ERI: Scalable Machine Learning Frameworks for Stability Enhancement in Inverter-Dominated Power Systems

This NSF ERI project aims to enhance the stability and resilience of modern power systems as they increasingly rely on inverter-based resources, such as distributed power generation and battery energy storage systems. While these technologies are essential for a sustainable energy future, they introduce fast and complex dynamics that make power grids more difficult to monitor and control. Traditional analysis tools are no longer sufficient due to limited visibility into how these devices operate and interact. The project will bring transformative changes by developing intelligent, data-driven methods that can estimate system behavior in real time and recommend actions to prevent instability. This will be achieved by combining machine learning with physical knowledge of power systems to create scalable and privacy-preserving solutions. The intellectual merit of the project includes the development of new learning frameworks that integrate physics-based modeling with advanced artificial intelligence for real-time system monitoring and control. The broader impacts of the project include improving national energy security by reducing the risk of large-scale power outages, enabling safer and more reliable grid operation, supporting workforce development through student training, and providing open-source tools for utilities and researchers. The project addresses two core technical challenges: accurate system modeling under limited data availability and real-time decision-making for stability enhancement. First, it develops a hybrid framework that combines physics-informed neural networks, which are machine learning models guided by physical laws, with federated learning, a distributed training approach that allows multiple devices to collaboratively build models without sharing sensitive data. This enables scalable, real-time estimation of system impedance, a key indicator of stability, using only local measurements. Second, the project designs a deep reinforcement learning framework, specifically based on the soft actor-critic algorithm, to learn optimal control strategies that improve system stability. The framework uses stability-related metrics, including damping and participation factors, to guide safe and effective decision-making. These methods will be validated using high-fidelity hardware-in-the-loop platforms that emulate real-world grid conditions. The expected contributions include a unified framework for real-time monitoring and control, improved stability margins in inverter-dominated systems, and practical tools for deployment in next-generation power grids. 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. NSF Award ID: 2552448 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: VAN HAI BUI | Institution: Regents of the University of Michigan - Dearborn, Dearborn, MI | Award Amount: $199,984 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2552448 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2552448.html