Species-based Methods For Dynamic Optimization Problems

Compared with the common optimization problems,the characteric of dynamic optimization problems(DOPs)is the state of the problem is time-varying,e.g.the objective function,constraint conditions.In order to capture the dynamics of the environment,algorithms need to locate and track the moving tracectory of optima.Due to the population-based searching strategy,Evolutionary Algorithms(EAs)is a suitable candidate to handle many difficult problems,e.g.DOPs.The speciation-based method is an effective techonology in EAs and has received wide attention in the field of dynamic optimization researches.The basic idea of the speciation-based method is to divide the population into several species and allow different species to exploit different areas simultaneously.In this way,multiple optima can be located parallelly,which contributes to the tracking for the global optima.The concerns of this paper include the combination of the speciation-based method with the memory strategy for DOPs,species identification for DOPs.The main contributions of this paper are listed as follows.(1)A species-based particle swarm optimizer enhanced by memory for DOPs is proposed,e.g.MSPSO.Different with existing methods,MSPSO has some features.Firstly,the number of the retrieved memory particles is related to the number of species and changes adaptively.Secondly,the number of particles replaced in each species is not more than one.Thirdly,the retrieved memory particles are classified with the purpose to improve the exploitation of existing species and enhance the exploration for potential optimal areas as well.MSPSO is tested in different benchmark problems and compared with other algorithms.Experimental results show that MSPSO is a competitive algorithm for DOPs.Besides,this paper also discusses and analyses the influence of the memory size to the results.(2)A new species identification method is proposed for DOPs,called psfNBC.Compared with the basic Nearest-Better Clustering(NBC),psfNBC has some improvements.Firstly,in the process of species seeds identification,only part of individuals is involved rather than all.Secondly,individuals are reassigned to the species by the principle of the nearest seed.Thirdly,scaling factor(?)uses random values instead of a constant.This paper proposes two methods for determining the number of outliers,namely fixed and adaptive.Moreover,a framework of species-based particle swarm optimization for DOPs is given.The psfNBC and three other typical methods are applied to the framework respectively and tested under MPB.Exprerimental results show that psfNBC outperforms other three methods in most cases.