A Local Pheromone Initialization Approach for Ant Colony Optimization Algorithm

Ant Colony Optimization Algorithms are the most successful and widely accepted algorithmic techniques based on the decentralized collaborative behavior of real ants when foraging for food. The initialization of pheromone in these algorithms is an important step because it dictates the speed of the system's convergence to the optimal solution. All the proposed initialization techniques in the literature use a single value to initialize the pheromone on all edges. In our paper, instead of using a constant or a pre-calculated value to initialize the pheromone the edges, we propose a local pheromone initialization technique that involves the ants initializing the edges, using local information, as they encounter the edges for the first time. We tested our proposed local initialization using the Ant Colony System algorithm to solve the Travelling Salesman Problem. Our approach, when compared to the standard initialization approaches, provided better results in more than 70% of the tested datasets. Also, our algorithm did not require an initialization for all edges. In general, our local pheromone initialization approach was successful in achieving a balance between the solution quality and the time required to construct that solution even in the cases in which it was not able to find the optimal path.