Energy production forecasting for a wind farm composed of turbines with different features

This paper presents models to predict the energy produced by the wind farm. Support Vector Machine and Artificial Neural Network models were used to predict the wind energy production of a wind farm in Galicia, Spain, with 24 turbines of 9 different models. The data used, to develop the models, presents some measurement errors caused by possible malfunctioning of the sensors, maintenance and other technical problems. The occurrence of these errors was revealed by lack of data in the sample or by data whose value was outside the expected range. So, the first stage was pre-processing of data. It consists in removing the data for which the wind speed is higher than the rated speed and remove the data where it is further away from the energy curve of the wind farm. In the following stage, a study was held about the distribution data where it was verified its Non-Gaussian distribution (i.e. non-parametric data). In the last stage a non-linear regression model was develop with supervised learning to predict the energy produced by the wind farm 1 hour ahead (Support Vector Machine with a Gaussian kernel and Artificial Neural Network).