Machine Learning Methods for Solar Irradiation Forecasting: A Comparison in a Mediterranean Site
Résumé
In this survey, several statistical and machine learning tools are analyzed and compared in view to forecast the solar irradiation in Ajaccio (Corsica, France, 41°55 N, 8°44 E, 4m asl). The forecasting horizon is from 1 to 6 hours with an hourly time granularity. Eleven forecasting models are compared: persistence, scaled persistence, ARMA, MLP, regression trees, boosted regression trees, bagged regression trees, pruned regression trees, random forest, Gaussian processes and support vector regression. The models are compared in term of error metrics: nRMSE (normalized root mean squared error), MAE (mean absolute error) and skill score related to the smart persistence.
Origine : Fichiers produits par l'(les) auteur(s)