Research On Fault Prediction Method Of ZPW-2000A Track Circuit Based On Data Drive

With the development of Chinese railway career and the maturity of railway automation technology,the automatic block system based on ZPW-2000 A track circuit has become a key component of railway transportation security equipment,it is very important to ensure the safe operation of uninsulated track circuit system.Based on the analysis of the electrification characteristics during the operation of the track circuit,this thesis briefly summarizes the parameters affecting the operation of the track circuit equipment and the impact of the equipment fault on the system,which lays the foundation for the ubsequent research on the fault prediction method of ZPW-2000 A track circuit.The data-driven fault prediction technology is to analyze the input and output data of the equipment based on the historical monitoring data of the equipment,and to establish the fault prediction model combined with the mutual connection between the equipment.The data-based fault prediction technology solves the problem that the mathematical model of complex components or systems is difficult to determine.HSMM(Hidden Semi Markov Model)has high efficiency in model building and training,and is widely used in power system,engine and other research fields.Deep confidence network obtains data characteristics by reconstructing data,in this thesis,the network structure,data input and output and the training algorithm of the deep confidence network are studied in depth.Based on the deep confidence network,the track circuit equipment operation state data feature extraction system is established to process the monitoring data and historical record data,and the processed data is combined with HSMM carries out fault prediction of track circuit.The main contents of this thesis are as follows:(1)Combined with the theory of fault tree,this thesis analyzes the equipment faults that affect the operating state of ZPW-2000 A uninsulated track circuit,and equates the transmission system of track circuit as a series model from transmitter to receiver.Combining with the theory of transmission line,analyze the transmission path of rail and tuning unit,and model it,therefrom,verify the validity of the model by comparing with the existing data;(2)With the assistance of data processing technology,the historical operation data of ZPW-2000 A uninsulated track circuit is extracted and equalized,and the feature data is extracted according to the lexical features and subject features of recorded text respectively.In order to avoid the under fitting phenomenon of unbalanced data in machine learning,SMOTE algorithm is used to balance the data;(3)Taking the recorded data of the transmitter as an example,the deep confidence network is used for feature extraction and hidden semi-Markov model trainng;(4)Expand the data set by constructing random linear interpolation,use 70% of the dataset to train the DBN-HSMM,and use the remaining data to verify the model test results;(5)Combined with the rail transmission characteristics of track circuit,the simulation test model and tuning unit model of rail transmission path of track circuit are established to locate the compensation capacitor leakage.Taking compensation capacitor C4 as an example,after analyzing the influence of compensation capacitance fault on normalized shunt current,the leakage of compensation capacitor capacitance value is predicted by calculating the fault data and KL(Kullback-Leibler Divergence)distance under standard operation state as health evaluation index.