Comparative performance study of DBN, LSTM, CNN and SAE models for wind speed and direction forecasting

Abstract

More evenly distributed renewable energy sources are emerging as an alternative to fossil fuels, especially for power generation in rural areas. This study compares four of the most widely used models, the Convolutional Neural Network (CNN), the Long Term Memory (LSTM), the Stacked Autoencoder (SAE) and the Deep Belief Network (DBN), for forecasting wind speed and direction in order to control the output of a wind turbine generator. The meteorological data used were downloaded from the NASA database for the period January 1, 2012 to December 31, 2021 on the Benin coast. The forecasting model developed combines one of the LSTM, CNN, DBN, SAE models and the Fast Fourier Transform, which was used to extract the most important frequencies in the data. The results of the comparison showed that the best-performing model was the LSTM, with respective root mean square errors (RMSE) and coefficients of determination of 0.36 m/s and 92% for wind speed, and 8.1 ° and 71% for wind direction.