Reinforcement Learning-Based Plating-Suppressive Fast Charging of Lithium-Ion Battery by Monitoring Strain Evolution

Charging speed of lithium-ion batteries is significant for widespread electric vehicle adoption. However, fast charging associated degradation reactions, particularly lithium plating, accelerate capacity decay and create safety risks. In this regard, this study developed a novel electrochemical-mechanical coupling model capable of predicting the evolution of battery thickness during lithium intercalation and plating. This model enables the precise estimation of the lithium-ion concentration gradients within graphite particles, thereby identifying the plating onset and quantifying the plating current. Leveraging these predictive capabilities, a deep reinforcement learning architecture is implemented to develop an offline-trained charging policy that eliminates the plating risks. Experimental results suggest that the strategy achieves comparable charging speed and plating suppression to model predictive control while it reduces the online computational demand greatly. The proposed charging framework balances the charging speed and safety for real-world deployment, delivering an efficient alternative to the commonly-used rule-based strategy.