关键词: T-S模糊系统建模, 公共特征选择, 联合优化, ADMM算法

Abstract: The traditional Takagi-Sugeno (T-S) fuzzy systems always have a poor interpretability because they use all the attributes of the samples in the rules. The existence of redundant features inevitably leads to the over fitting problem and further reduces the generalization performance of the fuzzy model. To this end, this paper proposes a novel joint sparse modeling method for Takagi-Sugeno fuzzy system called L2-common feature selection fuzzy inference systems (L2-CFS-FIS), which aims to improve the generalization performance and interpretability of the T-S fuzzy systems simultaneously. Taking advantage of the common feature information among different rules, and introducing the over fitting mechanism, this paper formulates a joint sparse optimization problem based on double regularization terms and further develops an alternating direction method of multipliers (ADMM) procedure to find the optimal solution. The experimental results demonstrate that the proposed method can not only obtain a satisfactory generalization performance of the fuzzy model, but also effectively mine the discriminant attributes from original data to ensure better interpretability of the model.

Key words: T-S fuzzy system modeling, common feature selection, joint optimization, alternating direction method of multipliers (ADMM) procedure