This project implements a federated learning approach using LSTM (Long Short-Term Memory) neural networks for time series forecasting. The system is built using TensorFlow and Flower (flwr) for federated learning.

This federated learning system allows multiple clients to collaboratively train an LSTM model without sharing their raw data. The model is designed to forecast time series data, specifically tailored for energy-related predictions.

1. LSTM Model: Defined in lstm_model.py
2. Federated Learning Client: Implemented in client.py
3. Data Preparation: Handled by prepare_data.py
4. Federated Learning Server: Set up in server.py
5. Main Execution Script: run_federated_learning.py

git clone https://github.com/KaavinB/Wind-Prediction-LSTM-Federated-Learning.git

pip install tensorflow pandas numpy scikit-learn flwr

3. Ensure jandata.csv is in the project directory.
4. Start the server: python server.py
5. Run clients in separate terminals: python run_federated_learning.py

• client.py: Federated learning client (Flower's NumPyClient)
• lstm_model.py: LSTM model architecture and utilities
• prepare_data.py: Data loading, preprocessing, and splitting
• run_federated_learning.py: Main script to start a client
• server.py: Federated learning server setup and execution

Use jandata.csv with columns:

• Datetime
• Region
• Grid connection type
• Offshore/onshore
• Most recent forecast (target variable)
• Other relevant features

• Adjust client numbers, LSTM architecture, or training parameters in respective files
• Modify data preprocessing in prepare_data.py for different datasets

Contributions welcome! Fork the repo and submit a pull request with your changes.

Built with Flower - a friendly federated learning framework.

Made with ❤️ by Kaavin