Evaluating LSTM and NARX neural networks for wind speed forecasting and energy optimization in Tetouan, Northern Morocco

Abstract

Generating electricity from renewable sources is crucial for advancing toward a low-carbon economy, with wind power playing a significant role. Effective wind energy management is essential for meeting societal needs and protecting the environment. This study aims to optimize wind power production by improving the accuracy of wind speed predictions. Building on previous research comparing MLP, NARX, and Elman models for Tetouan City, we introduce a novel comparison between the nonlinear autoregressive with exogenous inputs (NARX) model and the long short-term memory (LSTM) network. Utilizing MATLAB, we analyzed 12 years of meteorological data from Tetouan City to determine which model provides the most accurate predictions. Our results reveal that the LSTM model significantly outperforms the NARX model, achieving lower values for mean absolute error (MAE = 0.18855), mean squared error (MSE = 0.0666), and root mean squared error (RMSE = 0.25808). This demonstrates the LSTM network's superior capability to handle complex, long-term wind speed data. These findings offer valuable insights for enhancing wind energy management in Tetouan City and similar regions, highlighting the LSTM model's potential for improving energy optimization and efficiency.