Conceptual
Browse Instruct Curate

Reinforcement-Learning-Guided Heuristic Search for Hierarchical Arc Routing

A hybrid solver for the Hierarchical Directed Capacitated Arc Routing Problem that couples a fast constructive-plus-local-search heuristic with a reinforcement-learning policy. An Adjacency Matrix Attention Encoder embeds arc features and inter-arc distances, a pointer-network decoder produces a route sequence as a Markov decision process, and Proximal Policy Optimization trains the policy to select which swap operators local search should apply, accelerating convergence without degrading solution quality.