Decomposing and Analysing Hot Water Usage in an Icelandic District Heating System Using Hourly Smart Meter Data

This thesis analyses 398 days of hourly hot water smart meter data from 10,391 properties in the Reykjavík Capital Area to quantify how hot water consumption from a District Heating System (DHS) is divided between heating and non-heating uses, and how these components vary with building and behavioural factors. A key contribution is a scalable decomposition method capable of estimating heating and non-heating usage from total usage without labelled data or detailed building data. The method, based on a moving-average classifier, produced outputs consistent with known physical patterns - heating usage showed a strong negative correlation with outside temperature, while non-heating usage showed almost none. Outputs were consistent with operational assumptions and engineering estimates currently used by the DHS operator, including that 85-95% of total hot water usage is for heating. Individuals were found to consume approximately 70 L of non-heating hot water per day on average. From the decomposed outputs, a regression model was estimated to examine how residential heating demand varies with building volume, type and building regulations, and non-heating usage was clustered to identify key user groups. The regression model characteristics explained 67.1% of heating variation, with building regulations shown to have improved energy efficiency over time. The K-Means clustering algorithm revealed four user groups but cluster boundaries were not clear (silhouette score of 0.227). Overall, the research provides a methodology to decompose DHS smart meter data at scale, the results of which providing empirically-backed results for DHS operators, policy makers, and researchers.