Topology Identification in LV Distribution Network Based on SAE and Modified DBSCAN

The mistakes in topology records of low voltage distribution network influence the economic operation of the power grid. However, most existing methods to detect the customers who do not match the records are not always reliable and can’t realize automatic identification. This paper proposes an identification method based on stacked autoencoder (SAE) and modified Density-Based Spatial Clustering of Applications with Noise (DBSCAN). SAE is applied to reduce the data dimensions and extract nonlinear feature vectors from voltage time series. The modified DBSCAN can adjust parameters adaptively and cluster the feature vectors to identify the customers with wrong transformer connectivity, which improves the automation management level in distribution network. Case studies on real datasets have verified that the proposed method has high identification accuracy.