基于谱聚类和多元变分模态分解的风电机组功率预测  

Power Prediction of Wind Turbine Units Based on Spectral Clustering and Multivariate Variational Mode Decomposition

作　　者：徐睿麟 郑建勇[2] 梅飞[3] 解洋 XU Ruiin;ZHENG Jianyong;MEI Fei;XIE Yang(Suzhou Research Institute of Southeast University,Suzhou 215123,Jiangsu Province,China;School of Electrical Engineering,Southeast University,Nanjing 210096,Jiangsu Province,China;College of Electrical and Power Engineering,Hohai University,Nanjing 211100,Jiangsu Province,China;School of Cyber Science and Engineering,Southeast University,Nanjing 211189,Jiangsu Province,China)

机构地区：[1]东南大学苏州研究院,江苏省苏州市215123 [2]东南大学电气工程学院,江苏省南京市210096 [3]河海大学电气与动力工程学院,江苏省南京市211100 [4]东南大学网络空间安全学院,江苏省南京市211189

出　　处：《电网技术》2024年第5期2043-2053,I0066,共12页Power System Technology

基　　金：江苏省重点研发计划项目(BE2020027)。

摘　　要：传统的风电出力预测方法通常未能充分考虑机组之间的相似性和差异性,忽视了环境条件对风电出力的影响以及关键特征提取方法单一等问题。因此,提出了一种基于谱聚类和多元变分模态分解的人工神经网络风电出力预测方法。首先,为捕捉不同机组之间的相似性和差异性,对风速和风向进行谱聚类,构建风速-风向二维标签簇,并选取每个簇的中心机组以表征该簇的出力特征。接着,为更全面地描述出力与环境条件之间的关系,采用变分模态分解算法对聚类中心机组出力进行分解,同时将出力与风速、风向数据进行多元变分模态分解,得到不同频率的模态成分。最后,在预测阶段引入基于注意力机制的深度学习网络,对特征模态添加注意力机制后输入卷积长短期神经网络模型进行训练和预测,并通过误差修正模块得到同簇其他机组的预测结果。该方法相较传统方法在预测精确度上有明显提升,具有一定的实用性和有效性。Traditional wind power output forecasting methods need to fully consider the similarities and differences between units,neglect the influence of environmental conditions on wind power output,and have single-feature extraction methods.Therefore,this paper proposes an artificial neural network wind power output forecasting method based on spectral clustering and multivariate variational mode decomposition.Firstly,to capture the similarities and differences between different units,the wind speed and wind direction are clustered using spectral clustering,and a two-dimensional wind speed-wind direction label cluster is constructed.The center unit of each cluster is selected to represent the output characteristics of the cluster.Then,to comprehensively describe the relationship between output and environmental conditions,the variational mode decomposition algorithm is used to decompose the output of the clustering center units,and the output is decomposed into different frequency modal components together with wind speed and wind direction data.Finally,in the prediction stage,a deep learning network based on the attention mechanism is introduced,and the feature intrinsic mode functions are input into the convolutional long short-term memory neural network model after adding the attention mechanism for training and prediction.The prediction results of other target units in the same cluster are obtained through an error correction module.Compared with traditional methods,this method significantly improves prediction accuracy and has practicality and effectiveness.

关 键 词：风电预测 谱聚类 多元变分模态分解 卷积长短期神经网络 注意力机制 

分 类 号：TM614[电气工程—电力系统及自动化]