The Research On Ultra-Short-Term Wind Power Forecasting Based On Trend Point Model

In recent years, along with the manufacturing industry and equipments industry booming exponentially and the traditional fossil fuels exhausting dramatically, the scale of construction of wind farms has become larger than any other time in history, as well as its speed. Due to the uncertainty and intermittence of the nature of wind power, however, wind energy has brought great charge and challenge to the living power system. It is necessary to dispatch wind power reasonably in order to keep the whole power system in operation with safety, stabilization and economy. Furtherly, the dispatch of wind power lies in the prerequisite of prediction for the wind power. And a relatively accurate wind power forecasting plays a vital role not only in the control of wind driven generators and its preventive maintenance, but also in keeping the balance between active power and reactive power, and electric power bidding.Based on the data mining, a new method called trend point model for prediction is proposed in this article via an improvement on trend point state model. Aiming at extracting the useful message at utmost from a single historical data, trend point model divides the sequence of the past wind power into a series of sub-sequences, taking the Pearson’sCorrelation Coefficient to calculate the relativity between the latest sub-sequence close to the forecasting point and the rest sub-sequences respectively so that the trend points could emerge. Thanks to the ordinary least square, the trend points can be turned to the forecasting values. Last but not least, the final predicting value follows by taking the average of all the forecasting values before while the exceptional data get deleted through Pauta criterion.Besides being an independent algorithm for forecasting wind power, trend point model applies to the combination prediction models very well. Unlike the traditional combined forecast on the basis of the weights, trend point model could choose the historical data selectively when it comes to the weights in need instead of the blindness in the usual way.