Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Development of a metamodel to assess building thermal performance for naturally ventilated residential buildings

Rodolfo Kirch Veiga, Letícia Gabriela Eli, Amanda Fraga Krelling, Marcelo Salles Olinger, Ana Paula Melo, Roberto Lamberts 1
Federal University of Santa Catarina, Brazil

Abstract: The aim of this research was to develop a metamodel to predict the percentage of occupied hours within an operative temperature range (PHFT) for naturally ventilated apartments. PHFT is a thermal performance indicator from the Brazilian standard for performance of residential buildings (NBR 15575). The analysis focused on multi-story buildings and Brazilian climates. Five modeling techniques were investigated: Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), Support Vector Machine (SVM), Random Forest (RF) e Extreme Gradient Boosted Trees (XGBT). The methods were compared concerning their accuracy based on two indicators: Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The main conclusions show that the more accurate metamodeling technique was XGBT.
Keywords: Building Simulation, Accuracy, Simplified model
Pages: 1849 - 1856