基于多传感器融合声纹特征图谱的变压器铁芯松动故障诊断方法  

Fault diagnosis method for transformer core looseness based on multi-sensor fusion voiceprint feature map

作　　者：李楠 马宏忠[1] 段大卫 朱昊 何萍[2] LI Nan;MA Hongzhong;DUAN Dawei;ZHU Hao;HE Ping(College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China;State Grid Nanjing Power Supply Company,Nanjing 210008,China)

机构地区：[1]河海大学能源与电气学院,南京211100 [2]国网南京供电公司,南京210008

出　　处：《振动与冲击》2023年第15期129-137,198,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金(51577050);国网江苏省电力公司科技项目(J2021053)。

摘　　要：变压器铁芯轻微松动故障给变压器安全稳定运行留下巨大隐患,目前尚缺乏切实可靠的诊断方法。提出一种基于多传感器融合声纹特征图谱的变压器铁芯松动故障诊断方法。首先,利用4个传感器采集声纹时序序列,通过小波变换生成声纹特征图谱,利用熵权法确定不同传感器信号的权重分配,将4个声纹特征图谱加权融合,从而形成多传感器融合声纹特征图谱。其次,将融合声纹特征图谱输入优化后的ShuffleNetV2模型,通过分组卷积和通道混洗得到铁芯松动程度。最后,通过现场试验验证了方法的有效性。结果表明,所提方法对25%,50%,75%及100%的松动程度均能实现可靠诊断,平均准确率高达99.6%。与采用傅里叶频谱(fast Fourier transform, FFT)、格拉米角场(Gramian angular field, GAF)、马尔可夫变换场(Markov transform field, MTF)以及混沌特征(recurrence plot, RP)等传统声纹特征图谱的诊断相比,所提方法识别准确率提高了12.2%;与采用单传感器声纹特征图谱的诊断相比,所提方法识别准确度提高了5.8%;与采用AlexNet、MobilleNetV2、GoogleNet以及ResNet等卷积神经网络模型的诊断相比,所提方法识别准确率提高了2.7%。Slight looseness of transformer core leaves a huge hidden danger for its safe and stable operation,and there is currently a lack of practical and reliable diagnosis methods.Here,a fault diagnosis method for transformer core looseness based on multi-sensor fusion voiceprint feature map was proposed.Firstly,4 sensors were used to collect time series of voiceprints to generate voiceprint feature maps with wavelet transform.Weight assignments of different sensor signals were determined with entropy weight method,and the 4 voiceprint feature maps were weighted and fused to form a multi-sensor fusion voiceprint feature map.Secondly,the fused voiceprint feature map was input into the optimized ShuffleNetV2 model,iron core looseness degree was obtined through block convolution and channel shuffling.Finally,the effectiveness of the proposed method was verified through field tests.The results showed that the proposed method can reliably diagnose 25%,50%,75%,and 100%of looseness degrees with an average accuracy of 99.6%;compared with diagnoses using traditional voiceprint feature maps,such as,fast Fourier transform(FFT),Gramian angular field(GAF),Markov transform field(MTF),recurrence plot(RP),the recognition accuracy of the proposed method increases by 12.2%;compared with diagnosis using a single sensor voiceprint feature map,the proposed method’s recognition accuracy increases by 5.8%;compared with diagnoses using convolutional neural network models of AlexNet,MobileNetV2,GoogleNet and ResNet,the proposed method’s recognition accuracy increases by 2.7%.

关 键 词：电力变压器 铁芯松动故障 声纹特征图谱 多传感器融合 卷积神经网络 

分 类 号：TM41[电气工程—电器]