Regression tree-based predictive control with variable separation for airport terminal HVAC optimization

Abstract

Heating, ventilation, and air conditioning (HVAC) systems in airport terminals are among the largest energy consumers, yet existing optimisation approaches are often constrained by their dependence on complex first-principles modelling, limiting scalability in real-world applications. To address this challenge, this study employs regression trees to circumvent the computational demands of traditional physics-based models while enhancing the interpretability of control strategies. The proposed framework features an improved tree construction method that optimises node splitting through a rank-consistent coefficient criterion, significantly reducing computational overhead and enabling efficient processing of high-dimensional, large-scale datasets. Additionally, the validated control algorithm integrates a variable separation mechanism and partition adjustment learning to mitigate abrupt output discontinuities inherent in conventional regression trees, thereby improving system stability and operational performance. Comparative evaluations demonstrate that the framework outperforms existing methods in energy optimisation, providing a robust, data-driven solution for building energy management that ensures computational efficiency, transparency, and adaptability.

Practical Application

This study presents a transparent, data-driven approach for optimising HVAC operations in large-scale buildings, such as airport terminals, facilitating a reduction in energy consumption exceeding 15% while ensuring thermal comfort. By replacing computationally intensive physics-based models with interpretable regression trees, the proposed framework substantially reduces model construction time and enables real-time adjustments of HVAC setpoints. The hierarchical tree-based structure enhances operational transparency, allowing engineers to efficiently validate and adapt control strategies across diverse building configurations. Validated through terminal simulations, this approach effectively balances energy efficiency, occupant comfort, and system stability, providing a scalable solution for dynamic environments without necessitating costly retrofits or specialised domain expertise.