Wind Power Prediction Method Based On Hybrid Neural Network

With the rapid development of economy, the continuous increase of the degree ofpollution and the proposal and implementation of economic development strategy on energysaving and emission reduction, wind power has become an important component of renewableenergy. Moreover, wind power grid-connected technology and its related theory has become amajor research field of science and technology that attracts much attention and enough efforts.Wind is the power of wind turbine. The characteristics of wind itself and the electricitygeneration principle of wind power determine that the power output are fluctuant andintermittent. Thurs, wind power grid-connected technology will have enormous impact onpower system. Wind power forecasting plays a vital role on wind power grid integration andthe safe operation of power grid. In order to improve the accuracy of short-term wind powerforecasting, the hybrid neural network forecasting model for the time series characteristics isbuilt, which is based on the deep and systematical research of power output time series ofwind farm.The hybrid neural network wind power forecasting model based on improvingHilbert-Huang transform is built after deep study on the time-frequency characteristic ofpower output time series of the wind farm. The wind farm power output time series areresolved by ensemble empirical mode decomposition and the Hilbert spectrum is gained byintroducing the Hilbert-Huang transform to describe each intrinsic mode functioninstantaneous frequency. Then input elements of the RBF Neural network are determined. Theeffectiveness and feasibility of the forecasting model have been verified by the measured dataof a northeast wind farm.The wavelet neural network wind power forecasting model based on the wavelettransform is built after deep study on the time-frequency localization method of the wind farmpower output time series. The wavelet transform is used as a pre-processing method of theoutput power time series in this model. Moreover, the topology of the wavelet neural networkand excitation function of the hidden layer are determined based on it. The model has goodforecasting performance, especially for the long time forecasting in the future, which has beenverified by the measured data. The chaotic neural network wind power forecasting model based on state spacereconstruction is built after deep study on the nonlinear dynamic characteristics of wind poweroutput time series. The chaos characteristic of the wind farm output power time series isverified by computing the Lyapunov exponent, and the phase space is reconstructed with thedetermination of the optimal delay time and embedding dimension. The topology of thechaotic neural network is determined based on the model. Finally the predication performanceis gained by training the model with the processed measured data. Taking the wind farms inNortheast China as an example, the results indicate that the wind power forecasting model hasa strong forecasting performance, especially for the power output peaks and troughs, but theforecasting for the time in the future is of poor performance.The hybrid neural network forecasting model based on the above three kinds offorecasting models after three wind power forecasting models are verified through examples,and BP neural networks are used as a tool for information fusion for the advantages anddisadvantages of each model. Verified by the measured data, the hybrid neural network playsthe advantages of the above three kinds of forecasting models, and can effectively improve thedefect of the original model. The hybrid neural network technology can improve the accuracyof the wind power forecasting.