The air conditioning system of electric vehicle (EV) consumes a lot of energy, and the traditional control strategy is difficult to adapt to the dynamic driving conditions and the individual needs of the occupants. A double-layer collaborative optimization control algorithm combining deep reinforcement learning (DRL) and adaptive PID is proposed. The upper layer employs a Proximal Policy Optimization (PPO) algorithm to quantify the Overall Thermal Comfort Score (OTS) and Local Thermal Comfort Indicators (LTS) based on a multi-physics coupled model. It dynamically adjusts the comfort-energy consumption optimization weights in real time according to driving conditions, battery SOC, and other states. The lower layer realizes fast and accurate tracking of regional temperature through fuzzy adaptive PID. The algorithm breaks through the traditional assumption of uniformity, introduces dynamic clothing thermal resistance and metabolic rate estimation, and brings non-uniform thermal environment into decision-making. The simulation results show that, under the comprehensive working conditions including urban congestion, high-speed cruise, intense driving and low power consumption, compared with the traditional PID and fixed weight MPC strategy, the standard deviation of the proposed algorithm PMV (Predicted Mean Vote) is reduced to 0.18, the proportion of local uncomfortable time is only 6.2%, the cumulative energy consumption is reduced by 23.8%, and the comprehensive performance score is 85.9, which verifies its remarkable advantages in dynamic adaptability and energy efficiency balance.
Jiao et al. (Sun,) studied this question.