More than four million people were without power Friday morning (Sept 27) after the enormous ring of wind and rain known as Hurricane Helene made landfall in Florida and moved north.
It is the latest storm to show utility companies’ increasing vulnerability to extreme weather events that are becoming more common and more intense under climate change.
“There are a lot of different signs of climate-related weather risks to infrastructure,” said Catie Hausman, a professor of public policy at the University of Michigan. Those risks include hurricanes and flooding, wildfires, heat waves and increased tornado risks or cold snaps in regions less used to them.
Extreme weather has increasingly strained the grid, and it is the No. 1 cause of major power outages in the United States. In some areas of the country, the risk of hurricane-induced power outages could become 50% higher in the coming decades as such storms get stronger.
Wind and rain are the dominant factors that can strain power grids, said Andrea Staid, a researcher in energy systems and climate analysis at the nonprofit Electric Power Research Institute. The institute’s models show that as more hurricanes affect the Gulf and Atlantic coasts, more power outages will occur if the grid does not change.
“Hurricanes are not just a coastal problem,” Staid said. “Our models show that these storms can travel pretty far inland with strong winds, so we need to make sure people and communities have the information and resources needed to prepare.”
Now, in an effort to better predict what storms are coming and how to transition out of a fossil-dependent grid, utilities are looking to a new generation of technologies driven by artificial intelligence.
Hausman said that while she did not know whether AI was the right way to harden the grid, modern data and computing were needed to understand the problems.
“Whether AI is giving us something new or a black box of mush is going to depend on the company and the tools they’re using,” Hausman said.
What is clear? Utilities need to spend a lot of money to update their power systems to deal with storms both present and future.
“We’re still thinking of the grid as we have for the past 100 years, and it’s increasingly obvious that needs to change,” said Mark Dyson, managing director of the carbon-free electricity program at the global research firm Rocky Mountain Institute.
“Extreme weather, aging infrastructure and new technology are coming together in a way that creates an opportunity to use better technology, including AI-driven software, to help us keep the lights on and keep the grid affordable,” Dyson said.
One of the newest players in the nascent industry is Rhizome, a company founded in 2022 that uses AI-driven technology to help electric utilities identify and plan for vulnerabilities that could cause power failures.
“Rhizome’s AI platform approach is to fundamentally understand the relationships between hurricane conditions and grid impacts,” said Mishal Thadani, Rhizome’s co-founder. The company’s data points from thousands of hurricane-related asset failures and what caused them can show utilities their long-term risks from hurricanes, helping them figure out where to harden poles, move power lines underground or cut vegetation.
“Ultimately, we’re able to project how many future potential hurricane-related outages will be reduced per dollar of utility investment,” Thadani said.
Figuring out how to harden the grid and expand the country’s transmission network could lower costs to consumers, bring more renewables online and reduce power outages, Hausman said.
Most utilities are already using some machine learning or artificial intelligence technology, said Booga Gilbertson, a former utility executive and an investor in Rhizome. Right now, the flashy types of AI programming are language models like ChatGPT from Microsoft or Gemini from Google. Machine learning has been combing through a mind-boggling amount of data for more than a decade, Gilbertson said.
“It seems like once a month there’s a new entrant into this space,” she said. “Products like this are relatively new to the market, and the advent of AI and computing power have made them more available. It’s a tool utilities can now put into their tool chest.”
Artificial intelligence is also a known energy hog, and could ramp up the nation’s electricity demand by as much as 20% by the end of the decade. But the United States, like other global powers, wants to lead the world in AI. In September, the White House convened a group of utilities, artificial intelligence companies and data center operators to strategise for the future.
As the country’s transmission system struggles beneath the weight of growing electrification and a changing generation mix, the growing set of AI-driven products could potentially pivot AI’s bad climate reputation.
“I suspect that the amount of energy we would spend more than pays for itself in terms of maintaining an affordable and reliable grid,” Dyson said. “But the data to support that suspicion doesn’t exist yet. The industry is so immature we don’t have an outlook yet for the types of sectors that would be most advantaged from AI.”
As new companies enter the space, AI climate modeling is still a bit of a Wild West.
Andre Coleman, chief scientist at Pacific Northwest National Laboratory, leads a team that developed a program called RADR-Fire to build risk prediction models for wildfires. He said such models are also available for other hazardous events like tropical cyclones, extreme rain and extreme heat, extending as far into the future as 2100.
“Particularly when we’re talking about assessing risk around disaster events, we have to be really, really sure about what those machine-learning models are doing, especially for unique or outlier events,” Coleman said.
Coleman wants cooperation to make sure the methods and models are the same, then make all that data accessible, especially for smaller utilities with fewer resources, like rural cooperatives.
“You end up with this huge mosaic of people who are doing this kind of risk modeling using all kinds of different data sets; some are appropriate, others aren’t,” he said. “Those are the things that concern me a lot.” – The New York Times
