摘要
为解决低温低湿造成风机叶片表面结冰的问题,基于空洞因果卷积网络与知识图谱的风机叶片表面结冰识别方法,根据风机叶片表面结冰机制与特点,利用知识图谱构建方法来建立叶片表面结冰特征空间。在空间内提取与结冰故障的关键特征,通过知识实体融合生成结冰故障特征空间,结合各结冰故障特征的互信息向量和流行度计算特征重要度,筛选结冰异常关键特征,采用空洞因果卷积网络构建结冰识别模型。通过离线训练和在线识别对结冰异常进行检测,完成风机叶片表面结冰识别方法的设计。实验结果表明:所提方法对风机叶片表面结冰进行检测识别,输出的结冰类型与实验数据集中的标注类别相一致,得到的风机叶片表面结冰识别结果的平均K appa系数高于0.4,对风机表面结冰的识别精度较高。
To solve the problem of surface icing of wind turbine blades caused by low temperature and humidity,with the wind turbine blade surface icing recognition method based on hollow causal convolutional network and knowledge graph,a knowledge graph construction method is used to establish a feature space for blade surface icing according to the mechanism and characteristics of surface icing on wind turbine blades.Key features related to icing faults are extracted from the space,and a feature space for icing faults is generated through knowledge entity fusion.By combining the mutual information vectors and popularity of various icing fault characteristics,the importance of features is calculated,the key features of icing anomalies are screened,and a hollow causal convolutional network is applied to construct an icing recognition model.Through offline training and online recognitione,icing anomalies are detected,and the method for identifying icing on the surface of wind turbine blades is designed.The experimental results show that the designed method can detect and identify surface Icing on wind turbine blades,the output icing type is consistent with the annotated categories in the experimental dataset,and the average K appa coefficient of the wind turbine blade surface icing recognition results obtained is higher than 0.4,indicating a high accuracy in identifying surface icing on wind turbines.
作者
戴立伟
DAI Liwei(Datang Liangshan New Energy Co.,Ltd.,Chengdu 615399,China)
出处
《机械制造与自动化》
2025年第4期189-193,共5页
Machine Building & Automation
关键词
空洞因果卷积网络
知识图谱
风机叶片
结冰识别
识别模型
hollow causal convolutional network
knowledge graph
fan blades
ice recognition
identification model
作者简介
戴立伟(1991-),男,内蒙古自治区兴安盟人,工程师,硕士,研究方向为计算机视觉算法、神经网络算法、目标检测、知识图谱、语义大模型等,xie4214100@163.com。
