PARK RIDGE, Ill. A team of engineers has concluded that the monumental blackout earlier this month over the northeastern United States and parts of the Midwest and Canada could have been avoided had a "smart" North American power grid been shored up with intelligence gathered via neural networks and fuzzy logic to predict local overloads before they happen.
The team, led by a Purdue University researcher, said the Aug. 14 blackout was inevitable under the current system. To prevent another such debacle, the research group recommended that electric utilities adopt not a stronger network of power lines, but an Internet-based strategy to monitor them.
"Unless you put more-significant information technology around the grid, you're going to have more problems," said team leader Leftari Tsoukalas, head of the School of Nuclear Engineering at Purdue University(West Lafayette, Ind.). "The utilities need a major infusion of the Internet at the local levels."
The comments come on the heels of reports that a low-tech telephone network didn't do enough to send out warnings of problems on the electric power grid prior to the blackout. Those problems started in Ohio and subsequently cascaded around Lake Erie and east into Pennsylvania and New York.
Engineers termed the blackout unsurprising. "We foresaw something like this as long as two years ago, and we were not alone," said Yung Liu, manager of energy technology for Argonne National Laboratory (Argonne, Ill.). "A lot of people knew that something like this was bound to happen."
Tsoukalas and other engineers from a research consortium believe that more troubles can be headed off if utilities employ electronic intelligence to predict such problems and quickly localize them. The consortium-which includes 23 researchers from Purdue, Fisk University (Nashville, Tenn.), the University of Tennessee (Knoxville), the Tennessee Valley Authority, Commonwealth Edison of northern Illinois and the Electric Power Research Institute-has developed a system that monitors and predicts electricity consumption, then communicates with utility substations to head off problems.
The key to the system is the use of so-called software agents, which employ neural networks and fuzzy logic to remember consumption patterns at locales such as hospitals, steel mills, big manufacturing plants and retailing concerns. After learning the consumption patterns, the researchers said, the software agents at those "local-area grids" would predict subsequent boosts in power demand minutes or even hours ahead of time, and then work with substation software to identify areas of less demand to balance the consumption.
Researchers said that their system, known as Telos (transmission entities with learning capabilities and online self-healing), uses a software package that incorporates neural nets, fuzzy logic and "wavelets" to make comparisons against known information. Such software could initially be linked to power meters through computer clusters in small municipalities or large businesses at a cost ranging from a few thousand dollars to $100,000. Ultimately, the researchers say, Telos could also be integrated into small embedded modules containing 8-bit microprocessors and Internet capabilities, and then used to communicate with power meters in single-family homes.
"The beauty of this system is that you can do it on a microprocessor that costs $2," Tsoukalas said.
The researchers claim that the costs of such information technology are dwarfed by the alternative, which is to invest in a national program of power line upgrades at about $500,000 per mile. Estimates for the total price tag of such infrastructure upgrades run as high as $100 billion.
By investing in information technology instead of more power lines, the researchers say that utilities could isolate problems before they occur.
"The goal is to identify problematic areas, determine how long the overload will last and then isolate it if necessary," Tsoukalas said. "Local problems are too pernicious to be left alone. They need to be isolated early on."
Indeed, electrical engineers believe that local problems were the origin of the big blackout two weeks ago. Because the power grid was originally designed for local generation and is sorely in need of upgrading, local glitches like the one in Ohio can sometimes cascade through the grid, knocking out power lines in neighboring regions and causing generators to temporarily shut down.
"Power loads have grown by 20 percent over the past 10 years, but the transmission systems have stayed about the same," noted Tim Skvarenina, a professor of electrical engineering technology at Purdue. "We could be headed for more problems in the next 10 years, because the load is again expected to grow another 20 percent to 25 percent."
The research team is currently trying out the "neurofuzzy" system at Argonne National Laboratory, which has unusual electrical consumption patterns for its powerful equipment, such as particle accelerators. There, engineers are "training" the software to recognize those patterns, and doing testing and validation on the system.
Similarly, Commonwealth Edison Co., a unit of Exelon Corp., is testing the technology on a fictitious "data town" that it has created by gathering information from local hospitals, steel mills and businesses. Using the information, Telos "data agents" anticipate electrical consumption and pass the information to transmission substations, which balance the needs of various agents. "This lets us know if a load at a particular substation is going to pick up in, say, five minutes, and allows us to do something before anything bad happens," said David Schooley, a senior engineer at Commonwealth Edison's transmission operations.
Schooley declined to comment on the effectiveness of the software in the context of the recent blackout because of ongoing investigations.
Engineers also said that other digital devices, such as protective relays and meters, could work in conjunction with the system by measuring, remembering and "teaching" the software about currents, voltages, power and frequencies at certain times.
"People are definitely looking for a higher level of intelligence on the grid, so that decisions that would normally be made by an operator can now be automated," noted Ed Schweitzer, president and founder of Schweitzer Engineering Laboratories (Pullman, Wash.), a maker of digital transmission protection systems.
Ultimately, researchers say that the system could even enable homeowners to turn off certain appliances when utilities raise prices during peak consumption hours, resulting in a real-time give-and-take relationship that could keep overloads from occurring.
"This would provide information tools across the entire power grid," Tsoukalas said. "With it, everybody becomes a stakeholder."