Thermal optimization and control in open-plan spaces using physics-informed graph neural network based optimal controller
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing HVAC control systems in open-plan spaces face challenges in achieving uniform thermal comfort and energy efficiency due to spatial variations in thermal states, air mixing, and neglect of wall and window surface temperatures, while current models are computationally expensive or lack scalability and physical interpretability.
Innovation Solution
A physics-informed graph neural network (PI-GNN) model is employed to divide the space into interconnected cells, capturing thermodynamic interactions and surface temperatures, using a time-resetting strategy to optimize temperature setpoints for multiple HVAC units, leveraging graph neural networks to handle inter-cell coupling and model predictive control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computational fluid dynamics (CFD) is used for precise thermal modeling, then measurement precision is improved, but productivity deteriorates due to computational impracticality for real-time control
Solution Approach 1:
The patent creates a simplified copy of the complex CFD thermal model using a neural network that replicates its predictive capabilities. The neural network is trained on CFD simulation data to learn the thermal dynamics, enabling real-time predictions without the computational burden of running full CFD simulations at each control step.
Solution Approach 2:
The patent replaces the mechanical CFD computation system with a neural network-based predictive model. This substitution transitions from solving complex partial differential equations in real-time to using a pre-trained neural network that provides rapid predictions, enabling real-time HVAC control while maintaining thermal state accuracy.
2Productivity
If multiple lumped nodes without partitions are used for spatial variation modeling, then productivity is improved through faster computation, but measurement precision deteriorates by neglecting wall and window surface temperatures
Solution Approach 1:
The patent applies different modeling approaches to different spatial zones within the room. Instead of uniform lumped-node modeling, it incorporates specific surface temperature modeling for walls and windows that are exposed to external conditions, while using simplified modeling for interior zones, thereby capturing critical thermal variations without excessive computational complexity.
Solution Approach 2:
The patent segments the thermal modeling into distinct components: interior air temperature, wall surface temperatures, and window surface temperatures. Each component is modeled with appropriate complexity based on its contribution to thermal comfort and energy consumption, allowing precise tracking of surface temperatures that affect radiant comfort without requiring full CFD-level detail everywhere.
3Reliability
If physics-informed neural networks (PINNs) are used for thermal modeling, then reliability is improved through physical constraints, but device complexity increases making scalability difficult
Solution Approach 1:
The patent extracts and incorporates only the essential physical constraints needed for HVAC thermal modeling into the neural network, rather than implementing the full complexity of physics-informed neural networks. This selective extraction maintains physical consistency in the thermal predictions while avoiding the excessive complexity that would hinder scalability to larger building systems.
Data Source
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 set-points. 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.