A GIS-Informed deep learning framework to assess the effect of urban morphology on energy demand

Abstract

Assessing the impact of urban morphology on building energy demand is essential for sustainable city development. Traditional approaches often oversimplify complex urban geographic data or rely on extensive calculations of urban morphological factors (UMFs), leading to significant uncertainties. Advances in geographic information systems (GIS) and deep learning offer solutions to these challenges. As a consequence, this study introduces a GIS-informed deep learning framework, employing high-resolution 3D point cloud data to process UMFs. It outputs 3D vectors representing cooling, heating, and electricity demand via a multi-channel convolutional neural network (CNN). The network classifies building age and function type using satellite imagery and point-of-interest (POI) data, generating twelve-dimensional feature vectors by integrating building orientation coordinates and normal vectors as CNN inputs. Applied to Xinwu District in Wuxi City, the model significantly enhances prediction accuracy over traditional multiple linear regression (MLR) based on morphological parameters across twelve urban forms, achieving an R² improvement exceeding 50%. These results demonstrate capability of proposed model to capture complex nonlinear relationships between detailed morphology and energy efficiency in various urban configurations. This study offers a scalable and reliable tool for designing and planning energy-efficient cities.