Research On Prediction Of Track Quality Index And Early Warning Of Potential Defect Based On Dynamic Detection Data

Track irregularity is intuitive reflection of the basic shape and and geometric dimension of track,and is a key factor related to the safety and quality of train operation.Under the periodic dynamic detection mode,real-time evaluation of track regularity state and targeted defect repairs have become the focus of railway engineering department.In order to guide scientific maintenance and repair,the spatio-temporal management research of track irregularity was carried out based on dynamic detection data.The prediction model of track quality index(TQI)was established based on the grey prediction theory and the relevance vector machine algorithm,which can provide basis for the determination of maintenance work opportunity from the perspective of "when to repair".The early warning mechanism of potential defect based on the k-means++clustering algorithm was constructed around the standard deviation management mode and the peak management mode,which further solves the engineering management problem of "where to repair",and helps carry out refined maintenance work focusing on prevention.The main work and achievements are as follows:(1)The classification and management measures of track irregularity were sorted out.Data pre-processing process centered on the amplitude change rate correction method and the cross-correlation analysis method was established for the abnormal values and the detection mileage errors in the raw waveform.By analyzing the evolution law of track irregularity,it’s confirmed that the track quality on the subgrade and bridge sections conforms to the linear evolution trend.And the deterioration rate of the irregularity on the bridge is more significant.(2)Combined with the evolution characteristics of TQI,the non-equal interval grey model was improved by sequence smoothness optimization,initial value optimization,background value optimization and sequence weight optimization.The particle swarm optimization algorithm was introduced to realize the adaptive optimization of optimization parameters.Then the non-equal interval grey optimization model was established.The verification results of section examples show that the model can capture random fluctuation terms while simulating the real-time development trend of TQI.The model has good prediction performance on TQI sequences with highly linear evolution law and strong randomness,and has prominent application advantages for oscillating sequences with high volatility.(3)Based on the relevance vector machine algorithm,a sample feature mapping mode was constructed,which took the preliminary prediction result of TQI as input and the measured value of TQI as output.The particle swarm optimization algorithm was used to optimize the kernel parameters of the combined kernel function.The relevance vector machine regression model for interval prediction of TQI was trained in combination with the 5-fold cross validation link.The verification results of section examples show that the model can generate the prediction interval while modifying the prediction results of points.By considering the boundary of interval,it can adapt to the random fluctuations in TQI sequence and further improve the reliability of prediction results.(4)Based on the k-means++ clustering algorithm,the early warning mechanism of potential defect was established,which took the defect identification features as the clustering samples,took the optimal number of cluster as the identification object,and took the elbow method and KL-index method as the comprehensive judgment method.The defect identification features were defined around the standard deviation management mode and the peak management mode.The application results on simulated defects and section examples show that the the established defect early warning mechanism can effectively identify and locate the section unit defects and local irregularity defects.And it can assist in the prevention and repair of potential defects.