Machine Learning Prediction of Battery Thermal Health in Electric Vehicles Using Real-World Driving Data

Abstract

This research explores the subject of battery heat management in BEVs by using ML to anticipate the Battery Health Factor (BHF) using actual driving data. A total of 65 data points were retrieved from an MG ZS EV powered by a lithium-ion battery and cooled by liquid systems. The vehicle was tested at varying speeds (30-100 km/h), weights (100-350 kg), and environmental temperatures (20-35°C). By utilizing GridSearchCV for model optimization and 10-fold cross-validation for validation, we were able to attain a R² score of 0.9003, as well as low RMSE and MAE. To find out which traits were most relevant for predicting BHF, we used SHAP (Shapley additive explanations). Important factors for the battery health indicator were found during this investigation to be SoC, MaxCh, BT, and BCL. Based on SHAP's findings, raising the SoC improves BHF, but raising the BT and BCL levels has the reverse impact. Merging ML with physics-based models can further enhance system performance and prediction accuracy, and ML models show tremendous promise for enhancing BEV battery heat management, according to this study.