Distributed Auto-Clustering for Residential Load Profiling Using AMI Data From the U.S. High Plains

Abstract

Managing demand and building customized services are essential for power retailers, distribution network operators (DNOs), and load-serving entities (LSEs) in a competitive retail environment. Therefore, understanding residential customer power demand behaviors is essential for accurate load profiling. However, the large volume of fine-grained smart meter data and the wide variety of residential behavior make load profile extraction extremely challenging, even further exacerbated by deterministic clustering’s inherent limitations. This paper develops a distributed auto-clustering approach (DAA) for load profiling using real smart meter data based on a divide-and-conquer scheme to tackle current load profiling challenges. To improve the load profiling accuracy, temperature-based categories (TBCs) are proposed to distinguish days with similar temperatures. A Heterogeneous Distributed Local Sites (HDLS) scheme is developed based on the raster map concept to distribute smart meter data among local sites while considering their size and density. Local sites send the load representatives to a central site for global clustering to significantly reduce clustering complexity and computational burden. To verify the proposed DAA, approximately 11,000 residential 15-minute smart meter data provided by a utility company from the U.S. high plains were utilized in this paper.