Model Predictive Control for Thermal Management Systems in Commercial Battery Electric Vehicles : Investigating controller performance with respect to prediction model and controller tuning in varying ambient conditions

Sammanfattning

Battery electric vehicles (BEVs) have become one of the main focuses for the green trans-formation. In order to ensure satisfactory performance of BEVs, it is important that the different components in the BEVs operate efficiently. One way to ensure an efficient operating point is by controlling the temperature of the components. This is done by the thermal management system (TMS). In order to ensure that the range of the BEVs does not suffer, model predictive control (MPC) is implemented with a goal to minimize energy spent by the components in the TMS. The TMS is divided into two systems: the electric motor thermal management system (EMTMS) and battery thermal management system (BTMS). The components connected to the EMTMS are a radiator fan and a coolant pump. The BTMS utilizes a coolant pump and a heater/chiller. The system model was validated against real data in order to gain trust in the simulation results. The validation was satisfactory despite several system simplifications. Two different system linearization methods were compared: linear time-invariant (LTI) and linear time-varying (LTV). These were simulated and compared using different MPC tunings and varying ambient conditions. The results show that having fixed weights on the MPC can result in sub-optimal and unre-alistic controller behaviour. Online tuning was implemented for the MPC which showed improved robustness for all simulated ambient conditions. The difference in state of charge(SoC) between the different tunings was close to negligible. Both the LTI-MPC and LTV-MPC managed to follow the reference closely when increasing the prediction horizon of theMPC, which is beneficial for the battery state of health (SoH). The controller that showed the best performance was the MPC with online tuning. There was little to no difference between the LTI-MPC and LTV-MPC approaches.