高精度配电网电气设备故障识别检测方法  

Fault identification and detection method with high precision for electrical equipment in distribution network

作　　者：赵欢 阳浩 何亮 魏恩伟 郑杰 ZHAO Huan;YANG Hao;HE Liang;WEI En-wei;ZHENG Jie(Shenzhen Power Supply Bureau Co.Ltd. , China Southern Power Grid, Shenzhen 518000, China)

机构地区：[1]中国南方电网深圳供电局电力科学研究院,广东深圳518000

出　　处：《沈阳工业大学学报》2021年第6期614-618,共5页Journal of Shenyang University of Technology

基　　金：中国南方电网项目(090000KK52170154).

摘　　要：针对电气设备红外图像边界模糊、噪声大等问题,结合卷积神经网络模型和图像识别技术,利用可见光图像与红外成像,实现了对配电网电气设备的高精度远程识别和发热诊断.采用卷积神经网络和边框回归算法完成了对识别对象的标记,基于灰度梯度信息矩阵提取了配电网红外图像的纹理信息特征参数,采用主成分分析的方法得到特征参数的主成分分量,并将其作为输入向量,对设备运行状态进行识别.结果表明,样本训练及测试的准确率能够分别达到95%、90%以上,设备发热故障识别准确率约为85%.Aiming at the problems of fuzzy boundary and high noise in infrared image of electrical equipment,in combination with convolutional neural network model and image recognition technology,and by using visible light image and infrared imaging,the high-precision remote recognition and exothermic diagnosis of electrical equipment in distribution network were realized.The convolutional neural network and border regression algorithm were used to mark the recognition object.The texture information feature parameters of infrared image of distribution network were extracted in terms of gray-gradient information matrix.A principal component analysis method was used to obtain the principal component of feature parameters,which was used as the input vector to recognize the running state of equipment.The results show that the accuracy of sample training and testing can reach 95%and 90%,respectively,and the accuracy of equipment exothermic fault recognition is about 85%.

关 键 词：卷积神经网络 图像识别 红外成像 灰度梯度信息矩阵 主成分分析 故障识别 边框回归算法 对象识别 

分 类 号：TM744[电气工程—电力系统及自动化]