Using data-driven models to estimate the energy use of buildings based on a building stock model

Abstract

In order to increase the renovation rate in Belgium, an approach is needed to refurbish clusters of buildings rather than individual buildings. To allow for a meaningful clustering of buildings, the energy performance of the existing buildings should be known. Nowadays all energy related data at building level in Belgium are confidential and cannot be shared with municipalities, private institutions or researchers. Crucial information regarding the energy use of the existing buildings is hence lacking to allow for such clustering. Using different machine learning techniques, i.e. decision trees, random forest models and k-NN models, these missing energy data for the buildings of the city of Leuven are predicted.