Der Fachzeitschriftenartikel ist unter folgendem Link erreichbar: https://ieeexplore.ieee.org/document/11638223

Abstract:

The efficiency and reliability of electric vehicles (EVs) are highly dependent on the performance of their thermal management systems (TMSs), which regulate the temperature of key components, including the battery pack, electric machines, power electronic converters, and the passenger cabin. This article presents a hybrid simulation framework designed to enhance coolant temperature prediction in EV TMSs using informed neural networks (NNs). The proposed architecture combines analytical models for components such as electric motors and radiators with a pre-prediction module that addresses the initial value problem commonly encountered in time series forecasting with NNs. Experimental results demonstrate that embedding physical vehicle information into the NN significantly improves prediction accuracy. Informed models achieve up to a 38.9% reduction in mean absolute error (MAE), with an average improvement of 26.9% across scenarios. This framework represents the first comprehensive approach to integrating physical modeling with data-driven techniques, offering a more accurate and reliable simulation of EV TMSs based on real-world data.

This paper appears in: IEEE Transactions on Industrial Electronics Print

ISSN: 0278-0046 Online ISSN: 1557-9948 Digital Object Identifier:

10.1109/TIE.2026.3711472