一种结合粒子群和粗糙集的聚类算法  

Clustering Algorithm Based on Particle Swarm Optimization and Rough Set

作　　者：张丹丹[1] 罗可[1] 

机构地区：[1]长沙理工大学计算机与通信工程学院,长沙410114

出　　处：《小型微型计算机系统》2016年第9期2040-2045,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(11171095;71371065)资助;湖南省科技计划基金项目(2013SK3146)资助;湖南省自然科学衡阳联合基金项目(10JJ8008)资助

摘　　要：针对传统K-means算法过度依赖初始聚类中心、易陷入局部最优、不能处理边界对象及聚类精度低等问题,提出一种结合粒子群和粗糙集的聚类算法.此算法首先利用密度和最大距离积法初始化粒子群;然后采用线性递减与随机分布相结合的惯性权重、动态调整的学习因子和引入的随机粒子来避免陷入局部极值,使算法快速收敛于全局最优;最后结合粒子群和粗糙集来优化K-means算法.通过对几个常用UCI标准数据集的仿真实验表明,提出的算法不仅能减少对初始聚类中心的依赖、有效抑制局部收敛,而且聚类准确率更高,稳定性更强.To deal with the deficiencies of the traditional K-means algorithm of its depending overly on the selection of the initial clus- tering centers ,easing to fall into local optimum, low efficiency of handling boundary objects and low clustering accuracy, this paper proposes a new algorithm based on particle swarm optimization and rough set. First of all, the density and maximum distances product method is used in this approach to generate initial particle swarm optimization. Then the method of combining linearly decreasing and random distribution is adopted to produce inertia weight. And by adjusting the learning factor dynamically and introducing random par- ticles, the approach can speed up the global optimal convergence. At last, the rough set and particle swarm optimization are combined to optimize K-means. The results of the simulation of several commonly used UCI benchmark datasets show that the new algorithm can not only reduce the dependence on the initial cluster center and suppress the local convergence effectively, but also obtain high accura- cy and strong stability.

关 键 词：K—means算法 粒子群 粗糙集 最大距离积法 随机粒子 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]