期刊
IEEE ACCESS
卷 9, 期 -, 页码 61739-61751出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/ACCESS.2021.3073995
关键词
Predictive models; Wind power generation; Training; Prediction algorithms; Upper bound; Logic gates; Artificial neural networks; Gated recurrent unit; lower upper bound estimation; clustered wind power prediction; devised loss functions; error interval coefficients
资金
- National Key Research and Development Program of China (Technology and application of wind power/photovoltaic power prediction for promoting renewable energy consumption) [2018YFB0904200]
- Complement Science and Technology Program (Research on short term wind power prediction technology of wind farm clusters based on deep learning methods) of Inner Mongolia Power (Group) Company Ltd. [DUKZZZ-YBHT-2019-JSC0405-0085]
A novel lower upper bound estimation model based on the gated recurrent unit was proposed for clustered wind power forecasting. The model directly realizes interval prediction and introduces an unsupervised learning strategy to construct error interval coefficients. Additionally, loss functions related to the characteristics of the prediction interval are designed and an effective gradient descent algorithm is adopted to optimize the model.
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.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据