Ultra-early prediction of lithium-ion battery performance using mechanism and data-driven fusion model

Forecasting the battery performance accurately in the ultra-early stage can avoid safety incidents, analyze degradation patterns, and prolong battery cycle life, which is crucially essential for battery management. In this work, a mechanism and data-driven fusion model is developed to predict charging capacity and energy curves over the full life cycle of batteries in the case of only knowing the planned cycling protocol without any usage history. The proposed method can achieve accurate and robust prediction of three types of batteries under different working conditions and ambient temperatures with the root-mean-square error (RMSE) of 73.7, 100.9, and 45 mAh. The maximum charging capacity and energy trajectory can be extracted further. Moreover, the proposed method can also detect battery faults without setting a safety threshold in advance due to the inconsistency of the voltage and capacity evolutions of normal and faulty batteries. [Display omitted] •A fusion model is developed to predict battery performance.•The simulated error of the p2D model can be corrected by deep learning methods.•Charging capacity and energy curves can be predicted accurately.•The charging capacity and energy trajectory can be extracted from predicted curves.•The ISC can be detected accurately without setting a threshold in advance.