基于特征选择聚类方法的稀疏 TSK 模糊系统

摘要

In order to solve the curse of dimensionality existing in fuzzy system identification and approximation, this paper proposes the FCA-sparseTSK fuzzy system by casting the Takagi-Sugeno-Kang( TSK ) fuzzy system identifica-tion into a block sparse representation problem.First,FCA -sparseTSK fuzzy system uses the fuzzy clustering algo-rithm ( FCA) to simplify sample features and generate fuzzy system dictionary.Then selects main important fuzzy rules and estimate the fuzzy ruleˊs consequent parameter vector by taking into account the block-structured informa-tion that exists in the TSK fuzzy model.The FCA-sparseTSK fuzzy system simplifies the fuzzy rules and the number of fuzzy rules at the same time and shows good performance in artificial datasets and real-world datasets.%为避免模糊系统建模和估计领域的＂维数灾难＂，将TSK（ Takagi－Sugeno－Kang）模糊系统建模转换为一个分块稀疏表示问题，提出 FCA稀疏TSK模糊系统（ FCA－sparse TSK）。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TSK模糊系统中的分块结构信息，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。该系统同时对模糊规则及模糊规则数进行化简，在合成数据集和真实数据集上都表现出较好的性能。