AI-driven passive design optimization for thermal comfort in courtyard buildings

Amid accelerating climate change, enhancing indoor and outdoor thermal comfort (IOTC) via passive design is critical, yet architectural-scale studies remain scarce. This study introduces an AI-driven framework combining a genetic algorithm (GA) and explainable machine learning to optimize spatial morphological parameters (SMPs) for three courtyard typologies (E-shape, C-shape, and enclosed). Using Rhino/Grasshopper for parametric modeling, Ladybug/Honeybee for thermal simulations, and Wallacei for multi-objective optimization, we evaluate 2,500 design variants against indoor Predicted Mean Vote (PMV) and outdoor Universal Thermal Climate Index (UTCI). The Pareto-optimal set reduces summer UTCI by 2.17–3.07 °C and PMV by 0.89–1.08, while increasing winter UTCI by 1.06–1.91 °C and PMV by 0.21–0.40. An ensemble of regressors identifies Extreme Gradient Boosting (XGBoost; R2 ≥ 0.95) as optimal; Shapley Additive exPlanations (SHAP) highlight Form Factor Shape Coefficient (FSC), Building Shape Index (BSI), and Building Courtyard Ratio (BCR) as primary drivers—FSC > BSI > BCR for E−/C-shapes, and BCR > FSC > BSI for enclosed cases. Nonlinear SHAP-derived intervals ensuring seasonal comfort include FSC 0.25–0.26, BSI 90–100, BCR ≥1.7 (E-shape); FSC 0.18–0.22, BSI 30–33, BCR 0.8–1.3 (C-shape); and, for enclosed courtyards, BCR 1.8–4.0 with FSC ≤0.22 in summer or FSC ≤0.35 with BSI ≥300 and BCR ≥1.2 in winter. This passive optimization strategy significantly improves IOTC, lowers energy demand, and guides early-stage morphological decisions for low-carbon, healthy building retrofits.