Research On ZPW-2000A Track Circuit Fault Classification Method Based On Stacked Autoencoder

ZPW-2000 A track circuit ensures the safe operation of the train,and is an indispensable part of the railway operation equipment.If the track circuit fails,it will affect the traffic safety,even lead to train collision and other accidents,resulting in casualties and economic losses.Through the timely maintenance of the equipment with track circuit fault,the train operation is safe and reliable,and then the transportation efficiency of the train is improved.At present,the actual detection and diagnosis of track circuit fault is still based on the experience of field maintenance personnel to test and analyze,which is too dependent on the field personnel,so that the fault judgment time is long,the efficiency is reduced,the discrimination error and other problems are easy to appear.In order to solve the above problems,this dissertation combines the Kernel Principal Component Analysis(KPCA)and Stack Autoencoder(SAE)theory,and according to the actual fault data of track circuit,constructs ZPW-2000 A track circuit fault classification model.The main contents of this dissertation are as follows:First of all,the basic principle and structure of ZPW-2000 A track circuit are analyzed.According to the actual historical fault monitoring data,the voltage value of each equipment in the track circuit is analyzed.The transmission power supply value,power output voltage,distribution panel voltage,rail surface voltage,rail surface current,receiving cable side voltage,receiving equipment side voltage,XGJ voltage,track relay voltage,main rail output voltage in the original data as a parameter of historical data.The feature parameter extraction is completed and the fault categories are determined.Then,for the problems of information redundancy and noise that may occur in the historical fault data collected at the site,KPCA algorithm is used to reduce the original data.The original data need to be standardized.The corresponding kernel matrix is calculated by the kernel function,and the kernel matrix is centralized.The eigenvalue and eigenvector of the kernel matrix are obtained.And further calculate the cumulative contribution rate of the principal component corresponding to the eigenvalue.According to the cumulative contribution rate,the principal component that can reflect the characterization of the original data to the greatest extent is selected to realize the dimensionality reduction of the original data.Secondly,the data reduced by KPCA algorithm is divided into training set and test set,which are used as the input of trestle self coding network,and the classification model is pre-trained by greedy and unsupervised training method layer by layer.The ZPW-2000 A track circuit fault classification model based on KPCA-SAE network is established,and the test set is classified and tested.Finally,the classification results of the classification model in this dissertation are compared with the fault classification model based on SAE network and some traditional classification algorithms.The experimental results show that,based on the same data set,the classification effect of the classification model in this dissertation is more prominent.The combination of Kernel Principal Component Analysis and Stack Autoencoder network in ZPW-2000 A track circuit fault classification has high theoretical value and practical application prospects.