具备特征优选功能的中心化TSK模糊系统  

Centralized TSK Fuzzy System with Feature Optimization Function

作　　者：张佳骕 王士同[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122

出　　处：《小型微型计算机系统》2016年第7期1493-1498,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61272210)资助;江苏省自然科学基金项目(BK2011417)资助

摘　　要：对于经典的TSK模糊系统,模糊规则往往存在冗余而且后件参数缺乏可解释性.本文将TSK模糊系统模型转化为中心化形式,并将其建模过程转换为一个分块稀疏表示问题,提出FCA-sparse CTSK模糊模型.首先运用模糊聚类算法(FCA)对样本特征进行化简,并产生模糊系统字典;再利用存在于中心化TSK(CTSK)模糊模型中的分块结构信息,选取重要的模糊规则并对所选模糊规则的后件参数进行估计.该模型通过中心化方法提高了模糊模型的可解释性,并对模糊规则及模糊规则数同时化简,在合成数据集和真实数据集上都表现出较好的性能.For the classical TSK fuzzy system,fuzzy rules tend to have redundancy and their consequent parameters are lack of interpretability. In this paper,we transform TSK fuzzy system into its centralized forms and cast the fuzzy system identification into a block sparse representation problem. First,the proposed FCA-sparse CTSK algorithm uses the fuzzy clustering algorithm（ FCA） to simplify the sample features and generate fuzzy system dictionary. Then it selects main important fuzzy rules and estimates the fuzzy rule consequent parameter vector by taking into account the block-structured information that exists in the centralized TSK fuzzy model. The proposed FCA-sparse CTSK algorithm obtains high interpretability by centralization and simplifies both the fuzzy rules and the number of fuzzy rules at the same time. The proposed FCA-sparse CTSK algorithm shows good performance in artificial datasets and real-world datasets.

关 键 词：中心化TSK模糊系统 特征优选 分块稀疏表示 模糊规则约简 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]