Project Grant 2154171

This $200,000 National Science Foundation project grant supports the development of data-driven power systems control with stability guarantees. Funded through the NSF Engineering program (CFDA 47.041), the award to Carnegie Mellon University will support three thrusts of collaborative research over a 30-month period ending February 2025.

The research aims to design a new framework integrating reinforcement learning algorithms with Lyapunov stability theory to provide stability guarantees for learning-based controllers applied to frequency and voltage control in power grids. Specifically, the work will leverage analytical models, machine learning, and neural networks to discover Lyapunov functions, engineer controller structures that ensure stability, and develop robust control policies accounting for modeling errors and network changes. The framework seeks to clear a hurdle to applying reinforcement learning to power systems by guaranteeing the stability of learned policies, with potential to serve as the basis for future smart grid control architectures.