Automatic Identification of Fuzzy Models with Modified Gustafson-Kessel Clustering and Least Squares Optimization Methods

摘要

An automated method to generate fuzzy rules and membership functions from a set of sample data is presented. Our method is based on clustering and uses a modified version of Gustafson-Kessel algorithm. The aim is to divide a product space into set of clusters for which the systems exhibits behavior close to linear. For each of the clusters we produce a fuzzy rule and generate a set of membership functions for the rule antecedent with use of an approach based on curve fitting. Weighted linear least-squares regression is used to obtain consequent functions for TSK-models.