# Smart Buildings Dataset SB1-SB11: HVAC Building Automation Time Series

## Facts

- Provider: Goldfeder, Dean, Jiang, Wang, Lipson, Sipple et al. (Google and Columbia University)
- License: CC-BY-4.0 (https://huggingface.co/datasets/Jgoldfeder/smart-buildings)
- Size: 331 files
- Format: JSON, NumPy, CSV (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/Jgoldfeder/smart-buildings
- Access: public, readable without an account
- Category: Energy
- Source: https://arxiv.org/abs/2410.03756
- Page: https://scigantic.com/archive/d179f0e1-a063-5118-9f5e-9e3073fc500d
- Keywords: Energy, time series, JSON, NumPy, CSV, Hugging Face

## About

Operational building automation system (BAS) data from 11 buildings, 2018 to 2026, for HVAC energy modelling, fault detection, forecasting and reinforcement-learning control (Goldfeder et al., arXiv 2410.03756). SB1 to SB3 are Google office buildings in Mountain View, CA at 5-minute steps (2019-2024, 3,136 to 13,610 channels each) with RL-style observations, actions, rewards and reward_info; SB4 to SB6 are academic, residential retrofit and laboratory buildings in Syracuse, NY (1 to 5 minute steps); SB7 to SB11 are campus academic, dining and residence-hall buildings in Needham, MA at 15-minute steps. Channels cover sensors, setpoints, commands and equipment status such as zone air temperature, supply air flow, damper and valve positions and chiller and boiler status, recorded in native units without resampling. Layout: buildings.csv, a shared channel table points.parquet, and SBx/timeseries/ / .parquet wide tables plus irregular and auxiliary tables, provenance.json and floor plans; 331 files, about 6.3 GB. Read with pandas or the included loader.py. Licensed CC BY 4.0 by the authors; SB1-SB3 derive from the Google Smart Buildings data, which should also be credited.

## Load the data

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="Jgoldfeder/smart-buildings", filename="SB1/metadata/floorplan.npy", repo_type="dataset")
print(path)
```

## Data structure

File types in sampled files: JSON, NumPy, CSV

Example file: `SB1/metadata/floorplan.npy`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/d179f0e1-a063-5118-9f5e-9e3073fc500d
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/d179f0e1-a063-5118-9f5e-9e3073fc500d
- Over MCP: `get_schema_card` with id `d179f0e1-a063-5118-9f5e-9e3073fc500d` at https://api.scigantic.com/mcp

## Related topics

- [JSON datasets](https://scigantic.com/topics/json)
- [CSV datasets](https://scigantic.com/topics/csv)
- [Time series datasets](https://scigantic.com/topics/time-series)
- [Energy datasets](https://scigantic.com/topics/energy)
- [NumPy datasets](https://scigantic.com/topics/numpy)
