Vidal Bezerra, F.D.; Pinto Marinho, F.; Costa Rocha, P.A.; Oliveira Santos, V.; Van Griensven Thé, J.; Gharabaghi, B. Machine Learning Dynamic Ensemble Methods for Solar Irradiance and Wind Speed Predictions. Atmosphere2023, 14, 1635.
Vidal Bezerra, F.D.; Pinto Marinho, F.; Costa Rocha, P.A.; Oliveira Santos, V.; Van Griensven Thé, J.; Gharabaghi, B. Machine Learning Dynamic Ensemble Methods for Solar Irradiance and Wind Speed Predictions. Atmosphere 2023, 14, 1635.
Vidal Bezerra, F.D.; Pinto Marinho, F.; Costa Rocha, P.A.; Oliveira Santos, V.; Van Griensven Thé, J.; Gharabaghi, B. Machine Learning Dynamic Ensemble Methods for Solar Irradiance and Wind Speed Predictions. Atmosphere2023, 14, 1635.
Vidal Bezerra, F.D.; Pinto Marinho, F.; Costa Rocha, P.A.; Oliveira Santos, V.; Van Griensven Thé, J.; Gharabaghi, B. Machine Learning Dynamic Ensemble Methods for Solar Irradiance and Wind Speed Predictions. Atmosphere 2023, 14, 1635.
Abstract
In this paper, solar irradiance and wind speed forecasts were performed considering time horizons ranging from 10 min to 60 min, under a 10 min time-step. Global horizontal irradiance (GHI) and wind speed were computed using four forecasting models (Random Forest, k-Nearest Neighbours, Support Vector Regression, and Elastic Net) to compare their performance against two alternative dynamic ensemble methods (windowing and arbitrating). Forecasting models and dynamic forecasting ensembles were implemented in Python for performance evaluation. The performance comparison between the prediction models and the dynamic ensemble methods was carried out by evaluating the RMSE, MAE, R² and MAPE, to evaluate whether the dynamic ensemble forecasting method obtained greater. According to the results obtained windowing dynamic ensemble method was the most efficient among the tested. For the wind speed data, by varying its parameter λ (from 1 to 100), a variable performance profile was obtained, where from λ =1 to λ = 74, windowing proved to be the most efficient, reaching maximum efficiency for λ = 19. Windowing was the best method for the GHI analysis, reaching its best performance for λ = 1. The efficiency gain using windowing was 0.56% when using the wind speed model and 1.96% for GHI.
Keywords
wind energy; solar energy; renewable energy; machine learning; forecasting ensembles
Subject
Engineering, Mechanical Engineering
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.