Reinforcement Learning for Health-Aware Charging Regulation in Lithium-Ion Batteries

Developing advanced fast-charging methodologies for lithium-ion batteries is critically important for electric vehicles, yet remains challenging due to accelerated degradation modes including lithium deposition and fracture of electrode particles. This paper presents a health-aware fast-charging framework based on reinforcement learning that simultaneously mitigates multiple degradation mechanisms while maintaining computational efficiency. A coupled electrochemical-thermal-mechanical model is developed within a phase-field framework to simulate the initiation and evolution of lithium plating and particle cracking. Multi-physics constraints are formulated to restrict lithium plating current and diffusion-induced stress, preventing both short-term capacity drop and long-term aging. A reinforcement learning algorithm is subsequently employed to optimize the charging policy, effectively decoupling the complex model from online operation and significantly reducing computational burden. Experimental findings confirm that the introduced strategy enables rapid charging while effectively suppressing degradation, outperforming conventional methods in both aging suppression and computational efficiency. The framework proposes a viable approach to attaining high-performance fast charging in practical implementation.