Geospatial clustering as a method to reduce the computational load in urban building energy simulation

Abstract

Since the recent birth of physics-based urban building energy modeling (UBEM), researchers have started tackling the issues characterizing this research field, mainly linked to the lack of extensive and standardized building information datasets and the necessity of simplifying the modeling process. Concerning the latter, geospatial clustering approaches seem to be plausible methods to reduce the computational load in urban simulation, and this work aims to test their suitability and performance.

For this purpose, a case study of almost 3800 buildings in Padova, Italy, is analyzed. The tendency analysis is first used to quantify the underlying clusters that could be present. The study of this metric reveals the organic morphology and the heterogeneity of building stock in European cities like Padova. Additionally, several clustering algorithms are applied to the location, use, envelope, and geometry variables to simulate building clusters and quantify the increase in geometric and heating/cooling demand uncertainty.

Results show that, for this case study, building clusters are characterized by lower volumes than when considering single buildings, which is also reflected in a lower heating and cooling demand prediction. Nonetheless, these errors are found to be in an acceptable range (less than 6%) for UBEM applications.