Can Large Language Model Agents Balance Energy Systems?

Code Author: Zekun Guo (https://drzekunguo.github.io/), Xinxing Ren (https://github.com/renxinxing123)

This repository contains the codebase for leveraging Large Language Model (LLM) Agents to assist in decision-making for power system operations. It includes Jupyter Notebook and MATLAB scripts for various tasks related to power system balance.

*ChatGPT o1 and o3-mini-high assisted with debugging and commenting on this code.

Repository Structure

• Prob Generator.ipynb: A Jupyter Notebook utilizing Microsoft's Autogen framework for probabilistic generation and system operation tasks.
• MATLAB Codes: Scripts that rely on optimization libraries for power system modeling and decision-making.

Updates on 30th March 2025

A new 10-trial experiment has been implemented to evaluate the performance of the LLM-assisted Stochastic Unit Commitment (LLM-SUC) framework. Experimental results demonstrate that the LLM-SUC approach achieves a cost reduction of approximately 1.1–2.7% and lowers load curtailment by around 26.3% compared to the traditional SUC method.

The probability generation code is now all in the 'Prob_MoreTest.py' file.

Getting Started

Setting Up the Environment for Prob Generator.ipynb

To run the Prob Generator.ipynb, follow these steps:

1. Create a Conda Environment:

   conda create -n autogen-env python=3.10

2. Install the autogen-agentchat Package:

   pip install autogen-agentchat~=0.2

3. Create a Jupyter Notebook Kernel:

   ipython kernel install --user --name=Autogen-kernel

4. Run the Notebook:

   • Open the Prob Generator.ipynb in Jupyter Notebook.
   • Ensure you select the Autogen-kernel before running the notebook.

Dependencies for MATLAB Codes

The MATLAB scripts in this repository require the following dependencies:

1. YALMIP: Installation instructions are available here: YALMIP Installation Guide

2. Gurobi Optimizer: Ensure you have a valid Gurobi license and have installed the Gurobi MATLAB interface. Visit the Gurobi website for more details.

The Senario Tree can be generated with the code from this repo: https://github.com/badber/StochasticUnitCommitment/tree/master/scenario_tree

Citation

If you find this repository useful, please consider citing the following paper:

@misc{ren2025largelanguagemodelagents,
      title={Can Large Language Model Agents Balance Energy Systems?}, 
      author={Xinxing Ren and Chun Sing Lai and Gareth Taylor and Zekun Guo},
      year={2025},
      eprint={2502.10557},
      archivePrefix={arXiv},
      primaryClass={eess.SY},
      url={https://arxiv.org/abs/2502.10557}, 
}

Contributing

Contributions are welcome! If you encounter any issues or have suggestions for improvements, feel free to open an issue or submit a pull request.

License

This repository is licensed under the MIT License. See the LICENSE file for more details.