Statistical data-driven analysis and modelling of total energy use in new or thoroughly renovated single-family houses

Abstract

The subject of this paper is to analyse how the regulatory calculated energy use relates to the real total energy use for new or thoroughly renovated Flemish single-family houses where electricity is the only energy carrier. Additionally, the authors determine whether statistical data-driven models can help inform current and future home owners and tenants about their energy use (and thus also potential energy savings when applying energy saving measures). These questions are investigated by using housing datasets from the Flemish energy performance database and real energy use data from the Belgian grid operator. The paper comprises outlined database cleansing and filtering choices and enlightening statistical database analyses and figures. The results clearly demonstrate that the regulatory calculation method poorly estimates the real energy use (RMSE-/MAE-results of respectively 7227 kWh/y and 5242 kWh/y), yet both are moderately correlated (τ = .548, p < .001). Further, the statistical regression models show good results at stock level for new or thoroughly renovated Flemish single-family houses (where electricity is the only energy carrier) (adj. R2 up to 65.3%). Nevertheless, their performance at individual building level is still limited and considered too poor for inference as a considerable part of the variance is left unexplained.

Keywords

• energy performance database
• data-driven regression modelling
• real yearly building energy use
• statistical data analysis