chandar-lab/remote_microgrid_data

Remote Microgrid Time-Series Dataset

This dataset provides one-minute–resolution time-series traces from a remote wind–diesel–battery microgrid. It contains normalized historical values and short-term forecasts for available wind power and electrical demand. The data are anonymized but preserve the temporal structure needed for research on microgrid control and forecasting.

---

Contents

Parquet files included:

• norm_avail_wind_power_data.parquet — Realized available wind power
• norm_avail_wind_power_forecast.parquet — Forecasts of available wind power (DateTime, DateTimeUpdated)
• norm_demand.parquet — Realized electrical demand
• norm_demand_forecast.parquet — Demand forecasts (DateTime, DateTimeUpdated)
• data_exploration.ipynb — Example notebook for loading and visualizing the data

All series are normalized; see the notebook for details.

---

Usage

import pandas as pd

wind = pd.read_parquet("norm_avail_wind_power_data.parquet")
demand = pd.read_parquet("norm_demand.parquet")
wind_fc = pd.read_parquet("norm_avail_wind_power_forecast.parquet")
demand_fc = pd.read_parquet("norm_demand_forecast.parquet")

---

Citation

@article{nekoei2025shielded,
  title   = {Shielded Controller Units for RL with Operational Constraints Applied to Remote Microgrids},
  author  = {Nekoei, Hadi and Blondin Massé, Alexandre and Hassani, Rachid and Chandar, Sarath and Mai, Vincent},
  journal = {arXiv preprint arXiv:2512.01046},
  year    = {2025}
}

---

Datasets licensed under the Creative Commons Attribution-NonCommercial 4.0 International License.