Research On Fault Diagnosis For Tuning Area Of Jointless Track Circuit Based On Stacked Denoising Autoencoder Network

Jointless track circuit is one of the ground equipment of railway signal system,which is used to realize the train occupancy inspection and the information transmission between the train and the ground.As an important part of jointless track circuit,the tuning area is to realize the electrical isolation of two adjacent sections.The failure of tuning area will directly affect the signal transmission in track circuit,reducing the driving efficiency and even threatening the driving safety.At present,the fault diagnosis of tuning area mainly depends on the fault detection vehicle,which is too dependent on labor and has a high cost.In order to improve the efficiency of tuning area fault diagnosis and reduce the maintenance cost,this thesis proposes a diagnosis method based on Stacked Denoising Autoencoder network.The main contents of this thesis are as follows:(1)Taking ZPW-2000 A jointless track circuit as the research object,combined with the theory of transmission lines,the theoretical models under the regulated state and shunted state are established respectively,and the model parameters are derived according to the distributed-parameter method.For the common failure modes in tuning area,the corresponding four-terminal network models are derived under shunted state,and the expression of shunt voltage function is obtained.(2)Under the shunted state,the rail surface voltage waveforms under the normal state of track circuit,different fault states of tuning area and compensation capacitor fault states are simulated respectively,and the variation law of rail surface voltage in different modes is analyzed.Empirical Mode Decomposition method,Ensemble Empirical Mode Decomposition method and Complementary Ensemble Empirical Mode Decomposition method are applied to decompose the rail surface voltage.After analysis and comparison,the decomposition effect of CEEMD method is the best,which can effectively extract the characteristics of different fault states in the tuning area.(3)According to the characteristics of fault data set in tuning area,a diagnosis model based on Stacked Denoising Autoencoder network is built.Stacked Denoising Autoencoder network uses the Autoencoder for unsupervised learning,and Back Propagation neural network is responsible for fine-tune the model,and makes the model converge by continuously adjusting the network parameters.By setting different number of hidden layer nodes,multiple hidden layer combinations are constructed.After many times of training,the best combination of hidden layer is determined to solve the problem of fault classification in tuning area,and the evaluation index is used to evaluate the model.The results show that the model based on Stacked Denoising Autoencoder network can effectively classify different fault modes in tuning area,and the diagnostic accuracy can reach over 97%.(4)To solve diagnostic accuracy problem caused by the random determination of network structure of classification model,this thesis proposes a Stacked Denoising Autoencoder based on Particle Swarm Optimization and Genetic Algorithm.The idea of crossover,mutation and elimination mechanism is introduced into Particle Swarm Optimization to realize the fusion of Particle Swarm Optimization and Genetic Algorithm,so as to avoid the algorithm falling into locally optimal solution and improve the convergence speed.Through the simulation experiment,the results verify that the optimized algorithm can quickly construct the network structure for the classification model,and the optimized diagnosis model has higher fault recognition rate than the shallow neural network,Support Vector Machine and original Stacked Denoising Autoencoder model.Figures 58,Tables 11,References 62.