Optimization framework for daylight and thermal environment of retractable roof natatoriums based on generative adversarial network and genetic algorithm

Abstract

This paper proposes a framework for multi-objective optimization of indoor daylighting and thermal comfort in natatoriums with retractable roofs, using genetic algorithms and building performance simulation. The goal is to balance daylight illumination and thermal comfort under different roof opening states. Efficiency of computational fluid dynamics simulation is improved by a multi-objective optimization framework based on non-dominated sorting genetic algorithms, integrating a generative adversarial network to learn the environmental map. The study analyses the impact of different roof opening ratios on environmental objectives and establishes regression equations to support optimization and accurate decision-making. In the optimal opening state, compared to a fully closed roof, the daylight factor and natural airflow velocity increase by up to 3.9 and 5.3 times, respectively, while the universal thermal climate index decreases by 1.0 °C. The introduction of a generative adversarial network proxy model significantly improves computational efficiency, accelerating natural airflow velocity calculation by 70–100 times and reducing simulation time by 98.6 %. This study expands predictive models from the urban scale to the building scale, enhancing simulation efficiency and promoting the application of deep learning models and genetic algorithms in climate-responsive building environment assessment and prediction.