Lithium battery health state assessment based on vehicle-to-grid (V2G) real-world data and natural gradient boosting model

Abstract

Accurate state of health (SOH) estimation of batteries in vehicle-to-grid (V2G) applications is critical for the safety of vehicles, chargers, and grids. In this paper, a novel SOH estimation model based on inverse ampere-time integration and natural gradient boosting (NGBoost) is proposed and verified by real operating data from V2G electric vehicles (EVs). Raw data is segmented, where segments of charging and grid-feeding with relatively smooth currents are selected to constitute the dataset for research. On top of this, five model inputs have been selected through Pearson correlation and physical meaning analysis. Hyperparameters of the model are optimized and model performance is compared with the other nine commonly used machine learning methods. The results show that the NGBoost model has the highest estimation accuracy with the mean absolute percentage error and root mean squared error of 1.484% and 2.302 A h, respectively. To increase the transparency of the model, the Shapley additive explanation (SHAP) method is utilized to provide a full explanation of its predictions. Furthermore, the model is validated to be robust to noise and shows great potential for integration into embedded battery management systems for fast and accurate SOH estimation in V2G applications.