A Random Forest Surrogate Model for Multi-Objective Thermal and Daylighting Evaluation: Coupling Parametric Simulation with Machine Learning for Early-Stage Office Design in a Hot Arid Climate
Evaluating conflicting early-stage design objectives, such as minimizing area-normalized cooling energy intensity (kWh/m2) and maximizing spatial Daylight Autonomy under static shading (sDA300/50%), is limited by the computational cost of building performance simulation. This study develops Random Forest surrogate models of Honeybee (EnergyPlus and Radiance) simulations for single-zone offices in the hot arid climate of Cairo, trained on 1037 parametric scenarios generated with a hybrid sampling strategy. In 10-fold cross-validation, the models reproduced the simulated cooling and heating intensities and four climate-based daylight metrics with an R2 of 0.943–0.990 and mean absolute percentage errors below 4% for all indicators except Annual Sunlight Exposure (7.9%), while reducing the evaluation time from about 55 s per simulation to about 0.3 s. Room depth and window-to-wall ratio were the most influential variables overall, and orientation dominated sunlight exposure. A surrogate-assisted multi-objective optimization, verified by 124 additional full simulations, showed that the surrogate overestimates daylight performance at the extremes of the design space; the verified Pareto front indicates that north- to north-east-facing rooms with the baseline glazing achieve an sDA above 90% at about 119 kWh/m2. The surrogate, deployed in an interactive tool, is best used for rapid screening, with final candidates confirmed by simulation.
Authors
- Kareem S. Galal (ORCID: https://orcid.org/0000-0001-6311-0352)
Institutions
- Beirut Arab University (LB)
Publication Details
- Journal
- Architecture
- Published
- 2026-11-18
- DOI
- https://doi.org/10.3390/architecture6040218
- Primary Topic
- Building Energy and Comfort Optimization
- Type
- article
- Field-Weighted Citation Impact
- 0.00