A method for estimating lithium-ion battery state of health based on physics-informed hybrid neural network

Abstract

Lithium-ion batteries (LIBs) are widely used in portable electronic devices and electric vehicles due to their high energy density and long cycle life. However, aging is inevitable during battery cycling, leading to capacity degradation and performance deterioration, which in turn affects the accuracy of state of health (SOH) estimation. To address these issues, this paper proposes a physics-information hybrid neural network (PIHNN). By integrating the electrochemical-thermal-mechanical-side reaction coupling (ETMS) aging model with data-driven methods, the proposed framework achieves accurate capacity loss prediction. The PIHNN framework innovatively introduces membrane resistance as a key health indicator. A physical constraint term, based on the monotonic relationship between membrane resistance and capacity loss, is embedded to enhance physical consistency and prediction accuracy. Additionally, Bayesian optimization algorithm (BOA) is employed for efficient hyperparameter tuning, further improving model performance and computational efficiency. The results demonstrate that under different operating conditions (1C, 0.5C, and 2C), the PIHNN significantly outperforms traditional models in terms of mean absolute error (MAE) and root mean square error (RMSE). The model exhibits superior predictive performance and robustness. In addition, validation on a publicly available dataset from Oxford University reduces MAE and RMSE to below 0.5 %.