Koopman-Based Model Predictive Control for Energy-Efficient Air Conditioning in Electric Vehicles

Abstract In this paper, the development of a model predictive control (MPC) framework based on Koopman operator theory for the control of vapor compression system and cabin temperature in light-duty electric vehicles is presented. The inherent nonlinear dynamics of vehicle air conditioning (AC) systems poses significant challenges for traditional control approaches, often leading to suboptimal performances in terms of energy efficiency and computational times. The linearization capabilities of Koopman-based algorithms that lift the original state space into a higher-dimensional space facilitate the design of an MPC that can effectively handle the system's complexities. The proposed Koopman-based MPC framework is designed to optimize the AC system's performance by predicting future states exploiting the approximate higher-dimensional linear model, and adjusting control inputs accordingly. This approach proves to be successful in maintaining the desired cabin temperature, as computed through a specific temperature comfort indicator, while reducing energy consumption compared to baseline controllers. The efficacy of the method is demonstrated through simulations on the high-fidelity nonlinear model, considering different driving cycles and compared against a baseline proportional-integral controller. Results show improvements in both energy efficiency (∼10% on each driving cycle) and computation times, between 20 and 170 times faster than real-time on the tested cases.