A Method for Estimating the State of Charge of Lithium-Ion Batteries Considering Temperature

Abstract

Ensuring the safe operation of electric vehicles relies heavily on accurately estimating the State of Charge (SOC) of the battery. However, traditional SOC estimation methods often struggle to maintain performance at low temperatures and complex operating conditions. This study proposes a novel temperature-inclusive SOC estimation approach. In this approach, a simplified Multilayer Perceptron (MLP) battery model replaces the Equivalent Circuit Model (ECM), with temperature as the key input variable. Moreover, to enhance the robustness of the model, the training dataset is augmented by introducing noise. Combining the refined battery model with the Unscented Kalman Filter (UKF), the MLP-UKF SOC estimation framework is devised. Performance evaluation, generalizability analysis, and temperature adaptability testing are conducted using independent datasets. The results illustrate that the proposed model rapidly converges with the reference curves at different temperatures and operational conditions, and the average error is less than 3%. Furthermore, compared with the ECM model employing Recursive Least Squares parameter identification (MAE: 0.03228, RMSE: 0.03457), the mean absolute error (MAE) and root mean square error (RMSE) of MLP-UKF method are 0.01273 and 0.01430, respectively, showing better performance. The results of the analysis confirm the applicability of the MLP-UKF SOC estimation method for electric vehicle applications, particularly in challenging conditions such as low temperature and complex operations.