Abstract: To compensate for the deficiency of converging to local optimum at the outset and slow convergence latterly，a hybrid Particle Swarm Optimization（PSO） algorithm employing improved k-means clustering analysis algorithm（CA-PSO） is proposed and applied to multi-dimension function searching.In CA-PSO，the current particles is first divided into multi sub-populations by improved k-means cluster mechanism，and employs PSO itself inherent parallelism advantage to enhance the capacity of searching optimal solutions.It not only exchanges more information among particles，restrains the tendency of premature，but also increases the converging rate and accuracy.It is proved theoretically that CA-PSO is endowed with stable convergence under given conditions.The comparative results indicate that CA-PSO is superior to original particle swarm optimization algorithm.

Key words: Particle Swarm Optimization（PSO）, k-means, clustering analysis, subpopulation, inherent parallelism

摘要： 对于多维函数的最优解搜索，粒子群优化算法存在前期易陷入局部最优，后期收敛速度缓慢的问题。将改进的k均值聚类分析策略与PSO相结合提出了一种混合粒子群优化算法CA-PSO。在算法中，利用改进的k均值聚类分析方法将粒子群划分成若干个子群，结合PSO的隐含并行搜索的优势增强了寻优性能。不仅增加了粒子间的信息交换，抑制了早熟收敛，并且提高了全局寻优速度和计算精度。理论证明，在一定条件下，CA-PSO具有稳定收敛性。仿真结果表明，CA-PSO性能优于基本粒子群优化算法。

关键词: 粒子群, k均值, 聚类分析, 子群, 隐含并行