Physics-based battery SOC estimation: A joint Data-driven and window-varying adaptive extended Kalman filter approach

Accurate state of charge (SOC) estimation is critical to the safe and efficient operation of lithium-ion (Li-ion) batteries. In this work, a novel method that integrates a Random Forest-Least Absolute Shrinkage and Selection Operator (RF-LASSO) with a window-varying adaptive extended Kalman filter (WVAEKF) is applied to a physics-based battery model to estimate SOC. The algorithm integrates the strengths of data-driven methods and adaptive filtering to enhance the precision of SOC forecasting. The proposed physics-based battery model is a novel equivalent circuit model with diffusion dynamics. This model integrates the diffusion dynamics principle to provide a detailed account of the voltage loss encountered by Li-ion batteries during the diffusion phase. The WVAEKF algorithm's estimation error due to inaccurately capturing the distribution changes of the error innovation sequence and inherent instability of battery model, particularly under dynamic load conditions, is compensated by the introduction of a data-driven RF-LASSO model to improve the accuracy of SOC estimation. The experimental results show that the maximum SOC error of the SOC estimation method proposed in this study is less than 1% under the Dynamic Stress Test (DST) and Federal Urban Driving Schedule (FUDS) conditions. Compared with the conventional AEKF, this method has higher SOC estimation accuracy. Under DST and FUDS conditions at 25 degrees Celsius , the peak error of SOC estimation is significantly reduced by 58.73% and 57.53%, respectively. This improvement highlights the reliability and accuracy of the method in SOC estimation.