Abstract
In the present paper, the comparison of three of the main estimation methods of solar radiation was performed: empirical models, Artificial Neural Network (ANN) and Support Vector Machine (SVM). Four classical empirical models were calibrated and validated in order to estimate hourly solar UV data in Botucatu, São Paulo State, Brazil. Taken the empirical models as reference of accuracy and set for input variables, the performance of ANN and SVM were assessed. Through the statistical parameters Mean Bias Error (MBE) and Mean Absolute Error (MAE) was confirmed the superiority of the SVM over the ANN and empirical models. The SVM is capable to generate better results than ANN using a less number of input variables. Among all estimation methods, SVM using the set of input variables {UV0, KT} is considered the best alternative due to the smaller number of input variables and relative precision.
Keywords:
solar radiation; machine learning; input variables selection
