Multi-temporal forecasting of wind energy production using artificial intelligence models

Bouabdallaoui, Doha
Haidi, Touria
Derri, Mounir
Hbiak, Ishak
El Jaadi, Mariam

International Journal of Renewable Energy Development
Vol 14, No 3 (2025): May 2025

In response to changing energy demands, electricity suppliers are increasingly turning to sustainable energy sources, with wind power emerging as a promising solution. This study aims to predict wind energy production over four time horizons: hourly, daily, weekly, and monthly, for a 12,300 kW wind farm located in Northamptonshire, UK. We employed three artificial intelligence (AI) techniques: an ensemble of bagged decision trees, artificial neural networks (ANNs), and support vector machines (SVMs). The paper provides a comparative evaluation of AI-based forecasting techniques for wind energy prediction, highlighting differences in model performance across time horizons while emphasizing the strengths and limitations of each method in addressing the temporal variability of wind energy production. The models were tested over various times using important performance measures, such as the correlation coefficient (R), the coefficient of determination (R²), mean absolute error (MAE), root mean squared error (RMSE), and bias. The results indicate that support vector machines achieve the highest accuracy for medium-term forecasts, with a coefficient of determination of 0.9722 and a mean absolute error of 44.91 kW. Artificial neural networks perform best in short-term forecasting, particularly at the daily level, with a coefficient of determination of 0.948 and a mean absolute error of 36.04 kW. In contrast, long-term predictions exhibit greater variability across models, with the coefficient of determination decreasing to 0.778, reflecting the increased complexity of extended forecasting. The ensemble of bagged decision trees demonstrates strong predictive capability but with slightly higher error margins compared to support vector machines. The obtained results could serve as a reference for selecting the most suitable models based on forecasting objectives and time constraints. Future improvements in forecasting accuracy could happen by combining these models with optimization algorithms, especially for medium- and long-term predictions, where making accurate forecasts is still very difficult.