一种并行自适应微粒群聚类算法  

A parallel cluster algorithm for self-adaptive particle swarm optimization

作　　者：王华秋[1] 廖晓峰[1] 冯晋[2] 

机构地区：[1]重庆大学计算机学院,重庆400030 [2]重庆工学院计算机学院,重庆400050

出　　处：《重庆大学学报（自然科学版）》2008年第6期652-657,共6页Journal of Chongqing University

基　　金：重庆市自然科学基金资助项目(CSTC2007BB2406)

摘　　要：大规模的数据挖掘如聚类问题迫切需要大量计算,提出了自适应微粒群优化的并行聚类算法。通过从多种群并行地开始搜索,基于群体搜索技术的微粒群优化算法减少了初始条件的影响,采用任务并行和部分异步通信策略,降低计算时间。结合并行微粒群算法的自适应参数动态优化特性,克服群体逐渐失去迁移性而停止进化的问题,保持群体多样性从而了避免种群退化。仿真实验证明,该算法在并行机群上运行时,加快了聚类算法的计算速度,提高了聚类质量。Full-scale data mining, such as in cluster problems, requires large numbers of computations. A parallel cluster algorithm for self-adaptive particle swarm optimization was proposed to deal with this problem. The proposed parallel particle swarm optimization algorithm reduced the impact of the initial conditions via parallel searches of the globally best position amongst a varied population. Task parallelization and partially asynchronous communication of the algorithm were employed to decrease computing time. Furthermore, if combined with the characteristics of self-adaptive and dynamical optimization parameters of the parallel particle swarm algorithm, the problems of particle mobility loss and the end of evolution could be dealt with successfully. When modified thusly, the algorithm maintains individual diversity and restrains degeneration. The simulation experiments indicate the algorithm helps increase computing speed and improve cluster quality.

关 键 词：并行聚类 自适应微粒群优化 任务并行 异步通信 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]