rl-opf-env-design

KO | EN

gitlite — search

#javascript #python #react #ai #llm #rust #linux #docker #nodejs #cli #kubernetes #pytorch #mcp #ios

rl-opf-env-design

(★ 9)

Repository for the publication "Learning the Optimal Power Flow: Environment Design Matters".

rl-opf-env-design 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Learning the Optimal Power Flow: Environment Design Matters
This is the accompanying repository to the publication "[Learning the Optimal Power Flow: Environment Design Matters](https://www.sciencedirect.com/science/article/pii/S2666546824000764)" by [Thomas Wolgast](https://orcid.org/0000-0002-9042-9964)
 and [Astrid Nieße](https://orcid.org/0000-0003-1881-9172). 
 
If you are interested in this research, you might also want to check our follow-up work where we expanded our research in all directions: "[A General Approach of Automated Environment Design for Learning the Optimal Power Flow](https://dl.acm.org/doi/full/10.1145/3679240.3734626)"

Note that most source code implemented for this paper can be found in 
https://gitlab.com/thomaswolgast/drl (RL algorithms) and 
https://github.com/Digitalized-Energy-Systems/opfgym 
(RL environments, warning: repository re-named and moved from GitLab to GitHub), 
which are work-in-progress and therefore continued in different repositories.

Especially the environment framework was significantly improved in the meantime.
Please visit the GitHub repository above if you want to work with the most 
recent version.

# Installation
All experiments were performed with python 3.8. In an virtualenv, run `pip install -r requirements.txt` to install all dependencies in the right version at publication time (not the most recent).

Note: [torch](https://pytorch.org/get-started/locally/) sometimes makes problems and needs to be installed manually before performing the previous step.

# Repository structure
- `run.sh`: A list of the commands performed to reproduce all experiments done for this publication. **Should not be run all at once!** Overall computation time will be multiple weeks. Use this file to copy-paste single commands from. Will automatically create a `data/` folder with the result files. 
- `LICENSE`: The license used for this work (MIT). 
- `requirements.txt`: Reference to the two previously mentioned repositories for simple installation. 
- `src/`: The source code to aggregate the results and create the figures of this exact publication. The source code for running the experiments is in external repositories.
- `compact_results/`: Collection of JSON files with compact representations of the results with different metrics. (created with `create_results.py` and used by `boxplots.py`)
- `results/`: **[Optional folder that can be downloaded from Zenodo](https://zenodo.org/records/13284446)**. Contains model weights, hyperparameter information, and agent performances in the course of training (e.g. MAPE every 30k steps). The model weights are required to run `create_results.py`. The training performances are required to plot performance during training, if required. Download this folder from Zenodo and place it in this repository to work with the data.

Attention: Sometimes, the VoltageControl environment is called QMarketEnv because of a name change during development. Both names are synonymous in the context of this work.

# Contact
For questions or feedback, contact the first author Thomas Wolgast (thomas.wolgast@uol.de).

rl-opf-env-design

rl-opf-env-design 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

rl-opf-env-design

rl-opf-env-design 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README