Power outages are frustrating for everyone involved, and large-scale outages—blackouts—can cripple a city, a region, or even an entire country for days. Electrical engineer Nilanjan Ray Chaudhuri, an expert on power grids, heads a Penn State team that won a $999,000 grant from the National Science Foundation to explore ways to prevent and deal with the “cascading failures” that lead to massive blackouts. The project combines Chaudhuri’s knowledge of power systems with the communications and networking expertise of colleagues Tom LaPorta and Ting He. Chaudhuri recently described some of the challenges involved, and how he and his colleagues plan to address them.
Could you give an example of “cascading failure?”
The power grid is more than a physical system. It is reliant on a cyber layer—sensor networks, communications, algorithmic layer, all talking to each other. A failure in any one of those could start a blackout.
In the 2003 blackout in the northeastern U.S., the energy management system software was supposed to be updated or upgraded, and it faced issues. It was out of service for hours. So the operators in the control centers had no clue that lines were getting overloaded, power was getting re-routed, and which way.
So the operators were not getting the data they needed to manage the system.
They need a near-real time snapshot of the system’s status. And they’re not getting it—and they don’t know they’re not getting it. And everything went wrong from there, because when the system came back on, the operators were taking all the wrong decisions. All these very complex interactions can spread, one failure causing the other. That’s how, from one region of failures, this can become a much bigger problem.
How are you and your colleagues approaching these issues?
We want to come up with a more realistic model of cascading failure in a power grid coupled with the communication network on which it depends. The three legs of the project are modeling, preventive control, and restoration. We perform different ‘what-if’ scenarios. What if a major component like a generator or transformer or transmission line fails? What if two such components fail at the same time? Or three? People have tried to use very simple models, which are easy to analyze but give you wrong results in later stages of a blackout. A quite detailed model is going to give you results that are closer to reality—if you could do the computations. Can we determine the stage of cascading failure up to which I can trust my simple model, and beyond that point, I will use a more complex model?
After cascading happens, you want to bring power back as soon as you can. There, our goal is to develop approaches where you can estimate the state of the system with partial visibility. You know your system is down; you do not know which parts of your system are down, because your sensors are also down.
They depend on the power that they’re supposed to be monitoring.
Interesting, right? What people have done is to provide backup power supplies, but of course, batteries also have a certain capacity. If they start dying, then you start getting blind.
The other aspect is how we can strategically place the control center, where you have all the information coming in. Then you need to find out the minimum number of communication links to the control center that need to be preserved no matter what, so that at least that information continues to come to your control center.
Are there any new dangers to the power grid that we need to consider?
One new dimension is cyber-attack on sensors, and how that can negatively impact a cascading failure, and then how we can have redundancies or appropriate placement of sensors that can help us avoid cascading failure in spite of cyber-attacks.
Another is natural disaster, which is becoming more important now because of climate change and the types of weather patterns we are observing. Previously, we did not consider lightning striking at the same time in many places in our what-if scenarios. Of late, these extreme events are becoming more and more common. We might have four weather fronts with significant thunderstorm potential across a region, and we’re having lightning strikes continuously. Or there is a flood or high gale which is physically damaging the network. Now you’re talking about whether you’re ready for three or more components going out just because of a weather front. We need to talk to weather experts and try to develop models that can take input from weather-related variables.
Nilanjan Ray Chaudhuri is assistant professor of electrical engineering and computer science. Tom LaPorta is William E. Leonhard Endowed Chair, Evan Pugh Professor, and director of the School of Electrical Engineering and Computer Science. Ting He is associate professor of computer science and engineering.
This story first appeared in the Fall 2019 issue of Research/Penn State magazine.