Data2BEM’ framework cuts modeling time by 90%, accelerating the path to net-zero buildings.
by Jie Lu
The building sector is a giant in global energy consumption, accounting for approximately one-third of the world’s energy use. To reach net-zero targets, improving the performance of existing buildings through retrofits is crucial. However, the first step in this process—creating a detailed, physics-based energy model to simulate potential savings—remains a significant bottleneck.
Traditionally, building energy modeling (BEM) has been a technically complex and labor-intensive task. It requires experts to manually interpret architectural drawings, configure thousands of parameters, and meticulously calibrate models against real-world data. This manual process limits the scalability of retrofits needed for large building portfolios.
Now, a collaborative research team from the Energy Efficient Cities Initiative at the University of Cambridge and Zhejiang University has developed a solution that fundamentally changes this paradigm.
Published in the journal iScience, the team presents Data2BEM, a novel framework that uses a multi-agent system powered by Large Language Models (LLMs) to automate the entire modeling workflow—from reading raw design documents to evaluating retrofit scenarios.
"Developing detailed energy models remains a labor-intensive and technically complex task... current practices still rely heavily on expert judgment and manual coordination, which limits the scalability and automation of building energy modeling for energy retrofits”Jie Lu, Lead Author, Dept of Engineering, University of Cambridge & Zhejiang University
The Power of Multi-Agent Collaboration
Unlike previous automation tools that function as isolated, single-purpose scripts, Data2BEM employs a team of four specialized AI agents: an Information Retriever, a Programmer, a Result Analyzer, and a Reviewer. These agents work together much like a human engineering team. The Information Retriever extracts geometric data and material properties from CAD drawings and design specifications. The Programmer translates this data into executable simulation code. Crucially, the Reviewer agent acts as a quality control mechanism, verifying outputs to ensure cross-stage consistency and detect errors.
Dramatic Reductions in Modeling Time
To validate the framework, the researchers applied Data2BEM to a real-world educational building on the University of Cambridge campus. The results were transformative.
While an experienced human modeler required approximately 8 hours to construct and calibrate the building model, and a junior modeler needed 32 hours, the AI-driven system completed the entire end-to-end process in just 48 minutes.
"Relative to professional practice, the system reduced total modeling time by over 90% with minimal human input... delivering end-to-end automation while accurately reflecting measured performance.”Research findings
Accurate and Actionable Insights
Speed did not come at the cost of accuracy. The automatically generated model met the rigorous calibration standards of ASHRAE Guideline 14, matching the building's actual historical heating consumption with high precision.
Furthermore, the system successfully conducted a retrofit analysis. It autonomously simulated replacing the building’s gas boilers with air-source heat pumps (ASHP). The AI agents predicted that this electrification strategy would cut annual energy costs by roughly 44% and reduce carbon tax obligations by nearly 5-fold.
These results indicate that LLM-driven multi-agent methods can lower the expertise barrier for practitioners, making rigorous energy analysis accessible and scalable. This innovation offers a promising pathway to accelerate the decarbonization of the built environment globally.
Publication:Jie Lu, Zeyu Zheng, Max Langtry, Monty Jackson, Yang Zhao, Chenxin Feng, Ruqian Zhang, Chaobo Zhang, Jian Zhang, and Ruchi Choudhary. ‘Automated building energy modeling for energy retrofits using a large language model-based multi-agent framework.’ iScience (2025). DOI: 10.1016/j.isci.2025.113867
Image credit, used with permission: Jie Lu, Lead Author
|
Jie Lu conducted her joint PhD training at the University of Cambridge and The Alan Turing Institute, focusing on data-driven strategies for energy-efficient building renovations. Her work is bridging the gap between energy system simulation, computer science, and data science. Jie’s primary research interests include building energy digital twins, generative simulation modeling, and the application of Large Language Models to advance sustainable urban energy solutions. |
