Physics-informed dual-stage network for lithium-ion battery state-of-charge estimation under various aging and temperature conditions

Abstract

Accurate state-of-charge (SOC) estimation is essential for ensuring the safe and efficient operation of lithium-ion battery-based applications. However, traditional SOC estimation methods exhibit limitations in generalizability across diverse aging and temperature conditions. To address this challenge, this study proposes a physics-informed dual-stage network (PIDN) that enables robust SOC estimation under various aging, temperature, and current conditions. The PIDN method extracts key parameters of the 1-RC equivalent circuit model using a forgetting factor recursive least-squares algorithm. These physics-informed parameters, along with terminal voltage, current, and temperature measurements, are used as inputs to a dual-stage network comprising an aging model and a temperature compensation model for SOC estimation. A Kalman filter is then employed to refine the estimated SOC by leveraging the recursive characteristics of SOC dynamics. The PIDN method is validated under various operating conditions, including different aging levels, temperatures, and dynamic current profiles, using the urban dynamometer driving schedule and US06 tests. The results demonstrate that the PIDN method achieves reliable estimation accuracy, with a root mean square error below 1.76 % and a maximum absolute error below 4.55 % under previously untrained conditions. Thus, the PIDN method effectively combines domain knowledge of lithium-ion batteries with deep learning techniques, offering generalizable performance for real-time SOC estimation in practical battery management systems.