A Physics-Informed Data-Driven Method for Identifying the Temperature-Dependent Degradation Using Sparse Aging Data at Discrete Temperature Points

The aging behavior critically affects the safety and service life of lithium-ion batteries. To overcome the challenges associated with time-consuming and costly aging tests, this study proposes a physics-informed data-driven framework that integrates an electro-aging model with deep learning techniques to enable efficient prediction of temperature-dependent battery degradation. The electro-aging model is first employed to simulate battery aging behavior over a wide practical temperature range, instead of extensive multi-temperature aging experiments. Based on the data generated by the physical model and experimental aging data measured at 5 temperatures, a transfer learning model is developed. Through a fine-tuning strategy, the model can allow efficient estimation of long-term degradation using only a single initial constant-current charging cycle at a specific temperature point. Validation using sparse experimental data demonstrates that the proposed model achieves root mean square errors below 5.43 Ah for a 280 Ah cell, corresponding to only 1.94% of the nominal battery capacity across a temperature range of 10–75°C.