Machine learning and the short-term Forecast for maximum wind speed

Abstract

Events of great intensity of winds, such as the passage of cold fronts, the occurrence of cyclones, bomb cyclones, among others, are frequent in southern Brazil. The region also stands out for the production of electricity from the winds, which depends on a stochastic resource and, therefore, there will always be uncertainty about the final production. Considering this scenario, the present work aimed to analyze the performance of machine learning models for predicting the maximum wind speed one hour ahead. Linear regression was the method with the lowest predictive capacity. The artificial neural network and extreme gradient boosting (XGBoost) performed better in terms of metrics. The results obtained demonstrated promising prediction capacity. These represent a possibility of innovation in the prediction of maximum wind speeds, especially when inserted in places where different phenomena with extreme winds occur.