Topology detection of distribution grid using wavelet-based Graph Attention Network

The distribution grid (DG) adjusts its topology to control power flow or isolate faults. Accurate detection of the DG’s topology is vital for many advanced applications such as power flow analysis, reconfiguration verification, etc. However, a large number of switches and lines widely distributed within the grid make it impractical, due to the high cost, to install devices for monitoring the topology. Furthermore, existing methods for topology detection overlook the interdependence among bus voltages and the information contained in the high-frequency band of these voltages, which reduces the accuracy of detection. To overcome these challenges, this paper proposes a wavelet-based Graph Attention Network (wavelet-GAT) approach that detects the topology based on voltage magnitude data from the smart meter of each bus. First, the node graph of the DG is transformed into a line graph to enable lines to aggregate information from their neighboring lines. Then, the wavelet transform is employed to extract low- and high-frequency components, which are used as features to the wavelet-GAT and for computing its attention coefficients, respectively. The method was tested on IEEE 33-bus, 118-bus, and 367-bus DGs, achieving detection accuracies of 99.01%, 96.57%, and 94.76% respectively. These results demonstrate the efficacy and robustness of the proposed method when compared with several state-of-the-art methods.