Ridge regression ensemble of machine learning models applied to solar and wind forecasting in Brazil and Spain

In recent years, with the rapid development of wind and solar power generation, some problems arise gradually and are often inherent to intermittency. Currently, one of the essential methods to solve these problems is the application of forecasting methodologies. Our article proposes an ensemble learning method based on crest regression (penalized L2) which integrates consolidated wind and solar forecasting methodologies applied to two locations with different latitudes and climatic profiles. From the simulations carried out, the methodology is efficient to improve the predictions performance of isolated methods and applicable to different locations around the world. For solar data from Brazil and Spain, the ensemble model achieves MAPE values of 14.191% and 11.261%, respectively; for the same data, the best model applied individually (CFBP) shows a higher MAPE of 24.207% and 12.465%, respectively. For wind data from Brazil and Spain, the ensemble model has a MAPE of 3.927% and 5.491%, respectively. The best model applied individually to Brazilian wind data is CFBP, with MAPE of 9.345%. For the Spanish data, the best individual model is MLP, with MAPE of 7.186%. The ensemble modeling reduces the forecast errors and can be useful in optimizing the planning for the use of intermittent solar and wind resources in the electrical matrices.

Automated summary

With the rapid development of wind and solar power generation, the issue of intermittency becomes more prominent, and ensemble learning methods can improve predictions and be applied to wind and solar data from different locations. The ensemble model achieves better performance in predicting solar data from Brazil and Spain, as well as wind data from Brazil and Spain, compared to individual models.