Abstract
In Europe, buildings account for 40% of final energy demand. Building stock models assess the impacts of technologies on energy consumption, greenhouse gases, policies, city planning, renewable energy, renovation strategies, and health effects. Two approaches, top-down and bottom-up, generate these models using real data and simulations. In the big data era, information about buildings is increasingly available, allowing real analysis of building stocks (top-down approach). In the bottom-up approach, models are estimated through simulations of building archetypes and aggregated at stock level. Unsupervised machine learning like clustering is widely used to find and group similar buildings. Centroid- and density-based algorithms are most popular but subsequent evaluation of clusters is essential. In this chapter we demonstrate two applications of clustering on different building stocks. In the first, the aim is to generate heat saving cost curves for the residential sector. These curves allow policy makers to choose renovations that save most energy per Euro invested. In the second application, clustering is applied to a building stock in Flanders to generate synthetic data allowing to simulate energy efficiency scenarios for buildings. The archetypal modeling approach used classifies buildings based on characteristics and scales the energy consumption up to the entire housing stock.