Towards A Federation of Surrogate Models For Early Building Design: Balancing Energy Efficiency Daylighting and Glare Risk

This research develops a federation of surrogate models to predict energy, daylight, and glare risk trade-offs, allowing for efficient parametric analysis. It then validates their application using a professional case study. For the daylighting and energy performance components, we trained a deep convolutional neural network and neural net algorithm, respectively. The results of the experiments reveal acceptable absolute errors compared to standard simulation tools, reduction of analytical model preparation, and directional accuracy while drastically reducing the computational time. The proposed approach for parametric analysis for multicriteria optimization is a promising alternative for adoption in professional practice. The ML approach enables designers to make more informed decisions early in the design process through feedback on demand.