432
Research Article
International Journal of Current Engineering and Technology
ISSN 2277 - 4106
© 2012 INPRESSCO. All Rights Reserved.
Available at http://inpressco.com/category/ijcet
Performance Improvement of Differential Evolutionary Algorithm: A Survey
Vanita G. Tonge
a*
and P.S.Kulkarni
a
a
Department of Information Technology, RCERT, Chandrpur
Accepted 4 Nov.2012, Available online 27Dec. 2012, Vol.2, No.4(Dec. 2012)
Abstract
Differential evolution is one of the latest evolutionary optimization algorithm applied to continuous optimization
problems. Differential evolution for solving flow shop problem that belongs to the class of scheduling problems. The
scheduling problems arise in diverse areas such as manufacturing systems, production planning, computer design,
logistics etc.. Only in very special cases there exist exact polynomial algorithms to reach optimal solution. To solve
PFSP problem many researchers have used different EVs algorithm such as ACO, PSO, GA. Out of such evolutionary
algorithm DE is the best solution.
Keywords: Differential Evolution, recombination, PSO, ACO, perturbation, PFSP.
1. Introduction
1
Evolutionary Computation (EC) uses ideas inspired from
the behavior of community-based animals such as ants,
bees and birds. Furthermore researchers proposed
algorithms with variants of DE. Differential Evolution
(DE) algorithm has emerged as a very competitive form of
evolutionary computing more than a decade ago.
First a donor vector (V
i,t
) is created in step a and then a
trial vector (U
i,t
) is created in step b by stochastically
combining elements from X
i,t
and V
i,t
. This combination is
commonly done using an exponential distribution with
crossover factor of CR. If the newly created child U
i,t
is
better compared to X
i,t
then U
i,t
is stored for updating
X
i,t+1
. It should be noted carefully that X
i,ts
are updated to
X
i,t+1s
after entire set of U
i,ts
are created. Once the
population is updated, the generation counter is
incremented and a termination criterion is checked.
Following properties of this DE should be noted: (i)
There is „elitism‟ at an individual level i.e. if the newly
created trial vector Ui,t is inferior compared to the
individual then individual is preserved as a child for the
next generation and Vi,t is ignored. (ii) The algorithm
follows a generational model i.e. the current population is
updated only after the entire offspring population is
created. One of the noticeable features of standard DE is
elitism at the individual level i.e. a child is compared with
its base parent (i.e. the individual at the index
corresponding to which child has been created), and only
the better of the two survives for the next generation. We
modified this Replacement scheme by always accepting
the newly created child i.e. without carrying out the
parent-child comparison. This resulted in significant
* Vanita G. Tonge is M.Tech Scholar and Prof.P.S.Kulkarni is the
Project Guide.
performance degradation in all three test problems with
respect to both the metrics, indicating that elitism in DE by
parent-child comparison is key to its performance. (Nikhil
et al,2010)
DE has several advantages: it can search randomly,
requires only fewer parameters setting, high performance
and applicable to high-dimensional complex optimization
problems. But similar to PSO, DE has several drawbacks
including unstable convergence in the last period and easy
to drop into regional optimum. Compared with else
evolutionary computation, DE and PSO have advantage
respectively (Ying-Chih et al,2009).
Performance of DE is based on its control parameters.
Researchers have provided the various techniques to
improve the performance of DE. We will see it in next
section. The remaining paper is organized as follows.
Section 2 introduces the differential evolution (DE)
algorithm. Section 3 present PFSP problems, Section 4
gives brief review on performance improvement of DE,
proposed plan is discussed in Section 5.Finally, Section 6
summarizes the concluding remark.
2. Differential Evolution (DE)
Differential Evolution is announced in 1995 by Price and
Storn, and its superior performance in solving complex
problems.( D. G. Mayer et al,2005).DE is based on
individual‟s difference, utilize random research in solution
space, further utilize the mechanism “mutation”,
“recombination”, “selection” to compute every individuals
to obtain appropriate individual. DE used the information
between the difference in individuals to lead to search,
thus the result of search is more unstable(Z.-F. Hao et,
2007).The advantage and weakness of DE is given as
follows.