Seed grant spotlight: AI-driven energy benchmarking

The net-zero conversation often revolves around conspicuous sources of carbon emissions like fuel-powered planes and automobiles, but a new project at Carnegie Mellon draws attention to architecture’s sizable carbon footprint. 

“Buildings are responsible for close to 40 percent of global carbon emissions,” said researcher Tian Li, who earned his Ph.D. at CMU’s School of Architecture under the guidance of Azadeh Sawyer, an assistant professor of building technology. 

With support from a Scott Institute seed grant made possible by the Rankin family and Trane Technologies, both Grand Challenge Partners, Sawyer, the project’s principal investigator, and her former Ph.D. student Tian Li are developing an optimization model for AI-driven benchmarking that accurately predicts a building’s energy consumption and greenhouse gas emissions at daily, weekly, and yearly intervals.

The benchmarking method boils down to comparing a building’s energy consumption and carbon emissions patterns with a peer group of similar buildings. If one building uses a lot more energy compared with a peer group, then end users such as engineers, utilities, and building owners can respond appropriately. 

“Our goal is to collect as much real-world data as possible to establish better baselines for how much energy a typical building should use,” said Sawyer. 

The social relevance of sustainable architecture cannot be ignored. 

“We know there are energy equity problems right now,” Sawyer continued. “Lower-income communities are paying up to 20 percent more for energy than higher-income communities, and the reason is because their homes and systems are less efficient.” 

Many factors can contribute to an increase in emissions — including inefficient buildings and systems. For example, a lot of older buildings tend to have poor insulation, which creates greater heating and cooling demands. Outdated HVAC and lighting systems and inadequate sealing around windows and doors exacerbate inefficiency, creating higher costs and environmental strain. 

Sawyer and Li hope their AI model will contribute to making benchmarking more accessible as a sustainability resource along behavioral, technical, social, and policy dimensions. Currently, less than 50 cities in the U.S. deploy energy benchmarking to reduce emissions, and just a fraction of them make annual benchmarking data available to the public. By creating an AI model capable of analyzing daily, weekly, and monthly values, this project has the potential to increase the level of detail that benchmarking offers while furthering AI’s contribution to sustainability goals.