Weighted error functions in artificial neural networks for improved wind energy potential estimation

This paper presents the application of the artificial neural network (ANN) to predict long-term wind speeds of a particular site, and to estimate the annual energy production of wind turbines using the predicted wind speeds. A major finding in this study is that an ANN trained with a conventional error measure may significantly underestimate the annual energy production. An accurate prediction of the mean wind speed does not guarantee an accurate prediction of the energy production when the variance of the wind speed is underestimated. To improve the accuracy in estimating the energy production, we proposed two ANNs that are based on weighted error functions. They use the frequency of the wind speed and the power performance curve to develop the weighted form of the error function. For the site and the turbine studied in this paper, the proposed ANNs showed 8-12% improvement in predicting the annual energy production compared to the conventional ANN.