README.md

metadata

pretty_name: TTM4HVAC – Training dataset (source-default)
tags:
  - ttm4hvac
  - hvac
  - time-series
  - energy
task_categories:
  - time-series-forecasting
license: mit
papers:
  - title: >-
      Transfer learning of building dynamics digital twin for HVAC control with
      Time-series Foundation Model
    url: https://doi.org/10.1016/j.mex.2026.103866
    authors: Ferran Aran Domingo

TTM4HVAC – Training dataset (source-default)

This dataset contains HVAC and weather time-series data collected under default building control schedules for the source domain.

It is used to train the gft/ttm4hvac-source-default model.

Check out the paper A reproducible method to generate multi-building, multi-climate HVAC operation datasets with a stochastic exploratory controller and visit the main repository ttm4hvac for further details.

Columns

• time
• Outdoor Air Temperature (C)
• Heating Setpoint (C)
• Cooling Setpoint (C)
• Room Air Temperature (C)
• Outdoor Humidity (%)
• Wind Speed (m/s)
• Direct Solar Radiation (W/m^2)
• HVAC Power Consumption (W)
• series_id
• is_default

Usage

from datasets import load_dataset

ds = load_dataset("gft/ttm4hvac-source-default-train")
df = ds["train"].to_pandas()
df.head()

✒️ Citation

If you use this model or datasets, please cite:

@article{aran_domingo_2026_hvac_dataset,
  title   = {A reproducible method to generate multi-building, multi-climate HVAC operation datasets with a stochastic exploratory controller},
  author  = {Aran Domingo, Ferran and
             Fraile Alonso, Pablo and
             Rius Torrentó, Josep and
             Agost Batalla, Oriol and
             Barri Vilardell, Ignasi and
             Vilaplana Mayoral, Jordi and
             Mateo Fornés, Jordi},
  journal = {MethodsX},
  volume  = {16},
  pages   = {103866},
  year    = {2026},
  doi     = {10.1016/j.mex.2026.103866},
  url     = {https://doi.org/10.1016/j.mex.2026.103866}
}