Optimizing battery performance through thermal management in electric vehicles using heterogeneous ensemble learning: a study on energy efficiency and safety

Abstract

This paper presents a novel heterogeneous ensemble learning (HEL) framework for optimizing battery thermal management in electric vehicles (EVs). The proposed approach integrates six machine learning models—linear regression, decision tree, random forest, neural network, XGBoost, and gradient boosting—to enhance temperature prediction accuracy and intelligent cooling decisions. Validated using real-world driving data from 72 trips of BMW i3 (60 Ah) across various environments, the HEL model achieved significant performance improvements, with the stacking ensemble reaching an R² score of 0.9999 and an RMSE of 0.0111. The framework leverages weighted averaging and stacking techniques to capitalize on the strengths of individual models while minimizing their weaknesses. By continuously monitoring the battery parameters, vehicle dynamics, and environmental conditions, the system dynamically adjusts cooling strategies to ensure optimal thermal regulation, energy efficiency, and safety. The modular architecture supports deployment in both cloud-based and vehicle-embedded systems, enabling real-time decision-making and adaptability to diverse operational scenarios. This research advances the development of intelligent, responsive, and comprehensive battery thermal management systems, paving the way for safer, more efficient, and reliable electric vehicles in the future.