Energy management strategy optimization of fuel cell vehicles based on long-term and short-term hydrogen consumption prediction

Hydrogen consumption is not only an evaluation metric for the economic performance of fuel cell vehicles (FCVs), but also one of the key optimization objectives in energy management strategies (EMS). However, EMS that rely on cumulative hydrogen consumption (HC-C) or hydrogen consumption per 100 km (HC-P100) as optimization objectives are limited by the inaccuracies and time delays in hydrogen consumption data. This study proposes an algorithmic framework called Proximal Policy Optimization with Hydrogen Consumption Prediction (PPO-HCP) to optimize energy management strategies for FCVs. First, establishing a dynamic system model for FCVs. Then, innovatively designing a novel reward function to enhance the adaptability of deep reinforcement learning (DRL)-based EMS under complex and dynamic conditions. This includes weighting different terms in the reward function, such as short-term real-time hydrogen consumption (HC-RT), long-term HC-C and HC-P100, fuel cell power, and battery state of charge (SOC). Finally, the proposed PPO-HCP algorithm is evaluated and compared with the conventional PPO algorithm under both training and random conditions. The results show that the energy consumption optimization effect of the PPO-HCP algorithm is more significant, with HC-P100 reduced by 5.3 % under training conditions and 7.9 % under random conditions.