A computational study on ant colony optimization for the traveling salesman problem with dynamic demands

Ant colony optimization (ACO) algorithms have originally been designed for static problems, where the input data is known in advance and not subject to changes over time. Later, long term memory of ACO proved effective reoptimization environment when extended deal with dynamic combinatorial problems (DCOPs). Among major proposals this kind, several adaptations procedures improve information reuse can be identified, as well a population-based algorithm (P-ACO) specifically DCOPs. Indeed, P-ACO drew attention research community due its ability faster process pheromone information, but few works assessing effectiveness adapted also are enough reach more general conclusions on current state-of-the-art optimization. In work, we conduct an extensive experimental campaign evaluate most common procedure identified literature, using underlying (MAX-MIN System, MMAS) relevant proposed (P-ACO). A variant traveling salesman problem demands (DTSP) used test benchmark, similarly investigations Besides carefully setup adopt, our work represents significant contribution for, at least, three reasons. First, first acknowledge that DCOPs require custom-configured parameter settings, use automatic configuration tools task. Concretely, show how hypervolume indicator configure anytime behavior Second, directly compare MMAS P-ACO, isolating local search factor. While proves indeed absence search, able consistently outperform it adopted. Finally, investigation DCOP-specific components ACO, once again search. Results those contribute very little performance allowed remarkably absence. fact, coupled DCOP components, outperforms large part setup.