Intelligent monitoring method and system for ballastless track roadbed settlement repair

Abstract

The present application belongs to the field of subgrade settlement repair, and discloses an intelligent monitoring method and system for subgrade settlement repair of a ballastless track, which comprises the following steps: arranging monitoring units on both sides of the track structure in the subgrade settlement area and collecting track multi-source data in real time, constructing a multi-factor error correction model, calculating the calibrated elevation after obtaining the prediction error through the error prediction model in combination with the original elevation data; building a settlement-lifting prediction model, outputting the lifting amount prediction value in the future preset time period, and guiding the dynamic adjustment of the grouting amount and pressure according to the deviation between the lifting amount prediction value and the target lifting amount; the data processing unit analyzes the track structure state in real time to realize safety warning; and the corresponding system is also disclosed. The present application solves the problem that single elevation monitoring cannot obtain the coupling influence of the environment and the subgrade parameters on the settlement repair, avoids the missing and omission of monitoring data, improves the comprehensiveness and accuracy of monitoring, and ensures the safe operation of trains.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of subgrade settlement repair, in particular to an intelligent monitoring method and system for subgrade settlement repair of ballastless track. BACKGROUND

[0002] The smoothness of the track is the basis for ensuring the safe and comfortable operation of high-speed railways. After a long time of operation, the subgrade of some lines has uneven settlement, which leads to uneven track alignment and seriously affects the safe operation of trains. In order to quickly restore the smoothness of the line, it is of great importance to study the monitoring method for subgrade settlement repair of high-speed railway ballastless track.

[0003] At present, the main method for repairing subgrade settlement is to use polymer grouting material for lifting repair. During the grouting lifting process, a level instrument is used to measure the elevation of the lifting point. Two people are needed to hold the staff, and one person is needed to measure. Since the staff is manually held, the perpendicularity of the staff cannot be guaranteed every time the staff is measured, resulting in a large measurement error. The use of a level instrument for point-by-point measurement has low grouting lifting construction efficiency. Usually, an 8-person work team can only complete about 30 meters of subgrade settlement repair work in one day window. At the same time, the grouting personnel and the measurement personnel work in the same area, which often causes the measurement line of sight to be blocked, seriously affecting the construction organization and progress. Moreover, the monitoring only covers the repair stage, lacks long-term tracking during the operation period, cannot predict the risk of secondary settlement, and is difficult to meet the high safety and high efficiency operation and maintenance requirements of high-speed rail. SUMMARY

[0004] The present application aims to solve the problems of low measurement accuracy, large construction interference, lack of long-term tracking during the operation period, and inability to predict the risk of secondary settlement in the existing technology of subgrade settlement repair of ballastless track.

[0005] In order to achieve the above-mentioned purpose, the present application relates to an intelligent monitoring method for subgrade settlement repair of ballastless track, comprising the following steps:

[0006] Step 1: Monitoring system deployment: monitoring units are arranged on both sides of the track structure in the subgrade settlement area, the monitoring units include static level instruments, temperature sensors, vibration acceleration sensors, air pressure sensors, soil moisture content sensors and compaction degree sensors, fixed monitoring points are set according to a preset rule, a data acquisition instrument and a data processing unit are configured in a matched manner, the data acquisition instrument is connected with each monitoring point through a communication cable or a wireless ad hoc network, and the data processing unit communicates with the data acquisition instrument through a network;

[0007] Step 2: Multi-source data acquisition: real-time track elevation data , track structure surface temperature , train vibration acceleration effective value , atmospheric pressure Moisture content of roadbed fill material and compaction degree The data acquisition device automatically collects and stores the above-mentioned multi-source data and uploads it synchronously to the data processing unit.

[0008] Step 3: The data processing unit performs dual-model collaborative processing: First, a multi-factor error correction model is constructed using a gradient boosting regression tree network. The vector consisting of the original, quadratic, and interaction terms of environmental factors is used as input. After obtaining the prediction error through the error prediction model, the calibrated elevation is calculated by combining it with the original elevation data. Then, a settlement-uplift prediction model is built based on a neural network. This model integrates the calibrated elevation sequence, grouting pressure sequence, and roadbed parameters to form an input vector. The model outputs the predicted uplift value for a future preset time period. The dual-model collaboration achieves accurate data calibration and uplift trend prediction.

[0009] Step 4: Construction Guidance: Based on the predicted lift amount and the target lift amount deviation It generates grouting parameter adjustment instructions to guide on-site dynamic adjustment of grouting volume and pressure;

[0010] Step 5: Full life cycle monitoring: After the track maintenance window is completed, the monitoring unit remains installed and continuously collects data during train operation. The data processing unit analyzes the track structure status in real time to achieve safety early warning.

[0011] Furthermore, in step three: first, a multi-factor error correction model is constructed using a gradient boosting regression tree network. The vector consisting of the original environmental factor terms, quadratic terms, and interaction terms is used as input. After obtaining the prediction error using the error prediction formula, the calibrated elevation is calculated by combining the original elevation data.

[0012] A multi-factor error correction model is constructed based on a gradient boosting regression tree network; a vector containing the original, quadratic, and interaction terms of environmental factors is used. For input, through Determine the prediction measurement error Then through Obtain calibrated elevation data , where K is the total number of regression trees (100≤K≤200). For the first Learning rate of trees , This is the output value of the k-th regression tree. For the first Parameters of the regression trees.

[0013] Further, in step three: based on the neural network, a settlement-lifting prediction model is built, the calibrated high sequence, the grouting pressure sequence and the roadbed parameters are integrated to form an input vector, the lifting amount prediction value of the future preset time period is output through the model, and the method of double model cooperation realizes accurate data calibration and lifting trend prediction:

[0014] Based on the LSTM neural network, a settlement-lifting prediction model is built, and the calibrated high sequence , the grouting pressure sequence and the roadbed parameters are reconstructed into an input vector , and the lifting amount prediction value of the future preset time period is output through , wherein n is the time series window length, is the prediction step, is the weight matrix, is the bias vector, is the activation function, is the activation function.

[0015] Further, the preset rules in step one are: based on the early track linear monitoring data, the region with a settlement gradient greater than 5mm / m is arranged with fixed monitoring points at an interval of 3-5m, the settlement gentle region is arranged at an interval of 10m, and a mobile monitoring unit is configured for dynamic supplementary measurement of abnormal grouting areas; the wireless ad hoc network adopts LoRa or WiFi6 Mesh technology, and the monitoring unit is equipped with a solar power supply module and a super capacitor to realize passive deployment.

[0016] Further, the training process of the multi-factor error correction model includes:

[0017] s1. Collect real-time track elevation original data (not affected by temperature, vibration, air pressure and other interference factors, can truly reflect the actual elevation state of the track), synchronously acquire temperature , train vibration acceleration effective value , atmospheric pressure and other environmental factor data, and calculate the real error with the benchmark true value measured by the high-precision laser interferometer ;

[0018] s2. Build input characteristic quantity containing environmental factor original term, quadratic term and interaction term , take the real error as the label, and build the GBRT model;

[0019] s3. Use mean square error loss function to iteratively optimize model parameters to determine the total number of regression trees (100≤K≤200), the first learning rate of the tree and the parameters ;

[0020] s4. In real-time monitoring, the current environmental factor data is substituted into the model, the prediction error is output through the formula , and then the calibrated high-precision elevation data is obtained.

[0021] Further, the training process of the settlement-lifting prediction model in the step includes: collecting historical grouting repair data to construct a training data set, including a historical elevation sequence, a historical grouting pressure sequence, roadbed parameters and a measured lifting amount label , using an error loss function , optimizing model parameters W and b through a back propagation algorithm, and the total number of training samples M is greater than or equal to 5000 groups.

[0022] Further, the grouting parameter adjustment rule in step four is: when the monitoring data exceeds the preset positive threshold, output an instruction to increase the grouting pressure or grouting amount; when the monitoring data exceeds the preset negative threshold, output an instruction to reduce the grouting pressure or grouting amount; when the monitoring data is within the preset threshold interval, output an instruction to maintain the current grouting parameters, realizing closed-loop control of the grouting process.

[0023] Further, the whole life cycle monitoring in step five further includes: constructing a roadbed settlement repair whole life cycle database, integrating multi-source monitoring data before, during and after repair, and using a Prophet time series prediction algorithm to analyze long-term settlement evolution trend and early warning of secondary settlement risk; the data processing unit is connected with a high-speed rail line comprehensive operation and maintenance platform, and the monitoring data is synchronized to the platform to provide data support for preventive maintenance.

[0024] Further, the data processing unit includes an edge computing module, the real-time data processing delay is less than or equal to 100 ms, the lifting amount prediction value and the grouting adjustment instruction are output once every 1-5 seconds, and the hierarchical warning information and operation suggestions are pushed to the on-site construction personnel through a visual terminal; based on the track design elevation and the current settlement amount.

[0025] The monitoring unit adopts modular design, supports quick plug-in installation, the shell adopts high-strength engineering plastic and buffer structure, meets the requirements of anti-vibration and anti-wind pressure during train operation, the installation process does not damage the track structure, and does not need to be removed outside the skylight period.

[0026] ​The data processing unit also supports multi-line data transverse comparison, constructs a case library and an algorithm library, stores repair monitoring data and optimization algorithms of different roadbed types and different settlement degrees, and provides technical references for similar projects.

[0027] As another aspect of the application, the application also relates to an intelligent monitoring system for ballastless track roadbed settlement repair, comprising:

[0028] The monitoring system deployment device is used for deploying monitoring units in the full range on both sides of the track structure in the roadbed settlement area, the monitoring units comprise static leveling instruments, temperature sensors, vibration acceleration sensors, air pressure sensors, soil moisture content sensors and compactness sensors, fixed monitoring points are arranged according to a preset rule, a data acquisition instrument and a data processing unit are configured in a matched manner, the data acquisition instrument is connected with each monitoring point through a communication cable or a wireless ad hoc network, and the data processing unit communicates with the data acquisition instrument through a network;

[0029] The multi-source data acquisition device is used for acquiring real-time track elevation data , track structure surface temperature , train vibration acceleration effective value , atmospheric pressure , roadbed filler moisture content and compactness in real time through the monitoring units, and automatically acquires and stores the above multi-source data through the data acquisition instrument and synchronously uploads the multi-source data to the data processing unit;

[0030] The double-model collaborative processing device is used for constructing a multi-factor error correction model through a gradient boosting regression tree network, taking a vector formed by original items, quadratic items and interaction items of environmental factors as input, calculating the calibrated elevation by combining the original elevation data after obtaining the prediction error through the error prediction model;

[0031] The settlement-lifting prediction model device is used for integrating the calibrated elevation sequence, the grouting pressure sequence and the roadbed parameters to form an input vector, outputting a lifting amount prediction value of a future preset time period through the model, and realizing accurate data calibration and lifting trend prediction through the double models.

[0032] The construction guidance device is used for generating grouting parameter adjustment instructions according to the deviation between the lifting amount prediction value and the target lifting amount , and guiding dynamic adjustment of the grouting amount and pressure on site;

[0033] The whole life cycle monitoring device keeps the monitoring units in the installed state after the skylight construction is completed, continuously acquires data during train operation, and realizes safety early warning by analyzing the track structure state in real time through the data processing unit.

[0034] Compared with the prior art, the application has the following effects:

[0035] (1) The intelligent monitoring method and system for subgrade settlement repair of ballastless track, by installing static leveling instruments in the full range of the track structure on both sides of the subgrade settlement area, automatically collecting data through the single-machine version acquisition software built-in the field acquisition box, storing in the field acquisition system, connecting with the network through wired or wireless communication, and then transmitting to the background network version software, automatic observation is realized, the elevation change of the grouting lifting point is mastered in real time, and the on-site construction is guided.

[0036] During the grouting lifting construction stage, the accuracy of each measurement data can be ensured by using the monitoring method, the number of measurement personnel can be reduced from 3 to 1, so that the human consumption is reduced. The static leveling instrument is used for measuring the elevation, and the automatic collection of data is not affected during the construction of the grouting personnel, so that the construction efficiency and precision can be greatly improved. After the skylight construction is completed, the installed static leveling instrument does not affect the driving safety, does not need to be removed, and during the daytime train operation, monitoring can also be carried out, which plays an important early warning role in ensuring the safe operation of the train.

[0037] (2) The intelligent monitoring method and system for subgrade settlement repair of ballastless track, the monitoring unit containing static leveling instruments, temperature sensors, vibration acceleration sensors, air pressure sensors, soil moisture content sensors and compactness sensors is arranged on both sides of the track structure in the full range of the subgrade settlement area, and fixed monitoring points are adaptively set according to the settlement gradient, and a mobile monitoring unit is also provided. The multi-dimensional data such as real-time elevation of the track, environmental factors and parameters of the subgrade itself are comprehensively and accurately collected, the problem that single elevation monitoring cannot obtain the coupling influence of environmental and subgrade parameters on settlement repair is solved, the missing and omission of monitoring data are avoided, and the comprehensiveness and accuracy of monitoring are improved.

[0038] (3) The intelligent monitoring method and system for subgrade settlement repair of ballastless track, double model cooperation (GBRT error correction + LSTM lifting prediction) is adopted to predict the lifting amount in advance and dynamically adjust the grouting parameters, so as to avoid over-lifting and under-lifting, and improve the construction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0039] Fig. 1 It is the arrangement flow schematic diagram of the preferred embodiment of the present application.

[0040] Fig. 2 It is the monitoring construction installation schematic diagram (initial installation state) of the preferred embodiment of the present application.

[0041] Fig. 3 It is the line shape comparison graph of high-speed railway subgrade settlement grouting repair before and after (the unit of the abscissa in the graph is meter, indicating the distance from the starting point K83).

[0042] In the figure, the reference numerals represent: 1 - control circuit, 2 - liquid storage tank; DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0044] It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that, unless otherwise specified, all technical and scientific terms used in the present application have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0046] Please refer to Figs. 1-2 The embodiments of the present application relate to an intelligent monitoring method for repairing the settlement of a ballastless track subgrade, comprising the following steps:

[0047] Step one: deployment of the monitoring system: monitoring units are arranged on both sides of the track structure in the settlement area of the subgrade, the monitoring units include static leveling instruments, temperature sensors, vibration acceleration sensors, air pressure sensors, soil moisture content sensors and compactness sensors, fixed monitoring points are set according to a preset rule, a data acquisition instrument and a data processing unit are configured in a matched manner, the data acquisition instrument is connected with each monitoring point through a communication cable or a wireless ad hoc network, and the data processing unit communicates with the data acquisition instrument through a network;

[0048] Step two: multi-source data acquisition: real-time track elevation data , track structure surface temperature , train vibration acceleration effective value , atmospheric pressure , moisture content of subgrade filler and compactness are collected in real time by the monitoring units, the above multi-source data are automatically collected and stored by the data acquisition instrument, and are synchronously uploaded to the data processing unit;

[0049] Step three: Because environmental interference (temperature, vibration, air pressure, etc.) will directly change the liquid density of the static level instrument, the container deformation or the measurement reference, such as high temperature causing the expansion of the liquid, train vibration causing the fluctuation of the liquid surface, making the reading deviate from the true value, so the embodiment adopts a data processing unit for double model collaborative processing: first, a multi-factor error correction model is constructed through a gradient boosting regression tree network, a vector composed of original terms, quadratic terms and interaction terms of environmental factors is input, a predicted error is obtained through the error prediction model, and then the calibrated elevation is calculated combined with the original elevation data; a settlement-lifting prediction model is built based on a neural network, an input vector is formed by integrating the calibrated elevation sequence, the grouting pressure sequence and the roadbed parameters, the lifting amount prediction value of the future preset time period is output through the model, and the double model collaborates to realize accurate data calibration and lifting trend prediction;

[0050] Step four: construction guidance: according to the deviation of the lifting amount prediction value and the target lifting amount , generate grouting parameter adjustment instructions to guide the dynamic adjustment of the grouting amount and pressure on site;

[0051] Step five: whole life cycle monitoring: after the skylight construction is completed, the monitoring unit remains in the installed state, the data processing unit analyzes the track structure state in real time during the train operation to realize safety warning.

[0052] In some preferred embodiments, in step three: first, a multi-factor error correction model is constructed through a gradient boosting regression tree network, a vector composed of original terms, quadratic terms and interaction terms of environmental factors is input, a predicted error is obtained through the error prediction formula, and then the calibrated elevation is calculated combined with the original elevation data.

[0053] A multi-factor error correction model is constructed based on a gradient boosting regression tree network; a vector containing original terms, quadratic terms and interaction terms of environmental factors is input, a predicted measurement error is determined through , , , , wherein K is the total number of regression trees (100≤K≤200), is the learning rate of the th tree, , is the output value of the kth regression tree, is the parameter of the th regression tree.

[0054] ​In some preferred embodiments, in step three: based on the neural network, a settlement-lifting prediction model is built, the calibrated high sequence, the grouting pressure sequence and the subgrade parameters are integrated to form an input vector, the lifting amount prediction value of the future preset time period is output through the model, and the method of double model cooperation realizes accurate data calibration and lifting trend prediction.

[0055] Based on the LSTM neural network, a settlement-lifting prediction model is built, and the calibrated high sequence , the grouting pressure sequence and the subgrade parameters are reconstructed into an input vector , and the lifting amount prediction value of the future preset time period is output through , wherein n is the time series window length, is the prediction step, is the weight matrix, is the bias vector, is the activation function, is the activation function.

[0056] During the subgrade lifting process, the track elevation change and the grouting pressure adjustment have strong time correlation (such as the current lifting amount being affected by the pressure change and the elevation cumulative change in the previous 10-15 seconds), and the LSTM can accurately capture the long and short term dependence of time series data through the gating mechanism (forget gate, input gate and output gate). The internal law of the high sequence and the grouting pressure sequence evolving over time can be effectively mined. The LSTM model has high inference efficiency and can quickly respond in the calculation unit, and can quickly output the future lifting amount prediction value based on the updated real-time data, provide timely decision support for dynamic adjustment of grouting parameters, and meet the short window period and fast construction rhythm of ballastless track repair scene.

[0057] In some preferred embodiments, the preset rules in step one are: based on the early track linear monitoring data, the areas with a settlement gradient greater than 5 mm / m are arranged with fixed monitoring points at an interval of 3-5 m, the settlement gentle areas are arranged at an interval of 10 m, and a mobile monitoring unit is configured for dynamic supplementary measurement of abnormal grouting areas; the wireless ad hoc network adopts LoRa or WiFi6Mesh technology, and the monitoring unit is equipped with a solar power supply module and a super capacitor to realize passive deployment.

[0058] In some preferred embodiments, the training process of the multi-factor error correction model includes:

[0059] s1. Collecting real-time track elevation original data (not affected by temperature, vibration, air pressure and other interference factors, and can truly reflect the actual elevation state of the track), synchronously acquiring temperature Train vibration acceleration effective value Atmospheric pressure Real error Real error ;

[0060] s2. Build input feature vectors containing environmental factor raw terms, quadratic terms, and interaction terms Build a GBRT model with real error as the label; the influence of environmental factors such as temperature, vibration, and atmospheric pressure on the measurement results of the static level instrument is not linear, and the original term cannot represent the "difference in error rate with factor change"; the quadratic term can effectively fit this non-linear relationship, allowing the model to identify that the greater the factor intensity, the more significant the error increase, and avoid error prediction bias caused by linear assumptions; in addition, in actual monitoring scenarios, measurement errors are often the result of multiple environmental factors acting together, rather than a single factor affecting independently. For example, in a high-temperature environment, the liquid surface fluctuation caused by train vibration will be more intense (temperature rise reduces liquid viscosity, vibration interference is amplified), and this temperature + vibration synergistic interference cannot be represented by a single or original term; the interaction term can quantify the interference intensity after multiple factor superposition, allowing the model to learn the "specific temperature and vibration combination corresponding to a specific error increment" rule, avoiding missing key interference coupling effects; after introducing quadratic terms and interaction terms, the feature vector has more dimensions and more comprehensive information, and the GBRT model can accurately fit the complex mapping between error and environmental factors through the combination of multiple regression trees; the generation of interaction terms must be based on environmental factor data collected at the same time to avoid timestamp misalignment that prevents the interaction term from truly reflecting multiple factor synergistic interference.

[0061] Specific generation method of interaction term: first, synchronize the original data of multiple core environmental factors through corresponding sensors: for example, track structure surface temperature, train vibration acceleration effective value, and atmospheric pressure, to ensure that the three types of data are completely aligned by timestamp; combine the three types of original data in pairs and generate interaction terms through multiplication operation; the specific combination rule is: temperature and vibration interaction term: calculate the temperature original value at the same time multiplied by the vibration acceleration effective value; temperature and pressure interaction term: calculate the temperature original value at the same time multiplied by the pressure original value; vibration and pressure interaction term: calculate the vibration acceleration effective value at the same time multiplied by the pressure original value.

[0062] s3. Use mean square error loss function Iterative optimization of model parameters to determine the total number of regression trees (100≤K≤200), the learning rate of the th tree and parameters ;

[0063] s4. In real-time monitoring, the current environmental factor data is substituted into the model, the prediction error is output by the formula , and the calibrated high-precision elevation data is obtained after .

[0064] In some preferred embodiments, the training process of the above settlement-lifting prediction model includes: collecting historical grouting repair data to construct a training data set, including a historical elevation sequence, a historical grouting pressure sequence, roadbed parameters, and a measured lifting amount label , using an error loss function , and optimizing the model parameters W and b through a back propagation algorithm.

[0065] In some preferred embodiments, the above grouting parameter adjustment rule can be: when the monitoring data exceeds the preset positive threshold, output an instruction to increase the grouting pressure or grouting amount; when the monitoring data exceeds the preset negative threshold, output an instruction to reduce the grouting pressure or grouting amount; when the monitoring data is within the preset threshold interval, output an instruction to maintain the current grouting parameters, realizing closed-loop control of the grouting process.

[0066] In some preferred embodiments, the whole life cycle monitoring further includes: constructing a roadbed settlement repair whole life cycle database, integrating multi-source monitoring data before, during and after repair, using a Prophet time series prediction algorithm to analyze long-term settlement evolution trend and early warning secondary settlement risk; the data processing unit is connected with a high-speed rail line comprehensive operation and maintenance platform, and the monitoring data is synchronized to the platform to provide data support for preventive maintenance. After accumulating a large amount of data, case data of ballastless track roadbed settlement repair in different regions and different roadbed types (permafrost area, coastal soft soil area, loess roadbed, etc.) can be integrated, the case data includes roadbed basic information (type, geological condition, regional climate), settlement characteristic data (settlement gradient, maximum settlement amount, settlement distribution range), construction parameter data (grouting material type, grouting pressure / amount curve, monitoring point layout scheme) and repair effect data (lifting precision, secondary settlement amount in operation period, track smoothness index), forming a standardized case data set;

[0067] Based on the case data set, the correlation relationship network of subgrade type-settlement feature-grouting parameter-repair effect is established by knowledge graph technology, and the discrete data of each case is converted into structured knowledge nodes and correlation edges to clarify the mapping rules of parameter combination and repair effect in different scenarios. For example, taking the subgrade type as the core first-level node, the secondary nodes of settlement features are derived (such as freeze-thaw cycle type settlement and settlement gradient ≤3mm / m in permafrost area, and consolidation compression type settlement and settlement gradient ≥5mm / m in coastal soft soil area); the settlement feature nodes and grouting parameter nodes establish directional correlation edges; the grouting parameter nodes and repair effect nodes establish quantitative correlation edges, and the lifting accuracy interval corresponding to different parameter combinations is marked to form a traceable and inferable knowledge network.

[0068] In another aspect, considering the pain point that the traditional knowledge graph scheme only focuses on the construction stage and ignores the long-term operation effect, on the basis of the original core nodes, construction stage nodes (such as initial grouting, mid-term adjustment, and late-stage supplementary grouting) and operation period environmental dynamic nodes (such as seasonal changes, train load fluctuations, and geological settlement trends) can be added; the correlation edges add time sequence dependent attributes, such as the correlation edge of “soft soil subgrade + initial grouting pressure 1.0MPa and secondary settlement amount ≤2mm after 3 years of operation”, which marks: time span = 3 years, environmental impact weight = 0.3, and other dynamic parameters.

[0069] Then, similarity matching (such as cosine similarity and Jaccard coefficient) is adopted to calculate the multi-dimensional similarity of the scene features (subgrade type, settlement data, and regional environment) of the new project and the historical cases in the case library, and the optimal matching case is selected; combined with the correlation rules of the knowledge graph, the optimal repair scheme (grouting parameter combination) and monitoring scheme (monitoring unit layout interval and core monitoring parameter priority) suitable for the new project are inferred and generated, and are automatically pushed to the field operation terminal.

[0070] In some preferred embodiments, the data processing unit includes an edge computing module, the real-time data processing delay is ≤100ms, the lifting amount prediction value and grouting adjustment instruction are output once every 1-5 seconds, and the hierarchical early warning information and operation suggestions are pushed to the field construction personnel through the visual terminal; Based on the track design elevation and the current settlement amount.

[0071] When the ballastless track is applied in the tunnel section, the temperature sensor and the vibration acceleration sensor can be deleted due to the stable temperature and small vibration interference in the tunnel, and only the static level gauge, the air pressure sensor, and the subgrade moisture content sensor are retained to reduce the equipment cost; in the bridge-subgrade transition section, an inclination sensor (measuring track smoothness change) needs to be added to strengthen the settlement monitoring of the transition section;

[0072] Please refer to Fig. 3, a high-speed railway line with a total length of 301 km and a design speed of 350 km / h was officially opened for operation in July 2010, and the main line was laid with CRTSI double-block ballastless track. According to the track alignment monitoring data in April 2023, it was found that the subgrade section from K83+300 to K83+850 had uneven settlement to varying degrees, and the maximum settlement of the up and down lines reached 39 mm and 33 mm respectively, which urgently needed repair treatment.

[0073] The relative deviation of the track elevation after grouting lifting repair from the design elevation is between-0.94 mm and 0.83 mm, with an average difference of 0.7 mm, which can be adjusted by ordinary fasteners to meet the requirements of line opening. The track structure state is stable after grouting lifting repair, and the line alignment is smooth. The line has resumed normal operation at normal speed in June 2023, and no abnormalities have been found so far.

[0074] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0075] As those skilled in the art can easily understand, the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent monitoring method for ballastless track subgrade settlement repair, characterized in that, Comprise the following steps: Step one: monitoring system deployment: the monitoring unit is arranged on both sides of the track structure in the subgrade settlement area, which includes static level gauge, temperature sensor, vibration acceleration sensor, air pressure sensor, soil moisture content sensor and compaction sensor, fixed monitoring points are set according to the preset rule, data acquisition instrument and data processing unit are configured, the data acquisition instrument is connected with each monitoring point through communication cable or wireless ad hoc network, and the data processing unit communicates with the data acquisition instrument through network; Step two: multi-source data acquisition: real-time acquisition of track real-time elevation data by monitoring unit , track structure surface temperature , train vibration acceleration effective value , atmospheric pressure , roadbed filler moisture content and compaction degree , automatically collect and store the above multi-source data by the data acquisition instrument, and synchronously upload to the data processing unit; Step three: the data processing unit carries out double model collaborative processing: first, a multi-factor error correction model is constructed through gradient boosting regression tree network, the vector composed of original items, quadratic items and interaction items of environmental factors is input, the predicted error is obtained through the error prediction model, and the calibrated elevation is calculated combined with the original elevation data; Step four: a settlement-lifting prediction model is built based on neural network, the input vector is formed by integrating the calibrated elevation sequence, grouting pressure sequence and subgrade parameters, the lifting amount prediction value of future preset time period is output through the model, and the double model cooperates to realize accurate data calibration and lifting trend prediction; Step five: construction guidance: according to the predicted value of the lifting amount deviation from the target lifting amount Generate grouting parameter adjustment instructions to guide dynamic adjustment of grouting amount and pressure on site;​ Step six: whole life cycle monitoring: after the construction of the skylight is completed, the monitoring unit remains in the installed state, data is continuously collected during train operation, the data processing unit analyzes the track structure state in real time, and safety warning is realized. 2.The intelligent monitoring method for ballastless track subgrade settlement repair according to claim 1, characterized in that, In the step three: first, a multi-factor error correction model is constructed through gradient boosting regression tree network, the vector composed of original items, quadratic items and interaction items of environmental factors is input, the predicted error is obtained through the error prediction formula, and the calibrated elevation is calculated combined with the original elevation data; A multi-factor error correction model is constructed based on a gradient boosting regression tree network; a vector containing the original, quadratic, and interaction terms of environmental factors is used. For input, through Determine the prediction measurement error Then through Obtain calibrated elevation data Where K is the total number of regression trees. For the first The learning rate of each tree, of which , This is the output value of the k-th regression tree. For the first Parameters of the regression trees. 3.The intelligent monitoring method for ballastless track subgrade settlement repair according to claim 1, characterized in that, In the step four: a settlement-lifting prediction model is built based on neural network, the input vector is formed by integrating the calibrated elevation sequence, grouting pressure sequence and subgrade parameters, the lifting amount prediction value of future preset time period is output through the model, and the double model cooperates to realize accurate data calibration and lifting trend prediction; A subsidence-uplift prediction model was constructed based on an LSTM neural network, and the calibrated high-order series was used. Grouting pressure sequence and roadbed parameters Reconstructed into the input vector ,pass Output the predicted elevation over a future preset time period, where n is the time series window length. To predict the step size, This is the weight matrix. For bias vectors, for Activation function for Activation function.

4. The method of claim 1, wherein, The preset rule in step one is: based on the previous track alignment monitoring data, the fixed monitoring points are arranged at an interval of 3-5m in the area with a settlement gradient greater than 5mm / m, and the mobile monitoring unit is arranged at an interval of 10m in the area with gentle settlement, and dynamic supplementary measurement is carried out for the abnormal grouting area; the wireless ad hoc network adopts LoRa or WiFi6 Mesh technology, the monitoring unit is equipped with a solar power supply module and a super capacitor, and passive deployment is realized. 5.The intelligent monitoring method for ballastless track foundation settlement repair according to claim 2, characterized in that, The training process of the multi-factor error correction model comprises: s1. Collecting track real-time elevation raw data , synchronously acquiring temperature , train vibration acceleration effective value , atmospheric pressure environmental factor data, the benchmark true value measured by the high-precision laser interferometer Calculate real error ; s2. Constructing input feature quantities including environmental factor original terms, quadratic terms, and interaction terms , with real errors as labels, constructing a GBRT model; s3. Employing a mean squared error loss function Iteratively optimizing model parameters to determine total number of regression trees , 100 ≤ K ≤ 200, 1st Learning rate for each tree and parameters ; s4. When monitoring in real time, the current environmental factor data is substituted into the model, and the formula is used to output the prediction error, and then the calibrated high-precision elevation data is obtained. 6.The intelligent monitoring method for ballastless track subgrade settlement repair according to claim 1, characterized in that, The training process of the settlement-uplift prediction model includes: collecting historical grouting repair data to construct a training dataset, which includes historical elevation sequences, historical grouting pressure sequences, subgrade parameters, and labels for measured uplift. Using the error loss function The model parameters W and b are optimized using the backpropagation algorithm, where It is the first Predicted elevation values ​​for the group sample. It is the first Measured values ​​of elevation in the group of samples. This represents the total number of training samples. 7.The intelligent monitoring method for ballastless track subgrade settlement repair according to claim 1, characterized in that, The grouting parameter adjustment rule in step five is: when the monitoring data exceeds the preset positive threshold, the instruction of increasing grouting pressure or grouting amount is output; when the monitoring data exceeds the preset negative threshold, the instruction of reducing grouting pressure or grouting amount is output; when the monitoring data is within the preset threshold interval, the instruction of maintaining the current grouting parameters is output, and closed loop control of the grouting process is realized. 8.The intelligent monitoring method for ballastless track subgrade settlement repair according to claim 1, characterized in that, The whole life cycle monitoring in step six further includes: constructing a whole life cycle database for subgrade settlement repair, integrating multi-source monitoring data before, during and after repair and operation, analyzing long-term settlement evolution trend by using a Prophet time series prediction algorithm, and early warning of secondary settlement risk; the data processing unit is connected with a high-speed rail line comprehensive operation and maintenance platform, and the monitoring data is synchronized to the platform to provide data support for preventive maintenance.

9. The intelligent monitoring method for ballastless track foundation settlement repair according to any one of claims 1-8, characterized in that, The data processing unit comprises an edge computing module, real-time data processing delay is less than or equal to 100 ms, a lifting amount prediction value and a grouting adjustment instruction are output once every 1-5 seconds, and hierarchical early warning information and operation suggestions are pushed to on-site construction personnel through a visual terminal; Based on the track design elevation and the current settlement amount calculation determination.

10. An intelligent monitoring system for ballastless track subgrade settlement repair, characterized in that, It comprises: A monitoring system deployment device: used for deploying monitoring units in the whole range of the track structure on both sides of the subgrade settlement area, the monitoring units including static leveling instruments, temperature sensors, vibration acceleration sensors, air pressure sensors, soil moisture content sensors and compaction degree sensors, fixed monitoring points are set according to a preset rule, a data acquisition instrument and a data processing unit are configured in a matched manner, the data acquisition instrument is connected with each monitoring point through a communication cable or a wireless ad hoc network, and the data processing unit communicates with the data acquisition instrument through a network; Multi-source data acquisition device: used for real-time acquisition of track real-time elevation data by monitoring unit , track structure surface temperature , train vibration acceleration effective value , atmospheric pressure , roadbed filler moisture content and compaction degree , automatically collect and store the above multi-source data by data acquisition instrument, and synchronously upload to data processing unit; A double-model collaborative processing device: a multi-factor error correction model is first constructed by using a gradient boosting regression tree network, a vector composed of original terms, quadratic terms and interaction terms of environmental factors is taken as input, after the predicted error is obtained through the error prediction model, the calibrated elevation is calculated in combination with the original elevation data; A settlement-lifting prediction model device: used for integrating the calibrated elevation sequence, the grouting pressure sequence and the subgrade parameters to form an input vector, outputting the lifting amount prediction value in a future preset time period through the model, and realizing accurate data calibration and lifting trend prediction by the double models in collaboration; Construction guidance device: according to the lifting amount prediction value The deviation of the target lifting amount Generate grouting parameter adjustment instructions to guide the dynamic adjustment of grouting amount and pressure on site;​ A whole life cycle monitoring device: after the skylight construction is completed, the monitoring units remain in the installed state, data is continuously collected during train operation, the data processing unit analyzes the track structure state in real time, and safety early warning is realized.

Images