Durability‐Oriented Energy Management for Multi‐Stack Fuel Cell Hybrid Vehicles

ABSTRACT Energy management strategies (EMSs) for multi‐stack fuel cell systems (MFCSs) rarely co‐optimize fuel economy and durability: offline optimizers cannot track real‐time changes in stack state of health, and most learning‐based EMSs target single‐stack systems, leaving inter‐stack aging consistency unaddressed. This study proposes a two‐layer hierarchical EMS that jointly considers hydrogen consumption and inter‐stack performance consistency. In the upper layer, a causal sliding‐window Daubechies‐4 wavelet scheme decomposes the power demand into low‐ and high‐frequency components in real time without future information; the battery absorbs the high‐frequency component, shielding the stacks from the rapid load fluctuations that accelerate degradation. The proposed ARP‐SC‐NSGA‐III algorithm—improving convergence and suppressing oscillatory power commands over conventional NSGA‐III through adaptive reference‐point regeneration and a smoothness constraint—then yields a compromise total fuel‐cell power reference that balances hydrogen consumption, SOC deviation, and preferred‐range operation. In the lower layer, a trained twin delayed deep deterministic policy gradient (TD3) agent allocates the upper‐layer fuel cell power demand among stacks with different aging states to reduce inter‐stack degradation imbalance. Hardware‐in‐loop (HIL) results on the untrained HWFET and WLTP cycles show that the proposed strategy reduces SOC‐corrected equivalent hydrogen consumption by up to 9.71% and mean degradation inconsistency by up to 93.84% relative to fuzzy control, and lowers hydrogen consumption by 4.55% under WLTP relative to the equivalent consumption minimization strategy (ECMS), while maintaining the battery state of charge (SOC) within ±0.05 of its target value.