Load profiling using a two-step clustering framework: Capturing electricity usage variability based on temporal and ambient factors
Carl Flygare, Marcus Nystrand, Robert Eriksson, Valeria Castellucci · International Journal of Electrical Power & Energy Systems · 2025
Load profiles derived from the growing amount of smart meter electricity consumption data are increasingly being sought. They can be valuable to public entities and grid operators for purposes such as operational planning, analyzing the co-location of users with smoothing profiles, dimensioning distribution grids, procuring distributed energy resources (DER) such as batteries or PV arrays, and implementing demand response projects. The challenge lies in analyzing the vast amount of available data in ways that make it useful — preferably to a broader audience. This article presents a straightforward and adaptable framework for creating typical load profiles (TLPs). The framework applies two normalizations across two clustering steps: k -Means, to identify prominent patterns in specific data subsets, and k -Modes, to recombine the identified subset patterns into a yearly-coherent and re-scalable TLP in an innovative way. The framework is demonstrated using real data of roughly 60,000 daily time series from around 40 public elementary schools in Uppsala Municipality, Sweden. The results showed that the framework could efficiently differentiate the load behaviors of the studied users over the year. Their largest load differences occurred midday and during weekends, although most studied users exhibited a similar behavior. The load magnitude was also shown to have a usable linear relationship with the schools’ heated indoor area, enabling the use of TLPs to estimate the load of a new arbitrary school. Due to its design and versatility, the presented framework can serve as a valuable tool for identifying prominent patterns and supporting relevant decision-making processes.