文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 自然科学 > 天文/地理 > A differential evolution algorithm with self-adapting strategy and control parameters
侵权投诉

A differential evolution algorithm with self-adapting strategy and control parameters

差分算法相关文献

A differential evolution algorithm with self-adapting strategy and control parameters

A differential evolution algorithm with self-adapting strategy and control parameters

Computers&OperationsResearch38(2011)394–408

ContentslistsavailableatScienceDirect

Computers&OperationsResearch

journalhomepage:doc.guandang.net

Adifferentialevolutionalgorithmwithself-adaptingstrategyandcontrolparameters

Quan-KePana,P.N.Suganthanb,Ã,LingWangc,LiangGaod,R.Mallipeddib

a

CollegeofComputerScience,LiaochengUniversity,Liaocheng252059,PRChina

SchoolofElectricalandElectronicEngineering,NanyangTechnologicalUniversity,50NanyangAvenue,Singapore639798,Singaporec

TsinghuaNationalLaboratoryforInformationScienceandTechnology(TNList),DepartmentofAutomation,TsinghuaUniversity,Beijing100084,PRChinad

StateKeyLaboratoryofDigitalManufacturingEquipment&TechnologyinHuazhongUniversityofScience&Technology,Wuhan430074,PRChina

b

articleinfo

Availableonline23June2010Keywords:

DifferentialevolutionEvolutionaryalgorithm

GlobalnumericaloptimizationParameteradaptationStrategyadaptation

Continuousoptimization

abstract

ThispaperpresentsaDifferentialEvolutionalgorithmwithself-adaptivetrialvectorgenerationstrategyandcontrolparameters(SspDE)forglobalnumericaloptimizationovercontinuousspace.IntheSspDEalgorithm,eachtargetindividualhasanassociatedstrategylist(SL),amutationscalingfactorFlist(FL),andacrossoverrateCRlist(CRL).Duringtheevolution,atrialindividualisgeneratedbyusingastrategy,F,andCRtakenfromthelistsassociatedwiththetargetvector.Iftheobtainedtrialindividualisbetterthanthetargetvector,theusedstrategy,F,andCRwillenterawinningstrategylist(wSL),awinningFlist(wFL),andawinningCRlist(wCRL),respectively.Afteragivennumberofiterations,theFL,CRLorSLwillbere lledatahighprobabilitybyselectingelementsfromwFL,wCRLandwSLorrandomlygeneratedvalues.Inthisway,boththetrialvectorgenerationstrategyanditsassociatedparameterscanbegraduallyself-adaptedtomatchdifferentphasesofevolutionbylearningfromtheirprevioussuccessfulexperience.Extensivecomputationalsimulationsandcomparisonsarecarriedoutbyemployingasetof19benchmarkproblemsfromtheliterature.ThecomputationalresultsshowthatoveralltheSspDEalgorithmperformsbetterthanthestate-of-the-artdifferentialevolutionvariants.

&2010ElsevierLtd.Allrightsreserved.

1.Introduction

Thedifferentialevolution(DE)algorithmisasimpleyetpowerfulsearchtechniqueintroducedbyStornandPrice[24]forsolvingcomplexcontinuousnonlinearfunctions.Startingfromapopulationofrandomlyinitializedsolutions,theDEalgorithmemployssimplemutationandcrossoveroperatorstogeneratenewcandidatesolutions,andutilizesaone-to-onecompetitionschemetodeterministicallydecidewhethertheoffspringwillreplacetheirparentsinthenextgeneration.Duetoitssimplicity,easyimplementation,andfastconvergence,theDEalgorithmhasgainedmuchattentionwithsuccessfulapplicationsinmechanicalengineering,sensornetworks,scheduling,patternrecognitionandinotherdomains[8,9,14,15,21,25,28,32].

ThereexistmanydifferentDEtrialvectorgenerationstrate-gies.Thesestrategiesoftenpossessdifferentsearchingcapabil-itiesinvarioussearchphasesoftheevolutionprocess.Moreover,theirassociatedcontrolparameters,namelythescalingfactor,F,andthecrossoverrate,CR,maysigni cantlyin uencethesearchingaccuracyandconvergencespeedoftheDEalgorithm

ÃCorrespondingauthor.

E-mailaddress:epnsugan@ntu.edu.sg(P.N.Suganthan).

0305-0548/$-seefrontmatter&2010ElsevierLtd.Allrightsreserved.doi:10.1016/j.cor.2010.06.007

[3,4,10,18,30].Therefore,itisofsigni cancetodeterminesuitabletrialvectorgenerationstrategiesandtheirassociatedparametervaluesfortheDEalgorithmwhenitisusedtosolverealproblemsinscienti candengineering elds.Severalempiricalguidelinesexistinliteratureforchoosingsuitabletrialvectorgeneratorstrategiesandparametersettings.Forexample,StornandPrice[17,23]statedthatthecontrolparametersoftheDEalgorithmwerenotdif culttochoose,andsuggestedthatareasonableNPvalueshouldbebetween5nand10n,andaneffectiveFvalueshouldbeintherange[0.4,1],whereNPisthepopulationsize,andnisdimensionalityoftheproblemtobesolved.Iftheproblemisnearunimodalorfastconvergenceisdesired,CR¼0.9isagoodinitialchoice.Furthermore,Price[16]recommendsthatitisgoodtosetNP¼20n,K¼0.5,andF¼0.8whenthetrialvectorgenerationstrategyDE/Current-to-rand/1isused,whereKisascalingfactorinDE/Current-to-rand/1.However,basedonparametersettingsfortheDEalgorithmonSphere,RosenbrockandRastriginfunctions,Gamperle[4]reportedthatchoosingthepropercontrolparametersfortheDEalgorithmwasmoredif cultthanexpected,andtheyadvisedthatNPshouldbeintherange[3n,8n],andFshouldbeequalto0.6andCRisbetween0.3and0.9.Recently,Ronkkonenetal.[22]suggestedusingFvaluesin[0.4,0.95],andtheCRvaluein[0,0.2]forseparablefunctionswhileCRin[0.9,1]fordependentfunctions.Obviously,therulesfor

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email