Lithium Battery State-of-Health Estimation Based on Physics-informed Neural Network and Feature Extraction
Estimating the State-of-Health (SOH) of lithium-ion batteries is critical for electric vehicle safety and performance, yet it remains challenging due to complex, nonlinear aging mechanisms. This paper proposes a SOH estimation method integrating health feature extraction and a Physics-informed Neural Network (PINN). Twenty health features are extracted from voltage characteristics, incremental capacity curves, capacity degradation trends, and efficiency indicators. The PINN model embeds physical constraints—monotonic decay and smoothness of SOH evolution—as soft penalties in the loss function. Validated on the publicly available Xi’an Jiaotong University aging dataset using leave-one-out cross-validation, the proposed model achieves a minimum MAE of 0.159% and R2 of 0.998 on 2C-batch batteries. Across various operating conditions, the PINN yields a standard deviation of MAE below 0.15, outperforming CNN and ResNet in both accuracy and robustness, demonstrating the effectiveness of incorporating physical priors for reliable SOH estimation.