一种融合K-means算法和人工鱼群算法的聚类方法

摘要

Aiming at the problems of K-means clustering being subject to local optimum and sensitive to initial value,and the problems of artificial fish swarm algorithm having high convergence rate,being insensitive to initial values and self-organising behaviour,we propose a clustering method which combines K-means and artificial fish swarm algorithm.The algorithm first slightly improves the standard artificial fish swarm algorithm with self-adaptive strategy:i.e.in early iteration of artificial fish swarm algorithm it uses the fixed visual perspective,with the increase of iteration times,is adopts the self-adaptive decreasing visual perspective value.Based on this,it integrates K-means algorithm to artificial fishes of the improved artificial fish swarm algorithm,part of the artificial fishes randomly generated will go through the iteration of K-means once after finishing each iteration in artificial fish algorithm.Experimental results prove that the new algorithm is obviously superior to the particle swarm optimisation,K-means and the improved AFSA,and it will be effectively applied in data clustering.%针对K-means易收敛于局部最优以及对初始值敏感和人工鱼群算法收敛速度快，对初始值不敏感及自组织行为的问题，提出一种K-means和人工鱼群算法融合的聚类方法。该算法先将标准人工鱼群算法用自适应策略加以改进，即在人工鱼群算法早期迭代中使用固定视野，随着迭代次数的增加，采用自适应减少的视野值。在此基础上将K-means算法融入到改进的人工鱼群算法中人工鱼中，随机产生的部分人工鱼在每次完成人工鱼群算法的迭代后，进行一次K-means算法的迭代。实验结果证明融合后的新算法明显地优于粒子群优化（PSO）、K-means及改进的人工鱼群算法（IAFSA），它将有效地被应用于数据聚类中。