A Self-Tuning Method of Fuzzy Rules Based

Abstract

As a kind of learning approach for a fuzzy system model, so-called the self-tuning method of fuzzy rules by Dalta rule proposed by H. Nomura et al. is well known. While the method has high generalization capability, the number of tuning parameters increases quickly for fuzzy system models with multiple-input variables and the shape of the fuzzy rule table is destroyed after the learning. Moreover, sometimes the case of a non-firing occurs for evaluating input data. To improve the above problems, we suggest a new tuning algorithm for learning fuzzy models based on the gradient descent method. In this learning algorithm, the number of tuning parameters is fixed in the learning process and the shape of fuzzy rule table never changes even after the learning. Furthermore, we compare the suggested method with the self-tuning method, and illustrate the efficiency of the suggested method for tuning fuzzy reles by means for several numerical examples.