Identification of low voltage distribution transformer–customer connectivity based on unsupervised learning

Abstract

The increased penetration of smart meters in the low voltage distribution network provides valuable insights that distribution network operators can leverage to produce detailed analytics for decision making. This includes enhanced network observability that can support network planning, phase balancing and planned outage management. In this paper, we propose three unsupervised learning algorithms to model the problem of identifying the transformer–customer connectivity relation at the low voltage network using smart meter voltage measurements of residential units as a classification problem. The algorithms are tested on real-world smart meter datasets and can detect the correct transformer–customer connectivity with 95% to 100% accuracy. This allows distribution network operators to effectively detect inconsistencies between customer data reports and actual physical connection of residential units to distribution transformers to comply with legal regulations.