This repository includes Python notebooks to build and train deep-learning models for prediction of the power generated by wind turbines, using data from 4 turbines at ‘La Haute Borne’ wind farm. The original dataset was pre-processed to remove non-physical and anomalous data points, and a reduced dataset with the interest variables $(v,\beta,\omega,T,P,C_p)$ is available at the Dataset directory. Pre-trained regression models for the power coefficient, mechanical torque and generated power can be found within the Models folder. The implementation and training of the models can be consulted at the Scripts directory, containing notebooks for standard artificial neural networks (NNs), physics-informed neural networks (PINNs) and neural networks with an evidential layer for uncertainty quantification.

While NN models learn only from data, PINNs are able to reproduce both data and some physical constraints, expressed by the equations:

Indroducing an evidential output layer provided efficient and solid uncertainty quantification of the predictions, making possible the definition of confidence intervals in the power curve:

Alfonso Gijón, Miguel Molina-Solana, Juan Gómez-Romero

🔗 https://arxiv.org/abs/2307.14675

@article{Gijon2023_WindTurbines,
  title={Prediction of Wind Turbines Power with Physics-Informed Neural Networks and Evidential Uncertainty Quantification},
  author={Gij{\'o}n, Alfonso and Pujana-Goitia, Ainhoa and Perea, Eugenio and Molina-Solana, Miguel and G{\'o}mez-Romero, Juan},
  journal={arXiv preprint arXiv:2307.14675},
  year={2023},
  url={https://arxiv.org/abs/2307.14675}
}