State-Of-Health Estimation for Lithium-Ion Batteries Using Supervised Machine Learning: XGBoost

Abstract

Batteries are energy storage systems used in almost every aspect. As the battery will degrade over time as it is used, the Battery Management System (BMS) needs to be able to monitor its health so that the right battery replacement time can be predicted. Several State-of-Health (SoH) estimation methods have been studied, one of which is the data-driven method. This paper proposed SoH estimation for universal lithium-ion batteries using supervised machine learning:
XGBoost that trained with only one battery material, which is Lihtium-NickelCobalt-Alumunium (NCA). From Model evaluation results show that the model is able to predict well on other data with the same material as the training data with RMSE 1.4320% MAE 0.9174% and MAPE 0.0100%. However, to make predictions on other types of material data, the model has difficulty because XGBoost is not able to make good predictions outside of the training data.