Systems Seminar - ECE
Learning and Data based Decision Making with an Application to Power System Planning under Uncertainty
Add to Google Calendar
Making optimal decisions in an uncertain environment is a challenging task. In the same time, advances in many engineering disciplines have led to a huge amount of easily accessible data. This "big data" trend, leads to a paradigm shift in control and optimization, rendering data driven control an alternative to deterministic or robust techniques. In the first part of the talk we focus on the problem of designing a decision policy that optimizes a certain criterion and is immunized against data uncertainty. We show that the resulting decision can be accompanied with a performance certificate that is provided a-priori to the decision maker and encodes the confidence with which the decision maintains its robustness properties against uncertainty realizations other than those included in the data. We relate the issue of certificate provision with "learning and generalization" paradigms in machine learning and analyze the implications of this approach to optimization problems with certain structural characteristics. The second part of the talk will illustrate how the tools we have developed can be employed to revisit planning problems in power systems with renewable energy sources, providing a-priori guarantees regarding the satisfaction of the system constraints. This is in contrast to earlier approaches to such problems, that are typically restricted to rule based or ad-hoc methodologies.
Kostas Margellos received the Diploma in electrical and computer engineering from the University of Patras, Greece, in 2008 and the Ph.D. in automatic control from ETH Zurich, Switzerland, in 2012. In 2013 he was a post-doctoral researcher at ETH Zurich, and since January 2014 he continues his post-doctoral research in the Department of Industrial Engineering and Operations Research at UC Berkeley. His research interests include analysis and control of hybrid systems, data based optimization and applications to stochastic decision making problems in power networks with uncertainty.