Selecting Representative Net Load Profiles of Solar Homes With K-Means Clustering

Abstract

Increasing numbers of residential solar installations create challenges for distribution network management in terms of voltage management and feeder hosting capacity. As smart meter technology proliferates, large datasets become available to describe residential loads and PV generation on the network. This paper details a process by which solar generation and load data can be used to select representative residential net load profiles for use in economic and technical analyses of residential solar systems while retaining representation of the variance in net load profiles across a large dataset. In doing so, it is acknowledged that a single average net load response is insufficient to observe the behaviour of numerous different residential PV systems. The method clusters net load data of solar homes to create a set of synthetic profiles to which existing systems are matched. K-means, agglomerative hierarchical clustering, and self-organising map (SOM) clustering approaches were considered, with a silhouette analysis showing k-means had superior clustering quality for this dataset and was therefore used to generate the synthetic net load profiles. The representative profiles selected through this process had normalised mean-absolute errors of 4% on average, and normalised root-mean-square errors of 5% on average when comparing them to their corresponding synthetic profiles. This indicates net load profile shape information is retained through the representative system selection process. Since existing literature does not address the problem of finding representative solar home net load profiles, our study’s findings can be used as a point of reference for future work in this area.