Improving The Performance Of Ant Colony Optimisation Algorithms Using Biased Visibility

When used to solve traveling salesperson problems (TSP), Ant Colony Optimisation algorithms use the proximity of two nodes to determine the weight or cost of the edge connecting them. The weights counterpoint the pheromone in the state transition rule, used to determine the desirability of a certain next move. The proximity does not always reflect the desirability of the possible edge correctly, because some vertices are at a further distance from their nearest neighbours than others. In the Ant System and Ant Colony System algorithms nodes whose shortest connecting arcs are comparatively long tend to be chosen by the ants at a very late stage in the tour. Thus unfavourable tours are made, which do not contribute to finding a tour close to the optimum. We propose an improvement which introduces a node-specific bias to ACOs. The biased visibility improves the performance of the AS and ACS algorithms on TSP.