Machine learning–driven temperature prediction of fast-charging lithium-ion batteries under diverse protocols and C-rates in electric vehicles

Fast charging of lithium-ion batteries is essential for electric vehicles but may lead to excessive temperature rise and thermal instability. In this study, battery temperature prediction is performed for a Sanyo 18650 lithium-ion battery used in a BMW i3 electric vehicle using machine learn- ing techniques. A total of 70 real driving trips under different operating conditions were analyzed. Several machine learning models, including linear regression, K-nearest neighbors (KNN), support vector machines, decision tree, XGBoost, random forest, and artificial neural network, were developed and compared. The results show that the KNN model provides the highest prediction accuracy, achieving an R 2 of 0.9999, a mean ab- solute error (MAE) of 0.035 °C, and a root mean squared error (RMSE) of 0.05 °C. A comparative evaluation of fast-charging protocols (MCC, CCCV, CPCV, and boost) indicates that the CCCV protocol offers the most stable thermal behavior. Under fast-charging conditions at a 2.6 C rate, the maximum battery temperature reached approximately 36 o C, remaining within the safe operating range without any active cooling system. These results demonstrate the effectiveness of machine learn- ing–based temperature prediction for evaluating fast-charging strategies in electric vehicle applications.