From point cloud to energy model: Constructing high-resolution urban building model for energy simulation and evaluation using UAV oblique imagery

Urban building energy modeling has become a critical tool for analyzing building energy use in different scales, however, one practical obstacle that constrains its wide application is how to quickly construct highly accurate simulation models in complex existing urban areas. This study proposes an automated method utilizing open-source electronic maps and unmanned aerial vehicle oblique imagery to address three-dimensional (3D) modeling issues caused by incomplete databases. A Level of Detail (LOD) 2.1 of urban 3D building models was created and Energyplus engine was used for energy simulation. Further explorations of key parameters that influence model construction and energy results were conducted. Key findings include building height identification and roof type prediction both reach 92 % accuracy, with building height errors within 1 m, and roof type prediction achieving an AUC of 0.93. Average construction time of 2.5 s per building. Higher LOD models appropriately increase simulation times while significantly reducing simulation errors. Although simulation times for LOD 2.1 increased by 6.5 s compared to LOD1.2mean, the results show that the mean absolute percentage error for energy use intensity decreased from 3.02 % to 0.94 %, while the root mean square error reduced from 4.41 kWh/m2 to 2.46 kWh/m2 compared to the baseline model. Accurate height has a significant impact on simulation results, especially for estimating heating energy demand. This study advances the development of an energy simulation model at the neighborhood scale, consisting of multiple blocks with unified building types. The model is built using an efficient and automated construction method, which enhances urban planning, energy system design, and provides new perspectives on urban morphology.