Analysis of solar direct irradiance models under clear-skies: Evaluation of the improvements for locally adapted models

Abstract

Direct solar irradiance has to be determined for the design of many energy applications such as photovoltaic systems and concentration systems, and the generation of solar potential maps for energy use. Knowledge of the accurate values of radiation components in a local area will allow optimal sizing of solar energy conversion systems. The estimated values of direct solar irradiance from models are still necessary at those sites where no measurements are available. In this work, different models used for the estimation of the direct component of solar irradiance are analyzed. First, an evaluation of the performance of eight existing original models was performed from which three were selected. Second, the selected models were calibrated to adapt them to our studied geographical area, which is the important aspect of this work, and an assessment of performance improvements for locally adapted models is reported. Experimental data consisted of hourly horizontal global, direct, and diffuse solar irradiance values, provided by the National Meteorological Agency in Spain (AEMET) for Madrid. Long-term data series, corresponding to a total period of time of 32 years (1980–2011), have been used in this study. Only clear sky models were treated at present. The three selected models were adapted to the specific location of Madrid, and root mean square error (RMSE) and mean-biased error were determined. By comparing the performance in the direct horizontal irradiance estimation from existing original and the corresponding locally adapted models, it is found that the values of RMSE decreased from 9.9% to 5.7% for the Louche model, from 7.8% to 7.4% for the Robledo-Soler model, and finally from 8.8% to 6.7% for the European Solar Radiation Atlas model. Thus, significant improvements can be reached when parametric models are locally adapted. In our case, it is up to approximately 4% for the Louche model. It is expected that calibrated algorithms presented in this work will be applicable to regions of similar climatic characteristics.