Researchers have developed software to predict where blackouts are most likely to happen when storms hit, which could help authorities cut the amount of time people are in the dark after disasters like Hurricane Sandy.
Sandy wreaked havoc in 2012, causing as much as nearly $50 billion in damage, making it the second-costliest hurricane to hit the United States.
At its peak, it left roughly 8.5 million people without power.
“As large storms increase in frequency and intensity in the United States and worldwide due to changing climate, getting profiles of where places are vulnerable to damage and investing in infrastructure to eliminate those vulnerabilities is integral to maintaining a well-operating power grid,” says Steven Fernandez, a national security issues researcher at Oak Ridge National Laboratory in Tennessee.
As hurricanes approach, forecasts about their paths and the wind speeds areas might experience are often fairly accurate up to three days before landfall.
Analysts combine this data with knowledge concerning how vulnerable power lines are — for example, if they have a 50 percent chance of failure in 65 mph winds — to predict which parts of the electrical grid might go out when storms hit. Emergency responders can then prepare where to focus their efforts.
However, storms change over time, and researchers want predictions that are as up to date as possible. The problem is that it can take analysts an hour to prepare each estimate based on a weather advisory, and hurricanes that hit the United States typically lead to more than 40 weather advisories, with updates issued every two to three hours around the clock — Sandy alone spawned 64 such advisories.
Manually developing estimates based on each update would consume time the few analysts that are typically available might not have, and such tedious, repetitive labor is prone to error.
Manual to automatic
Instead of relying on analysts, Fernandez and his colleagues have developed software that fully automates the process of determining how storms might knock out power.
“This software takes these predictions to the next level,” Fernandez says. “It takes weather models from the National Hurricane Center and flood information from the U.S. Geological Survey. It combines this with national-level models of where power grid components are and where people associated with those grids are from the LandScan program to figure out county by county what percentage of people might lose power, where they are and how long power is going to be out.”
All the extra variables the system analyzes “reduces problems of false positives and false negatives — that is, you don’t predict areas will stay in operation when they actually will get knocked out, or lose power when they actually ride the storm out,” he says. “This will help emergency responders get crews and supplies out to places that need power the most when these big storms occur, to make a difference about whether people survive and to minimize the amount of suffering that happens.”
The software can finish power outage analysis within 15 minutes of receiving a weather advisory. After the software frees analysts up from this labor, these experts will be able to concentrate on refining estimates further by integrating power utility reports and other information from the ground during storms.
It is also possible “to combine these models with climate change predictions to see how electrical grid outages might actually evolve over time as weather changes, where water will be and how people move in reaction to these changes,” he says. “We can make judgments as to what areas in the future will be more or less vulnerable to these extreme events.”
More robust models coming
The research team hopes to make this software available to mobile users. They also plan to expand their model to include factors such as tree density, as well as to analyze storms worldwide instead of just the United States.
“We’d like to more definitively identify vulnerabilities within the power grid — for instance, how disruptions cascade through the infrastructure,” Fernandez adds. “This might help reveal ways to short-circuit those disruptions before they lead to large-scale blackouts by hardening certain transmission lines or power substations.”
More detailed knowledge about conditions areas face is key to improving predictions. For instance, while analysts could accurately predict when 80 percent of people who lost power would get electricity restored, “the 20 percent who waited much longer to get power back were much more dependent on local conditions — say, a tree hit a piece of equipment that took a long time to replace, or it took a longer time for a place to dry out after flooding than what models predicted,” he says.
Future models could model other natural disasters that can impact the electric grid, such as earthquakes and wildfires.
“With wildfires in the West, we can get infrared signatures of those systems and overlay those over infrastructure to predict outages,” he says. “With developing thunderstorm cells, we can look at potential damage associated with tornadoes, which are much more surgical in their damage profiles than hurricanes, but the damage where they do hit is much more complete. We can also look at flooding from stream gauge data and predict what places will get inundated and probably de-energized.”