Characterising temporal aspects of residential electricity consumption using statistical learning

Achieving ambitious climate change mitigation targets requires a comprehensive transformation of the energy system. As part of this transformation, early, rapid, and full decarbonisation of the electricity sector is essential. Residential buildings contribute substantially to electricity consumption, so pathways for electricity system decarbonisation similarly depend on accelerated change in the residential sector. Delivering further and faster reductions in emissions from residential buildings requires not only reductions in electricity demand but also more responsive demand. Flexibility in the residential sector can support the supply-side transition to renewable energy while also helping reduce the costs of electricity provision during times of peak demand. Understanding the potential contributions of residential buildings to ambitious mitigation pathways through demand reduction and flexibility requires more detailed characterisations of demand. In particular, it necessitates stronger evidence on the structural and occupant factors that shape electricity demand and its resulting emissions. This thesis contributes to these research areas and is guided by three primary research questions. The first focuses on improving methods for characterising residential electricity consumption, the second considers advancing our knowledge of factors influencing consumption, and the third relates to the policy implications of having deeper insight into the temporal aspects of demand. The three research questions are: • What statistical learning approaches can improve efforts to characterise residential electricity consumption patterns? • What factors explain electricity consumption in residential buildings at annual, daily, and hourly temporal resolutions? • How does a more detailed understanding of the temporal aspects of residential electricity consumption inform energy efficiency and flexibility interventions? This thesis consists of four original research papers, each of which makes methodological, empirical, and policy contributions in answering these research questions. The papers’ methodological contributions are the development and application of methods from statistical learning and the sub-fields of regularisation and sparsity. These techniques are the subject of a rich literature in statistics disciplines but are not commonly used in energy research. The papers show the utility of these techniques for analysing complex data sets and addressing challenges inherent to building energy modelling. The papers make empirical contributions to the literature on determinants of electricity consumption through statistical analyses of factors that influence consumption at different temporal resolutions in households. The variety in types of data considered, such as interval electricity, household-level survey, and time-use data, facilitates a deep investigation of the factors that shape residential electricity consumption. In this way, the papers contribute to the evidence base on effective intervention policies for achieving further demand reduction and flexibility in the residential sector. Demand-side solutions are underrepresented in the literature on climate change mitigation, but these solutions have considerable potential to achieve ambitious emissions reduction targets. They therefore require a more thorough assessment. This thesis research presents new analytical methods, data, and evidence to investigate how residential buildings can contribute to accelerated decarbonisation pathways.