🌬️ SpatioTemporal Wind Forecasting

An advanced Deep Learning pipeline for spatio-temporal wind speed forecasting using ConvLSTM, PredRNN, and a state-of-the-art Transformer model (PredFormer Fac-T-S). This project handles the entire workflow from data preprocessing to model training, hyperparameter tuning, and evaluation.

(Note: Please update the image link to a valid one for it to display correctly)

---

✨ Features

• Multi-Model Architecture: Implements and compares multiple deep learning models against a classical baseline:
  • ARIMA (Baseline)
  • FCNN (Fully Connected Neural Network)
  • ConvLSTM (Recurrent-Convolutional)
  • PredRNN (Spatiotemporal LSTM)
  • PredFormer (Fac-T-S) (State-of-the-art pure Transformer)
• End-to-End Pipeline: Automates the entire workflow, including data loading, preprocessing, training, and evaluation.
• Advanced Training Techniques: Utilizes modern methods for high performance and stability:
  • Automatic Mixed Precision (AMP) for faster training.
  • Gradient Clipping to prevent exploding gradients.
  • Configurable Optimizers (Adam, AdamW) and Schedulers (CosineAnnealing, ReduceLROnPlateau).
• Experiment Tracking: Fully integrated with MLflow for comprehensive logging of parameters, metrics, and visual artifacts.
• Rigorous Validation: Employs Time Series Cross-Validation to ensure robust model evaluation and prevent data leakage.
• Centralized Configuration: All experiments are managed through a single, easy-to-use modelos.json file.

---

📈 Performance Results

After a full cycle of training and hyperparameter optimization, PredFormer (Fac-T-S) emerged as the top-performing model on the test set, demonstrating superior accuracy.

Rank	Model	Best RMSE	Best MAE
🏆 1ˢᵗ	PredFormer (Fac-T-S)	1.30760	0.93202
🥈 2ⁿᵈ	PredRNN	1.33124	0.94050
🥉 3ʳᵈ	ConvLSTM	1.33959	0.94973
🏅 4ᵗʰ	FCNN	1.86136	1.35341
🏅 5ᵗʰ	ARIMA	2.57494	1.90345

---

🛠️ Technologies Used

• Python 3.8+
• PyTorch
• MLflow
• Xarray & NetCDF4
• Scikit-learn
• Pandas & NumPy
• Einops

---

🚀 Getting Started

Follow these steps to set up and run the project on your local machine.

1. Clone the Repository

git clone [https://github.com/seu-usuario/SpatioTemporal-Wind-Forecasting.git](https://github.com/seu-usuario/SpatioTemporal-Wind-Forecasting.git)
cd SpatioTemporal-Wind-Forecasting

2. Create Environment and Install Dependencies

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate

# Install the required packages
pip install -r requirements.txt

3. Configure the Experiment

• Open the modelos.json file.
• Set "ativo": true for the models you wish to train.
• Adjust hyperparameters as needed for your experiment.

4. Run the Main Pipeline

python main.py

5. Visualize Results with MLflow

• In a new terminal (from the same project directory), launch the MLflow UI:

mlflow ui

• Open your web browser and navigate to http://127.0.0.1:5000 to view and compare your experiment runs.

---

🔮 Future Work

• Containerization: Create a Dockerfile to package the environment and code for easy deployment.
• Inference API: Develop a FastAPI script to serve the champion model (.pth) via a REST API.
• CI/CD Pipeline: Automate retraining and deployment on a cloud platform (e.g., AWS, GCP, Azure).

---

📄 License

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