Neural Network Ensemble Method Study for Wind Power Prediction

Abstract

Wind power prediction is of great importance for the safety, stabilization and economic efficiency of electric power grids, especially when the wind power penetration level of the gird is high. ANN (Artificial Neural Network) is an appropriate method for wind power prediction. But the generalization of common ANN is poor and the prediction precision is not stable. Neural network ensemble can enhance the generalization ability of neural network remarkably. Neural network ensemble has two key problems: one is how to build individual neural network, and the other is how to synthesize the outputs of the individual networks. According to wind power prediction characteristic, a new method was used to build individual neural network, the different individual neural network can be given specific physical meaning. ANN was used to synthesize the outputs of the individual networks. The calculation example showed that the difference scale between each individual neural network was higher and the prediction precision was greatly improved compared to that of the single neural network.