Study On Data Analysis And Predict Based On Rail Wear

With the passenger and freight traffic increased and train run speed improved greatly, therail wear on a line became more and more serious. Orbital check-out car collected manyspecies of rail wear data, but a single type of rail wear data had large amount of data, normalwear data and transfinite wear data mixed together, lacked the corresponding method extractthe transfinite wear data and use it effective.Separated the transfinite data by use the effective method, according to the transfinitedata took prediction of rail wear. Not only could be able to grasp the trends of rail accuratelyas a guide for the rail replacement or rail grinding, but also the economic losses and thesecurity risk could be both reduced.In combination with track geometry car collected inspection data and referenced relatedsections wear situation, as the gauge wear and the vertical wear were the main research objectof data analysis and forecasting, analyzed the main influence factors of rail wear, summarizedthe travel speed, axle load and some factors which were used in numeric expressed asinfluencing parameters. Used the One-Class Support Vector Machine analysis wear data,separated the transfinite data as the reference to abrasion prediction of rail wear. Used twodifferent ways to predict rail wear, with the abrasion of influence parameters establish a railwear prediction nonlinearity model, used the RBF fuzzy neural network running on the railwear prediction model, combined with the actual data proceed simulation experiments.According to the combination forecast algorithm have the higher generalization ability andforecasting accuracy, used wavelet analysis and particle swarm optimization least squaressupport vector machine combination forecast algorithm in the rail wear forecast, used waveletdecomposition technology according to the different frequency decomposed some datafrequency component, combined particle swarm optimization least squares support vectormachine predicted data component, used wavelet reconstruction technique to get the completeforecast data, combined with the actual data to make simulation prediction.The simulation results showed that one class support vector machine based on wear dataanalysis can achieve the data analysis and own a fast speed and a high precision inclassification. The combination of the simulation results of the RBF fuzzy neural network andthe combinatorial prediction algorithm showed that both of these two algorithms can achievethe rail abrasion prediction function. The method of combinatorial prediction algorithm wasnot affected by the wearing of factors and had higher prediction accuracy.