A New Emigrant Creation Strategy with Randomized Sources for Parallel Artificial Bee Colony Algorithm

Abstract

Dividing the whole population into subpopulations or subcolonies then evaluating them simultaneously is one of the most commonly used parallelization approaches to utilize the computational power of the current systems. However, this type of parallelization strategy decreases the population diversity because of the division of the entire population and needs migrations between subpopulations to maintain the solution diversity until the end of the iterations. In this study, we proposed a new emigrant creation strategy in which the parameters of the best food source being migrated to the neighbor subpopulation is modified with the more appropriate parameters of the randomly determined solution or solutions and investigated its effect on the performance of the parallel Artificial Bee Colony (ABC) algorithm. Experimental studies showed that newly proposed emigrant creation strategy based on randomized solutions significantly improved the convergence performance and solution qualities of parallel ABC algorithm compared to the its standard serial and ring neighborhood topology based parallel implementation for which the best solutions are directly used as emigrants.