Selection schemes with spatial isolation for genetic optimization

Abstract

We tested genetic algorithms with several selection schemes on a massively multimodal spin-lattice problem. New schemes that introduce a spatial separation between the members of the population gave significantly better results than any other scheme considered. These schemes slow down considerably the flow of genetic information between different regions of the population, which makes possible for distant regions to evolve more or less independently. This way many distinct possibilities can be explored simultaneously and a high degree of diversity can be maintained, which is very important for most multimodal problems.