Identification and simplification of T-S fuzzy neural networks based on incremental structure learning and similarity analysis

Abstract

Listen

Translate article

This paper proposes a novel identification method and a simplification scheme for T-S fuzzy neural networks, which consists of two steps. The first step refers to the structure design based on an incremental clustering approach whose basic ideas are that the structure identification of fuzzy neural networks is guided by the attenuation of output approximation error in each cluster and processed by a recursive refined clustering iteration with the input space clustering and sub-clustering as the main steps. Once the structure of a T-S fuzzy neural network is identified by the incremental clustering approach, its parameters are further learned and refined by the Levenberg-Marquardt optimization algorithm. The second step refers to the structure simplification including removing redundant fuzzy rules and merging highly similar fuzzy rules. Furthermore, the performances of several similarity calculating methods are analyzed and discussed, which provides a basis for the selection of the appropriate similarity analysis and effective calculating method for the merging of fuzzy rules and system simplification. That is, the given performance analysis provides the methodology basis and design guide for structure simplification based on similarity analysis and merger of fuzzy rules. Several experiments are implemented to illustrate the feasibility and effectiveness of the proposed approach.