基于M-LSTM与Resnet并联网络的同轴线缆开路故障定位方法

Open-circuit Fault Location Method of Coaxial Cable Based on M-LSTM and Resnet Parallel Network

为了同轴线缆的高效维护和修理,精确、有效地定位同轴线缆开路故障非常关键,提出一种基于M-LSTM与Resnet并联网络的开路故障定位方法 SROCFL。该方法利用HFSS软件建立单端带有激励信号源的同轴线缆开路故障模型,通过对故障模型仿真,获取了在阻抗不连续和有噪声情况下的电压幅值序列数据;提出矩阵式LSTM (M-LSTM),用于充分提取开路故障电压幅值序列的时域特征;再对电压幅值序列做希尔伯特-黄变换获得时频域分布谱图,引入Resnet,提取时频域分布谱图的局部时频域特征。该方法利用M-LSTM与Resnet并联网络捕获不同层次的特征信息,能增强局部时域特征,减少时频域变换带来的信息损失。相比于现有方法,实验结果表明,SROCFL方法在机载同轴线缆开路故障定位中表现出更好的性能。

In order to maintain and repair the coaxial cable efficiently,it is very important to locate the open circuit fault of the coaxial cable accurately and effectively.An open circuit fault location method SROCFL based on M-LSTM and Resnet parallel network is proposed.In this method,HFSS software is used to establish an open circuit fault model of coaxial cable with an excitation signal source at one end.Matrix LSTM(M-LSTM)is proposed to fully extract the time-domain characteristics of open circuit fault voltage amplitude sequence.Then Hilbert Huang transform is performed on the voltage amplitude sequence to obtain the time-frequency distribution spectrum.Resnet is introduced to extract the local time-frequency characteristics of the time-frequency distribution spectrum.The method uses M-LSTM and Resnet parallel networks to capture different levels of feature information,enhances local time-domain features,and reduces the information loss caused by time-frequency transformation.Compared with the existing methods,the experimental results show that SROCFL method has better performance in the open circuit fault location of airborne coaxial cables.