A railway track irregularity potential disease identification method based on feature clustering

Abstract

The present application relates to railway track irregularity disease identification technical field, especially in kind based on feature clustering's railway track irregularity potential disease identification method, including the following steps: S1: the track irregularity history dynamic detection data of target section is integrated;S2: the track irregularity potential disease identification index system is established;S3: the disease identification index value is extracted from the history dynamic detection data;S4: the disease identification index value is standardized, and the sample data set to be clustered is obtained;S5: based on k-means++ algorithm, the KL clustering index of sample data set is calculated;S6: determine whether the target section exists disease, if not, prompt no disease, if exists, then execute S7;S7: the sample data set is clustered and divided into two categories, the position information of abnormal sample is extracted, and the potential disease positioning is marked and completed.The present application integrates potential disease identification index and clustering algorithm, realizes the automatic positioning of potential disease, and can guide to carry out line preventive repair operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of railway track irregularity disease identification, in particular to a railway track irregularity potential disease identification method based on feature clustering. BACKGROUND

[0002] Track irregularity is an intuitive reflection of track geometry and spatial position, has a significant impact on wheel-rail dynamic response and vehicle dynamic behavior, and is a key element for ensuring train operation safety and comfort. In order to ensure the state of railway line equipment, the railway maintenance department improves track quality and enhances the level of line smoothness through periodic dynamic detection and maintenance and repair operations. Under the principle of "prevention first" of line maintenance, identifying and positioning the potential diseases of track irregularity conforms to the maintenance concept of preventive repair and precise repair, and is of great significance to optimizing the allocation of maintenance resources, improving the level of maintenance technology, and ensuring the service state of the line.

[0003] In the existing method, the early warning of track irregularity diseases is mainly based on signal analysis and feature extraction, and focuses on identifying irregularity diseases that have exceeded the limit, which cannot realize the advanced control of potential safety risks. SUMMARY

[0004] The present application provides a railway track irregularity potential disease identification method based on feature clustering, which extracts disease identification indicators to mine hidden information in dynamic detection data, and identifies track irregularity potential diseases with the help of feature clustering, realizes the automatic processing of the disease identification and positioning process, overcomes the lag of the existing method in disease identification, and can be used to guide the line preventive repair operation for potential diseases.

[0005] To achieve the above purpose, the technical solution adopted by the present application is:

[0006] A railway track irregularity potential disease identification method based on feature clustering, comprising the following steps:

[0007] S1: obtaining track irregularity historical dynamic detection data of a target section;

[0008] S2: establishing a track irregularity potential disease identification index system, which includes absolute amplitude mean, time domain amplitude standard deviation and average sliding standard deviation;

[0009] S3: extracting disease identification index values from the historical dynamic detection data;

[0010] S4: standardizing the disease identification index values to obtain a sample data set to be clustered;

[0011] S5: calculating the KL clustering index of the sample data set based on the k-means++ algorithm;

[0012] S6: determining whether the target section has potential disease, prompting no potential disease if not, and performing S7 if yes;

[0013] S7: dividing the sample data clustering into two categories, extracting the position information of abnormal samples, marking and completing the potential disease positioning.

[0014] Further, in step S5, according to formula (1) and k The values 1 and 2 are respectively calculated as the KL clustering index:

[0015] ;

[0016] Wherein, d is the data dimension of the feature sample; SSE k-1 and SSE k are the within-group sum of squares of error corresponding to the clustering category number taking k- 1 and k ; DIEF k and DIEF k+1 are the within-group distance change values corresponding to the clustering category number taking k- 1 and k ; KL k is the KL clustering index when the clustering category number is k ;

[0017] Further, in step S6, whether the target section has potential disease is determined according to the following method: if the KL clustering index of the sample data set is KL 2> KL 1, the target section has potential disease, otherwise the target section has no potential disease, wherein, KL 2 is the KL clustering index when the clustering category number is 2, KL 1 is the KL clustering index when the clustering category number is 1.

[0018] Further, in step S1, the track irregularity history dynamic detection data is obtained by a track detection vehicle.

[0019] Further, in step S3, the potential disease identification index value is extracted from the history dynamic detection data according to formula (2):

[0020] ;

[0021] Wherein, n is the total number of detection time points to which the history dynamic detection data belongs; y tp ist Time p Original amplitude of track irregularity at mileage position; is p Average of original amplitude of track irregularity at mileage position; is p Absolute value of average of original amplitude of track irregularity at mileage position; σ p is p Standard deviation of time domain amplitude at mileage position; S tp is t Sliding standard deviation with p Mileage position as the center of sliding window at time; S p is p Average of sliding standard deviation at mileage position.

[0022] Further, the width of the sliding window is 20m, and the sliding step is 0.25m.

[0023] Further, in step S4, the potential disease identification index value is standardized according to formula (3), and the standardized data is taken as the sample feature value:

[0024] ;

[0025] In the formula, x jf is the sample value of the disease identification index value f ; j is the sample value of the disease identification index value after standardization; f is the maximum value of all sample values of the disease identification index value j ; is the minimum value of all sample values of the disease identification index value f . f

[0026] Further, in the two-class sample data set clustering divided in the step S7, the relatively less one is taken as the abnormal sample, and the position information corresponding to the abnormal sample is the corresponding mileage of the sampling point. The positioning result is distinguished and marked on the mileage graph.

[0027] Compared with the prior art, the railway track irregularity potential disease identification method based on feature clustering has the following beneficial effects:

[0028] ​​1. The application provides a railway track irregularity potential disease identification method based on feature clustering, which establishes a track irregularity potential disease identification index system, extracts the identification index value of the track irregularity original data as the object, mines the key evolution information while compressing the historical dynamic detection data volume, realizes the effective extraction and representation of the track irregularity attribute, and enhances the distinguishability of the potential disease.

[0029] 2. The method provided by the application adopts the k-means++ algorithm to complete the clustering process of the samples, uses the KL clustering index as the standard to determine whether there is a potential disease, has strong operability and easy programmability, relies on the unsupervised clustering process to automatically divide the sample data set into two categories, extracts the position information corresponding to the abnormal samples to complete the identification and positioning of the potential disease, and realizes the rapid and automatic identification of the track irregularity disease. BRIEF DESCRIPTION OF DRAWINGS

[0030] The application has the following drawings:

[0031] Figure 1 is a flowchart of the railway track irregularity potential disease identification method based on feature clustering of the application;

[0032] Figure 2 is a framework diagram of the track irregularity potential disease identification index system based on feature clustering of the application;

[0033] Figure 3 is a track irregularity dynamic detection result diagram of a section of ballast track obtained by a track inspection vehicle in an embodiment of the application;

[0034] Figure 4 is a potential disease identification and positioning result diagram of track irregularity in an embodiment of the application; KL

[0035] Figure 5 is a potential disease identification and positioning result diagram of track irregularity in an embodiment of the application. DETAILED DESCRIPTION

[0036] The application will be further described in detail below with reference to the drawings.

[0037] Overall, as shown in Figure 1 The application discloses a railway track irregularity potential disease identification method based on feature clustering, which comprises the following steps:

[0038] S1: integrating track irregularity historical dynamic detection data of a target section;

[0039] S2: establishing a track irregularity potential disease identification index system, which comprises absolute amplitude mean value, time domain amplitude standard deviation and average sliding standard deviation; ​

[0040] S3: Extract disease identification index values ​​from historical dynamic monitoring data;

[0041] S4: Standardize the disease identification index values ​​to obtain the sample dataset to be clustered;

[0042] S5: Calculate the KL clustering index of the sample dataset based on the k-means++ algorithm;

[0043] S6: Determine whether there are potential defects in the target section. If there are no potential defects, indicate that there are no potential defects. If there are potential defects, proceed to S7.

[0044] S7: Cluster the sample dataset into two classes, extract the location information of abnormal samples, label and complete the location of potential diseases.

[0045] Specifically, unevenness within a 500m continuous measurement section of a certain ballasted high-speed railway was used as a potential defect target identification section.

[0046] According to step S1, the historical dynamic detection data of track irregularities measured by the track inspection vehicle is extracted as the basic data, and the historical dynamic detection results of elevation and sag irregularities are as follows: Figure 3 As shown in the figure. The horizontal axis represents the measurement mileage at a sampling frequency of 0.25m, and the vertical axis represents the elevation variations at different measurement mileage locations over different periods. From... Figure 3 It can be seen that in the historical dynamic detection results of unevenness amplitude, there are two sections with relatively obvious fluctuations, with unevenness amplitudes of 3mm and 3.3mm respectively. According to the "Maintenance Rules for Ballasted Track Lines of High-Speed ​​Railways (Trial)" (TG / GW 116-2013), the above values ​​have not yet reached the alarm threshold (5mm) for unevenness deviation, and are considered potential defects.

[0047] Specifically, such as Figure 2 As shown, a potential track irregularity identification index system is established according to step S2, including the absolute amplitude mean, time-domain amplitude standard deviation, and average moving standard deviation.

[0048] According to step S3, potential disease identification index values ​​are extracted from historical dynamic monitoring data. The calculation method for potential disease identification indicators is as follows:

[0049] ;

[0050] in, n This represents the total number of detection times to which the historical dynamic detection data belongs; y tp for t time p The original amplitude of track irregularities at the mileage location; for pThe average value of the original amplitude of track irregularities at the mileage location; for p The absolute value of the original average amplitude of track irregularities at the mileage location; σ p for p Standard deviation of time-domain amplitude at mileage location; S tp for t At present p The mileage position is the sliding standard deviation of the center of the sliding window, the sliding window length is 20m, and the sliding step size is 0.25m; S p for p The average of the sliding standard deviations at mileage locations.

[0051] According to step S4, the potential disease identification index values ​​are standardized, and the calculation formula is as follows:

[0052] ;

[0053] in, x jf Disease identification index values f Below j Sample values; Standardized disease identification index values f Below j Sample values; Disease identification index values f The maximum value among all sample values; Disease identification index values f The minimum value among all sample values.

[0054] It should be noted that the three-dimensional features corresponding to each measurement mileage are obtained by calculating the absolute amplitude mean, time-domain amplitude standard deviation, and average sliding standard deviation based on historical uneven data and then standardizing them respectively.

[0055] According to step S5, consider respectively k =1 and k In cases where the value is 2, the k-means++ algorithm is used to cluster the sample dataset of potential diseases, and the corresponding values ​​are calculated for each case. KL Clustering index, calculated as follows:

[0056] ;

[0057] in, d The data dimension of the feature samples; SSE k-1 and SSE krespectively, the number of clustering categories k- 1 and k corresponding to the within-group sum of squares of errors; DIEF k and DIEF k+1 respectively, the number of clustering categories k- 1 and k corresponding to the within-group distance change value; KL k corresponding to the KL clustering index when the number of clustering categories is k .

[0058] According to step S6, the clustering index of the target section sample data set is KL . KL 2> KL 1 as a unified discrimination condition, and when the condition is met, it indicates that the target section has potential diseases.

[0059] The KL clustering index calculation result for the potential diseases is shown in Figure 4 . As can be seen from Figure 4 , the KL clustering index of the target section sample data set is KL 1=2.50201, KL 2=23.41742, and since KL 2> KL 1, it indicates that the target section has potential diseases.

[0060] According to step S7, the sample data set clustering is divided into two categories, and the relatively smaller one of the two categories of data samples is taken as an abnormal sample, the position information of the extracted abnormal sample is marked, and the potential disease positioning is completed. The corresponding position information is the specific mileage of the sampling point, and the positioning result is distinguished and marked as a "disease section" and a "normal section" on the mileage diagram.

[0061] The identification and positioning result of the potential diseases are shown in Figure 5 . As can be seen from Figure 5 , two potential diseases are identified in the 500m section range, and the corresponding relative measurement mileages are 140~160m and 220~240m, respectively.

[0062] According to the identification result in the embodiment, the railway track irregularity potential disease identification method based on feature clustering provided by the application can complete the identification and positioning of potential diseases. In combination with Figures 3-5 , in particular, the railway track irregularity potential disease identification method based on feature clustering provided by the application can be applied to identify track irregularity potential diseases that have not yet reached the management limit value, thereby guiding the accurate repair operation under the preventive repair strategy and controlling potential safety hazards.

[0063] That which is not described in detail in the specification is considered to be of prior art by those skilled in the art.

Claims

1. A method for identifying potential defects in railway track irregularities based on feature clustering, characterized in that, Includes the following steps: S1: Obtain historical dynamic detection data of track irregularities in the target section; S2: Establish an index system for identifying potential track irregularities. This index system includes the mean absolute amplitude, the standard deviation of the time-domain amplitude, and the average moving standard deviation. S3: Extract disease identification index values ​​from historical dynamic monitoring data; S4: Standardize the disease identification index values ​​to obtain the sample dataset to be clustered; S5: Calculate the KL clustering index of the sample dataset based on the k-means++ algorithm; S6: Determine whether there are potential defects in the target section. If there are no potential defects, indicate that there are no potential defects. If there are potential defects, proceed to S7. S7: Cluster the sample dataset into two classes, extract the location information of abnormal samples, label and complete the location of potential diseases; In step S5, according to equation (1) and k Calculate the KL clustering index using values ​​1 and 2 respectively: ； in, d The data dimension of the feature samples; SSE k-1 and SSE k Take the number of cluster categories respectively k- 1 and k The sum of squared errors within the group corresponding to the time; DIEF k and DIEF k+1 Take the number of cluster categories respectively k- 1 and k The corresponding change in intra-group distance at that time; KL k For the number of cluster categories k KL clustering index at that time; In step S6, the presence of potential defects in the target section is determined using the following method: if the KL clustering index of the sample dataset... KL 2> KL If 1, then the target section has potential defects; otherwise, the target section has no potential defects. KL 2 represents the KL clustering index when the number of clusters is 2. KL 1 is the KL clustering index when the number of clusters is 1.

2. The method for identifying potential defects in railway track irregularities based on feature clustering as described in claim 1, characterized in that: In step S1, the historical dynamic detection data of track irregularities is obtained by the track inspection vehicle.

3. The method for identifying potential defects in railway track irregularities based on feature clustering as described in claim 1, characterized in that: In step S3, potential disease identification index values ​​are extracted from historical dynamic detection data according to formula (2): ； in, n This represents the total number of detection times to which the historical dynamic detection data belongs; y tp for t time p The original amplitude of track irregularities at the mileage location; for p The average value of the original amplitude of track irregularities at the mileage location; for p The absolute value of the original average amplitude of track irregularities at the mileage location; σ p for p Standard deviation of time-domain amplitude at mileage location; S tp for t At present p The mileage position is the sliding standard deviation of the center of the sliding window; S p for p The average of the sliding standard deviations at mileage locations.

4. The method for identifying potential defects in railway track irregularities based on feature clustering as described in claim 3, characterized in that: The width of the sliding window is 20m, and the sliding step length is 0.25m.

5. The method for identifying potential defects in railway track irregularities based on feature clustering as described in claim 1, characterized in that: In step S4, the potential disease identification index values ​​are standardized according to equation (3), and the standardized data are used as sample feature values: ； In the formula, x jf Disease identification index values f Below j Sample values; Standardized disease identification index values f Below j Sample values; Disease identification index values f The maximum value among all sample values; Disease identification index values f The minimum value among all sample values.

6. The method for identifying potential defects in railway track irregularities based on feature clustering as described in claim 1, characterized in that: In the two types of sample datasets clustered in step S7, the relatively smaller type of data samples is designated as the abnormal samples. The location information corresponding to the abnormal samples is the mileage of the sampling points, and the positioning results are distinguished and marked on the mileage map.

Images