A CGRU multi-step wind speed forecasting model based on multi-label specific XGBoost feature selection and secondary decomposition. (February 2023)

• Record Type: Journal Article
• Authors: Jiang, Zheyong; Che, Jinxing; He, Mingjun; Yuan, Fang
• Is Part Of: Renewable energy. Volume 203(2023)
• Page Start: 802
• Page End: 827
• Publication Date: 2023-02
• Subjects: Secondary decomposition -- Sample entropy -- Multi-label XGBoost feature selection -- Convolutional gated recurrent unit network -- Multi-step wind speed forecasting; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2022.12.124
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: Accurate wind speed forecasting is beneficial to reduce the risk of power system caused by wind power uncertainty, which is of great significance for wind power grid connection. However, the randomness and intermittence of wind speed bring great challenges to the accurate wind speed forecasting. In this study, a CGRU multi-step wind speed forecasting model based on secondary decomposition (SD) and multi-label specific XGBoost feature selection method is proposed, which achieves good forecasting performance. By using the secondary decomposition and sample entropy analysis, the original wind speed series is decomposed into multiple sub-series, and the frequency division characteristics of wind speed fluctuations are further extracted. Previous studies rarely focused on feature selection for multi-step wind speed forecasting, the feature selection method suitable for multi-step wind speed forecasting is proposed in this study. By using this method, the optimal input features of each sub-series are selected, which simplifies the data structure and improves the modeling efficiency. An efficient CGRU forecasting model is developed in this study, which can capture the depth characteristics and obtain the time-dependent relationship of wind speed. Experiments conducted in four seasons show that the model can adapt to long-term dependence and extract effective information from original data.