Establishment and Analysis of Correlation Between Li-Ion Battery Model Accuracy and Kalman Filter-Based SOC Estimation

Battery management systems (BMS) rely heavily on estimating the state of charge (SOC) as the foundation for controlling other functionalities. Significant advancements in the use of non-linear Kalman filters (KFs) aim to address the growing need for model-based state estimation in BMS. The KFs are robust enough to handle the effects of process and sensor noise. However, KFs cannot entirely eradicate the inherent battery model error. Consequently, the precision of SOC estimation heavily relies on the model’s accuracy. Despite this importance, the precise quantitative correlation between the model and SOC estimation accuracy still needs to be discovered. This article outlines and validates three equivalent-circuit battery models, followed by SOC estimation using the extended Kalman filter under real-time operating conditions. The 2RCH battery model outperforms the others, achieving root-mean-square errors (RMSE) that are 1mV, 9mV, and 12mVlower than those of the 2RC model across three different drive profiles. Correlation and regression analyses of the normalized RMSE and standard deviation are performed to compare the model error and SOC estimation error. The Pearson correlation coefficients of 0.9087 and 0.9175 in the first and second cases, respectively, reveal a strong linear relationship between these parameters. This study examines two metrics to determine how model accuracy affects SOC estimation accuracy: overall error level and the dispersion of the error frequency distribution. The analytical expressions established in this work provide significant information for reliability evaluation to implement a robust control plan in BMS. Additionally, it can be conveniently adapted to evaluate and predict the SOC estimation error when KF-based SOC estimation is employed. The novelty of this work is in quantitatively establishing the link between battery model accuracy and SOC estimation accuracy in a Kalman Filter–based framework, which has not been widely reported.