Guided mutation operation based on search degree and fitness estimation

Abstract

Mutation is a fundamental operation in genetic algorithm (GA) for it has a significant impact on global search ability and convergence rate. Traditional mutation operation of GA changes chromosomes randomly (or blindly), which would waste a lot of computational cost in searching less promising regions or those have been searched frequently. To address these drawbacks, this paper proposes a novel guide mutation. The proposed guide mutation makes use of history search experience to estimate the average fitness and search degree of sub-regions in the search space. New chromosomes generated by the guide mutation are more likely to be in regions with higher average fitness and less search degree. In this way, the search efficiency can be improved and the algorithm can have a stronger ability of jumping out of local optima. The proposed guide mutation is incorporated into a simple GA, forming a guided mutation GA (GMGA). The GMGA is validated by testing 23 benchmark functions and the experimental results reveal that the proposed guide mutation is very effective in improving the performance of GA.