Multi-Step Wind Speed Forecasting Based on Hybrid Deep Learning Model and Trailing Moving Average Denoising Technique

Abstract

Nowadays the world is in desperate need for an alternative energy, due to the climate changes and the environmental damages caused by the abused use of traditional energy resources. In this paper we focused on the wind speed prediction for the wind energy generation purpose. A hybrid wind speed model was developed based on the Trailing Moving Average pre-processing method (TMA) and the Encoder-Decoder-Convolutional Neural Network-Long Short Term Memory (CNN-LSTM-ED) prediction model. The TMA technique was used for smoothing the wind speed data followed by the CNN-LSTM-ED hybrid deep learning network for the wind speed forecasting. The experimental results proved that the proposed hybrid model outperformed the benchmarks models being the LSTM, CNN, CNN-LSTM, LSTM-ED and CNN-LSTM-ED models respectively and provided the best performance and the more accurate wind speed forecast.