Paper ID: 2407.04522

Graph Reinforcement Learning for Power Grids: A Comprehensive Survey

Mohamed Hassouna, Clara Holzhüter, Pawel Lytaev, Josephine Thomas, Bernhard Sick, Christoph Scholz

The rise of renewable energy and distributed generation requires new approaches to overcome the limitations of traditional methods. In this context, Graph Neural Networks are promising due to their ability to learn from graph-structured data. Combined with Reinforcement Learning, they can serve as control approaches to determine remedial network actions. This review analyses how Graph Reinforcement Learning (GRL) can improve representation learning and decision making in power grid use cases. Although GRL has demonstrated adaptability to unpredictable events and noisy data, it is primarily at a proof-of-concept stage. We highlight open challenges and limitations with respect to real-world applications.

Submitted: Jul 5, 2024