Answers are grounded in this paper and cite their sources.
Simulation study demonstrates machine-learning-driven daylight optimization for heritage spaces, indicating reversible ceiling reflectors improve illuminance without worsening glare.
Adaptive-reuse heritage spaces require improved daylight conditions, while façade modifications are constrained by minimal intervention and reversibility. This study presents a simulation-based workflow supported by machine learning surrogate models to optimise reversible ceiling-mounted reflectors while considering the daylight–glare trade-off. A reflector is defined by six vertical control points, each taking five discrete Z-levels, yielding 15,625 geometries. Reflector placement was fixed through an XY pre-screening study ( n = 119) using annual spatial daylight autonomy (sDA) and mean workplane illuminance, after which a location was selected by two-objective Pareto screening. Annual ClimateStudio simulations are completed for a 7500-sample subset of the factorial space, producing sDA and mean illuminance under consistent room, material and sensor-grid settings. Multiple surrogate families (random forest, gradient-boosted trees and multilayer perceptron) are benchmarked using hold-out evaluation with error metrics root mean square error (RMSE), mean absolute error (MAE), coefficient of determination ( R 2 ) and ranking metrics Spearman correlation, top-K recall (Recall@ TopK) to reflect the design-screening objective. The best-performing surrogate is retrained on all simulated samples and used to predict performance for the un-simulated candidates, enabling full-space Pareto screening to shortlist high-performing geometries. A confirmation set of simulations, including top-ranked and randomly selected un-simulated candidates, is used to verify surrogate reliability before selection. The workflow incorporates annual glare screening using spatial disturbing glare (sDG) via a stratified glare subset and glare surrogate, and demonstrates scalable two-reflector screening through shortlist-based shape pairing and controlled 2D placement sweeps. Validation simulations showed that the selected single-reflector design increased mean workplane illuminance from 375 to 403.6 lx and sDA300/50% from 43.1% to 47.3%. The selected two-reflector configuration achieved 412 lx and 53.2% sDA, while annual glare remained comparable to the baseline, with sDG values of 8.0% for the single-reflector case, 8.8% for the two-reflector case and 8.6% for the baseline.
No takes yet. Share an insight, caveat, or question.
Catay et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: