摘要
针对ZPW-2000R型无绝缘轨道电路的多样性和复杂性造成故障诊断准确率低的问题,从故障特征提取和特征分类两方面出发,将栈式稀疏自编码器(SSAE,Stacked Sparse Auto-Encoder)和深度神经网络(DNN,Deep Neural Networks)相结合,提出了基于SSAE-DNN模型的故障诊断方法。采用SSAE对故障数据以无监督的方式进行降维和特征提取,获得最优网络参数,从而挖掘无绝缘轨道电路不同故障特征信息,并将SSAE提取的特征样本导入DNN,得到SSAE-DNN模型,据此进行无绝缘轨道电路的故障分类识别。试验结果表明,该模型对故障数据进行了降维,减少了故障诊断时间,且获取了故障数据的深层特征;对ZPW-2000R型无绝缘轨道电路的故障具有较高的诊断准确率,仅有少数故障出现误判情况。通过与反向传播(BP,Back Propagation)神经网络、卷积神经网络(CNN,Convolutional Neural Network)和支持向量机(SVM,Support Vector Machine)的对比试验,进一步验证了该方法的有效性和优越性。
In response to the low accuracy of fault diagnosis caused by the diversity and complexity of ZPW-2000R uninsulated track circuits,this paper proposed a fault diagnosis method based on the SSAE-DNN model by combining Stacked Sparse Autoencoder(SSAE)and Deep Neural Networks(DNN)from two aspects of fault feature extraction and feature classification.The paper used SSAE to unsupervised dimensionality reduction and feature extraction of fault data,obtained the optimal network parameters to mine different fault feature information of uninsulated track circuits.The feature samples extracted by SSAE were imported into DNN to obtain the SSAE-DNN model for fault classification and recognition of uninsulated track circuits.The experimental results show that the model reduces the dimensionality of fault data,reduces fault diagnosis time,and obtains deep features of fault data,has a high diagnostic accuracy for faults in ZPW-2000R uninsulated track circuits,with only a few faults experiencing misjudgment.The effectiveness and superiority of this method were further verified through comparative experiments with backpropagation(BP)neural networks,convolutional neural networks(CNN),and support vector machines(SVM).
作者
王国保
蔡水涌
杨红刚
谢本凯
黄春雷
WANG Guobao;CAI Shuiyong;YANG Honggang;XIE Benkai;HUANG Chunlei(School of Management Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450046,China;Heilongjiang Ruixing Technology Co.Ltd.,Harbin 150030,China)
出处
《铁路计算机应用》
2023年第12期6-12,共7页
Railway Computer Application
基金
河南省软科学研究计划项目(212400410099)。
关键词
自编码器
深度神经网络
特征提取
故障诊断
无绝缘轨道电路
autoencoder
deep neural network
feature extraction
fault diagnosis
uninsulated track circuit
