Physics-informed neural network modelling of wind power production

Abstract

With the large-scale construction of wind power production facilities, accurate prediction of wind power production has become increasingly important. Traditional data-driven methods struggle to effectively incorporate physical boundaries and handle noisy data, while real-world wind power systems are fraught with uncertainty due to factors such as air quality and wind speed. Machine learning models offer advantages in speed and accuracy in prediction, and the physics-informed neural network (PINN) framework maintains physical consistency while exhibiting considerable predictive performance. This bachelor’s thesis investigates the application of PINNs for wind power prediction and presents a MATLAB-based PINN example. This simplified MATLAB-based PINN model is used for accurate wind power production prediction, comparing it with standard neural network models. This approach provides a solution that balances computational accuracy and physical interpretability for improving wind power grid connection capabilities, laying a solid foundation for supporting energy transition goals.