Machine learning-based predictive control of thermal management system in battery electric vehicles

This study presents a predictive control logic based on machine learning (ML) for the thermal management system (TMS) of battery electric vehicles (BEVs), aiming for cost-effective energy consumption and response time. The developed methodology consists of six critical steps. The first step is to collect data from virtual integrated models or vehicle tests. The second step requires the selection of the components to be controlled, such as an air conditioning (AC) compressor. Then, the nonlinear autoregressive model with exogenous inputs (NARX) metamodel is built to predict key thermal parameters of cabin temperature, battery temperature, and compressor power consumption, for both heating and cooling processes. Unlike more computationally intensive models such as long short-term memory (LSTM) networks, the NARX framework offers a low-complexity, real-time compatible solution, making it particularly well-suited for embedded vehicle controllers. It is shown that the developed ML model aligns well with the experimental test data gathered under heating conditions. In the cooling case, the NARX-based model is embedded in an optimization framework to minimize AC compressor power subject to predetermined temperature limits. The results indicate that the ML-based control logic achieved an 18 % reduction in compressor energy consumption, and a 10 % saving in total TMS auxiliary load consumption. Under summer operation conditions, this translates to an approximate 0.6 % increase in vehicle range. The main goal of this study is to demonstrate that machine learning-based predictive control can improve the efficiency of BEV thermal management systems; the proposed framework proved its real-time potential by completing 750 optimization runs in only 7 min, showing measurable benefits in energy savings, range extension, and the development of smarter, more sustainable vehicle architectures.