deep learning-driven optimization of fenestration for daylighting in hot-arid climates: a hybrid evolutionary framework

This study articulates a pioneering, integrated methodology for the optimization of fenestration and spatial configuration in residential living rooms, tailored specifically to the exigencies of isfahan's hot-arid climate. the overarching objective is the maximization of daylighting performance through the simultaneous evaluation of three pivotal climate-based metrics: spatial daylight autonomy (sda), annual sunlight exposure (ase), and useful daylight illuminance (udi). distinctively, this research synergizes a deep learning (dl) predictive model with a genetic algorithm-based multi-objective optimization framework (galapagos), transcending the limitations of traditional static modeling techniques. the deployed feedforward neural network exhibited exemplary predictive fidelity, yielding r² coefficients of 0.97 for udi and ase, and a perfect 1.00 for sda. subsequent interpretability analyses underscored the critical impact of room depth and window-to-wall ratio (wwr) on luminous performance. the optimization protocol culminated in a definitive design archetype for a south-facing volume (4m width × 5m depth), oriented at a -1-degree azimuth. this configuration, featuring a 30% wwr distributed across two vertical apertures with a 0.90m sill height and devoid of external shading, achieved an optimal equilibrium: 100% sda, 43% ase, and 72% udi. consequently, this work establishes a robust, data-driven framework for sustainable architectural practice, offering precise parametric guidelines for daylighting efficacy in challenging climatic zones.