Developing urban residential reference buildings using clustering analysis of satellite images

Built-up areas tend to comprise a variety of buildings with diverse and complex shapes, functions and construction characteristics. This variety is the source of significant challenges when calculating building energy use at the building stock level. Moreover, the process of developing stock models usually requires large amounts of data that are frequently scarce, nonexistent or at least not publicly available. Under these circumstances, defining a limited set of reference buildings representing the stock is useful to study the actual energy consumption and the potential effects of different energy conservation measures. This paper presents a new method for developing typical residential reference buildings at district level for bottom-up energy modeling purposes. By means of widely and freely available satellite images, an information database of building shapes is created and a clustering analysis of the geometrical features is performed to define a number of archetypes representative of the heating and cooling energy demand of the district. The method is tested and demonstrated through the case study of the Yuzhong District in Chongqing (China) by comparing the Energy Use Intensity (EUI) of the archetypes derived in this way against detailed dynamic simulations. Results show very small differences in the estimated stock energy consumption (+0.03% in heating energy consumption and +2.97% in cooling energy consumption).