基于MPI的并行PSO混合K均值聚类算法  

Parallel PSO combined with K-means clustering algorithm based on MPI

作　　者：吕奕清[1] 林锦贤[1] 

机构地区：[1]福州大学数学与计算机科学学院,福州350108

出　　处：《计算机应用》2011年第2期428-431,437,共5页journal of Computer Applications

基　　金：福建省高校科研专项重点项目资助项目(JK2009002)

摘　　要：传统的串行聚类算法在对海量数据进行聚类时性能往往不尽如人意,为了适应海量数据聚类分析的性能要求,针对传统聚类算法的不足,提出一种基于消息传递接口(MPI)集群的并行PSO混合K均值聚类算法。首先将改进的粒子群与K均值结合,提高该算法的全局搜索能力,然后利用该算法提出一种新的并行聚类策略,并将该算法与K均值聚类算法、粒子群优化(PSO)聚类算法进行比较。实验结果表明,该算法不仅具有较好的全局收敛性,而且具有较高的加速比。The performance of traditional serial clustering algorithm cannot meet the needs of data clustering of the huge amounts of data.To enhance the performance of clustering algorithm,a new clustering algorithm combining parallel Particle Swarm Optimization（PSO） with K-means based on MPI was proposed in this paper.Firstly,the improved PSO was combined with K-means to enhance the capacity of global search,and then a new parallel clustering algorithm was proposed,which was compared with K-means and PSO clustering algorithms.The experimental results show that the new algorithm has better global convergence,and also has higher speed-up ratio.

关 键 词：消息传递接口集群 粒子群优化算法 K均值算法 并行聚类 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP311.1[自动化与计算机技术—计算机科学与技术]