Managing Uncertainty in Future Smart Grid: An Online-Algorithmic Approach towards Robust and Efficient Decisions

Xiaojun Lin
Associate Professor of Electrical and Computer Engineering
Purdue University
SERC 306
Wednesday, April 15, 2015 - 11:00
How to respond to uncertainty is one of the primary challenges facing future power systems, which must operate under significant uncertainty both in the renewable supply (wind/solar) and in the demand patterns. Such uncertainty is often revealed sequentially in time, and thus the decision at each instant must be adjusted based on the information that has already been revealed, and yet be prepared for the remaining uncertainty towards the future. Further, the nature of the power systems often dictates that robust performance guarantees must be ensured even at the worst-case uncertainty, e.g., the energy supply must always meet the demand, and otherwise the entire power grid may fall apart. Thus, there is a pressing need to develop sequential decision algorithms that can achieve robust worst-case performance against future uncertainty. In this talk, we argue that competitive online algorithms could be a useful framework for solving this type of sequential decision problems in future smart grid. In the typical CS literature, an optimal competitive online algorithm, which achieves the smallest possible worst-case competitive ratio compared to the offline solution, can be found even when there is absolutely no prior information about the future input. However, in power systems, such competitive results could be quite pessimistic because it does not exploit any partial (yet inaccurate) future information that may be available. Instead, in this work our goal is to develop computationally-efficient online algorithms that are both robust (in terms of worst-case guarantees), and efficient (in terms of exploiting any partial future information that becomes available).