State of Charge Estimation of Power Lithium-ion Battery Based on a Variable Forgetting Factor Adaptive Kalman Filter

Due to the lack of direct measurement, how to accurate estimate the State of charge (SoC) becomes one of the most crucial tasks in the battery management system recently. In this paper, a linear model with the Variable Forgetting Factor Adaptive Kalman Filter is proposed for the SoC estimation. Firstly, Multiple Linear Regression and Adaptive Kalman Filter are used to predict the initial values of model parameters and determine their threshold. Then, Variable Forgetting Factor Adaptive Kalman Filter (VFFAKF) is proposed for the first time, which makes full use of posterior measurement correction rather than just the current error. Numerical experiments demonstrate the effectiveness of our linear model in estimating the SoC Regardless of whether the exact terminal current is known or not, which is better than the traditional Rint and Thevenin Models. The traditional Rint and Thevenin Models can only obtain acceptable estimation results when the exact terminal current is known. The RMSE of SoC estimation results with the proposed method in this paper is less than 1.4%. The RMSE of the Rint model is larger than 3.6% and the RMSE of the Thevenin model is larger than 2.1% at 0 degrees C with the traditional Extended Kalman Filter (EKF). When the temperature reaches to 25 degrees C and 45 degrees C, the slight increase of the RMSE of our linear model can be compensated by the significantly reduced execution time, which implies a good balance between the estimation accuracy and the computation burden. When the terminal current is unknown exactly, the linear model can reach an acceptable results, the maximum error is less than 5% in FUDS, 25 degrees C. However, neither Rint model nor Thevenin model can obtain good estimation results, especially at the tail end of discharge.