A curriculum-guided deep reinforcement learning framework for health-aware energy management of hybrid electric aircraft

The electrification of aviation propels the urgent need for complex energy management strategies (EMSs) in hybrid electric aircraft (HEA), where the trade-off between fuel economy and battery life presents a critical multi-objective challenge. To address the unstable training and sparse reward issues inherent in health-aware optimization, this paper proposes an enhanced soft actor-critic (SAC) strategy integrated with a soft curriculum learning (SCL) framework and dynamic reward re-labeling (DRR) mechanism. Specifically, a high-fidelity dual-state state of health (SOH) model is established to decouple irreversible degradation from dynamic recovery, incorporating electrochemical relaxation and internal resistance coupling effects. Building on this physics-informed environment, the SCL framework employs cosine annealing to smoothly evolve constraint coefficients from a relaxed to a strict state, effectively mitigating the cold start problem. Crucially, the DRR mechanism combined with a Huber robust loss function is introduced to correct the distribution shift caused by curriculum evolution, ensuring unbiased value estimation during experience replay. Comprehensive simulations and systematic ablation studies on real flight data demonstrate that the proposed strategy improves convergence speed by 81.68 % compared to the standard SAC algorithm and reduces maximum irreversible SOH degradation by 19.54 %. Furthermore, hardware-in-the-loop (HIL) tests validate the strategy’s real-time feasibility and its ability to achieve fuel economy nearing the global optimal solution. This work provides a robust theoretical and engineering paradigm for prolonging the lifecycle of airborne energy storage systems.