改进作用力的多智能体聚类算法  

An Improved Data Clustering Algorithm Using a Flock of Agent

作　　者：林玉琪[1] 余春艳[1] 

机构地区：[1]福州大学数学与计算机科学学院,福州350108

出　　处：《小型微型计算机系统》2013年第11期2588-2593,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60805042)资助;福建省自然科学基金项目(2010J01329;2011J05150;2012J01262)资助;福建省重大产学合作项目(2010H6012;2011H6014)资助

摘　　要：针对多智能体聚类算法(FClust)中存在相异智能体朝相同方向一致运动及各智能体朝边缘散化不集中这两个问题,提出一种改进智能体间作用力并增加向心力的多智能体聚类算法AIFClust(Attraction-added and Influence-improved FClust).该算法通过改进智能体间的作用力使得相异智能体相互排斥,加大了相异智能体之间的区分度,并通过增加向心力作用使各智能体具有朝中心运动的趋势,提高了智能体之间的相遇概率,同时降低了智能体运动的随机性.实验结果表明,改进后的AIFClust算法不仅有效解决了FClust算法中存在的问题,且在提高算法收敛速度的基础上,提高了算法的稳定性、算法聚类的准确率以及算法发现类簇的能力.To avoid the two problems of the flocks of agent-based clustering and data visulization algorithm（Faust） , dissimilar agents moving in the same direction, and each agent moving toward the edge, an attraction-added and influence-improved FClust（AIFClust） is proposed. The algorithm through improving the acting force between agents makes dissimilar agent mutually exclusive, so as to in- crease the discrimination of alien agents. Meanwhile it makes each agent has the trend moving toward the center by adding the cen- tripetal force, and thus increases the encounter probability and reduces the moving randomness. The experiments indicate that AIF-Clust efficiently solves the problems, not only accelerates the convergence velocity of iteration, but also improves the algorithm＇s sta-bility, the clustering accuracy and the clusters found ability.

关 键 词：无监督聚类 数据可视化 多智能体 群智能 FClust 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]