Clustering household electricity and gas load profiles in the UK using smart meter and survey data

Conference item

The transition to Net Zero energy systems requires a comprehensive understanding of household energy consumption patterns. This understanding is crucial for developing simulation models that accurately reflect real-world building energy consumption and inform the design and implementation of effective energy policies and demand-side management interventions. This paper presents a real-world dataset collected from 139 UK households over a three-year period, consisting of smart meter and socio-demographic survey data. We apply machine learning techniques, including the Kmeans clustering algorithm, to identify and analyze distinct clusters for household gas and electricity load profiles. We also developed an XGBoost machine learning model to learn the relationship between household socio-demographics, building characteristics, and energy savings habits, using the model to interpret the clusters. Our findings reveal that the households' self-reported annual energy spend (i.e., utility bills) and the number of rooms (i.e., house size) are the two most important variables for inferring the households’ electricity and gas clusters. Additionally, household annual income, specific housing types (i.e., Bungalow), and certain energy-saving habits also play smaller but significant roles during cluster assignment. The results offer a granular understanding of household energy consumption in the UK, facilitating the development of targeted energy policies and personalized demand response strategies. This research underscores the importance of integrating large-scale data acquisition with advanced analytics to support the implementation of equitable and effective Net Zero policies, ultimately contributing to smarter and more resilient urban energy systems.

Authors: Tekler, ZD

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: International Building Performance Simulation Association
• Host title: Proceedings of uSim Conference 2024: 4th uSim Conference of IBPSA-Scotland
• Article number: 3A-1
• Series: uSim Conference Proceedings
• Series number: 4
• Publication date: 2024-11-25
• Acceptance date: 2024-11-25
• Event title: uSIM2024 – Shaping Net Zero Policies with Building Simulation
• Event location: Edinburgh, UK
• Event website: https://eng.ed.ac.uk/about/events/20241125-0900/usim2024-ibpsa-scotlands-biennial-usim-conference
• Event start date: 2024-11-25
• Event end date: 2024-11-25
• EISBN: 9781914241833
• Language: English
• Keywords: electricity and gas load; household consumption; large-scale smart meter data; machine learning; clustering