An application of a feed-forward neural network model for wind speed predictions

Since wind energy covers an important part of the increasing electricity demands, accurate wind speed and direction forecasts are essential for the optimal operation of both the wind farms and the grid. Here we present the set-up and optimisation of a selected feed-forward neural network model in order to forecast, 24-hour ahead, the wind speed at specific points of three wind farms, using past data measured at the same locations. The model combines the k-fold cross-validation method for selecting the appropriate input data consisted either of hourly or 10-min average input values, while the impact of the length of the input time series is also investigated. Our method is tested against persistence method forecasts using the RMSE and MAPE statistics.