A Robust State of Health Estimation Method for Lithium-Ion Batteries Based on Physics-Informed Enhancement

Accurate and reliable state of health (SOH) estimation is essential for the safe and stable operation of lithium-ion batteries. However, achieving highly accurate and robust SOH estimation remains challenging due to battery inconsistency, complex and variable working conditions, and the limited ability of models to characterize aging mechanisms. Thus, a deep physics-informed neural network method is proposed that integrates a self-attention mechanism with a Koopman neural operator (KNO) for SOH estimation of batteries. On the one hand, the self-attention mechanism enhances multidimensional health indicators, improving the model’s generalization ability. On the other hand, KNO solves the battery dynamic degradation equations, overcoming the challenges of long-term and complex nonlinear degradation estimation. Finally, the proposed method achieves reliable SOH estimation. Experimental results on the Northeast Electric Power University (NEEPU), Xi’an Jiaotong University (XJTU), and Tongji University (TJU) datasets show that the proposed method achieves the lowest mean absolute error of 0.0032, the mean absolute percentage error of 0.0036, and the root mean square error of 0.0043. The fitting coefficient is as high as 99.07%, representing a 47.25% improvement in performance compared to the existing PINN method. These results indicate the effectiveness and universality of the proposed method, providing a reliable solution for advanced battery health management.