power forecast system Stefano Alessandrini*, Goffredo Decimi*, Dario Ronzio*, Paolo Bonelli* European Wind Energy Conference and Exhibition (EWEC) Polonia, 20-23 Aprile 2010 POSTER SESSION * ERSE SpA Most of wind power forecast systems (WPFS) are based on mesoscale meteorological models supplying wind forecast at their resolution. Usually different post-processing techniques are used to downscale the wind reducing systematic errors. In this paper most common statistic procedures, usually called MOS, are compared with more physical ones, like diagnostic models (CALMET) or CFD models (Windsim). The application of a WPFS, based on the meteorological model RAMS and the different post processing procedures, is described. The wind and power measurements of a wind farm located in the south of Italy, in a complex topography situation, are used as a benchmark. The Wind Power Forecast system developed at ERSE, whose application has been already presented in the previous EWEC conference, is based, as many other forecast tools, on a mesoscale meteorological model like RAMS providing forecast. The wind forecast are supplied with a resolution that usually is not able to take into account for the small topographic or land use features surrounding the wind farm. Our WPFS in the past has adopted a statistical approach (MOS) to reduce systematic errors caused by this lack of details. In this paper are also applied more physical downscaling techniques as diagnostic models (CALMET) or CFD models (Windsim). In these cases, the low-resolution (1-4 km) meteorological fields computed by RAMS is used as a background data input for CALMET or Windsim, that are able to increase the horizontal resolution until less than 100 m. The hourly wind and power measurements, for all 2008 year, of a wind farm located in the south of Italy, in a complex topography situation, are used as a benchmark. The meteorological model RAMS has been first applied using the ECMWF meteorological forecast as boundary conditions. The prediction time horizon is until the day 3 after the analysis date. A simulation, using the ECMWF analysis data as boundary condition, has been also carried on in order to understand the amount of uncertainty due to the forecast errors (coming from the boundary conditions). The computational domain consists of two nested grids with a horizontal resolution of 12 x 12 km 2 and of 4 x 4 km 2 and a vertical resolution of 40 m near the ground. Only the outputs on the smaller grid are then post processed by the different modules described above. In the case of the MOS application, the corrected wind data are used as an input for the wind farm power curve, statistically computed by using historical wind and power measurements. In the case of CALMET or Windsim the wind data computed at the turbine level has been given as an input to the nominal power curve. The different approach are evaluated using the hind cast techniques, therefore 3 months of 2008 are used as a test period while the remaining part of the year as a training period.