Evaluation on State of Charge Estimation of Li-Ion Battery with Extended Kalman Filter Compared to Random Forest and Gradient Boosting Models

Using an estimator is essential to obtain information about the state of charge (SoC) of the battery at any time because, in direct measurement, the battery capacity of an electric car is unattainable. In this paper, we have set up our real data acquisition circuit of the battery, and we chose the first-order Thevenin model to model the battery and identify the parameters of the model, then we did a discharge and a charge of the model of the battery with a current profile for verifying the performance of an extended Kalman filter (EKF) to estimate the state of charge. The estimation error does not exceed 5%. We have a good estimate of the state of charge despite the errors due to the linearization of the nonlinear function of the model and the errors accumulated during the integration of the current. We also make a comparison of the state of charge and the estimation error with the “Random Forest” and “Gradient Boosting” methods. We performed a battery discharge with a constant current profile. It was noted that the error in estimating the state of charge by EKF does not exceed 2%. On the other hand, we have seen that the estimation error by Random Forest does not exceed 4%, and the estimation error of the state of charge by Gradient Boosting does not exceed 5%. This indicates that the extended Kalman filter (EKF) allows good prediction of the state of charge and error rejection.