Assessing solar irradiance on facades is challenging due to complex urban features. SolarGAN helps by introducing a Deep Learning model that generates stochastic solar irradiance time series from simple fisheye images, offering a faster, data-driven alternative to labor-intensive simulations for robust BIPV energy system design.
Illustration of the detailed model architecture and the entire training process in 3 separate stages. Lines and texts in red indicate the loss function terms involved. Source: Zhang et al. (2023)
Rhino Grasshopper implementation
https://github.com/architecture-building-systems/SolarGAN (to be released soon)