Performance Analysis of Hybrid Machine Learning Techniques for Wind Speed Forecasting and Modeling: The Case of the Bodele Triangle

Accurate wind speed prediction is crucial for managing wind power generation systems. However, the stochastic nature of wind complicates the estimation of optimal intervals. This work analyzes the performance of hybrid machine learning techniques for modeling wind speed. Two deep learning models, Large Language Memory Long Short-Term Memory and Large Language Memory Convolutional, are proposed, along with two hybrid models from the literature, Bidirectional LSTM and Convolutional LSTM, for four-season forecasting in the Bodele low-pressure area. Meteorological data come from the NASA Power/Dav site. Data processing includes removal of outliers and imputation of missing values by mean, median, or predictive models, performed with Python. The four hybrid models use the Adam algorithm to optimize predictions. The predicted values calculate wind turbine power, efficiency, and storage energy. Results show that performance indicators vary: MAE from 0.020 to 0.586, RMSE from 0.027 to 0.848, and R² from 0.902 to 0.966. Energy predictions for a 5 MW wind turbine range from 4.91 MWh in winter to 0.89 MWh in summer. The CL-LSTM and LLM-LSTM models give high wind speeds in summer and winter, providing insights for developing efficient models for similar applications, both for researchers and companies.