A GMM-Based Phase Group Identification for Residential Low Voltage Networks

Having adequate low voltage (LV) network electrical models is the first step for distribution companies to be able to assess the effects of distributed energy resources (DER) as well as potential DER management solutions. One challenging part, however, is the identification of the phase group to which customers are connected to. Fortunately, the widespread adoption of smart meters combined with machine learning algorithms opens new avenues. This paper proposes a clustering technique, based on the Principal Component Analysis (PCA) method and Gaussian Mixture Model (GMM), that only employs historical smart meter voltage measurements of customers. The effectiveness of the proposed approach is demonstrated using four realistic Australian LV networks with dozens of single-phase residential customers and solar photovoltaic (PV) systems. The findings show that the approach can identify all customer phase groups correctly.