An Electrochemical Aging-Informed Data-Driven Approach for Health Estimation of Lithium-Ion Batteries With Parameter Inconsistency

Abstract

Accurate estimation of the state of health (SOH) for lithium-ion batteries is crucial for maintaining their safety, reliability, and sustainability. This article presents an electrochemical aging-informed data-driven approach for battery SOH estimation by integrating physics-based electrochemical model with deep learning model. In addition, electrochemical parameter inconsistencies resulting from manufacturing differences can cause variations in battery aging rates, a factor often overlooked in traditional SOH prediction methods. The proposed method addresses inconsistency by leveraging the initial cyclic state to improve prediction accuracy and adaptability. Furthermore, a physics-informed dual neural network (PIDNN) is developed to estimate electrochemical parameters and the Li$^+$ concentration in both the solid phase and the electrolyte to calculate battery capacity fade. A gradient normalization strategy is utilized to train the model effectively. The prediction performance of the proposed method is assessed using three metrics: mean absolute error, root mean square error (RMSE), and the coefficient of determination (R$^{2}$). Notably, the RMSE remains below 0.556%, 0.310%, 0.187%, and 0.486% across four real-world battery datasets, even when trained with just 1% of the total data. Furthermore, PIDNN effectively simulates Li$^+$ concentration dynamics in both the electrode and electrolyte, demonstrating the exceptional interpretability and accuracy of the proposed method.