Performance evaluation and intelligent control of electric vehicle R290 heat pump/air conditioning system based on reinforcement learning

Abstract

The heat pump air conditioning (HPAC) system constitutes a dominant energy-consuming subsystem in electric vehicles (EVs), under winter operating conditions, its energy consumption can account for up to 50% of the total energy consumption of the vehicle. As an environmentally friendly and thermos-dynamically superior alternative to R134a, R290 emerges as a promising development trend for next-generation automotive HPAC systems. This study proposes a dynamic modeling method and an intelligent control strategy based on reinforcement learning (RL) for the R290 heat pump air conditioning system. The established thermal model was validated via bench tests. The performance of R290 and R134a in the thermal model was compared, and the dynamic responses of three control algorithms: RL, proportion integration differentiation (PID), and model predictive control (MPC). The results show that the optimal filling volume of R290 is only 39% of that of R134a. R290 outperforms R134a in terms of coefficient of performance (COP) in both cooling and heating conditions, and the lower the speed, the greater the advantage. Under the 45 °C condition, the RL algorithm reduces the temperature peak by 70% compared to the PID algorithm and by 51% compared to the MPC algorithm. It also increases the COP by 3.9% and reduces energy consumption by 4% compared to PID. Furthermore, it reduces the compressor rotational speed and features a lower rotational speed change rate, which is highly beneficial for extending the compressor service life. Under the -10 °C conditions, the RL algorithm reduces overshoot by 80% and 63% compared to PID and MPC, respectively. It also increases COP by 5.8% and 3%, and reduces energy consumption by 4.7% and 1.6%.