A Long-Term Value Estimation Based Ant Colony System for Traveling Salesman Problem
Year: 2026
Type: conference-paper
Abstract: Ant Colony System (ACS) is an effective metaheuristic for solving the Traveling Salesman Problem, which is a classical NP-hard problem widely studied in combinatorial optimization. However, ACS depends solely on short-term heuristic information and pheromone feedback, resulting in myopic path construction decisions. To address this issue, this paper proposes a long-term value estimation based ACS, which incorporates long-term future return estimation into the solution construction process, together with two novel designs. First, the State-Action-Reward-State-Action algorithm is adopted to estimate long-term action values, and the learned values are integrated into the ACS transition rule through a multiplicative fusion strategy with pheromone and heuristic information. Second, a diversity-and-stagnation co-driven adaptive regulation strategy is designed to adaptively balance exploration and exploitation during the search. Experimental results on TSPLIB benchmarks demonstrate that the proposed method improves solution quality and convergence stability over standard ACS, with more significant gains on large-scale instances.
Authors: Jia-Ying Liu, Jian-Yu Li, Zhi-Hui Zhan
Institution: Nankai University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201848292