Files

SEGAN-D-25-01740.pdf
  • Open Access
  • Adobe PDF
  • 10.11 MB
Download

Details

Authors
Abstract
Building electricity consumption highly varies due factors such as occupancy behavior, weather conditions, and operational practices. In this study, we develop a novel statistical framework to model these complex consumption patterns using an extensive set of probability distributions. Specifically, we systematically evaluate twenty candidate models -- including fourteen canonical continuous probability distributions, Gaussian mixture models, 2-parameter Weibull mixture models, and kernel-density estimation using Gaussian kernels -- across three distinct temporal clustering granularities. Our analysis is conducted on three sets of datasets that cover UK households and two Renewable Energy Communities from Brussels, Belgium, encompassing both residential and business energy consumption profiles. For each data cluster, probability distributions are fitted using maximum likelihood estimation or the Expectation-Maximization algorithm. We assess the goodness-of-fit of each model by combining multiple metrics: log-likelihood, Akaike information criterion, likelihood-ratio tests, and a parametric bootstrap-based Kolmogorov–Smirnov test that adjusts for parameter estimation uncertainty. Our findings indicate that clustering granularity plays a crucial role in the effectiveness of probability distribution fitting for electricity consumption data. At coarser granularities, mixture models (MMs) consistently outperformed other methods. However, as granularity became finer, the performance landscape became more varied. While MMs maintained strong performance, other models, such as the 3-parameter Weibull and gamma distributions, also demonstrated competitive results in specific scenarios. Our approach enhances electricity consumption modeling accuracy while offering valuable insights for demand management and energy policy. By integrating diverse datasets and a rigorous evaluation framework, we set a new benchmark for analyzing building energy use and supporting more efficient energy systems.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

de Schietere de Lophem, M., Misselyn, L., Veroneze, R., Verhaeghe, H., & Legay, A. (2025). Modeling Electricity Consumption Patterns in Buildings Using Probability Distribution Functions. https://hdl.handle.net/2078.5/270748