Abstract
We propose a parallel genetic algorithm with sexual selection. In genetic algorithms with sexual selection with one population, females keep their traits around local optima by using lower mutation rate than males’ one, while males change their traits actively. When a runaway process takes place, the transitions of males’ traits are biased toward a certain direction which is decided by the bias of females’ preferences. If the population size is large, the search converges quickly. The large population size, however, causes the decrease of the search performance. In the proposed method with parallelization, the population size of each sub-population is kept adequately, and each sub-population searches its own direction of evolution independently. As a result, the proposed method makes a search converge quickly because the runaway process which leads to the intermittent evolution tends to take place more quickly than one population model. We applied the proposed method to some test problems. In these problems, while the performance of conventional genetic algorithms decreased by parallelization, the proposed method revealed better performance by parallelization. Moreover, the performance of the proposed method was better than the ones of conventional methods. This availability of parallelization is characteristic of the sexual selection.
