基于Tri-SE-CNN的风电机组叶片结冰检测研究  

RESEARCH ON ICING DETECTION OF WIND TURBINE BLADES BASED ON Tri-SE-CNN

作　　者：孙坚[1,2] 杨宇兵 Sun Jian;Yang Yubing(School of Electrical and New Energy,China Three Gorges University,Yichang 443002,China;Hubei Provincial Collaborative Innovation Center for New Energy Microgrid,China Three Gorges University,Yichang 443002,China)

机构地区：[1]三峡大学电气与新能源学院,宜昌443002 [2]新能源微电网湖北省协同创新中心(三峡大学),宜昌443002

出　　处：《太阳能学报》2024年第12期360-369,共10页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(52077120);三峡大学科学基金(KJ20A016)。

摘　　要：针对现有风力机叶片结冰检测方法未能充分利用无标签数据,且分类性能差的问题,提出一种基于改进的三重训练和卷积神经网络(Tri-SE-CNN)的结冰检测方法。首先建立基于最优加权策略的三重训练(Tri-training)模型,对无标签样本的状态进行判别,用以扩充训练集;接着将压缩与激励(SE)模块嵌入到卷积神经网络(CNN)中,并用SE-CNN分类器学习扩充后的样本集。结合提取的叶片结冰主控特征,以2017年工业大数据创新竞赛平台中15号和21号风力机数据为例进行仿真,并用云南某风场历史数据进行验证。实验结果表明,所提方法的准确度优于CNN、支持向量机等方法,在15号风力机上达到99.96%,可为风力机叶片结冰预警提供有益参考。Aiming to address the issue of existing icing detection methods for wind turbine blades,which fail to fully utilize unlabeled data and have poor classification performance,we propose an approach called Tri-SE-CNN based on improved Tri-training and convolutional neural networks.Firstly,establish a Tri-training model based on an optimal weighted strategy to distinguish the state of unlabeled samples and expand the training set.The expanded sample set is learned by the SE-CNN model that embeds the Squeeze-and-Excite(SE)module into Convolutional Neural Networks(CNN).Combined with the strong correlation characteristics of blade icing,this paper utilizes the data from wind turbines No.15 and No.21,which were provided by the Industrial Big Data Innovation Competition Platform in China in 2017,for simulation.Additionally,data from a wind farm in Yunnan,China,are used for verification.The experimental results show that the proposed method achieves higher accuracy than CNN,support vector machine and other methods.Specifically,it achieves an accuracy of 99.96%on the No.15 wind turbine,which can provide valuable references for early warning systems regarding wind turbine blade icing.

关 键 词：风电机组叶片 无标签数据 卷积神经网络 三重训练 压缩和激励网络 结冰检测 

分 类 号：TM315[电气工程—电机]