A novel integrated strategy for coordinated optimization of energy management and air supply control based on multi-agent reinforcement learning for fuel cell hybrid electric vehicle

Precise coordination between energy management (EM) and air supply control (ASC) is critical for enhancing the efficiency and durability of fuel cell hybrid electric vehicles (FCHEVs). Most existing studies optimize EM and ASC in isolation and subsequently connect them through hierarchical control structures, thereby neglecting their intrinsic dynamic coupling effects. This separation compromises global optimality and can induce oxygen-starvation-related power undershoot, undermining both energy efficiency and stack safety. To address this challenge, we propose an integrated air supply and energy management (IASEM) strategy based on multi-agent deep reinforcement learning (MARL), which achieves real-time cooperative optimization of equivalent hydrogen consumption and air supply stability. The underlying coupled dynamics between proton exchange membrane fuel cell (PEMFC) current and compressor motor voltage, and their coupled effect on PEMFC power output and air supply condition, are systematically elucidated. Building on this insight, a heterogeneous multi-agent Soft Actor-Critic (MASAC) framework is designed, where EM and ASC agents coordinate adaptively under dynamic driving conditions. Simulation results demonstrate that the MASAC-IASEM strategy effectively suppresses PEMFC power undershoot and compressor power fluctuations, reducing the standard deviation of oxygen excess ratio by up to 73.5%. More importantly, compared to conventional hierarchical strategies, it achieves up to 8.58% reduction in equivalent hydrogen consumption across three representative driving cycles, thereby significantly improving both system energy efficiency and PEMFC stack longevity. These findings establish MASAC-IASEM as a promising framework for integrated optimization in FCHEVs, bridging the gap between energy management and air supply control.