Hybrid energy management strategy combining adaptive frequency separation and deep reinforcement learning for fuel-cell hybrid electric vehicles

• A hybrid enegy management strategy combining Adaptive Frequency Separation (AFS) and Deep Reinforcement Learning (DRL) is proposed for fuel-cell hybrid electric vehicles. • The adaptive frequency layer dynamically splits the power demand, assigning high-frequency components to the supercapacitor and low-frequency components to the fuel cell and battery. • A Deep Q-Network (DQN) optimally allocates the low-frequency power between the fuel cell and the battery based on multi-objective cost minimisation. • The cost function integrates hydrogen consumption, component degradation, and SOC deviation to achieve efficient and durable operation. • The proposed AFS–DQN controller reduces hydrogen consumption by 6.3% and maintains SOC within ± 2% of the reference under dynamic driving conditions. • A comprehensive comparative study between Dynamic Programming (DP), Reinforcement Learning (RL), and Deep Q-Network (DQN) algorithms is performed under multiple standard driving cycles (NEDC, UDDS, and HWFET). FCHEV powertrain structure. This study develops a hybrid energy management strategy (EMS) for fuel-cell hybrid electric vehicles (FCHEVs) by combining Adaptive Frequency Separation (AFS) and Reinforcement Learning (RL). The supercapacitor compensates for high-frequency power fluctuations, ensuring dynamic stability. At the same time, the low-frequency demand is optimally distributed between the fuel cell (FC) and the battery through a Deep Q-Network (DQN) agent. The control objective minimises a composite cost function that accounts for hydrogen consumption, fuel-cell and battery degradation, and state-of-charge (SOC) deviation. Detailed physical models of the FC system (including compressor dynamics), lithium-ion battery ageing, and supercapacitor behaviour are implemented to ensure realistic simulation. Comparative evaluations on three standard driving cycles—NEDC, UDDS, and HWFET—demonstrate that the proposed hybrid FD–DQN controller reduces hydrogen consumption by 6.3% on HWFET, achieves stable SOH profiles for both the FC and the battery, and maintains SOC within ± 2% of its reference compared with conventional rule-based EMS, while ensuring smoother FC power trajectories. Although Dynamic Programming (DP) achieves the global optimum, the proposed method offers a practical balance between fuel economy, component durability, and real-time implementability for next-generation FCHEVs.