Application of Attention Model for the Traveling Repairer's Problem

In this article, we address the Traveling Repairman Problem, whose objective is to minimize customer waiting time by determining the best sequence of visits. It is an NP-hard routing problem, with great relevance in the fields of logistics and e-commerce, for which a heuristic solution is widely used, but still presents a time bottleneck to solve the problem. Machine learning-based solutions have been applied to routing problems so that, once trained, the time to find a solution is shorter than with heuristics. To solve it, we use a neural network model based on attention mechanisms, capable of identifying patterns in sequential data and generating high-quality solutions. Our models were evaluated on standard benchmark instances from the literature and on real cases from TSPLIB, demonstrating the ability to find efficient solutions almost instantly and to generalize well to problems of different sizes.