Estimation of battery state of charge based on changing window adaptive extended Kalman filtering

Adaptive extended Kalman filter (AEKF) is commonly used for lithium-ion battery state of charge (SOC) estimation. However, it overlooks the impact of changes in the distribution of error innovation sequence (EIS) on the noise covariance, resulting in inaccurate state of charge estimates. To address this issue, this paper introduces a novel approach called changing window adaptive extended Kalman filter (CW-AEKF) algorithm. This algorithm uses variance ratio and Levene test to identify the change in the distribution of error innovation sequence, and adaptively updates the optimal noise window length based on the judgment result to achieve accurate noise estimation. Subsequently, the proposed algorithm is combined with a temperature-corrected second-order RC equivalent circuit model for state of charge estimation. The results of dynamic stress test (DST) at different temperatures show that the changing window adaptive extended Kalman filter algorithm can obtain higher accuracy in state of charge estimation results than other algorithms, with state of charge estimation errors remaining within 1 %. Finally, the state of charge estimation of the changing window adaptive extended Kalman filter algorithm in publicly available datasets is analyzed. The results demonstrate that the proposed algorithm maintains strong generalization ability when facing various working conditions.