Short-term Wind Speed Forecasting using Machine Learning Algorithms

This paper evaluates the performance of several machine learning algorithms for short-term wind speed forecasting. The algorithms evaluated include: Long Short-Term Memory, Extra-Tree, Gradient Boosting Tree, Extreme Gradient Boosting Tree, Voting Averaged, Multi-layer Perceptron, K-Nearest Neighbors, and Support Vector Machine. The performance of the algorithms was evaluated with different error metrics using real wind speed and meteorological data collected from the city of Maceio, Brazil. First, pre-processing methods are applied in the large database to deal with outliers, noisy and missing values. Then, variable selection technique is employed to select the most significant set of variables and their lag-values as input to the forecast algorithm. Results show Voting Averaged algorithm performs better for all forecast time horizons considered, which are 1 hour, 2 hours and 3 hours ahead.