一种求解最大团问题的并行交叉熵算法.pdfVIP

ISSN 1000—9825，CODEN RUXUEW Em·ail：jOS@iscas．ac．cn JournalofSoftware，Vo1．19，No．11，November2008，PP．2899—2907 http：／／www．jos．org．cn DOI：l0．3724／SP．J．1OO1．2008．02899 T l／Fax：+86．10 ◎2008byJournalo／Software．Allrightsreserved． 木 一 种求解最大团问题的并行交叉熵算法 吕强 ，柏战华 ，夏晓燕 (苏州大学 计算机科学与技术学院，江苏 苏州 215006) (江苏省计算机信息处理重点实验室，江苏 苏州 215006) Leader-BasedParallelCrossEntropyAlgorithm forM aximum CliqueProblem El3Qiang，，BAIZhan．Hua， XIAXiao．Yan (‘SchoolofComputerScienceandTechnology，SoochowUniversity，Suzhou215006，China) (ProvincialKeyLaboratoryforComputerInformationProcessingTechnology，Suzhou215006，China) +Correspondingauthor：E-mail：qian~：@suda．edu．cn LfiQ，BaiZH，XiaXY．Leader-Basedparallelcrossentropyalgorithmformaximum cliqueproblem．Journal ofSoftware，2008，19(11)：2899—2907．http：／／www．jos．org．cn／1000—9825／19／2899．htm Abstract： TheCrossEntropymethodisanew searchstrategyofrcombinatorialoptimizationproblems．However， itusuallyneedsconsiderablecomputationaltimetoachievegoodsolutionquality．ThispaperintroducesaCross Entropyalgorithmforsolvingmaximum cliqueproblem (MCP)．TomaketheCrossEntropyalgorithm faster,this paperproposesaleader·basedcooperativeparallelstrategy．Unlikethewidelyusedcoarse-grainedparallelstrategy, ourmethod hasa leader，who can move around the parallelprocessorsand collectdata actively，and several followerswhosemainjobaresimplytosamplethecliquesguidedbytheleaderviatransitionmatrix．Toevaluate theperformanceofthealgorithm，thispaperimplementsthealgorithm usingOpenMPIonMIMD architecture，and appliesitontheMCPbenchmarkproblems．Thespeedupandefficiencyareanalyzed，andtheresultsarecompared withthoseobtainedbyfourotherbestheuristicalgorithms．Theresultsshow thatthepresentedmethodhasachieved goodperformanceamongthosebestpopulation—basedheuristics． Keywords：