Thermal Management Optimization in Battery Electric Vehicles Using Hierarchical Nonlinear MPC and Machine Learning Techniques

The landscape of global transportation is undergoing a significant transformation with the widespread adoption of electric vehicles (EVs). The reasons behind this major shift in vehicle technology have become widely recognised and understood at this point in history, as are the challenges. One significant problem to address on this matter is maximizing travel distance by optimizing the use of the batteries' limited capacity, while ensuring passengers' comfort. The aim of this thesis is to improve the performance of the control strategy proposed by previous works over the Thermal Management System of a Battery Electric Vehicle by applying Machine Learning techniques. This is approached with the objective of maximising battery lifespan and minimising energy consumption. The control strategy is based on Two-level Hierarchical Nonlinear Model Predictive Control, which enables the adjustment of control actions by anticipating and responding to predicted future conditions. This strategy is employed across two separate levels: (i) the reference generator and (ii) the online feedback loop. Firstly, the reference generator takes into account information about the environment provided by Vehicle-to-Everything (V2X) connectivity and builds an ideal control command that is optimised over the whole planned trip. Secondly, the online feedback loop follows the generated instructions and adjusts the course based on real-time needs and information. The benefits of this control strategy have been constrained by the accuracy of the model used to predict the system's behavior. In light of this, this work leverages Recurrent Neural Networks (RNNs) and experimental data for system identification. By means of a training algorithm developed by the System Identification and Control (SIC) research group at DAUIN, Politecnico di Torino, and exploiting the recurrent nature of RNNs that well fits the feedback behaviour of dynamic systems, a new robust model is constructed and validated through simulation. With the integration of this newly developed model into the control scheme, enhancements in the vehicle's energy management performance have been achieved, leading to energy savings in both high and low external temperature scenarios.