Paper Conference
Proceedings of BauSim Conference 2022: 9th Conference of IBPSA-Germany and Austria
![]() ![]() ![]() ![]() |
MACHINE LEARNING FOR IMAGE-BASED RECOGNITION OF BUILDING AGE FOR URBAN ENERGY SIMULATION-TESTING AND VALIDATION ON AN EXEMPLARY CITY QUARTER
Alexander Benz, Mara Geske, Conrad VoelkerDOI: https://doi.org/10.26868/29761662.2022.4Abstract: Data acquisition for urban energy simulations is time consuming and usually associated with incomplete data sets which enhanced by subjective approaches (e.g., estimation of building age and a probable construction). The analysis of exterior views based on machine learning algorithms enables an automated and reproducible recognition of building age classes. Therefore, convolutional neural networks (CNNs) were trained in a supervised process with approximately 3,700 images of residential buildings from different German building age classes. The highest accuracy obtained (56 % correct) exceeds the human prediction accuracy (37 % correct). A real life test and validation of the trained CNNs is conducted on an inhomogeneous city quarter with predominantly residential use. Paper:bausim2022_Benz_Alexander