EAGER-DynamicData: A Scalable Framework for Data-Driven Real-Time Event Detection in Power Systems

点击翻译按钮获取中文摘要

Electricity is the lifeblood of our society; therefore providing a reliable and efficient electricity supply is vital for ensuring human welfare and sustainable economic growth. A pivotal need in ensuring reliable operation of the US power grid is the development of sophisticated and robust tools for monitoring and anomaly detection. To this end, this research project aims to develop robust and scalable data-driven inference algorithms for detecting and isolating the occurrence of undesirable events that could threaten the integrity of the grid. In this regard, the combination of tools and methods on which the project will rely, namely (i) power system reliability modeling and analysis, and (ii) statistical signal processing and detection, and estimation theory, will result in a unique interdisciplinary collaboration program.The proposed framework relies on large datasets obtained with phasor measurement units (PMUs) located across the system. By exploiting the statistical properties of voltage phase angle measurements obtained from the aforementioned PMUs, algorithms will be developed to detect and identify undesirable events in power grids, e.g., outages in transmission lines and other assets, in near real-time. Specifically, the ultimate objective of this research is to develop a data-driven framework for real-time detection of undesirable events in power systems that is robust and highly scalable. The framework builds on existing powerful tools from the theory of quickest change detection (QCD), and will provide techniques for partitioning the graph describing the connectivity of a power system, and PMU placement to allow these QCD-based tools to be exploited in large scale systems such as the US power grid. Additionally, the research will explore the challenging problem of explicitly incorporating the sparsity structure of the undesirable events in our QCD-based algorithms to make them scaleable to multiple events.