Collaborative Research: Online Data Stream Fusion and Deep Learning for Virtual Meter in Smart Power Distribution Systems

合作研究：智能配电系统中虚拟电表的在线数据流融合和深度学习

• 批准号: 1933212
• 负责人: Tianbao Yang
• 金额: $ 10.94万
• 依托单位: University of Iowa
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-09-01 至 2022-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1933212&HistoricalAwards=false
• 关键词: Collaborative; Research; Online; Data; Stream

With ever growing deployment of information and communication technologies in engineering systems online steaming of data becomes available. Online learning algorithms can utilize such high value data to enhance operation of national critical infrastructures such as power grids. Power distribution systems, unlike transmission power grids, lack extensive direct online measurement through sensing infrastructures. This makes an accurate monitoring of power distribution systems, which is crucial for reliable operation of the system, a challenging task, specifically in power distribution systems with massive integration of intermittent renewable energy sources that increase the variability of the aggregated load-generation values. The proposed research enables reliable monitoring of power distribution systems with massive integration of renewable energy which has economic and social impacts on the public. The proposed online optimization techniques, which will be investigated in this project, can be applied to a variety of learning tasks over data streams beyond power engineering. Research and teaching will be integrated through development of interdisciplinary educational modules on machine learning and smart power grids. The smart grid technologies will be promoted among high school seniors by defining and providing mentorship for projects that intersect power systems and computer science. Talented students from under-represented groups in STEM will be actively engaged in the project through the Washington State University and University of Iowa mentorship engineering programs. Installing new sensors/meters at every node of the power distribution network, which may include thousands of nodes, is an expensive and a multi-year planning task. Also, the required sensors/meters redundancy for achieving reliable sensing platforms in facing possible failure or loss of sensors/meters cannot be fulfilled with such a scarce sensing infrastructure. Our proposed solution to this challenging real-world problem is analytical methodologies in the form of 'Virtual Meter'. The proposed "Virtual Meter" is not an actual physical device; rather it is a co-modeling paradigm that fuses data-driven and physics-based models in a closed loop setting with online bidirectional interactions. We propose a class of coherent, holistic, and feasible stream processing and online learning algorithms with provable quality guarantees and incur learning cost that enables such an online interaction, forging the co-modeling framework. First, we will create a class of ad-hoc data fusion algorithms that can exploit and extract reliable values from heterogeneous data streams. Second, the project will devise a class of online learning algorithms including online deep learning to estimate virtual measurements. The third major contribution of the project is that the proposed 'Virtual Meter' closes the loop of interactions between data-driven and physics-based models in an online setting creating a co-modeling framework to enhance the real-time monitoring of power distribution systems.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.

随着信息和通信技术在工程系统中的不断发展，数据的在线流变得可用。在线学习算法可以利用这些高价值数据来增强国家关键基础设施（如电网）的运行。 与输电网不同，配电系统缺乏通过传感基础设施进行的广泛的直接在线测量。这使得配电系统的准确监测成为一项具有挑战性的任务，这对于系统的可靠操作至关重要，特别是在具有间歇性可再生能源的大规模集成的配电系统中，间歇性可再生能源增加了聚合负载生成值的可变性。 拟议的研究能够对配电系统进行可靠的监测，并大规模整合可再生能源，对公众产生经济和社会影响。建议的在线优化技术，这将在这个项目中进行调查，可以应用到各种学习任务的数据流超越电力工程。研究和教学将通过开发机器学习和智能电网的跨学科教育模块来整合。智能电网技术将在高中高年级学生中推广，为电力系统和计算机科学交叉的项目定义和提供指导。 来自STEM代表性不足群体的有才华的学生将通过华盛顿州立大学和爱荷华州大学的导师工程项目积极参与该项目。在可能包括数千个节点的配电网络的每个节点处安装新的传感器/仪表是一项昂贵的多年规划任务。 此外，在面对传感器/仪表的可能故障或丢失时，用于实现可靠的感测平台所需的传感器/仪表冗余不能用这种稀缺的感测基础设施来实现。 我们提出的解决这个具有挑战性的现实世界的问题是分析方法的“虚拟仪表”的形式。所提出的“虚拟仪表”不是一个实际的物理设备;相反，它是一个共同建模的范例，融合数据驱动和基于物理的模型在一个闭环设置与在线双向交互。我们提出了一类连贯的，整体的，可行的流处理和在线学习算法，具有可证明的质量保证，并产生学习成本，使这样的在线交互，锻造共同建模框架。首先，我们将创建一类ad-hoc数据融合算法，可以利用和提取可靠的值从异构数据流。其次，该项目将设计一类在线学习算法，包括在线深度学习，以估计虚拟测量。该项目的第三个主要贡献是，拟议的“虚拟电表”在在线设置中关闭了数据驱动和基于物理的模型之间的交互循环，创建了一个共同建模框架，以增强配电系统的实时监控。该奖项反映了NSF的法定使命，并通过使用基金会的知识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

Communication-Efficient Distributed Stochastic AUC Maximization with Deep Neural Networks

• 发表时间: 2020-05
• 作者: Zhishuai Guo;Mingrui Liu;Zhuoning Yuan;Li Shen;Wei Liu;Tianbao Yang

An Online Method for A Class of Distributionally Robust Optimization with Non-convex Objectives

• 发表时间: 2020-06
• 作者: Qi Qi-Qi;Zhishuai Guo;Yi Xu;Rong Jin;Tianbao Yang

Stability and Generalization of Stochastic Gradient Methods for Minimax Problems

• 发表时间: 2021-05
• 作者: Yunwen Lei;Zhenhuan Yang;Tianbao Yang;Yiming Ying

A Decentralized Parallel Algorithm for Training Generative Adversarial Nets

• 发表时间: 2019-10
• 期刊: arXiv: Optimization and Control
• 作者: Mingrui Liu;Wei Zhang;Youssef Mroueh;Xiaodong Cui;Jarret Ross;Tianbao Yang;Payel Das

Optimal Epoch Stochastic Gradient Descent Ascent Methods for Min-Max Optimization

• 发表时间: 2020-02
• 期刊: arXiv: Optimization and Control
• 作者: Yan Yan-Yan;Yi Xu;Qihang Lin;W. Liu;Tianbao Yang