TROY, N.Y. (NEWS10) — Winter weather shocked people across Texas, many still recovering after days-long blackouts. At last check Friday afternoon, some 150,000 are still without power, down from the peak at around 4 million. There are also 13.6 million still affected by water service interruptions across the lone star state.
Engineering students and one New York company may just have the answer to get power grids ahead of any future disasters.
“Make it more resilient so that things like this can be avoided, like what is happening in Texas, to be seen way ahead of time and have a mitigation plan in advance,” explains Purnel Kharel, CEO and president of K&A Engineering.
K&A works with several New York utility companies and many more around the nation, plus an office in Nepal. A new partnership with the Rensselaer Polytechnic Institute Design Lab let them bring in students for a special task — find a way to apply neural networks to electrical substations.
“Trying to essentially train a model, so you give it a bunch of data and you train it as if it were a human and it naturally picks up on relationships,” explains senior student Palmer Feinburg, who was one of the first students to join the collaborative in Fall 2020.
“I always tell my students at the beginning of each semester, don’t think of this as a class you need to pass to graduate. You can’t just cram the night before. Think of it as a semester project that will carry you into the real world, it’s a months-long effort,” says Professor Kathryn Dannemann, the RPI Design Lab director.
A neural network could be used to detect power changes, problems and failures much faster and computers would have an automated design to start reacting immediately. The AI also creates a completely virtual copy of the substations, which Kharel calls a “digital twin”.
“The model can let you plan and play with all these different scenarios. What if the weather is terrible? Too hot or too cold? What it a Category 5 hurricane hits it or a tornado? You can design a resilient grid and test it against these scenarios so that you can build your network to handle situations like this,” he explains to NEWS10’s Mikhaela Singleton.
“There are also the safety hazards to consider. You cannot go to a live substation, there is the high voltage but also if you do something wrong, you could cause a major blackout. There are things at a substation you cannot even touch or things you can’t open,” Kharel goes on to say. “What the neural network and the digital twin do is run constructability analysis, make sure everything works, everything fits, and you can make adjustments remotely.”
Both Kharel and Feinburg agree in engineering, your result can only be as good as your data. So first, Kharel says one of K&A’s New York power company clients graciously shared real data from one of its substations. Then the RPI students got to work to see if integrating the artificial intelligence was even possible.
“It was pretty much trial and error, because there wasn’t a lot of research. We were all learning this as we went, but then the guys working with the program worked out a working model and it really took off from there,” Feinburg explains.
Kharel says while neural networks are not new to businesses working to innovate for the future, he says applying them to power grids has never been done before.
“The very traditional industry that is utility was lagging in that space. We saw a tremendous need for really investing in technology, and then what’s the best place to go? RPI already has a design lab and they have the right and talented students,” he says.
Feinburg says he’s now looking forward to an internship with K&A and keeping his real-world education going.
“It was definitely way different from learning in the classroom, and I definitely learned a lot more and a lot shorter time,” he says.
“We want to take these young minds together with the best, most experienced people so that we can come up with solutions to solve future problems,” Kharel says.