State-of-Charge Estimation for Lithium-Ion Batteries Across Wide Temperature Range and Multiple Working Conditions Based on Improved Open-Circuit Voltage Model and Adaptive Square Root Unscented Kalman Filter Algorithm

Accurately estimating lithium-ion batteries’ state of charge (SOC) is critical. While existing research primarily focuses on optimizing battery models, parameter identification, and filtering algorithms, this paper introduces an open-circuit voltage (OCV) model. Firstly, an innovative OCV-SOC mapping correction strategy is designed for a wide temperature range. Additionally, an open-circuit voltage model based on support vector machine regression (SVR) is developed to accurately capture varying behaviors of batteries at different temperatures, thereby enhancing the stability, robustness, and initial error correction capability of SOC estimation. The impact of various Kalman filter algorithms and OCV models on SOC estimation across a wide temperature range and multiple working conditions is discussed. By integrating a second-order RC equivalent circuit model, dynamic forgetting factor recursive least squares (DFFRLS), and adaptive square root unscented Kalman filter (ASRUKF), the proposed approach achieves SOC-accurate estimation. In the wide temperature range tests of the non-complex conditions, the SOC estimation based on ASRUKF achieves a maximum error, mean absolute error (MAE), and root mean square error (RMSE) of below 0.94%. In contrast, at 0℃ or lower temperatures, SOC estimations based on unscented Kalman filter (UKF) and square root unscented Kalman filter (SRUKF) reach an RMSE of 2.36%. Under complex conditions across a wide temperature range, all three filtering algorithms’ maximum RMSE and MAE exceed 3%. Although the estimation time of SVR lags by 0.3 seconds behind extreme gradient boosting (XGBoost) and polynomial regression (PR), indicating a more complex computational process, it remains within acceptable bounds for real-time SOC applications. With an initial deviation of 20%, the SOC estimation based on SVR upholds an RMSE and MAE below 1.8% across a wide temperature range and multiple working conditions, ensuring that errors are contained within 1.2% under the non-complex operating conditions and below 4.3% under the complex operating condition. At 0℃ or lower ambient temperatures, the accuracy of SOC estimation under the complex condition based on PR significantly deteriorates, with both RMSE and MAE exceeding 3%. The SOC estimation based on XGBoost completely diverges under complex condition across a wide temperature range. Results: (1) UKF, SRUKF, and ASRUKF display the potential to adjust initial discrepancies in fluctuating temperatures and operational conditions. ASRUKF is better at showing the mitigating divergence towards the estimation process's end. (2) OCV models are unsuitable for state-space model-based methods in SOC estimation. Improving and establishing high-accuracy OCV models can effectively enhance the performance of state-space model-based methods. (3) The proposed method achieves real-time, high-accuracy SOC estimation across a wide temperature range and multiple working conditions without significantly increasing algorithmic complexity, exhibiting strong stability, robustness, and initial error correction capability. © 2024 China Machine Press. All rights reserved.