An air balancing method for ventilation systems considering random fluctuations: Flow prediction and operating condition optimization

This study develops an air balancing method combining Multi-task Gaussian Process Regression (MTGPR) and Genetic Algorithm (GA) to overcome two shortcomings in ventilation system control. Firstly, conventional methods rely on time-averaged velocity measurements for flow modeling, which neglect to account for the phenomenon of random fluctuations in airflow. Secondly, existing methods employ implicit flow prediction models, which can only predict the damper adjustment angles based on target flow rates, while failing to provide explicit branch flow predictions. The proposed MTGPR-GA method addresses these shortcomings through two core advancements: (1) an explicit flow prediction model that accurately estimates time-averaged velocities based on instantaneous velocities from a single sample, and (2) GA-driven optimization of damper configurations and fan operations to achieve precise airflow balancing coupled with energy efficiency enhancement. To validate the effectiveness of the MTGPR-GA method, Computational Fluid Dynamics (CFD) simulation tests were conducted in both three-dimensional and two-dimensional ventilation systems. In the three-dimensional simulation, the average errors of the branch flow rates under four operating conditions were 4.04 %, 4.95 %, 2.64 %, and 3.78 %, with a fan pressure error of 5.01 %. In the two-dimensional simulation, under 30 sets of operating conditions, the average errors of the branch flow rates were 3.50 %, 4.45 %, 4.90 %, and 5.07 %, with a fan pressure error of 2.10 %. In all conditions, the damper of the most unfavorable loop remained fully open, ensuring that the fan operated with minimal energy consumption. The results demonstrate the effectiveness of the MTGPR-GA method in achieving air balancing and its energy-saving potential.