Research On The Fault Diagnosis Of Jointless Track Circuit Tuning Area And Compensation Capacitor Based On SPBO-SDAE

Jointless Track Circuit(JTC),as one of the important equipment in railway signal foundation,plays an important role in ensuring the safety of train running and improving the efficiency of train running.Jointless track circuit has been widely used in China’s railroad signal system.Its faults mainly include compensation capacitance fault and tuning area equipment fault.Compensation capacitor failure and tuning area equipment failure will cause signal crossing between track circuits,and even “red-light-strap”,locomotive “code-missing”and other phenomenon.Aiming at the problems of low efficiency and high cost of JTC fault diagnosis,this paper proposes a diagnosis method based on student psychology optimized stacked noise-reducing self-encoder network.Firstly,the four-terminal network model of jointless track circuit in the tuned and tapped states,respectively,is established according to the two-port theory,which can simulate the four-terminal network model of JTC normal state,compensation capacitor fault and tuned zone equipment fault,and derive the expression of the tapped current amplitude envelope.Secondly,to accurately extract the fault feature information of tuning area and compensation capacitor of ZPW-2000 A track circuit,a fault feature extraction method based on optimization variational mode decomposition of Genetic Algorithm with envelope entropy as the optimization objective was proposed.The minimum envelope entropy of modal component in the VMD method was taken as the objective function,and the optimal combination parameters were determined by searching the minimum value of the objective function through genetic algorithm.The original fault signal was decomposed into several intrinsic modal components by the VMD under the optimal combination to verify the effectiveness of the algorithm.Compared with EEMD and traditional VMD methods,it is concluded that GA-VMD method proposed in this paper has stronger ability to extract fault information,which avoids the mode aliasing problem of EEMD method and the over-decomposition and under-decomposition problems of traditional VMD method in the process of signal decomposition for accurate extraction of complex signals.Finally,the Stacked Denoising Auto Encoders network is introduced into JTC fault diagnosis to address the low accuracy of Auto Encoder network and Denoising Auto Encoder network for fault diagnosis of complex data samples.By setting the network hyperparameters such as the number of implied layer layers,the number of implied layer nodes,the sparsity factor,and the additive noise ratio,the optimal number of implied layers of SDAE network for JTC fault diagnosis is determined.The experiment shows that when other parameters remain unchanged,the fault accuracy rate is the highest when the number of hidden layers is3.However,in view of the problem that the setting of super parameters in SDAE network will affect the generalization ability and reconstruction accuracy of the network,the Student Psychology Based Optimization algorithm is proposed to optimize the super parameters of SDAE network.Finally,accuracy rate,recall rate,false alarm rate and F-1 value were used as evaluation indexes of the network,and SPBO-SDAE was compared with Random Forest,Support Vector Machine and SDAE accuracy.The validity of the proposed method is proved.