A New Emigrant Utilization Strategy for Parallel Artificial Bee Colony Algorithm

Abstract

Artificial bee colony (ABC) algorithm is one of the most important swarm intelligence based metaheuristics that models the foraging behavior of real honey bees. Like other swarm intelligence based optimization algorithms, ABC algorithm is intrinsically suitable for parallelization by using extensive computational power of the distributed or shared memory based architectures. In the vast majority of the studies, the whole bee colony is divided into equally sized subcolonies and evaluated concurrently for the parallelization purposes. However, when an algorithm is parallelized, some mechanisms should be modified or new techniques should be introduced. In this paper, a new emigrant creation utilization strategy also called swap model is introduced. The main idea lying behind the swap model is based on directly using the information sent by the topological neighbor to change the best solution of the current subcolony. For investigating possible contributions of the swap model on the performance of the parallel ABC algorithm, a set of experimental studies with different benchmark problems, number of subcolonies and migration periods was carried out. The results obtained from the experiments compared with the serial ABC algorithm and its some variants in addition to the conventional parallel implementation of the same algorithm. From the comparisons, it is concluded that the parallelization of the ABC with the swap model significantly improves the convergence speed of the algorithm while protecting the qualities of the solutions, speedup and efficiency values.