AMPS: Dynamics-Aware Algorithms for Real-Time Structured Fault Detection in Power Systems

美国电网正处于自成立以来最根本的变革之中。在减排需求的推动下，发电组合正从传统的、可靠的能源转向不稳定和不确定的可再生能源。与此同时，传感和监测设备的数量、质量和种类都出现了前所未有的增长。从相量测量单元(PMU)到智能电表，电网很快就会经历数据溢出，这些数据有可能在传输和分配级别提供超大数量的信息。然而，尽管数据可用性激增，但仍然缺乏能够利用最新可用的遥测来帮助运营商面对可再生能源带来的范式转变的分析工具。此外，利用网络资源传输信息以进行处理和分析的监测系统的现代化带来了新的挑战和威胁。如果没有适当的工具来纠正和验证收集的数据，未检测到的错误或恶意修改的数据可能会误导操作员，并使系统走向停电。这项工作通过开发考虑到测量的内在属性的新型算法工具来解决这些挑战。该项目开发了一个理论框架和相关算法，可以可靠地利用新获得的测量来提供有用的实时信息，使运营商能够更好地利用资源并对不可预见的事件做出反应。更准确地说，PI寻求结合统计、动态系统和优化的工具来开发数据分析方法，以识别和防止网络物理攻击、纠正丢失和损坏的数据、识别网络结构变化，并识别供需失衡的局部突然变化。这项研究在现有的电力系统监测文献中是独一无二的。首先，它承认并利用了这样一个事实，即由底层动力系统(电网)产生的测量之间存在空间和时间相关性。其次，它建立了一个统一的建模框架，可以联合捕获电网测量如何受到以下因素的影响：(A)网络中的拓扑变化，(B)供需的突然变化，以及(C)测量误差。第三，开发了一种新的算法框架，该框架利用不同网络扰动的结构稀疏性来区分和识别给定电网暂态行为的来源。本研究还将搭建一个大规模的仿真试验台，以评估所设计算法的准确性和可扩展性，从而弥合理论模型和实际电力系统之间的差距。

The U.S. power grid is in the midst of its most fundamental transformation since its inception. Spurred by the need to reduce emissions, the electric generation mix is drifting away from traditional, reliable sources, towards volatile and uncertain renewable sources. Simultaneously, there is an unprecedented increase in the quantity, quality, and variety of sensing and monitoring devices. From phasor measurement units (PMUs) to smart meters, the power grid is soon to experience an overflow of data that has the potential of providing an extraordinary amount of information at the transmission and distribution levels. However, despite this burst in data availability, there is still a lack of analytic tools that can leverage the newly available telemetry to help operators face the paradigm shift that renewable sources pose. Moreover, the modernization of monitoring systems that use cyber resources to transmit information for processing and analysis begets new challenges and threats. Without the proper tools to correct and validate the collected data, undetected errors or maliciously modified data can mislead operators and bring the system towards blackouts. This work addresses these challenges by developing novel algorithmic tools that take into account intrinsic properties of the measurements.This project develops a theoretical framework and associated algorithms that can reliably utilize the newly available measurements to provide useful real-time information that can allow operators to use resources better and react to unforeseen events. More precisely, the PIs seek to combine tools from statistics, dynamical systems, and optimization to develop a data analytic approach to identify and prevent cyber-physical attacks, correct missing, and corrupted data, identify network structural changes, and recognize abrupt local changes in supply- demand imbalance. This research is unique within the existing the literature in power system monitoring in several ways. Firstly, it acknowledges and leverages the fact that there are spatial and temporal correlations between the measurements generated by the underlying dynamical system (the power grid). Secondly, it builds a unifying modeling framework that can jointly capture how grid measurements are affected by (a) topology changes in the network, (b) abrupt changes in supply or demand, and (c) measurement errors. Thirdly, it develops a novel algorithmic framework that exploits structural sparsity properties of the different network disturbances to discriminate and identify the source of a given grid transient behavior. The research will also build a large-scale simulation testbed to assess the accuracy and scalability of the designed algorithms and in this way bridge the gap between theoretical models and actual power systems.