E4Cast approach in energy forecasting couples a numerical meteorological model with a statystical model based on Artificial Neural Networks. Such technology represents the state of the art in wind energy forecasts and grants excellent results thanks to the benefit of the two techniques.
The NWP model is the WRF model (Weather Research and Forecasting) developed at NCAR. The WRF model is a next-generataion mesoscal numerica prediction system dedigned and developed to serve both research and operational forecasting. This model has been implemented and tuned by out team for the European region at a spatial resolution of 12km and for the Central and South Italy at a spatial resolution of 4km.
Our implementation in based on initial and boundary condition from GFS general circulation model (Global Forecasting Model) . The two nested domains are evaluated in parallel with active feedback between the processes. Microphisics processes are parametrized using the WSM3 scheme (Hong, Dudhia and Chen, 2004), shortwave and longwave radiations are parametrized through respectively the Dudhia and the Rrtm schemes. The planetary boundary layer scheme follows the Yonsei University Scheme (Hong, Dudhia, 2003) and finally the convective parametrization scheme used for convection processes within the grid element is the Kain Fritsch Scheme (Kain, Fritsch, 1993).
The wind forecast from WRF model in then used as input for our Statistical model, based on an Artificial Neural Network. The NN is an adaptive system and must be trained comparing forecasts and real measurements from the wind farm.
This step has the fundamental role of finding the best match between forecasts and real energy production, letting us to tune the forecast sistem for a specific site.
The final wind & energy forecast is then postprocessed in several delivery formats for the final user (txt, csv, gif) and distributed via e-mail, ftp or shared though our web portal E4Cast.