New Coupled Wavelet-Random Forest Method for Wind Speed Prediction

Abstract

Accurate prediction of wind speed records is an important task in various disciplines including agriculture, meteorology, climatology, navy, energy studies, wind power, etc. Although some traditional models have been suggested and applied for wind forecast, machine learning (ML) approaches can be suitable alternatives for such models due to their successful performance inn wide range of subjects and phenomena. On the other hand, using ML techniques alone might not be suitable/successful in all cases especially when the studied data series has strong time dependency and the series show clear periodicity. So, applying wavelet transform to resolve the issue might be a good choice to increase the generalization ability of the ML techniques. The present study aimed at assessing the performance of the random forest (RF) method for predicting daily wind speed records at four sites in Iran. The wavelet transform was used for producing new sub series of data and make the new wavelet- random forest (WR) models. Both the RF and WR models were fed with the previously recorded wind speed values with different lag times. The obtained results revealed that the WRF has improved the performance of the RF model inn all studied locations, considerably.