Estimation of daily, weekly and monthly global solar radiation using ANNs and a long data set: a case study of Fortaleza, in Brazilian Northeast region

Abstract

A 14-year-long data set containing daily values of meteorological variables was used to train three artificial neural networks (ANNs) for daily, weekly averaged and monthly averaged global solar radiation prediction for Fortaleza, in the Brazilian Northeast region. Local climate is semiarid coastal. Day of the year, maximum temperature, minimum temperature, irradiance, precipitation, cloudiness, extraterrestrial radiation, relative humidity, evaporation and wind speed were adopted as predictors. The ANNs were developed by an in-house code and trained with the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm. Besides the lack of explicit predictors able to model El Niño and La Niña phenomena, which have strong influence on local weather, the accuracy of the predictions was considered excellent according to its values of normalized root-mean-square error (nRMSE) and good relative to mean absolute percentage error (MAPE) values. Both error metrics presented the smallest values for the monthly case study.