规则递归T-S模糊模型及其辨识方法

针对传统T-S模糊模型不能较好描述系统时变特性的问题,提出了一种基于递归策略的动态T-S模糊模型及其辨识方法.规则递归T-S模糊模型在传统T-S模糊模型基础上,增加了具有一定权重的反馈环节,该环节对当前激励强度与前一时刻激励强度进行加权和得到当前时刻新的规则激励强度,从而实现动态递归变化,有效描述了系统的动态过程.为使规则递归T-S模糊模型具有较少的规则数量和较好的泛化能力,前件参数采用一种基于规则激励强度的模糊聚类算法获得,而后件和递归环节参数则采用一种由支持向量机和粒子群优化算法组成的联合辨识方法获得.Box-Jenkins煤气炉的仿真结果表明,规则递归T-S模糊模型及其辨识方法具有较好的动态描述能力,与混合聚类方法相比,均方差降低了1.2％.

A T-S Fuzzy Model with Recurrent Rule and Its Identification Method

A dynamic T-S fuzzy model with a recurrent rule structure(TFM-RR) and its identification are proposed to improve the problem that conventional T-S fuzzy models can not exactly describe the time-varying characteristics of systems.A weighted feedback component that bases on the traditional T-S fuzzy model,is introduced in TFM-RR,and produces a new firing strength of the current rule from the weighted sum of the current firing strength and the previous firing strength.Thus,the firing strength of a rule varies dynamically and recursively,and effectively describes the dynamic process of the system.In order to make TFM-RR has fewer rules and good generalization capabilities,parameters of the antecedent of a rule are achieved using a fuzzy clustering algorithm that bases on the firing strength of the rule,while parameters of the consequent and the recursion are achieved by an integrated identification method that combines the support vector machine and a particle swarm optimization algorithm.Simulation results and comparisons with the hybrid clustering method on Box-Jenkins gas furnace show that the TFM-RR and its identification algorithm significantly reduce the mean variance by 1.2%,and show a better dynamic description ability.