摘要
金属氧化物避雷器外壳受灰尘、盐、碱等污染物长期沉积的影响,会导致其过早失效,因此,准确地检测金属氧化物避雷器表面的污染状态是非常重要的。基于此,面向饱和积污提出了一种利用红外热成像技术和迁移学习检测金属氧化物避雷器外壳污染严重程度的新方法。该方法通过拍摄不同污染状态下金属氧化物避雷器的红外热图像,经过适当的预处理后,捕获的红外热图像被送入预训练卷积神经网络“ResNet50”进行自动特征提取。提取的深度特征被输入到4个机器学习分类器,即k-最近邻分类器、支持向量机分类器、贝叶斯分类器和随机森林分类器中进行分类。其中,使用随机森林分类器获得了最佳性能。最后,通过仿真分析,结果表明所提方法可精确地感知金属氧化物避雷器表面的污染严重程度。
The metal oxide arrester shell is affected by the long-term deposition of dust,salt,alkali and other pollutants,which will lead to its premature failure.Therefore,it is very important to accurately de⁃tect the contamination state of the metal oxide arrester surface.Based on this,this paper proposes a new method to detect the pollution severity of metal oxide arrester shell based on infrared thermal imaging technology and transfer learning.The method takes infrared thermal images of metal oxide arresters under different pollution conditions.After proper preprocessing,the captured infrared thermal images are sent to the pre-training convolutional neural network ResNet50 for automatic feature extraction.The extrac⁃ted depth features are input into four machine learning classifiers,namely k-nearest neighbor classifier,support vector machine classifier,Bayesian classifier and random forest classifier.The best performance is obtained by using random forest classifier.Finally,the simulation results show that the proposed meth⁃od can accurately perceive the pollution severity of metal oxide arrester surface.
作者
沈顺群
宫旻
刘凯
万教智
李冰然
傅洪全
SHEN Shunqun;GONG Min;LIU Kai;WAN Jiaozhi;LI Bingran;FU Hongquan(State Grid Hubei Electric Power Co.,Ltd.,Shennongjia Power Supply Company,Shennongjia 442400,China;Skill Training Center of State Grid Jiangsu Electric Power Co.,Ltd.,Suzhou 215000,China)
出处
《电瓷避雷器》
CAS
2024年第1期1-8,共8页
Insulators and Surge Arresters
基金
国家自然自然科学基金面上项目(编号:51977123)。
关键词
红外热成像
卷积神经网络
迁移学习
金属氧化物避雷器
污染严重程度分类和状态监测
Infrared thermal imaging
convolutional neural network
transfer learning
metal oxide surge arrester
pollution severity classification and condition monitoring
