Forecast And Application Of Railway Track Irregularity State

With the implementation of "One Belt One Road Initiative",as the artery of economy of China,the railway will become an important tool promoting the goal.How to ensure the safety of railway transportation is an essential prerequisite for railway operation and the core work of relevant railway departments.In order to meet the prerequisite,the maintenance plan of railway track should be made under the help of studying on track irregularity data and mining the development trend of track state.Focusing on the following two quality problems of track irregularity data in the first place: the outlier and the mileage drift,the absolute mean value method and the data variation calibration algorithm based on trends similarity are utilized to solve them respectively in the preprocessing stage.The result shows that the outliers can be identified and corrected accurately as well as the mileage drift can be calibrated relatively,which can contribute to providing reliable data for supporting later prediction research.Furthermore,for meeting requirements about delicacy management and macro-control of railway departments,a local irregularity prediction model and section irregularity prediction model are established during the prediction of track state.On the one side,considering two features of track irregularity data in time dimension: smallness of sample and stochastic fluctuation,the prediction model of track quality state with Regression Automata(RA)has been established based on the theory of syntactic pattern recognition.The drawback of lacking track irregularity samples in time dimension can be compensated by assembling space information of adjacent tracks.In this thesis,RA model which can handle the effect of uncertainty in track system is applied to solve the problem of random variation.On the other side,taking advantage of wavelet transform,it is used to solve the prediction problem about non-stationary time series.Firstly,a non-stationary time series of track irregularity can be decomposed into several stationary signals by means of wavelet transform.Furthermore,optimal fitting model is sought for each stationary signal.At last,the final prediction result of non-stationary time series can be received by recomposing the prediction data of all stationary signals.The experiment result shows that the two models proposed in this thesis both have higher prediction accuracy compared with GM(1,1)model.Finally,the experimental results of track state prediction models mentioned above are used to assist relevant railway departments in maintenance planning,which can handle questions like "Whether","When","Where" and "How" about repairing.A few examples areused to explain the procedure of drawing up preventive maintenance plan for normal track and the weak track respectively.