基于双尺度相似性和改进DBSCAN算法的低压配电台区相序识别方法OA北大核心CSTPCD

To address the problem of low-quality smart meter collection data in low-voltage distribution station areas and poor phase identification results owing to inconspicuous differences in voltage similarity among users,a phase identification method based on dual-scale similarity and improved density-based spatial clustering of applications with noise (DBSCAN) for low-voltage distribution station areas is proposed. First,a data processing method based on matrix completion is proposed to address the negative impact of missing user data on the accuracy of phase sequence identification in low-voltage distribution station areas. Second,a dual-scale similarity metric is proposed,which first adopts the Euclidean distance to measure the overall distribution characteristics of voltage curves and subsequently uses the first-order difference and the dynamic time warping (DTW) distance to measure the local dynamic characteristics of the voltage curves,thereby improving the similarity measure of DBSCAN in the whole and local contexts and mitigating the issue of a high misjudgment rate when the voltage curves are close to each other inherent to the conventional method. Finally,the sparrow search algorithm (SSA) is used to identify the optimal initial DBSCAN parameters,which improves the robustness of the algorithm. Simulation experiments show that matrix complementation improves 1. 6 to 2. 3 times in accuracy relative to conventional interpolation complementation. Using dual-scale similarity with the improved DBSCAN algorithm,100％ of the phase sequences of all the users in the station area can be identified. The application of the SSA to determine the optimal values for Eps and MinPts enables DBSCAN to obtain the best evaluation index and effectively addresses the impediments involved in selecting the initial parameters manually.