In 2020, residential and commercial buildings in the U.S. accounted for 40 percent of all energy consumption in the country – and with climate change rapidly accelerating, enormous cuts will be necessary in the building sector. With tens of millions of buildings in the U.S., though, even understanding the dynamics of buildings’ electricity use is a herculean task. Now, researchers at Oak Ridge National Laboratory are leveraging Argonne National Laboratory supercomputing to create digital twins for all of the 129 million buildings in the U.S.
The modeling suite is called AutoBEM, for Automatic Building Energy Modeling. AutoBEM, which has been in development since 2015, amalgamates energy use data and publicly available building data (such as satellite imagery and street view data) to identify key characteristics of each building (its materials, floors, roof type, etc.) and predict, from those variables, a building’s energy characteristics.
“To build AutoBEM, we looked at many of the available data sources and created partnerships with the people that are in the business of generating this data, like satellite, aerial and street view imagery, LiDAR data and other types of databases,” said Joshua New, the ORNL computer scientist leading the AutoBEM team. “We had to use or extend algorithms to turn the data into descriptions, creating a 3D footprint of the building[.]”
Automating these predictions allows urban planners and energy researchers to quickly identify areas with high building energy footprints and the technologies that could reduce those footprints. “Utilities spend billions of dollars a year on energy efficiency and demand-response programs in the U.S.,” New said. “This spending is based on a signature analysis of electrical profiles. Before this program, no one had the capability to perform that analysis with detailed, building-specific energy modeling at this scale. Individual utilities now have the capability to perform modeling to show the potential of reducing demand and greenhouse gas emissions.”
Recently, the AutoBEM team at ORNL set out to create these digital twins for 178,337 buildings in Chattanooga, Tennessee. Using the local electric utility’s information on each building’s energy use – granular to 15-minute intervals – and the publicly available building data, they generated the digital twins and projected the impacts of a series of energy efficiency measures (such as improved insulation and lighting changes). Those results were published in the December 2020 issue of Energies.
Running the AutoBEM suite at that scale, of course, is computationally intensive. For these initial large runs, the ORNL team first used their in-house Titan supercomputer – a 17.6 Linpack petaflops system that was decommissioned in August 2019 – then turned to their colleagues at Argonne National Laboratory, who provided access to their Theta supercomputer, a 6.9 Linpack petaflops system with over 4,000 nodes.
“Theta is one of the most powerful supercomputers in the U.S. in terms of CPUs,” New said. “When we got to Theta, AutoBEM scaled so beautifully. We regularly use over 80 percent of Theta. … We’ve been able to scale up to running annual simulations of over a million buildings in one hour on Theta. That really unlocks a lot of potential that you wouldn’t see otherwise.”
“If you’re trying to understand energy requirements over a lot of different scenarios in a city and wanted to look at all the knobs you could turn, that’s thousands of simulations – not one big one,” added Katherine Riley, the Argonne Leadership Computing Facility’s director of science, in an interview with Argonne’s Christina Nunez. “For a project like that, you need a system that’s capable of managing a very dynamic workflow. That’s what ALCF can support.
For New, these initial simulations are just a first step toward a brighter – and cleaner – future.
“What we do in buildings will have a long-lasting impact,” he said. “Creating a more sustainable and resilient building stock will have an impact that I might not see in my lifetime, but my grandchildren’s grandchildren will be thankful we got that right.”