WEST LAFAYETTE, Ind. The Consortium for the Intelligent Management of the Electric Power Grid launched its flagship effort this month with a plan intended to prevent power failures by using a self-healing regulatory system that smartens its management decisions by drawing on fuzzy logic and neural learning. "We want to make the power grid self-regulating, so that it anticipates future demands by learning from past usage patterns," said Lefteri Tsoukalas, an associate professor specializing in fuzzy logic and neural learning at Purdue University's School of Nuclear Engineering.
The consortium, one of six nationwide groups designed to modernize power distribution in the United States, includes Purdue, Commonwealth Edison Co., the University of Tennessee, Fisk University, the Tennessee Valley Authority and the Electric Power Research Institute, an organization of electric utilities). The consortium described its smart-power management prototype dubbed Telos, for Transmission Entities with Learning and Online Self-healing at a meeting earlier this month.
The group's purpose is to make the grid smart enough to prevent the brownouts and other power failures currently plaguing major cities, especially during summer heat waves. "A system that can learn to accurately predict power needs for the coming day could automatically meet those demands, but so far we have been unable to forecast usage patterns accurately enough there are just too many variables," said Tsoukalas. "In order to capture the complexity of the grid, we need to learn more about its customers."
To accomplish that, Purdue will employ neural learning to browse through the past usage patterns of utility customers and correlate them with the weather and other pertinent environmental factors on a day-by-day basis. From that knowledge, the group will craft a fuzzy-logic controller that implements the model learned by the neural network. A test neighborhood to give the system a trial run De Kalb, Ill. was chosen because it contains many different types of users, from industrial plants to residential neighborhoods, and even a university with its own auxiliary power generators.
In De Kalb, an automated system will be put in place that controls small sections of a service area with energy resources that can be quickly redistributed to meet the varying demands in different parts of town. Its fuzzy-logic controller will only run in parallel as an adviser to a human operator it won't take control of the actual grid until testing is completed. The city has slated a demonstration of the Telos prototype for the fourth quarter of 2001.
Unfortunately, even with neural learning it is sometimes impossible to redistribute enough power to meet everybody's needs, especially in unusually hot weather. Electricity cannot be stored for significant periods, so it cannot be stockpiled at night, for instance.
"Universities and hospitals already have emergency generators in place that take over for them when the power goes out, but if we could make deals with them to contribute their power to the grid when the power is not down, then we could use them to meet needs that can't be met by redistribution," said Tsoukalas.
However, to manage such local generators, a local power management system is needed, what Tsoukalas calls a "local-area grid." With demands varying dramatically from neighborhood to neighborhood, the local-area grid smooths supply variations by drawing on local resources as necessary.
Tsoukalas and his colleagues at Purdue believe that neural learning and fuzzy logic can manage not only the grid but also the local generators, to smooth the flow and meet varying demands even during times of peak energy consumption. "The human brain makes effective decisions by evaluating the overall accuracy of a solution, without requiring fine precision in each part," he said. "Likewise, fuzzy logic can effectively manage power resources without having to calculate precisely the perfect solution. These problems are too complex to calculate everything; we have to build systems that have the ability to summarize."
Telos will work from a mathematical model that predicts changes in demand by comparing past patterns to current demands and environmental conditions. Neural learning will match historical patterns to current conditions, while fuzzy logic combines real-time power demand profiles with constraints on power grid redistribution and local generator capabilities.
By automatically tracking and compensating for new conditions, Telos will "self-heal" by matching local demands with supply. The ultimate goal is to eliminate the need for large, centrally located power grid management systems.
The local-area grids will be prototyped in places like De Kalb. If successful, the smart-power management system will be put online there by 2003, after which it will require a 10-year testing period. The local-area grid could be implemented nationwide by 2015.