Wavelet Transform and LSTM to Improve Wind Energy Forecasting

The present work deals with the improvement of short-term wind energy forecasting techniques by combining time series decomposition techniques (Wavelet Transform) and Deep Learning recurrent models (LSTM). The wavelet transform is used to decompose the wind speed into several more stationary components and then the LSTM model is used to forecast each component individually. The inverse wavelet transform is then used to reconstruct the predicted wind speed. Once the wind speed has been predicted, a mathematical model for estimating wind energy can be used to obtain the corresponding wind power. This combination of Wavelet Transform and LSTM gave better results (RMSE of 0.7 m/s and R2 of 0.81) than using the LSTM alone. In addition, it was found that the more the 0 component from the wavelet transforms is successively decomposed, the more accurate the model becomes. For this study, velocity data at 50 m above ground level were downloaded from the NASA database for the Benin coast and the models were implemented using Python 3.