“The aim of the hybrid method we have developed is to predict the wind speed in each of the wind turbines on a wind farm”, explained Sancho Salcedo, an engineer and co-author of the study, published in the on-line journal Renewable Energy.
The new model incorporates information covering the entire planet with a resolution of approximately 100 kilometres provided by the Global Forecasting System of the US National Center for Environmental Prediction. Researchers are also able to make more precise predictions by integrating the so-called ‘fifth generation mesoscale model’ (MM5), from the US National Center of Atmospheric Research, designed to enhance resolution to 15x15 kilometres.
However, this information is still not detailed enough to predict the wind speed of a particular wind turbine and therefore, Salcedo explains, the researchers at the UAH and the UCM decided to apply artificial neural networks. These networks are automatic information learning and processing systems that simulate the workings of animal nervous systems. In this case, they use the temperature, atmospheric pressure and wind speed data provided by forecasting models, as well as the data gathered by the turbines themselves.
Once the system has been “trained”, these data enable predictions regarding wind speed to be made between one and 48 hours in advance. Wind farms are obliged by law to supply these predictions to the Spanish electricity grid operator, Red Eléctrica Española.
Salcedo says the technique has been tried and tested and has already been used very successfully at the wind farm in Fuentasanta, in
Possible cost savings are enormous
The team of scientists is working to hone the method and recently proposed the use of a suite of global forecasting models instead of just one, according to an article published this year in Neurocomputing. As a result, several sets of observations are obtained, which are then applied to banks of neural networks to achieve a more accurate prediction of wind speeds around wind turbines. Results indicate that predictions could be improved by up to 2% using this more exhaustive approach. “Although this may seem like a small improvement, it is really substantial, as we are talking about an improvement in predicting energy production that could be worth millions of euros”, Salcedo concluded.