Sie sind auf Seite 1von 2

A Modified Binary Differential Evolution Algorithm

Ling Wang, Xiping Fu, Muhammad Ilyas Menhas, and Minrui Fei

Shanghai Key Laboratory of Power Station Automation Technology,


School of Mechatronics and Automation, Shanghai University,
200072, Shanghai, China
wangling@shu.edu.cn

Abstract. Differential evolution (DE) is a simple, yet efficient global optimiza-


tion algorithm. As the standard DE and most of its variants operate in the
continuous space, this paper presents a modified binary differential evolution
algorithm (MBDE) to tackle the binary-coded optimization problems. A novel
probability estimation operator inspired by the concept of distribution of esti-
mation algorithm is developed, which enables MBDE to manipulate binary-
valued solutions directly and provides better tradeoff between exploration
and exploitation cooperated with the other operators of DE. The effectiveness
and efficiency of MBDE is verified in application to numerical optimization
problems. The experimental results demonstrate that MBDE outperforms the
discrete binary DE, the discrete binary particle swarm optimization and the bi-
nary ant system in terms of both accuracy and convergence speed on the suite
of benchmark functions.

Keywords: Differential Evolution, Binary Encoding, Probability Estimation


Operator, Multidimensional Knapsack Problem.

1 Introduction
Differential Evolution (DE), an emerging population-based stochastic optimization
technique first proposed by Storn and Price in 1995 [1], has become a new hotspot in
evolutionary computation. The standard DE algorithm, which is simple yet efficient in
global optimization, has been successfully applied in scientific and engineering fields.
As a versatile evolutionary algorithm, DE does not need any gradient information so
that it is capable of solving non-convex, nonlinear, non-differentiable and multimodal
problems. Moreover, there are only two control parameters in the update formulas of
DE, thus it is easy to implement and tune parameters. Literatures have reported that
DE is superior to particle swarm optimization (PSO) and genetic algorithm (GA) in
some real-world applications [2-4].
Due to its simplicity and effectiveness, DE has attracted much attention in recent
years, and a number of improved variants have been proposed [5-7]. However, the
standard DE and many of its improved variants operate in the continuous space,
which are not suitable for solving discrete combinational optimization problems.
Therefore, several binary DE algorithms are proposed to tackle this drawback. In-
spired by angle modulated PSO algorithm [8], Pampará [9] proposed a new binary DE

K. Li et al. (Eds.): LSMS/ICSEE 2010, Part II, LNCS 6329, pp. 49–57, 2010.
© Springer-Verlag Berlin Heidelberg 2010
50 L. Wang et al.

called angle modulated DE (AMDE). In AMDE, standard DE was adopted to update


the four real-coded parameters of angle modulated function and the angle modulated
functions with different parameter values were used to generate the binary-coded
solution by sampling till the global best solution was found. So it actually operated in
the continuous space. He[10] presented a binary DE based on artificial immune sys-
tem(AIS-DE), in which the scaling factor were treated as a random bit-string and trial
individuals were generated by Boolean operators. Further, an extra control parameter
was introduced, which weakened the flexibility of algorithm as it was significantly
dependent on the problem. Gong et al. [11] proposed a binary-adapted DE (BADE)
where the scaling factor was regarded as the probability of the scaled difference bit to
take on “1”. However, both AIS-DE and BADE discarded the update formulas of the
standard DE and generated the new individuals based on different Boolean operating.
Chen [12] developed a discrete binary differential evolution (DBDE) where the
sigmoid function used in DBPSO [13] was borrowed to convert the real individuals to
bit strings. DBDE directly search in the binary space so that it is easy to implement,
but it is very sensitive to the setting of control parameters. Moreover, the value trans-
formed by the sigmoid function is not symmetrical resulting in deterioration of global
searching ability.
In this work, an improved binary DE (MBDE) is proposed which develops a novel
probability estimation operator to generate the offspring individuals. MBDE reserves
the updating strategy of the standard DE so that the characteristics of DE, such as ease
of implementing and tuning parameters are well maintained as well. This paper is
organized as follows. Section 2 gives a brief review of the standard DE. The proposed
MBDE is introduced in detail in section 3. Section 4 presents the application of
MBDE into numerical optimization, where the comparison of MBDE with DBDE,
DBPSO and the binary ant system (BAS) [14] on a suite of 16 well-know benchmark
functions is conducted. Finally, conclusions are remarked in section 5.

2 The Standard DE
DE is a population-based evolutionary algorithm for global optimization, whose popu-
lation is consisted of a group of floating-point encoded individuals randomly initialized
in the continuous space. Three commonly used evolutionary operators, i.e., mutation
operator, crossover operator and selection operator are used for DE to update the popu-
lation. In the evolutionary process, mutation operator and crossover operator are used
to generate the new trail individual, while selection operator chooses the better one for
the next generation by comparing it with the target individual.

Mutation: There are several mutant schemes in DE, where “DE/rand/1”, as Eq. (1),
is the most popular one. The mutated individual ui is produced according to Eq. (1),

uij t +1 = xrt 1, j + F *( xrt 2, j − xrt 3, j ) (1)

where F, a positive real constant, is the scaling factor ; t is the index of generation;
xr1, j , xr 2, j and xr 3, j are three random chosen individuals with indexes r1≠r2≠r3≠i.

Das könnte Ihnen auch gefallen