Study Of Short-term Wind Power Forecasting Method

Along with the exhaustion of primary energy and the growing need of environmental protection, new energy such as wind power is developing rapidly. And the proportion of wind power in the grid is continuously increasing. However, because the wind power is intermittent and uncontrollable, large-scale wind power integrated into power system will bring severe challenges of power system safety operation and power quality. Wind power forecasting technology is one of the key technologies in coping with the problems. As it can guide the dispatching of grid and the production planning of wind farm effectively, it is playing a more and more important role when wind power integrated into power system.This paper studies short-term wind power forecasting technology based on the demands that the practical application call for. The main work is as following:1. Based on the traditional time series ARMA model, an improved prediction model ARMAX-GARCH is proposed. This method considers several influencing factors and corrects prediction results with prediction errors. Making an example and the result shows that the model can improve the prediction accuracy effectively.2. Study a model of BP neural network, a short-term forecasting for wind power based on wavelet decomposition and BP neural network is proposed. By use of wavelet decomposition, the problem of poor data fluctuation regularity is solved. Meanwhile, in order to overcome the shortcomings which BP network is easy to fall into minimum and the selection of hidden layers nodes lack of guidance, the BP algorithm has been improved. Verified with a example and the result shows that the proposed method can improve the prediction accuracy effectively.3. A method of condition probability density distribution is used to build the uncertainty prediction model for wind power prediction. This model can be used for estimating confidence interval under different confidence levels. The effectiveness of the method is validated by practical examples.4. Based on fuzzy clustering analysis theory, a method of continuous multi-step prediction model is proposed. By using this method, the predictions of next four hours can be achieved in the form of every 15 minutes. Also the predictions of next 24 hours with the form of one hour can be achieved. This method solves the problem that it difficult to do continuous prediction without the data of NWP.