Wind Power Forecasting Based On Wavelet Analysis And Support Vector Machine (SVM)

After the "eleventh five-year plan" of rapid development, China’s wind power installed capacity showing exponential growth, the influence to the power system is more and more serious. Wind power is affected by the uncontrollable wind speed and direction present a randomness and volatility, large-scale wind power grid to the power system result in a serious challenge to the safe and stable operation. Therefore, wind power forecasting research has important theoretical and practical significance.Foreign wind power forecasting techniques started more earlier, there are a lot of sophisticated forecasting system has been successfully operating. Domestic research is still in its infancy, although some systems are available, but to there is still much room for improvement in prediction accuracy and reliability.This paper summarizes the existing prediction methods, for the combination of wind power prediction accuracy prediction algorithm based on time series forecasting models, two types of combination forecasting method based on wavelet analysis and neural network, at the same time, compile the short-term wind power and forecast system. The main work is as follows:Firstly, for the traditional neural network learning process easily falling into the local minimum, the wavelet neural network is successfully applied to wind power forecasting, and by adding an biased wavelet in the hidden layer, adaptive optimization network structure parameters, effectively reducing the redundancy of the wavelet transform. Through a wind field of wind power prediction research shows that, the biased wavelet neural network can avoid into the local minimum, and effectively improve the prediction accuracy of wind power forecasting.Secondly, putting forward an combined prediction method based on lifting wavelet transform and support vector machine, and applied to study the sequence of wind power. For wind power sequence of nonlinear time-varying characteristic, using lift wavelet transform the original sequence into different frequency detail signals and approximate signals, then by support vector machines (SVM) for these series regression forecast, superposed their output for forecasting wind power. And using cross validation to optimize the SVM penalty parameter and kernel parameter, effectively improve the prediction accuracy. Select a wind farm wind power data as an application case, results show that the lifting wavelet transform can reflect the variation of the wind power data and provide accurate samples for SVM learning predicts, the LWT-SVM forecasting accuracy is better than the existing prediction methods.Thirdly, putting forward the wavelet packet-support vector machine (SVM) method use for wind power forecasting, using SVM to predict the sub-sequences in the wavelet packet decomposition, then superimposed the predictions to get the wind power curve. While the noise reduction technology to handle high-frequency components of the wavelet packet decomposition, to avoid the inclusion of noise in the signal prediction results. The method is applied to wind power data, and the results show that the noise reduction treatment can effectively improve the prediction accuracy.In this paper, using VC6.0developed a software to achieve a short-term forecasting of wind power. Through to the wind power series prediction results show that the system is stable and reliable, user friendly interface, to the realization of high precision wind power forecast.