A simple yet powerful dimensionality reduction method for annual daylighting prediction and its inverse process via pix2pix

Predicting annual daylighting performance is an important task in building science, as daylighting impacts both occupant comfort and building energy consumption. Traditional simulation methods, such as Climate-Based Daylight Modeling (CBDM), require detailed 3D modeling of the testing environment and intensive computational resources, which limit their practical application. This research proposes a simple yet powerful dimensionality reduction method that simplifies the process by transforming complex 3D information of a room, such as room dimensions and window locations, into a 2D image. Utilizing this approach, a series of pix2pix-based prediction models were trained to predict annual daylighting performance of target spaces using different annual daylighting performance metrics, including Daylight Autonomy (DA), Continuous Daylight Autonomy (cDA), and Useful Daylight Illuminance (UDI). This research also introduces an inverse process, where models were developed to predict window configurations based on desired daylighting performance results, which can be very useful in design projects that prioritize daylighting performance. The models demonstrate reasonable accuracy, achieving a structural similarity index measure (SSIM) of 0.627 for the UDI prediction model with under 500,000 training steps, and 0.611 for the inverse UDI prediction model. In addition, this method is much more efficient than the traditional CBDM method. This research also proposes a novel evaluation method for pix2pix-based daylighting prediction models, which gives more accurate results during the evaluation process. • A daylighting prediction model with a novel dimensionality reduction method. • An inverse model to design openings based on desired daylighting outcomes. • Enables reliable daylighting predictions from simple 2D input, bypassing 3D modeling. • Evaluates model performance by isolating irrelevant background details. • Demonstrates high computational efficiency over traditional daylighting simulations.