Data-driven Load Estimation in Low-Voltage Networks Based on Limited Smart Meter Coverage

Improving observability in low-voltage networks is critical for enabling the integration of low-carbon technologies and ensuring efficient network operation. This paper presents a data-driven methodology for improving LV network (LVN) observability, assuming only partial adoption of smart meters (SMs) by end users. The methodology uses a machine learning (ML) technique to estimate the load consumption of nodes without SMs, using SM data at nodes with known SM measurements. Subsequently, a probabilistic model is employed to refine these estimations, accounting for uncertainties and variability in consumption patterns. The effectiveness of the method is validated through state estimation (SE), which utilizes realistic LV distribution network with known topology, appropriate network model, and available SM measurements, including active power (P) and voltage magnitude (V). Results demonstrate that the proposed framework significantly enhances the accuracy and reliability of estimation of load consumption, providing a practical solution for improving LVN observability under limited measurement conditions.