A Hybrid Learning Algorithm for Extracting and Optimizing Fuzzy Rule Bases

This paper proposes a hierarchical learning algorithm for optimizing fuzzy rule bases. In this algorithm, Takagi-Sugeno fuzzy neural network is used for the parametric learning of a fuzzy rule base with steepest descent method. Then, the structure of fuzzy rule bases is optimized with the binary coding method, in which a rule corresponds to a gene bit, and a rule base to a chromosome. This hierarchical learning algorithm can reduce the redundant rules and simplify the fuzzy rule base. A computer simulation verifies the effectiveness of the proposed algorithm.