GNN4GC—Graph Neural Networks for Grid Control
摘要
The increasing electrification of various sectors and industries and challenges of the energy transition presents complex tasks for transmission system operators. Transmission systems serve as the core infrastructure for transmission electricity across regions, nations, and even continents, playing a crucial role in ensuring a steady and save power supply. However, operators face significant challenges due to the increasing power demand and the fact that the current speed of network expansion cannot keep up with the changed market situation. Additionally, the fluctuating power generation from renewable energy causes congestion in the transmission grid and overloading of transmission lines. Currently, operators resort to measures like redispatch to alleviate these issues, which proves to be unfavorable due to high expenses and regulatory challenges. The GNN4GC (Graph Neural Networks for Grid Control) project, a collaborative effort by Fraunhofer IEE, the University of Kassel, and transmission system operators (TenneT TSO GmbH, 50Hertz, TenneT TSO B.V), aims to address these challenges by leveraging Artificial Intelligence, particularly Graph Neural Networks (GNNs) and Deep Reinforcement Learning (DRL). The goal of GNN4GC is to develop GNN models to optimize load flow calculations and subsequently select optimal topology configurations. By combining GNNs with DRL, the project aims to create self-learning agents capable of recommending optimal topology changes, enhancing grid control and efficiency to support network operators with specific action recommendations. GNNs are particularly well-suited for this task due to their compatibility with network structures, enabling rapid grid calculations and power flow approximations [