EAGER-DynamicData: A Hierarchical Approach to Dynamic Big Data Analysis in Power Infrastructure Security

这项研究将解决一个与国家紧急利益相关的动态大数据问题：需要有效的方法来诊断电力网络等关键互联基础设施中的故障和攻击。此外，该项目还将研究从复杂数据流中提取知识的新方法，这些数据流来自基础设施系统中的各种传感器及其行为模型。该项目的结果和发现将通过行业认可的电力系统仿真器进行验证，并将有助于电力行业提高基本电力基础设施的安全性、稳定性和安全性。该项目将促进多学科研究，涉及大数据分析、机器学习、安全、电力系统和控制系统方面的专业知识。这项研究将在理论和实际应用之间架起一座强大的桥梁，同时成为加州大学河滨分校多样化的新一代工程师的培训平台，加州大学河滨分校是美国种族最多元化的研究密集型机构之一。该项目将促进使用多分辨率数据驱动方法，以电力网络为重点，对关键动力基础设施中的异常情况进行检测和分类。该项目有三个新颖、创新和潜在的变革性技术要素：(1)一个全面的统计模型，作为现有基于物理的模型的替代，使用动态贝叶斯网络和条件随机场对容易发生故障和攻击的复杂基础设施进行建模；(2)基于机器学习概念的分层检测和分类方法，以驯服和利用由空间分布的传感器收集的海量和多样性的动态多分辨率数据；(3)利用电力系统和控制理论中基于模型和分析的知识来训练和告知数据驱动方法的系统方法，以建立可扩展的电力基础设施安全检测和分类机制。

This research will address a dynamic big data problem that is of urgent national interest: the need for efficient methods to diagnose faults and attacks in critical interconnected infrastructures, such as electricity power networks. Additionally, this project will investigate new methodologies to extract knowledge from the complex streams of data that come from various sensors in infrastructure systems and the models of their behavior. Results and findings in this project will be validated via industry-accredited power system simulators, and will be useful to the power industry in enhancing the safety, stability, and security of essential power infrastructure. This project will promote multi-disciplinary research involving expertise in big data analysis, machine learning, security, power systems, and control systems. This research will provide a powerful bridge between theory and real-world applications while serving as a training platform for a diverse new generation of engineers at the University of California, Riverside, one of America's most ethnically diverse research-intensive institutions. This project will foster the use of multi-resolution data-driven methods for the detection and classification of anomalies in critical dynamical infrastructures, with focus on power networks. This project has three novel, innovative, and potentially transformative technical elements: (1) A comprehensive statistical model, as an alternative to existing physics-based models, using Dynamic Bayesian Networks and Conditional Random Fields to model complex infrastructures subject to failures and attacks; (2) A hierarchical detection and classification method based upon machine learning concepts to tame and leverage the vast amount and diversity of dynamic multi-resolution data collected by spatially distributed sensors; (3) A systematic method to train and inform data-driven methodologies from model-based and analytical knowledge that come from power systems and control theory to build scalable and performing detection and classification mechanisms in power infrastructure security.