Abstract: This Paper presents a new Clustering-based niche technique which can efficiently locates multiple optimal solutions in multimodal problems and convergences rapidly. Cognition only model PSO uses only a particle’s personal best position found thus far in the velocity update, and each particle effectively performs an individual search in its local area. This paper proposed an improved method, which made particles convergent rapidly to areas near the optimal solution. At last each particle’s best experiences generate some clusters, which can be gotten through clustering algorithm. So multimodal problems has turn into unimodal problems in some clusters in which the optimal solution is guaranteedly located through the guaranteed convergence particle swarm optimizer. Experimental results show that the proposed algorithm can successfully locates all maxima on a small set of test functions which are circular topological structure during all simulation runs.

摘要： 本文提出了一种基于聚类的小生境技术，可以有效地解决多模态问题并获得多个最优解，并且有较快的收敛速度。认知模式微粒群优化器只利用了每个粒子的认知信息从而在局部区域进行搜索，每个粒子在局部区域寻优并趋向区域最优解, 且存在收敛速度慢等问题。为此，本文提出了一种改进算法，可以让粒子迅速收敛到局部最优解附近。最终每个粒子经历过的最优位置形成了若干个簇，通过对其聚类获得每个簇中的粒子信息。此时问题已转化为多个簇的单模态问题，在各个簇中再利用保收敛微粒群优化器获得每个簇的最优解。最后给出了实验，证明了该方法在圆形拓扑环境中的有效性。