非均衡原型结构模式模糊聚类方法研究

Fuzzy Clustering Method Research Based on Disequilibrium Prototype Pattern

针对传统FCM算法通常只能检测大小近似相等呈球形或椭球分布的样本子集,而对样本结构、类型、密度分布不均衡的数据集聚类效果不理想等问题,提出一种基于原型初始化的样本隶属度分配方法.首先运用数学形态学理论对聚类原型初始化以获得模糊聚类的原型先验知识,在此基础上设计一种样本隶属分配方法进行样本聚类.理论分析和实验表明,该方法不但可以解决样本集内原型结构差异悬殊的数据集聚类问题,而且具有求解速度快、易于实现等优点.

Traditional FCM algorithm can usually only detect approximately equal size or spherical or elliptical distribution of a subset of the sample set, which is not ideal to detect the structure,type and the density distribution disequilibrium of the sample. An improved fuzzy clustering method is put forward. Firstly it uses mathematical morphology theory to initialize clustering prototype and gets the priori knowledge of clustering prototype, then designs a new membership function based on prototype initialization to process samples clustering. Theoretical analysis and experiment show that this method can not only solve the problem of disequilibrium prototype structure in data sets. but also has its advantages such as high solution speed, facile imolementation, etc.