Sensory Data Analytics for Securing Wind Farm Generation Against Disruptive Events

点击翻译按钮获取中文摘要

Integrating higher levels of wind power is a critical step to building a secure and sustainable energy infrastructure for the nation. While being green and free, wind power generation can be quite volatile and intermittent, which opposes grand challenges for harvesting wind energy in a reliable and on-demand manner. This proposal aims to address the fundamental challenges of securing reliable power output from wind farms, through the development of online and in-situ sensory data analytics tools. By leveraging the rich information contained in the diverse measurements collected from dispersed meteorological and power sensors widely deployed at wind farms, the tools developed by the proposed project would assist in preventive controls and decision making against impending disruptive weather events or extreme grid conditions that would otherwise cause dramatic power ramps or oscillations. By bridging wind power engineering and sensory data analytics for research and education, the proposed work is interdisciplinary and transformative. The integrated research and education activities of the proposed project would train future engineers with diverse expertise for a thriving wind energy industry, and contribute to fulfilling the DOE's goal of 20% wind energy by 2030 as well as the renewable portfolio standards legislated in many states of the U.S.The proposed project studies wind farm sensory data analytics in two thrust areas: 1) detection and quantification of impending front-induced ramp events by using turbine-level power measurements, and 2) detection and mode analysis of sub-synchronous interactions by using synchrophasor data. Particularly, front-induced ramp event detection is formulated as a new class of change detection problems for multiple time series with spatial dependencies, in which the movement of weather front and the induced wind power ramp are quantified by using wind farm's geographical layout information together with the information extracted from turbine-level power measurements. One intellectual merit of the proposed work arises from the unique and significant insight into discovering the signatures of front-induced ramp events from multiple correlated turbine-level power measurements. Another intellectual merit of the proposed work is the design of model-less and non-parametric methods for sub-synchronous interaction detection using synchrophasor data and synchrosqueezing transform. The proposed approaches to both problems are data-driven, and thus provide novel avenues for non-intrusive condition monitoring and disruptive event detection for reliable wind farm operations, which constitutes the transformative aspect of the proposed work.