A control strategy based on deep reinforcement learning for air-condition system of electric vehicles integrating intelligent sensing of passenger's thermal preference

In order to make the AC (air-conditioning) system of the electric vehicle work smartly, improve the passenger's thermal comfort and reduce the energy cost of the AC system for a larger range, an intelligent control strategy is established for the AC system. This algorithm includes the automatic setting of the cabin target temperature based on the intelligent identification of passenger's thermal habit and intelligent control of the AC system based on the reinforce learning algorithm TD3 (twin delayed deep deterministic policy gradient algorithm). The former component can intelligently identify the passenger's thermal preference and automatically set it as the target temperature of the cabin, making the automatic AC system evolve into the intelligent AC system. The latter component can realize the optimal operation of the AC system by learning its historical data, improve the passenger's thermal comfort and save the energy. To demonstrate the proposed strategy can improve the performance of the AC system, the historical target temperature of the cabin from a certain subject is firstly applied to verifying the identification precision of the thermal preference. The average error is only 0.5 °C. Then, the proposed strategy is compared to the traditional ON-OFF and PID (proportional-integral-derivative) strategies. Because the TD3-based strategy keeps the compressor working at a lower speed and uses higher fan speed to adjust the cabin temperature, it achieves a smaller fluctuation in the cabin temperature and lower energy consumption. Under the test condition, its average control error of the cabin temperature is 55% and 12% lower than those of ON-OFF and PID, and its energy consumption is 3% and 8.5% smaller.