Self-Adaptation in Ant Colony Optimization

Leonor Melo, University of Coimbra, Portugal

Abstract

ACO is a global metaheuristic loosely inspired by the behavior of social ants. Several variants were proposed over the past two decades and, throughout this period, they have been successfully applied to solve difficult combinatorial optimization problems. Notwithstanding its relevance in optimization, ant colony algorithms have several well-known drawbacks. One important limitation is that they tend to be particularly sensitive to parameterization and different settings may obtain significantly different results in the same situation. Also, they have strong greedy components that can easily lead to the loss of diversity and to premature convergence. We will be discussing two self-adaptive ant algorithms. Both approaches rely on the coexistence of heterogeneous groups of ants within a single optimization framework, each set with its own search strategy. Moreover, the search strategy is not fixed and, instead, the algorithms can autonomously adapt their behavior to the different stages of the optimization problem being solved. On-line self-adaptation has two crucial advantages: it frees the practitioner from having to carefully define settings for each specific optimization situation and it grants the algorithm the ability to adjust its behavior in accordance with the structure of the search landscape.