Regression models for hourly diffuse solar radiation

• Eight models for estimating hourly diffuse solar radiation are proposed.
• Data measured in Timişoara, Romania during 2009 and 2010 were used.
• Hourly relative sunshine is introduced as a new predictor.
• The model based on clearness index and relative sunshine performs best (R2 = 0.93).

Accurate information about the available solar radiation is a key factor in designing efficient photovoltaic systems. In general, the meteorological stations measure global solar irradiance on the horizontal surface. In order to estimate the total solar irradiation on surfaces with different spatial orientations, the knowledge of all components on the horizontal plane is required. This paper proposes eight new models for estimating hourly diffuse solar irradiation at the ground level. The estimates of the models are statistically significant at least 97.4% confidence level. The models estimate the diffuse fraction on the basis of the traditional predictor clearness index and various other astronomical and meteorological predictors. The hourly relative sunshine is introduced as a new predictor. The study was conducted on data measured in Timisoara, Romania. The proposed models were compared against twelve different empirical models for the hourly diffuse fraction previously reported by different authors. The results show that the best performing model includes two predictors: clearness index and relative sunshine. The performance of this new model (R2 = 0.930) is comparable to the one reported for models based on more numerous and more sophisticated predictors.