Modelling Residential Electricity and Gas Demand at the Dutch Postcode-6 Level Using Publicly Available Data

Abstract

This thesis investigates how publicly available data can be used to estimate and explain average annual residential electricity consumption and gas consumption per dwelling across residential PC6 areas in the Netherlands. Energy-consumption and socioeconomic data from CBS are combined with detailed housing-stock information from VNG/DEGO and aggregated to a common PC6-level dataset. Interpretable regression models are developed and evaluated using model-fit diagnostics and 10-fold cross-validation, with Random Forest models used as non-parametric predictive benchmarks. The selected log-log regression models explain approximately 72% of the observed variation in net-grid electricity consumption and 64% of the variation in gas consumption within the selected gas sample. The results highlight dwelling floor area as an important predictor for both energy carriers, while household size is particularly relevant for electricity consumption and building age and energy-label composition are more strongly associated with gas consumption. Random Forest benchmarks provide moderate improvements in predictive accuracy, indicating that some additional nonlinear structure remains while supporting the regression models as an interpretable primary framework. Finally, annual consumption estimates are combined with Dutch Standard Load Profiles to construct representative temporal electricity and gas demand profiles. Overall, the thesis demonstrates that publicly available Dutch datasets can support spatially detailed, interpretable, and reproducible residential energy-demand modelling at the PC6 level.