Short-term Wind Power Prediction Method Based on Dynamic Wind Power Weather Division of Time Sequence Data

It is an effective measure for improving meticulousness of short-term wind power forecasting to establish wind power forecasting models based on weather patterns. The traditional method of wind power prediction based on weather pattern clustering often utilizes feature extraction to adapt to the clustering method. But it is difficult to reflect the change of complex weather completely, thus weakening reliability of the weather division. In order to retain the dynamic nature of all wind power weather types, a wind power prediction method based on dynamic weather division is proposed in this paper. The t-distributed stochastic neighbor embedding (t-SNE) for nonlinear dimensionality reduction is employed to reduce the number of meteorological variables. And the hierarchical clustering method of panel data based on combination of absolute distance, increment speed distance and fluctuation distance is used to segment historical weather. According to different wind power weather types, wind power prediction models based on support vector regression (SVR) are established. With the data of a wind farm group in Gansu as example, it is proved that accuracy of the proposed method is significantly improved compared with the model based on feature extraction, and effectiveness of the improved wind power prediction method is verified. © 2019, Power System Technology Press. All right reserved.