基于邻域影响的改进粒子群算法的聚类算法

Clustering Algorithm Based on Improved Particle Swarm Optimization

K-均值算法是一种传统的聚类分析方法,具有思想与算法简单的特点,因此成为聚类分析的常用方法之一。但K-均值算法的分类结果过分依赖于初始聚类中心的选择,对于某些初始值,该算法有可能收敛于一般次优解,在分析K-均值算法和粒子群算法的基础上,提出了一种基于邻域影响的改进的粒子群算法的聚类算法,通过对粒子群算法的改进来优化与K-均值结合的聚类算法。该算法将局部搜索能力强的K-均值算法和全局搜索能力强的粒子群算法结合,提高了K-均值算法的局部搜索能力、加快收敛速度,有效阻止了早熟现象的发生,达到那些离群的孤立点。实验表明该聚类算法有更好的收敛效果,一方面聚类所用的时间更短,另一方面聚类的准确率更高。

K-mean algorithm, a traditional clustering method with simple characteristic of thought and algorithm, has therefore be- come one of the methods commonly used in cluster analysis. But the K-means algorithm classification results depend on the initial cluster centers choice. For some initial value, the algorithm may converge to the general sub-optimal solution. This paper proposes a clustering algorithm based on the influence of neighborhood improvement particle swarm optimization （PSO）. Through the im- proved PSO algorithm, we can optimize the combination of K-means clustering algorithm. Both the K-mean, which has strong ca- pacity of local searching, and the PSO, which has power global search ability, are combined. It not only improves the K-mean＇ s local searching capacity, accelerates the convergence rate, but also effectively prevents the premature convergence. The experiments show that this clustering algorithm has better convergence. On the one hand the use of clustering is shorter; on the other hand the accuracy is higher.