Energy management strategy for ammonia-hydrogen hybrid vehicles based on prior knowledge-informed deep reinforcement learning

To address the challenge of coordinating the ammonia hydrogen volume ratio and excess air ratio under complex operating conditions for ammonia‑hydrogen hybrid vehicles, this paper proposes a deep deterministic policy gradient (DDPG) energy management strategy that systematically incorporates prior experimental knowledge. Initially, bench tests of the ammonia‑hydrogen internal combustion engine (AHICE) were conducted to investigate the effect of the ammonia-to‑hydrogen volume ratio on engine power across different rotational speeds and excess air ratios. This established the engine's optimal operating range and delineated its optimal operating curve. Subsequently, a vehicle dynamics model was developed in MATLAB/Simulink, incorporating prior knowledge into the training environment. A DDPG model was constructed using Python and validated under WLTC and CLTC conditions. Compared with the power-following strategy, the proposed DDPG strategy reduces equivalent hydrogen consumption by 9.19% and 19.11% under WLTC and CLTC, respectively. It achieves 98.42% and 97.48% of the globally optimal Dynamic Programming benchmark under the two cycles, respectively, and yields lower equivalent hydrogen consumption than the Equivalent Consumption Minimization Strategy and Model Predictive Control. Meanwhile, the strategy maintains the battery SOC deviation within ±2% of the target value and limits the vehicle-speed tracking error to below 0.5 m/s, indicating satisfactory charge-sustaining performance and drivability under the tested cycles.