ANN-Based High-Dimensional Multi-objective Optimal Design for Natural Lighting in Large-Span Buildings

In order to build a high-dimensional multi-objective comprehensive balance optimization system for natural lighting, and to solve the limitations of the building performance simulation (BPS) based optimization design method in terms of the number of optimization objectives and the feedback time of results when large-span buildings involve high-dimensional multi-objective optimization design. The study proposes an architect-oriented high-dimensional multi-objective optimization design method for natural lighting based on an artificial neural network (ANN) meta-model coupled with a high-dimensional multi-objective optimization algorithm, based on the Rhino + Grasshopper parametric design technology platform, using a large-span building medium-sized gymnasium as a practical research case to carry out the study. The empirical results show that: 1) the mean square error (MSE) value domain of ANN meta-model construction schemes are all less than 0.0054, and the Pearson correlation coefficient R value domain are all greater than 0.81. 2) the Pareto front solution search time for high-dimensional multi-objective optimal design is 7s. 3) the Daylight Autonomy (DA) and Useful Daylight Illuminance (UDI) values are improved by 0.82%–109.62% and 3.08%–26.71%, respectively, and the Energy Use Intensity (EUI) and Predicted Mean Vote (PMV) values are reduced by 3.81%–13.78% and 0.21%–13.77%, respectively, for the non-dominated solution optimized design schemes.