Multi-caste Ant Colony Optimization Algorithms

In this paper we present a multi-caste ant colony optimization approach, where each caste has its own set of parameters. Two variants are proposed: in the first, the composition of the castes remains fixed throughout the optimization, whilst the other allows ants to move from one caste to another. Results obtained in several traveling salesperson problem instances reveal that the adoption of a multi-caste framework increases the robustness of ACO algorithms. In concrete, we show that the existence of different castes removes the need to carefully define q0, an essential parameter for the success of Ant Colony System.