Research on Daylight Inside Buildings Based on Machine Learning

Daylight assessment ensures the quality of indoor environments and is crucial for health and thermal comfort. Traditional daylight simulation tools require significant computational resources and time. In the foreseeable future, more efficient and robust machine learning models should be used to develop more convenient programs. In this study, a machine learning model was developed to predict indoor illuminance in buildings, further considering the real-world feature of window transmittance for different building layouts. A method for locating planar vertices was also developed, improving accuracy. All data used for training and prediction were generated by the simulation software Honeybee. The results indicate that introducing more real-world features enhances the model's generalization capability, and the planar vertex location method can replace the uniform ray location method to achieve more accurate predictions.