A Hybrid Method Combing Reinforcement Learning and Heuristics in Solving Two-Echelon Vehicle Routing Problem with Backhauls

Addressing the rising importance in urban logistics networks, solving the two-tier capacitated vehicle routing problem (2E-CVRP-B) has become crucial. This intricate challenge involves fleets for linehaul and backhaul demands in a dual-tier logistics setup. The primary fleet transports goods between the depot and satellites, while the feeder-line fleet serves satellites and customers. The NP-hard nature of the vehicle routing problem (VRP) is compounded in 2E-CVRP-B due to intricate interactions between the echelons, especially in large-scale applications. To overcome this, we've developed a hybrid algorithm, combined with heuristics and deep reinforcement learning, called RePart-DRL, which effectively dissects dual-echelon optimization issues. Evaluation using 2E-CVRP benchmark instances and newly generated real-world instances in Birmingham further validates its effectiveness.