Enhancing Focused Ant Colony Optimization for Large-Scale Traveling Salesman Problems Through Adaptive Parameter Tuning

Ant Colony Optimization (ACO) is a well-known family of nature-inspired metaheuristics, capable of finding approximate solutions to difficult optimization problems. The Focused ACO is a state-of-the-art, ACO-based algorithm for solving large instances of the Traveling Salesman Problem (TSP) with hundreds of thousands of nodes. In the current work, we propose four candidate methods for automatically setting a crucial Focused ACO parameter that directly influences the extent to which newly constructed solutions differ from previously generated ones. Computational experiments on a diverse set of TSP instances, ranging from \(11\,849\) to \(104\,815\) nodes, show that two of the proposed methods, which are based on solutions to the multi-armed bandit problem, significantly improve the convergence of Focused ACO to high-quality solutions in 75% of the cases.