Neural networks for wind power generation forecasting: A case study

This paper uses data collected in a southern Italy wind farm to develop a neural network based prediction of the power produced by each turbine. First, some characteristics of wind turbine power generation are investigated. Then a careful data preprocessing is proposed to detect and remove outliers and to deal with damping, i.e. the effect of smoothing of wind speed caused by presence of other turbines. Besides, two different training algorithms for the most popular model, the multilayer perceptron, are analyzed, i.e. backpropagation and extreme learning machine (elm). The latter, when utilized together with a proposed data preprocessing technique, demonstrates to achieve better and more stable performance, despite its greater sensibility to overfitting.