Managing search in a partitioned Search space in GA

Abstract

Converging to suboptimal solutions in genetic algorithms prevents the search from reaching the global optima. Search space could have several suboptimal but one optimal solution. As the suboptimal solutions are within the search space, dividing the search space would bound them in different divisions. Thus, searching in each division separately would increase the probability of reaching the global optima. In other words, the optimal solution would be bounded in one of the divisions and then searching that division would result in finding the optimal solution. Although, the suboptimal solutions could be in the same division as optimal solution but the chance of finding the optimal solution in this case would be more compared to the cases that have no division. The proposed methodology divide the search space into partitions called regions. Individuals will be assigned to each region. The search continues while each set of individuals are focused in searching a region. Preliminary results shows a fair improvement in the performance and efficiency compared to genetic algorithm.