A Generative Adversarial Network-based Method for Sunshine Duration Prediction in China

In this paper, we propose a model based on the Generative Adversarial Network (GAN) aimed at providing fast and effective predictions of sunshine duration for architectural design in urban planning. Given the increase in building heights and the decrease in the distance between buildings in urban areas of China, effective prediction of sunshine duration is crucial for complying with local government building standards. Our method initially utilizes the sunshine simulation module of the Revit software to generate sunshine metadata, which is then transformed into image data through a linear transformation. These image data are subsequently used to train the GAN model, where the generator is responsible for producing the predicted sunshine duration images, and the discriminator differentiates them from the actual sunshine duration images. Experimental results demonstrate that our approach can effectively predict sunshine duration within a 4-minute error margin and is 100 times faster than the traditional Revit software. The innovation of this study lies in the application of the GAN model for predicting sunshine duration in the building environment field, offering a more efficient solution compared to existing software.