A high-resolution prediction method for track irregularities

Abstract

This invention discloses a high-resolution prediction method for track irregularities, comprising: organizing track irregularity data of the same nominal mileage in chronological order to form an original data sample; performing mileage matching on the original data sample, and using the matched data as training samples; performing Gaussian process regression on the training samples to train a Gaussian process regression model; and using the trained Gaussian process regression model to predict track irregularities. This invention, by performing mileage matching on track irregularity data of the same nominal mileage to construct training samples and using a Gaussian process regression model for prediction, can achieve single-point (high-resolution) prediction of track irregularities under conditions of few samples. This method overcomes the shortcomings of existing technologies that can only perform low-resolution prediction at the unit level (approximately 200 m), and can achieve point-by-point early warning of track over-limit risks between two adjacent detections, providing a more refined decision-making basis for track maintenance.

Description

Technical Field

[0001] This invention relates to the field of track detection technology, and more specifically to a high-resolution prediction method for track irregularities. Background Technology

[0002] Existing track irregularity prediction methods are mainly based on dynamic detection data from track inspection vehicles. They establish deterioration models around the Track Quality Index (TQI) of a unit section (e.g., 200m) and focus on the overall deterioration trend within adjacent maintenance cycles. Based on the modeling principles, existing methods can be divided into three categories: mechanical theory-based, statistical, and machine learning-based.

[0003] Among these methods, mechanical theory-based approaches are mainly based on theories such as track dynamics, vehicle dynamics, and vehicle-track coupling dynamics. They predict the development of track irregularities by using field measurement data and indoor experimental simulations of the interaction between the vehicle and the track. Statistical methods primarily predict track irregularity development through the statistical analysis of historical monitoring data. In recent years, with the development of data analysis technology and the improvement of computing power, methods for predicting track irregularities based on machine learning methods such as probabilistic models, neural networks, support vector machines, grey models, and deep learning algorithms have been developed.

[0004] However, due to factors such as track foundation, rolling stock, operating conditions, and climate, theoretical modeling or statistical analysis results are difficult to apply in track maintenance. Existing machine learning methods generally calculate degradation trends using TQI (Traffic Quality Index), resulting in low resolution (approximately 200m) and prediction accuracy dependent on long-term historical time series. For track maintenance, the focus is often on the risk of exceeding limits at specific locations between two track inspections and whether intervention is necessary. Therefore, there is an urgent need to develop a high-resolution track irregularity prediction method capable of single-point prediction under limited sample conditions. Summary of the Invention

[0005] Therefore, it is necessary to provide a high-resolution prediction method for track irregularities to address the problems of low prediction resolution and large training sample size in existing technologies.

[0006] A high-resolution prediction method for track irregularities includes: Organize the track irregularity data of the same nominal mileage in chronological order to form the original data sample; The original data samples are matched for mileage, and the matched data is used as training samples. Gaussian process regression is performed on the training samples to train a Gaussian process regression model; Predict track irregularities using a trained Gaussian process regression model.

[0007] Furthermore, the track irregularity data with the same nominal mileage refers to track irregularity data with the same line name, class, and start and end mileage of the sample. The term "organized in chronological order" refers to arranging the data samples with irregular tracks in an orderly manner according to the order of collection time.

[0008] Furthermore, mileage matching is performed on the original data samples, specifically including: Mileage alignment is performed based on the correlation or similarity between the original data samples, and a mapping table is established between different original data samples.

[0009] Furthermore, Gaussian process regression is performed on the training samples to train a Gaussian process regression model, specifically including: Select a kernel function, wherein the kernel function includes at least one of the squared exponential kernel, the Matern kernel, and the rational quadratic kernel; Gaussian process regression is performed on the training samples to determine the hyperparameters of the model, thereby training a Gaussian process regression model.

[0010] Furthermore, the trained Gaussian process regression model is used to predict track irregularities, specifically including: Determine the future time point to be predicted and use the future time point as the input time coordinate; Substitute the input time coordinates into the trained Gaussian process regression model to calculate the mean of the predicted track irregularities and its confidence interval corresponding to the input time coordinates. The predicted mean is used as an estimate of the track irregularity at that future time point for early warning of track over-limit risks.

[0011] Furthermore, the track irregularity data is collected by a track inspection instrument or a track inspection vehicle.

[0012] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs training samples by matching track irregularity data with the same nominal mileage, and then uses a Gaussian process regression model for prediction. This enables single-point (high-resolution) prediction of track irregularity conditions even with a small sample size. This method overcomes the limitations of existing technologies, which can only perform low-resolution predictions at the unit level (approximately 200m). It allows for point-by-point early warning of track over-limit risks between adjacent detections, providing more refined decision-making support for track maintenance. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the high-resolution prediction method for track irregularities provided in an embodiment of the present invention.

[0014] Figure 2This is a schematic diagram of the original data sample in an embodiment of the present invention.

[0015] Figure 3 This is a schematic diagram illustrating the unevenness of locations with potential for exceeding limits in an embodiment of the present invention.

[0016] Figure 4 This is a schematic diagram illustrating the prediction of track irregularities using a Gaussian process regression model in an embodiment of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0018] Please see Figure 1 This invention provides a high-resolution prediction method for track irregularities. This method, applied to track maintenance, enables point-by-point early warning of track over-limit risks between adjacent inspections, providing more refined decision-making support for track maintenance. The high-resolution prediction method for track irregularities includes steps S11-S14.

[0019] Step S11: Organize the track irregularity data of the same nominal mileage in chronological order to form the original data sample.

[0020] To comprehensively perceive and accurately evaluate the track irregularities of high-speed railways, my country has established a track inspection system that combines dynamic and static inspections, and periodically collects track irregularity data according to requirements such as the "High-Speed ​​Railway Line Maintenance Rules" (TG / GW115-2023). Track irregularity data consists of track geometric dimension observation data recorded in mileage order, and the recorded uncorrected mileage is the nominal mileage (also known as the marked mileage). The track maintenance section uses the track irregularity data for peak and balance management, arranging temporary repairs or comprehensive maintenance for locations exceeding peak limits or TQI (Traffic Quality Index) limits. However, track condition deterioration is a process. To prevent peak limits from exceeding limits before the next inspection, multiple track irregularity data samples with the same line name, class, and origin and end mileage (i.e., the same nominal mileage) should be arranged in order of collection time for comparative analysis to achieve reliable early warning.

[0021] Please see Figure 2This is a schematic diagram of the original data samples in an embodiment of the present invention. The data source is a routine inspection of a high-speed railway line going south. The nominal mileage of each sample is K443+800~K444+300, and the collection times are January 15, February 6, March 9, April 8, May 8, June 8 and July 7, 2021, respectively.

[0022] Step S12: Perform mileage matching on the original data sample and use the matched data as training samples.

[0023] Because different observations often have different mileage benchmarks, mileage wheel errors and wear, and wheel-rail slippage, different samples with the same nominal mileage may have certain mileage errors. Mileage errors will lead to prediction failure. Therefore, currently, the TQI of a unit segment (e.g., 200 m) is often used as input to build prediction models, which means that only the track unit's uniformity can be predicted. To predict future track irregularity peak states from raw data samples, mileage alignment should first be performed and a mapping table between different raw data samples should be established. Mileage alignment can be implemented based on the correlation or similarity between raw data samples. Specifically, time-delay cross-correlation or dynamic time warping can be used for mileage alignment.

[0024] Please refer to Table 1, which shows the training samples and mapping table in this embodiment of the invention.

[0025] Table 1

[0026] Preferably, this embodiment selects the dynamic time warping method. Dynamic time warping is an elastic measure of sequence similarity, capable of handling local mileage scaling, robust to nonlinear distortions, and handling sequences of unequal lengths, thus achieving higher accuracy in complex mileage drift scenarios. Specifically, the right elevation and elevation data of a certain high-speed railway's southbound line from K443+800 to K444+300 on January 15, February 6, March 9, April 8, May 8, and June 8, 2021, are aligned to obtain... D train ={( t 1, y 1), ( t 2, y 2), ( t 3, y 3), ( t 4, y 4), ( t 5, y 5), ( t 6, y 6)}, where t i and y iThe corresponding time coordinates for each measurement cycle and the uneven right elevation between K443+860 and K443+900 for each measurement cycle are shown. i For date indexing, i =1,2,3,4,5,6, which correspond to the six dates mentioned above.

[0027] Please see Figure 3 This is a schematic diagram of the unevenness at a location with a risk of exceeding limits in an embodiment of the present invention. It can be seen that the right elevation difference at the nominal mileage K443+887.88 measured on June 8, 2021 was -1.97mm, which is close to the acceptance management value of ±2mm for unevenness. Before the inspection on July 7, there was a risk of exceeding the limits.

[0028] Step S13: Perform Gaussian process regression on the training samples to train a Gaussian process regression model.

[0029] Gaussian Process Regression (GPR) is a nonparametric regression method based on Bayesian theory. It first establishes the model's prior function in the form of a probability distribution. Then, within the Bayesian framework, it transforms the prior function into a posterior function and can deduce the hyperparameters of the kernel function. Compared to prediction models such as LSTM, GM(1,1), CNN, and ARIMA, GPR has advantages such as strong small-sample capability, clear physical meaning of parameters, and the ability to provide confidence intervals in addition to outputting predicted values. It is particularly suitable for modeling and predicting track irregularities with small samples, nonlinearity, and high noise.

[0030] Specifically, assuming t i and y i Relationship It is a Gaussian process, that is:

[0031] in, It is a mean function. Let be the covariance function.

[0032] but ; In the formula, For independent Gaussian white noise, .

[0033] Preferably, to account for both the trend of orbital irregularities and local fluctuations, the Matern 5 / 2 kernel is selected to construct the prediction model. The training samples from step S12 are then used... D train Input a Gaussian process regression model and determine the model's hyperparameters. and Thus, a prediction model is obtained. That is, a well-trained Gaussian process regression model.

[0034] Step S14: Use the trained Gaussian process regression model to predict the track irregularity state.

[0035] Specifically, the future time point to be predicted is determined to be July 7, 2021, and this future time point is used as the input time coordinate. t pred Substitute the input time coordinates into the trained Gaussian process regression model. Calculate the mean predicted track irregularity corresponding to the input time coordinate. y pred and its 95% confidence interval y int The predicted mean y pred As an estimate of the track irregularity on July 7, 2021, it is used for early warning of track over-limit risks.

[0036] Please see Figure 4 This is a schematic diagram illustrating the prediction of track irregularities using a Gaussian process regression model in an embodiment of the present invention. The Gaussian process regression model predicts the mean value of the right-side elevation irregularity at nominal mileage K443+887.88 measured on June 8, 2021, up to July 7. y pred =-1.74mm, 95% confidence interval y int =[-2.43 mm, -1.04 mm]. Based on probability, the elevation will not exceed the operational acceptance management value before the next inspection. On July 7, 2021, the right elevation difference at this location was measured to be -1.71 mm, confirming the model's prediction.

[0037] In summary, the high-resolution prediction method for track irregularities according to the above embodiments has the following beneficial effects: This invention constructs training samples by matching track irregularity data with the same nominal mileage, and then uses a Gaussian process regression model for prediction. This enables single-point (high-resolution) prediction of track irregularity conditions even with a small sample size. This method overcomes the limitations of existing technologies, which can only perform low-resolution predictions at the unit level (approximately 200m). It allows for point-by-point early warning of track over-limit risks between adjacent detections, providing more refined decision-making support for track maintenance.

[0038] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A high-resolution prediction method for track irregularities, characterized in that, Includes the following steps: Organize the track irregularity data of the same nominal mileage in chronological order to form the original data sample; The original data samples are matched for mileage, and the matched data is used as training samples. Gaussian process regression is performed on the training samples to train a Gaussian process regression model; Predict track irregularities using a trained Gaussian process regression model.

2. The high-resolution prediction method for track irregularities according to claim 1, characterized in that, The track irregularity data with the same nominal mileage refers to track irregularity data with the same line name, class, and start and end mileage of the sample. The term "organized in chronological order" refers to arranging the data samples with irregular tracks in an orderly manner according to the order of collection time.

3. The high-resolution prediction method for track irregularities according to claim 1, characterized in that, The mileage matching of the original data sample specifically includes: Mileage alignment is performed based on the correlation or similarity between the original data samples, and a mapping table is established between different original data samples.

4. The high-resolution prediction method for track irregularities according to claim 1, characterized in that, Gaussian process regression is performed on the training samples to train a Gaussian process regression model, specifically including: Select a kernel function, wherein the kernel function includes at least one of the squared exponential kernel, the Matern kernel, and the rational quadratic kernel; Gaussian process regression is performed on the training samples to determine the hyperparameters of the model, thereby training a Gaussian process regression model.

5. The high-resolution prediction method for track irregularities according to claim 1, characterized in that, Predicting orbital irregularities using a trained Gaussian process regression model, specifically including: Determine the future time point to be predicted and use the future time point as the input time coordinate; Substitute the input time coordinates into the trained Gaussian process regression model to calculate the mean of the predicted track irregularities and its confidence interval corresponding to the input time coordinates. The predicted mean is used as an estimate of the track irregularity at that future time point for early warning of track over-limit risks.

6. The high-resolution prediction method for track irregularities according to claim 1, characterized in that, The track irregularity data is collected by a track inspection instrument or a track inspection vehicle.

Images