Improving solar radiation forecasts using advanced statistical post-processing methods

Abstract

The increased usage of solar energy places additional importance on forecasts of solar radiation. Solar panel power production is primarily driven by the amount of solar radiation and it is therefore important to have accurate forecasts of solar radiation. Accurate forecasts that also give information on the forecast uncertainties can help users of solar energy make better solar radiation based decisions related to stability of the electrical grid. To achieve accurate forecasts of global radiation together with information about the uncertainty, we apply statistical post-processing techniques to the deterministic forecast from the Numerical Weather Prediction model HARMONIE, which runs at KNMI. We use regression based methods that determine relationships between observations of global radiation (made within the Netherlands network of automatic weather stations) and forecasts of various meteorological variables from HARMONIE. Those relationships are used to produce probabilistic forecasts of global radiation. We compare parametric methods that make assumptions on the distribution of the global radiation and non-parametric methods without any assumptions on the distribution. We find that both types of methods are able to generate probabilistic forecasts that improve the raw global radiation forecast from HARMONIE according to the root mean squared error (on the median) and the potential economic value. We also compare the different regression methods using various scoring metrics like the continuous ranked probability skill score, the Brier skill score and reliability diagrams. We find that quantile regression and generalized random forests generally perform best and the artificial neural network worst. Additionally we show that the non-parametric approaches use information from all predictors, whereas the parametric approaches look only at a limited number of predictors.