Topology Identification of Distribution Networks Based on Physics-Informed Latent Graph Attention Network

In modern society, the distribution network plays a critical role as one of the essential energy infrastructure systems. Its stable and secure operation is vital for ensuring reliable electricity consumption. Therefore, it is crucial to develop a method that accurately estimates the topology of the distribution network. This paper presents an intelligent approach for recognizing the topology of the distribution network, utilizing the Physics-Informed Latent Graph Attention Network (PLGAN). The PLGAN incorporates attention weights during the node aggregation process, providing a powerful and flexible framework for handling graph-structured data. The proposed method begins by abstracting the distribution network topology into a line graph representation. Then, the abstracted line graph’s adjacency matrix and node features are used to train the PLGAN, resulting in a distribution network topology recognition model. To validate the effectiveness, superiority, and robustness of the approach, extensive case studies are conducted on an enhanced IEEE-33 node system. The results demonstrate the method’s capability to accurately identify distribution network topologies using only time-sectional measurement data, making it applicable to both radial and looped networks. Specifically, our proposed method achieves a high accuracy of 98.15% in identifying the distribution network topology. Furthermore, the advantages of PLGAN in distribution network topology recognition are further confirmed through a comparison with existing methods. The experimental results reveal that PLGAN surpasses other techniques in terms of accuracy, convergence speed, and stability.