, American Electric Power
Pages: p. 57
The electric power industry has a relatively rich history of AI research and development that goes back to the mid-1970s with early isolated attempts in pattern-classification applications for power system security assessment. In the early 1980s, the advent of AI expert systems spurred numerous academic and industrial projects in the field. Unfortunately, a vast majority of those projects never passed the prototyping stage. In the mid-1980s, the power industry's focus shifted from symbolic AI to connectionist AI, and researchers attempted to apply neural networks to various power system problems. In this era, load-forecasting applications received much attention, with very encouraging results.
Throughout the past decade, researchers have reported a number of results for distributed-problem-solving, hybrid-symbolic-connectionist, and machine-learning systems. However, not many of these systems scaled up to industrial-grade software products. With the advent of deregulation in the power industry, industry experts anticipate that distributed AI and various flavors of game theory will gain in popularity.
Over the next several issues, IEEE Expert's special track on AI in power systems will give you an idea of where the field is and where it's headed. The track features a rich set of articles spanning several countries and both academia and industry:
Richard Christie, G.W. Rosenwald, Chen-Cheng Liu, and I provide an overview of the field.
Sarosh Talukdar, former director of CMU's Power Engineering Program, discusses trends in AI research in power systems in this issue's interview.
David Leahy, Jonathan Wallace, Maurice Mulvenna, and John Hughes discuss an intelligent assistant for contract compliance in a deregulated utility environment in the United Kingdom. They explain how heuristic or knowledge-based systems can profitably replace the traditional operations-research methods (such as dynamic or integer programming) for unit commitment, to handle large contracts containing several thousand pages of text, tables, and graphics.
D. Atanackovic, D. McGillis, F. Galiana, J. Cheng, and L. Loud describe a system-planning model that is designed to resemble the actual workflow in a power system-planning workgroup. Because it assumes vertical integration, its command-and-control structure might need to be revised to reflect the more distributed nature of the post-deregulation environment, but it still conveys valuable information.
Zita Vale, A. Machado e Moura, M. Fernanda Fernandes, Albino Marques, and Couto Rosado describe an on-line expert system that helps manage operational alarms in a Portuguese substation control center, thereby helping the human operator make decisions.
Jeffrey Bann, Guillermo Irisarri, Daniel Kirschen, Bradley Miller, and S. Mokhtari describe the design of an evolved, production-grade architecture that serves as an interface between coexisting AI applications and an energy-management system supporting an AI-specific power system model. They report that their AI-EMS interface is simple, flexible, and cost-effective.
Finally, Louis Wehenkel addresses the application of symbolic machine learning to the security assessment of electric power systems. This system extracts operationally useful knowledge from a power system database. His article provides a good overview of the relevant system-security concepts.
Some people in the power industry believe that AI is a technological solution looking for problems. Hopefully, the practical applications described in these articles will help overcome that misperception.