Using Real Smart Meter Data to Construct Three-Phase Low Voltage Network Models

Abstract

The increasing uptake of residential solar PV and electric vehicles calls for advanced planning and operational techniques to ensure the integrity of low voltage (LV) distribution networks. This, however, requires distribution companies to run power flow analyses and, therefore, to have adequate three-phase LV network models. Although, in practice, these models are often inaccurate or do not exist, the growing adoption of smart meters around the world brings about new opportunities. This paper proposes a comprehensive approach to construct a three-phase LV network model solely exploiting real smart meter data. Two multiple linear regression models combined with clustering techniques and Spearman correlation are employed to identify the network topology, line impedance, and phase groups. The proposed approach is tested on a real LV network (14 customers) from Victoria, Australia, and a set of 60 days of real smart meter data (${\bm{V}},{\bm{\ I}}$) with 5-min resolution. Results demonstrate that the construction of the LV network model can be done with a mean error of $ \pm $1 Volts (measurements vs power flow simulations). This suggests that the proposed smart meter-driven approach can help distribution companies to construct accurate models for any planning or operational application.