关键词: 荷电状态估计；扩展卡尔曼滤波；数据驱动模型；锂离子电池；XGBoost 模型

Abstract: Accurately estimating the state of charge (SOC) is challenging due to the changing environmental conditions in which batteries operate. Accurate SOC estimation not only prevents battery overcharging and overdischarging, but also provides accurate range predictions and extends battery life. In this paper, a SOC estimation method for lithium-ion batteries based on a combination of extended Kalman filter (EKF) and data-driven is adopted. Firstly, the recursive least squares method with forgetting factor (FFRLS) is used to dynamically track the characteristic parameters of the battery model. Secondly, the results of the EKF algorithm are input into the Extreme Gradient Boosting (XGBoost) model as training data. With the machine learning and prediction capabilities of XGBoost, combined with the EKF algorithm, high-precision estimation of battery SOC is realized.The simulation results show that the combined algorithm adopted in this paper is better than the EKF and traditional machine learning algorithms in terms of convergence and accuracy. In addition, the results show that the method can achieve accurate SOC estimation and provide accurate and stable SOC values by verifying under different working conditions.

Key words: state of charge estimation; extended Kalman filtering; data-driven model; lithium-ion batteries; XGBoost model