A double-subpopulation variant of the bat algorithm

Abstract

The bat algorithm (BA), which has been demonstrated to be competitive with some conventional nature-inspired algorithms, such as particle swarm optimization (PSO) and harmony search (HS), was recently invented by Yang in 2010. However, BA may be poor in balancing exploitation and exploration for certain problems and thus may become trapped in local optima with loss of population diversity. In this paper, by introducing a double subgroup (external exploration subgroup and internal exploitation subgroup) with a dynamic transition strategy to improve the global exploring ability and local exploiting ability of BA, we propose an improved double-subpopulation Lévy flight bat algorithm called DLBA. The external subgroup updates positions using a dynamic weight model and the internal subgroup uses a Lévy flight model. To mitigate a loss of diversity, DLBA enables mutation with mutation probability Mp in the external subgroup when the diversity drops below a given threshold. Several other improvements, such as selection strategy and loudness updating formulae, are also introduced. Our results from tests on a set of numerical benchmark functions indicate that DLBA can outperform other algorithms in most of our experiments.