Modular and hierarchical evolutionary design of fuzzy systems

Abstract

The paper introduces a modular, hierarchical evolutionary method to design fuzzy systems. The method uses genetic algorithms to evolve a population of rule-based fuzzy models. For this purpose, three hierarchical modules are defined namely, partition, population of granules, and population of fuzzy graphs. The evolutionary process is designed to find the best set of fuzzy rules induced by the granularity required, the form of the membership functions and the inference procedure. This hierarchical configuration guides to the implementation of an effective process of cooperation and competition among rules, responsible for the small cardinality of the resulting set of rules. Simulation results show that the method does increase flexibility to design fuzzy models, preserves the computational tractability, and improves the descriptive nature of the final solution.