Answers are grounded in this paper and cite their sources.
Urban-scale building envelope optimization is constrained by the computational burden of iterative simulation. This study proposes a Retrieval-Augmented Generation (RAG)-enhanced framework that replaces the search phase of conventional optimization: literature-screened envelope solutions are retrieved from a curated corpus and then cross-checked through physics-based EnergyPlus simulation, rather than located through exhaustive parametric search. The corpus comprises 20,247 initially identified publications screened to a core of 338 studies, together with an experimentally characterized database of 502 envelope materials, of which 65 were used in the assemblies evaluated here. A three-stage progressive retrofit strategy — conventional high-performance insulation; vacuum insulation panels, aerogel and phase change materials (PCMs); and optimized PCM hybrid assemblies — was applied to seven reference buildings on Yonsei University's 86-building campus in Seoul, Korea. A systematic literature review established that whole-building field-measured savings are systematically 0.25–0.75× of simulation-only upper bounds, a correction embedded directly in the retrieval thresholds and screening criteria. The seven reference buildings produce a simulated mean building-level EUI reduction of 21.5%, ranging from 9.5% in high window-to-wall-ratio modern buildings to 39.6% in thermally deficient heritage structures, while campus-wide analysis of all 86 buildings yields 15.7% (124.4 to 104.9 kWh/m 2 ·year). EnergyPlus models were not calibrated against sub-metered operational data, so all reported reductions are simulation-based screening estimates. The framework offers a scalable, computationally efficient route to envelope retrofit planning for large institutional portfolios and urban decarbonization programs.
No takes yet. Share an insight, caveat, or question.
Han et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: