Wei Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Wei Wu has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
Ming Qu
1 shared publication- Qingchun Li
Bowen School of Construction, Purdue University, West Lafayette, IN 47906, USA
1 shared publication Yongfei Li
1 shared publication- Yu Hou
Department of Construction Management Western New England University 1215 Wilbraham Rd 01119 Springfield MA
1 shared publication Zhe Wu
1 shared publication
Latest Publications
- Open Access
- 22 August 2026
- Published by Elsevier in SSRN Electronic Journal
Geo2UBEM: 3D Geometry Abstraction and LLM Multi-Agent Parameter Inference for Automated Urban Building Energy Modeling
Urban building energy modeling (UBEM) is critical for city-scale decarbonization planning. However, most existing calibration workflows, which are deterministic, Bayesian, and machine learning-based, rely on building-level metered energy data as calibration targets. For most urban building stock, such data are unavailable, inconsistent, or inaccessible, excluding most buildings from rigorous, standards-based calibration. To address the gap, Geo2UBEM is introduced as the first end-to-end, meter-free workflow integrating automated scan-to-IDF 3D geometry reconstruction with an LLM-driven multi-agent system (LLM-MAS) for energy parameter inference. Urban 3D geometry is generated from street-level imagery, with orientation-specific window-to-wall ratios refined through facade-aware enhancement. In place of the metered data conventionally required under ASHRAE Guideline 14, CBECS 2018 national median benchmarks are used: the LLM-MAS diagnoses simulation-benchmark discrepancies and iteratively narrows the Bayesian optimization search space until convergence. Applied to six calibrated campus buildings across four types at Purdue University (101,170m2), Geo2UBEM achieves ASHRAE-consistent convergence (|NMBE| ≤5%; CV(RMSE) ≤15%) without any metered input, a result not previously demonstrated for a multi-building cohort. Independent validation against actual metered EUIs confirms all six models fall within CBECS sector ranges, at an average inference cost of $0.09 per building.A three-layer sensitivity and uncertainty analysis (Morris screening, Sobol decomposition, Monte Carlo propagation) reveals a cohort-level observability gap: equipment density accounts for 80--99% of annual energy-use variance, while envelope parameters, most directly recoverable from geometry, contribute negligibly. This challenges common geometry-driven UBEM assumptions, indicating calibration effort is better spent characterizing operational loads than refining envelope parameters. Collectively, these results establish a scalable, low-cost pathway to standards-consistent UBEM calibration for urban stocks lacking metered data.
- 22 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGeo2UBEM: 3D Geometry Abstraction and LLM Multi-Agent Parameter Inference for Automated Urban Building Energy Modeling
Urban building energy modeling (UBEM) is critical for city-scale decarbonization planning. However, most existing calibration workflows, which are deterministic, Bayesian, and machine learning-based, rely on building-level metered energy data as calibration targets. For most urban building stock, such data are unavailable, inconsistent, or inaccessible, excluding most buildings from rigorous, standards-based calibration. To address the gap, Geo2UBEM is introduced as the first end-to-end, meter-free workflow integrating automated scan-to-IDF 3D geometry reconstruction with an LLM-driven multi-agent system (LLM-MAS) for energy parameter inference. Urban 3D geometry is generated from street-level imagery, with orientation-specific window-to-wall ratios refined through facade-aware enhancement. In place of the metered data conventionally required under ASHRAE Guideline 14, CBECS 2018 national median benchmarks are used: the LLM-MAS diagnoses simulation-benchmark discrepancies and iteratively narrows the Bayesian optimization search space until convergence. Applied to six calibrated campus buildings across four types at Purdue University (101,170m2), Geo2UBEM achieves ASHRAE-consistent convergence (|NMBE| ≤5%; CV(RMSE) ≤15%) without any metered input, a result not previously demonstrated for a multi-building cohort. Independent validation against actual metered EUIs confirms all six models fall within CBECS sector ranges, at an average inference cost of $0.09 per building.A three-layer sensitivity and uncertainty analysis (Morris screening, Sobol decomposition, Monte Carlo propagation) reveals a cohort-level observability gap: equipment density accounts for 80--99% of annual energy-use variance, while envelope parameters, most directly recoverable from geometry, contribute negligibly. This challenges common geometry-driven UBEM assumptions, indicating calibration effort is better spent characterizing operational loads than refining envelope parameters. Collectively, these results establish a scalable, low-cost pathway to standards-consistent UBEM calibration for urban stocks lacking metered data.
- Open Access
- 22 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 374-384
Scan2UBEM: An AI-driven End-to-End Pipeline Towards Automated Urban Building Energy Modeling from 3D Scans
- 22 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 374-384
Open AccessScan2UBEM: An AI-driven End-to-End Pipeline Towards Automated Urban Building Energy Modeling from 3D Scans