A Robust Algorithm for Wind Power Forecasting Based on Projection Pursuit and Back Propagation Neural Network

Abstract

This paper has proposed a wind farm generation output forecasting model based on projection pursuit (PP) and back propagation neural network (BPNN), in order to eliminate the influence of the bad points and mutations on and enhance robustness of the forecasting model. A median absolute deviation is used as projection index function, effectively avoiding the influence of the outlier. Firstly, Extract the principal components of each factor by PP. Then, input the principal components to the BPNN for training the network. Finally, forecast the wind farm generation output via the trained network. The simulation shows that the proposed approach is of higher accuracy.