A hybrid methodology using VMD and disentangled features for wind speed forecasting

Abstract

Wind energy is gaining worldwide attention due to its renewable and ecological characteristics. The accurate prediction of wind speed presents a challenge due to its unpredictable and stochastic nature. In this study, a novel method for predicting wind speed is proposed. The proposed approach is called VMD-CoST-SVR and it involves the use of variational mode decomposition (VMD), contrastive learning of seasonal-trend representations (CoST), and support vector regression (SVR). The study utilizes the VMD technique to denoise the wind speed data. The CoST model, which employs contrastive learning, is then applied to the denoised wind speed data to extract the disentangled trend and seasonal features. The concatenated feature map of trend and seasonal features is subsequently fed to the SVR to predict the future wind speed. As the time interval increases, there is a corresponding decrease in the performance of predictive models. Addressing this, the proposed methodology has been subjected to training and validation for wind speed predictions at various time intervals, namely 5-min, 10-min, 15-min, 30-min, 1-hour, and 2-hour ahead using two experiments. This study utilizes wind speed data obtained from wind farms situated in Leicester and Portland. This validation has been accomplished through the use of seven individual models and seven hybrid models. The results of two experiments indicate that the proposed approach exhibited superior performance, resulting in a significant improvement across all time intervals evaluated.