Turnout Fault Prediction Based On Rough Set Theory And Grey Prediction Model

Turnout is an important equipment for safe operation of trains.With the rapid development of high-speed railway in China,turnout becomes more and more important.Once a fault occurs,it will lead to serious accidents such as derailment and rear-end collision.Different from common-speed railway turnouts,high-speed railway turnouts require higher safety and reliability because of their high speed and short maintenance time.Therefore,it is of great significance to predict the faults of high-speed railway turnouts so as to achieve "on condition repair" for ensuring traffic safety and improving transportation efficiency.Because it is rough to extract the time-domain feature directly from the whole switching action current curve,which is not conducive to the subsequent classification of faults.While the switching action is naturally divided into different time zones,so the switching action could be separated into five parts: starting,unlocking,converting,locking and slow-releasing.First of all,the time-domain features of the recorded current curve by monitoring system are extracted according to different parts,and the optimal features are selected based on Fisher criterion.Because of the large difference of the selected optimal feature values,it is difficult to obtain the minimum decision table directly by using knowledge reduction method,and it is not conducive to subsequent fault prediction.Therefore,a "replacement" method is proposed to convert the specific value with large difference into the feature with only three different states.The decision table is reduced to get the minimum decision table which is based on the knowledge reduction method of Rough Set Theory.According to the minimal decision table,the minimal diagnosis rules can be obtained,and the rule base can be provided for subsequent fault prediction.Different "replacement" ranges and reduction methods are analyzed.The results show that the change of "replacement" range will affect the decision table,but the change of reduction method will not.Then,the data provided are used to calculate the required parameters.Given the initial sequence and the calculated parameters,the grey prediction model is established according to the recursive formula,and the specific prediction value is obtained by using the restored formula.However,the predicted results are specific data so that the fault types of turnouts can't be directly obtained.Therefore,the prediction model can't be used alone and the decision table is needed.After replacing the predicted value,the final predicted fault type can be obtained by comparing them with the state features in the minimum decision table.And the error between the actual value and the predicted value is calculated.This method can solve the problems of small amount of data and high accuracy requirement of data for turnout faults.And it does not require subjective experience.Finally,in order to make the prediction more objective,a method for calculating the probability of predicted fault of turnout is proposed.The accuracy of turnout fault diagnosis is fused,and the number of different fault types with different conclusions under the same conditions is added.Based on the above,the probability of the fault is finally obtained by quantitative calculation.The prediction model can predict faults in real time,and get specific fault types with decision table.It has significance in engineering application.