Wind Power Ramp Events Prediction Considering Time-frequency Characteristics

Abstract

Accurate prediction of wind power ramp events plays an important role in the operation and dispatch of power systems with high wind power penetration. Aiming at the problem that the failure to effectively decompose and refine the high-frequency components of wind power affects wind power prediction and thus reduces the ramp events identification accuracy, a novel ramp events prediction model using wavelet packet transform (WPT) to describe the high-frequency characteristics of wind power is proposed in this paper. Firstly, WPT is adopted to decompose the historical wind power sequence. Then, extreme learning machine (ELM) is used to predict each power component. Finally, by analysing the characteristics of different ramp events identification definitions, a combination definition considering wind power time-frequency characteristics is proposed and a ramp events prediction model of WPT-ELM is established. Extensive tests using actual power data of a wind farm in northern China demonstrate that the proposed prediction model has higher prediction accuracy and can effectively identify wind power ramp events.