Large Language Models for HVAC Operations in Building Energy Systems: A Critical Review of Methods, Applications, and Deployment Readiness
A systematic review of 66 peer-reviewed studies on LLMs for HVAC operations published between 2023 and March 2026 found that 32 papers focus on building energy modelling, only 4 studies reach pilot-level evidence, none reports sustained operational deployment, and no study was classified as ready-now for industry adoption.
Building automation systems generate rich sensor data but remain insight-poor due to heterogeneous point naming, missing metadata, and fragmented documentation. This review classifies 66 studies across five application families and three LLM method families, assessing evidence realism, deployment readiness, and responsibility boundaries. The corpus is concentrated in building energy modelling (32 of 66 papers), while load forecasting remains too sparse for subfield-level conclusions. Only four studies reach pilot-level evidence, and none reports sustained operational deployment. No study was classified as ready-now for industry adoption; three were near-term and 63 research-only. Several bounded, human-in-the-loop uses merit near-term trials, including point-name normalisation, document-grounded operator support, BEM workflow assistance, and advisory interfaces around physics-based controllers.
The review identifies five application families and three LLM method families, with a concentration in building energy modelling. Evidence realism and deployment readiness are low, with only four pilot-level studies and no sustained deployments. The responsibility boundary between LLM and physical HVAC decisions is a key assessment dimension, suggesting a need for human-in-the-loop designs.
Despite the lack of ready-now solutions, near-term trials are recommended for bounded, human-in-the-loop uses such as point-name normalisation, document-grounded operator support, BEM workflow assistance, and advisory interfaces around physics-based controllers. This indicates a cautious but tangible path for industry adoption.
Potential business value lies in reducing manual effort for point-name normalisation, improving operator support through document-grounded LLMs, and streamlining BEM workflows. However, current evidence does not support immediate ROI, and adoption should be limited to low-risk advisory roles.
Observable next signals include publication of pilot results from the three near-term studies, emergence of load forecasting research, and industry trials of the recommended bounded use cases. Sustained operational deployments would mark a significant shift from research-only status.