A voltage sag homology detection method based on the Siamese network

• A voltage sag fault will trigger waveform recording at multiple locations. • The propagation form of voltage sag changes after passing through the transformer. • Identifying sag data from one fault as same-source data is voltage sag source detection. • Build a deep model to learn sag transmission patterns through the transformer. • Innovated clustering algorithms to efficiently complete homologous detection. A voltage sag may change its shape as it propagates through a power grid, especially after it crosses transformers. The traditional method based on the transformer transfer matrix to solve the homology problem of voltage sag is generally complicated so it cannot be applied in actual scenarios. Therefore, this paper proposes a voltage sag homology detection method that uses the Siamese network to extract the homology correlation of the recorded voltage sag data, and the waveform type is obtained by combining the graph construction algorithm.Then uses the waveform type, start time, and location of the recorded voltage sag as the input to an improved DBSCAN to obtain the final clustering result for homology detection. Experiments show that compared with other voltage sag homology detection methods, the proposed method can improve the robustness and accuracy of voltage sag homology detection, thus can be applied in complex scenarios in real-life applications.