This repository contains the code for the paper "Online Event-Triggered Switching for Frequency Control in Power Grids with Variable Inertia", submitted to Transactions on Power Systems.

This project implements a Neural-PI controller framework for frequency control in power grids with variable inertia. The framework uses reinforcement learning to train controllers that adaptively switch based on system events, ensuring stability and performance under varying conditions.

• online-event-triggered-switching-neural-pi.ipynb: Jupyter notebook containing the main code for training and evaluating the Neural-PI controllers.
• data/: Directory containing the simulation data, such as the IEEE 39-bus system data.
• checkpoints/: Directory containing trained model checkpoints.

To run the code in this repository, you will need the following libraries:

• matplotlib
• numpy
• tensorflow
• networkx
• scipy
• mat4py
• pickle
• time

You can install the required libraries using pip:

pip install matplotlib numpy tensorflow networkx scipy mat4py

1. Set Up the Environment: Ensure you have a suitable Python environment with the required libraries installed.

2. Load the Notebook: Open the online-event-triggered-switching-neural-pi.ipynb notebook in Jupyter.

3. Run the Cells: Execute the cells in the notebook to train the Neural-PI controllers and evaluate their performance on the provided simulation data.

The Frequency class in the notebook defines the environment for the frequency control problem. It initializes system parameters and defines the state transition dynamics.

Simulation data is loaded from the IEEE 39-bus system using the mat4py library. Key parameters include system inertia, damping, and network topology.

The notebook provides a detailed implementation of the training procedure for the Neural-PI controllers, including the reinforcement learning setup and evaluation metrics.

The notebook includes plots and metrics to demonstrate the performance of the trained controllers under different inertia modes. The best performance for each mode is highlighted in bold in the results section.

If you use this code in your research, please cite our paper once it is published.

This project is licensed under the MIT License - see the LICENSE file for details.