A selectionless two-society multiple-deme approach for parallel genetic algorithms

Abstract

A novel multi-deme parallel genetic algorithm approach that eliminates the use of the selection operator by using multiple populations separated into two societies is introduced. Each individual population contains two subpopulations, one in each society, and individuals in one society are superior in fitness to the ones in the other and the size of subpopulations in each society is dynamically determined based on the average fitness value. The fitness-based division of individuals into two social subpopulations is based on the fact that, due to fitness-based selection procedures, most of the recombination operations take place among individuals with an above-average fitness value. Unidirectional synchronous migration of individuals is carried between populations in the same society and in the two societies. The proposed algorithm is applied for the solution of hard numerical and combinatorial optimization problems, and it outperforms the standard genetic algorithm implementation in all trials.