Using a Surrogate Model to Analyze the Impact of Geometry on the Energy Efficiency of Buildings

Parametric exploration and optimization of building geometry is a powerful tool for designing energy efficient buildings. However, in practice this process is computationally expensive and time-consuming. In this research, we explore the use of surrogate models, i.e. efficient statistical approximations of expensive physics-based building simulation models, to lower the computational burden of large-scale building geometry analysis. For this purpose, we developed a novel dataset of 38,000 residential building models derived from real world floor plans from (Wu et al. (2019)) and train a surrogate model to emulate their simulated annual energy performance. We extract up to 20 parameters as surrogate model inputs to represent the building geometry and show that the trained surrogate model reaches a high accuracy (R-2 score = 0.999, MSE = 0.007 and RMSE = 0.022) on test data. The current setup forms the basis for further research where the complexity of the building models will be increased.