Physics-Informed Graph Neural Network for Open-Plan HVAC Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for thermal optimization and control in open-plan spaces face challenges in achieving uniform thermal comfort and energy efficiency due to the lack of scalable and fast thermal modeling, particularly in large open-plan offices, where air-mixing and MRT-induced discomfort are not adequately addressed by existing HVAC control systems.
Innovation Solution
The use of Physics-Informed Graph Neural Networks (PI-GNNs) to model thermodynamic interactions in open-plan spaces, dividing the room into interconnected cells and incorporating time-resetting strategies to handle air-mixing and surface temperatures, enabling real-time optimization of HVAC units for uniform comfort and energy efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computational fluid dynamics (CFD) is used for thermal modeling, then modeling precision is improved, but computation time increases making it impractical for real-time control
Solution Approach 1:
The open-plan space is divided into multiple spatial zones or cells, each represented as a node in the graph neural network. This segmentation allows the complex continuous thermal field to be approximated by discrete zones with representative thermal states, reducing computational complexity while maintaining acceptable accuracy for control applications.
Solution Approach 2:
The patent replaces the traditional CFD mechanical computation system with a machine learning-based prediction system. The GNN model learns thermal dynamics from training data and directly predicts future thermal states, substituting heavy numerical simulations with faster neural network inference that achieves real-time control capabilities.
2Manufacturing precision
If multiple lumped nodes without partitions are used, then thermal comfort optimization is improved, but explicit modeling of wall and window surface temperatures is neglected
Solution Approach 1:
The patent applies different thermal modeling approaches to different spatial locations. Perimeter zones adjacent to walls and windows incorporate surface temperature nodes to capture radiant asymmetry effects, while core zones use simpler air temperature representation. This local differentiation maintains thermal comfort accuracy where it matters most without uniformly increasing model complexity everywhere.
Solution Approach 2:
The patent extends the traditional single-dimensional air temperature modeling by adding surface temperature dimensions for walls and windows. Instead of only modeling air temperature T_air, the system incorporates surface temperatures T_surface as additional state variables, enabling calculation of mean radiant temperature and improving thermal comfort prediction in perimeter zones.
3Loss of energy
If conventional thermal models are used for HVAC control, then energy efficiency is improved, but uniform thermal comfort across open-plan spaces deteriorates due to spatial variations
Solution Approach 1:
The control system applies different temperature setpoints and control strategies to different spatial zones within the open-plan space. Perimeter zones with higher radiant asymmetry receive different treatment compared to core zones, allowing each region to achieve optimal thermal comfort with minimized energy consumption tailored to its specific thermal characteristics.
Solution Approach 2:
The system continuously monitors thermal states across multiple zones and uses this feedback to dynamically adjust HVAC control actions. The GNN-based predictor forecasts future thermal conditions, enabling model predictive control that proactively adjusts setpoints to maintain uniform comfort while optimizing energy usage based on actual measured and predicted spatial variations.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Optimal control of multiple heating, ventilation, and air-conditioning units in an open-plan space demands fast and accurate thermodynamic modeling. Prior methods lack scalability required for effective control in large open-plan offices primarily due to air-mixing interactions. The present disclosure describes a physics-informed graph neural network (PI-GNN) to overcome these challenges. Specifically, thermodynamic interactions are modeled as edges between nodes that represent cells. Further, a modeling approach is used that allows explicit modeling of wall and window surface temperatures which are commonly ignored. The method of present disclosure utilizes PI-GNN as a state-estimator that employs a receding-horizon approach for optimal HVAC control. PI-GNNs are adapted for building HVAC control by incorporating a time-resetting strategy to handle time-dependent ambient conditions and therefore setpoints. The method of the present disclosure outperforms a regular PINN model and other baseline control strategies on thermal model accuracy, computation time, energy consumption, and user comfort.