Application of NNT in wind speed forecasting

In this paper several architectures of neural network multilayer artificial are used to forecast mean daily wind speed at a target station. For this purpose, meteorological variables of reference stations are chosen as exogenous inputs. The mean daily wind speed and direction of 88 measuring stations, from 2005 to 2008 were used. From these stations we will identify zones with similar wind patterns to a target station, using a method based on clustering techniques. Data of target station were acquired from a unit located in Southern Andalusia (Peñaflor, Sevilla), with a soft orography (10 minutes between measurements). The network inputs are basically historical values of the predicted variable as well as a number of support variables. Different feed-forward models and Radial Basis networks have been elected with the aim of carrying out the treatment of the data. These models are compared to the persistence model to select which prediction models are better than persistence model in a short-time prediction.