Preparing communities for power outage risks

In February 2021, winter storms plunged millions of Texans into darkness when the power grid failed.  More than 240 people died as a result of the storms, including some because they lost electricity to power life-sustaining medical equipment or regular heating sources.

What if officials had known the risk of extreme weather events beforehand? Could they have used information about grid vulnerabilities to strengthen electrical infrastructure and prepare residents for an outage?

These questions motivated Tao Sun, a postdoctoral researcher in civil and environmental engineering at Stanford, to develop an AI-based tool to help communities predict outage risk. He used physics-based and data-driven methods to build a unified extreme weather dataset and collaborated with utility companies to analyze decades of power outage data.

From power plants to climate science

Sun has long had an interest in electrical systems. Growing up in China, he saw his parents work on the construction of the Jiaxing Power Plant, one of Asia’s largest coal-fired power plants. He earned bachelor’s and master’s degrees in electrical engineering from Shanghai Jiao Tong University before coming to Stanford for his doctorate in civil and environmental engineering, which he completed in 2025.

“I came to this department with my power systems background, and that grew my interest in energy and the impacts of climate change and extreme weather,” Sun said. “I wanted to draw connections between climate, extreme weather, and power systems.”

Weather is a leading cause of power outages across the U.S., and extreme weather events are increasing in frequency. By combining his knowledge of power systems with climate research, Sun wants to help communities better understand and prepare for these risks.

Decades of data

By collaborating with public utility commissions, government entities, and individual utilities, Sun and others in civil and environmental engineering professor Ram Rajagopal’s Sustainable Systems Lab built a database that spans from 1988 to 2024, while previous datasets only reached back to 2013.

Four decades of data provides a more nuanced picture of outage risks nationwide.

“The trend is significant – but it’s not as stark if you start from 2013,” Sun said. “One year, you could see an average of 10 hours of outages, but the next year it could be just three hours. The overall trend is increasing, but it shows substantial year-to-year variability.”

From data centers to home generators

Sun’s research is focused on power outage risk – not specifically prevention – and the applications are wide-ranging.

For example, investors and developers could use the model to evaluate locations for artificial intelligence data centers. Utility districts could explore how to strengthen infrastructure based on particular types of weather threats. With climate change, damaging ice storms may decrease in some regions, while windstorms and wildfires increase in others – each requiring different preparation strategies.

Residents in outage-prone areas can use the data to make informed decisions about backup generators, solar battery systems, or portable storage devices.

“If you have outages frequently, how should you prepare for them?” Sun said. “It varies from region to region and household to household on which option is the most economical.”

Ultimately, the work could save lives.

“Being able to make decisions based on past experiences is good, but if you know what the risks are into the future, that’s even better,” Sun said. “Now we have this model for past outages, and we can combine that with future extreme weather modeling to understand what our future risks look like.”