干旱区生态环境知识资源中心

This work proposes an alternative for generation of synthetic series of daily solar irradiation from spatial interpolation using artificial neural networks (ANN). The estimations were obtained using geographic variables and meteorological variables, that are easily available. The data were measured between 2001 and 2006 in seven Brazilian localities of semi-arid tropical climate, in Northeast Region of the country. The estimations were carried out for the months of November, one of the months of higher incidence of solar irradiation in the region. For each locality a daily temporal series was estimated founded on what was learned from the experimental sequences of the other six localities. The daily solar irradiation calculated series when compared with the experimental series produced RMSE between 11% and 19%, depending on the local. The auto correlations of first order also are well reproduced with maximum deviation of 17%. Finally, the monthly accumulated irradiation values, calculated and experimental, have maximum deviation of 8%, thus, demonstrating their aptitude as input for long term solar system simulation.