator calculate Optimal fuel cost and Pgen of units and STOP.
In order not to loose the good solutions from the population Else
elitism operator is used. This operator prevents all population 13. Apply elitism, Reproduction (RWS), crossover and
members from undergoing crossover operation. If a crossover mutation and generate new population from old one.
probability of Pc is used, 100Pc% of population undergoes 14. Update generation count.
crossover, whereas 100(1-Pc)% of population is copied to 15. Check if Generation count > maximum generations?
next population without any modification. This operator 16. If yes, print ‘Problem not converged in maximum number
preserves the best fit chromosomes in the current population, of generations’, STOP.
and prevents them from being modified during the crossover Else
operation. 17. Repeat from step 3.

351

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

For ELD using Fast GA incremental fuel cost of the identify the chromosome with minimum negative error, and
generators i.e. Lambda is encoded in the chromosome. The set the lambda_act of this chromosome to be Lambda_min.
algorithm for implementing ELD without losses using Fast This will largely reduce the search space from wide
GA is as follows. Lamdba_max, Lambda_min to small region. The remaining
steps are just same as for simple GA. Since the search space is
1. Read population size, chromosome length, unit data, now a very narrow region between the maximum and
Pdemand, Probability of Elitism, Crossover and mutation minimum lambda values, the algorithm converges in two or
2. Randomly generate population of chromosomes. three generations. For ELD considering transmission losses, in
3. Decode the chromosomes using eqn. (6). step 5 and 6 of the algorithms the losses are accounted for
4. Lambda_act = Lambda_min + (Lambda_max - while calculating the real power output of generators and then
Lambda_min) * Decoded lambda error of each chromosome is evaluated as [(sum of Pgen)-
5. Use the Lambda_act and cost coefficients of the Pdemand-Ploss]. The other steps remain same. The present
generators, and calculate real power output of the paper uses loss coefficients for calculation of losses.
generators (Pgen).
6. Calculate the error of each chromosome as (Sum of Pgen) V. RESULTS AND DISCUSSION
– Pdemand. The developed algorithm is tested on standard 3 generator,
7. Identify Minimum positive error and corresponding 6generator, 20 generator and 38generator systems.
chromosome no.(I1) For every case the chromosome length, population size,
8. Identify Minimum negative error and corresponding Probability of crossover, Mutation and elitism considered with
chromosome no.(I2) Simple GA and Fast GA are same. Roulette Wheel Parent
9. Set Lambda_max= Lambda_act(I1) selection technique, Uniform crossover and bit-wise mutation
10. Set Lambda_min=Lambda_act(I2) are used in all cases. Data for 38 generator system is presented
11. For all chromosomes calculate Lambda_act(chrom) = in the appendix.
Lambda_min+(Lambda_max-Lambda_min)*Decoded Tables I to IV present the test system results without
lambda consideration of system losses. Tables V to VIII provide the
12. Calculate Pgen from Lambda_act(chrom) test system results while taking system losses into
13. Error(chrom)=(Sum of Pgen)- Pdemand consideration.
14. Fitness(i) of each chromosome is calculated as Tables I presents the results of standard 3 generator system
1/(1+Error(i)/Pdemand). [9] for a power demand of 850MW without considering
15. Arrange the chromosomes in the descending order of transmission losses. The results obtained are in agreement
their fitness. with the results shown in [9]. Further, for the same GA
16. Check if error(1) ≤ 0.0001*Pdemand parameters FGA locates the optimal solution in just one
17. If yes STOP and calculate Optimal fuel cost and Pgen of iteration. The GA parameters considered for this case are
units Population size: 30, Chromosome length: 16 bits, max no of
Else generations: 100, Elitism probability: 0.15, Crossover
18. Check if Fitness(1)=Fitness(last chromosome) probability: 0.9, Mutation probability: 0.01.
19. If yes print ‘All chromosomes have equal values’,
calculate Optimal fuel cost and Pgen of units and STOP. TABLE I
Else 3 GENERATOR SYSTEM WITHOUT LOSS
20. Apply elitism, Reproduction (RWS), crossover and Simple GA Fast GA
mutation and generate new population from old one. Pdemand (MW) 850
21. Update generation count. Pg1 (MW) 393.26 393.09
22. Check if Generation count > maximum generations? Pg2 (MW) 334.67 334.54
23. If yes, print ‘Problem not converged in maximum Pg3 (MW) 122.25 122.2
FC ($/hr) 8196.0 8192.7
number of generations’, STOP. λ($/MWh) 9.148 9.148
Else Repeat from step 3. No. of 2 1
Generations
Upto step 6 both algorithms are same. In the fast GA, Time (Sec.) 0.078 0.062
after the errors of all chromosomes are evaluated, the search
space is restricted. This is done by identifying the
chromosome which has minimum positive error, and setting
the lambda_act of this chromosome to be Lambda_max. Then,

352

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

TABLE II Table IV presents results obtained for 38 generator


6GENERATOR SYSTEM WITHOUT LOSS
system for power demands of 7500, 8600 MW without losses.
Simple GA Fast GA The population size: 60, string length: 16, max no of
Pdemand generations: 100, elitism probability: 0.12, crossover
(MW) 283.4 probability: 0.92, mutation probability: 0.01. For the two cases
Pg1 (MW) 46.908 46.84
considered FGA provides better optimal solution in just one
Pg2 (MW) 19.134 19.11
Pg3 (MW) 10 10
generation.
Pg4 (MW) 10 10
TABLE IV
Pg5 (MW) 12 12
38 GENERATOR SYSTEM WITHOUT LOSSES
Pg6 (MW) 185.6 185.24
FC ($/hr) 768.33 766.88 Method Simple GA Fast GA Simple GA Fast GA
λ($/MWh) 3.3918 3.3892 Pdemand 7500 8600
No. of 9 1 Pg1 550 550 550 550
Generations Pg2 550 550 550 550
Time (Sec.) 0.109 0.016 Pg3 500 500 500 500
Pg4 500 500 500 500
Pg5 500 500 500 500
Tables II presents the results of standard 6 generator system Pg6 500 500 500 500
[10] for a power demand of 283.4MW. The GA parameters Pg7 500 500 500 500
Pg8 500 500 500 500
considered for this case are population size: 40, string length: Pg9 225.5833 222.9809 398.6468 398.5425
16, max no of generations: 100, elitism probability: 0.15, Pg10 225.5833 222.9809 398.6468 398.5425
Pg11 233.3335 230.8836 396.2643 396.1661
crossover probability: 0.95, mutation probability: 0.001.
Pg12 247.5648 244.9654 420.4329 420.3287
Tables III presents the results of 20 generator system [11] for Pg13 110 110 272.7205 272.5453
power demands of 1200, 2500 and 3600MW. The GA Pg14 90 90 176.0789 175.9355
Pg15 82 82 166.0387 165.909
parameters for this test system (without losses) are population Pg16 325 325 325 325
size: 40, string length: 16, max no of iterations: 100, elitism Pg17 192.5738 191.8624 239.8835 239.855
probability: 0.12, crossover probability: 0.98, mutation Pg18 65 65 65 65
Pg19 65 65 65 65
probability : 0.01. In all the cases the FGA shows reliable Pg20 272 272 272 272
convergence in just 1 iteration. Pg21 272 272 272 272
Pg22 260 260 260 260
Pg23 190 190 190 190
TABLE III Pg24 11.52467 11.48934 13.87377 13.87236
Pg25 125 125 125 125
20 GENERATOR SYSTEM WITHOUT LOSSES
Pg26 110 110 110 110
Pg27 65.26482 64.66595 75 75
SGA FGA SGA FGA SGA FGA
Pg28 20 20 20 20
Pd 1200 2500 3600
Pg29 20 20 20 20
Pg1 150 150 600 600 600 600
Pg30 20 20 20 20
Pg2 50 50 131.22 131.20 200 200
Pg31 20 20 20 20
Pg3 50 50 50 50 75.146 74.3918
Pg32 20 20 20 20
Pg4 50 50 50 50 167.69 166.709
Pg33 25 25 25 25
Pg5 50 50 91.215 91.213 160 160
Pg34 18 18 18 18
Pg6 20 20 20 20 100 100
Pg35 8 8 8 8
Pg7 65.53 65.56 125 125 125 125
Pg36 25 25 25 25
Pg8 50 50 50 50 114.20 113.597
Pg37 31.84392 31.74845 38 38
Pg9 50 50 112.67 112.67 200 200
Pg38 29.86693 29.78338 38 38
Pg10 30 30 45.928 45.925 150 150
FC($/hr) 10824602.06 10809797.2 12322082.6 12320766.9
Pg11 154.72 154.77 287.11 287.11 300 300 λ($/MWh) 1234.9 1231.2 1479.785 1479.64
Pg12 228.28 228.36 433.28 433.27 500 500 No. of 1 13 1
Pg13 47.962 47.989 122.72 122.72 160 160 generations 12
Pg14 20 20 73.027 73.023 130 130 Time 0.452 0.483 0.109
Pg15 25 25 93.760 93.756 185 185 (Sec) 4.087
Pg16 27.495 27.498 36.420 36.420 45.76 45.695
Pg17 30 30 30 30 85 85
Pg18
Tables V presents the results of standard 3 generator system
30 30 36.909 36.907 120 120
Pg19 40 40 78.483 78.481 120 120 [9] for a power demand of 850MW while considering
Pg20 30 30 30 30 63.836 63.201 transmission losses. The GA parameters considered for this
FC ($/hr) 35585.7 35589.0 82145.85 82082.65
60109.1 60108.1
λ($/MWh 18.175 18.176 20.77 20.76
system are population size: 40, string length: 16, max no of
) 19.446 19.446 generations:100, elitism probability: 0.15, crossover
No. of 6 1 3 1
generatio
ns 7 1
probability:0.95, mutation probability:0.001. The results
Time 0.343 0.016 1.638 0.062 1.076 0.094 obtained are in agreement with the results shown in [9].

353

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

TABLE V (MW)
3GENERATOR SYSTEM WITH LOSSES Pg1 (MW) 189.52 189.613
Simple GA Fast GA Pg2 (MW) 47.7244 47.745
Pdemand (MW) 850 Pg3 (MW) 19.5719 19.5761
Pg1 (MW) 435.219 435.199 Pg4 (MW) 13.8642 13.8752
Pg2 (MW) 299.983 299.970 Pg5 (MW) 10 10
Pg3 (MW) 130.667 130.660 Pg6 (MW) 12 12
FC ($/hr) 8344.96 8192.7 FC ($/hr) 799.384 799.823
λ($/MWh) 9.52844 9.5283 λ($/MWh) 3.80709 3.77345
No. of 10 2 No. of 8 2
Generations Generations
Loss 15.8304 15.829 Loss(MW) 9.68247 9.68968
Time (Sec.) 0.28 0.015 Time (Sec.) 0.483 0.125

18 25
FGA
FGA
16 SGA

SGA
14 20

12

10
15
Error

8
Error

6
10

2
5
0
1 2 3 4 5 6 7 8 9 10
No. of generations

0
1 2 3 4 5 6 7 8
Fig. 1. Variation of error of best fit chromosome with number of generations No. of generations

for 3 generator system with loss

Further, for the same GA parameters FGA locates better Fig. 2. Variation of error of best fit chromosome with number of generations
optimal solution in just 2 generations compared to 10 for 6generator system with loss
generations with SGA. Fig.1 shows the variation of error of
the best fit chromosome with number of generations. The Table VII provides the optimal generator outputs and
figure clearly demonstrates the superiority of the developed optimal fuel cost with losses for 20 generator system [11]. The
FGA over SGA. considered GA parameters are population size: 40, string
Table VII presents the generator power outputs, optimal length: 16, max no of generations: 100, elitism probability:
fuel cost and loss obtained for 6 generator system. The GA 0.15, crossover probability: 0.95 (Pd=1200), crossover
parameters considered are population size: 40, string length: probability: 0.9 (Pd=2500), crossover probability: 0.92
16, max no of generations: 100, Elitism probability: 0.15, (Pd=3600), mutation probability:0.001.
crossover probability: 0.95, mutation probability: 0.001. From the table it is very clear that, for all the cases
From the table it is clear that, for the same GA parameters considered, FGA provides the optimal solution in minimum
the proposed FGA presents the optimal solution in just 2 time. Further, for a power demand of 2500MW, SGA and
generations compared to 8 generations with SGA. Fig. 2. FGA provide a Fuel cost of 62153.8 ($/hr) and 62153.04($/hr)
shows the variation of error of the best fit chromosome with against 62456.63($/hr) as reported in [11].
respect to number of generations. From the figure it is obvious
that, in case of FGA algorithm the error is considerably
reduced in the first generation itself, so this leads to very fast TABLE VII
convergence to optimal solution. 20 GENERATOR SYSTEM WITH LOSSES

Gen in MW
TABLE VII SGA FGA SGA FGA SGA FGA
Pd 1200 MW 2500MW 3600MW
6GENERATOR SYSTEM WITH LOSSES
Pg1 150 150 393.8 393.8 600 600
Simple GA Fast GA Pg2 50 50 177.51 177.50 200 200
Pdemand 283.4 Pg3 50 50 50 50 169.78 169.52
Pg4 50 50 82.860 82.856 200 200
354

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

Pg5 50 50 110.04 110.0 160 160 VI. CONCLUSION


Pg6 20 20 49.755 49.72 100 100
Pg7 85.801 85.813 125 125 125 125
The paper presented a Fast computation algorithm for
Pg8 50 50 70.803 70.76 150 150 Economic Load dispatch problem. For the same Genetic
Pg9 50 50 169.72 169.7 200 200 operators and parameters, the proposed algorithm shows better
Pg10 30 30 99.97 99.96 150 150 and faster convergence over simple GA. Large computational
Pg11 152.07 152.086 245.54 245.5 300 300 times involved with GAs can be overcome by assessing the
Pg12 500
217.38 217.412 355.06 355.0 500 variation of error with the solution at hand. A number of test
Pg13 66.309 66.3193 128.11 128.1 160 160
cases have been studied and the algorithm showed reliable
Pg14 20 20 119.61 119.60 130 130
Pg15 185
convergence. The developed algorithm can be used for real
25 25 113.43 113.03 185
Pg16 30.734 30.736 39.833 39.83 57.315 57.277
time ELD problems, as the solution accuracy is achieved in
Pg17 30 30 45.319 45.32 85 85 minimum time. The only limitation of the algorithm is, if the
Pg18 30 30 78.269 78.27 120 120 solution represents a single variable (lambda in this case), the
Pg19 40 150 105.51 105.5 120 600 error variation can be easily assessed with respect to this
Pg20 30 50 30 30 100 200 single variable. But if the chromosome represents a number of
FC ($/hr) 36107.9 36108.3 62153.8 62153.04 86612.1 86605.3
λ($/MWh) 18.79 18.79 20.139 20.138 22.8 22.74
variables, the error variation with respect to all the variables
No. of 4 2 3 2 cannot be predicted and it becomes difficult to restrict the
generations 3 2
Loss (MW) 27.99 28 115.68 115.68 211.717 211.716 search space to small regions.
Time 0.172 0.076 0.125 0.086 0.109 0.09
APPENDIX
38 GENERATOR SYSTEM DATA
160
FGA Gen.No. a b c Pmin Pmax
SGA
140 1. 64782 796.9 0.3133 220 550
2. 64782 796.9 0.3133 220 550
120 3. 64670 795.5 0.3127 200 500
4. 64670 795.5 0.3127 200 500
100 5. 64670 795.5 0.3127 200 500
6. 64670 795.5 0.3127 200 500
7. 64670 795.5 0.3127 200 500
Error

80

8. 64670 795.5 0.3127 200 500


60 9. 172832 915.7 0.7075 114 500
10. 172832 915.7 0.7075 114 500
40 11. 176003 884.2 0.7515 114 500
12. 173028 884.2 0.7083 114 500
20 13. 91340 1250.1 0.4211 110 500
14. 63440 1298.6 0.5145 90 365
0
1 1.5 2 2.5 3 3.5 4 4.5 5 15. 65468 1290.8 0.5691 82 365
No. of generations
16. 72282 190.8 0.5691 120 325
17. 190928 238.1 2.5881 66 315
Fig. 3. Variation of error of best fit chromosome with number of generations
18. 285372 1149.5 3.8734 65 315
for 20 generator system with loss, Pd=1200
19. 271376 1269.1 3.6842 65 315
Fig. 3 and 4 display the variation of error of the best fit 20. 39197 696.1 0.4921 120 272
chromosome with number of generations for a power demand 21. 45576 660.2 0.5728 120 272
of 1200MW and 2500Mw respectively. Even here the FGA 22. 28770 803.2 0.3572 110 260
shows its superiority over SGA. 23. 36902 818.2 0.9415 80 190
24. 105510 33.5 52.123 10 150
14
FGA 25. 22233 805.4 1.1421 60 125
12
SGA 26. 30953 707.1 2.0275 55 110
27. 17044 833.6 3.0744 35 75
10
28. 81079 2188.7 16.765 20 70
8
29. 124767 1024.4 26.355 20 70
30. 121915 837.1 30.575 20 70
Error

6 31. 120780 1305.2 25.098 20 70


32. 104441 716.6 33.722 20 60
4
33. 83224 1633.9 23.915 25 60
2 34. 111281 969.6 32.562 18 60
35. 64142 2625.8 18.362 8 60
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 36. 103519 1633.9 23.915 25 60
No. of generations
37. 13547 694.7 8.482 20 38
Fig. 4. Variation of error of best fit chromosome with number of generations 38. 13518 655.9 9.693 20 38
for 20 generator system with loss, Pd=3600

355

© 2009 ACADEMY PUBLISHER


RESEARCH PAPER
International Journal of Recent Trends in Engineering Vol. 1, No. 1, May 2009,

REFERENCES
[1] Kalyanmoy Deb, “Multi-objective Optimization using Evolutionary
algorithms”, John Wiley and Sons, 2001.
[2] Anastasios G.Bakirtzis, Pandel N.Biskas, Christoforos E.Zoumas,
Vasilios Petridis, “Optimal Power Flow by Enhanced Genetic
Algorithm”, IEEE Transactions on Power Systems, Vol.17, No.2, May
2002, pp.229-236.
[3] Sheble, G.B. and K. Brittig, “Refined genetic algorithm-economic
dispatch example”, IEEE paper 94 WM 199-0 PWRS, presented at the
IEEE/PES 1994 winter meeting.
[4] Walters, D.C. and G.B.Sheble, “Genetic algorithm solution of Economic
dispatch with valve point loading”, IEEE Trans. on Power Systems, Vol.
8, No. 3, pp. 1325-1332, August 1993.
[5] Po-Hung Chen and Hong-Chan Chang, “Large-Scale Economic
Dispatch by Genetic Algorithm”, IEEE Trans. On Power Systems, Vol.
10, No.4, November 1995.
[6] Chao-Hong Chen and Ying-ping Chen , “Real Coded ECGA for
Economic Dispatch”, GECCO ’07, July 7-11, 2007, London, England,
United Kingdom.
[7] P.Venkatesh, R.Gnanadass and Narayana Prasad Padhy, “Comparison
and Application of Evolutionary Programming Techniques to combined
Economic Emission Dispatch with Line flow Constraints”, IEEE
Transactions on Power Systems, Vol 18, No.2, May 2003.
[8] A.Laoufi, A.Hazzab and M.Rahli, “Economic Dispatch using Fuzzy-
Genetic Algorithm”, International Journal of Applied Engineering
Research”, ISSN 0973-4562 Vol.1, No.3 (2006).
[9] Allen J.Wood and Bruce F.Wollenberg, “Power Generation Operation
and control”, A wiley-Interscience Publication, John Wiley & Sons, INC,
1996.
[10] Alsac,O. and Stott.B, “Optimal Load Flow with Steady State Security”,
IEEE Transactions on Power Apparats and Systems, Vol.93, No.3, 1974.
[11] Ching-Tzong Su and Chien-Tung Lin, “New approach with a Hopfield
Modeling Framework to Economic Dispatch”, IEEE Transactions on
Power Systems, Vol. 15, No. 2, May 2000.

356

© 2009 ACADEMY PUBLISHER

You might also like