Daily UV radiation modeling with the usage of statistical correlations and artificial neural networks

LEAL SS, TÍBA C Y PIACENTINI RD

RENEWABLE ENERGY

PERGAMON-ELSEVIER SCIENCE LTD

Año: 2010 vol. 36 p. 3337 - 3344

0960-1481

The information regarding solar UV radiation (UVA þ UVB) in Brazil and around the world is scarce with low spatial and temporal coverage. This information scarcity, due to the small number of measuring stations, has directed some researchers towards the creation of computational parametric models or the generation of statistical models for the estimation of the UV radiation from the measurement of the global radiation. Information about UV irradiation is expanded for other places where there is only global radiation data. Thus, two stations were set up, in 2008, one in the city of Pesqueira and the other in Araripina, both in the state of Pernambuco, for simultaneous measurements of daily global solar and UV radiation. Another station is being set up in Recife-PE, completing a group of stations that are located between latitudes 8 and 10º and longitudes 34 - 38º W, representing the typical climate of the region. The daily values of G global and UV ultraviolet radiation (A þ B) striking the horizontal plane in Pesqueira and Araripina during the time period (2008-2010) were measured, analyzed and compared. The collected data enabled the generation of three different statistical models for estimating the daily UV solar radiation from the daily global radiation: a) linear correlation between global and UV radiation (model 1), b) polynomial correction of the average fraction of UV irradiation, as a function of the transmittance index of global solar irradiation (model 2) and c) the UV atmospheric transmittance index is obtained by multiple regression of the air mass hmriand hKTi (model 3). Besides, they were modeled by two artificial neural networks: a) estimative of the (FUV), considering the same physical variables of model 2 (model ANN1) and b) estimative of (KTUV) from the same physical variables of model 3 (model ANN2). The statistical models and the artificial neural networks displayed a good statistical performance with RMSE% inferior to 5% and MBE between 0.4% - 2%. All the models can be used for estimating the UV radiation in places where there is only global irradiation data.