Combined prediction of wind power generation in multi-dimension embedding phase space

In order to diminish the effect of reconstructed parameters to prediction of chaotic system, a combined model for wind power prediction based on multi-dimension embedding is proposed. The combined model respectively makes use of linear weighted method and neural network method to achieve combination of several neural network models based on phase space reconstruction, which can synthesize information and fuse prediction deviation in different embedding dimensions, resulting in forecast accuracy improved. Simulation is performed to the real power time series from Fujin wind farm. The results show that the combined prediction model is effective, and the prediction error of neural network combination is less than 7%.