Large Language Models for HVAC Operations in Building Energy Systems: A Critical Review of Methods, Applications, and Deployment Readiness
Large language models for HVAC operations in building energy systems; none ready for industry deployment.
Key Findings
Methodology
This paper systematically reviews 66 studies on large language models for HVAC operations, categorizing them into five application areas and three method families. It assesses evidence realism, deployment readiness, and the responsibility boundary between LLMs and physical HVAC decisions.
Key Results
- Of the 66 papers, 32 focus on building energy modeling, with only 4 reaching pilot-level evidence and 63 limited to research stage.
- No study is deemed ready for industry adoption; three are near-term.
- Conventional ML and MPC remain dominant for high-frequency control.
Significance
The study shows LLMs primarily serve as semantic and workflow layers rather than autonomous controllers in HVAC, with potential applications in point-name normalization, document support, and BEM workflow assistance.
Technical Contribution
The paper provides a framework linking evidence realism, readiness, and responsibility boundaries, identifying bounded roles suitable for near-term trials.
Novelty
This is the first systematic review of LLM applications in HVAC operations, emphasizing the role of semantic interfaces rather than direct control.
Limitations
- Only a few studies reach pilot-level evidence, lacking sustained operational deployment.
- Research is concentrated on building energy modeling, with insufficient data for load forecasting.
Future Work
Future research should prioritize field-validated benchmarks, orchestration evaluation under operational constraints, and LLM-MPC/RL architectures with verifiable safety properties.
AI Executive Summary
This paper systematically reviews the application of large language models (LLMs) in HVAC operations within building energy systems. Despite the rich sensor data generated by building automation systems, operational use is hindered by heterogeneous point naming, missing metadata, and fragmented documentation. The study analyzes 66 peer-reviewed studies, categorizing them into five application areas and three method families, assessing evidence realism, deployment readiness, and the responsibility boundary between LLMs and physical HVAC decisions. The research is concentrated in building energy modeling, while data for load forecasting is insufficient for subfield-level conclusions. Only four studies reach pilot-level evidence, with none reporting sustained operational deployment. No study is classified as ready-now for industry adoption; three are near-term and 63 research-only. However, several bounded, human-in-the-loop uses merit near-term trials, including point-name normalization, document-grounded operator support, BEM workflow assistance, and advisory interfaces around physics-based controllers. Conventional machine learning, model predictive control, and reinforcement learning remain more adopted for high-frequency control, short-horizon numerical forecasting, and well-posed ontology mapping, while autonomous agentic operation and unvalidated occupant proxies remain research-stage.
Deep Analysis
Background
Building automation systems generate rich sensor data, but operational use is hindered by heterogeneous point naming, missing metadata, and fragmented documentation. Advances in LLMs in other domains bring new possibilities for HVAC operations.
Core Problem
The core problem is how to leverage LLMs to support HVAC operations, addressing issues of heterogeneous point naming, missing metadata, and fragmented documentation.
Innovation
This paper is the first systematic review of LLM applications in HVAC operations, proposing a framework linking evidence realism, readiness, and responsibility boundaries.
Methodology
- �� Systematic review of 66 studies
- �� Categorization into five application areas and three method families
- �� Assessment of evidence realism, deployment readiness, and responsibility boundaries
Experiments
The study analyzes 66 peer-reviewed studies, categorizing them into five application areas and three method families, assessing evidence realism, deployment readiness, and the responsibility boundary between LLMs and physical HVAC decisions.
Results
The research is concentrated in building energy modeling, while data for load forecasting is insufficient for subfield-level conclusions. Only four studies reach pilot-level evidence, with none reporting sustained operational deployment.
Applications
LLMs have potential applications in point-name normalization, document support, and BEM workflow assistance.
Limitations & Outlook
Research is concentrated on building energy modeling, with insufficient data for load forecasting. Only a few studies reach pilot-level evidence, lacking sustained operational deployment.
Plain Language Accessible to non-experts
Imagine a complex building system with many devices and sensors. Each device has its own name and data, but this information is scattered like a messy bookshelf. A large language model acts like a smart librarian, quickly finding the information you need, organizing it, and making it easier to understand and use. This is the role of large language models in HVAC systems.
ELI14 Explained like you're 14
Imagine you're playing a complex game with lots of characters and tasks. Each character has its own name and skills, but this information is scattered everywhere. A large language model is like a smart assistant that helps you quickly find the information you need, organize it, and make it easier to complete tasks. That's what large language models do in HVAC systems.
Glossary
Large Language Model (LLM)
An AI model capable of understanding and generating natural language.
Used to improve information access and workflows in HVAC systems.
Building Energy Modeling (BEM)
A technique for simulating and analyzing building energy consumption.
32 of the reviewed papers focus on this area.
Model Predictive Control (MPC)
An advanced method for controlling systems by optimizing control based on future predictions.
Remains widely adopted in traditional methods.
Semantic Interface
An interface for connecting and interpreting different information sources.
The primary role of LLMs in HVAC systems.
Point-Name Normalization
Standardizing the names of different devices for easier management and analysis.
One potential application of LLMs.
Open Questions Unanswered questions from this research
- 1 Lack of sustained operational deployment studies.
- 2 Insufficient data for subfield-level conclusions in load forecasting.
Applications
Immediate Applications
Point-Name Normalization
Using LLMs to standardize the names of different devices for easier management and analysis.
Long-term Vision
Autonomous HVAC Control
Developing LLMs capable of autonomously controlling HVAC systems, requiring solutions for safety and responsibility issues.
Abstract
Building automation systems generate rich sensor data yet remain insight-poor because heterogeneous point naming, missing metadata, and fragmented documentation obstruct their operational use. This systematic review analyses and codes 66 peer-reviewed studies on large language models (LLMs) for HVAC operations published between 2023 and March 2026. Each study is classified across five application families and three LLM method families and assessed for evidence realism, deployment readiness, and the responsibility boundary between the LLM and physical HVAC decisions. The corpus is concentrated in building energy modelling (BEM, 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. Nevertheless, 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. Conventional machine learning (ML), model predictive control (MPC), reinforcement learning (RL) and ontology-based tools remain more adopted for high-frequency control, short-horizon numerical forecasting, and well-posed ontology mapping, while autonomous agentic operation and unvalidated occupant proxies remain research-stage. Current evidence therefore supports LLMs primarily as semantic and workflow layers rather than autonomous HVAC controllers. Future work should prioritise field-validated benchmarks, orchestration evaluation under operational constraints, and LLM-MPC/RL architectures with bounded latency and verifiable safety properties.