Fuzzy associative memory optimization using genetic algorithms

Recently genetic algorithms (GAs) have been used to search high performance fuzzy membership functions in increasing number of control applications. They differ primarily in the ways the search space is encoded and the objective function is targeted. In this paper, we focus on the effectiveness of GAs to optimize the fuzzification step in the fuzzy rule generation method of Wang-Mendel (1991). The method generates a fuzzy rule base from a set of numerical training data organizing it into a fuzzy associative-memory (FAM) bank, and produces a mapping from input space to output space based on the FAM bank using a defuzzifying procedure. The fuzzification step in the method partitions both the input and output spaces into fuzzy regions. Results show that by optimizing the membership functions via GAs, the total error between desired output data and defuzzified output data can be reduced by as much as 50%