基于模糊权和有效性函数的演化聚类算法  

A Fuzzy Weighted Sum Validity Function for Clustering with a Mixed Strategy Evolutionary Algorithm

作　　者：董红斌 黄厚宽[1] 周成[1] 何军[1] 尚文倩[1] 

机构地区：[1]北京交通大学计算机与信息技术学院

出　　处：《电子学报》2007年第5期964-970,共7页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60443003);黑龙江省自然科学基金(No.F200605);北京交通大学科技基金(No.2003SZ003)

摘　　要：本文改进了Sheng的权和有效性函数,将XB、PE、PC和PBMF等模糊聚类有效性函数集成为一种新的模糊聚类有效性度量函数—模糊权和有效性函数FWSVF,从而提高了聚类有效性函数的性能.为了有效的实现聚类,将混合策略演化算法与传统的模糊C均值算法(FCM)相结合,将改进的模糊权和有效性指标作为适应度函数,提出了一种混合策略演化聚类算法MSECA.人工数据集和真实数据集的仿真实验表明,MSECA算法可以正确发现聚类簇的数量,避免了局部极值问题,比其他算法具有更好的性能.Clustering is inherently a difficult problem, with respect to both construction of adequate objective functions and optimization of the objective functions.In this paper, we propose a novel objective function called the Fuzzy Weighted Sum Validity Function （FWSVF）, which is a merged weight from the several fuzzy cluster validity functions,including XB,PE,PC and PBMF. The new validity function has more efficient quality than old ones. Furthermore, we present a Mixed Strategy Evolutionary Clustering Algorithm （MSECA）, which is merged from Mixed Strategy Evolutionary Algorithm and Fuzzy C-means Algorithm and could be applied to optimization of FWSVF. The improved validity function could improve the confidence of clustering solutions and achieve more accurate and robust results. Moreover, MSECA could automatically evolve the proper number of clusters as well as appropriate partitions of the data set, and avoid local optimum. In the experiments, we show the effectiveness of MSECA. In comparison with other genetic clustering algorithms, the MSECA can consistently and efficiently converge to the best known optimum corresponding to given data in concurrence with the convergence result.

关 键 词：模糊C均值算法 演化规划 聚类有效性 混合策略 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]