Influência dos fenômenos climáticos do el niño e da la niña na previsão da média diária de irradiação global na cidade de Fortaleza

Abstract

In this work, predictions of the daily average of global solar irradiation were obtained by the application of machine learning algorithms in two data sets formed by exogenous variables (insolation, air temper-ature, precipitation), endogenous variables (time series of the daily average of irradiation global solar) and temporal variables (year, month and day of measurement). The difference between the data sets is related to the fact that in one the intensities of El Niño and La Niña weather phenomena are consid-ered as predictors for the learning models used, while in the other it is not considered. In this way, it was possible to evaluate whether the addition of the predictor related to El Niño/La Niña contributes to a better prediction accuracy by the applied models: Minimum Learning Machine, Support Vector Re-gression, Random Forests, K-Nearest Neighbors and a regression tree using Bootstrap. The error met-rics Mean Absolute Error, Mean Bias Error, Root Mean Square Error, Root Mean Square Error and Forecast Ability were used to analyze the performance of the algorithms. The arithmetic mean of the Root Mean Square Error and the Prediction Ability for the case in which El Niño/La Niña were consid-ered as attributes were 40.78 W/m² and 7.87%, respectively. For the case in whichsuch predictors are not considered, the values obtained were 40.86 W/m² and 7.69%, indicating that the use of these pre-dictors increases the prediction accuracy of the algorithms in question.