Adaptive mutation using statistics mechanism for genetic algorithms

It has long been recognized that mutation is a key ingredient in genetic algorithms (GAs) and the choice of suitable mutation probability will have a significant effect on the performance of genetic search. In this paper, a statistics-based adaptive non-uniform mutation (SANUM) is presented within which the probability that each gene will subject to mutation is learnt adaptively over time and over the loci. As a search algorithm based on mechanisms abstracted from population genetics, GAs implicitly maintain the statistics about the search space through the population. SANUM explicitly makes use of the statistics information of the allele distribution in each gene locus to adaptively adjust the mutation probability of that locus. To test the performance of SANUM, it is compared to traditional bit mutation operator with a number of “standard” fixed mutation probabilities suggested by other researchers over a range of typical test problems. The results demonstrate that SANUM performs persistently well over the range of test problems while the performance of traditional mutation operators with fixed mutation probabilities greatly depends on the problem under consideration. SANUM represents a robust adaptive mutation operator that needs no prior knowledge about the fitness landscape of the problem being solved.