Lawrence Berkeley National Laboratory

Researching the dynamics of residential electricity consumption at finely-resolved timescales is increasingly practical with the growing availability of high-resolution data and analytical methods to characterize them. One methodological approach that is popular for exploring consumption dynamics is load profile clustering. Despite an abundance of available algorithmic techniques, clustering load profiles is challenging because clustering methods do not always capture the temporal aspects of electricity consumption and because clusters are difficult to explain without additional descriptive household data. These challenges limit the use of cluster analysis to better understand behavioral and other drivers of electricity usage patterns. We address these challenges by applying a novel clustering approach to a unique data set of high-resolution electricity and occupant time-use data from UK households. We cluster cumulative rather than raw load profiles to capture their full shape. Our clustering approach identifies two distinct patterns of electricity consumption during evening weekdays (5–9 p.m.), which are primarily differentiated by the timing of their peak demand. Next, we apply several classification algorithms to assess the potential for using time-use activity data to predict membership in these distinct usage clusters. The methods we use are suited to this predictive modeling context and are able to identify key activities driving patterns of electricity demand. We discuss how such an approach can inform more targeted strategies for residential peak demand reduction and response interventions as well as improve our understanding of constraints and opportunities for demand-side flexibility in the residential sector.