| 引用本文: | 杨茂,韩超,张薇.基于深度图聚类和特征重构的风电集群功率短期预测方法[J].电力自动化设备,2025,45(4):53-59 |
| YANG Mao,HAN Chao,ZHANG Wei.Short-term power prediction method for wind farm cluster based on deep graph clustering and feature reconstruction[J].Electric Power Automation Equipment,2025,45(4):53-59 |
|
| |
|
|
| 本文已被：浏览 1427次 下载 551次 |
码上扫一扫！ |
|
|
| 基于深度图聚类和特征重构的风电集群功率短期预测方法 |
|
杨茂, 韩超, 张薇
|
|
东北电力大学 现代电力系统仿真控制与绿色电能新技术教育部重点实验室，吉林 吉林 132012
|
|
| 摘要: |
| 针对当前短期风电集群功率预测方法难以充分提取时空特征实现高精度集群预测的问题，提出一种基于深度嵌入式图注意力聚类、改进自适应噪声完备集合经验模态分解和长短期时间序列网络的风电集群功率短期预测方法。基于地理位置信息构建图注意力网络，指导深度嵌入式图注意力聚类算法通过预报风速实现有效的集群划分，通过自适应噪声完备集合经验模态分解算法分别对每个类别的风电功率和风速进行分解；根据各分量的排列熵将分解后的风电功率序列和风速序列分别重构为随机分量、振荡分量和趋势分量；通过长短期时间序列网络模型得到预测结果。将所提方法应用于中国东北部某大规模风电集群，结果表明，所提预测方法的均方根误差、平均绝对误差和准确率分别为0.063 76、0.052 31和93.62%，优于对比方法，验证了所提方法的有效性。 |
| 关键词: 风电功率预测 图注意力网络 集群划分 深度学习 特征重构 |
| DOI：10.16081/j.epae.202412025 |
| 分类号:TM614 |
| 基金项目:国家重点研发计划项目(2022YFB2403000) |
|
| Short-term power prediction method for wind farm cluster based on deep graph clustering and feature reconstruction |
|
YANG Mao, HAN Chao, ZHANG Wei
|
|
Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education, Northeast Electric Power University, Jilin 132012, China
|
| Abstract: |
| Aiming at the problem that it is difficult for the current short-term wind power cluster prediction methods to sufficiently extract spatiotemporal features to achieve high-precision cluster prediction. A short-term power prediction method for wind power clusters is proposed based on deep attentional embedded graph attention clustering, improved complete ensemble empirical mode decomposition with adaptive noise and long-term and short-term time series network. Based on geographic location information, the graph attention network is constructed to guide deep attentional embedded graph clustering algorithms to achieve effective cluster partitioning by predicting wind speed. The improved complete ensemble empirical mode decomposition with adaptive noise algorithm is used to decompose the wind power and wind speed separately of each category. The decomposed wind power sequences and wind speed sequences based on the arrangement entropy of each component are reconstructed respectively into random components, oscillation components and trend components. The prediction results are obtained through the long-term and short-term time series network model. The proposed method is applied to a large-scale wind farm cluster in northeastern China, the results show that the root mean square error, mean absolute error and accuracy rate of the proposed prediction method are 0.063 76,0.052 31 and 93.62% respectively, which are superior to the comparison method, and the effectiveness of the proposed method is verified. |
| Key words: wind power prediction graph attention network cluster partitioning deep learning feature reconstruction |
|
