Research On Compensation Capacitor Fault Diagnosis Method Of ZPW-2000 Joint-less Track Circuit

Under the circumstance that the high-speed railway is speeding up in an all-around way,the detection,diagnosis,and localization of track circuit faults are increasingly important to ensure the safe and reliable operation of railway traffic.Compensation capacitors are important components of track circuit outdoor equipment.Due to the complex outdoor environment,missing installation,disconnection,or poor contact of lead wire often occur,resulting in faults such as "red light strip" of track circuit,mutation of train locomotive signal or failure of the receiver to work normally,which eventually leads to the reduction of train operation efficiency and even affects the operation safety.Therefore,detecting,diagnosing,and locating compensation capacitor faults is one of the key issues in track circuit fault handling,which is of great significance and value.Since the rail circuit has the form of a twoport network,a two-port network model of the rail circuit can be built,based on which a data set that can represent the state of the compensation capacitor of the rail circuit can be constructed,and the fault diagnosis of the compensation capacitor can be realized by combining deep learning algorithms.The main research work of this thesis is as follows:(1)Firstly,based on the structure principle of ZPW-2000 A type jointless track circuit,transmission line theory,and two-port theory,build a two-port network model of the track circuit.Secondly,the shunt current data of the track circuit is obtained through model simulation.Finally,the shunt current data is visualized and the function of the compensation capacitor in the signal transmission process of the track circuit and the influence of fault are analyzed.(2)A fault diagnosis method of compensation capacitor based on recurrent neural network is proposed.Specifically include:(1)Construct numerical data set: The discrete current data obtained by simulation is constructed into a function curve,and the characteristics of the shunt current curve are extracted by the method of piecewise integral,and then,the characteristic data that can represent the state of compensation capacitance in a track circuit section is obtained.(2)Fault diagnosis: The recurrent neural network is applied to the fault diagnosis task of the compensation capacitor.By comparison,the Long Short-Term Memory(LSTM)neural network model is better than the ordinary recurrent neural network(RNN)in processing the compensation capacitor fault data is better.The experimental results show that when the length parameters of a track circuit are fixed,LSTM has high accuracy in diagnosing compensation capacitor faults.(3)A fault diagnosis method of compensation capacitor based on improved GAFAlex Net is proposed.Specifically include:(1)Construct GAF data set: The discrete current data obtained by simulation is encoded from one-dimensional data to two-dimensional images by using Gramian Angular Field(GAF)coding.Each GAF image contains the state characteristics of compensation capacitors in a track section.(2)Fault diagnosis:Convolutional Neural Networks(CNN)are used to diagnose the fault of the GAF data of compensation capacitors.According to the spatial correlation of the compensation capacitor fault data and the similarity between different fault types,the Alex Net convolutional neural network combined with dilated convolution,experiments show that the improved Alex Net model has higher fault diagnosis accuracy on the compensated capacitance GAF dataset.The improved GAF-Alex Net method proposed in this thesis can still achieve fault diagnosis when the track circuit segment length parameters are different,which improves the applicability of the compensation capacitor fault diagnosis method.