KL Divergence-Based Pheromone Fusion for Heterogeneous Multi-Colony Ant Optimization

Abstract

In this paper we present a heterogeneous multi-colony ant optimization with a novel interaction strategy named pheromone fusion to balance the search ability and the convergence speed of the conventional ant colony optimization. The pheromone fusion performs interaction directly and effectively by the interchange of the pheromone matrices. It could exploit the benefits of pheromone distribution and take full use of the advantages of heterogeneous sub-colonies. There are also two states defined in this study to control the interaction. The global state based on KL divergence determines which sub-colonies should interact with each other, while the local state based on information entropy decides when a sub-colony starts interaction. These two states greatly improve the adaptability and ensure the effectiveness of the interaction. In addition, a reward and punishment strategy is introduced to adjust the pheromone distribution and facilitate the interaction. The experimental results on the Traveling Salesman Problem demonstrate that the proposed algorithm outperforms the multi-colony algorithms presented in some recent works. The studies also indicate that the proposed algorithm could improve the solution quality and accelerate the convergence compared with single-colony algorithms.