CAREER: Physics-informed Graph Learning for Anomaly Detection in Power Systems

这个NSF CAREER项目旨在改善电力系统运营商的状态估计和异常检测问题。该项目将为运营商检测系统状态和异常事件（包括网络攻击）的方式带来革命性的变化。这将通过使用自适应图形和物理方法来实现。该项目的智力优势包括：1）开发用于状态估计和异常检测的时空图神经网络（ST-GNN），2）将物理学集成到ST-GNN中，3）IT/OT事件的相关性，以及4）大规模状态估计和异常检测的开发。该项目更广泛的影响包括通过克拉克森的地平线计划激发未来女工程师对可持续能源和网络物理安全的兴趣，支持研究中代表性不足的本科生，加强国家实验室和大学之间的合作，并通过每年为公用事业学生举办的研讨会提高对公用事业的认识。 及时识别和缓解新出现的网络物理系统（CPS）风险需要电网运营商采用更复杂和更有弹性的状态估计和异常检测工具。该CAREER项目的主要目标是通过自适应基于图形和基于物理的方法来提高电力系统运营商的状态估计和异常检测能力。这一目标将通过精心制定的工作计划来实现，该计划旨在（1）使用图形上的物理信息学习来提高状态估计的精度和鲁棒性，以及（2）通过混合和分布式方法增强异常检测，同时实现可扩展性。这些目标将通过四项关键研究活动来实现：1）开发时空图神经网络以增强状态估计，2）建立物理信息框架以处理不平衡数据，3）将信息技术（IT）数据与操作技术（OT）数据集成以支持异常检测，以及4）创建分布式方法以实现可扩展的状态估计和异常检测。在教育方面，目标是通过鼓励妇女参与该领域，培训下一代电力工程师，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。

This NSF CAREER project aims to improve state estimation and anomaly detection problems for power system operators. The project will bring transformative changes in how operators detect the system states and abnormal events, including cyber-attacks. This will be achieved by using adaptive graph-based and physics-based methods. The intellectual merits of the project include 1) Development of spatial-temporal graph neural networks (ST-GNN) for state estimation and anomaly detection, 2) Integration of physics into the ST-GNN, 3) Correlation of IT/OT events, and 4) Development of state estimation and anomaly detection at scale. The broader impacts of the project include spurring the interests of future female engineers in sustainable energy and cyber-physical security through Clarkson's Horizons Program, supporting underrepresented undergraduates in research, enhancing collaboration between national labs and universities, and enhancing awareness of utilities through annual workshops for utility students. Timely identification and mitigation of emerging cyber-physical system (CPS) risks necessitate more sophisticated and resilient tools for state estimation and anomaly detection employed by grid operators. The primary objective of this CAREER project is to improve the state estimation and anomaly detection capabilities of power system operators through adaptive graph-based and physics-based methods. This objective will be realized through a meticulously crafted work plan aimed at (1) improving the precision and robustness of state estimation using physics-informed learning on graphs and (2) enhancing anomaly detection while achieving scalability through hybrid and distributed approaches. These goals will be pursued through four key research activities: 1) Developing spatial-temporal graph neural networks for enhanced state estimation, 2) Establishing a physics-informed framework to handle imbalanced data, 3) Integrating information technology (IT) data with operational technology (OT) data to bolster anomaly detection, and 4) Creating distributed methods for achieving scalable state estimation and anomaly detection. On the educational front, the aim is to broaden and strengthen the power engineering workforce by encouraging women to participate in the field, training the next generation of power engineers, and elevating the skills and awareness of current power engineers to effectively navigate emerging CPS threats in power system operations.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.