Abstract:Aiming at the problem of inaccurate identification of user-transformer relationship in distribution networks，which affects the normal operation of distribution networks and fault diagnosis and localisation，a multivariate data-driven user-transformer relationship identification method based on multivariate data is proposed. Firstly，the multidimensional fluctuation characteristic parameter index of voltage is defined，secondly， the t-distributed stochastic neighbourhood embedding （t-SNE） algorithm is used to downscale the voltage fluctuation characteristic matrix，and then the BIRCH clustering algorithm is used to cluster the downsized voltage fluctuation characteristics to form virtual users，and finally，based on the assumption of the user’s power balance relationship with the higher-level meter，user-transformer relationship identification is carried out，and user-transformer relationship identification is completed in the distribution network. The results show that compared with the previous identification methods，the identification accuracy is improved.