Eco-driving and battery sizing for automated plug-in hybrid electric harbor-trucks considering uncertain load

Eco-driving involves the optimization of speed planning and energy management, offering significant potential for energy-saving, safety, and transportation efficiency for automated plug-in hybrid electric harbor-trucks (A-PHEHTs). Since truck load, which affects speed and required power, varies frequently, mass uncertainty must be considered to ensure energy efficiency and system adaptability. Furthermore, the battery capacity directly determines the charging/discharging rate and consequently influences eco-driving. In this context, this paper proposes a deep reinforcement learning method, namely twin delayed deterministic policy gradient (TD3), to coordinate eco-driving for an A-PHEHT under uncertain mass. Monte Carlo random sampling (MCRS) is employed to approximate the probability distribution of truck mass and to determine the optimal battery capacity. To assess the proposed method, TD3 is compared with a hierarchical method (TD3-DP) and a co-optimization method based on deep Q-network (DQN). The results indicate that the proposed method can determine the optimal battery capacity, and that TD3 outperforms both TD3-DP and DQN in speed planning and energy saving. Specifically, for different distribution parameters, the total cost of TD3 is 1.09%–3.57% lower than that of TD3-DP and 103.07%–163.61% lower than that of DQN. Finally, the practicality of TD3 is verified by an STM32F407-based microcontroller.