Fault Diagnosis Method Of Speed Measurement And Location System Of Metro Train Based On Time Series

Speed measurement and location system of metro train is an important part of the train control system.It undertakes the task of obtaining the train speed and position information,and guarantees safety and efficiency of train operation.It is also a high-risk area.At present,the research on the fault diagnosis method for the speed measurement and location system of metro train is mainly focused on the text information,and there are few studies on the train driving log that stores on-board equipment device information by time period.This leads to the current fault analysis of the train driving log still relying on expert experience and manual processing,which increases the workload of maintenance personnel.This thesis analyzes the fault train driving log of the speed measurement and location system.The abnormal data caused by faults and stored in the train driving log is called the fault time series.The fault time series are mined from the train driving log.Then the fault diagnosis model based on LSTM neural network is built to extract the spatio-temporal features of the fault time series for multi-dimensional time series classification,to realize the fault diagnosis of the speed measurement and location system and to reduce the maintenance cost of the subway site.The research contents of this thesis are listed as follows:(1)The faults related to the speed measurement and location system and the time series performance of the abnormal data items in the train driving log are analyzed.Combined with the analysis method of the train driving log by the maintenance personnel,this thesis expounds the idea of fault diagnosis based on the fault information contained in the fault time series.For the problem that the data items related to the faults in the driving log are redundant and useless,association rule analysis based on Apriori algorithm is used to reduce the data items of the driving log.The adaptive minimum support is used to avoid the influence of unbalanced on-site faults distribution on the data items reduction.(2)For the problem of missing on-site fault logs,the speed measurement and location system and train position uncertainty are modeled,including train state model,speed measurement and error model of OPG sensor,speed measurement and error model of Doppler speed radar and train position uncertainty calculation model.This thesis also models the common faults of the speed measurement and location system under different train driving conditions,generates the train driving simulation log of faults,and obtains the fault time series as the data support for the fault diagnosis model training.The T-test and K-S test are used to verify that the overall mean and overall distribution of the simulation log and the on-site log are not significantly different.(3)The fault diagnosis model based on LSTM neural network is constructed,and the training speed and generalization ability of the model are compared by using three training optimization algorithms,Adagrad,Rmsprop and Adam,and different hyperparameters.By comparing the model with FCNN and RNN fault diagnosis models,the validity of the model in dealing with fault time series is verified.The accuracy and availability of the model for fault diagnosis in the speed measurement and location system are verified by the on-site train driving log of faults and the test log of the faulty equipment.(4)For the actual needs of on-site maintenance personnel,the multi-thread-based offline fault diagnosis software for speed measurement equipment is developed.This software can realize offline fault recurrence,fault diagnosis and fault verification of the speed measurement equipment,improve the work efficiency of maintenance personnel and reduce the maintenance cost.