A COMBINATION OF GENETIC ALGORITHM AND SIMULATED ANNEALING

FOR OPTIMAL DG ALLOCATION IN DISTRIBUTION NETWORKS

M. Gandomkar M. Vakilian M. Ehsan
Saveh Azad University Sharif University of Technology Sharif University of Technology
IRAN IRAN IRAN
email: [email protected] email: [email protected] email: [email protected]

Abstract storage energy devices such as batteries. Both the distribution

company and/or the customer can, in principle, invest in and
This paper presents a new algorithm based on integrating operate units. Due to the availability of such a flexible option
the use of genetic algorithms and simulated annealing methods of DG as an energy source at the distribution voltage level, the
to optimal allocation of distributed generation resources in distribution network is now being transformed from a passive
distribution networks. Through this algorithm a significant network to an active one. The DG as an energy source in the
improvement in the optimization goal is achieved. With a distribution network will play a significant role in operation,
numerical example the superiority of the proposed algorithm is structure, design and up gradation issues. DG technologies,
demonstrated in comparison with the simple genetic algorithm. their benefits and concepts, and their valuable effect on the
electricity market make it a credible alternative in the
Keywords: distributed generation, distribution networks, distribution system planning. The importance of DG is now
genetic algorithm, simulated annealing. being increasingly accepted and realize by power engineers.
From distribution system planning point of view, DG is a
feasible alternative for new capacity especially in the
1. Introduction competitive electricity market environment and has immense
benefit such as [2-3]:
The utilities requirements regarding flexible electric - Short lead-time and low investment risk since it is built in
systems, energy saving, loss reduction and environmental modules.
impact are providing impetus to the development of Distributed - Small-capacity modules that can track load variation more
Generation (DG). DG includes the application of small closely.
generators, scattered throughout a power system, to provide the - Small physical size that can be installed at load centers and
electricity service required by customers. The DG allocation does not need government approval or search for utility
can be obtained by a complete enumeration of all feasible territory and land availability.
combinations of sites and sizes of DGs in the network. Since - Existence of a vast range of DG technologies
the number of alternatives could be very large, the load flow For these reasons, the first signs of a possible technological
analysis should be performed for every feasible network change are beginning to arise on the international scene, which
combination. Of course the optimized solution among these could involve in the future the presence of a consistently
alternatives should be selected [1-3]. generation produced with small and medium size plants
The artificial intelligence techniques are the most widely directly connected to the distribution network (LV and MV)
employed tool for solving most of the optimization problems. and characterized by good efficiencies and low emissions. This
These methods (e.g. genetic algorithm, simulated annealing will create new problems and probably the need of new tools
and tabu search) seem to be promising and are still evolving. and managing these systems.
The published are on the DG allocation by application of
genetic algorithm (GA)[4,5]. Tabu search (TS) algorithm is 3. Problem Formulation
used for DG allocation in distribution systems[6]. A new
hybrid algorithm for evaluation of the DG site and size in MV The problem is to determine allocation and size of DGs
networks is proposed. The GA and SA are employed for DG which minimizes the distribution power losses for a fixed
allocation. The results showed that the proposed combined number of DGs and an specific total capacity of DGs.
GA-SA method is better than the simple GA in terms of Therefore the following assumptions are employed in this
solution quality and number of iteration. formulation [6, 7]:
- The maximum number of installable DGs is given (D ) .
2. Distributed Generation
- The total installation capacity of DGs is given (Q ) .
DG can be defined as a small-scale generating unit located - The possible locations for DG installation are given for each
close to the load being served. A wide variety of DG feeder.
technologies and types exists: renewable energy source such as - The upper and lower limits of node voltages are given.
wind turbines, photovoltaic, micro-turbines, fuel cells, and - The current capacities of conductors are given.

0-7803-8886-0/05/$20.00 ©2005 IEEE 645 Authors Absent - Paper Not Presented

CCECE/CCGEI, Saskatoon, May 2005

Authorized licensed use limited to: UNIVERSIDADE TECNICA DE LISBOA. Downloaded on May 15, 2009 at 10:43 from IEEE Xplore. Restrictions apply.
The objective function in this optimization problem is: C : total injected dispersed generation for network,
n 'V : maximum permissible voltage deviation,
f ¦P i 1
i (1)
I i : current of section i ,
Where Pi is the nodal injected power at bus i , and n is the I imax : maximum current capacity of section i .
total number of buses.
If the total injected power of distributed generation was 4. Genetic Algorithm
constant as C MW, this equality constraint should be
expressed in form of a penalty function as shown [7]: Genetic Algorithm is a general-purpose search techniques
n L based on principles inspired from the genetic and evolution
f ¦ P  D (¦ P
i 1
i
k 1
k  C) (2) mechanisms observed in natural systems and populations of
living beings. Their basic principle is the maintenance of a
Constraints: population of solutions to a problem (genotypes) as encoded
information individuals that evolve in time [8].
Maximum number of DGs: Generally, GA comprises three different phases of search:
M N phase 1: creating an initial population; phase 2: evaluating a
¦¦ n
l 1 g 1
gl dD (3) fitness function; phase 3: producing a new population.
A genetic search starts with a randomly generated initial
Total capacity of DGs: population within which each individual is evaluated by means
M N of a fitness function. Individual in this and subsequent
¦¦ G
l 1 g 1
g .n g l d Q (4) generations are duplicated or eliminated according to their
fitness values. Further generations are created by applying GA
One DG per installation position: operators. This eventually leads to a generation of high
N performing individuals [9].
¦n
g 1
gl d1 (5) There are usually three operators in a typical genetic
algorithm [8]: the first is the production operator (elitism)
Upper and lower voltage limits: which makes one or more copies of any individual that posses a
high fitness value; otherwise, the individual is eliminated from
Vi d Vn r 'V (6)
the solution pool; the second operator is the recombination
Current capacity limits: (also known as the ' crossover' ) operator. This operator selects
I i d I imax (7 ) two individuals within the generation and a crossover site and
carries out a swapping operation of the string bits to the right
(k 1,2,..., L, l 1,2,..., M , g 1,2,..., N ) hand side of the crossover site of both individuals. Crossover
Where, operations synthesize bits of knowledge gained from both
Pi : nodal injection of power at bus i , parents exhibiting better than average performance. Thus, the
probability of a better offspring is greatly enhanced; the third
Pk : load power of bus k , operator is the 'mutation' operator. This operator acts as a
Vi : magnitude of voltage of bus i , background operator and is used to explore some of the
invested points in the search space by randomly flipping a 'bit'
Vn : nominal magnitude of voltage in the network, in a population of strings. Since frequent application of this
operator would lead to a completely random search, a very low
G g : capacity of g th DG, probability is usually assigned to its activation.
n g l : 0-1 variable for determining whether one DG with
5. Simulated Annealing
g th capacity is allocated at l th location(1: allocated, 0: not
allocated), Simulated Annealing (SA) is a metaheuristic which has been
L : total number of load buses , successfully applied to solve a variety of difficult optimization
M : total number of DG location candidates, problems [10]. It is based on an analogy between statistical
mechanics systems and combinatorial optimization problems.
N : total number of capacity types of DGs, The term annealing refers to the process of cooling after
Q : total installation capacity of DGs, heating, in order to toughen and temper the material. SA is a
D : maximum number of installable DGs , Monte Carlo approach that simulates this process. A solution to
a combinatorial optimization problem is seen as a possible state
D : penalty weight of equality constraint, of a thermo dynamical system which is submitted to slow down
the cooling, starting from an initial high temperature.

646

Authorized licensed use limited to: UNIVERSIDADE TECNICA DE LISBOA. Downloaded on May 15, 2009 at 10:43 from IEEE Xplore. Restrictions apply.
In SA the cost of a solution is equivalent to the energy of the
START
physical state, and the temperature, although it has no physical
meaning, can be seen as controlling the entropy of the system.
Create initial population
At high temperature, all solutions to the optimization problem
are equally likely while, at low temperature only the minimal
Load flow& Power
cost solutions are accepted. Following an appropriate cooling Loss calculation
schedule, SA has the potential to avoid local minima and
converges to the global optimal solutions within a reasonable Calculate objective
function
computing time [10].
In the implementation of SA, starting from an initial
Applying Crossover
configuration, new ones are proposed through local changes,
called moves, and accepted within the probability
­1 if f x d f Yes
Is accepted
according to eq.8?
No
°
P ® § f  f x· x
(8)
°exp ¨¨ T ¸¸ if f ! f New chromosomes=child New chromosomes=parents

¯ © ¹
Applying Mutation

where f and f x are the costs of the current and proposed

Generate next generation
configurations, respectively, and T is the temperature. The
algorithm is run until a stopping condition is reached, typically
a minimum temperature value, specified as part of the Is stopping criteria
No Update
Parameters
satisfied?
annealing schedule.
Yes

END
6. The Proposed Algorithm
The major steps of the proposed algorithm are: Fig. 1. Flowchart of algorithm.
Step1: Initialize the variables of GA and SA
Step2: Creating an initial population by randomly generating a
set of feasible solutions (chromosomes); Selection of chromosomes for applying various GA
Step3: Evaluating each chromosome by running the load flow operators is based on their scaled fitness function in accordance
program; to the roulette wheel selection rule. The roulette wheel slots are
Step4: Determining the fitness function of each chromosome in sized according to the accumulated probabilities of reproducing
the population (1/Power Losses); each chromosome. Crossover and mutation operators are
Step5: Apply the crossover operator, the child chromosomes carried with the pre-specified probabilities, but the resultant
x chromosomes of crossover operator are accepted according to
are accepted according to eq. 8, where f and f are the
step5 in the proposed algorithm steps. Then the total number of
fitness values of parent and child chromosomes; installable DGs and the maximum capacity of installable DGs
Step6: Apply the mutation operator to the new population; are checked, and subsequently the necessary corrections will
Step7: Let the current population be the new population; be made.
Step8: If the convergence criterion is satisfied, stop. Otherwise
go to step 3.
Figure 1 shows flowchart of the proposed algorithm. 7. Numerical Example
The solution of the problem is represented by a binary
coding. As in the example, 4 types of DGs are used for each
MV candidate bus, 3 bits are considered for coding: one for In order to test the proposed algorithm, the 34-node IEEE
presenting the DG on bus and 2 bits for type of DG. distribution test feeder has been considered [12]. A number of
The fitness value of each chromosome is the reverse of tests on the performance of the proposed algorithm have been
power loss for the related network. So the proper load flow carried out on the example to determine the most suitable GA
program [11] is prepared and it is run for each network. Output and SA parameters setting. Table 1 shows the control
of the program is the fitness value of chromosome (1/Power parameters which have been chosen after running a number of
Loss). simulations.

647

Authorized licensed use limited to: UNIVERSIDADE TECNICA DE LISBOA. Downloaded on May 15, 2009 at 10:43 from IEEE Xplore. Restrictions apply.
 Table 1. GA& SA setting parameters
Population size 50 Table 2. Comparison of SGA &GA-SA results
Crossover Probability 0.82 SGA GA-SA
Mutation Probability 0.04 Total Capacity of Installed DGs (kW) 200 200
Initial Temperature 10000 Power Losses (kW) with DG 213.2 201.8
Percent of Reduced Power Losses 21.9 26.1
we assumed DG can be installed on all of the nodes in the
example network. Furthermore, maximum total capacity
(Q ) and maximum number ( D ) of DGs assumed about 0.1 References
power demand and one-third of nodes number for each [1] Willis, H.L., and W.G. Scott, DISTRIBUTED POWER
network. Therefore, Q, D are considered 200 kW and, GENERATION. New York: Marcel Dekker, 2000.
11respectively. [2] N. Hadisaid, J.F. Canard, and F. Dumas, "Dispersed
Different experiments with different random number seeds generation impact on distribution networks," IEEE
were carried out to investigate the performance of the proposed Transactions on Computer Applications on Power, vol. 12,
algorithm. It was found that the proposed algorithm performs no. 2, pp. 22-28, April, 1999.
in a better form , in terms of both solution quality and number [3] CIGRE WG 37-23, "Impact of increasing contribution of
of iterations, than the individuals SGA . Table 2 shows the
comparison of the results for GA-SA and SGA for the test distributed generation on the power system," Final Report,
network. According to [12] the test network has 273 kW power Electra, 1998.
losses without DG. [4] G. Celli, and F. Pillo, "Optimal distributed generation
Figure 2 shows the convergence process of the GA-SA and allocation in MV distribution networks," Proceedings of
SGA when employed to solve the optimization problem of this the IEEE International Conference on Power Engineering
test network. Society, pp. 81-86, May, 2001.
[5] A. Silversti, and S. Buonaao, "Distributed generation
8. Conclusion planning using genetic algorithm," Proceedings of the IEEE
International Conference on Power Tech, pp. 257,
In this paper the results of application of GA-SA algorithm September, 1997.
to the optimal allocation of DGs in distribution network is [6] K. Nara, Y. Hayashi, K. Ikeda, and T. Ashizawa,
presented. The effectiveness of the proposed algorithm to solve "Application of Tabu Search to optimal placement of
the DG allocation problem is demonstrated through a
distributed generators," Proceedings of the IEEE Power
numerical example. The IEEE 34-node distribution test feeders
Engineering Society, vol. 2, pp. 918-923, February, 2001.
have been solved with the proposed algorithm and, the simple
genetic algorithm. The results demonstrated the better [7] J.O. Kim, S.W. Nam, S.K. Park, and C. Singh, "Distributed
characteristics of the GA-SA algorithm in comparison with the generation planning using improved hereford ranch
SGA specially in terms of; solution quality and number of algorithm," Electric Power System Research, vol. 47, pp.
iterations. 47-55, 1997.
[8] Goldberg D.E., GENETIC ALGORITHM IN SEARCH,
OPTIMIZATION AND MACHINE LEARNING, MA:
280
Addison Wesley, 1989.
270 [9] Davis L., HANDBOOK OF GENETIC ALGORITHM,
260
New York: Van Nostrand, 1991.
GA-SA [10] Laarhoven, P.J.M.V. and E.H.L. Aarts, SIMULATED
Power Losses(kW)

250 ANNEALING: THEORY AND APPLICATIONS,

GA
240
Dordrech:Reidel Publishing Company, 1987.
[11] C.S. Cheng, and D. Shirmohammadi, "A three-phase
230 power flow method for real-time distribution system
220 analysis," IEEE Transactions on Power Delivery, vol. 10,
no.2, pp. 671-679, May 1995.
210
[12] IEEE distribution test feeders;http://ewh.ieee.org
200 /soc/pes/dsacom/testfeeders.htm.
0 100 200 300 400 500 600 700 800 900 1000
Generation Number

Fig. 2. Power losses reduction with generation number

648

Authorized licensed use limited to: UNIVERSIDADE TECNICA DE LISBOA. Downloaded on May 15, 2009 at 10:43 from IEEE Xplore. Restrictions apply.