基于同步压缩小波变换和ResNet的变压器放电故障诊断方法

Method of transformer discharge fault diagnosis based on synchrosqueezed wavelet transform and ResNet

为实现电力变压器运行状态的智能监测和有效辨识,文中提出一种基于同步压缩小波变换图谱与残差神经网络(ResNet)的变压器放电故障诊断方法。利用同步压缩小波变换技术将采集得到的原始声纹进行相应的时频变换,进而得到不同状态下的时频图谱数据集;然后利用残差神经网络实现电力变压器不同状态的辨识;最后,搭建包含三种电力变压器设备典型放电故障的试验对系统进行模拟测试。试验结果表明:所提方法不仅能够有效表征变压器不同的工作状态,而且辨识精度有显著提升,相比于常规方法提升约10%。

In order to realize the intelligent monitoring and effective identification of power transformers operation state,a method of transformer discharge fault diagnosis based on synchrosqueezed wavelet transform and residual neural network(ResNet)is proposed.The collected original voiceprints are subjected to corresponding time⁃frequency transformations by means of the synchrosqueezed wavelet transform,thereby obtaining time⁃frequency atlas data sets in different states.The ResNet is used to identify different states of the transformers.The experimental simulation testing system including three typical discharge faults of transformer equipment was established.The experimental results show that this method can not only effectively characterize the different working states of the transformer,but also significantly improve the recognition accuracy by about 10%compared with the traditional methods.