A reinforcement learning-optimized charging strategy for lithium-ion batteries with plating mitigation based on multi-physics modeling

Fast charging of lithium-ion batteries (LIBs) represents an important enabler for mass electric vehicle deployment. Nevertheless, accelerated degradation and safety hazards arise from parasitic reactions during fast charging, especially lithium plating. To overcome this limitation, we introduce a phase-field electrochemical-thermal model to predict the temperature rise and plating current during charging and plating processes. This model precisely quantifies lithium-ion concentration gradients within graphite particles, enabling prediction of plating initiation and current magnitude. Utilizing this model, we employ deep reinforcement learning to derive an offline-trained charging protocol eliminating plating risks. Validation confirms comparable charging speed and plating mitigation to model predictive control, while reducing online computation by 35-fold. The proposed charging strategy delivers a practical solution balancing charging speed and safety, outperforming conventional approaches.