Research On Fault Diagnosis Technology Of 25Hz Phase Sensitive Track Circuit Based On Data Fusion

At present,25 Hz phase-sensitive track circuits are mainly used in stations in our country.It is necessary to ensure their stable operation to make trains run safely and efficiently.However,due to various natural weather or man-made reasons,the 25 Hz phase-sensitive track circuit will not be able to operate normally.Once a fault occurs,it will cause delays in trains or reduce driving efficiency,resulting in property losses.,so the fault diagnosis of the 25 Hz phasesensitive track circuit requires accuracy and high timeliness,but now the on-site judgment of the fault of the 25 Hz phase-sensitive track circuit mainly relies on manual detection and expert experience,which is inefficient and poor in real-time;And most of the fault diagnosis methods need fault data,and the fault data is very scarce,resulting in insufficient training of the fault diagnosis system;in addition,most of the current fault diagnosis mainly relies on an intelligent algorithm,which may lead to one-sided fault discrimination,resulting in a low recognition rate.Therefore,to deal with the issues raised in the foregoing paragraphs,this thesis goes about the following research content:(1)This thesis first introduces the equipment included in the 25 Hz phase-sensitive track circuit,and briefly describes its working principle through its equipment composition.On this basis,it leads to three working states of the track circuit,and the causes of the induced faults are summarized based on data review and on-site learning.According to the uniform transmission line theory,firstly,each device in the 25 Hz phase-sensitive track circuit is analogous to the four-port network model,and calculated its transmission matrix and transmission formula.According to these,the total four-port network is finally formed.Divide it into different blocks and relevant parameters are obtained by simulating faults through the model.Compare and verify the simulated data with the measured data.After confirming its validity,the fault simulation of the track circuit equivalent model based on the multi-port network established in the fourth chapter of this thesis is executed,and the fault data is derived from it,which lays the foundation for the later fault detection.(2)Use the sine function to improve the Whale Optimization Algorithm(WOA)optimization algorithm so that it can converge adaptively,and optimize the weights and thresholds of the Back propagation(BP)neural network with the improved WOA algorithm;and using the deep belief network(deep belief network,DBN)algorithm,the gray relational analysis method and the BP algorithm optimized by the improved WOA to construct three kinds of preliminary fault diagnosis models of 25 Hz phase-sensitive track circuit,after the The training and calculation of the DBN preliminary fault diagnosis model,the gray relational degree analysis method preliminary fault diagnosis model and the IWOA-BP preliminary fault diagnosis model respectively obtain their fault diagnosis results;and input or calculate the same fault samples for the three preliminary fault diagnosis models It is found that there are some conflicts in the output results of the three preliminary fault diagnosis models,which also shows that the single intelligent algorithm still has some limitations and one-sidedness.(3)Analyze the preliminary diagnosis results of the single intelligent algorithm in the previous part,and use the evidence theory to fuse them.Due to conflicts,the weight assignment method is selected to re-construct the probability of the initial basic probability assignment,and the conflict measurement is introduced.Conflict measurement is used to judge its degree of trust;the accuracy of the preliminary diagnosis results of a single intelligent algorithm is used to construct a weight;finally,a proportional weight is obtained by combining the weight and trust degree,and the existing conflict is eliminated through weight reconstruction.Using the newly constructed probability distribution,data fusion decision-making through evidence theory and a determination of the fault diagnosis classification is achieved.The conclusion shows that the equivalent model of the model can make up for the lack of field fault data and supplement the fault data,and the fault diagnosis results after data fusion can not only correct the wrong results of a single intelligent algorithm but also the overall the accuracy and diagnostic effect of the method are better than a single fault diagnosis method.