Genetic Algorithm with Fuzzy Inference for Convergence Acceleration

Abstract

Genetic algorithms (GA) with a fuzzy inference methodology for convergence acceleration is proposed. Mutation operation is to select bits to flip in a chromosome can achieve the convergence acceleration. For classical GA, a chromosome is considered as a bit string and the mutation of bit strings is to ensue through bit flips at random positions. In this paper, three fuzzy rules are proposed to select a bit to flip in a bit string. The idea of the fuzzy rules is that the higher fitness values may select the lower significant bit to flip and the lower fitness values may choose the higher significant bit to flip in a chromosome. Experiments for testing the performance include two numerical cases and a positron emission tomography (PET) liver image segmentation problem. Experimental results demonstrate our approach has better performance of convergence than classical genetic algorithm. The proposed fuzzy inference rules for mutation of genetic algorithms (GA) reduce the iteration number when the average value of fitness function converges.