Submitted Abstract
Forecasting the photovoltaic (PV) power is of increasing importance for grid operators as well as energy providers, considering the PV growth rates within the EU and in Luxemburg (factor 4, between 2011 and 2016 in Lux.; (ILR 2018)). Several methods exist to perform forecasting on regional and single site scale. Each method has its benefits and might outperform the other, depending on the forecast horizon and other factors, such as weather conditions or regional scale (Antonanzas, et al. 2016). In the common research project “Combi-Cast”, LIST and Electris will combine different forecast schemes to make use of the best performance for the respective forecast horizon and (weather) conditions. The combined methods are very complementary and will be a. based on numerical weather predictions (NWP), b. based on cloud motion vectors (CMV) and c. using persistence forecasting – which have proven benefits at different forecast horizons (Sobri, Koohi-Kamali und Rahim 2018). The model will use machine-learning algorithms to weight the respective model in an optimal manner and would derive probabilistic forecasts to meet the requirements of the modern energy market (Alessandrini, et al. 2015). A further innovative aspect is the estimation of CMVs based on smart meter data from PV systems, to be contributed by Electris, at high temporal resolution (up to 10s). The estimation of cloud motions is already being used to forecast solar irradiation, but mainly based on satellite data or using specific additional equipment (e.g. sky-imager cameras), whereas Combi-Cast is based on data and methods accessible and applicable by the stakeholders themselves.