Forecasting of wind speed using multiple linear regression and artificial neural networks

In this paper, two methods are developed for the prediction of wind speed, namely, the Multiple Linear Regression (MLR) and Artificial Neural Networks (ANNs) in north and south regions of Morocco for three years (i.e., 2011-2012-2013). The first method consists of determining the parameters which most significantly influence the wind speed in order to build a regression model between the predictors and the dependent variable. The second proposed approach is ANNs where the neural network chosen is the multilayer perceptron that uses the back-bropagation (BP) as a supervised learning technique for training. The results show that both MLR and ANNs models predict the wind speed at an acceptable correlation coefficient between the actual and predicted wind speed. However, the ANNs perform better in terms of statistical errors notably in terms of mean absolute error, mean absolute percentage error and mean square error.