结合模糊(C+P)均值聚类和SP-V-支持向量机的TSK分类器  

TSK classifier based on fuzzy(C+P) means clustering and SP-V-SVM

作　　者：徐明亮[1,2] 王士同[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122 [2]无锡环境科学与工程研究中心,江苏无锡214063

出　　处：《光学精密工程》2016年第3期643-650,共8页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.61202311;No.61272210);江苏省青蓝工程资助项目(2014)

摘　　要：为获得具有模糊规则自适应约简性能和较好的泛化性能的TSK分类器,本文提出了一种结合模糊(C+P)均值聚类(FCPM)算法和SP-V-支持向量机(SVM)分类算法来构建TSK(Takagi-Sugeno-Kang)分类器的方法。该方法首先用FCPM聚类算法对训练数据进行聚类;然后根据聚类结果确定TSK分类器的模糊规则前件中的高斯隶属度函数的中心和宽度参数;最后采用成组稀疏约束SP-V-SVM算法对模糊规则后件参数进行学习,该算法不仅改善了系统的泛化性能,还使系统具有模糊规则自适应约简功能,使得系统更为紧凑。与相关算法在UCI和IDA标准数据集分类实验中的模糊规则数和分类性能对比表明:用提出的分类算法所构造的TSK分类器不仅具有较好的分类性能,而且模糊规则数少,有利于构建更为紧凑的模糊分类系统。To obtain a TSK （Takagi-Sugeno-Kang） classifier with well generalization performance and a capability of rule adaptive deduction, the method based on the FCPM （ Fuzzy （C＋P） Means） clustering and the SP- V-SVM（Support Vector Machine） is proposed to construct the TSK classifier. The method firstly clusters train data by FCPM clustering algorithm, and then identifies the centers and widths of the Gaussian membership functions of the antecedent part in TSK classifier based on the clustering result. Finally it uses the SP-V-SVM algorithm with the group sparsity constraints to learn the parameters of the consequent part of the fuzzy rule to improve the generalization performance. Moreover, the attribute reduction capacity of the SP-V-SVM algorithm is used to reduce the number of fuzzy rules and to make the TSK classifier compacted. The classification error and the number of fuzzy rules of the proposed algorithm are compared with several standard data sets from UCI and IDA repository, and results show that the proposed TSK classifier not only gets better performance but also has a few number of fuzzy rules and is suitable for structuring compact fuzzy classifying systems.

关 键 词：TSK分类器 模糊规则 规则约简 模糊(C+P)均值聚类(FCPM) SP-V-支持向量机(SVM) 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]