Distributed genetic algorithms with randomized migration rate

Abstract

Discusses the effect of randomization of migration rate in distributed genetic algorithms (DGAs). DGAs are extended algorithms of GAs that can be performed in parallel. In DGAs, the total population of genes is divided into subpopulations called islands. In each island, a simple GA is performed and some of the individuals in each island are moved to another island after a certain migration interval. The number of genes in the DGA is determined by the migration rate. Although DGAs can find optimum solutions even when there are several peaks in objective functions, they require more parameters than simple GAs and are therefore more time intensive. We describe a new DGA in which the migration rate is randomized (DGA/rmr). This algorithm is evaluated using two numerical simulations: the Rastrigin function and the Rosenbrock function. We show that optimal parameters exist in these systems and obtain solutions with the proposed approach. The solutions are not optimal, but are better than those obtained using a DGA with fixed migration rate. DGA/rmr may therefore be a less time intensive alternative to conventional DGAs.