Wind parameter forecasting for wind turbines

Abstract

In recent years, the importance of clean and renewable energy has increased due to the rise of pollution and environmental degradation. In that context, alternatives such as wind, solar, wave, green hydrogen, or biomass stand out. Specifically, wind energy has been considered as one of the most promising alternatives. Nevertheless, wind energy is a quite unstable source, due to the continuous variation and random nature of wind and wind speed. The uncertainty generated by this energy production clearly affects its stability, and increases the cost of its productions. That is why accurate forecasting of wind positively affects wind energy development and its trade in certain markets. In this context, we have analyzed if gradient boosting (XGBoost), a very recent and powerful intelligent technique, can be used to obtain great results in wind prediction. Therefore, in this project we aim to model some wind features using XGBoost, and then compare them with the performance of other algorithms, such as Support Vector Regression (SVR), Gaussian Process Regression (GPR) and neural networks (NN) to see if it is a promising algorithm to consider. In this project we show that the results obtained confirmed our theory: XGBoost is in fact a good option to consider in wind forecasting. However, we have to keep in mind that the further we want to predict, the worse the accuracy of each model gets. Due to this, these algorithms are better for short-term forecasts.