A Wavelet, AR and SVM based hybrid method for short-term wind speed prediction

Wind speed modelling and prediction has been gaining importance because of its significant roles in various stages of wind energy management. In this paper, we propose a hybrid model, based on wavelet transform to improve the accuracy of the short-term forecast. The wind speed time series are split into various frequency components using wavelet decomposition technique, and each frequency components are modelled separately. Since the components associated with the high- frequency range shows stochastic nature, we modelled them with autoregressive (AR) method and rest of low-frequency components modelled with support vector machine (SVM). The results of the hybrid method show a promising improvement in accuracy of wind speed prediction compared to that of stand-alone AR or SVM model.

Wind speed forecast using random forest learning method

Wind speed forecasting models and their application to wind farm operations are attaining remarkable attention in the literature because of its benefits as a clean energy source. In this paper, we suggested the time series machine learning approach called random forest regression for predicting wind speed variations. The computed values of mutual information and auto-correlation shows that wind speed values depend on the past data up to 12 hours. The random forest model was trained using ensemble from two weeks data with previous 12 hours values as input for every value. The computed root mean square error shows that model trained with two weeks data can be employed to make reliable short-term predictions up to three years ahead.