Simulation data and frozen-model analysis code for cross-city transfer of residential cooling and heating loads
Description
This dataset supports the study "A Shanghai-based model for cross-city transfer of residential cooling and heating loads with building-climate interactions". It provides simulation data and analysis code for the Mechanism-Informed Cross-city Transfer model (MICT). The data include Shanghai source-domain training and test sets with 2,048 and 512 building configurations; development data for 50 cities with 256 shared building configurations per city; and external evaluation data for 15 city-weather cases with 256 configurations per case (3,840 cases per load target). Cooling and heating are separate targets for the same cases. External cases include weather-record changes at represented locations and cases from additional climate subzones. External load labels are used only for evaluation after model freeze. Annual ideal cooling and heating loads were simulated with EnergyPlus and are expressed in kWh per square metre per year. The release also includes climate descriptors for 66 weather cases, Morris screening summaries for 14 cities, locked outer validation partitions, numerical frozen-model parameters, archived predictions, supplementary tables, and source data and scripts for Figures 1-13 and Supplementary Figure S1. The Python code reproduces frozen MICT correction inference for the 256 archived building configurations, recalculates external performance metrics and regenerates the figures. Shanghai source-model predictions for these configurations are supplied as data. The release does not rerun complete EnergyPlus simulations, source-model fitting, benchmark training or nested feature selection. Third-party EPW files are identified through source metadata and repository links but are not redistributed. README.txt and DATA_DICTIONARY.md explain the data and reproduction scope; FILE_MANIFEST.csv describes individual archived files. Data tables, numerical model parameters, results and documentation are licensed under CC BY 4.0. Original Python source files in code/ and figures/ are licensed separately under MIT, as specified in LICENSE_CODE_MIT.txt.
Files
Steps to reproduce
1. Extract MICT_Data_and_Code.zip and preserve its folder structure. Read README.txt, DATA_DICTIONARY.md, FILE_MANIFEST.csv and code/README.md. 2. Use Python 3.12. From the extracted release root, install the analysis dependencies: python -m pip install -r code/requirements.txt 3. Reproduce the frozen MICT predictions and verify the external evaluation metrics: python code/validate_results.py Results and the verification report are written to results/recomputed/. Benchmark metrics are recalculated from archived predictions; benchmark models are not retrained. 4. Install the figure dependencies and regenerate Figures 1-13 and Supplementary Figure S1: python -m pip install -r figures/requirements.txt python figures/reproduce_all.py --output-dir reproduced_figures 5. Preserve the source, development and external data roles. External simulated loads are evaluation labels only. The README specifies the reproduction scope; these commands do not rerun the complete original simulation and model-selection pipeline.