This repository contains a Jupyter Notebook demonstrating Non-Intrusive Load Monitoring (NILM) using Federated Learning. The goal is to disaggregate household energy consumption data while preserving user privacy by leveraging decentralized machine learning techniques.

• Implementation of Federated Learning for NILM through FedAVG, FedSR, FedADG, and FedDA.
• Uses real-world energy datasets.
• Privacy-preserving approach by keeping user data on local devices.
• Training across multiple clients using PyTorch/Federated Learning frameworks.
• Performance metrics and visualization for model evaluation.

The notebook supports various publicly available NILM datasets such as:

• IRISE
• REFIT

Users can modify the dataset path in the notebook to work with different datasets.

To run the notebook, ensure you have Python installed along with the required dependencies. You can set up the environment using:

# Clone the repository
git clone https://github.com/yourusername/NILM-Federated-Learning.git
cd NILM-Federated-Learning

# Create a virtual environment (optional)
python -m venv env
source env/bin/activate   # On Windows use `env\Scripts\activate`

# Install dependencies
pip install -r requirements.txt

Start Jupyter Notebook and open the .ipynb file:

jupyter notebook

Run the cells step-by-step to train and evaluate the federated learning NILM model.

The notebook implements a deep learning model using CNNs for NILM disaggregation.

• The notebook provides visualizations for training loss, accuracy, and disaggregation results.
• Federated Learning performance is compared with centralized NILM models.
• Metrics include Mean Absolute Error (MAE), Normalized Root Mean Sqruared Error (NRMSE), and Energy Disaggregation Accuracy.

Contributions are welcome! Please fork the repository, create a branch, and submit a pull request with improvements.

For any queries or discussions, feel free to reach out via GitHub Issues or email Fares.alkhawaja@gmail.com.