How breakthrough statistical models are helping us reduce our carbon footprint, one building at a time

When buildings on the Stanford campus sat empty during the COVID pandemic, MS&E Professor Peter Glynn and colleagues from the Stanford Doerr School of Sustainability saw an opportunity.

Today's buildings worldwide are becoming increasingly "smart" through Building Automation Systems (BAS) that proactively seek to optimize energy efficiency, water conservation, air quality and other factors that reduce the environmental impact of commercial infrastructure we’ll be using for the next several decades. Using sensors, real-time building data, and algorithms, these systems solve mathematically complex decision problems, such as how to adjust indoor lighting based on the natural light coming into a space or adapt indoor temperatures to external weather conditions to save energy. Since the building sector accounts for 40% of the world’s electric energy consumption, even a 5% reduction in resource usage worldwide can have a massive impact on carbon emissions alone.

One energy-intensive and expensive byproduct of climate change that is vexing facilities managers is more prevalent, extreme weather events such as heat waves. Climate scientists expect the frequency and intensity of these events to continue to increase, so it's essential that we find ways to deal with them that are energy efficient and not cost-prohibitive. The challenge, however, is that current BAS decision models are not built to handle the fluid forecasts for these events, and it's often difficult to know how long they'll last.

Universities, hospitals, military bases, corporate campuses and government entities that manage clusters of buildings often have a central energy facility. Among its features is a cooling capability that stores and then pumps chilled water through buildings to keep temperatures down on hot days. If too much of that water gets used during the first day or two of a heat wave, there is nothing left as the heat wave unexpectedly continues. When patients, critical scientific experiments and defense systems depend on temperate buildings, a non-functioning air conditioning system leads to disaster. The solution today is to mitigate that risk by beefing up the central energy facility, an investment of tens of millions of dollars to accommodate, for example, the rare Day 4 of a heatwave. But what if it's possible, instead, to use more advanced decisioning models to get through them with the existing infrastructure by simply being flexible and making use of better information?

Prior to the pandemic, Glynn had been in conversations with two faculty in the since-formed Stanford Doerr School of Sustainability: Jacques de Chalendar, an adjunct professor in the Energy Science and Engineering Department, and Sally Benson, a professor in the Department of Energy Resources Engineering. They were exploring how to bring together mathematical modeling involving new statistical techniques, technical engineering, and software engineering to answer this question and better optimize overall energy efficiency. Changing the temperature in large buildings is like trying to turn a ship—you need a lot of lead time. If it were possible to optimize heating/cooling decisions before extreme conditions hit, facilities managers could dramatically reduce risk and energy usage. If they could also dynamically adjust decisions as the forecast changes by the day, even better. Current models didn't do this.

De Chalendar had been cultivating a relationship with the university's Lands, Buildings and Real Estate (LBRE) department. While the buildings sat unused in 2020, the team obtained access to a few buildings to run experiments that are difficult to do when occupants are present.

Explains Glynn, "The core research issue was whether it would be possible to incorporate weather forecast data hyper-localized to a given building into algorithms and create dynamic decision models to take preemptive energy-reduction action."

For example, they thought it could be possible to make that chilled water last the duration of an extended heat wave, even when the heatwave unexpectedly persists. They wanted to experiment with raising the temperature of a building early in the heat wave while keeping it comfortable. Where there were storage rooms adjacent to occupied floorspace, they wondered if it might be possible to cool these at night or in the morning to limit the energy required to keep adjacent areas cool. They also foresaw the ability to give people in certain buildings advance notice to work from home during the heat wave.

Empty buildings allowed them to run controlled experiments with simple statistical models for predicting a building's thermal inertia over a given period of time (i.e., its natural resistance to changes in temperature). For example, if you cooled the building by five degrees in the morning, how long would it take for the building to heat up and to what degree? How might you heat or cool the building more aggressively when energy demand is relatively low and then ride that wave to moderate temperatures later in the day? If you can draw more energy when the grid is green and less at peak times when it’s heavily supplemented by carbon sources, you reduce the carbon footprint and costs. This process is called temporal shifting.

At the center of all of this is an underlying question with broad applicability: Can we incorporate changing forecasts into mathematical models in the presence of uncertainty over time in order to make the models more accurate? Would it be possible to use a Markov Decision Process—a sequential decision making problem in which, for example, a decision on Day 1 affects Day 2 outcomes and so on? If so, there are wide-ranging implications. You could take advantage of localized weather information to tell a particular farmer when to plant, water, and harvest—based on the location of that farm, not the town 15 miles away. There is applicability in forest management, commodities trading, airport logistics…any scenario in which you need to choose a strategy based on a forecast that gets updated throughout your decision process. These new techniques have the potential to add a new level of precision to dynamic decision making.

"Uncertainty is everywhere in climate change, and as a society, we need to be working on this. MS&E as a discipline is an excellent complement to the earth sciences and can offer a lot to advance academic work in these areas," says Glynn.