When it comes to the energy transition, one thing is certain: the field is marked by profound uncertainty.
Uncertain technological constraints. Uncertain societal priorities. Uncertain climate conditions.
Yet despite these unknowns, energy planners must continue making decisions and moving systems forward. Fortunately, as the three speakers at the Irving Institute's May 1 Faculty Seminar demonstrated, the academic community is developing new tools to help navigate that complexity.
How do you plan long-term energy transformation projects, which can take decades to complete, when societal goals are changing every few years?
That was the timely question posed by Erin Mayfield, Hodgson Family Assistant Professor of Engineering at the Thayer School and Principal Investigator at the Sustainable Transitions Lab. Her Faculty Seminar presentation was entitled "Designing Net-Zero Emissions Energy Systems to Address Dynamic Societal Objectives."
If anything, "dynamic" may be an understatement. "We're very uncertain regarding what's going to happen three days from now, let alone 30 years," Mayfield said. "Sometimes it's unclear what the objectives even are."
You'd know about this first-hand if you've considered buying an electric vehicle recently. Because the $7,500 federal tax credit for buyers of new electric vehicles, introduced in 2022 as part of then-President Biden's Inflation Reduction Act, is no longer available. It was eliminated by President Trump's One Big Beautiful Bill, approved by Congress last year.
Setbacks like this can be frustrating, as Mayfield conceded when describing a collaborative study by the Sustainable Transitions Lab. The study found that had the Inflation Reduction Act been fully implemented, one expected result would have been a 43% to 48% reduction in U.S. emissions by the year 2030.
Another Lab projection shows that achieving net-zero emissions targets by mid-century could prevent as many as 400,000 premature deaths from the reduction in air pollution. But here, too, societal goals have shifted. Consider the Paris Agreement, where long-term international temperature goals were set. The U.S. joined the Agreement, then withdrew from it, then joined again, and then, as of this past January, withdrew a second time.
For this and other reasons, designing net-zero energy systems is "inherently multidisciplinary, inherently multi-scale, and inherently multi-objective," Mayfield said.
So how are energy planners supposed to plan? With humility, Mayfield said—and a sense of humor.
Another area of uncertainty lurks in the computer models used to understand factors including global warming, carbon budgets and climate-change mitigation. Some of these models are huge, comprising hundreds of thousands to millions of lines of code. And most of these complex models show that for a given level of emissions, a certain amount of global warming can be expected. That got the attention of Ashwin Seshadri, another speaker at the May Faculty Seminar.
Seshadri, an Associate Professor at the Indian Institute of Science in Bengaluru, titled his presentation "From Global Carbon Budgets to Regional Energy Transitions." Carbon budgets are used to help set fair and effective targets for reducing greenhouse-gas emissions.
"One doesn't need a straight-line relationship to construe and implement a carbon budget," Seshadri said. "What one really needs is the broader idea of path independence."
To offer that, Seshadri and his colleagues have built computer models that are much smaller—just a few hundred lines of code. Unlike the bigger models, these smaller and simpler models allow researchers to ask "Why" questions about complex issues. For example, one topic Seshadri has explored is when path independence works, and when it fails.
In general, path independence works because of what Seshadri calls a "separation of time scales." Some climate factors respond quickly, while others shift far more gradually. For example, over the past century, cumulative carbon emissions have doubled every few decades. That's fast. But other variables, such as ocean temperatures or airborne fraction of carbon dioxide, have changed much more slowly. As a result, Seshadri said, there's a one-to-one relationship between emissions and warming and it's what allows carbon budgets to function as a planning tool. However, if global emissions are brought down substantially the separation of timescales would break down–and with it, path independence. At that point, additional factors would need to be accounted for, and the simple link between a carbon budget and a temperature target would no longer hold.
The day's most technical presentation came from Junbo Zhao, the Todd M. Cook and Elizabeth Donohoe Cook Associate Professor of Engineering at Dartmouth's Thayer School of Engineering. Like the other speakers, Zhao focused on a central challenge of the energy transition: understanding and managing uncertainty, particularly as it affects electric-grid security and reliability.
Joining the seminar by live video, Zhao delivered a talk titled, "Physics-Informed Data Analytics for Power System Risk Assessment and Control."
Today's electric grid, he explained, is becoming increasingly dynamic and increasingly exposed to disruptive forces, including extreme weather events such as wildfires and storms. These conditions introduce substantial uncertainty into grid risk assessment. The question, Zhao said, is how planners and operators can evaluate those risks accurately enough to make reliable decisions.
Machine learning has emerged as one promising tool for addressing uncertainty in grid operations. But Zhao emphasized that AI-based solutions cannot function as black boxes. To be useful in real-world power systems, they must respect physical laws and provide predictions with quantified confidence to support decision-making.
The challenge is even greater because many of the uncertainties affecting the grid are difficult to represent mathematically. Zhao described these as "huge uncertainties," noting that "it's really, really challenging for existing AI to deal with these."
His proposed remedy is to embed physical knowledge directly into the machine-learning training and decision-making process. Zhao's research team has developed AI-powered surrogate models that preserve essential physical constraints while reducing the computational complexity of large-scale grid analysis. In effect, these models make highly complex risk-assessment problems more tractable without sacrificing the underlying physics that govern system behavior.
Zhao's team has also translated these methods into real-world applications. One example involves large-scale electric-grid risk assessment of the ISO New England system under uncertain renewable generation and fluctuating demand. The computational gains are substantial: Analyses that can take conventional models hours to complete can be performed by Zhao's approach in less than several minutes.
Learn more: Watch a video replay of the Irving Institute's May 1 Faculty Seminar.
Peter Krass is a contributing writer and editor to the Irving Institute.