Solving Heterogeneous Capacitated Vehicle Routing Problem Assisted with Deep Reinforcement Learning

The heterogeneous capacitated vehicle routing problem (HCVRP) is an NP-hard problem with broad applications and growing importance. Traditionally, exact algorithms and heuristic algorithms have been the two primary approaches for solving this problem, but they face computational efficiency bottlenecks when handling large-scale instances. To address this challenge, this paper proposes a hybrid method by integrating deep reinforcement learning with an existing heuristic algorithm, SISRs. Experimental results indicate that the proposed hybrid method successfully combines the strengths of both heuristic and DRL algorithms, outperforming SISRs and DRL-based method in terms of route length while achieving significant reduction of computation time, compared to SISRs.