---
license: mit
language:
- en
pretty_name: MAPDN
---

This dataset is for simulating the multi-agent reinforcement learning (MARL) environment for distributed active voltage control problem, a key research area to guarantee the safety of voltage in the future power grid. The GitHub repository for the environment is [MAPDN](https://github.com/Future-Power-Networks/MAPDN). If you would like to run the MARL environment, you require to first download the dataset held in this dataset storage.

Briefly speaking, the user load data is modified based on the real-time Portuguese electricity consumption accounting for 232 consumers for 3 years, while the PV data is collected from Elia group, i.e. a Belgiums power network operator for 3 years. To distinguish among different solar radiation levels in various regions, the 3-year PV generations from 10 cites/regions are collected and PVs in the same control region possess the same generation profiles.

One motivation for us to upload the data to Hugging Face is that we are curious about whether LLMs are capable of dealing with the time-series forecasting for the real-world user electricity consumption which are beneficial for us to understand users behaviours, and the real-world PV generation which is beneficial for us to estimate intermittent renewable power generation influenced by weather. Success of both directions can help develop more convincing control/AI algorithms for regulating future power grids.


The `voltage_control_data.zip` include the PV (solar panel) generator data and user load data for MARL simulation. 
The `traditional_control_data.zip` with the same data is for replicating the droop control and OPF methods in MATLAB.

If you either use the dataset or run the MARL environment, please cite the following paper published in NeurIPS 2021:
```
@inproceedings{NEURIPS2021_1a672771,
 author = {Wang, Jianhong and Xu, Wangkun and Gu, Yunjie and Song, Wenbin and Green, Tim C},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {M. Ranzato and A. Beygelzimer and Y. Dauphin and P.S. Liang and J. Wortman Vaughan},
 pages = {3271--3284},
 publisher = {Curran Associates, Inc.},
 title = {Multi-Agent Reinforcement Learning for Active Voltage Control on Power Distribution Networks},
 url = {https://proceedings.neurips.cc/paper/2021/file/1a6727711b84fd1efbb87fc565199d13-Paper.pdf},
 volume = {34},
 year = {2021}
}
```

If you have any question, please feel free to contact `jianhong.wang@bristol.ac.uk`.