期望因子驱动下的K-means初始聚类中心优化算法研究

打开文本图片集

关键词：初始聚类中心；优化算法；K-means；期望因子；网格划分；权重系数中图分类号：TN911.1-34；TP301.6 文献标识码：A文章编号：1004-373X（2026）06-0089-05

Research on K-means initial clustering centers optimization algorithm drivenbyexpectationfactor

FENG Xin， TAN Ding，LI Mingfeng （SchoolofGeomaticsScienceand Technology，Nanjing TechUniversity，Nanjing211816，China）

Abstract：ReasonableinitialclusteringcentersarethekeytoimprovetheclusteringefectofK-meansalgorithmandavoid localoptimization.Inordertodeterminereasonableinitialclusteringcenters，aK-meansinitialclusteringcentersoptiization algorithmdrivenbyexpectationfactorisproposed.Thegriddivisioncriteriadrivenbyexpectationfactorisdesignedtomeasure thesamplepointsdensityfactor，andtheEuclideandistanceisusedtomeasurethesamplepointsdistancefactor.Theweightis introduced toconstraindensityfactoranddistancefactor，andthetwofactorsarecomprehensivelyconsideredtooptimizethe selectionofinitialclusteringcentersenhanetheglobalsearchabilityandimprovetheclusteringefect.Theconceptofthesum ofcentersdistance isproposedtomeasuretheoptimization efectof initialclusteringcenters.Comparativeexperimentswere carriedutinIris，SeedsandWinedatasets.ComparedwiththetradionalK-meansalgorithm，thecentersdistanceof the proposed algorithm was reduced by 75% ， 52% and 58% respectively，the sum of squares due to error was reduced by 15% ， 7% （204 and 6% respectively，and the clustering accuracy was increased by 20% ，19%and 24% respectively，which are better that those ofotherimprovedalgorithms.Theexperimentalresultsshowthattheproposedalgorithmefectivelyoptimizetheinitialclustering centers，improve the clustering effect and the stability of the clustering results.

Keywords：initialclusteringcenters；optimizationalgorithm;K-means;expectationfactor;griddivision;weightcoficient

0 引言

聚类等领域。（剩余6945字）