A data-driven framework for characterising building archetypes: A mixed effects modelling approach

Abstract

Building archetypes are a common solution to study the energy demand of cities and districts. These are generally based on building information such as construction year and function. However, there can be large differences in the energy demand of buildings of the same archetype due to factors such as the preferences of occupants, quality of the building construction, and unrecorded renovations. This work uses a non-linear mixed effects model to capture these random differences. The model uses weather measurements to generate the daily heating load of buildings for the whole year. The model is generated and tested using data from 56 Norwegian apartments. Results show that 91% of measurements from an out-of-sample test set fall inside the 95% prediction interval. Additionally, the model allows us to compute a proxy of the heat loss coefficient, which characterises the heating performance of the population of apartments. Finally, two sub-categories of apartments are identified by clustering the model estimates for the studied population. The model is general, computationally light and uses existing data that are commonly collected in many buildings. The suggested method offers a more robust and reliable method to segment building archetypes using only weather data and energy demand.