Energy-oriented Routing Strategy of Electric Vehicle: An End-to-End Reinforcement Learning Approach

Abstract

Electrification of trucks and vans is an effective way to reduce the cost of last-mile logistics due to their high maneuverability and lowered carbon emissions. However, the limited driving range of electric vehicle (EV) and constrained charging stations bring great challenge to delivery services' routing problem. In this paper, we propose an efficient reinforcement learning based routing method for electric logistic vehicles to reduce the energy consumption while meeting some constraints. First, the EV energy consumption model is presented and an electric vehicle routing problem (EVRP) with time, load and recharging station constraints is formulated. Second, we propose an end-to-end deep reinforcement learning algorithm with pointer network. The algorithm generated instances sampled from a given distribution, and trained a model by applying a policy gradient method while keeping the route feasible. Finally, several numerical experiments show the proposed method can effectively reduce the total energy consumption of the electric logistics fleet with high real-time performance.