Abstract

The rise of renewable energy and distributed generation requires newapproaches to overcome the limitations of traditional methods. In this context,Graph Neural Networks are promising due to their ability to learn fromgraph-structured data. Combined with Reinforcement Learning, they can serve ascontrol approaches to determine remedial network actions. This review analyseshow Graph Reinforcement Learning (GRL) can improve representation learning anddecision making in power grid use cases. Although GRL has demonstratedadaptability to unpredictable events and noisy data, it is primarily at aproof-of-concept stage. We highlight open challenges and limitations withrespect to real-world applications.

 

Quick Read (beta)