基于微粒群优化聚类数目的K-均值算法  

K-mean algorithm for optimizing the number of clusters based on particle swarm optimization

作　　者：巩敦卫[1] 蒋余庆[1] 张勇[1] 周勇[2] 

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221008 [2]中国矿业大学计算机科学与技术学院,江苏徐州221008

出　　处：《控制理论与应用》2009年第10期1175-1179,共5页Control Theory & Applications

基　　金：江苏省自然科学基金资助项目(BK2008125);教育部新世纪优秀人才支持计划资助项目(NCET-07-0802)

摘　　要：K-均值算法是广泛使用的聚类算法,但该算法的聚类数目难以确定,且聚类结果对初始聚类中心比较敏感.本文提出一种基于微粒群优化聚类数目的K-均值算法,该算法采用聚类中心的坐标和通配符表示微粒位置,通过定义微粒更新公式中新的加减运算符,动态调整聚类中心的数目及坐标,此外,以改进的聚类有效性指标Davies-Bouldin准则作为适应度函数.5个人工和真实数据集的聚类结果验证了所提算法的优越性.K-mean algorithm is a widely used clustering method, but it is difficult to determine the number of clusters; and the clustering result is sensitive to initial cluster centers. We present a novel K-mean algorithm for optimizing the number of clusters based on particle swarm optimization. The algorithm denotes the position of a particle with the coordinates of cluster centers and wildcards. The coordinates of cluster centers are dynamically adjusted by defining the new plus and new minus operators in the particle update formula. In addition, an improved Davies-Bouldin index is employed to evaluate the efficiency of a clustering result. Experimental results of 5 sets of artificial and real-world data validate the advantages of the proposed algorithm.

关 键 词：聚类 K-均值算法 微粒群优化 微粒更新 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP301[自动化与计算机技术—控制科学与工程]