摘要
现有瓷支柱绝缘子振动声学检测仪的检测结果仅依赖信号的功率谱特征,单一特征携带的信息量有限,有缺陷的绝缘子振动信号样本少,信号特征干扰因素多,且人工判断容易引起漏报、误报。针对以上问题,提出一种包含时域特征、功率谱峰值特征和各频段小波包能量比组合特征的随机森林故障自动识别模型。通过在绝缘子上、下端人工制作不同程度的缺陷,采集多组数据并进行特征筛选和模型参数调优,实验结果表明,使用该方法检测35 kV绝缘子,故障识别准确率达到96.53%,并可区分上端缺陷和下端缺陷。
The detection results of the existing porcelain post insulator vibration acoustic detector only rely on the power spectrum characteristics of the signal,and the information carried by a single feature is limited.There are few samples of defective insulator vibration signals,many interference factors of signal characteristics,and artificial judgment is easy to cause missing and false positives.Aiming at the above problems,this paper proposes a random forest fault automatic identification model including time domain features,power spectrum peak features and the combined features of the energy ratio of wavelet packets in each frequency band.By artificially creating different degrees of defects on the upper and lower ends of the insulator,collecting multiple sets of data and performing feature screening and model parameter tuning,the experimental results show that using this method to detect 35 kV insulators,the fault identification accuracy rate reaches 96.53%,and can be distinguished upper and lower end defects.
作者
赵洲峰
赵志勇
邹君文
裘吕超
杨斌
吕福在
ZHAO Zhoufeng;ZHAO Zhiyong;ZOU Junwen;QIU Lüchao;YANG Bin;LV Fuzai(Hangzhou Yineng Power Technology Co.,Ltd,Hangzhou,Zhejiang 310000,China;Polytechnic Institute,Zhejiang University,Hangzhou,Zhejiang 310015,China;Zhejiang Electric Boiler Pressure Vessel Inspection Institute Co.,LTD,Hangzhou,Zhejiang 310000,China;Electric Power Research Institute,State Grid Zhejiang Electric Power Co.,LTD,Hangzhou,Zhejiang 310000,China;Hangzhou Zheda Jingyi Electromechanical Technology Co.,Ltd.,LTD,Hangzhou,Zhejiang 311100,China;School of Mechanical Engineering,Zhejiang University,Hangzhou,Zhejiang 310013,China)
出处
《广东电力》
2022年第12期93-100,共8页
Guangdong Electric Power
关键词
瓷支柱绝缘子
振动声学
故障识别
组合特征
随机森林
porcelain post insulator
vibration acoustics
fault diagnosis
combination characteristics
random forest