Wind Power Forecasting Using Neural Network Ensembles With Feature Selection

In this paper, a novel ensemble method consisting of neural networks, wavelet transform, feature selection, and partial least-squares regression (PLSR) is proposed for the generation forecasting of a wind farm. Based on the conditional mutual information, a feature selection technique is developed to choose a compact set of input features for the forecasting model. In order to overcome the nonstationarity of wind power series and improve the forecasting accuracy, a new wavelet-based ensemble scheme is integrated into the model. The individual forecasters are featured with different mixtures of the mother wavelet and the number of decomposition levels. The individual outputs are combined to form the ensemble forecast output using the PLSR method. To confirm the effectiveness, the proposed method is examined on real-world datasets and compared with other forecasting methods.