Application of "Analog" method for wind power ensemble forecasting

Wind energy, being variable and uncertain, raises several issues for the security of the power grids operations. In particular, Transmission System Operators (TSOs) need as accurate as possible forecasts. Concerning to this, it is also important to provide information about the accuracy of a deterministic forecast. A new method called "Analog" Ensemble (AnEn, originally suggested by NCAR) is applied for the first time to wind power forecasting and is able to provide ensemble wind power prediction. The AnEn method can be briefly resumed as follows. For each forecast lead time the ensemble set of forecast of a certain variable is constituted by a set of its measurements of the past. These measurements are those concurrent to the past forecasts at the same lead time, chosen across the past runs more similar to the current forecast. The variable to be predicted is not necessary an output of the meteorological model but must be dependent from at least one of them. In the current application wind power production must be predicted. The possibility of producing more accurate deterministic forecasts together with an estimate of their accuracy among other methods is analyzed. To support this, an application to a real case wind farm is explored showing evaluation indexes and graphs. A comparison of the performances obtained by another deterministic forecast model based on Neural Network (NN) is also carried out. Furthermore, analyzing reliability diagram, continuous ranked probability scores, RMSE (computed using the ensemble mean against the power measurements) and other indexes, it's clear that, quite at every lead time in this case, the performances of AnEn ensemble are better than those of other systems (based on the use on meteorological ensemble prediction model like the ECMWF one).