A retrieval-augmented generation (RAG)-enhanced framework for knowledge-guided campus-scale building retrofit design
Year: 2026
Type: article
Abstract: 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.
Source: Applied Energy
Authors: Jung Min Han, Deunkyeol Chung, Sanghwan Choi
Institution: Yonsei University
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,430
OpenAlex ID: W7236432959
