检测SF_(6)分解产物的半导体传感器筛选与识别方法研究  

Research on Screening and Identification Methods of Semiconductor Sensors for Detecting SF_(6) Decomposition Products

作　　者：金海勇 周启义 黄小泵 JIN Haiyong;ZHOU Qiyi;HUANG Xiaobeng(Shanghai Roye Electrical Co.,Ltd.,Shanghai 201802,China;State Grid Jinzhou Electric Power Co.,Ltd.of Hubei Province,Hubei Jingzhou 434000,China)

机构地区：[1]上海乐研电气有限公司,上海201802 [2]国网湖北省电力有限公司荆州供电公司,湖北荆州434000

出　　处：《高压电器》2024年第11期201-208,共8页High Voltage Apparatus

基　　金：国网湖北省电力有限公司科技项目(5215J0230002)。

摘　　要：六氟化硫(SF_(6))因其优越的电气性能和化学稳定性成为气体绝缘开关设备(gas insulated switchgear,GIS)等高压电气设备的首选绝缘介质。当SF_(6)电气设备内部发生局部放电等故障时,SF_(6)气体会发生分解反应,产生SO_(2)、H_(2)S等物质。而SF_(6)气体分解产物的组分体积分数反映了电气设备的运行状态。因此,监测设备内部的气体组分能够有效评估SF_(6)电气设备的运行状况,从而对设备内部可能的故障起到预警作用。文中通过搭建气敏测试平台,研究来自中日知名传感器制造企业12种商业传感器在SF_(6)背景下对H_(2)S和SO_(2)气体的响应特性,筛选出4种对两种气体选择性最好的传感器,即MP-5、TGS2611、TGS2612和TGS2618。在实验数据的基础上,构建SF_(6)气体分解产物数据集,对比研究K近邻(KNN)、支持向量机(support vector machines,SVM)、朴素贝叶斯(Naive Bayes)、梯度提升(Gradient Boosting)和极限树(Extra Tree)5种分类算法对SF_(6)气体分解产物的识别准确率。结果表明,极限树具有最高的识别准确率(83.33%),采用极限树算法能够有效识别SF_(6)气体的分解产物。Sulphur hexafluoride(SF_6) is the preferred insulating medium for such high-voltage electrical equipment as gas insulated switchgear(GIS)due to its superior electrical properties and chemical stability.In case of such faults as partial discharge inside SF_(6) electrical equipment,SF_(6) will undergo a decomposition reaction and produce such substances as SO_(2) and H_2S.While,the component content of SF_(6) decomposition products has reflected the operating status of the electrical equipment.Therefore,monitoring the gas components inside the equipment can effectively assess the operating status of SF_(6) electrical equipment,thus providing an early warning of possible failures inside the equipment.In this paper,a gas-sensitive test platform is set up to study the response characteristics of 12 commercial sensors from Japan and China for H_(2)S and SO_(2) gases in the SF_(6) background.Such four types of sensors as MP-5、TGS2611、TGS2612 and TGS2618,which are best for the choice of two kings of gasses,are selected.On the basis of the experimental data,the SF_(6)gas decomposition product data set is constructed to compare the identification accuracy of such five identification algorithms as KNN,support vector machines(SVM),Naive Bayes,Gradient Boosting and Extra Tree.The results show that extra tree has the highest identification accuracy(83.33%),and the use of the extra tree algorithm can effectively identify the decomposition products of SF_(6).

关 键 词：SF_(6)气体 气体传感器阵列 分解产物识别 机器学习 

分 类 号：TM213[一般工业技术—材料科学与工程] TP212[电气工程—电工理论与新技术]