Optimizing temperature prediction in automotive lithium-ion battery discharge cycles using machine learning and SHAP analysis

Abstract

Effective thermal control is essential for lithium-ion batteries used in electric vehicles and energy storage systems, as uncontrolled heat generation accelerates degradation, reduces performance, and increases safety risk. Conventional temperature-prediction studies often report high accuracy using random data splitting; however, such evaluation may not adequately represent performance on unseen battery cells. This study aims to develop an accurate, stable, and interpretable machine-learning framework for predicting lithium-ion battery temperature during discharge cycles under unseen-battery conditions. The NASA lithium-ion battery ageing dataset was used to train and evaluate Linear Regression, Random Forest, and Light Gradient Boosting Machine models. Five input features were considered: measured voltage, measured current, power, cumulative discharged energy, and voltage drop, with measured temperature as the target. Battery-wise validation was performed using cells B0005, B0006, B0007, and B0018, where each battery was used once as an independent test set. Model performance was assessed using root mean square error, mean absolute error, and coefficient of determination. Uncertainty analysis, residual diagnostics, and SHapley Additive exPlanations were used to examine prediction stability, error behaviour, and feature influence. Light Gradient Boosting Machine achieved the strongest overall generalization, with a mean root mean square error of 1.070 °C and a mean coefficient of determination of 0.910. Random Forest showed closely comparable performance and achieved the lowest mean absolute error of 0.893 °C, while Linear Regression showed weaker predictive performance. SHAP analysis identified power, cumulative discharged energy, and measured voltage as the dominant predictors. The proposed framework provides a reliable and interpretable approach for battery temperature prediction and supports data-driven thermal monitoring in battery management systems. Future work should validate the model using broader datasets, real-time sensor data, ageing indicators, and physics-informed learning under dynamic operating conditions.