Energy production forecast from renewable sources is of great interest for its optimal exploitation. In this paper we present and compare results obtained using different approaches for the prediction of power production of a turbine located in the Island of Borkum (Germany), in the Northern Sea, for times of forecast till 48 hours ahead. We use as prognostic meteorological parameter the wind speed forecast from both deterministic and ensemble ECMWF models. In particular we face here the problem of not having wind speed data observations to verify the quality of wind speed forecast and then had to rest our training and verification procedures entirely on relatively short length time series of power production measurements. This mode of operation does not allow to separate errors on power production forecast due to wind forecast errors from errors due to power turbine production measurements, and the turbine operational power curve must be extracted by fitting power data measurement directly with wind speed forecast. Results presented here, for a period of one year, show that an approach based on probability distribution mapping between wind speed forecast and measured power production are comparable with those obtained using standard approaches and are superior when operating with limited historic data sets.