ADAPT: Physics-Aware Diffusion-based World Models for Adaptive Predictive Transferable HVAC Control
Researchers have developed a new AI model called ADAPT for controlling heating, ventilation, and air conditioning (HVAC) systems in buildings. The model uses physics-aware diffusion-based world models to predict indoor climate conditions and optimize energy consumption. It addresses the challenges of partial observability, thermal inertia, and cumulative prediction errors by incorporating a learnable multi-zone heat-balance regularizer. Experiments show that ADAPT reduces HVA
Researchers have developed a new AI model called ADAPT for controlling heating, ventilation, and air conditioning (HVAC) systems in buildings. The model uses physics-aware diffusion-based world models to predict indoor climate conditions and optimize energy consumption. It addresses the challenges of partial observability, thermal inertia, and cumulative prediction errors by incorporating a learnable multi-zone heat-balance regularizer. Experiments show that ADAPT reduces HVAC energy consumption by 7.3% and occupant discomfort by 30.2% compared to state-of-the-art baselines. The model's robustness is demonstrated in out-of-distribution control scenarios, where it maintains performance with only marginal degradation.
---
Why it matters: This matters because building climate systems consume a significant portion of global energy and contribute to CO2 emissions. Optimizing these systems can help mitigate urban climate challenges and align with UN Sustainable Development Goals.
Source: https://arxiv.org/abs/2608.19804
This article was originally published at: https://arxiv.org/abs/2608.19804