Intelligent control of electric vehicle air conditioning system based on deep reinforcement learning

Abstract

The existing AC system of EVs consumes a lot of energy in summer, resulting in reduced range. Additionally, the cabin temperature (Tcabin) fluctuates due to varying driving conditions. In this paper, an innovative cooling strategy is designed to address the above drawbacks using deep reinforcement learning algorithms. The primary objective of this strategy is to achieve high-precision temperature control while concurrently minimizing the energy consumption of the AC system. The strategy combines compressor speed (Ncompressor) and blower airflow (Bairflow) control to optimize the AC system's performance. The RL controller minimizes energy consumption by reducing Ncompressor while increasing Bairflow, thus ensuring a constant cooling capacity (Q). And it effectively reduces temperature fluctuations within the cabin. Finally, the RL control strategy was compared with on/off control and proportional integral differential (PID) control. The results showed that RL control performed well in the multi-objective optimization of the AC system. In terms of temperature control, it effectively reduces the amount of overshooting, and the minimum temperature fluctuation achieved through RL control is merely 0.07 °C, exhibiting an 84.4 % and 86.5 % decrease compared to traditional control methods. The average absolute temperature error stands at 0.06–0.07℃, maintaining a remarkable precision in preserving the target temperature. Furthermore, RL control not only ensures superior energy efficiency but also reduces energy consumption by up to 5.98 % and 7.65 % compared with traditional control methods.