基于进化规划的聚类算法研究

《基于进化规划的聚类算法研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、北京交通大学硕士学位论文基于进化规划的聚类算法研究姓名：周成申请学位级别：硕士专业：计算机应用技术指导教师：黄厚宽20061201ABSTRACTAlongwiththerapiddevelopmentofmodeminformationtechnologysuchastheIntemet,peoplemustfacemassiveinformationeveryday．Howtoextractusefulinformationfrommassivedatahasincreasinglybecomeahottopicofconcern．Asabas

2、icmoal坞ofinformationprocessing,clust盯analysistechnologyhasbecomepeople'soonccr／linrecentyears．ClusteranalysishasalsogainedawiderangeofresearchandappHcationinmachinelearning,patternrecognition,datamining,informationretrievalandmanyotherfields，．Theclusteringalgorithmmainlyinclud

3、espartition-basedclusteringalgorithmsandhierarchicalclusteringalgorithm．Partition-basedclusteringalgorithmsagethemostcommonlyuseddataminingalgorithms,Asimportantpartition-basedclusteringalgorithms，K-MeansfuzzyC-MeansclusteringalgorithmffCM)iswidelyusedinpgactice．However,therea

4、rethreedrawbacksinthealgorithms：thenumberofclustercentersmustbespecifiedinadvance；thealgorithmstendtoconvergetothelocalmillilnumorsaddlepoint；clusteringresultsareimpactedmuchbyinitialclustercenters．Aimedatsolvingtheseflaws，thispaperpresentsK-meansclusteringmethodbasedonevoluti

5、onaryprogrammingand吻C-meansclusteringmethodbasedonevolutionaryprogrammingwhichisnamedKEPandEPFCMalgorithm,BytheoptimmationabilityofevolutionaryprogrammingKEPcanavoidtheflaw仃appinginlocalminimaandtheimpactofinitialclustercenters．ExperimentsshowthatKEPhavebetterclusteringresults

6、comparedwithK-meansanditismorefasterandmoreprecisethanK-meansalgorithmbasedonGA(KGA)．InEPFCM,risingclustervalidityindexfortheassesslnentandoptimizationabilityoftheoptimizationabilityofevolutionaryprogramming,usersdon’tneedtospccifythenumberofclustercenters．EPFCMcanautomaticall

7、ysearchthebestnumberofcentersandtheoptimalclusterstructure．Tospeeduptheconvergenceprocess，wetakeFCMiterationintotheevolutionaryprogrammingprocess；Tosearchtheoptimalnumberofclustercenters，wemodifythenmnberofcentersdynamically．ExperimentsshowthatEPFCMalgorithmtailgainbestcluster

8、centersandoptimalclusterstructures．andtheprobabilityoffalling