基于短时傅里叶变换和深度学习的牵引网过电压辨识  

Traction network overvoltage identification based on short time Fourier transform and deep learning

作　　者：贾君宜 吴命利[1] 宋可荐[1] 王琪 JIA Junyi;WU Mingli;SONG Kejian;WANG Qi(School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044)

机构地区：[1]北京交通大学电气工程学院,北京100044

出　　处：《电气技术》2021年第10期1-10,共10页Electrical Engineering

摘　　要：牵引网过电压严重影响电气化铁路正常运行,对牵引网过电压进行类型辨识有利于提高牵引供电系统的可靠性。针对牵引网过电压的非线性和不稳定性,本文利用短时傅里叶变换将过电压时域波形转化为二维的时频图;先通过局部特征提取和设置阈值,实现对铁磁谐振过电压的快速识别;再利用卷积神经网络的自学习能力挖掘时频图特征与牵引网过电压信号的深层次关系,实现对机车进出分相、断路器开闭操作过电压和高频谐振过电压的识别。实验结果表明,该方法的准确度在90%以上。Traction network overvoltage affects the normal operation of electrified railways.Identification of traction network overvoltage is helpful to improve the reliability of traction power supply system.In view of the nonlinearity and instability of traction network overvoltage,the short-time Fourier transform is used to convert the time-domain waveform of overvoltage into two-dimensional time-frequency diagram.Fast identification of ferromagnetic resonance overvoltage is realized by feature extraction and threshold setting.Then the self-learning ability of convolutional neural network is used to analyze the deep relationship between the time-frequency diagram characteristics and the overvoltage of traction network.The convolutional neural network realizes the identification of into/out neutral-section overvoltage,vacuum circuit breaker overvoltage and high-frequency resonance overvoltage.The test result shows that the accuracy of this method is over 90%.

关 键 词：过电压 短时傅里叶变换 深度学习 时频图 

分 类 号：U223[交通运输工程—道路与铁道工程]