A reinforcement learning-based segmented cooperative air balancing control method for multiple dampers

Abstract

Air balancing is a latent energy-saving technology in heating, ventilation, and air conditioning (HVAC) system, which ensures accurate airflow delivery to satisfy indoor air quality. Due to the complex and diverse structure of ventilation duct systems and the strong coupling between associated branches, the existing air balancing methods suffer from slow convergence speed and low accuracy. This paper proposes a reinforcement learning-based segmented cooperative air balancing control method (RLSC-AB) for multiple dampers. This method designs a Markov property-based control process to accelerate the convergence speed of air balancing and a fine-adjustment to enhance the accuracy. It features two control models: reinforcement learning model and dynamic fine-adjustment model. Firstly, reinforcement learning model is trained by a dynamic target approach across multiple terminal shapes, which enhances the generalization on both diverse target airflow levels and different shape terminals, and it is employed to rapidly converge the airflow within the ASHRAE standard when the air balancing system deviates from the standard. Subsequently, dynamic fine-adjustment model is conducted to further enhance convergence accuracy when the air balancing system falls within the standard. The method performance is validated on an experimental platform and the results demonstrate that the proposed RLSC-AB method can control the air balancing error within 3.17%, and exhibits excellent general performance for various airflow levels and different shape terminals.