The State of Charge (SOC) of the lithium-ion battery plays a vital role in monitoring and optimizing the performance of the battery management system (BMS). Traditional Kalman filter (KF) algorithm requires an accurate understanding of the dynamic model of the system and usually contains unknown statistical noises, which can make the SOC estimation inaccurate. To overcome the problem, this paper proposes a modified Kalman filter, namely Kalman-LSTM, which integrates the Long Short-Term Memory (LSTM) into the KF framework. By incorporating a neural network, the method preserves the data efficiency and interpretability of traditional algorithms while simultaneously learning the dynamic behavior of the system. The accuracy of the Kalman-LSTM method is tested using four datasets: DST, BJDST, FUDS, and US06. The SOC estimation results are then compared with different Kalman filter variants, including EKF, UKF, and AKF. The experimental results demonstrate that the proposed model has superior accuracy compared to benchmark models across various working conditions.
