FEWER HYPER-ELLIPSOIDS FUZZY RULES GENERATION USING EVOLUTIONAL LEARNING SCHEME

Abstract

Fuzzy rules generation is known an important task in designing fuzzy systems. This article applies an evolutionary fuzzy rules learning scheme to approach desired fuzzy systems having a lower fuzzy rules. The proposed learning scheme overcomes limitations of conventional fuzzy rules generation and completes the complex searching problems to extract the desired fuzzy system. In this article, aggregations of hyper-ellipsoids fuzzy partitions with different sizes and different positions are suggested to approximate the knowledge rule base of fuzzy systems whose membership functions are arbitrarily shaped and flexibly tuned in parameters searching space. Several corresponding parameters in defining the region of such hyper-ellipsoids type membership functions are efficiently selected based on the simple rule extracting technology. Furthermore, the constructed fuzzy system with only two fuzzy rules can be automatically extracted by the evolutional genetic algorithms (GAs) learning scheme with the guide of special fitness function. Finally, both inverted pendulum balance and nonlinear modeling problems are used to illustrate the effectiveness of the proposed method.