Energy-efficient management of an electric vehicle heat pump system: nonlinear model predictive control versus control allocation

To increase the driving range and enhance the occupants’ thermal comfort in very cold ambient conditions, modern battery electric vehicles are increasingly equipped with energy-efficient heat pump systems. As these systems are featured by redundant actuators and complex thermal energy flows, optimal coordination of control actions is essential to exploit the full energy management potential. The paper presents and compares two characteristic vehicle cabin optimal thermal management strategies for an indirect heat pump system: (i) a hierarchical control allocation approach and (ii) a nonlinear model predictive control strategy. The control allocation approach relies on offline-optimized control input allocation maps governing cabin inlet air temperature and blower airflow based on heating power demand commanded by a superimposed proportional-integral cabin temperature controller and coordinating auxiliary pump speed and front radiator power inputs. On the other hand, the superimposed nonlinear model predictive control strategy provides online optimization of the cabin inlet air temperature and blower airflow control trajectories on a receding horizon, while accounting for preview information of disturbance variables such as vehicle velocity and the system thermal dynamics. The two strategies are comparatively evaluated by using high-fidelity simulations under heat-up transient and quasi-steady-state conditions. The control allocation approach is additionally verified through in-vehicle testing, thus also providing indirect validation of the system simulation model. The evaluation results demonstrate that for a comparable thermal comfort, the nonlinear model predictive control strategy reduces energy consumption by up to 3% during heat-up transient and around 9% in quasi-stationary conditions when compared with the control allocation approach. These energy efficiency gains come at the expense of significantly increased computational requirements, which need to be supported through sufficiently capable hardware to enable real-time deployment.