Inajara Rutyna
supervisor: Paweł Piotrowski
Forecast models used to predict wind energy are considered chaotic by nature due to their dependence on weather conditions, which can imply in complex formulations.
The formulation of wind forecasting models is based on historical data obtained from Numerical Weather Predictions (NWP) and time series, represented by features directly connected to the wind in different steps of time.
The classical approaches used for wind forecasting are categorized by naive methods, physical models, statistical methods and artificial intelligence techniques.
Machine learning as a subset of artificial intelligence is considered the most effective technique for wind forecasting. It is able to obtain higher accuracy in predictions, by mimicking human behavior without a predefined mathematical model, and is used to generate a model for future generation. However, machine learning techniques also have its disadvantages, such as, low convergency speed, overfitting, computational complexity and problem generalization.
To avoid these problems, ensemble and hybrid methodologies combined with machine learning techniques are able obtain best prediction performance, by aggregating different methodologies in weak spots of the model.