基于多核机群的混合群智能并行算法分析-analysis of hybrid swarm intelligence parallel algorithm based on multicore cluster.docxVIP

ABSTRACTInrecentyears,SwarmIntelligenceOptimizationAlgorithmisanewkindofbionicoptimizationalgorithm,moreresearchabouttwoswarmintelligencealgorithmsareArtificialFishSwarmAlgorithm(AFS)andArtificialBeeColonySwarmAlgorithm(ABC).Becauseofthesetwoalgorithmshavestrongrobustness,simpleparametersetting,easytoimplementandsoon.Hasextensiveresearchandapplicationinevolutionarycomputation,geneticcontroloptimizationstrategy,it’sbecomeveryactivefrontiertopicsininterdisciplinaryscience.Asanewevolutionarycomputationtechnique,swarmintelligencealgorithm,inthesimulationanimalsampling,takingadifferentapproachtosolvingcomplexproblemsintheprocessofcollectingfood.AFSandABChavestrongglobalsearchcapability,butduetopoorpost-finesearchcapabilities,easytofallintolocaloptimum.So,firsttoimproveoptimizationstrategyofAFSandABC,uniformlydistributedontheinitialpopulation,thenrandomlydividedintotwopopulations,anduseinteractivelearningstrategiestoacceleratetheconvergencerate.InpreliminaryalgorithmbyAFSandABC,getrangeoptimization.ThelateadoptionofimprovedbyDN-AFSandRP-ABCalgorithmtogetthesolutionforlocalpre-finesearch.Finally,buildMPI+OpenMP+STMparallelprogrammingmodelforparallelalgorithmdesignandanalysis.Throughsimulation,toprovethefeasibilityandeffectivenessofthehybridswarmintelligenceparallelalgorithm.Themainworkincludes:Becausetheweaklate-searchcapabilityofAFSA,reducedpopulationdiversity,easytofallintolocaloptimum.Inthesearchprocessofartificialfishadoptdynamicweightingfactorstrategyanddynamicallyadjustthestepsizeofthefieldofview,balancingtheglobalsearchcapability;byconstantthresholdεdeterminethedegreeofaggregationoffish,introductionofnichemechanism,thenanalysisofparallelalgorithms,putforwardanewartificialfishswarmParallelDynamicweighNichesArtificialFishSwarm(PDN-AFS)algorithm.SimulationexperimentsshowtheeffectivenessofPDN-AFSalgorithminmulti-coreclusterenvironment.Byanalyzingthecharacteristicsofartificialbeecolonyalgorithm,whenitdealingwithcomplexfunctionoptimizationconvergenceisslow,easiertofallint