一种改进的K-means蚁群聚类算法

An Improved K-means Ant Colony Clustering Algorithm

现有的K-means蚁群聚类算法,首先进行K-means聚类算法操作,快速、粗略地确定初始聚类中心,接着根据上一步获得的聚类中心再进行蚁群算法聚类操作,有效地解决蚁群聚类算法收敛速度过慢的问题。研究发现,现有的Kmeans蚁群聚类算法并没有改善算法在迭代后期易出现收敛于非全局最优的缺陷。针对这一问题,提出一种改进的Kmeans蚁群聚类算法。每次迭代结束时,随机选择一个或多个簇,再从选中的簇里选择含有信息素最小的节点进行变异操作,把选中的节点变异到其他簇,计算评价值判断变异是否进行。仿真实验结果表明,用F值表示的平均值和最差结果都比原有的算法较好,有效解决了原有算法易收敛于非全局最优及早熟问题,但由于变异操作使算法运行时间相对较长。

Existed K-means ant colony clustering algorithm carries out K-means algorithm operation, fast and roughly determines the clustering center, then according to rough clustering center, ant colony clustering algorithm is conducted again to solve the problem of low convergence speed effectively. The research shows that the existed K -means any colony clustering algorithm doesn＇ t improve the defect of converging to non-global optimal in late iteration. In order to solve this problem, a modified K -means ant colony clustering algorithm is presented. At the end of each iteration,randomly select one or more clusters,and then choose the point from the selected cluster with minimum pheromones for mutation, the mutation selecting node to another cluster, evaluation value is calculated to judge whether to mu- tate. Experimental results show that the average and worst results indicated by F value are better than the original algorithm, effectively solving the problem that is easy to converge to non-globa/optimal and premature, but it takes a longer running time.