A differentialevolutionalgorithmwithself-adaptingstrategyand control parameters

This paperpresentsa Differential Evolutionalgorithmwith self-adaptivetrialvectorgeneration strategyandcontrol parameters(SspDE)forglobalnumericaloptimizationovercontinuousspace.In the SspDEalgorithm,eachtargetindividualhasanassociatedstrategylist(SL),amutationscalingfactor F list (FL),andacrossoverrate CR list (CRL).Duringtheevolution,atrialindividualisgeneratedbyusing a strategy, F, and CR taken fromthelistsassociatedwiththetargetvector.Iftheobtainedtrialindividual is betterthanthetargetvector,theusedstrategy, F, and CR will enterawinningstrategylist(wSL), a winning F list (wFL), andawinning CR list (wCRL), respectively.Afteragivennumberofiterations,the FL, CRL or SL will berefilledatahighprobabilitybyselectingelementsfrom wFL, wCRL and wSL or randomlygeneratedvalues.Inthisway,boththetrialvectorgenerationstrategyanditsassociated parameterscanbegraduallyself-adaptedtomatchdifferentphasesofevolutionbylearningfromtheir previoussuccessfulexperience.Extensivecomputationalsimulationsandcomparisonsarecarriedout by employingasetof19benchmarkproblemsfromtheliterature.Thecomputationalresultsshowthat overalltheSspDEalgorithmperformsbetterthanthestate-of-the-artdifferentialevolutionvariants.