Comparison of Extended Kalman Filter and Long Short-Term Memory Neural Network for State of Charge Estimation of Lithium-Ion Battery

Abstract

The state of charge (SoC) estimation for lithium-ion batteries is an essential function of the battery management system (BMS) for ensuring reliable operation of electric vehicles. The nonlinear behavior of the battery makes estimating of SoC a challenging task. In this paper, SoC estimation based on model-based and data-driven methodology are implemented and compared. In the case of model-based estimation, static OCV-SOC test, AC impedance measurement and system identification technique are utilized to obtain accurate third order equivalent circuit model of the battery. Subsequently, SoC was estimated using Extended Kalman Filter (EKF). In the case of data-driven estimation, long short-term memory recurrent neural network (LSTM-RNN) is adopted. The Input that fed into the network are terminal voltage and current, while output is SoC. The training set is WLTP Class 3 driving cycle, and the test set is WLTP Class 2 driving cycle. The performance of state estimation based on EKF and LSTM-RNN is evaluated through the RMSE. Considering only the accuracy of estimation, the SoC estimation using EKF is more accurate than LSTM-RNN, demonstrating the effectiveness of model-based state estimation.