Uncertainty-driven adaptive estimation with applications in electrical power systems

Zhang, Jinghe

Download PDF

Request Version for Screen Reader

Last Modified

March 22, 2019

Creator

Zhang, Jinghe
- Affiliation: College of Arts and Sciences, Department of Computer Science

Abstract

From electrical power systems to meteorology, large-scale state-space monitoring and forecasting methods are fundamental and critical. Such problem domains pose challenges from both computational and signal processing perspectives, as they typically comprise a large number of elements, and processes that are highly dynamic and complex (e.g., severe nonlinearity, discontinuities, and uncertainties). This makes it especially challenging to achieve real-time operations and control. For decades, researchers have developed methods and technology to improve the accuracy and efficiency of such large-scale state-space estimation. Some have devoted their efforts to hardware advances---developing advanced devices with higher data precision and update frequency. I have focused on methods for enhancing and optimizing the state estimation performance. As uncertainties are inevitable in any state estimation process, uncertainty analysis can provide valuable and informative guidance for on-line, predictive, or retroactive analysis. My research focuses primarily on three areas: 1. Grid Sensor Placement. I present a method that combines off-line steady-state uncertainty and topology analysis for optimal sensor placement throughout the grid network. 2. Filter Computation Adaptation. I present a method that utilizes on-line state uncertainty analysis to choose the best measurement subsets from the available (large-scale) measurement data. This allows systems to adapt to dynamically available computational resources. 3. Adaptive and Robust Estimation. I present a method with a novel on-line measurement uncertainty analysis that can distinguish between suboptimal/incorrect system modeling and/or erroneous measurements, weighting the system model and measurements appropriately in real-time as part of the normal estimation process. We seek to bridge the disciplinary boundaries between Computer Science and Power Systems Engineering, by introducing methods that leverage both existing and new techniques. While these methods are developed in the context of electrical power systems, they should generalize to other large-scale scientific and engineering applications.

Date of publication