RICHLAND, Wash.-Buildings researchers at the Department of Energy's Pacific Northwest National Laboratory have released a new AI-driven, autonomous bot that could help speed up the energy modeling process for commercial building construction.
Long before construction begins, commercial building teams evaluate a building's expected energy consumption to inform design decisions, estimate operating costs and comply with relevant state and local energy codes.
For this evaluation, building teams use building energy modeling software, which simulates how a proposed building will use energy under different design options. A range of modeling tools exist for the building industry, including software developed or supported by DOE. But developing and interpreting these models requires specialized expertise, including training in energy modeling and building systems.
To simplify the process and save resources for designers, PNNL researchers created Building Energy Model AI, an autonomous bot that can quickly produce a complete building energy model for users in every stage of the modeling process. The researchers describe BEM-AI in a recent paper in Energy and Buildings. In tests, the bot successfully modeled reductions in energy use for typical commercial buildings in Florida.
"With BEM-AI, we are broadening the potential access to these complicated tools that require a lot of subject matter expertise," said Weili Xu, a building modeling expert and lead author on the new paper. "Access to a well-developed autonomous energy modeling bot helps all stakeholders in the life cycle of building construction, from designers, architects, engineers or code officials."
BEM-AI is open source and available to use now, and the research team wants to encourage users to experiment with the platform and expand on it.
Energy modeling 101
Let's say you want to bake a loaf of bread, but have zero baking knowledge, ingredients or appliances. So you first hire an engineer to build an oven and an electrician to hook up the power. But now you need a farmer to grow the wheat for flour and a meteorologist to forecast the ideal growing conditions. Finally, you need a chemist to understand the reactions between the ingredients.
Baking your loaf would take months of work. Building energy modeling similarly can take months.
A building's energy consumption depends on a slew of factors: the number and orientation of windows, the kind of heating and air conditioning systems, the materials used in construction, the local climate, occupancy schedules, what components of the building need to operate during different times of the day or night and more.
"The workflow for creating a building energy model is pretty straightforward, although the details can get complicated very quickly. You're working with all these different experts on calculations, on data analysis," Xu said. He started to think about how AI, with its powerful ability to synthesize information, could help simplify this process
Xu identified several pain points in energy modeling that an autonomous bot could help solve. One example is that building and energy data comes in many different formats (like PDFs or spreadsheets), and an autonomous bot could potentially translate those into a format that the existing modeling software can process.
Another pain point Xu identified is constructing the energy model itself, or translating design intent into a model that can be "read" by an existing program like DOE's OpenStudio or EnergyPlus. This pain point became the focus of BEM-AI.
Testing BEM-AI
BEM-AI uses "agentic AI," an autonomous system that can work independently with any number of separate "agents." The bot essentially breaks down a larger task, such as "estimate the energy use for a building with these parameters" and independently assigns tasks to multiple agents that themselves can complete those tasks independently.
To test BEM-AI, the researchers gave it several prompts. They started with a three-story building with 53,000 square feet of conditioned space built in a hot and humid weather condition, submitting the prompt below:
I want to evaluate the energy savings achieved by reducing the window-to-wall ratio by 10% for a medium office building designed in accordance with energy code standards in Tampa, Florida.
One of BEM-AI's agents, the "planner," analyzes the request and breaks it down into subtasks. The "orchestrator" then assigns those tasks to other agents that each have specific bodies of knowledge, such as about walls, roofs, lighting, or the energy modeling software itself. These specialized agents gather all the necessary data-such as seasonal temperatures in Tampa, building materials or local energy codes-and run an energy simulation.
Once those subtasks have been completed, the orchestrator agent forwards everything to a "summary" agent, which provides the requested energy savings result.
BEM-AI successfully calculated energy savings for several prompts for buildings in Florida.
For now, the bot can supply information about buildings in Florida. The researchers hope that the energy modeling community can help expand its capabilities. The team's software is open source, which means anyone can build upon it-and that's what the researchers hope happens.
"We need a lot more data from architects, construction consultants and other experts to expand the model," Xu said. "No two commercial buildings are the same. So we are hoping the community can help us by providing more examples."
This work is funded by the Building Technologies Office within DOE's Office of Critical Minerals and Energy Innovation.
