| 摘 要: 为提高风电功率多步预测精度,提出了一种融合长期与短期特征的深度学习预测模型。首先,采用变分模态分解(VMD)将风电功率数据分解为高频和低频序列;随后,将得到的序列输入双向时间卷积网络(BiTCN)和双向门控循环单元(BiGRU)组成的并联结构中,并行提取短期和长期特征;最后,通过交叉注意力机制加权融合提取到长短期特征。实验结果表明,模型在风电功率的4步、8步、16步和24步预测中的 R2 分别为0.992、0.976、0.969和0.933,均优于对比模型,验证了该模型在风电功率多步预测中的高精度和实用性。 |
| 关键词: 风电功率 变分模态分解 时间卷积网络 门控循环单元 交叉注意力 |
|
中图分类号: TP183
文献标识码: A
|
|
| Multi-step Wind Power Prediction Integrating Long-Short-Term Features of Variational Modes |
|
CHEN Xiao1,2, YAN Hao1,2
|
(1.School of Electronics & Information Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China;
2.Nanjing Xiaoyang Electronic Science and Technology Co., Ltd, Nanjing 211500, China)
chenxiao@nuist.edu.cn; 202212490240@nuist.edu.cn
|
| Abstract: To improve the accuracy of mult-i step wind power prediction, a deep learning prediction model integrating long-term and shor-t term features is proposed. Firstly, Variational Mode Decomposition (VMD) is employed to decompose the wind power data into high-frequency and low-frequency series. Subsequently, the obtained series are input into a parallel structure composed of a Bidirectional Time Convolutional Network (BiTCN) and a Bidirectional Gated Recurrent Unit (BiGRU) to extract shor-t term and long-term features in parallel. Finally, the extracted long-and shor-t term features are weighted and fused via a cross-attention mechanism. Experimental results show that the model’s R2 values for 4-step, 8-step, 16-step, and 24-step wind power predictions are 0.992, 0.976, 0.969, and 0.933, respectively,all superior to those of the comparison models. This verifies the high accuracy and practicality of the proposed model in mult-i step wind power prediction. |
| Keywords: wind power Variational Mode Decomposition (VMD) time convolutional network gated recurrent unit cross-attention |
