A Multi-View Clustering based Dynamic Partitioning Method for Distribution Network

Abstract

Aiming at solving the division problem of the area-centralized layout in the power distribution network, a dynamic partitioning method of distribution network area based on a multi-view clustering algorithm is proposed. Firstly, a mathematical model is established to calculate the optimal number of clusters considering communication quality and communication cost. Secondly, the Laplacian matrix of distribution network structure and other perspectives, such as the geographic location and the administrative area of distribution network stations are introduced to the distribution network area division by multi-view clustering. Thirdly, one of the stations is selected as the edge computing center to ensure efficient edge computing by combining the clustering center and the actual situation. Finally, the proposed method realizes the effective partitioning of the distribution network and the automatic area adjustment when the structure of distribution network changes. Based on the network structure calculation of 145 stations in a local distribution network, the experimental simulation results verify that the proposed partitioning method is practical and feasible.