一种基于改进PSO的K—means优化聚类算法

针对传统的K—means算法对初始聚类中心的选取敏感、容易收敛到局部最优的缺点，提出一种基于改进粒子群优化算法（PSO）的K—means优化聚类算法。该算法利用PSO算法强大的全局搜索能力对初始聚类中心的选取进行优化：通过动态调整惯性权重等参数增强PSO算法的性能；利用群体适应度方差决定算法中前部分PSO算法和后部分K—means算法的转换时机；设置变量实时监控各个粒子和粒子群的最优值变化情况，及时地对出现早熟收敛的粒子进行变异操作，从而为K—means算法搜索到全局最优的初始聚类中心，使聚类结果不受初始聚类中心影响，易于获得全局最优解。实验结果表明文中提出的改进算法与传统聚类算法相比具有更高的聚类正确率、更好的聚类质量及全局搜索能力。

An Optimized K -means Clustering Algorithm Based on Improved Particle Swarm Optimization

Aiming at the shortcomings of traditional K - means algorithm which is sensitive to initial clustering centers and easy to converge to local optima, an optimized clustering algorithm of K-means based on improved PSO algorithm is proposed. It takes advantages of the powerful global searching capability of PSO algorithm to improve the selection of the initial centers： updating parameters dynamically, for the inertia weight, so as to strengthen the global searching capability of PSO ; determining the occasion of the improved algorithm transferred from PSO to K-means by the fitness variance of the particle swarm; using variants to monitor the optimal value condition of every particle and the particle swarm in real time and executing mutation operations on those particles that converge prematurely on time, which protects the clustering results from being influenced by the initial clustering centers and thus achieves the global optima solution. Experimental results show that the proposed method has higher accuracy rates, better clustering quality and global searching capability.