粒子群K-means聚类算法的改进

摘要

Combining particle swarm with K-means algorithm is one of the important methods in data mining, but all methods almost ignore the empty class problem which the particle update causes. This paper proposes a PK-means clustering algo-rithm based on multi-subswarms particle swarm and pseudo means, then is compared with both PSOK-means and K-means. The theory analysis and experiments show that the algorithm not only avoids empty clustering class but also has well global convergence and the local optimization, overcomes local minimum better, has a great effect on isolated data.%粒子群（PSO）与K-means结合是聚类分析中的重要方法之一，但都未考虑粒子更新导致的空类问题。提出基于多子群粒子群伪均值（PK-means）聚类算法，为该问题的解决提供一种有效途径，并与粒子群K均值（PSOK-means）， K-means算法进行比较。理论分析和实验表明，该算法不但可以防止空类出现，而且同时还具有非常好的全局收敛性和局部寻优能力，并且在孤立点问题的处理上也具有很好的效果。