An hierarchical genetic algorithm for learning Beta fuzzy system from examples

The aim of the work was the full design of a Beta fuzzy system for the modeling of nonlinear processes. We propose a hierarchical genetic algorithm with two nodes connected together. Each node is a real coded genetic algorithm allowing the migration of individuals between each other. These two algorithms are based on a Pittsburgh-style approach where each chromosome encodes a set of knowledge bases. Our main contribution results in the genetic representation wherein each individual is coded as a two-dimension matrix, the number of lines is equal to the number of input variables whereas four columns of the matrix represent one fuzzy rule. One distinguishing feature of this approach is that it gives a standard solution to building a fuzzy logic system and neural networks.