Here’s how AI could transform the US power grid
Much of the US energy grid is more than 25 years old, and electricity demand is expected to grow nearly 20% by 2050.
By Hayden Field and Grace Donnelly
Whether you’re reading this at home, your office, or a coffee shop, you’re likely not far from an unassuming gray box that charts your every move—at least, as far as energy usage goes.
These familiar electric meters send energy-usage information to the utilities operating the power grid. At the moment, the devices—and the utilities themselves—can chart how much energy is used and then bill you for it, but despite all this data, many utilities may be unable to tell if a local transformer is close to overloading.
Insight into real-time demand is becoming even more necessary as intermittent renewable sources of electricity, like wind and solar, make up a larger portion of electricity supply—and as electricity is projected to become a bigger share of the overall power mix. Electricity demand is expected to grow nearly 20% by 2050, in large part due to the adoption of clean tech like electric vehicles. Increasingly frequent extreme weather events are also ramping up the pressure on the grid.
Some startups, tech giants, and researchers believe it’s time for AI to play a bigger role.
“It’s a confluence of things: It’s Mother Nature, it’s the market, and it’s customers being like, ‘I’m sick and tired of losing power all the time,’” Josh Brumberger, CEO of Utilidata, a grid-edge technology company, told us. He added, “If you take all the storms and the frequency of outages, and all the complexity of the solar and the EV ramp-up that’s about to happen—and then you layer on the incredible amount of federal spend that’s about to be levied onto the system—I just think the time is now right for these step-change technologies and opportunities.”
Over 70% of the US energy grid is more than 25 years old, and the Biden administration plans to invest $20 billion to modernize grid infrastructure—the largest US investment of its kind to date. AI could eventually help make that modernization possible. Ideally, the technology could allow utilities to make more accurate decisions about how much power to generate, improve efficiency, and avoid outages.
“Our grid system is getting more and more complicated,” Kibaek Kim, a computational mathematician at Argonne National Lab, told us. He added, “We need a lot more computing power to solve and operate complex systems. And with these challenges, AI is providing lots of tools and approaches to advance—and make [computation] faster.”
Traditionally, utilities (think: PG&E or ConEd) have generated electricity from a mix of sources, which is delivered to customers via transmission and distribution systems, and they have to communicate anticipated demand to power generators ahead of time to ensure enough electricity is produced. Since the inception of the grid, that supply has flowed in only one direction: Electricity is created by the utilities’ assets and used up by customers.
The increasing adoption of on-site solar, bidirectional EV charging, and batteries for energy storage and backup power complicate that equation, since these distributed energy resources can be power-generation assets on the consumer side.
Add that to growing electricity demand overall, and suddenly the industry needs more granular, accurate predictions for power generators and faster decision-making tools for power distribution—not to mention better monitoring for potential outages due to equipment failures or extreme weather events.
These are the sorts of tasks AI could be equipped to handle.
Nvidia, one of the world’s largest chipmakers, is helping Utilidata design a smart grid chip powered by its Jetson Edge GPU and implemented via a meter adapter.
Right now, utilities can essentially use meters to turn power on/off, accurately bill users, and receive data in ~15-minute intervals. Nvidia’s chips could allow meters to collect and process data on power needs in real time, to allow utilities to direct resources more efficiently.
“[Utilities have] years of data collection. The thing is it’s in smaller intervals—it’s not as real-time as I think what’s going to be required. A lot can change in 15 minutes, with cloud overhead and other things,” Marc Spieler, Nvidia’s head of energy, told us.
Utilidata is already deploying smart meters in the field to gather training data for AI models, focusing mostly on North America. Spieler anticipates nearly 1,000 units being used by the end of December; by the end of 2023, tens of thousands; and by 2024, large-scale rollouts. To allow utility companies to effectively utilize the smart meters, Utilidata is building a smart meter platform on top of Nvidia’s software.
For now, many AI tools for grid-modernization, like Utilidata’s software, focus on reliability and outage predictions, rather than giving algorithms the authority to determine power distribution. Kim and other researchers at ANL are working on systems that could allow for such machine-learning-based grid control, but they’re relatively far off.
Spieler thinks the future of grids could be akin to an autonomous car. “It’s got a powerful chip in there that allows it to make real-time decisions…and then it uses that network to update the model, share what it’s learned, and ingest what other people have learned”—but that likely won’t happen for a while.
Ultimately, though, developing the tech itself may not be as difficult as getting the green light from utilities and regulators.
“The utility industry is a very cautious industry,” Spieler said. “It’s kind of like IT—when everything works, everybody’s happy. When it doesn’t work, it’s a disaster.”
As a result, convincing utilities to embrace—and invest heavily in—an AI overhaul isn’t easy. (Emerging Tech Brew reached out to utilities like PG&E and Evergy to talk about AI projects, and they either declined to talk or did not respond.)
“You don’t really have to re-create the wheel, from a distributed AI technology perspective,” Brumberger told us. He added, “This is incredibly complex stuff, but it’s not nuclear fusion. This is really around a system embracing existing technologies that have already revolutionized countless industries and bringing it into this industry at the dawn of what is being called the electrification of everything.”