Revistes Catalanes amb Accés Obert (RACO)

Experiments writh variants of ant algorithms

Thomas Stützle, Sebastian Linke

Resum


A number of extensions of Ant System, the first ant colony
optimization (ACO) algorithm, were proposed in the literature. These
extensions typically achieve much improved computational results
when compared to the original Ant System. However, many design
choices of Ant System are left untouched including the fact that
solutions are constructed, that real-numbers are used to simulate
pheromone trails, and that explicit pheromone evaporation is
used. In this article we experimentally investigate adaptations of
ant algorithms to the traveling salesman problem that use
alternative choices for these latter features: we consider using
pheromones to modify solutions and different schemes for
manipulating pheromone trails based on integer pheromone trails
without recurring to pheromone evaporation.

Text complet: HTML