摘要
机械故障是气体绝缘组合电器(gas insulated switchgear,GIS)最主要的故障之一,尚缺乏有效的在线诊断方法。为此该文利用GIS断路器激振能力强的优势,以GIS断路器分合操作为激振源,根据振动信号所包含的设备机械特性信息对GIS各构件的机械状态进行评估。搭建了包含3种GIS典型机械故障的试验模拟测试系统,并提出了一种基于S变换D-SVMAlex Net模型的GIS机械故障诊断方法。利用S变换处理断路器动作激发的非平稳振动信号,得到含有设备机械特征的时频图谱;建立D-SVM AlexNet模型,使用预训练的Alex Net神经网络模型提取S变换图像特征作为预测变量,通过Fitcecoc函数拟合支持向量机(support vector machine,SVM)进行图像预分类,根据模糊矩阵显示的分类结果筛选出有效测点;将有效测点的时频图送入AlexNet进行迁移学习,获得经微调后的神经网络模型。实验验证发现,训练完毕的卷积神经网络的故障诊断训练准确率达到99%,验证准确率达到92%,具备较好的时频图像分类效果,可实现GIS机械故障的有效诊断。
Mechanical fault is one of the major faults in gas insulated switchgear(GIS),and there is still no effective method for online diagnosis of GIS mechanical faults.Therefore,in this paper,by taking advantage of the strong vibration ability of GIS circuit breakers,the split and close operation of GIS circuit breakers were adopted as the excitation source to evaluate the GIS mechanical state of each component according to the mechanical characteristic information of equipment contained in the vibration signals.Moreover,three kinds of GIS typical mechanical fault test simulation systems were set up,and a GIS mechanical fault diagnosis method based on S Transform D-SVM AlexNet model was proposed.The vibration signal excited by the operation of the circuit breaker was processed by the S Transform,and the time-frequency spectrum containing mechanical characteristics was obtained.The D-SVM AlexNet model was established and the pre-trained AlexNet neural network model was used to extract the S Transform image features as predictive variables.The support vector machine(SVM)was fitted with the Fitcecoc function for image pre-classification.According to the classification results of fuzzy matrix,effective measuring points were screened out.Time-frequency diagrams of effective measuring points were sent to AlexNet for transfer learning,and the fine-tuned neural network model was obtained.The experimental results show that the training accuracy of the model is 99% and the verification accuracy is 92%,which has a better time-frequency image classification effect and can realize the effective diagnosis of GIS mechanical faults.
作者
刘宝稳
汤容川
马钲洲
马宏忠
许洪华
LIU Baowen;TANG Rongchuan;MA Zhengzhou;MA Hongzhong;XU Honghua(College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China;Nanjing Power Supply Company,State Grid Jiangsu Electric Power Company,Nanjing 210008,China)
出处
《高电压技术》
EI
CAS
CSCD
北大核心
2021年第7期2526-2535,共10页
High Voltage Engineering
基金
江苏省自然科学基金青年基金(BK20190490)
中央高校基本科研业务费项目(B200202173)
中国博士后科学基金面上项目(2020M671318)。
作者简介
通讯作者:刘宝稳,1988-,男,博士,讲师,主要从事电力设备故障诊断、小电流接地系统保护与控制相关的研究工作,E-mail:lbw_5566@163.com;汤容川,1998-,男,硕士生,主要从事电力系统及其自动化方向的研究,E-mail:tangrc@hhu.edu.cn;马钲洲,1999-,男,博士生,主要从事电力系统及其自动化方向的研究,E-mail:js_mzz@163.com;马宏忠,1962-,男,博士,教授,博导,从事电气设备状态监测与故障诊断、磁悬浮承重技术、电力系统谐波分析研究,E-mail:hhumhz@163.com。
