Özet:

The fluctuated nature of wind makes it a very challenging phenomenon to track where making an accurate forecast of one of its parameters requires a robust and reliable model. In this study we will focus on the wind speed forecast for wind energy generation purpose which is a very delicate process that requires an accurate prediction results. The wind speed prediction is considered as one of the highest complexity time series problems where the studies proved the efficiency of Recurrent Neural Network (RNN) models and specifically the Long Short Term Memory (LSTM) model that provides accurate prediction with the capacity to handle long-term dependencies. In this paper an hourly wind speed forecasting model was proposed based on Fast Fourier Transform Filter and Encoder-Decoder-LSTM model (FFT-Encoder-Decoder-LSTM), the FFT Filter was used for Data Denoising pro-cess then Max-Min normalization technique was applied to standardize the data and finally the Encoder-Decoder-LSTM model was used for the wind speed prediction. The traditional MPL, Single-layer-LSTM, Encoder-Decoder-LSTM, FFT-MLP and FFT-Single Layer LSTM model were used as benchmark models. While accentuating the effectiveness of data prepro-cessing step in the forecasting process, the efficiency of the models is evalu-ated for 1-hour and 3-hours ahead wind speed forecasting where the FFT-Encoder-Decoder-LSTM showed the best and the more consistent results.

Anahtar Kelimeler: