A novel two-stage data-driven model for ultra-short-term wind speed prediction

Abstract

Accurate prediction of wind speed and its output power is playing an essential role in the planning and scheduling of wind power grid. This study presents a novel two-stage data-driven model for ultra-short-term wind speed prediction based on the smoothing spline preprocessing (SSP) method and error optimization theory (EOT). Firstly, high-resolution wind data observed from 39 wind turbines are collected and transformed according to the proposed SSP method to eliminate the non-Gaussian and non-stationary features. Then, several individual models are introduced to perform multi-step ahead wind speed prediction for the transformed wind data, and the prediction of transformed data should be recovered to wind speed. Finally, these single models are combined based on the proposed EOT theory for multi-step ahead wind speed prediction, and their accuracy and uncertainty are analyzed and compared with other existing models in depth. The results show that the proposed SSP method can reasonably identify non-Gaussian and non-stationary features of the original wind series, and the transformed data are more favorable for prediction. Furthermore, the suggested two-stage data-driven model can reduce prediction errors by 3%–20% compared with other models mentioned in this study, indicating that it is more effective and stable in terms of providing reasonable ultra-short-term wind speed prediction results.

Keywords

• Ultra-short-term prediction
• Wind speed
• Smoothing spline preprocessing
• Error optimization theory