An analytical comparison of four approaches to modelling the daily variability of solar irradiance using meteorological records

Temporal solar variability significantly affects the integration of solar power systems into the grid. It is thus essential to predict temporal solar variability, particularly given the increasing popularity of solar power generation globally. In this paper, the daily variability of solar irradiance at four sites across Australia is quantified using observed time series of global horizontal irradiance for 2003–2012. It is shown that the daily variability strongly depends on sky clearness with generally low values under a clear or overcast condition and high values under an intermittent cloudiness condition. Various statistical techniques are adopted to model the daily variability using meteorological variables selected from the ERA-Interim reanalysis as predictors. The nonlinear regression technique (i.e. random forest) is demonstrated to perform the best while the performance of the simple analog method is only slightly worse. Among the four sites, Alice Springs has the lowest daily variability index on average and Rockhampton has the highest daily variability index on average. The modelling results of the four sites produced by random forest have a correlation coefficient of above 0.7 and a median relative error around 40%. While the approach of statistical downscaling from a large spatial domain has been applied for other problems, it is shown in this study that it generally suffices to use only the predictors at a single near point for the problem of solar variability. The relative importance of the involved meteorological variables and the effects of clearness on the modelling of the daily variability are also explored.

[1]  R Core Team,et al.  R: A language and environment for statistical computing. , 2014 .

[2]  Iain MacGill,et al.  Characterizing future large, rapid changes in aggregated wind power using Numerical Weather Prediction spatial fields , 2009 .

[3]  P. Ineichen,et al.  A new airmass independent formulation for the Linke turbidity coefficient , 2002 .

[4]  Leo Breiman,et al.  Random Forests , 2001, Machine Learning.

[5]  Yih-huei Wan,et al.  Dark Shadows , 2011, IEEE Power and Energy Magazine.

[6]  J. Gutiérrez,et al.  Reassessing Statistical Downscaling Techniques for Their Robust Application under Climate Change Conditions , 2013 .

[7]  J. Sáenz,et al.  Downscaling of surface moisture flux and precipitation in the Ebro Valley (Spain) using analogues and analogues followed by random forests and multiple linear regression , 2011 .

[8]  David S. Siroky.  Navigating Random Forests and related advances in algorithmic modeling , 2009 .

[9]  Robin Girard,et al.  Forecasting ramps of wind power production with numerical weather prediction ensembles , 2013 .

[10]  Pablo M. Granitto,et al.  Prediction of minimum temperatures in an alpine region by linear and non-linear post-processing of meteorological models , 2007 .

[11]  Nick Ellis,et al.  Predicting wind power variability events using different statistical methods driven by regional atmospheric model output , 2015 .

[12]  Andy Liaw,et al.  Classification and Regression by randomForest , 2007 .

[13]  V. Quintana,et al.  Hierarchical State Estimation Using a Fast Rectangular-Coordinate Method , 1989 .

[14]  Leon Freris,et al.  Renewable energy in power systems , 2008 .

[15]  R. Davy,et al.  Statistical Downscaling of Wind Variability from Meteorological Fields , 2010 .

[16]  H. Storch,et al.  The Analog Method as a Simple Statistical Downscaling Technique: Comparison with More Complicated Methods , 1999 .

[17]  D. R. Cutler,et al.  Utah State University From the SelectedWorks of , 2017 .

[18]  Joshua S. Stein,et al.  The Variability Index: A New and Novel Metric for Quantifying Irradiance and PV Output Variability. , 2012 .