Rail side wear prediction optimization method based on trend prediction

Abstract

The application provides a rail side wear prediction optimization method based on trend prediction, and relates to the technical field of rail transit monitoring.The application divides the sampling priority by adopting the grading threshold of curvature stress Qyl, calculates the sampling point number Cds per unit time to dynamically adjust the monitoring frequency, and solves the lag problem of traditional discrete sampling in capturing the instantaneous wear of a curve.Based on the sliding window, the time sequence wear characteristics, dynamic stress characteristics and working condition correlation characteristics are extracted, and the LSTM model with enhanced space-time attention mechanism is input, so that the predicted side wear Yc and wear acceleration Mj are output, and the dynamic risk index TC is calculated.According to the evaluation content of the dynamic risk index TC, a high-risk section is marked.Through the secondary determination of the side wear deviation PCy, the maintenance window suggestion is generated by combining multi-objective optimization, and the GIS-BIM three-dimensional heat map is used for artificial calibration and online update of the model weight, so that a monitoring-prediction-maintenance closed loop is formed, and the side wear diagnosis accuracy and maintenance efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rail transit monitoring, in particular to a rail side wear prediction optimization method based on trend prediction. BACKGROUND

[0002] With the development of urban rail transit and high-speed rail transportation, the problem of rail side wear is becoming increasingly serious, directly affecting the safety and stability of train operation. The traditional maintenance method relies on manual experience and periodic detection, which is low in efficiency and difficult to accurately predict wear trends, so it is urgent to develop an intelligent prediction optimization algorithm to improve the level of rail operation and maintenance.

[0003] In the prior art, a method and system for detecting alternating side wear of a rail are disclosed, and the three-dimensional profile of the rail profile in the abnormal shaking area of the subway is obtained, and the alternating side wear waveform at the side wear position is generated according to the three-dimensional profile of the rail. According to the alternating side wear waveform, a side wear identification model is used to determine the occurrence position of the rail alternating side wear phenomenon.

[0004] However, the above technical solution has the following two technical defects when applied to urban subways:

[0005] Since the three-dimensional profile is collected at discrete preset distance intervals, the dynamic monitoring of rail side wear, including sudden wear on curved tracks, is delayed;

[0006] And this lack of real-time further hinders the effective collaborative analysis of the side wear identification model and the real-time operation data of the train, including speed and axle load, in the complex working conditions of frequent starting and stopping and small radius curve sections. The model is inaccurate due to monitoring delays, which further causes a chain reaction of maintenance strategy deviation, ultimately affecting the accurate diagnosis of alternating side wear phenomena and maintenance efficiency.

[0007] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0008] The purpose of the present application is to provide a rail side wear prediction optimization method based on trend prediction to solve the problems raised in the background.

[0009] To achieve the above purpose, the present application provides the following technical solution:

[0010] The rail side wear prediction optimization method based on trend prediction includes the following specific steps:

[0011] S1, acquire the three-dimensional profile data of the steel rail in the target area, and simultaneously collect the real-time running data of the train in the multiple track sections to construct a dynamic correlation dataset; the track section includes a curved track and a small-radius curve section; a sampling priority is set, and the sampling frequency is automatically adjusted; the dynamic correlation dataset includes the curvature stress of each track section;

[0012] S2, based on the dynamic correlation dataset, relevant features are extracted and input into a long short-term memory network model for training to construct a rail side wear trend prediction model to output the predicted side wear Yc and the wear acceleration Mj in the future m period;

[0013] The predicted side wear Yc, wear acceleration Mj and curvature stress are combined and analyzed to construct a dynamic risk index TC, which is used to mark the risk degree of the current track section;

[0014] S3, according to the output result of the rail side wear trend prediction model, an optimization strategy is generated by establishing and combining a line maintenance rule library, including:

[0015] Triggering retesting for high-risk sections;

[0016] According to the deviation of the predicted side wear Yc and the actual monitoring data, the sensor sampling frequency is adjusted again;

[0017] Output a maintenance window suggestion coordinated with the train operation plan;

[0018] S4, visualization and feedback calibration, the predicted side wear trend, high-risk sections and optimized maintenance strategies are displayed through a visualization interface, and artificial calibration input is received to update the weight parameters of each parameter in the dynamic risk index TC in real time.

[0019] Further, based on the continuous sampling method, the real-time running data and the three-dimensional profile data of the steel rail are collected in real time by the vehicle-mounted sensor and the track sensor, and the space-time matching is performed to establish the dynamic correlation dataset;

[0020] By combining the real-time running data and the track geometric parameters, the curve radius and the curvature change gradient of each track section are analyzed and obtained, and according to the running speed and the axle load of the train, the curvature stress Qyl of each track section is calculated and obtained;

[0021] The priority grading threshold of the curvature stress is set, including Qa and Qb, and Qa>Qb; then the curvature stress Qyl is compared and evaluated with the priority grading threshold, and the sampling priority of each track section is determined, and the specific content is as follows:

[0022] When the curvature stress Qyl is greater than or equal to Qa, the current track section is determined as the first priority sampling section;

[0023] When Qb≤curvature stress Qyl<Qa, the track section is determined as a second priority sampling section;

[0024] When curvature stress Qyl<Qb, the track section is determined as a third priority sampling section.

[0025] Further, after the sampling priority of each track section is determined, the number of sampling points Cds in the current monitoring time period is calculated according to the priority of each section, and the sampling frequency of the track section is automatically adjusted;

[0026] The specific calculation formula of the number of sampling points Cds is as follows:

[0027]

[0028] In the formula, Cds represents the reference sampling point number under the standard curvature stress; Qb represents the reference curvature stress value;

[0029] The preset sampling point threshold includes Ca and Cb, and Ca>Cb;

[0030] The sampling point threshold and the number of sampling points Cds are compared and evaluated, and the specific content is as follows:

[0031] When the number of sampling points Cds>Ca, the sampling frequency of the current track section is increased;

[0032] When Cb<the number of sampling points Cds≤Ca, the sampling frequency of the current track section is maintained;

[0033] When the number of sampling points Cds≤Cb, the sampling frequency of the current track section is reduced.

[0034] Further, the data in the dynamic correlation data set is processed in a sliding window manner, and the time sequence wear characteristics, dynamic stress characteristics and working condition correlation characteristics in the dynamic correlation data set are extracted;

[0035] The time sequence wear characteristics, dynamic stress characteristics and working condition correlation characteristics are input into the long short-term memory network enhanced by the spatiotemporal attention mechanism for training. The output wear acceleration Mj and the predicted side wear Yc in the future m time periods are associated and fitted after being dimensionless processed with the curvature stress Qyl, and the dynamic risk index TC of the current track section is calculated, and the specific calculation formula is as follows:

[0036]

[0037] In the formula, a1, a2 and a3 represent the weight parameters of the wear acceleration Mj, the predicted side wear Yc and the curvature stress Qyl respectively, and a1+a2+a3=1.

[0038] 5. The rail side wear prediction and optimization method based on trend prediction according to claim 4, characterized in that: based on the statistical analysis of historical accident data and material fatigue limit, a dynamic risk threshold T is preset and compared with the dynamic risk index TC for evaluation, the specific evaluation content is as follows:

[0039] If the dynamic risk index TC is greater than or equal to the dynamic risk threshold T, the current track segment will be marked as a high-risk segment.

[0040] If the dynamic risk index TC is less than the dynamic risk threshold T, the current track segment is determined to be a safe segment, and the current monitoring mode is maintained.

[0041] Furthermore, based on historical track maintenance records, standardized maintenance specifications, and experience data on side wear evolution under different operating conditions, a track maintenance rule base is constructed.

[0042] When the current track section is marked as a high-risk section, the track maintenance rule library is combined with the track wear limit standard, safe operation requirements and historical track inspection data for comprehensive statistical analysis. The pre-set side wear warning threshold Yth is compared with the predicted side wear Yc in real time to assess whether the side wear of the current track section will exceed the standard in the next m time period.

[0043] The specific assessment content is as follows:

[0044] If the predicted side wear amount Yc is greater than or equal to the side wear amount warning threshold Yth, it is determined that the side wear amount of the current track section will exceed the standard in the next m time period, and a retest will be triggered at this time.

[0045] If the predicted side wear amount Yc is less than the side wear amount warning threshold Yth, it is determined that the side wear amount of the current track section will not exceed the standard in the next m time period, and the current detection frequency will remain unchanged.

[0046] Furthermore, after the retest is triggered, the actual profile side wear of the current track section is collected in real time within the m-period using trackside sensors. ;

[0047] Extract the predicted side wear amount Yc and the actual profile side wear amount A secondary determination is made by calculating the side wear deviation value PCy in real time. The specific calculation formula is as follows:

[0048]

[0049] In the formula, PCy∈[0,5];

[0050] When the side wear deviation value PCy=0, it indicates that the predicted side wear amount Yc of the current track section is accurate and matches the actual profile side wear amount. Completely consistent, meaning without any deviation;

[0051] When the side wear deviation value PCy > 0, it indicates that there is a certain deviation between the predicted value and the actual measured value of the current track section, and the greater the deviation value, the greater the prediction error;

[0052] The deviation threshold Pth is set, which is compared with the side wear deviation value PCy to determine whether the current track section is a significant deviation section, and the track side wear trend prediction result is obtained;

[0053] The specific content is as follows:

[0054] When the side wear deviation value PCy ≥ the deviation threshold Pth > 0, it is determined that the current track section is a significant deviation section, and is marked as an "emergency section" and is preferentially included in the high-frequency detection and maintenance plan;

[0055] When the deviation threshold Pth > the side wear deviation value PCy > 0, it is determined that the section is a deviation acceptable section, and is marked as a "regular section", and the current detection frequency is maintained.

[0056] Further, after determining that the current track section is a significant deviation section, the future available maintenance window period is read and queried based on the train operation plan data;

[0057] And combined with the track side wear trend prediction result and the operation time demand of the train operation density, the window optimization algorithm of multi-objective optimization scheduling is used for priority sorting and matching to generate and output maintenance window suggestion information, which is pushed to the maintenance scheduling platform through the wireless communication module.

[0058] Further, based on the GIS-BIM fusion engine, a three-dimensional thermal map of rail wear is generated to map the distribution of dynamic risk index TC by color gradient; train real-time positioning data and maintenance work order status are integrated to form a multi-layer collaborative visualization interface;

[0059] An interactive diagnosis panel is provided to support manual frame selection of high-risk sections and linked retrieval of multi-modal data including acoustic emission waveform and infrared thermal imaging, while providing a spatiotemporal comparison view of predicted side wear Yc and actual profile side wear Real-time historical data backtracking analysis is performed through a sliding time axis.

[0060] Further, the model misjudgment samples confirmed by manual confirmation in the rail side wear monitoring process are received, and the adversarial samples are automatically generated, and the model misjudgment samples specifically include false positive cases and false negative cases; then the model misjudgment samples are injected into the long short-term memory network for incremental retraining;

[0061] At the same time, the parameter weight in the calculation of the dynamic risk index TC is dynamically adjusted, and the specific logic is as follows:

[0062] If marked as "emergency section", increase the weight of wear acceleration Mj to quickly capture the risk of sudden wear; reduce the weight of curvature stress Qyl to focus on short-term risk response;

[0063] If marked as "regular section", increase the weight of curvature stress Qyl to strengthen the analysis of long-term cumulative effects; reduce the weight of wear acceleration Mj to reduce sensitivity to short-term fluctuations;

[0064] Online learning technology is adopted to fine-tune the model parameters of the rail side wear trend prediction model in real time and generate a feature weight drift analysis report.

[0065] Compared with the prior art, the beneficial effects of the present application are: the real-time performance of rail side wear monitoring is significantly improved through continuous sampling and dynamic priority adjustment mechanism; a dynamic correlation data set is constructed by synchronously collecting train operation data and track geometric parameters using on-board sensors and track sensors; the sampling priority section is automatically divided based on the grading threshold of curvature stress Qyl; the sampling frequency is dynamically adjusted through the calculation of the number of sampling points Cds in the current monitoring time period; this technical means effectively solves the problem of insufficient capture of instantaneous wear on curves in traditional fixed interval sampling

[0066] The present application realizes the cooperative optimization of monitoring data and operating conditions through spatio-temporal feature fusion and intelligent prediction model, specifically including: extracting time series wear features, dynamic stress features and working condition correlation features using a sliding window; outputting predicted side wear Yc and wear acceleration Mj through an LSTM model enhanced by a spatio-temporal attention mechanism; calculating a dynamic risk index TC combining the dimensionless curvature stress Qyl and evaluating it; marking high-risk sections according to the evaluation content of the dynamic risk index TC; this scheme overcomes the model misalignment problem caused by data fragmentation in traditional methods, and the prediction accuracy is significantly improved, especially under frequent start-stop and heavy load working conditions;

[0067] The present application also eliminates the chain reaction caused by monitoring delay through closed-loop feedback and adaptive maintenance strategy; mainly embodied in: establishing a secondary judgment mechanism for side wear deviation PCy; generating a maintenance window suggestion using a multi-objective optimization algorithm combined with the train operation plan; receiving manual calibration through a visual interface and updating the model weight in real time to form a "monitoring-prediction-maintenance" closed loop, ultimately improving the diagnosis accuracy and maintenance efficiency of alternating side wear. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 The present application is a whole method flowchart. DETAILED DESCRIPTION

[0069] To make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application is further described in detail below in combination with specific embodiments.

[0070] It should be noted that, unless otherwise defined, technical terms or scientific terms used in the present application shall have the common meaning understood by one of ordinary skill in the art to which the present application pertains. The terms "first", "second", and similar terms used in the present application do not indicate any order, number, or importance, but are only used to distinguish different components. The terms "include", "contain", and similar terms mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "upper", "lower", "left", "right", and the like are only used to indicate relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships can also change accordingly.

[0071] Embodiment one:

[0072] Please refer to Figure 1 The present application provides a rail side wear prediction optimization method based on trend prediction, and the specific steps include:

[0073] S1, obtain the three-dimensional profile data of the steel rails of a plurality of continuous track sections in a target area, simultaneously synchronously collect real-time running data of trains on the plurality of track sections, and construct a dynamic correlation data set; the track sections include curved tracks and small-radius curve sections; a sampling priority is set, and the sampling frequency is automatically adjusted; the dynamic correlation data set includes the curvature stress of each track section;

[0074] S2, based on the dynamic correlation data set, extract relevant features and input them into a long short-term memory network model for training, construct a rail side wear trend prediction model, and output the predicted side wear Yc and the wear acceleration Mj in the future m time periods;

[0075] The predicted side wear Yc, the wear acceleration Mj, and the curvature stress are combined and analyzed to construct a dynamic risk index TC, which is used for risk degree marking of the current track section;

[0076] S3, according to the output result of the rail side wear trend prediction model, establish and combine a line maintenance rule library to generate an optimization strategy, including:

[0077] Triggering re-measurement for high-risk sections;

[0078] According to the deviation of the predicted side wear Yc and the actual monitoring data, the sensor sampling frequency is adjusted again;

[0079] Output a maintenance window suggestion coordinated with the train operation plan;

[0080] S4, visualizing and feedback calibration, showing the predicted side wear trend, high-risk section and optimized maintenance strategy through the visual interface, and receiving manual calibration input, and updating the weight parameters of each parameter in the dynamic risk index TC in real time.

[0081] In this embodiment, the three-dimensional profile data of the steel rail and the real-time running data of the train are acquired by continuous sampling in step S1, and a dynamic correlation data set is constructed, which can realize high-frequency dynamic capture of the track wear evolution process, set sampling priority for curved tracks and small-radius curve sections, and automatically adjust the sampling frequency, effectively improving the response speed and monitoring accuracy of side wear abnormalities in high-risk sections;

[0082] Through step S2, the time series wear characteristics, dynamic stress characteristics and working condition correlation characteristics are extracted based on the dynamic correlation data set, input into the long short-term memory network model for training, and a steel rail side wear trend prediction model is constructed, which outputs the predicted side wear Yc and wear acceleration Mj in the future m time periods, and marks the high-risk section, which can early warning potential wear failure and support active maintenance decision;

[0083] Through step S3, according to the output result of the steel rail side wear trend prediction model, a line maintenance rule library is established and combined, and a maintenance window suggestion is generated for high-risk sections to trigger re-measurement, adjust the sensor sampling frequency according to the deviation value PCy of the predicted side wear Yc and the actual profile side wear, and output the maintenance window suggestion coordinated with the train operation plan, which can dynamically optimize the detection and repair resource allocation, reduce invalid maintenance and improve overall operation efficiency;

[0084] Through step S4, visualizing and feedback calibration are realized, the predicted side wear trend, high-risk section and optimized maintenance strategy are displayed, and manual calibration input is received to update the weight parameters of the side wear trend prediction model in real time, which can continuously improve the prediction accuracy and field adaptability of the model through human-machine cooperation.

[0085] Embodiment 2

[0086] Based on the continuous sampling method, the real-time running data and the three-dimensional profile data of the steel rail are collected in real time by the on-board sensor and the track sensor, and are matched in space and time to establish a dynamic correlation data set;

[0087] By combining the real-time running data and the track geometric parameters, the radius of the curved track and the curvature change gradient of each track section are analyzed and obtained, and the curvature stress Qyl of each track section is calculated according to the running speed and the axle load of the train;

[0088] Further, the curvature stress Qyl represents the bending stress of the curve section, and the specific calculation formula is as follows:

[0089]

[0090] In the formula, E represents the elastic wear resistance of the rail material; h represents the height of the rail cross section; and R represents the curve radius of the track section. This indicates the axle load of the train; v indicates the train speed.

[0091] The priority threshold for curvature stress is set based on historical statistical data and engineering experience.

[0092] A priority grading threshold is set for curvature stress, including Qa and Qb, where Qa > Qb. Then, the curvature stress Qyl is compared and evaluated against the priority grading threshold to determine the sampling priority of each track segment. The specific details are as follows:

[0093] When the curvature stress Qyl≥Qa, the current track segment is determined to be the first priority sampling segment;

[0094] When Qb≤curvature stressQyl<Qa, the track segment is determined to be the second priority sampling segment;

[0095] When the curvature stress Qyl < Qb, the track segment is determined to be a third-priority sampling segment.

[0096] After determining the sampling priority for each track segment, the number of sampling points Cds within the current monitoring time period is calculated based on the priority of each segment, and the sampling frequency of that track segment is automatically adjusted.

[0097] The specific formula for calculating the number of sampling points Cds is as follows:

[0098]

[0099] In the formula, This indicates the number of reference sampling points under standard curvature stress; Indicates the reference curvature stress value;

[0100] Furthermore, the standard for standard curvature stress is based on a comprehensive calibration of the fatigue limit of rail materials, allowable stress values ​​in track design specifications, and historical accident statistics. The formula for the number of sampling points, Cds, is derived from the curvature stress value QyI of the current section and the reference value. The ratio of the reference sampling points is dynamically adjusted. This determines the actual number of sampling points Cds to achieve frequency adaptation. The floor sign in the formula is used to round the calculated value up to the nearest integer.

[0101] The higher the curvature stress Qyl, the greater the stress in that section and the higher the risk. Therefore, the corresponding number of sampling points Cds in the current monitoring time period is also larger, and the sampling point density is automatically increased.

[0102] The lower the curvature stress Qyl is, the sampling point number Cds in the current monitoring time period is automatically reduced, and resources are saved.

[0103] Further, the sampling point number threshold is set according to historical monitoring data and engineering experience of the track.

[0104] The preset sampling point number threshold includes Ca and Cb, and Ca > Cb.

[0105] The sampling point number threshold and the sampling point number Cds are compared and evaluated, and the specific content is as follows:

[0106] When the sampling point number Cds > Ca, the sampling frequency of the current track section is increased.

[0107] When Cb < sampling point number Cds ≤ Ca, the sampling frequency of the current track section is maintained.

[0108] When the sampling point number Cds ≤ Cb, the sampling frequency of the current track section is reduced.

[0109] In this embodiment, by using the vehicle-mounted sensor and the track sensor to collect train operation data and track geometric parameters in real time and perform space-time matching, a dynamic correlation data set is constructed, which is different from the static or periodic sampling mode in the prior art, and can dynamically reflect the changes of train load and track working conditions; the radius of the curve and the curvature change gradient are calculated in combination with real-time data, and the curvature stress Qyl is calculated based on the train axle load Mv, the running speed v, the elastic modulus E of the rail material and the rail section height h, which can accurately quantify the stress level of the curve section and dynamically reflect the track fatigue state.

[0110] The curvature stress priority grading threshold Qa and Qb are set, the curvature stress Qyl is compared and evaluated with Qa and Qb to determine the sampling priority of the track section, and the differential monitoring of the high-risk curve section is realized; the sampling point number Cds in the current monitoring time period is calculated based on the reference sampling point number under the standard curvature stress and the reference curvature stress value, and the greater the curvature stress is, the higher the sampling density is, so as to timely capture the wear abnormality; the preset sampling point number threshold Ca and Cb are compared with Cds to dynamically adjust the sampling frequency, realize the reasonable allocation of resources and maximize the monitoring efficiency; this module undertakes the core functions of high-risk track section identification and adaptive sampling scheduling in the system, and provides a high-quality and high-correlation data basis for the subsequent rail side wear trend prediction model.

[0111] Embodiment 3

[0112] The data in the dynamic correlation data set is processed in a sliding window manner, and the time sequence wear feature, the dynamic stress feature and the working condition correlation feature in the dynamic correlation data set are extracted.

[0113] The time sequence wear characteristics, dynamic stress characteristics and working condition correlation characteristics are input into the long short-term memory network enhanced by the space-time attention mechanism for training. The output wear acceleration Mj and the predicted side wear Yc in the future m time periods are associated with the curvature stress Qyl, and after being dimensionless processed, the dynamic risk index TC of the current track section is obtained by calculation, and the specific calculation formula is as follows:

[0114]

[0115] In the formula, a1, a2 and a3 respectively represent the weight parameters of the wear acceleration Mj, the predicted side wear Yc and the curvature stress Qyl, and a1+a2+a3=1.

[0116] Based on the statistical analysis of historical accident data and material fatigue limit, the dynamic risk threshold T is preset, and compared with the dynamic risk index TC for evaluation, and the specific evaluation content is as follows:

[0117] If the dynamic risk index TC is greater than or equal to the dynamic risk threshold T, the current track section is marked as a high-risk section;

[0118] If the dynamic risk index TC is less than the dynamic risk threshold T, the current track section is determined as a safe section, and the current monitoring mode is maintained.

[0119] Based on the historical track maintenance records, the standardized maintenance specifications and the side wear evolution experience data under different operation conditions, a line maintenance rule library is constructed;

[0120] When the current track section is marked as a high-risk section, the line maintenance rule library is combined, the track wear limit standard, the safety operation requirement and the comprehensive statistical analysis of the track inspection data in previous years are adopted to preset the side wear warning threshold Yth, and the predicted side wear Yc is compared in real time to evaluate whether the side wear exceeds the standard in the future m time periods of the current track section;

[0121] The specific evaluation content is as follows:

[0122] If the predicted side wear Yc is greater than or equal to the side wear warning threshold Yth, it is determined that the side wear exceeds the standard in the future m time periods of the current track section, and at this time, the retest is triggered.

[0123] If the predicted side wear Yc is less than the side wear warning threshold Yth, it is determined that the side wear does not exceed the standard in the future m time periods of the current track section, and at this time, the current detection frequency is maintained unchanged.

[0124] In this embodiment, the data in the dynamic correlation data set is processed by a sliding window method, which can extract the time series wear characteristics, dynamic stress characteristics and working condition correlation characteristics, which is different from the static or simple analysis method in the prior art; the time series wear characteristics are based on the side wear rate change gradient calculated by the sliding window, which can dynamically reflect the development trend of the track wear;

[0125] The dynamic stress characteristics are obtained by calculating the curvature stress Qyl and the contact patch wear index through the wheel-rail contact mechanics model, which can accurately evaluate the track stress condition and wear risk; the working condition correlation characteristics combine the correlation coefficient of train start-stop frequency and curve passing frequency to provide more comprehensive working condition information for prediction;

[0126] These features are input into the long short-term memory network enhanced by the spatio-temporal attention mechanism for training, which can more accurately predict the rail side wear Yc and wear acceleration Mj in the future m period, and through dimensionless processing and correlation fitting with the curvature stress Qyl, the dynamic risk index TC is obtained to identify potential high-risk track sections in time;

[0127] This module effectively improves the accuracy of track section risk assessment, and through the combination of historical accident data and statistical analysis of rail material fatigue limit, it ensures the accurate coverage of critical risk state; based on this, combined with historical track maintenance records and standardized maintenance specifications, a line maintenance rule base is constructed to provide reliable basis for subsequent maintenance decision-making, optimize the maintenance strategy and improve the maintenance efficiency.

[0128] Embodiment 4

[0129] After the retest trigger, the actual profile side wear of the current track section is collected in the m period through the wayside sensor ;

[0130] The predicted side wear Yc and the actual profile side wear are extracted, and the side wear deviation value PCy is calculated in real time to make a second determination, and the specific calculation formula is as follows:

[0131]

[0132] In the formula, PCy∈[0,5];

[0133] Further, according to the Railway Line Repair Rules, the maximum allowable value of the side wear is 5 mm, and the deviation value increases with the increase of the prediction error, but in actual engineering, the deviation threshold Pth is set as the action boundary;

[0134] When the side wear deviation value PCy=0, it means that the predicted value of the predicted side wear Yc of the current track section is accurate, and the actual profile side wear is completely consistent, i.e. no deviation;

[0135] When the side wear deviation value PCy > 0, it indicates that there is a certain deviation between the predicted value and the actual measured value of the current track section, and the greater the deviation value, the greater the prediction error;

[0136] Further, if the side wear deviation value PCy < 0 occurs, there are usually the following possibilities:

[0137] 1) Formula error: If the absolute value is not taken, Yc− is directly used as the deviation value, then when the predicted value is less than the actual value, the deviation will be less than 0;

[0138] 2) Data processing error: Abnormalities occur in the process of extracting the predicted value or the actual measured value, such as data type, unit inconsistency, or negative value error;

[0139] 3) Abnormal working conditions: In extreme special cases, such as sensor failure or model abnormal output, the input data itself is unreasonable, which may also calculate a negative deviation;

[0140] 4) Deviation not defined by standard: If the deviation is defined as a direct difference rather than an absolute difference in some early algorithms or non-standard implementations, theoretically there will be positive and negative fluctuations.

[0141] At the same time, the setting of the deviation threshold Pth is based on historical detection data, engineering safety tolerance, and maintenance requirements;

[0142] The deviation threshold Pth is set to 5 > deviation threshold Pth > 3, and compared with the side wear deviation value PCy to determine whether the current track section is a significant deviation section;

[0143] The specific content is as follows:

[0144] When the side wear deviation value PCy ≥ deviation threshold Pth > 0, it is determined that the current track section is a significant deviation section, and is marked as an "emergency section", which is preferentially included in the high-frequency detection and maintenance plan;

[0145] When the deviation threshold Pth > side wear deviation value PCy > 0, it is determined that the section is a deviation acceptable section, and is marked as a "regular section", keeping the current detection frequency unchanged.

[0146] Further, for the significant deviation section, the sampling frequency is adjusted again through secondary calculation and according to the number of sampling points Cds in the current monitoring time period.

[0147] After determining that the current track section is a significant deviation section, read and query the future available maintenance window period based on the train operation plan data;

[0148] In combination with the track side wear trend prediction result and the operation density of the train, a multi-objective optimization scheduling is used to prioritize and match the window optimization algorithm to generate and output the maintenance window suggestion information, which is pushed to the maintenance scheduling platform through the wireless communication module.

[0149] In this embodiment, the actual profile side wear of the current track section is collected in real time after the retest is triggered , which can be compared with the predicted side wear Yc, and the prediction error can be accurately determined by calculating the side wear deviation PCy, which is different from the static evaluation method in the prior art;

[0150] The calculation of the side wear deviation PCy can identify the deviation between the prediction and the actual measurement, and when the deviation is positive, it means that the prediction has an error, and when the deviation is zero, it means that the prediction is accurate.

[0151] By setting the deviation threshold Pth based on historical detection data, engineering safety tolerance and maintenance requirements, the significant deviation section can be effectively identified and prioritized for high-frequency detection and maintenance planning, improving the accuracy and priority of detection.

[0152] If the deviation value is within the acceptable range, the current detection frequency is maintained, avoiding unnecessary resource waste; further combined with the train operation plan and the track side wear trend prediction result, the maintenance window is optimized through the multi-objective optimization scheduling algorithm, realizing the fine management of the maintenance plan and improving the track maintenance efficiency; this module makes the track maintenance more dynamic and personalized, and can make timely adjustments according to real-time data and historical analysis, reducing the impact of prediction error on maintenance decision-making, and making resource allocation more reasonable.

[0153] Embodiment 5

[0154] A three-dimensional thermal map of rail wear is generated based on a GIS-BIM fusion engine to map the distribution of dynamic risk index TC with color gradient; real-time positioning data of trains and maintenance work order status are integrated to form a multi-layer collaborative visualization interface;

[0155] An interactive diagnosis panel is provided to support manual framing of high-risk sections and linked retrieval of multi-modal data including acoustic emission waveform and infrared thermal imaging, while providing a spatio-temporal comparison view of the predicted side wear Yc and the actual profile side wear , and real-time historical data backtracking analysis through a sliding time axis.

[0156] Receiving model misjudgment samples confirmed by manual in the rail side wear monitoring process, and automatically generating adversarial samples, the model misjudgment samples specifically include false positive cases and false negative cases; then the model misjudgment samples are injected into the long short-term memory network for incremental retraining;

[0157] Meanwhile, the parameter weight in the calculation of the dynamic risk index TC is dynamically adjusted, and the specific logic is as follows:

[0158] If it is marked as an "emergency section", the weight of the wear acceleration Mj is increased to quickly capture the sudden wear risk, and the weight of the curvature stress Qyl is reduced to focus on the short-term risk response;

[0159] If it is marked as a "regular section", the weight of the curvature stress Qyl is increased to strengthen the analysis of long-term cumulative effects, and the weight of the wear acceleration Mj is reduced to reduce the sensitivity to short-term fluctuations;

[0160] Online learning technology is adopted to fine-tune the model parameters of the rail side wear trend prediction model in real time and generate a feature weight drift analysis report.

[0161] Further, the false positive case is specifically that the model predicts high risk, i.e., the dynamic risk index TC is greater than or equal to the dynamic risk threshold T, but the actual detection does not reach the side wear threshold of the section;

[0162] The false negative case is specifically that the model predicts safety, i.e., the dynamic risk index TC is less than the dynamic risk threshold T, but the actual occurrence of abnormal wear of the section.

[0163] The data sources of the false positive case and the false negative case are the abnormal section records marked by maintenance personnel in actual inspection or review, combined with historical maintenance logs for verification.

[0164] Based on the weight parameters a1, a2 and a3 of each parameter in the dynamic risk index TC, and a1+a2+a3=100%; The specific adjustment value is determined according to the formula:

[0165] If it is marked as an "emergency section", the weight of the short-term sudden risk response is increased, the weight of the wear acceleration Mj is increased, and the weight of the curvature stress Qyl is reduced;

[0166] Adjustment formula: If the initial weights a1=30%, a2=40%, and a3=30%, then after adjustment:

[0167] ;

[0168] ;

[0169] ;

[0170] wherein, and are adjustment coefficients, which are specifically determined by actual working conditions combined with expert experience, are used for dynamic scaling of the weight, and ∈[1.1,1.5], ∈[0.8,1.0];

[0171] If marked as "regular section", the long-term cumulative effect analysis is strengthened, the curvature stress Qyl weight is increased, and the wear acceleration Mj weight is reduced;

[0172] Adjustment formula: initial weight as above, after adjustment:

[0173]

[0174]

[0175]

[0176] In this embodiment, the three-dimensional thermal map of rail wear is generated based on the GIS-BIM fusion engine, the visualization of the dynamic risk index TC distribution is realized, and the multi-layer collaborative visualization interface is formed by combining the real-time positioning data of the train and the maintenance work order status, which improves the intuitiveness and timeliness of data analysis and decision-making;

[0177] The interactive diagnosis panel supports manual frame selection of high-risk sections and linkage to call multi-modal data, further improving the diagnosis and processing efficiency of abnormal sections; by receiving manually confirmed misjudgment samples, automatically generating adversarial samples and injecting long short-term memory network for incremental retraining, the model prediction accuracy is optimized;

[0178] By adjusting the weight coefficients of wear acceleration Mj and curvature stress Qyl in the calculation of dynamic risk index TC, the system can flexibly respond to the actual situation on site, and improve the pertinence of risk prediction; generate feature weight drift analysis report, form a closed-loop optimization mechanism of man-machine cooperation, and further improve the adaptive ability and real-time optimization ability of the model;

[0179] The current step effectively improves the prediction accuracy and decision support ability, and through the weight coefficient in the calculation of the dynamic risk index TC, the maintenance work is more flexible, fine and efficient.

[0180] It should be noted that all the calculation formulas in the application file use regression analysis including but not limited to machine learning algorithms to deeply analyze the collected relevant parameters, identify their natural trends and mutual relationships. Professional software such as Python's Scikit-learn library or R language is used to automatically generate mathematical models matching the data. Then, the performance of the model is objectively evaluated through methods such as cross-validation, and combined with continuous feedback and optimization to ensure that the created formula truly reflects the inherent law of the data, thereby ensuring its effectiveness and accuracy. In all the calculation formulas in the application, the parameters in each formula are processed by consistent range of dimensionless to ensure that different physical quantities are compared on the same scale; the dimensionless technique includes but is not limited to Min-Max normalization, Z-Score standardization;

[0181] The technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of each embodiment of the present application.

[0182] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, apparatus or device. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, apparatus or device or in conjunction with these instruction execution systems, apparatus or devices.

[0183] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

[0184] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A rail side wear prediction optimization method based on trend prediction, characterized in that, The specific steps include: S1, acquiring the three-dimensional profile data of the rails of a plurality of continuous track sections in a target area, synchronously collecting real-time running data of a train on the plurality of track sections, and constructing a dynamic correlation data set; the track sections include curved tracks and small-radius curve sections; a sampling priority is set, and a sampling frequency is automatically adjusted; the dynamic correlation data set includes the curvature stress of each track section; S2, based on the dynamic correlation data set, relevant features are extracted and input into a long short-term memory network model for training, a rail side wear trend prediction model is constructed, and a predicted side wear amount Yc and a wear acceleration Mj in a future m time period are output; The predicted side wear amount Yc, the wear acceleration Mj, and the curvature stress are combined and analyzed to construct a dynamic risk index TC, which is used to mark the risk degree of the current track section; S3, according to the output result of the rail side wear trend prediction model, an optimization strategy is generated by establishing and combining a line maintenance rule library, including: triggering re-measurement for a high-risk section; secondarily adjusting the sensor sampling frequency according to the deviation between the predicted side wear amount Yc and the actual monitoring data; outputting a maintenance window suggestion coordinated with a train operation plan; S4, visualization and feedback calibration, the predicted side wear trend, the high-risk section, and the optimization maintenance strategy are displayed through a visualization interface, and artificial calibration input is received to update the weight parameters of each parameter in the dynamic risk index TC in real time.

2. The rail side wear prediction optimization method based on trend prediction of claim 1, wherein: Based on a continuous sampling mode, real-time running data and three-dimensional profile data of rails are collected in real time by using on-board sensors and track sensors, and spatio-temporal matching is performed to establish a dynamic correlation data set; By combining real-time running data and track geometric parameters, the radius of each curved track section and the curvature change gradient are analyzed and obtained, and the curvature stress Qyl of each track section is calculated according to the running speed and axle load of the train; The priority grading threshold of the curvature stress is set, including Qa and Qb, and Qa>Qb; then the curvature stress Qyl is compared and evaluated with the priority grading threshold, and the sampling priority of each track section is determined, including: When the curvature stress Qyl is greater than or equal to Qa, the current track section is determined as a first-priority sampling section; When Qb is less than the curvature stress Qyl and the curvature stress Qyl is less than Qa, the track section is determined as a second-priority sampling section; When the curvature stress Qyl is less than Qb, the track section is determined as a third-priority sampling section.

3. The rail side wear prediction optimization method based on trend prediction of claim 2, wherein: After the sampling priority of each track section is determined, the sampling point number Cds in the current monitoring time period is calculated according to the priority of each section, and the sampling frequency of the track section is automatically adjusted; The specific calculation formula of the sampling point number Cds is as follows: ； In the formula, represents the reference sampling point number under the standard curvature stress; represents the reference curvature stress value; The preset sampling point number threshold includes Ca and Cb, and Ca> Cb; The sampling point number threshold is compared and evaluated with the sampling point number Cds, including: When the sampling point number Cds is greater than Ca, the sampling frequency of the current track section is increased; When Cb is less than the sampling point number Cds and the sampling point number Cds is less than or equal to Ca, the sampling frequency of the current track section is maintained; When the sampling point number Cds is less than or equal to Cb, the sampling frequency of the current track section is reduced.

4. The rail side wear prediction optimization method based on trend prediction of claim 3, wherein: The data in the dynamic correlation data set is segmented and processed in a sliding window manner, and the time sequence wear characteristics, dynamic stress characteristics and working condition correlation characteristics in the dynamic correlation data set are extracted; The time sequence wear characteristics, dynamic stress characteristics and working condition correlation characteristics are input into a long short-term memory network enhanced by a spatio-temporal attention mechanism for training, the output wear acceleration Mj and the predicted side wear Yc in the future m time periods are associated and fitted after being dimensionless processed with the curvature stress Qyl, and the dynamic risk index TC of the current track section is obtained by calculation, and the specific calculation formula is as follows: ； In the formula, a1, a2 and a3 respectively represent the weight parameters of the wear acceleration Mj, the predicted side wear Yc and the curvature stress Qyl, and a1+a2+a3=1.

5. The rail side wear prediction optimization method based on trend prediction of claim 4, wherein: Based on statistical analysis of historical accident data and material fatigue limit, the dynamic risk threshold T is preset, and compared with the dynamic risk index TC for evaluation, and the specific evaluation content is as follows: If the dynamic risk index TC is greater than or equal to the dynamic risk threshold T, the current track section is marked as a high-risk section; If the dynamic risk index TC is less than the dynamic risk threshold T, the current track section is determined as a safe section, and the current monitoring mode is maintained.

6. The rail side wear prediction optimization method based on trend prediction of claim 5, wherein: Based on historical track maintenance records, standardized maintenance specifications and side wear evolution experience data under different operation conditions, a line maintenance rule library is constructed; When the current track section is marked as a high-risk section, the side wear early warning threshold Yth is preset by using track wear limit standard, safety operation requirement and comprehensive statistical analysis of track inspection data in recent years, and compared with the predicted side wear Yc in real time to evaluate whether the side wear exceeds the standard in the future m time periods; The specific evaluation content is as follows: If the predicted side wear Yc is greater than or equal to the side wear early warning threshold Yth, it is determined that the side wear exceeds the standard in the future m time periods, and the retest is triggered at this time; If the predicted side wear Yc is less than the side wear early warning threshold Yth, it is determined that the side wear does not exceed the standard in the future m time periods, and the current detection frequency is maintained unchanged at this time.

7. The rail side wear prediction optimization method based on trend prediction of claim 6, wherein: After the retest trigger, the actual profile side wear of the current track section is collected in real time by the trackside sensor within m period ; extracting a predicted side wear Yc and an actual profile side wear by calculating a side wear deviation value PCy in real time, and performing secondary determination, and the specific calculation formula is as follows: ； In the formula, PCy∈[0,5]; When the side wear amount deviation value PCy = 0, it indicates that the predicted value of the predicted side wear amount Yc of the current track section is accurate, and the actual profile side wear amount Completely identical, i.e. no deviation; When the side wear deviation value PCy is greater than 0, it indicates that there is a certain deviation between the predicted value and the actual measured value of the current track section, and the greater the deviation value, the greater the prediction error; The deviation threshold Pth is set to 5, and compared with the side wear deviation value PCy to determine whether the current track section is a significant deviation section and obtain the track side wear trend prediction result; The specific content is as follows: When the side wear deviation value PCy is greater than or equal to the deviation threshold Pth and greater than 0, it is determined that the current track section is a significant deviation section and is marked as an "emergency section", which is preferentially included in the high-frequency detection and maintenance plan; When the deviation threshold Pth is greater than the side wear deviation value PCy and greater than 0, it is determined that the section is a deviation acceptable section and is marked as a "normal section", and the current detection frequency is maintained unchanged.

8. The rail side wear prediction optimization method based on trend prediction of claim 7, wherein: After it is determined that the current track section is a significant deviation section, the train operation plan data is read and based on which the future available maintenance window period is queried; Combined with the track side wear trend prediction results and the operation density of trains, the multi-objective optimization scheduling is used to prioritize and match the window optimization algorithm, generate and output the maintenance window suggestion information, and push it to the maintenance scheduling platform through the wireless communication module.

9. The rail side wear prediction optimization method based on trend prediction of claim 8, wherein: Based on the GIS-BIM fusion engine, a three-dimensional thermal map of rail wear is generated to map the distribution of dynamic risk index TC with color gradient; real-time positioning data of trains and maintenance work order status are integrated to form a multi-layer collaborative visualization interface; An interactive diagnostic panel is provided to support manual bounding box drawing on high risk sections and to trigger multi-modal data retrieval including acoustic emission waveform and infrared thermography, while providing a time-space comparison view of predicted side wear Yc and actual profile side wear with real-time historical data back analysis by sliding time axis.

10. The rail side wear prediction optimization method based on trend prediction of claim 8, wherein: Receive the model misjudgment samples confirmed by artificial in the process of rail side wear monitoring, and automatically generate adversarial samples, the model misjudgment samples specifically include false positive cases and false negative cases; then inject the model misjudgment samples into the long short-term memory network for incremental retraining; At the same time, the parameter weight in the calculation of the dynamic risk index TC is dynamically adjusted, and the specific logic is as follows: If it is marked as "emergency section", increase the weight of wear acceleration Mj to quickly capture the mutation wear risk; reduce the weight of curvature stress Qyl to focus on short-term risk response; If it is marked as "regular section", increase the weight of curvature stress Qyl to strengthen the analysis of long-term cumulative effect; reduce the weight of wear acceleration Mj to reduce the sensitivity to short-term fluctuations; Online learning technology is used to fine-tune the model parameters of the rail side wear trend prediction model in real time and generate a feature weight drift analysis report.

Images