基于样本点能量扩散模型的聚类  

Clustering based on energy diffusing model of sample points

作　　者：曾昭贤[1] 张茂军[1] 王炜[1] 熊志辉[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院,长沙410073

出　　处：《计算机应用》2011年第9期2534-2537,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60705013);中国博士后科学基金特别资助项目(200902665);中国博士后科学基金资助项目(20070410977);湖南省自然科学基金资助项目(08JJ4018)

摘　　要：聚类问题是一个复杂的问题,尽管目前聚类方法多种多样,但是仍然存在诸多不足之处,如聚类收敛速度慢,聚类效果不理想,聚类需要人为提供某些参数等。为此尝试提出一种全新的聚类思路:首先认为每一个类都存在一个(或多个)类中心;其次将每个样本点视为一个能量辐射源,以一个合理的模型向空间中辐射能量,空间点根据其得到能量的多少,确定出某些点为类中心;最后将样本点聚类到不同的类中心,达到聚类的目的。实验结果表明,该方法具有收敛速度快、可扩展性强、适合自然聚类的特点,可以达到与很多经典聚类算法相同的聚类效果。Clustering is a complex issue. Although there is a variety oi clustering methods, many shortcomings still exist, such as slow clustering convergence, unsatisfactory clustering results, requiring certain parameters provided by people. To solve these problems, a new idea of clustering was put forward. Firstly, the authors supposed each cluster had a cluster center. Secondly, each sample point was considered as an energy source, eradiating energy to the clustering space with a reasonable physical or mathematical diffusing model. Cluster center was confirmed by the total energy that each point gained. Finally, as a result, sample points could be easily clustered to their cluster centers. The experimental results demonstrate that this clustering approach has the characteristics of fast convergence, strong extendibility, and being suitable for natural clustering. Additionally, it can obtain the same results of many classic clustering methods.

关 键 词：聚类 类中心 能量模型 概率 极值 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP18[自动化与计算机技术—计算机科学与技术]