Advancing Graph Signal Processing Techniques for Monitoring and Control of Electric Distribution Power Systems

先进的图形信号处理技术用于配电电力系统的监测和控制

• 批准号: 2210012
• 负责人: Anna Scaglione
• 金额: $ 36万
• 依托单位: Cornell University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-07-01 至 2025-06-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2210012&HistoricalAwards=false
• 关键词: Advancing; Graph; Signal; Processing; Techniques

This NSF project aims to incorporate the physical modeling of the electric grid in the theory of machine learning algorithms, to benefit the monitoring and control of energy delivery systems. While the basic theory we will develop applies broadly to transmission and distribution systems, the section of the grid we are focusing on is the one that is undergoing the greatest transformation, which is the distribution grid. This section of the system not only poses unique modeling challenges, it is also undergoing significant changes because of the integration of distributed energy resources and the control of responsive demand and storage. These are the key ingredients to sustainable energy delivery, and the project will bring transformative changes to the digital technology and machine intelligence that can accelerate this transition. More specifically, the proposal explores a novel mathematical approach for the analysis of grid signals, rooted in fundamental power systems graph-based methods and born out of interpreting the system state as an instance of graph signals. The goal is to use the insights that come from Graph Signal Processing (GSP) and from graph Fourier analysis, to extract signals features that allow to improve data driven inference and decision algorithms. At this time, GSP machine learning tools are designed for real signals and are not physics based. The project will fill this gap, by providing the underpinning for a theory of grid graph signals. This entails extending the GSP tools to tackle complex signals, incorporating the grid system parameters in the algorithm and considering realistic power measurements systems. The goal is to have a better representation of the spatial-temporal characteristics of the data as compared to generic machine learning algorithms and advance the theoretical tools in GSP which are not based on systems whose properties are captures by the signals envelopes and on the physics of the grid. The tools developed will be made available open source. In addition to the advances in GSP, the project will have broader impact through its outreach to New York City public schools and create a short program for K-12 students, supported by an illustrated book, explaining how energy is delivered and the path that advanced societies need to follow to achieve the goal of a decarbonized economy.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

该 NSF 项目旨在将电网的物理建模纳入机器学习算法理论中，以利于能源输送系统的监测和控制。虽然我们将开发的基本理论广泛适用于输电和配电系统，但我们关注的电网部分是正在经历最大变革的部分，即配电网。系统的这一部分不仅提出了独特的建模挑战，而且由于分布式能源的整合以及响应需求和存储的控制，它也正在发生重大变化。这些是可持续能源输送的关键要素，该项目将为数字技术和机器智能带来变革，从而加速这一转变。 更具体地说，该提案探索了一种用于分析电网信号的新颖数学方法，该方法植根于基本电力系统基于图的方法，并将系统状态解释为图信号的实例。目标是利用来自图信号处理 (GSP) 和图傅里叶分析的见解来提取信号特征，从而改进数据驱动的推理和决策算法。目前，GSP 机器学习工具是为真实信号设计的，而不是基于物理的。该项目将通过为网格图信号理论提供基础来填补这一空白。这需要扩展 GSP 工具来处理复杂信号，将电网系统参数纳入算法并考虑实际的功率测量系统。目标是与通用机器学习算法相比，更好地表示数据的时空特征，并改进 GSP 中的理论工具，这些工具不基于其属性由信号包络捕获的系统，也不基于以下物理特性：网格。开发的工具将开源。除了 GSP 方面的进步外，该项目还将通过向纽约市公立学校推广并为 K-12 学生创建一个短期项目，并附有一本插图书，解释能源如何输送以及推进的路径，从而产生更广泛的影响。社会需要遵循以实现脱碳经济的目标。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Reinforcement Learning using Physics Inspired Graph Convolutional Neural Networks

使用物理启发的图卷积神经网络进行强化学习

• DOI: 10.1109/allerton49937.2022.9929321
• 发表时间: 2022-09-27
• 期刊: 2022 58th Annual Allerton Conference on Communication, Control, and Computing (Allerton)
• 作者: Tong Wu;A. Scaglione;D. Arnold
• 通讯作者: D. Arnold

Spatio-Temporal Graph Convolutional Neural Networks for Physics-Aware Grid Learning Algorithms

用于物理感知网格学习算法的时空图卷积神经网络

• DOI: 10.1109/tsg.2023.3239740
• 发表时间: 2022-03-31
• 期刊: IEEE Transactions on Smart Grid
• 影响因子: 9.6
• 作者: Tong Wu;Ignacio Losada Carreño;A. Scaglione;D. Arnold
• 通讯作者: D. Arnold