Semantic obstruction detection for improved solar energy potential modeling in urban areas

Abstract Achieving precise and scalable solar potential estimation in urban settings is challenging due to the presence of a wide variety of obstructions. To address this issue, we developed a novel urban solar potential modeling method based on an improved 2-phase daylight model. Utilizing a dynamic graph convolutional neural network semantic segmentation model to process urban point cloud data, our method distinguishes between different types of solar obstructions, assigning specific simulation hyperparameters accordingly. Demonstrated through experiments, our method significantly outperforms traditional models by avoiding the underestimation of shading impacts—by up to 60% for monthly solar irradiation potential and 40% for annual PV yield potential. Moreover, our method accurately accounts for complex solar transmission through tree canopies, avoiding underestimation of PV energy potential by up to 7% compared to its predecessor (Pyrano 1.0). These improvements offer substantial benefits for managing PV shading risks, configuring PV systems, and managing renewable resources, especially in urban areas with complex geometries and dynamically changing shading conditions. Our findings underscore the method’s potential to enhance decision-making in sustainable urban development and renewable energy integration.