Decarbonizing the energy supply requires extensive use of renewable generation. Their intermittent nature requires to obtain accurate forecasts of future generation, at short, mid and long term. Wind Energy generation prediction is based on the ability to forecast wind intensity. This problem has been approached using two families of methods one based on weather forecasting input (Numerical Weather Model Prediction) and the other based on past observations (time series forecasting). This work deals with the application of Deep Learning to wind time series. Wind Time series are non-linear and non-stationary, making their forecasting very challenging. Deep neural networks have shown their success recently for problems involving sequences with non-linear behavior. In this work, we perform experiments comparing the capability of different neural network architectures for multi-step forecasting in a 12 h ahead prediction. For the Time Series input we used the US National Renewable Energy Laboratory’s WIND Dataset [3], (the largest available wind and energy dataset with over 120,000 physical wind sites), this dataset is evenly spread across all the North America geography which has allowed us to obtain conclusions on the relationship between physical site complexity and forecast accuracy. In the preliminary results of this work it can be seen a relationship between the error (measured as R2R2 ) and the complexity of the terrain, and a better accuracy score by some Recurrent Neural Network Architectures.

Peer Reviewed