Deep Reinforcement Learning for Large-Scale Efficient Routing of Electric Vehicles

Future smart transportation is anticipated to be much enhanced by autonomous electric vehicles, but efficient routing remains a challenge due to uncertainties and time constraints. Exact solutions for vehicle routing problem (VRP) can not provide a timely solution when the problem size increases. Heuristic and meta-heuristic strategies enhance the exact methods, particularly for mid-sized problems. However, the need for accurate and timely solutions is still an open issue, especially dealing with large-scale VRPs. We propose a novel approach that uses deep reinforcement learning addressing the VRP for electric vehicles with stochastic uncertainties. The problem is mathematically formulated and represented as a Markov Decision Process and a deep Q-Network with custom-designed reward is proposed. The performance of the proposed approach is assessed against Google-OR tools, IBM CPLEX, and state-of-The-Art reinforcement learning-based algorithms. The evaluations show that the proposed methods can handle large-scale instances with uncertainties and exceed them in many cases.