Research On Time Series Prediction Model Of Track Dynamic-detected Data

In recent years,with the rapid development of machine learning and data mining,intelligent processing and analysis of time series data has been widely used in various industries.Industry time series data presents the characteristics of large data volume,high complexity and non-linearity,so it is also facing great challenges in its discrimination and prediction.The railway track dynamic-detected data is a multiple time series data collected by the inspection equipment such as the track inspection vehicle.It not only reflects the current service status of the railway infrastructure,but also contains rich state evolution information,and applies advanced machine learning and data mining techniques to the basis.In the state prediction task of the track dynamic-detected data,it provides technical support for the daily maintenance and repair of the railway engineering department,which has important theoretical significance and practical value for ensuring the safety of railway operation and improving the utilization of data resources.Based on the research of time series prediction model,this paper focuses on the task of railway engineering department's prediction of high-speed track deformation,focusing on the bridge creep prediction model and CPIII differential settlement prediction model.The main work and research results of this paper include:(1)Aiming at the classification problem of bridge data classification,a polynomial logistic bridge classification method based on track dynamic-detected height irregularity data is proposed.Firstly,the track dynamic-detected height irregularity data waveform is segmented.The segmented waveform is extracted by feature,and finally classified by polynomial logistic regression algorithm.The validity of the proposed classification and extraction of bridge data is proved by the classification verification experiment on the verification set.(2)Aiming at the prediction problem of high-speed railway bridge creeping camber,a bridge prediction model based on LSTM neural network is proposed.Firstly,the bridge data set in time dimension is constructed,and then a cross-bridge is calculated.The time-varying upper camber value is used to construct the time series data of the bridge creeping upper camber.Finally,the prediction task is realized by the LSTM neural network bridge creeping camber prediction model.The verification experiments on the 24m and 32m bridge datasets demonstrate the effectiveness of the LSTM neural network-based prediction model for predicting arch creep prediction.(3)Aiming at the differential settlement prediction of CPIII in high-speed railway,a CPIII differential settlement prediction model based on Siamese neural network is proposed.Firstly,the two-input sample of Siamese neural network is constructed.It is necessary to analyze the two dynamic data of CPIII differential settlement.As input,the two sub-networks of the Siamese neural network are used to transform the two into the feature space.Finally,the metrics in the network structure are used to realize the"distance" metric of the two,and the CPIII differential settlement prediction is realized.The effectiveness of the proposed CPIII differential settlement prediction model based on Siamese neural network is proved by experiments on the verification set.