科研成果详情

Thanks to the technological advances, electric vehicles (EVs) are becoming more and more economically competitive especially when commercial use scenarios are in concern. However, unlike internal combustion engine vehicles, EVs tend to have shorter range when traveling at higher speed, which would undoubtedly impact how commercial EV should be operated, where time efficiency plays an important role to the business. In this study, a new variant of EV routing problem is established, which explicitly considers such 'speed-varying range' (SVR) of EVs and en-route partial re-charging. In view of the complexity of the problem, a deep reinforcement learning approach is tailored, leveraging automated entropy regularisation to enhance exploration. Experiment results show that the policies found by the proposed approach outperform the OR-Tools based ones by better coping with the SVR, and exploiting the en-route charging to improve the overall delivery efficiency.