Graph Neural Network Based Deep Reinforcement Learning for Volt - Var Control in Distribution Grids

Auteurs

Ma A., Cao J., Cortes P.R.

Référence

2024 IEEE 15th International Symposium on Power Electronics for Distributed Generation Systems, PEDG 2024, 2024

Description

Voltage reactive power control (VVC) plays a vital role in future-oriented smart grids, which not only ensures the stable operation of the power system but also optimizes energy distribution and consumption. However, with the widespread access to distributed renewable power generation resources, the structure and operation mode of the power grid, especially the distribution network, are becoming more and more complex. The traditional VV C method is difficult to cope with this new situation due to its inherent limitations. challenge. These limitations include, but are not limited to, the static nature of the control strategy, lack of adaptability to dynamic changes in the grid, and computational inefficiency when dealing with large-scale networks. With the widespread use of artificial intelligence in smart grids, some deep-reinforcement learning algorithms have been used to solve this problem. However, because the algorithm is not good enough for controlling complex environments, in order to overcome these challenges, this paper proposes innovative methods based on deep reinforcement learning (DRL) and graph neural networks (GNN). GNN is used to extract graphical representations from topological maps of power systems and is trained with data from the VVC environment. Improve the efficiency and reliability of VV C by improving algorithms and optimization strategies.

Lien

doi:10.1109/PEDG61800.2024.10667392