基于放电声音识别的高压电器绝缘监测研究  

High Voltage Apparatus Insulation Monitoring Based on Discharge Acoustic Signal Recognition

作　　者：马文婧 郑欣 鲍克磊 张伟欣 Ma Wenjing;Zheng Xin;Bao Kelei;Zhang Weixing(Guangzhou Power Supply Bureau, Guangdong Power Grid Company, Guangzhou Guangdong 511400, China)

机构地区：[1]广东电网有限责任公司广州供电局,广东广州511400

出　　处：《电气自动化》2021年第2期24-26,共3页Electrical Automation

基　　金：中国南方电网有限责任公司科技项目(GZHKJXM20170135)。

摘　　要：为了监测开关设备耐压试验过程绝缘状态情况,提取试验过程中异常声音信号的语谱图,导入卷积神经网络识别有无放电信号,并基于得到的准确率优化网络模型。使用卷积神经网络能自动提取输入图像特征,建立相应的识别判据,并且能较大提高识别准确率。试验结果表明,卷积神经网络能较好地完成识别任务,平均识别准确率能达到94%以上。In order to monitor the insulation state of switchgear in the withstand voltage test,the spectrogram of abnormal acoustic signal during the test was extracted.The convolutional neural network was imported to recognize whether there was a discharge signal,and the network model was optimized based on the obtained accuracy.The use of convolutional neural network could automatically extract the input image features,establish the corresponding recognition criteria,and greatly improve the recognition accuracy.The experimental results show that the convolutional neural network can complete the recognition task well,and the average recognition accuracy can reach more than 94%.

关 键 词：高压电器 卷积神经网络 在线监测 放电信号识别 语谱图 

分 类 号：TM85[电气工程—高电压与绝缘技术]