You are here
Defending the U.S. Power Grid from Malicious Hackers (Data Science Institute)
Source: Data Science Institute
The sophisticated attack on Ukraine’s power grid last year plunged 225,000 people into darkness for days, and put America’s own aging, sprawling system under the spotlight.
The U.S. Defense Advanced Research Projects Agency (DARPA) has now launched a $77 million program to prevent a similar attack here. Called RADICS, the goal is to develop automated defense systems that operate independently of utilities to identify an attack, isolate vulnerable equipment and quickly get the system running again.
Columbia Engineering professors Gil Zussman, Daniel Bienstock, Dan Rubenstein, and Vishal Misra, all members of the Data Science Institute, will contribute to RADICS under a recent $8 million award to Vencore Labs.
“The grid is extremely vulnerable to cyber and physical attacks as recent events have shown,” said Zussman. “We have spent years studying its theoretical vulnerabilities and are excited to now put this research to practical use.”
The U.S. grid is divided into three regions. Each has its own interconnected network of power stations, power lines, electrical-transmission substations and control centers. If a key piece fails, it can put the entire region at risk for cascading failures. A physical attack, or even a combined cyber-physical attack like the one Russia is thought to have orchestrated in Ukraine, could have devastating consequences.
The system, of course, has failed spectacularly before without outside meddling. In 2003, a heat-stressed power line in northern Ohio set off a chain of failures that shut power to 55 million people from Ontario to New York City for up to two days. In 2011, a technician accidentally cut a power line in the Arizona desert knocking out power to 7 million people from San Diego to Tijuana. Similar events, in India, Turkey and elsewhere have highlighted the extent of grid vulnerabilities globally.
Researchers have analyzed blackouts like these to understand how to better predict and prevent them. Previously, researchers had tried to simulate blackouts using a model of how epidemics spread, but those models failed to recreate the magnitude and observed patterns of actual events.
Looking closely at San Diego’s cascading failures, Bienstock and Zussman demonstrated in a 2014 study that they could successfully simulate the blackout’s evolution using a so-called Direct Current power-flow model. As in real-life, the power lines in their model did not need to be touching for problems like a mismatch between power supply and demand to knock out nodes in the network.
Bienstock, who has literally written the book on cascading failures—Electrical Transmission System Cascades—will continue to build on this work for DARPA, improving the model and its underlying algorithms to capture the physics of how electricity moves over a network of power lines. The eventual goal is to predict massive cascading failures in real-time and to identify locations on the grid that require extra close monitoring.
The project builds on computational methods Bienstock has developed to identify weaknesses an attacker could exploit. He is currently using machine-learning algorithms to sift through a database of simulated attacks to find subtle vulnerabilities.
Still, pulling off an attack would not be easy, says Bienstock. “It would require industry knowledge and access to good data. Even if hackers managed to take down the grid, they would leave a digital trail that would probably lead investigators to their door. Nonetheless, the system is vulnerable, and a successful attack could be crippling.”
Zussman will lead the design of a second component focused on detecting and recovering from an attack— whether from a hacker injecting code into the system’s computer controls, or someone physically attacking the system as snipers did at a PG&E substation near San Jose, Calif. in 2013.
Building on prior research, he will develop algorithms that divide the grid into virtual, strategic sub-areas to compare power flows and other data gathered in one area with those collected in other areas. Inconsistencies in the data would trigger a warning that a cyber attack on the equipment that measures grid power flows could be in progress.
“If the equipment reports a high flow on a power line in an area with little incoming flow and no major power station, that should raise a red flag,” said Zussman. “Our algorithms would pick up more subtle anomalies.”
In a third area of research, Columbia researchers Misra and Rubenstein will contribute to the DARPA project by building a system to monitor data beyond the grid itself, such as Twitter feeds and communications networks, to identify regions without power. If the grid monitoring system were attacked, these outages could go undetected.
The three components are part of an $8 million situational awareness tool dubbed MANTESSA, for Machine-Intelligence for Advance Notification of Threats and Energy-Grid Survivable Situational Awareness that will incorporate an understanding of physical power flows to flag unusual activity indicating a possible attack. Vencore Labs will lead its development, with assistance from researchers at Columbia, Princeton and Carnegie Mellon universities.
— Kim Martineau