Abstract: In order to improve the control efficiency of double duct variable air volume, the deep reinforcement learning algorithm was introduced, and the HVAC system in a building was taken as an example to design and study the adaptive control method of double duct variable air volume regulation. On the basis of considering many factors such as air flow, temperature and humidity, the HVAC dual duct ventilation model is established. The comfort level of indoor environment is calculated. According to the air supply volume of each branch, the branch air supply volume adjustment based on deep reinforcement learning is designed. According to the deviation of indoor environment parameters (such as temperature and humidity) and the adjustment amount of branch air supply, the double air duct air volume to be adjusted is calculated, and the closed-loop adaptive control of variable air volume adjustment is performed. The experimental results show that the designed method can realize the adaptive control of variable air volume regulation in practical application, and the variable air volume can be stabilized at the target value after control.