Improved Prediction of Wind Speed using Machine Learning

Abstract

The prediction of wind speed plays a significant role in wind energy systems. An accurate prediction of wind speed is more important for wind energy systems, but it is difficult due to its uncertain nature. This paper presents three artificial neural networks namely, Back Propagation Network (BPN), Radial Basis Function (RBF) and Nonlinear AutoRegressive model process with eXogenous inputs (NARX) with Mutual Information (MI) feature selection for wind speed prediction. The MI feature selection identifies the significant features and reduces the complexity of wind speed prediction model without loss of information content. The performance of prediction model is evaluated in terms of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The results show that the performance of all three neural network models are highly satisfied. Moreover, NARX model with mutual information feature selection is more accurate in dealing with wind speed prediction.