UAV-based automated 3D modeling framework using deep learning for building energy modeling

Abstract

Many approaches to the Urban Building Energy Model (UBEM) have been developed to analyze urban-scale energy demand patterns. The main goal of UBEM is to minimize manual work and improve modeling accuracy when building 3D modeling procedures that deal with vast amounts of data. To do so, it is important to build an automated process and increase the modeling efficiency throughout the process. This study proposes a new framework for automatically generating 3D models for building energy modeling. This framework collects geographic coordinate system (GCS) data by applying algorithms based on a convolutional neural network (CNN), Haversine formula, unmanned aerial vehicle (UAV), and geographic information system (GIS) information. The collected GCS data were used to generate a 3D model using EnergyPlus, resulting in a 3D model capable of a Level of Detail 3. Subsequently, we compared and verified the size and energy performance of the actual building with those of the generated model. The model size errors generated without drawing information are as follows: buildings, 3.69%, windows1 16.75%, and windows2 19.43%. The error range evidenced through the energy performance evaluation indicator showed that the MBE values for the cooling and heating energies were 5.54% and 5.77%, respectively.