• First to study new post-seismic Ayiwang dwellings and design optimization path. • Room-specific seasonal energy rhythms define objectives, avoiding mismatch. • First SHAP-based sensitivity analysis for Ayiwang dwellings' unique needs. • First application of XGBoost in Ayiwang dwelling design and optimization. • GA-XGBoost achieves 64.9% energy saving and 31.46% less discomfort. To address high energy consumption and poor thermal comfort of seismic-retrofitted traditional Ayiwang dwellings in Southern Xinjiang’s sandy-windy region, China, this study optimizes typical cases via the GA-XGBoost framework, achieving three core innovations: first, first applying the GA-XGBoost framework to traditional residential design (previously mainly limited to public buildings); second, introducing the SHAP algorithm for interpretable sensitivity analysis of Ayiwang dwellings’ unique spatial morphology (distinct from conventional residences), identifying facade window-to-wall ratio (WWR) as the key factor regulating comprehensive building performance; third, tailoring energy optimization goals aligned with their energy-use patterns via the dwellings’ seasonal space-use characteristics. Specifically, a parametric simulation workflow was built with Rhino-Grasshopper and EnergyPlus. The XGBoost surrogate model performed well on the test set, with R² values of 0.8956 (EUI), 0.9758 (PPD), and 0.9317 (UDI). SHAP analysis further confirmed WWR as the core performance-regulating factor. After multi-objective optimization of 9 key design variables via genetic algorithm (GA), the final scheme achieved 64.9% lower EUI, 31.46% reduced PPD, and 5.9% higher UDI. This study offers a reliable technical path for green performance optimization and energy-efficient design of Southern Xinjiang’s traditional Ayiwang dwellings.
Wang et al. (Sun,) studied this question.