Robust learning-based real-time load estimation using sparsely deployed smart meters with high reporting rates

Abstract

With increasing renewable generation and demand response, the load profiles of distribution feeders become more fluctuating and uncertain, requiring real-time load estimation (RTLE) with high temporal granularity. Smart meters (SM) provide new data sources that have the potential to enable RTLE. However, it is cost prohibitive to communicate and process real-time high-resolution data from a massive number of SMs. To address the challenge, this paper proposes a novel solution to RTLE using High-Reporting-Rate SMs (HRRSMs) installed at a sparsely selected subset of customers in the feeder. The first step is to select customers for installing HRRSMs based on clustering, such that load profiles can best represent those of the others and the whole feeder. Then, a state-of-the-art Deep Learning (DL) model is trained to capture the relation between the historical load profiles of the selected customers and that of the feeder. Finally, real-time HRRSM data from the selective customers is fed to the trained model to perform RTLE with high resolution. The method is also robustified to address anomalies in real-time HRRSM data streams. The proposed method is validated on a large real-world SM dataset. Simulation results show that even with a small number of HRRSM installation, the proposed method can track feeder loads with much improved accuracy and temporal granularity compared with conventional methods based on historical data of regular SMs, providing a cost-effective solution to the monitoring of distribution feeder loads.