Performance Comparison of Battery SoC Estimation Methods

Abstract

Accurate prediction of the State of Charge (SoC) is essential for the efficient operation of electric vehicles through Battery Management Systems. But it is difficult to make accurate SoC estimates for lithium-ion batteries because of their nonlinear chemical nature and dependency on factors like temperature and aging. This study evaluates three different approaches for SoC estimation: Feedforward Neural Network (FNN), XGBoost Regressor, and Extended Kalman Filter (EKF). To do so, the models are trained by employing battery datasets collected under different driving cycles and temperatures. Evaluation of the models is based on errors including Mean Absolute Error (MAE), Mean Square Error (MSE), and Mean Percentage Error (MPE).