Modeling forecasting errors of fluctuating renewables and electrical loads

The integration of large amounts of fluctuating renewable energy into the power system, for instance in central western Europe raises a need for accurate handling of forecasting error in power system modeling. Especially, fundamental power system models should incorporate these forecasting errors to supply robust results. This paper presents an approach to generate scenarios of day-ahead wind, photovoltaic and load forecasting errors for a stochastic unit commitment approach. The stochastic character is sampled using an ARMA-GARCH-approach, after deterministic components of underlying time series have been filtered. Subsequently, the clustering of the simulated samples is performed to account for parsimonious model selection. We show, that a scenario set clustered to 20 time-series has the best possible trade-off between accuracy and parsimony.