| 引用本文: | 杨茂,代博祉,马志远,王勃,王钊,苏欣.基于多尺度特征融合与多任务学习的短期风电功率预测方法[J].电力自动化设备,2026,46(5):110-117 |
| Yang Mao,Dai Bozhi,Ma Zhiyuan,Wang Bo,Wang Zhao,Su Xin.Short-term wind power forecasting method based on multi-scale feature fusion and multi-task learning[J].Electric Power Automation Equipment,2026,46(5):110-117 |
|
| 摘要: |
| 为解决广域空间尺度下高维气象特征导致的风电场集群功率预测的高复杂度建模问题，提出一种基于多尺度特征融合与多任务学习的短期风电功率预测方法，其中多尺度特征融合实现了全局、集群、场站3种尺度的特征融合。从全局、集群、场站3个角度分别构建特征提取模块，并基于全局块级注意力与动态混合的稀疏图注意力降低模型的复杂度，提高运算效率；将集群预测任务分解为直接预测、间接预测、融合预测3个任务，通过多任务学习实现集群预测。将所提方法应用于吉林省某风电集群，结果表明，所提方法可以有效提升模型的运算效率和风电功率的预测精度。 |
| 关键词: 风电 风电场集群 功率预测 特征提取 多任务学习 Transformer 图注意力 |
| DOI：10.16081/j.epae.202601010 |
| 分类号:TM614 |
| 基金项目:国家自然科学基金青年科学基金资助项目(基于全局气象特征感知的省级风电可靠供电能力预测)(52307151) |
|
| Short-term wind power forecasting method based on multi-scale feature fusion and multi-task learning |
|
Yang Mao1, Dai Bozhi1,2, Ma Zhiyuan3, Wang Bo4, Wang Zhao4, Su Xin1
|
|
1.Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education, Northeast Electric Power University, Jilin 132012, China;2.Zhongshan Power Supply Bureau of Guangdong Power Grid Co.,Ltd.,Zhongshan 528400, China;3.Electric Power Research Institute of State Grid Inner Mongolia East Electric Power Co.,Ltd.,Hohhot 010020, China;4.National Key Laboratory of Renewable Energy Grid-Integration of China Electric Power Research Institute, Beijing 100192, China
|
| Abstract: |
| In order to solve the problem of high-complexity modeling for wind farm cluster power forecasting caused by high-dimensional meteorological features in wide-area spatial scale, a short-term wind power forecasting method based on multi-scale feature fusion and multi-task learning is proposed, in which, the multi-scale feature fusion realizes the feature fusion of global, cluster and station scales. The feature extraction modules are constructed from global, cluster and station perspectives respectively, and on the basis of global block-wise attention and dynamic hybrid sparse graph attention, the model complexity is reduced and the computational efficiency is improved. The cluster forecasting task is decomposed into three tasks of direct prediction, indirect prediction, and fused prediction, and the cluster forecasting is realized through multi-task learning. The proposed method is applied in a wind farm cluster of Jilin Province, and the results show that the proposed method can effectively improve the model computational efficiency and wind power forecasting accuracy. |
| Key words: wind power wind farm cluster power forecasting feature extraction multi-task learning Transformer graph attention |
