Competition Controlled Pheromone Update for Ant Colony Optimization

Pheromone information is used in Ant Colony Optimization (ACO) to guide the search process and to transfer knowledge from one iteration of the optimization algorithm to the next. Typically, in ACO all decisions that lead an ant to a good solution are considered as of equal importance and receive the same amount of pheromone from this ant (assuming the ant is allowed to update the pheromone information). In this paper we show that the decisions of an ant are usually made under situations with different strength of competition. Thus, the decisions of an ant do not have the same value for the optimization process and strong pheromone update should be prevented when competition is weak. We propose a measure for the strength of competition that is based on Kullback-Leibler distances. This measure is used to control the update of the pheromone information so that solutions components that correspond to decisions that were made under stronger competition receive more pheromone. We call this update procedure competition controlled pheromone update. The potential usefulness of competition controlled pheromone update is shown first on simple test problems for a deterministic model of ACO. Then we show how the new update method can be applied for ACO algorithms.