One day ahead prediction of wind speed class by statistical models

This paper deals with the clustering of daily wind speed time series based on two features, namelythe daily average wind speed and the corresponding degree of fluctuation. Daily values of the feature pairs are first classified by means of the fuzzy c-means unsupervised clustering algorithm and then results are used to train a supervised MLP neural network classifier. It is shown that associating to a true wind speed time series a time series of classes allows performing some useful statistics. Further, the problem of predicting the class of daily wind speed 1-step ahead is addressed by using both theHidden Markov Models (HMM) and theNon-linear Auto-Regressive (NAR) approaches. The performances of the considered class prediction models are finally assessed in terms of True Positive rate (TPR) and True Negative rate (TNR), also in comparison with the persistent model.