State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
Qiyou Xu
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
Hongyu Wang
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
Jianxun Lang
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
Wenze Li
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China
Interval prediction is essential to improve the scheduling and planning of wind power systems. In this study, a novel lower upper bound estimation model based on the gated recurrent unit was proposed for the clustered wind power forecasting. Different from existing research, the proposed model directly realizes interval prediction based on the point prediction results and the corresponding error interval coefficients, and an unsupervised learning strategy is introduced to construct the error interval coefficients. In addition, loss functions related to the characteristics of the prediction interval are designed, and an effective gradient descent algorithm is adopted to optimize the entire model. In the comparative experiments, two clustered data were collected as experimental data, and seven representative models were selected as benchmark models, which fully proved the superiority of the proposed model.
