面向網格的并行演化算法研究.pdf

1、上海大學碩士學位論文面向網格的并行演化算法研究姓名：冷平申請學位級別：碩士專業(yè)：計算機系統(tǒng)結構指導教師：雷詠梅20060101上海大學碩士學位論文AbstractG“d—enabledalgorithmspamlleliza“ontechnologyongridenVimnmenthasbeenresearchedGrid—enabledscalableparallelcomputingmodelhasbeenresearchedand

2、gridenabledparallelev01utionaryalgorithms(EAs)frameworkhasbeenconstructedBasedonimplementa“onofcoeVolutiontypeparallelEAs(CoPEA)，thisthesisconslructsandimplementsspacedecompositionbasedparallelEAs(Sp—PEA)，whichsupponsdyn

3、amicre—sourcea110cationong“denvironmentCo—PEAtakesmasterslavecommunicationmode，dataexchangeoperationinmasterprocesstoimplementindiVidualmi盯ationSp—PEAadoptssplicing／decomposableencodingwhichisanadaptiVeencoding，usesmulti

4、plecrossoveroperatorsandmutationopemtorwhichissuitableforfunctionoptimizationComparedwithCo—PEA，Sp—PEAproVidesbetcerconve唱enceproperty，scaIabilityandismoresuitableforgridenVironmentCoPEAandSpPEAareimplementedonclusterenv

5、ironment，andnumericalcxperimentiscarryoutonZiqiall93000ofShan曲aiUniversityParallelemciencyoftwoalgo一“thmsisanalyzedIt’sshownthat，withtheincreaseofprocessorscale，CoPEAcaTlgetgoodspeedupwithinsuitableprocessorscale，andSpPE

6、A’sprecisionofresuJtisimproVedobviouslMwhichisVeryValuableinpractic“applicationandillustrateSpPEAcanwellsupportdynamicrcsourcealIocationongridenvironmentExperimentshowsSp—PEAhasbetterco九ve唱encepmpertyandscalability，andis

7、moresuitableforg“denVimment，incontrasttoCo—PEAThisthesisfurtherparallelizethesealgorimmsongrid，putforwardtheproblemsthosewilloccurt11ereandgivethemethodstesolvethemSpPEAisimproVedtosupportdynamicresourcea110cationong“dEx

8、ecutionprocessongridisdescribedExperimentresultofdynamicresourceallocationofSpPEAispresentedThefutureworkisIistedinthe1astchapterofthisthesisComputingserVicecanbeprovidedintheformofgridserViceong^dplatformExploitingthcfu

9、nctionofdatamanagementofgrid，wecanconservetherunningparametersandresuItsinthedatabaseofgrid，“willbeusefulforanalyzingthealgo“thms_TheworkonclusterenvirorlmentinthisthesishasbeenimplementedonZiqian93000hi曲performancecom芻u