Evolutionary Approaches to Rule Extraction for Fuzzy Logic Controllers

Abstract

We first discuss some limitations of Chan et al.’s [5] method and propose some modifications on their ‘optimized fuzzy logic controller’ (OFLC) to eliminate those limitations. Then we propose a new method to reduce the number of rules in a symmetric rulebase which reduces the search space as well as the design time. Our fitness function can reduce the number of rules maintaining the performance of the rulebase. It requires no prior knowledge about the system. Applying this procedure to the inverted pendulum problem, we get a rulebase containing less than 3% of all possible fuzzy rules and it takes about 42 steps on average to balance over the entire input space. Our results are compared with those of Lim et al.’s [4] method.