Data and Code for “Cross-Building Transferability of Occupant-Centric HVAC Digital Twins: Generalization, Uncertainty-Aware Adaptation, and Multi-Objective Control”
Description
This repository contains the processed analysis-ready datasets, reproducible analysis code, derived numerical outputs, and supplementary material supporting the study “Cross-Building Transferability of Occupant-Centric HVAC Digital Twins: Generalization, Uncertainty-Aware Adaptation, and Multi-Objective Control.” The operational analysis uses six pre-frozen educational buildings from Building Data Genome Project 2 spanning three site/climate contexts and evaluates all 30 directed source-to-target cross-building transfers. The repository supports the matched 300-tree XGBoost H1 analysis, physics-guidance H2 ablation, recent-window target adaptation, pooled-source learning, conformal uncertainty calibration and rolling recalibration, seasonal and extreme-condition tests, forecast-horizon sensitivity, Temporal Transformer and CORAL benchmarks, and deployment-safe domain-distance analysis. A separate processed CU-BEMS layer supports measured thermal modelling, effective 1R1C and RC+ML hybrid evaluation, cross-floor transfer, and EKF-style state synchronization. Reduced-order closed-loop simulations compare rule-based, scheduled, deterministic hard-constrained MPC, and uncertainty-robust MPC under normal, heatwave, occupancy-surge, ventilation-restriction, and model-uncertainty scenarios, with energy, thermal comfort, CO₂-related IAQ, peak-demand, cycling, and load-ramp outcomes. The repository includes processed analytical subsets, final code, result tables, supplementary material, environment specifications, data dictionary, reproduction instructions, provenance notes, and SHA-256 integrity checks. The complete upstream raw BDG2 and CU-BEMS datasets are not redistributed; source citations and DOI information are provided in the repository. Controller energy and indoor-condition results are reduced-order simulated closed-loop outcomes and should not be interpreted as live-building savings. CO₂ is used as an IAQ-related proxy rather than a complete characterization of indoor air quality.