Ant Colony Optimization - Optimal Number of Ants

Sammanfattning: The focus of this thesis paper is to study the impact the number of ants has on the found solution of the Ant Colony Optimization (ACO) metaheuristic when solving the Traveling Salesman Problem. The goal was to find out how the length of the computed tours change for different amounts of ants within a limited number of iterations. To study this, three well known versions of the ACO algorithm were implemented and tested: Min-Max Ant System (MMAS), Elitist Ant System (EliteAS) and Ranked Ant System (RankedAS). The results showed trends that were consistent over several test cases. EliteAS and RankedAS which both utilize specialist ants showed clear signs that the number of specialists had a large influence on the length of solutions. Meanwhile, normal ants did not affect the solutions as much. MMAS and EliteAS only had a small variation on the answer, with lower amount of ants being more favorable. On the other hand, RankedAS performed better by a large margin when working with five specialists and a number of ants equaling the number of cities in the problem.