Forecasting hourly global solar irradiation using simple non-seasonal models

The solar radiation on ground level has cyclic variations due to the rotation and revolution movements of the Earth. This deterministic variation is modulated by weather, which superimposes a random component. The subject of this paper is the forecasting of hourly global solar irradiation using simple statistical models, without exogenous inputs. The study is conducted based on data recorded at Timisoara, Romania, during 2009 and 2010. Since the time series of hourly solar irradiation exhibits a natural seasonality at frequency of 1/24 h, a seasonal decomposition of data is required. First, results referring to the dynamic properties of the seasonal indices are discussed. Second, a comparison of the accuracy of five non-seasonal models applied to seasonally adjusted data was performed. Overall results demonstrate that the random walk model performs best. Third, an adaptive procedure to fit the model parameters (in which the model parameters are refitted for every day) was tested against the standard "frozen parameters" approach (in which the model parameters are obtained by a onetime fitting of the equations on historical data and then used in every forecast in the future). The results show that the proposed adaptive procedure leads to better performance than the frozen parameter approach. (C) 2013 AIP Publishing LLC.