Mutation Rates in the Context of Hybrid Genetic Algorithms

Abstract

Traditionally, the mutation rates of genetic algorithms are fixed or decrease over the generations. Although it seems to be reasonable for classical genetic algorithms, it may not be good for hybrid genetic algorithms. We try, in this paper, the opposite. In the context of hybrid genetic algorithms, we raise the mutation rate over the generations. The rationale behind this strategy is as follows: i) The perturbation rate of crossover decreases over the generations as the chromosomes in the population become similar; ii) Local optimization algorithms can undo a considerable level of perturbation and return the offspring to one of the parents; iii) Thus, we rather need stronger mutation at a later stage of a hybrid genetic algorithm. Experimental results supported our strategy.