CRII: SaTC: GridAI: Physics-based Machine Learning Models and Algorithms for Attack-Resilient Smart Grid Infrastructure

保护网络网络物理系统，如电网，已经成为国家和经济安全的头等大事。攻击者利用网络和物理领域的漏洞，部署隐蔽的多级网络攻击，破坏智能电网的可靠运行。该项目将利用基于物理的机器学习和深度学习算法来构建具有攻击弹性的网络物理智能电网基础设施，旨在为智能电网提供先进水平的网络物理系统安全性，同时增强系统的弹性和可靠性。该项目利用基于物理的机器学习和深度学习算法来检测数据完整性、拒绝服务和分布式拒绝服务网络攻击。该项目还对研究生和本科生进行多学科教育，包括代表性不足的少数民族。该项目的两个主要目标是：i)使用基于物理的机器学习算法检测测量和控制信号的异常情况，以防止网络物理智能电网应用中的数据完整性网络攻击，并随后为系统运营商制定合适的缓解策略。ii)利用基于物理的深度学习算法重构正常和扰动电网条件下拒绝服务和分布式拒绝服务网络攻击导致的缺失数据，确保为网络物理智能电网应用提供可靠和连续的数据框架。提出的基于物理的机器学习和深度学习攻击弹性算法将使用硬件在环云基础设施进行设计、训练、测试和评估。该项目将利用与公用事业公司的合作，培训和推断实用的机器学习和深度学习模型。该奖项反映了美国国家科学基金会的法定使命，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。

Securing networked cyber-physical systems, such as an electric power grid, has emerged as paramount to national and economic security. Adversaries exploit the vulnerabilities in both cyber and physical domains and deploy stealthy multistage cyberattacks to disrupt the smart grid’s reliable operation. This project will leverage physics-based Machine Learning and Deep Learning algorithms to build an attack-resilient cyber-physical smart grid infrastructure, aiming for advanced levels of cyber-physical system security for the smart grid while simultaneously enhancing the system resiliency and reliability. The project leverages physics-based Machine Learning and Deep Learning algorithms to detect data-integrity, denial-of-service, and distributed denial-of-service cyberattacks. The project likewise pursues multidisciplinary education of graduate and undergraduate students, including underrepresented minorities.The project’s two primary goals are -- i) Detecting anomalies on both measurement and control signals using physics-based machine learning algorithms against data-integrity cyber attacks for cyber-physical smart grid applications, and subsequently develop suitable mitigation strategies for system operators. ii) Reconstructing the missing data due to denial-of-service and distributed denial-of-service cyber attacks under both normal and perturbed grid conditions using physics-based deep learning algorithms to ensure reliable and continuous data frames provided for cyber-physical smart grid applications. The proposed physics-based machine learning and deep learning attack-resiliency algorithms will be designed, trained, tested, and evaluated using hardware-in-the-loop cloud infrastructure. The project will leverages collaboration with utility companies for training and inferring practical machine learning and deep learning models.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.