With roughly one-quarter of global greenhouse gas emissions produced by burning fossil fuels for power and heat, there’s an urgent need to decarbonize the world’s electricity supply. This requires integrating renewable energy sources into power grids, explains Priya Donti, an assistant professor of electrical engineering and computer science at MIT. Her mission: design machine learning algorithms to streamline the process.

“On a power grid, you have to make sure that for any given node, the flow in is equal to the flow out,” she says. This was simpler in the past, when the electricity supply was less affected by weather, and, due to the uniformity of power supplies, there was more time to restore balance if anything went wrong. But as we rely on an increasing diversity of power sources, including renewable energy, maintaining this balance gets trickier, and the time available to correct imbalances before things go awry shrinks. “The underlying physics is really nasty,” she says.

To address these challenges, Donti has developed a demand forecasting model in use by the U.K.’s national grid operator, and an algorithm that estimates the grid’s voltage being piloted in some parts of the U.S. Her research group has also created datasets and benchmarks to foster innovation in the ecosystem, plus algorithms to optimize power flow in the grid.

Donti is a co-founder and chair of Climate Change AI, a climate tech nonprofit that runs a summer school with 16,000 registered participants, a workshop series, and an innovation grants program. She is also a lead author on a seminal paper recognized for defining the field. “Empowering people to be equipped to use AI where it’s helpful to accelerate their work, is hugely important,” she says. 

While some of her work—like using planning algorithms to guide how renewable energy is added to power grids—could become more widely deployed in the next few years, Donti emphasizes that effort should be put into accelerating the process of adoption, not just designing new algorithms.  “These are not theoretical problems in terms of how you balance power grids with renewables,” she says. The power sector “is facing them literally right now—and yesterday.”