Predictive Modeling of Variable Air Volume Terminal Valve Performance Using a Multi-Channel Convolutional Neural Network
Year: 2026
Type: article
Abstract: Accurate prediction of the Variable Air Volume (VAV) terminal valve opening is critical for maintaining balanced airflow distribution and optimizing energy efficiency in Heating, Ventilation, and Air Conditioning (HVAC) systems. However, the nonlinear and time-varying interactions among occupancy behavior, environmental conditions, and actuator dynamics make this task highly complex. Variations in these interdependent factors often lead to unstable airflow regulation and reduced system performance, particularly in under-actuated zones where ventilation demand fluctuates rapidly. To address this challenge, this study introduces a Multi-Channel Convolutional Neural Network (MC-CNN) framework to predict VAV terminal valve opening by processing heterogeneous input data across multiple feature channels. Each channel independently learns temporal and contextual patterns, and a fusion mechanism then integrates all representations to generate accurate predictions. Experimental results show that the proposed model achieves outstanding predictive performance, with RMSE = 0.4812, MAE = 0.3739, and R² = 0.9993, surpassing state-of-the-art approaches. The findings demonstrate that the multi-channel architecture enhances cross-domain feature interpretation, model stability, and adaptability, providing a scalable foundation for intelligent and energy-efficient airflow control within HVAC systems.
Institutions: Trilogi University, Asia e University
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7202392380