This talk describes a new approach, using ``Random Chemistry'' sampling, to efficiently estimate the risk of large cascading blackouts triggered by multiple contingencies. On a model of the Polish electrical grid, the new approach finds the expected value of large-blackout sizes (a measure of risk) two orders of magnitude faster than Monte Carlo sampling, without introducing measurable bias. We also derive a method to compute the sensitivity of blackout risk to individual component-failure probabilities, allowing one to quickly identify low-cost strategies for reducing risk. For example, we show how a 1.9% increase in operational costs reduced the overall risk of cascading failure in a 2383-bus test case by 61%. An examination of how risk changes with load yielded a surprising decrease in cascading failure risk at the highest loadings, due to increased locality in generation and less long-distance transmission. Finally, we propose new visualizations of spatio-temporal patterns in cascading failure risk that could provide valuable guidance to system planners and operators.
Bio of Dr. Eppstein:
"Dr. Margaret (Maggie) Eppstein has highly interdisciplinary educational background and interests with a B.S. and graduate study in Biology, an M.S. in computer science, a Ph.D. in environmental engineering, and continuing education in complex systems. She is currently the Chair or Computer Science Department, Professor of Computer Science and was the founding Director of the Complex Systems Center at the University of Vermont (2006-2010). Most of her early research focused on developing novel Bayesian computational methods for large-scale, nonlinear, multi-scale tomographic inverse image reconstruction problems in three-dimensional subsurface hydrology, geophysics, and deep-tissue imaging. However, over the past decade her research has shifted towards modeling and analysis of complex adaptive systems in general. Current or recent projects include developing, studying, and using novel bio-inspired computational approaches (including evolutionary algorithms, agent-based modeling, and artificial neural networks) for a wide range of important problems, including design of watershed management plans, plant species’ invasiveness in ecological communities, biological speciation, the impact of spatial topologies on information flow through complex interaction networks, identifying nonlinear interactions between single nucleotide polymorphisms that predispose for complex disease traits, agent-based integrated assessment modeling of transportation energy alternatives, analyzing non-linearly interacting outages that cause cascading failures in electrical networks, studying the evolution, structure, and function of a world-wide network of neonatal intensive care units, and exploring the effectiveness of alternative search strategies used by hospital teams in seeking improvements in health care. In a more general sense, she is interested in understanding evolvability and emergent properties of dynamical processes on complex networks."
RACC Seminar 10-27-2014: Dr. Margaret Eppstein - Assessment, Visualization, and Mitigation of Cascading Failure Risk in Power Systems
RACC Seminar 10-27-2014: Dr. Margaret Eppstein - Assessment, Visualization, and Mitigation of Cascading Failure Risk in Power Systems