Multi-objective optimization of window retrofit for dormitory buildings under material circularity uncertainties: A robust surrogate-assisted framework

Escalating climate risks necessitate deep decarbonization of the building sector. As a vital lever of building decarbonization, window retrofits in existing buildings face complex trade-offs between embodied impacts, operational savings, costs, and occupant comfort. These challenges are intensified by uncertainties in future material circularity. To address this, this study proposes a robust surrogate-assisted multi-objective optimization framework. Applied to a representative university dormitory in China, the workflow integrates high-fidelity machine learning models with evolutionary algorithms to navigate the design space efficiently. The optimization targets four competing objectives: carbon payback period, life cycle carbon emission variation, life cycle cost, and useful daylight illuminance, under three distinct aluminum recycling scenarios (30 %, 50 %, 100 %). Through multi-criteria decision-making analysis, results reveal that increased material circularity reshapes the optimal solution landscape, shifting from 30 % to 100 % aluminum circularity achieves a 20.64 % increase in life cycle carbon savings (reaching 2.63 × 10⁵ kgCO₂eq). The identified robust archetypes demonstrate exceptional viability, achieving carbon payback periods of 1.4 years at most while maintaining high visual comfort comparable to the benchmark, effectively decoupling decarbonization from cost and comfort penalties. Crucially, a “material reversal” phenomenon is identified: the optimal frame material shifts from PVC in low-circularity scenarios to high-performance recycled aluminum in future circular economies. This study validates the pivotal role of supply chain circularity and provides a scalable methodology for selecting resilient, low-carbon retrofit strategies.