A multi-objective reinforcement learning-based velocity optimization approach for electric trucks considering battery degradation mitigation
提出一种深度强化学习方法，优化电动卡车速度并最小化电池退化，在安全、效率、舒适和电池寿命间取得平衡，相比人类驾驶可降低电池容量损失2.4%-8.3%，循环退化减少27.7%-29.6%，能耗降低35.3%-39.8%。
Electrification of commercial vehicles for more sustainable logistic systems has been promoted in the past decades. This study proposes a deep reinforcement learning method for velocity optimization and battery degradation minimization during operation for battery-powered electric trucks (BETs), aiming to achieve a safe, efficient, and comfortable driving control policy for BETs. To obtain an optimal solution considering both calendar and cyclic battery degradation, Deep Deterministic Policy Gradient and Twin Delayed Deep Deterministic Policy Gradient (TD3) approaches are integrated within a simulation environment. To optimize overall BET velocity performance, a trade-off among safety, efficiency, comfort, and battery degradation is incorporated into the reward function of reinforcement learning using Mixture of Experts (MoE) model. The results indicate that the proposed TD3-MoE model achieves safe, efficient, and comfortable car-following control while optimizing total battery degradation. Specifically, the model achieves reductions in total battery capacity loss ranging from 2.4% to 8.3% at different states of charge (SoC) of battery compared to human-driven scenarios. Moreover, despite calendar battery degradation being inevitable, the cyclic battery degradation is effectively mitigated by 27.7% to 29.6% compared to the same SoCs in human-driving data. Furthermore, the TD3-MoE model achieves significant energy consumption reductions , ranging from 35.3% to 39.8% compared to real car-following trajectories.