A box culvert jacking support and settlement construction monitoring method under normal operation of railway

Abstract

The present application provides a kind of box culvert jacking support and settlement construction monitoring method under normal operation of railway, it is related to railway construction monitoring technical field, the method includes: obtaining the multi-source monitoring time series data of box culvert jacking construction area under normal operation of railway and current box culvert jacking real-time construction control parameter;Multi-source monitoring time series data is executed adaptive filtering denoising and timestamp alignment processing, and high-fidelity synchronous monitoring data set is obtained;Based on high-fidelity synchronous monitoring data set, track settlement dynamic characteristics, jacking attitude evolution characteristics and the stress-strain characteristics of current support system are extracted respectively;Track settlement dynamic characteristics, jacking attitude evolution characteristics and the stress-strain characteristics of current support system are respectively mapped into space geometric vector under three-dimensional orthogonal coordinate system, and multi-dimensional space-time cubic feature space is constructed along time axis extension.The present application improves the safety guarantee capability and construction efficiency of underpass engineering of operating railway.

Description

Technical Field

[0001] This invention relates to the field of railway construction monitoring technology, and in particular to a method for monitoring the jacking support and settlement construction of box culverts under normal railway operation. Background Technology

[0002] With the continuous densification of transportation infrastructure networks, the demand for new highways and municipal roads to pass under existing operating railways is increasing. The box culvert jacking method is widely used due to its advantages of minimal interference with railway traffic and high construction efficiency. Taking the construction of a box culvert under a highway on an existing railway trunk line as an example, the railway traffic density in this section is extremely high. During construction, the track geometry must be controlled within millimeter-level precision; otherwise, it will directly endanger traffic safety. Traditional monitoring methods typically employ decentralized approaches such as manually collecting track settlement point elevations at regular intervals, independently monitoring the stress on the support structure with sensors, and independently monitoring the box culvert's attitude with attitude measurement equipment. These methods lack a unified time reference and the data is isolated, making it difficult to reveal the multi-factor coupled evolution law between train dynamic load, box culvert jacking, support system deformation, and track settlement.

[0003] Existing methods can only trigger over-limit alarms after settlement occurs, lacking the ability to predict settlement trends based on multi-source information fusion. This often results in emergency reinforcement measures being initiated only after the track geometry has approached or even exceeded the safety threshold. The response window left for grouting compensation or jacking parameter adjustment is extremely limited, which can easily lead to consequences such as railway speed restrictions and train interruptions. This has become a technical bottleneck restricting the safe and efficient implementation of underpass projects for operating railways. Summary of the Invention

[0004] This invention provides a method for monitoring the jacking support and settlement construction of box culverts under normal railway operation, which improves the safety assurance capability and construction efficiency of railway underpass projects.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for monitoring the jacking support and settlement construction of box culverts during normal railway operation, the method comprising: Step 1: Obtain multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation and the current real-time construction control parameters of the box culvert jacking; perform adaptive filtering and noise reduction and timestamp alignment processing on the multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset; Step 2: Based on the high-fidelity synchronous monitoring dataset, extract the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system. Map the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system into spatial geometric vectors in a three-dimensional orthogonal coordinate system, and extend them along the time axis to construct a multi-dimensional spatiotemporal cube feature space. Step 3: Calculate the volume change rate and centroid offset distance of the projection trajectory of the spatial geometric vector in the feature space of the multidimensional spatiotemporal cube to obtain the comprehensive safety state coupling index. Step 4: Input the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory. Step 5: Dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result; Step 6: When the settlement early warning judgment result is a triggered early warning, based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index, the support deformation compensation requirement of the current support system is calculated in reverse to obtain the adaptive monitoring and control command.

[0006] Secondly, a monitoring system for box culvert jacking support and settlement construction during normal railway operation includes: The acquisition module is used to acquire multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation and the current real-time construction control parameters of the box culvert jacking; it performs adaptive filtering and noise reduction and timestamp alignment processing on the multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset. The mapping module is used to extract the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system based on the high-fidelity synchronous monitoring dataset. The dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system are mapped into spatial geometric vectors in a three-dimensional orthogonal coordinate system, and a multi-dimensional spatiotemporal cube feature space is constructed by extending along the time axis. The calculation module is used to calculate the rate of change of the volume of the projection trajectory of the spatial geometric vector in the feature space of the multidimensional spatiotemporal cube and the centroid offset distance, so as to obtain the comprehensive safety state coupling index. The training module is used to input the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory. The comparison module is used to dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result; The determination module is used to calculate the support deformation compensation requirements of the current support system based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index when the settlement early warning determination result is a triggered early warning, and obtain adaptive monitoring and control instructions.

[0007] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0008] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0009] The above-described solution of the present invention has at least the following beneficial effects: This invention systematically solves four core technical challenges in monitoring the jacking construction of box culverts under operating railways: isolated multi-source data, inability to quantify safety status as a whole, passive and delayed settlement early warning, and lack of quantitative basis for control decisions. It adopts a full-chain technical path of multi-source synchronous preprocessing, multi-dimensional spatiotemporal feature fusion, coupled index quantitative evaluation, settlement trend prediction and early warning, and adaptive control. It constructs a closed-loop intelligent monitoring and control system covering the entire process of perception, characterization, evaluation, prediction, early warning, and control, and realizes a dual leap from passive alarm to active prediction and from experience-based control to adaptive decision-making. This reduces the safety risks of railway speed limits or train interruptions caused by track deformation exceeding limits induced by construction. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a method for monitoring the jacking support and settlement construction of box culverts under normal railway operation, provided by an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of a railway culvert jacking support and settlement monitoring system under normal operation, provided by an embodiment of the present invention. Detailed Implementation

[0012] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0013] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a method for monitoring the jacking support and settlement construction of box culverts under normal railway operation. The method includes the following steps: Step 1: Obtain multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation and the current real-time construction control parameters of the box culvert jacking; perform adaptive filtering and noise reduction and timestamp alignment processing on the multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset; Step 2: Based on the high-fidelity synchronous monitoring dataset, extract the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system. Map the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system into spatial geometric vectors in a three-dimensional orthogonal coordinate system, and extend them along the time axis to construct a multi-dimensional spatiotemporal cube feature space. Step 3: Calculate the volume change rate and centroid offset distance of the projection trajectory of the spatial geometric vector in the feature space of the multidimensional spatiotemporal cube to obtain the comprehensive safety state coupling index. Step 4: Input the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory. Step 5: Dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result; Step 6: When the settlement early warning judgment result is a triggered early warning, based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index, the support deformation compensation requirement of the current support system is calculated in reverse to obtain the adaptive monitoring and control command.

[0014] In this embodiment of the invention, noise interference and timing deviations from multi-source data are eliminated by adaptive filtering and timestamp alignment, resulting in a high-fidelity synchronous data foundation. Through three-dimensional orthogonal coordinate system mapping and spatiotemporal cube construction, a deep fusion expression of three types of heterogeneous data—track settlement, box culvert attitude, and support stress—is achieved. By calculating the volume change rate of the projected trajectory and the centroid offset distance, a comprehensive safety state coupling index that can quantify the evolution law of multi-factor coupling is obtained. Through a settlement trend prediction model jointly driven by the coupling index and construction parameters, proactive prediction before settlement exceeds the limit is achieved, significantly extending the control response window. By inversely calculating the prediction deviation and the coupling index, quantitative adaptive control commands are generated, forming a closed-loop monitoring and control system, thereby improving the construction safety assurance capability and efficiency.

[0015] In a preferred embodiment of the present invention, step 1 above may include: Step 11: Based on the multi-source monitoring time-series data, extract local vibration waveform segments under train dynamic load; based on the extreme value distribution of the local vibration waveform segments, calculate the geometric variance threshold within the sliding window. Specifically, this includes: extracting local vibration waveform segments under train dynamic load based on the multi-source monitoring time-series data; performing short-time energy detection on the track vertical acceleration response signal in the multi-source monitoring time-series data; and setting the frame length of the short-time energy detection to [value missing]. Frame, frame shift The sampling point, the first Short-time energy of a frame Calculate using the following formula: ; In the formula, For the first The short-time energy value of the frame signal, in units of ; The frame length is 256 sampling points. For the first The acceleration amplitude at each sampling point, in units of ; The frame number has a value of According to short-time energy Set the energy trigger threshold to , The value is taken as 3 times the average short-time energy of the background noise. At that time, the judgment of the first The frame contains the train's dynamic load vibration signal, and the signal extends forward from this frame as the starting boundary. Frame, backward expansion A frame is a segment of local vibration waveforms corresponding to a single train passing event. The value is 20 frames. The value is 30 frames. Within a local vibration waveform segment, the sequence of local maxima of the signal is extracted. and local minimum sequence The criterion for determining local extrema is: when and hour, It is a local maximum; when and hour, To find local minima, merge the local maxima sequence with the local minima sequence, and take the absolute values ​​of each extreme value to form the set of extreme amplitudes of the local vibration waveform: ; In the formula, It is the set of extreme value amplitudes; The number of local maxima; This represents the number of local minima. For the first The absolute value of the amplitude at each extreme point, in units of Based on the set of extreme amplitudes of local vibration waveform segments Calculate the geometric variance threshold within the sliding window, and define the half-width of the sliding window as... One sampling point, The value is the sampling frequency. Divide by the train's characteristic frequency The integer part of the quotient, i.e. For sampling frequency Train characteristic frequency The working conditions The value is 100, and the set of extreme amplitudes within the window is: Calculate the arithmetic mean of the extreme values ​​within the window: ; In the formula, For the first The arithmetic mean of the extreme value amplitudes within the window corresponding to each sampling point, in units of ; Half the width of the sliding window; For the first in the window Each extreme value amplitude, in units of Calculate the first Geometric variance threshold corresponding to each sampling point: ; In the formula, For the first The geometric variance threshold corresponding to each sampling point is dimensionless. For the first The squared deviation of the extreme value amplitude from the mean; The mean square is used for normalization to eliminate the influence of the signal amplitude magnitude on the threshold. The reason for using geometric variance instead of arithmetic variance is that geometric variance, through mean square normalization, makes the threshold insensitive to the overall scaling of the signal amplitude, reflects the relative dispersion of the extreme value distribution within the window, and is suitable for scenarios with large differences in the amplitude range of multi-source sensors. Based on the geometric variance threshold, the geometric variance threshold sequence corresponding to each sampling point and the extreme value distribution characteristic parameters of the local vibration waveform segment are obtained.

[0016] Step 12: Based on the geometric variance threshold, perform adaptive median filtering on the multi-source monitoring time series data to obtain preliminary denoised time series data. Based on the preliminary denoised time series data, identify the timestamp breakpoints of data collected by different monitoring devices. Specifically, this includes: performing adaptive median filtering on the multi-source monitoring time series data based on the geometric variance threshold sequence. The filtering window width of the adaptive median filter is dynamically determined by the geometric variance threshold, and its calculation formula is as follows: ; In the formula, Indicates the first The width of the adaptive filtering window for each sampling point is an odd number and is dimensionless. This represents the minimum filter window width, with a value of 3. This represents the maximum filter window width, with a value of 15. Indicates the first The geometric variance threshold corresponding to each sampling point; This represents the upper limit threshold of the geometric variance, with a value of 0.85. When the geometric variance threshold is small, it indicates that the vibration signal near the sampling point is stable, and a smaller window is used to preserve signal details. When the geometric variance threshold is close to or exceeds the upper limit threshold, it indicates that there is strong non-stationary noise in the area, and the window is automatically increased to the maximum window to enhance the denoising effect. Based on the determined filtering window width, a median filtering operation is performed on the original multi-source monitoring time series data, and its calculation formula is as follows: ; In the formula, Indicates the filtered first... The values ​​of each sampling point; Indicates the original data number The values ​​of each sampling point; This represents the median operator, which takes the median value of all values ​​in the window after sorting. The integer part representing the half-width of the window is processed by adaptive median filtering. In areas with strong vibration and noise, the filtering window is automatically increased to effectively suppress non-stationary noise; in areas with stable signals, the filtering window is decreased to preserve detailed features, resulting in preliminary denoised time-series data. Based on this preliminary denoised time-series data, the timestamp breakpoints of data collected by different monitoring devices are identified. For the preliminary denoised time-series data of each monitoring channel, the time interval between adjacent sampling points is calculated using the following formula: ; In the formula, Indicates the first The and the first The time interval between sampling points is in seconds; Indicates the first Timestamp of each sampling point; Indicates the first The timestamp of each sampling point is set, and the sampling period benchmark is set to the maximum value of the nominal sampling period of each monitoring channel, which is 0.5 seconds. The breakpoint judgment coefficient is set to 1.8. When the time interval between adjacent sampling points exceeds the product of the breakpoint judgment coefficient and the sampling period benchmark, it is determined that there is a timestamp breakpoint between the two sampling points. All sampling points of all monitoring channels are traversed, and the start and end timestamps of all breakpoints and their corresponding monitoring channel identifiers are recorded to construct a timestamp breakpoint set.

[0017] Step 13: Based on the timestamp breakpoints, numerical imputation is performed on the missing time series nodes using a preset piecewise linear interpolation rule to obtain the imputed time series nodes. Based on the imputed time series nodes, the time base of each monitoring data is unified to obtain a high-fidelity synchronous monitoring dataset. Specifically, this includes: based on the timestamp breakpoint set, numerical imputation is performed on the missing time series nodes using a preset piecewise linear interpolation rule. For each breakpoint interval in the breakpoint set, the imputed value for any target time within that interval is calculated according to the following formula: ; In the formula, Indicates the target time The dimensions of the supplementary values ​​are consistent with the dimensions of the data from the corresponding monitoring channel. This represents the known denoising value at the start time of the breakpoint interval; The known denoised value represents the time when the breakpoint interval ends; Indicates the start timestamp of the breakpoint interval, in seconds; Indicates the end timestamp of the breakpoint interval, in seconds; Indicates the time when the target is filled, satisfying The unit is seconds. With a sampling period of 0.5 seconds as the step size, the fill value for each intermediate time point within the breakpoint interval is calculated sequentially until all sampling nodes within the breakpoint interval have obtained fill values, resulting in a filled time series node sequence. Based on the filled time series nodes, the time reference for all monitoring data is unified. Global Navigation Satellite System (GNSS) time synchronization is adopted as the unified time reference. The timestamps of the filled time series nodes for each monitoring channel are mapped from local clock records to the unified timestamp, and the mapping relationship is as follows: ; In the formula, Represents a unified timestamp, expressed in the Global Positioning System week-second format; This indicates the timestamp recorded by the local clock of the monitoring device; This represents the fixed clock deviation of the monitoring equipment's local clock relative to a unified time reference, measured in seconds. It is obtained by performing clock synchronization calibration on each monitoring device before construction. After unifying the time reference, the supplemented data from each channel are aligned and arranged according to a unified timestamp, constructing a two-dimensional data matrix with the unified timestamp as the row index and the monitoring channel identifier as the column index. In this two-dimensional data matrix, the... The row corresponds to the first The unified timestamp node, the first Column corresponding to the first There are 10 monitoring channels, and the matrix elements are the monitoring values ​​of the corresponding channels at the corresponding times. This two-dimensional data matrix is ​​defined as a high-fidelity synchronous monitoring dataset.

[0018] In this embodiment of the invention, targeted techniques such as adaptive median filtering based on geometric variance threshold of local vibration waveform extreme value distribution under train dynamic load, timestamp breakpoint identification, and piecewise linear interpolation filling are used to effectively overcome the problems of poor monitoring data quality and inconsistent time references among multiple devices under strong vibration interference in operating railways. In the feature expression stage, deep fusion of three types of heterogeneous data—track settlement, box culvert attitude, and support stress—is achieved through three-dimensional orthogonal coordinate system mapping and spatiotemporal cube construction. In the evaluation and prediction stage, a leap from passive alarm to active prediction is achieved through the calculation of coupled indexes of the rate of change of projected trajectory volume and centroid offset distance, driven by a pre-trained model. In the control stage, adaptive instructions are generated by reverse calculation of prediction deviation, constructing a closed-loop intelligent monitoring and control system covering the entire process of perception, characterization, evaluation, prediction, early warning, and control, thereby improving the safety assurance capability and construction efficiency of box culvert jacking projects under operating railways.

[0019] In a preferred embodiment of the present invention, step 2 above may include: Step 21: Based on the high-fidelity synchronous monitoring dataset, perform curvature integral calculation on the track geometric deformation data to extract the curvature extrema points of track alignment changes and obtain the dynamic characteristics of track settlement. Specifically, this includes: extracting a subset of track geometric deformation data from the high-fidelity synchronous monitoring dataset, including the distribution sequence of track surface elevation along the mileage direction, and performing curvature integral calculation on the track surface elevation sequence. Using mileage as the independent variable and track surface elevation as the dependent variable, the first and second derivatives are calculated using the three-point central difference scheme. The formula for calculating the first derivative is as follows: ; In the formula, Indicates mileage location The first derivative of the track surface elevation represents the longitudinal slope of the track surface and is dimensionless. Indicates mileage location The elevation value of the rail surface at the specified location, in meters; Indicates mileage location The elevation value of the rail surface at the specified location, in meters; The distance between mileage samples is represented by a value of 1.0 meter. The formula for calculating the second derivative is as follows: ; In the formula, Indicates mileage location The second derivative of the track surface elevation represents the rate of change of track surface curvature, expressed in units of meters (1 / m). Indicates mileage location The track surface elevation value, in meters, is used to calculate the curvature value at each mileage location using the first and second derivatives. The calculation formula is as follows: ; In the formula, Indicates mileage location The curvature value of the track surface profile, in meters; This represents the absolute value of the second derivative, ensuring that the curvature is non-negative; The normalization factor for the curvature formula is used to account for the influence of the longitudinal slope on the curvature calculation, and to extract the curvature extrema from the curvature sequence. The first difference of the curvature sequence is calculated using the following formula: ; In the formula, Indicates curvature at kilometer position The first-order difference at the point, in units of meters; The curvature value represents the next mileage position. When the first difference of curvature changes sign at a certain position, the position is determined to be a curvature extremum point. The mileage position, curvature value, and corresponding elevation value of the curvature extremum point are recorded to construct a set of track curvature extremum points. The set of track curvature extremum points contains the spatial distribution information of the most significant track alignment deformation positions. The larger the curvature value, the more severe the track bending at that position and the higher the risk of settlement. The dynamic characteristics of track settlement are obtained, including the set of curvature extremum points and the complete track surface elevation sequence and curvature sequence.

[0020] Step 22: Based on the high-fidelity synchronous monitoring dataset, perform rigid body transformation decomposition on the box culvert jacking attitude data to extract the pitch, yaw, and roll angle change sequences of the box culvert in three-dimensional space, obtaining the jacking attitude evolution characteristics. Specifically, this includes: extracting a subset of box culvert jacking attitude data from the high-fidelity synchronous monitoring dataset, including three-dimensional displacement monitoring data at the four corners of the box culvert; performing rigid body transformation decomposition on the box culvert jacking attitude data; defining an initial reference coordinate system for the box culvert; this coordinate system uses the designed jacking direction of the box culvert as the first coordinate axis, the horizontal direction perpendicular to the jacking direction as the second coordinate axis, and the vertical upward direction as the third coordinate axis. The spatial attitude change of the box culvert during the jacking process is described by translation vectors and rotation matrices. The rotation matrix is ​​decomposed according to the Euler angle rotation order of the third coordinate axis, the second coordinate axis, and the first coordinate axis, that is, first rotating around the third coordinate axis, then around the second coordinate axis, and finally around the first coordinate axis. The expression of the rotation matrix is ​​as follows: ; In the formula, The rotation matrix of the box culvert is a 3x3 orthogonal matrix; Indicates rotation about the third coordinate axis The rotation component matrix of the angle; Indicates rotation about the second coordinate axis The rotation component matrix of the angle; Indicates rotation about the first coordinate axis The rotation component matrices of the angle are as follows: ; ; ; In the formula, This represents the rotation angle of the box culvert around the first coordinate axis, i.e., the pitch angle, in degrees. A positive value indicates that the front end of the box culvert is raised, and the value range is usually from 3° to +3°. This represents the rotation angle of the box culvert around the second coordinate axis, i.e., the yaw angle, in degrees. A positive value indicates that the front end of the box culvert yaws to the right, and the value range is usually from 5° to +5°. This represents the rotation angle of the box culvert around the third coordinate axis, i.e., the roll angle, in degrees. A positive value indicates that the right side of the box culvert has subsided, and the value range is usually from 2° to +2°. Based on the coordinate changes of the monitoring points at the four corners of the box culvert, the attitude parameters at each time moment are solved using the least squares method. Let the first... If the coordinates of each monitoring point at the initial reference time are the initial coordinate vectors, and the coordinates at the current time are the current coordinate vectors, then the rigid body transformation equation is: ; In the formula, Indicates the first Each monitoring point at the current time The coordinate vector, in meters; This represents the rotation matrix determined by the pitch angle, yaw angle, and roll angle; Indicates the first The coordinate vectors of each monitoring point at the initial reference time, in meters; Indicates the current time The translation vector, in meters, contains translational components in three directions. The four monitoring points at the four corners of the box culvert provide 12 scalar equations, with 6 parameters to be solved. The rotation angle sequence is arranged in time, and the pitch angle change sequence, yaw angle change sequence, and roll angle change sequence of the box culvert in three-dimensional space are extracted to form a complete evolution description of the box culvert's jacking attitude. The evolution characteristics of the jacking attitude are obtained, including the three-dimensional rotation angle change sequence and the translation vector sequence.

[0021] Step 23: Based on the high-fidelity synchronous monitoring dataset, perform contour topology analysis on the deep stress and deformation data of the current support system to extract the closed topological loop boundaries of the stress concentration region and obtain the stress-strain characteristics of the support structure. Specifically, this includes: extracting a subset of the deep stress and deformation data of the current support system from the high-fidelity synchronous monitoring dataset, including multi-dimensional stress-strain monitoring data such as the deep horizontal displacement of the support piles, the axial force of the steel supports, and the surface strain of the retaining structure; performing contour topology analysis on the deep stress and deformation data of the support system to construct a two-dimensional scalar field of the stress distribution of the support system on the horizontal profile; and constructing a continuous stress field on the time cross-section based on the known stress values ​​of a finite number of monitoring points using the Kriging space interpolation method. Assume there are a total of... There are 1, whose coordinates are respectively The measured stress value is The stress estimate at any interpolation target point is calculated using the following formula: ; In the formula, Indicates the interpolation target point The estimated stress at the location is expressed in kilopascals. This indicates the number of known monitoring points involved in the interpolation; Indicates the first The kriging interpolation weighting coefficients of each monitoring point to the target point are dimensionless. Indicates the first The measured stress values ​​at each monitoring point are in kPa. The Kriging weighting coefficients are obtained by solving a system of linear equations constrained by a semivariogram, which uses a spherical model and is expressed as follows: ; In the formula, Represents the semivariogram value, characterizing the spatial distance. The semivariance of the stress values ​​between two points; This indicates spatial distance, expressed in meters. This represents the value of the nugget, with a value of 0.02. This represents the base value, which is 1.0. The range is 5.0 meters. On the constructed continuous stress field, a sequence of stress contour values ​​is defined, increasing in increments of 10 kPa at equal stress intervals. For each value, a contour path satisfying the stress value equal to that value is traced on a two-dimensional regular grid. The connection relationships of contour segments are extracted to form a contour family. Closed topological loop boundaries of stress concentration regions are identified from the contour family. For each closed contour line, its enclosed area and internal average stress value are calculated. The enclosed area of ​​the closed contour line is calculated using Green's formula. ; In the formula, Indicates closed contour lines The area enclosed by [the symbol], in square meters; Indicates along closed contour lines The loop integral, where , The coordinates of points on the contour line are given. The internal average stress value is the arithmetic mean of the stress values ​​of all grid nodes within the area enclosed by the closed contour line. The upper limit threshold for the closed loop area is set to 2.0 square meters, and the stress concentration discrimination threshold is set to 60% of the yield strength of the corresponding support structure material. When a closed contour line simultaneously satisfies the conditions that the enclosed area is less than the upper limit threshold and the internal average stress value is greater than the stress concentration discrimination threshold, the closed contour line is determined to be the boundary of the closed topological loop of the stress concentration region. The boundary coordinate sequence of all closed topological loops and the stress statistical characteristics of their enclosed areas are recorded to constitute the stress-strain characteristics of the support structure, including the set of closed topological loop boundaries and the continuous stress field distribution.

[0022] Step 24: Based on the dynamic characteristics of track settlement, the evolution characteristics of jacking attitude, and the stress-strain characteristics of the support structure, their values ​​are mapped to initial spatial geometric vectors in a three-dimensional orthogonal coordinate system. Specifically, this includes: based on the dynamic characteristics of track settlement, the evolution characteristics of jacking attitude, and the stress-strain characteristics of the support structure, their values ​​are mapped to initial spatial geometric vectors in a three-dimensional orthogonal coordinate system. A three-dimensional orthogonal coordinate system is defined, with its three coordinate axes as follows: the first coordinate axis along the track settlement dimension, the second coordinate axis along the jacking attitude dimension, and the third coordinate axis along the support stress-strain dimension. The three coordinate axes are orthogonal to each other, forming a complete three-dimensional feature expression space. Three quantitative indicators are extracted from the dynamic characteristics of track settlement. The first quantitative indicator is the maximum value of the track curvature extremum, representing the severity of the track surface deflection at the current moment, measured in meters. The second quantitative indicator is the maximum track surface elevation settlement, representing the maximum subsidence of the track surface relative to the initial reference plane at the current moment, measured in millimeters. The third quantitative indicator is the distribution density of curvature extremum points along the track, defined as the ratio of the number of curvature extremum points within a mileage interval to the length of the interval, calculated using the following formula: ; In the formula, Indicates the current time The distribution density of curvature extrema points along the path, in units of points per meter; This represents the total number of curvature extrema detected within the monitored mileage interval at the current moment; The total length of the monitoring mileage interval is represented by 100 meters. The three quantitative indicators are mapped to the first spatial geometric vector in a three-dimensional orthogonal coordinate system. The three quantitative indicators are extracted from the jacking attitude evolution characteristics: pitch angle deviation (the absolute value of the difference between the current pitch angle and the design pitch angle, with the design pitch angle set at 0.5 degrees), yaw angle deviation (the absolute value of the difference between the current yaw angle and the design yaw angle, with the design yaw angle set at 0 degrees), and roll angle deviation (the absolute value of the difference between the current roll angle and the design roll angle, with the design roll angle set at 0 degrees). The unit of all three is degrees.

[0023] The three quantitative indicators are mapped to a second spatial geometric vector in a three-dimensional orthogonal coordinate system. Three quantitative indicators are extracted from the stress-strain characteristics of the support structure. The first quantitative indicator is the maximum value of the average stress within the closed topological loop, taken as the maximum value of the average stress within all closed topological loops at the current moment, expressed in kilopascals. The second quantitative indicator is the rate of change of the area enclosed by the closed topological loops, defined as the percentage change in the total area of ​​the closed topological loops at the current moment relative to the total area of ​​the closed topological loops at the initial moment. The third quantitative indicator is the maximum value of the stress field gradient modulus, defined as the maximum value of the gradient modulus at each point in the continuous stress field, calculated using the following formula: ; In the formula, Indicates the current time The maximum value of the stress field gradient modulus, expressed in kilopascals per meter (kPa / m); This represents the partial derivative of the stress field along the first coordinate axis. The partial derivative of the stress field in the direction of the second coordinate axis is represented. The three quantitative indicators are mapped to the third spatial geometric vector in the three-dimensional orthogonal coordinate system. The three spatial geometric vectors represent the construction safety status from three independent dimensions: track settlement, jacking attitude and support stress, respectively. The three initial spatial geometric vectors and their coordinate value sequences in the three-dimensional orthogonal coordinate system are obtained.

[0024] Step 25: Based on the initial spatial geometric vectors, continuously stack and extend them along a unified time axis to construct a multi-dimensional spatiotemporal cube feature space containing both spatial and temporal dimensions. Specifically, this includes: based on three initial spatial geometric vectors, continuously stack and extend them along a unified time axis to construct a multi-dimensional spatiotemporal cube feature space containing both spatial and temporal dimensions. Define a time axis whose positive direction points from the start of construction to the current monitoring time. The time axis is orthogonal to the three spatial axes of the three-dimensional orthogonal coordinate system, forming a four-dimensional spatiotemporal coordinate system. A synchronous sampling period of 0.5 seconds is used as the time step. Within the time interval from the start of construction to the current monitoring time, a total of [number missing] features are defined. At each time step node, the nine coordinate components of the three spatial geometric vectors are arranged column-wise to construct a time series matrix of the spatial geometric vectors. The number of rows in this matrix equals the total number of time steps, and the number of columns equals nine. Each column of the time series matrix is ​​expanded along the time axis to form nine spatial feature trajectory curves. Each trajectory curve describes the continuous path of the corresponding spatial coordinate component as it evolves over time. The nine trajectory curves together enclose a four-dimensional hypercube region in four-dimensional spacetime. The cross-section of this region at any time point is a spatial triangle formed by the endpoints of the three spatial geometric vectors. Extending along the time axis, it forms a spacetime conduit-like geometry, which is defined as the multi-dimensional spacetime cube feature space. The multi-dimensional spacetime cube feature space realizes the fusion expression of three types of features—track settlement, jacking attitude, and support force—within a unified spacetime framework, resulting in the multi-dimensional spacetime cube feature space and its nine spatial feature trajectory curves.

[0025] In this embodiment of the invention, extreme points of track alignment changes are extracted by curvature integral calculation, the three-axis attitude angle sequence of the box culvert is obtained by rigid body transformation decomposition, and the boundary of the closed topological loop of support stress is identified by contour topology analysis. The dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of support structure are accurately extracted from the high-fidelity synchronous dataset. By mapping the three-dimensional orthogonal coordinate system and continuously stacking and extending the time axis, the three types of heterogeneous feature data are uniformly transformed into spatial geometric vectors in the multi-dimensional spatiotemporal cube feature space. This overcomes the technical defects of traditional methods, such as the isolation of the three types of monitoring data (track settlement, box culvert attitude, and support stress), the fragmentation of information, and the difficulty in characterizing the multi-factor coupling relationship between train dynamic load, box culvert jacking, support deformation, and track settlement.

[0026] In a preferred embodiment of the present invention, step 3 above may include: Step 31: Based on the multidimensional spatiotemporal cube feature space, extract the set of three-dimensional coordinate points of spatial geometric vectors within adjacent time steps. Specifically, this includes: extracting the set of three-dimensional coordinate points of spatial geometric vectors within adjacent time steps based on the multidimensional spatiotemporal cube feature space. In the multidimensional spatiotemporal cube feature space, for adjacent time steps with a time step interval of 0.5 seconds, extract the coordinates of three spatial geometric vectors at the start and end points of that time period. The start point coordinate set contains three three-dimensional coordinate points: the coordinates of the track settlement feature vector, the jacking attitude feature vector, and the support stress-strain feature vector at the start point. The end point coordinate set also contains three three-dimensional coordinate points: the coordinates of the three types of feature vectors at the end point. Merge the start point coordinate set and the end point coordinate set to form the set of three-dimensional coordinate points within the adjacent time step, which contains a total of 6 three-dimensional coordinate points. Each coordinate point is a vector in a three-dimensional orthogonal coordinate system, containing 3 coordinate components. Therefore, the set contains a total of 18 scalar values. Traverse all adjacent time steps and extract the set of coordinate points for each time period in turn to construct a sequence of coordinate point sets.

[0027] Step 32: Based on the set of three-dimensional coordinate points, construct the spatial geometric envelope corresponding to adjacent time steps, and calculate the volume difference of the spatial geometric envelope to obtain the rate of change of the projected trajectory volume. Specifically, this includes: Based on the set of 3D coordinate points, a spatial geometric envelope corresponding to adjacent time steps is constructed, and the volume difference between adjacent time steps is calculated to obtain the volume change rate of the projected trajectory. For each time step, six 3D coordinate points are used to calculate the minimum convex hull using a fast 3D convex hull algorithm. The algorithm's steps are as follows: First, select points with maximum and minimum values ​​in three coordinate directions from the six coordinate points, and select four points that form the largest volume tetrahedron as the initial convex hull. Second, assign each remaining point that is not inside the current convex hull to its visible outer facet. Third, for each outer facet with assigned points, select the point farthest from the facet's visible point set as a new vertex, delete all existing faces visible from that vertex, and connect the vertex to the deleted boundary edge to form a new facet. Repeat the above steps until all points are located inside or on the convex hull surface to obtain the spatial geometric envelope. Based on the spatial geometric envelope, calculate the directed volume of the spatial geometric envelope. The surface of the spatial geometric envelope is composed of several triangular facets. The formula for calculating the volume of the envelope is as follows: ; In the formula, Indicates the first The volume of the spatial geometric envelope at each time step is expressed in cubic units of coordinate units. This represents the total number of triangular facets on the surface of the envelope. Indicates the first The coordinate vector of the first vertex of a triangular facet; Indicates the first The coordinate vector of the second vertex of each triangular facet; Indicates the first The coordinate vector of the third vertex of each triangular facet; construct spatial geometric envelopes for each of two adjacent time steps, calculate their respective volumes, and calculate the rate of change of the projected trajectory volume, the formula of which is as follows: ; In the formula, Indicates the first The time step to the The rate of change of the projected trajectory volume over each time step, expressed as a percentage; Indicates the first The volume of the envelope at each time step; Indicates the first The volume of the envelope at each time step; This represents the absolute value of the volume difference between adjacent envelopes. When the volume of the envelope at the reference time step is zero, the rate of change of the projected trajectory volume is directly set to zero to avoid the division-to-zero anomaly. The rate of change of the projected trajectory volume characterizes the magnitude of change of the comprehensive state of the three types of feature vectors in the multidimensional feature space within a time step. The larger the value of this rate of change, the more drastic the change in the coupling state between track settlement, jacking attitude and support stress occurs in a short time, and the worse the stability of the construction safety state. The projected trajectory volume change rate sequence is obtained.

[0028] Step 33: Based on the set of three-dimensional coordinate points, calculate the coordinates of the mass distribution center of the spatial geometric vector within a preset time window, and solve for the Euclidean distance between the mass distribution center coordinates and the initial reference coordinates to obtain the centroid offset distance. Specifically, this includes: calculating the coordinates of the mass distribution center of the spatial geometric vector within a preset time window based on the set of three-dimensional coordinate points, solving for the Euclidean distance between the mass distribution center coordinates and the initial reference coordinates to obtain the centroid offset distance, defining a preset time window with a width of 20 time steps, corresponding to a physical time length of... Seconds, with the first The window starts at the nth time step, and the time window covers from the nth time step. One to the first A time interval of length is defined. Within this time window, the spatial geometric vector coordinates of all moments are extracted, forming a set of coordinate points within the window. This set contains a total of _____. Given a set of three-dimensional coordinate points, calculate the coordinates of the center of mass distribution for all coordinate points within the window. The center of mass distribution is the arithmetic mean of the positions of all coordinate points within the window, and its calculation formula is as follows: ; In the formula, Indicates the first The coordinates of the center of mass distribution within a preset time window starting from a time step are a three-dimensional vector with dimensions consistent with the three-dimensional orthogonal coordinate system. This represents the number of time steps within the time window, with a value of 20. This represents each three-dimensional coordinate point in the set of coordinate points within the window; Indicates the first A set of coordinate points within a window with a time step as the starting point is obtained, and the initial reference coordinates are acquired simultaneously. The initial reference coordinates are taken as the coordinates of the mass distribution center of the three-dimensional coordinate point set corresponding to the construction start time, that is, the arithmetic mean position of the three spatial geometric vectors at the construction start time, which represents the safety state reference point at the start of construction. The Euclidean distance between the mass distribution center coordinates and the initial reference coordinates is calculated to obtain the centroid offset distance, and the calculation formula is as follows: ; In the formula, Indicates the first The centroid offset distance within a preset time window starting from a time step, with dimensions consistent with the three-dimensional orthogonal coordinate system; Indicates the first The coordinates of the center of mass distribution within a preset time window starting from a time step are a three-dimensional vector with dimensions consistent with the three-dimensional orthogonal coordinate system. Represents the initial reference coordinates, whose components are: , , ; The vector magnitude symbol is represented by the centroid offset distance, which quantitatively describes the overall cumulative drift in the three-dimensional feature space. The larger the value of this offset distance, the further the combined state of track settlement, jacking attitude, and support stress deviates from the initial safety benchmark, and the more significant the degradation of the safety state during construction. The windows are slid sequentially along the time axis with a sliding step of 1 time step, and the centroid offset distance at the starting point of each window is calculated to obtain a complete centroid offset distance sequence.

[0029] Step 34: Based on the projected trajectory volume change rate and centroid offset distance, calculate the dynamic coupling weights between each feature vector using a preset multi-dimensional spatial distance weighted fusion rule to obtain a comprehensive safety state coupling index. Specifically, this includes: calculating the dynamic coupling weights between each feature vector based on the projected trajectory volume change rate sequence and centroid offset distance sequence using a preset multi-dimensional spatial distance weighted fusion rule to obtain a comprehensive safety state coupling index; and normalizing the projected trajectory volume change rate and centroid offset distance to eliminate differences in their numerical dimensions and magnitudes. The formula for calculating the normalized projected trajectory volume change rate is as follows: ; In the formula, This represents the rate of change of the normalized projected trajectory volume, with a value ranging from 0 to 1, and is dimensionless. Indicates the first The rate of change of the volume of the original projected trajectory at each time step; This represents the maximum value in the volume change rate sequence after removing the first 1% extreme values. It is used to suppress the interference of outliers on normalization. The formula for calculating the centroid offset distance normalization is as follows: ; In the formula, This represents the normalized centroid offset distance, with a value ranging from 0 to 1, and is dimensionless. Indicates the first The original centroid offset distance for each time step; This represents the maximum value in the centroid offset distance sequence after removing the top 1% extreme values. Since the length of the centroid offset distance sequence is less than the length of the volume change rate sequence, the common time index range of the two sequences is used during the fusion calculation to ensure that the two indicators are time-aligned. The dynamic coupling weights of each feature dimension are calculated, and the standardized offset distance of each dimension at the current moment is defined. The formula for calculating the standardized offset distance of the orbital settlement dimension is as follows: ; The formula for calculating the standardized offset distance in the jacking attitude dimension is as follows: ; The standardized offset distance calculation formula for the support stress-strain dimension is as follows: ; In the formula, Indicates the orbital settlement dimension at the th Standardized offset distance for each time step, dimensionless; Indicates the jacking attitude dimension at the 1st position. Standardized offset distance for each time step, dimensionless; Indicates the support stress-strain dimension in the first place. Standardized offset distance for each time step, dimensionless; Indicates the current time The spatial geometric vector of orbital settlement; This represents the initial spatial geometry vector of the track settlement. Based on the standardized offset distances of the three dimensions, the dynamic coupling weights for each dimension are calculated. The formula for calculating the dynamic coupling weights of the track settlement dimension is as follows: ; The formula for calculating the dynamic coupling weights in the jacking attitude dimension is as follows: ; The formula for calculating the dynamic coupling weights in the stress-strain dimension of the support is as follows: ; In the formula, The dynamic coupling weights representing the orbital settlement dimension are dimensionless. Represents the dynamic coupling weights in the jacking attitude dimension, which are dimensionless; The dynamic coupling weights, representing the stress-strain dimension of the support structure, are dimensionless. Represents a very small positive number, with a value of This is used to prevent the denominator from being zero; the sum of the three coupling weights is approximately equal to 1. The physical meaning of the dynamic coupling weight is: the larger the standardized offset distance of a certain dimension, the more serious the deviation of the state of that dimension from the initial baseline, and the larger its corresponding coupling weight, so that the anomaly of that dimension is reflected more prominently in the comprehensive index. The normalized projected trajectory volume change rate and the centroid offset distance are weighted and fused according to the dynamic coupling weight to obtain the comprehensive safety state coupling index, the calculation formula of which is as follows: ; In the formula, For the first The comprehensive safety status coupling index has a time step and a value range of 0 to 1. The weighted contribution of the orbital settlement dimension to the instantaneous rate of change; This represents the weighted contribution of the jacking attitude dimension to the cumulative offset. The term represents the weighted coupling term of the support stress-strain dimension with respect to the instantaneous rate of change and the geometric mean of the cumulative offset. The judgment rule for the comprehensive safety state coupling index is as follows: when the index approaches 0, it indicates that the coupling state of the three types of features is close to the initial safety benchmark, and the construction safety state is good; when the index approaches 1, it indicates that the coupling state of the three types of features deviates significantly from the initial safety benchmark, and the construction safety risk increases significantly; when the index is in the middle range of 0.4 to 0.7, it indicates that the safety state is in the transitional degradation stage, thus obtaining the comprehensive safety state coupling index sequence.

[0030] In this embodiment of the invention, a set of three-dimensional coordinate points of spatial geometric vectors at adjacent time steps is extracted from the feature space of a multi-dimensional spatiotemporal cube. A spatial geometric envelope is constructed, and the volume difference between adjacent time steps is calculated to obtain the volume change rate of the projected trajectory. At the same time, the Euclidean distance between the coordinates of the mass distribution center and the initial reference coordinates within a preset time window is calculated to obtain the centroid offset distance. The dynamic coupling weights between each feature vector are solved by a preset multi-dimensional spatial distance weighted fusion rule to obtain a comprehensive safety state coupling index. This method quantifies the overall evolution of three types of features—track settlement, box culvert attitude, and support stress—in the spatiotemporal cube into two core metrics: volume change rate and centroid offset distance. This overcomes the shortcomings of existing technologies that lack multi-dimensional overall quantitative assessment of construction safety status. The contribution ratio of each feature vector is adaptively adjusted through dynamic coupling weights.

[0031] In a preferred embodiment of the present invention, step 4 above may include: Step 41: Based on the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking, feature dimensions are spliced ​​to construct a multi-dimensional feature fusion input vector. Specifically, this includes: based on the comprehensive safety status coupling index sequence and the current real-time construction control parameters of the box culvert jacking obtained from the construction control system, feature dimensions are spliced ​​to construct a multi-dimensional feature fusion input vector. The current real-time construction control parameters of the box culvert jacking are read in real time from the box culvert jacking construction control system with a synchronous sampling period of 0.5 seconds, including the following four types of parameters: The first type of parameter is the jacking speed: defined as the speed along the jacking direction of the front face of the box culvert. The first parameter is the instantaneous thrust rate in the direction of the jacking, measured in millimeters per minute, which is measured in real time by the stroke sensor of the jacking jack. The second parameter is the total thrust of the jacking jack: defined as the sum of the thrust of all the jacking jack cylinders, measured in kilonewtons, which is calculated by multiplying the oil pressure measured by the oil pressure sensor by the effective area of ​​the piston. The third parameter is the grouting pressure of the support: defined as the grout pressure at the outlet of the grouting hole between the outer wall of the box culvert and the soil, measured in kilopascals, which is measured in real time by the pressure transmitter on the grouting pipeline. The fourth parameter is the grouting volume of the support: defined as the cumulative grouting volume from the start of construction to the current moment, measured in cubic meters.

[0032] The above four types of construction control parameters and comprehensive safety status coupling indicators are concatenated at the corresponding time steps to form a multi-dimensional feature fusion input vector containing five feature components. To eliminate the dimensional differences between the feature components, each component of the multi-dimensional feature fusion input vector is standardized. For each feature component, its original value is subtracted from the mean of that component in the training dataset, and then divided by the standard deviation of that component in the training dataset to obtain the standardized feature value. This process is repeated for all time steps to construct a complete sequence of multi-dimensional feature fusion input vectors.

[0033] Step 42: Based on the multi-dimensional feature fusion input vector, input it into the pre-trained settlement trend prediction model to extract the temporal evolution law and obtain the initial settlement prediction sequence. Specifically, this includes: inputting the multi-dimensional feature fusion input vector sequence into the pre-trained settlement trend prediction model to extract the temporal evolution law and obtain the initial settlement prediction sequence. The pre-trained settlement trend prediction model adopts a bidirectional long short-term memory network architecture based on an attention mechanism. This network architecture consists of four levels connected in series: an input layer, a bidirectional long short-term memory encoding layer, a multi-head self-attention layer, and a fully connected output layer. The first level is the input layer: the input layer receives the multi-dimensional feature fusion input vector sequence within a sliding time window with a preset sequence length of 64 time steps, corresponding to a physical time length of [missing information]. Seconds, for the first The input with a time step of the window ending point contains the input sequence from the nth time step. One to the first The first level is a bidirectional long short-term memory (LSM) encoding layer, which integrates all multidimensional features from each time step into the input vector. The second level consists of a forward LSM network and a backward LSM network connected in parallel. The hidden layer dimensions are all 128. The forward LSM network processes the input sequence in forward chronological order, while the backward LSM network processes it in reverse chronological order. The forward and backward hidden states are concatenated at each time step, resulting in an encoded output vector with a dimension of 256. The third level is a multi-head self-attention layer, which integrates all multidimensional features from the bidirectional LSM encoding layer into the input vector. The step output serves as the input to the self-attention mechanism. The multi-head self-attention output is calculated as follows: the input is transformed into a query matrix, key matrix, and value matrix through three different linear transformations; the product of the query matrix and the transpose of the key matrix is ​​calculated, divided by a scaling factor, and then processed by a normalized exponential function to obtain the attention weight matrix; the attention weight matrix is ​​multiplied by the value matrix to obtain the self-attention output. Multi-head self-attention uses eight attention heads computed in parallel. The outputs of each head are concatenated and transformed linearly to obtain the fused attention vector. The calculation formula for the self-attention mechanism is as follows: ; In the formula, The query matrix is ​​formed by transforming the output vector of the bidirectional long short-term memory encoding layer into a first learnable linear transformation weight matrix. Mapped to obtain, The dimension is 256 rows and 64 columns, query matrix The dimension is the sequence length multiplied by 64, and each element is a real number with no dimension. The key matrix is ​​formed by transforming the output vector of the bidirectional long short-term memory coding layer into a weight matrix via a second learnable linear transformation. Mapped to obtain, The dimension is 256 rows and 64 columns, key matrix The dimension is the sequence length multiplied by 64, and each element is a real number with no dimension. The value matrix is ​​formed by transforming the output vector of the bidirectional long short-term memory encoding layer into a third learnable linear transformation weight matrix. Mapped to obtain, The dimension is 256 rows and 64 columns, value matrix The dimension is the sequence length multiplied by 64, and each element is a real number with no dimension. This represents the transpose of the key matrix, with dimensions of 64 multiplied by the sequence length, and each element being a real number; This represents the scaling factor, which is the dimension of the key vector for each attention head, and has a value of 64. Represents the square root of the scaling factor; The exponential function is normalized by row. The fourth level is the fully connected output layer: the fused attention vector is input into the fully connected network, which contains two hidden layers. The first hidden layer has a dimension of 128 and uses the modified linear unit activation function; the second hidden layer has a dimension of 64 and also uses the modified linear unit activation function. The output layer has a dimension of 40, corresponding to the predicted settlement value for the next 40 time steps. The calculation formula for the fully connected output layer is as follows: ; In the formula, Indicates the first The initial settlement prediction sequence corresponds to each time step, with dimension [ ]. The unit is millimeters; Indicates the first A fusion attention vector with a time step; This represents the weight matrix of the first hidden layer, with dimension 1. ; This represents the weight matrix of the second hidden layer, with dimension 1. ; This represents the weight matrix of the output layer, with dimension 1. ; This represents the bias vector of the first hidden layer; This represents the bias vector of the second hidden layer; This represents the bias vector of the output layer; This indicates that the modified linear unit activation function maps negative inputs to zero. The training data for this pre-trained model comes from historical monitoring data of similar railway underpass box culvert jacking projects. The mean square error is used as the loss function, and gradient descent training is performed using an adaptive moment estimator optimizer. The training iterations are 200 rounds, the batch size is 32, and the initial learning rate is 0.001.

[0034] Step 43: Based on the initial settlement prediction sequence, a smoothing process is performed using a preset spline curve extrapolation rule to eliminate local abrupt noise in the sequence, thus obtaining the predicted track settlement trajectory. Specifically, this includes: The preset spline curve extrapolation rule uses cubic basic spline curves for smoothing. These cubic basic spline curves are determined by a set of control points and possess the smooth characteristic that the function values, first derivative, and second derivative are all continuous. This effectively eliminates high-frequency abrupt noise in the sequence while maintaining the overall smoothness of the settlement trend. For the 40 predicted values ​​in the initial settlement prediction sequence, their corresponding prediction time nodes are constructed into a time series. Within the prediction time interval, eight control points are selected at uniform intervals. Each control point corresponds to a control point settlement value to be determined. The control point settlement value is solved by minimizing a weighted objective function, the expression of which is as follows: ; In the formula, This represents the weighted objective function, expressed in square millimeters. Indicating the first [number]th [unit] in the initial settlement prediction sequence One predicted value, in millimeters; This represents the time interval of a cubic basic spline curve. The function value at that location is in millimeters. Indicates the first The fitting weights for each data point are calculated using an exponential decay function, where the weight decreases as time progresses. ,in The value is 1.5. To predict the end point of the time interval; This represents the smoothing penalty coefficient, with a value of 0.05, which controls the balance between fitting accuracy and smoothness. The second derivative of a spline curve is represented by . The curvature penalty term is the integral of the square of the second derivative over the prediction time interval, used to suppress excessive curvature of the spline curve. This represents the total number of predicted values, with a value of 40. The expression for the cubic basic spline curve is determined by the control point settlement values ​​and the basic spline basis functions, and its expression is: ; In the formula, This indicates that the spline curve at time t is The smoothed fit value, in millimeters; Indicates the first Settlement values ​​for each control point, in millimeters; Indicates the first The cubic fundamental spline basis functions at time 1 The value; The total number of control points is 8. By solving the minimization problem of the weighted objective function, the optimal estimate of the control point settlement value is obtained. The control point settlement value is substituted into the spline curve expression, and dense sampling is performed at intervals of 0.1 seconds within the prediction time interval to generate a smooth and continuous track settlement prediction trajectory. After three basic spline smoothing fits, the track settlement prediction trajectory eliminates the local high-frequency abrupt noise points in the initial prediction.

[0035] In this embodiment of the invention, by splicing the comprehensive safety status coupling index with the current real-time construction control parameters of the box culvert jacking, a multi-dimensional feature fusion input vector that simultaneously includes structural safety status and construction driving factors is constructed. This overcomes the shortcomings of traditional prediction methods that rely on a single settlement monitoring sequence and ignore the driving role of construction control parameters on settlement evolution. The fusion vector is then input into a pre-trained settlement trend prediction model to extract the temporal evolution law, enabling a forward-looking inference of the future trend of track settlement. By using a preset spline curve extrapolation rule to smooth the initial prediction sequence, local abrupt noise is eliminated, resulting in a continuous and smooth track settlement prediction trajectory. This achieves a leap from passive monitoring and alarm to active prediction and judgment, significantly extending the response window for safety control.

[0036] In a preferred embodiment of the present invention, step 5 above may include: Step 51: Based on the predicted track settlement trajectory, extract the extreme values ​​of predicted settlement at each time node within a future preset time window. Specifically, this includes: extracting the extreme values ​​of predicted settlement at each time node within a future preset time window based on the predicted track settlement trajectory. The future preset time window is defined as a prediction interval extending backward for a preset duration from the current moment. This preset duration is 120 seconds, covering the settlement prediction range for the next 2 minutes. Within this time window, the predicted settlement values ​​at each time node in the predicted track settlement trajectory constitute a set of predicted settlement values. Extract the extreme values ​​of predicted settlement from this set of predicted settlement values, i.e., the maximum value of predicted settlement within the window, representing the maximum possible settlement deformation of the track during the prediction period. Simultaneously, record the time node corresponding to the extreme value of predicted settlement. The smaller the value of this time node, the shorter the predicted settlement peak will arrive in a shorter time.

[0037] Step 52: Calculate the settlement deviation rate based on the predicted extreme settlement value and the preset track geometry safety threshold. Specifically, this includes: calculating the settlement deviation rate based on the predicted extreme settlement value and the preset track geometry safety threshold. The preset track geometry safety threshold is determined according to the operating railway line grade and train speed grade: For a conventional railway trunk line with a speed of 120 km / h, the track vertical geometry safety threshold (i.e., the allowable deviation of rail surface elevation) is 10 mm. For a high-speed railway with a speed of 200 km / h, the safety threshold is 6 mm. For a high-speed railway with a speed of 300 km / h and above, the safety threshold is 4 mm. Taking a conventional railway trunk line as an example, the safety threshold is 10 mm. The settlement deviation rate is calculated based on the predicted extreme settlement value and the track geometry safety threshold, using the following formula: ; In the formula, For the first The settlement deviation rate over time step, expressed as a percentage; This represents the predicted extreme value of settlement, in millimeters. This represents the safety threshold for track geometry, in millimeters. When the extreme value of the predicted settlement is less than or equal to the safety threshold, the settlement deviation rate is less than or equal to zero, indicating that the predicted settlement is within the safety threshold range and no warning needs to be triggered. When the extreme value of the predicted settlement is greater than the safety threshold, the settlement deviation rate is greater than zero, indicating that the predicted settlement will exceed the safety threshold. The larger the deviation rate, the more serious the exceedance and the higher the urgency of the warning.

[0038] Step 53: Compare the settlement deviation rate with the preset warning trigger threshold. When the settlement deviation rate exceeds the warning trigger threshold, generate a flag to trigger a warning, thus obtaining a settlement warning judgment result. Specifically, this includes: comparing the settlement deviation rate with the preset warning trigger threshold to generate a settlement warning judgment result. The preset warning trigger threshold adopts a graded setting strategy, divided into three levels according to different response levels of construction safety management: Level 1 warning trigger threshold: 60%. When the settlement deviation rate reaches or exceeds 60%, it indicates that the predicted settlement has reached 60% of the safety threshold, triggering a yellow warning, prompting on-site technicians to increase monitoring frequency and verify the status of the support system. Level 2 warning trigger threshold: 85%. When the settlement deviation rate reaches or exceeds 85%, it indicates that the predicted settlement is close to 85% of the safety threshold, triggering an orange warning, prompting on-site technicians to prepare to initiate grouting compensation measures and adjust jacking parameters.

[0039] The third-level warning trigger threshold is set at 100%. When the settlement deviation rate reaches or exceeds 100%, it indicates that the predicted settlement has reached or exceeded the safety threshold, triggering a red warning. This requires an immediate halt to the jacking operation and the initiation of emergency support reinforcement. The warning triggering logic is as follows: the settlement deviation rate is compared sequentially with the third-level warning trigger threshold. When the deviation rate is less than the first-level threshold, a warning flag of 0 is generated, corresponding to a green safety state, and no warning is triggered. When the deviation rate is greater than or equal to the first-level threshold but less than the second-level threshold, a warning flag of 1 is generated, corresponding to a yellow warning level. When the deviation rate is greater than or equal to the second-level threshold but less than the third-level threshold, a warning flag of 2 is generated, corresponding to an orange warning level. When the deviation rate is greater than or equal to the third-level threshold, a warning flag of 3 is generated, corresponding to a red warning level. The warning flag, the corresponding warning level information, the deviation rate, and the extreme value of the predicted settlement are encapsulated into the settlement warning judgment result.

[0040] In a preferred embodiment of the present invention, step 6 above may include: Step 61: Based on the settlement early warning judgment result, when the judgment result is a trigger warning, extract the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index. Specifically, this includes: based on the settlement early warning judgment result, when the warning flag bit in the judgment result is greater than or equal to 1, extract the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index, and read the warning flag bit from the settlement early warning judgment result. When the warning flag bit is 0, it indicates that the current safety status is good, no control is required, the current monitoring cycle ends, and return to step 1 to enter the next monitoring cycle. When the warning flag bit is greater than or equal to 1, trigger the adaptive control process. First, extract the deviation of the track settlement prediction trajectory, that is, the absolute deviation value of the predicted settlement extreme value exceeding the track geometric safety threshold. Its calculation formula is as follows:

[0041] In the formula, Indicates the first The deviation of the predicted trajectory settlement over a time step is expressed in millimeters. A positive value indicates that the predicted settlement exceeds the safety threshold. This represents the predicted extreme value of settlement, in millimeters. The deviation is expressed as a safety threshold for track geometry, in millimeters. When the deviation is non-positive, it is set to 0.1 millimeters to ensure the continuity of the control calculation. At the same time, the coupling index value corresponding to the current time step is extracted from the comprehensive safety state coupling index sequence. This coupling index value reflects the comprehensive coupling deviation of the three dimensions of track settlement, jacking attitude and support force. Its magnitude determines the strength of the control. The extracted deviation and the coupling index value are combined into a control input parameter pair.

[0042] Step 62: Based on the deviation of the predicted track settlement trajectory and the comprehensive safety status coupling index, the required adjustment of the jacking speed and the required grouting compensation for the current support system are calculated using a preset reverse mapping rule to obtain the support deformation compensation requirement. Specifically, this includes: based on the control input parameter pair, the required adjustment of the jacking speed and the required grouting compensation for the current support system are calculated using a preset reverse mapping rule to obtain the support deformation compensation requirement. The preset reverse mapping rule is established based on the mechanical response model of the support system. This mechanical response model describes the suppression effect of changes in jacking speed and grouting compensation on the track settlement deviation. Based on the statistical analysis of monitoring data from a large number of similar projects, the following linearized reverse mapping relationship is established. The mapping formula for the jacking speed adjustment is as follows: ; In the formula, Indicates the first The jacking speed adjustment is measured in millimeters per minute for each time step. A positive value indicates a reduction in the jacking speed to mitigate disturbance to the formation. This represents the jacking speed-settlement sensitivity coefficient, with a value of 2.5 per minute. Its physical meaning is the jacking speed reduction required for each millimeter of settlement deviation. This indicates the amount of track settlement deviation, expressed in millimeters. This represents the coupling enhancement coefficient, which has a value of 1.8 and is dimensionless. It is used to amplify the speed adjustment range under high coupling risk conditions. The comprehensive safety status coupling index is dimensionless, and the mapping formula for the support grouting compensation amount is as follows: ; In the formula, Indicates the first The grouting compensation amount for each time step, in cubic meters, represents the amount of grouting that needs to be added based on the current grouting rate. This represents the grouting-settlement sensitivity coefficient, with a value of 0.15 cubic meters per millimeter. Its physical meaning is the amount of additional grouting required to compensate for each millimeter of settlement deviation. This indicates the amount of track settlement deviation, expressed in millimeters. This represents the baseline grouting compensation coefficient, which is 0.3 and dimensionless. It is used to ensure that there is still a basic amount of grouting compensation when the coupling index is low. The comprehensive safety status coupling index is dimensionless. The design principle of the two mapping formulas is as follows: the adjustment amount of the jacking speed is directly proportional to the settlement deviation. The larger the settlement deviation, the more the jacking speed needs to be reduced to reduce the disturbance to the surrounding rock. At the same time, the coupling index amplifies the speed adjustment amount. The higher the coupling index, the greater the risk of multi-factor coupling, and the more the jacking speed needs to be reduced. The grouting compensation amount is also positively correlated with the settlement deviation amount and increases with the increase of the coupling index. This is because when the risk of multi-factor coupling increases, it is not enough to control the settlement by reducing the jacking speed alone. The grouting amount must be increased simultaneously to fill the stratum voids and compensate for stratum losses. The jacking speed adjustment amount and the support grouting compensation amount are combined to form the support deformation compensation requirement, which includes the jacking speed adjustment amount and the support grouting compensation amount.

[0043] Step 63: Based on the support deformation compensation requirements, a set of control parameters including the suggested adjustment amount of the jacking speed and the suggested grouting compensation amount are encapsulated and generated to obtain adaptive monitoring and control instructions. Specifically, this includes: converting the compensation requirements into absolute control suggested values ​​based on the jacking construction control parameters at the current moment; reading the current jacking speed and current grouting rate from the construction control system; calculating the adjusted suggested jacking speed and suggested grouting rate; the suggested jacking speed is the current actual jacking speed minus the jacking speed adjustment amount; the grouting rate is the current actual grouting rate plus the grouting compensation amount divided by the compensation implementation time, where the compensation implementation time is set to 10 minutes, indicating that the compensation grouting amount is applied uniformly within 10 minutes; applying engineering constraint boundaries to the suggested control values ​​to ensure that the suggested values ​​are within the capacity range of the construction equipment; the constraint boundary for the jacking speed is a lower limit of 5 mm / min to an upper limit of 50 mm / min; the constraint boundary for the grouting rate is a lower limit of 0.01 m / min to an upper limit of 0.5 m / min. When the recommended value exceeds the constraint boundary, the recommended value is clamped to the boundary value. The constrained recommended control value, along with information such as the warning level, deviation, and timestamp, is encapsulated into a structured adaptive monitoring and control instruction. This instruction includes the following fields: instruction generation timestamp, warning level, recommended jacking speed, recommended grouting rate, settlement deviation, and comprehensive safety status coupling index. The adaptive monitoring and control instruction is sent in real time to the box culvert jacking construction control system and grouting control system via on-site industrial Ethernet or wireless data transmission link, automatically executing jacking speed and grouting rate adjustments. Simultaneously, the instruction content is pushed to the on-site monitoring terminal display interface for construction technicians to view and confirm in real time.

[0044] In this embodiment of the invention, by using the settlement early warning judgment result as the trigger condition, when an early warning is triggered, the deviation of the predicted track settlement trajectory and the comprehensive safety status coupling index are automatically extracted. This achieves seamless connection between the early warning signal and the control response, overcoming the shortcomings of existing technologies where the early warning and control links are disconnected and rely on manual experience-based decision-making. Through a preset reverse mapping rule, the settlement deviation and coupling index are quantitatively converted into the jacking speed adjustment and support grouting compensation required by the current support system. This establishes a closed-loop mapping relationship from safety status deviation to construction control parameter correction, solving the problem that traditional control methods lack quantitative reverse derivation basis and rely on experience-based determination of control intensity. Furthermore, by encapsulating the calculation results into a standardized control parameter set containing jacking speed adjustment suggestions and support grouting compensation suggestions, adaptive monitoring and control instructions that can directly guide on-site construction operations are generated. This method realizes a fully automatic closed-loop control link of early warning triggering, deviation extraction, reverse mapping, and instruction generation, enabling the construction monitoring system to have complete autonomous decision-making capabilities from perception, assessment, prediction, early warning to control, shortening the time delay from discovering safety hazards to implementing control measures.

[0045] like Figure 2 As shown, an embodiment of the present invention also provides a monitoring system for the jacking support and settlement construction of box culverts under normal railway operation, comprising: The acquisition module is used to acquire multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation and the current real-time construction control parameters of the box culvert jacking; it performs adaptive filtering and noise reduction and timestamp alignment processing on the multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset. The mapping module is used to extract the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system based on the high-fidelity synchronous monitoring dataset. The dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system are mapped into spatial geometric vectors in a three-dimensional orthogonal coordinate system, and a multi-dimensional spatiotemporal cube feature space is constructed by extending along the time axis. The calculation module is used to calculate the rate of change of the volume of the projection trajectory of the spatial geometric vector in the feature space of the multidimensional spatiotemporal cube and the centroid offset distance, so as to obtain the comprehensive safety state coupling index. The training module is used to input the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory. The comparison module is used to dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result; The determination module is used to calculate the support deformation compensation requirements of the current support system based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index when the settlement early warning determination result is a triggered early warning, and obtain adaptive monitoring and control instructions.

[0046] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0047] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0048] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0049] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring the jacking support and settlement construction of box culverts during normal railway operation, characterized in that... The method includes: Step 1: Obtain multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation and the current real-time construction control parameters of the box culvert jacking; perform adaptive filtering and noise reduction and timestamp alignment processing on the multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset; Step 2: Based on the high-fidelity synchronous monitoring dataset, extract the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system. Map the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system into spatial geometric vectors in a three-dimensional orthogonal coordinate system, and extend them along the time axis to construct a multi-dimensional spatiotemporal cube feature space. Step 3: Calculate the volume change rate and centroid offset distance of the projection trajectory of the spatial geometric vector in the feature space of the multidimensional spatiotemporal cube to obtain the comprehensive safety state coupling index. Step 4: Input the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory. Step 5: Dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result; Step 6: When the settlement early warning judgment result is a triggered early warning, based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index, the support deformation compensation requirement of the current support system is calculated in reverse to obtain the adaptive monitoring and control command.

2. The method for monitoring the jacking support and settlement construction of box culverts under normal railway operation as described in claim 1, characterized in that, Step 1: Obtain multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation conditions, as well as the current real-time construction control parameters for box culvert jacking. Adaptive filtering and denoising, along with timestamp alignment, are performed on the multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset, including: Based on multi-source monitoring time-series data, local vibration waveform segments under train dynamic load are extracted; based on the extreme value distribution of local vibration waveform segments, the geometric variance threshold within the sliding window is calculated. Based on the geometric variance threshold, adaptive median filtering is performed on the multi-source monitoring time series data to obtain preliminary denoised time series data; based on the preliminary denoised time series data, the timestamp breakpoints of data collected by different monitoring devices are identified. Based on timestamp breakpoints, missing time series nodes are numerically filled using a preset piecewise linear interpolation rule to obtain filled time series nodes. Based on the filled time series nodes, the time reference of each monitoring data is unified to obtain a high-fidelity synchronous monitoring dataset.

3. The method for monitoring the jacking support and settlement construction of box culverts under normal railway operation as described in claim 2, characterized in that, Step 2 includes: Based on the high-fidelity synchronous monitoring dataset, curvature integral calculation is performed on the track geometric deformation data to extract the curvature extreme points of track linear changes and obtain the dynamic characteristics of track settlement. Based on the high-fidelity synchronous monitoring dataset, rigid body transformation decomposition was performed on the box culvert jacking attitude data to extract the pitch angle, yaw angle and roll angle change sequences of the box culvert in three-dimensional space, and the jacking attitude evolution characteristics were obtained. Based on the high-fidelity synchronous monitoring dataset, contour topology analysis is performed on the deep stress and deformation data of the current support system to extract the closed topological loop boundary of the stress concentration area and obtain the stress-strain characteristics of the support structure. Based on the dynamic characteristics of track settlement, the evolution characteristics of jacking attitude, and the stress-strain characteristics of the support structure, their numerical values ​​are mapped to initial spatial geometric vectors in a three-dimensional orthogonal coordinate system. Based on the initial spatial geometric vector, the vectors are continuously stacked and extended along a unified time axis to construct a multidimensional spatiotemporal cube feature space containing spatial and temporal dimensions.

4. The method for monitoring the jacking support and settlement construction of box culverts under normal railway operation as described in claim 3, characterized in that, Step 3: Calculate the volume change rate and centroid offset distance of the projected trajectory of the spatial geometric vector in the multidimensional spatiotemporal cube feature space to obtain the comprehensive safety state coupling index, including: Based on the feature space of the multidimensional spatiotemporal cube, extract the set of three-dimensional coordinate points of the spatial geometric vector within adjacent time steps; Based on the set of three-dimensional coordinate points, construct the spatial geometric envelope corresponding to adjacent time steps, calculate the volume difference of the spatial geometric envelope, and obtain the volume change rate of the projected trajectory. Based on the set of three-dimensional coordinate points, calculate the coordinates of the mass distribution center of the spatial geometric vector within a preset time window, and solve the Euclidean distance between the mass distribution center coordinates and the initial reference coordinates to obtain the centroid offset distance; Based on the rate of change of the projected trajectory volume and the centroid offset distance, the dynamic coupling weight between each feature vector is calculated by using a preset multi-dimensional spatial distance weighted fusion rule, and a comprehensive safety state coupling index is obtained.

5. The method for monitoring the jacking support and settlement construction of box culverts under normal railway operation as described in claim 4, characterized in that, Step 4: Input the integrated safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory, including: Based on the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking, feature dimensions are spliced ​​together to construct a multi-dimensional feature fusion input vector; The multi-dimensional feature fusion input vector is input into the pre-trained settlement trend prediction model to extract the temporal evolution law and obtain the initial settlement prediction sequence. Based on the initial settlement prediction sequence, a smoothing fitting process is performed using a preset spline curve extrapolation rule to eliminate local abrupt noise in the sequence and obtain the track settlement prediction trajectory.

6. The method for monitoring the jacking support and settlement construction of box culverts under normal railway operation as described in claim 5, characterized in that, Step 5: Dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result, including: Based on the predicted trajectory of track settlement, extract the extreme values ​​of predicted settlement at each time node within the preset future time window; The settlement deviation rate is calculated based on the predicted extreme settlement value and the preset track geometry safety threshold. The settlement deviation rate is compared with the preset warning trigger threshold. When the settlement deviation rate is greater than the warning trigger threshold, a flag is generated to trigger the warning, and the settlement warning judgment result is obtained.

7. The method for monitoring the jacking support and settlement construction of box culverts under normal railway operation as described in claim 6, characterized in that, Step 6: When the settlement early warning determination result is a triggered early warning, based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index, the support deformation compensation requirement of the current support system is calculated in reverse to obtain adaptive monitoring and control instructions, including: Based on the settlement early warning judgment result, when the judgment result is a trigger warning, the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index are extracted. Based on the deviation of the predicted track settlement trajectory and the comprehensive safety status coupling index, the required adjustment of the jacking speed and the grouting compensation of the current support system are calculated through the preset reverse mapping rules, so as to obtain the support deformation compensation requirements. Based on the requirements for support deformation compensation, a set of control parameters is encapsulated and generated, including the suggested amount of jacking speed adjustment and the suggested amount of support grouting compensation, to obtain adaptive monitoring and control instructions.

8. A monitoring system for the jacking support and settlement construction of box culverts under normal railway operation, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire multi-source monitoring time-series data of the box culvert jacking construction area under normal railway operation conditions, as well as the current real-time construction control parameters of the box culvert jacking. Adaptive filtering and denoising, along with timestamp alignment, are performed on multi-source monitoring time-series data to obtain a high-fidelity synchronous monitoring dataset. The mapping module is used to extract the dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system based on the high-fidelity synchronous monitoring dataset. The dynamic features of track settlement, the evolution features of jacking attitude, and the stress-strain features of the current support system are mapped into spatial geometric vectors in a three-dimensional orthogonal coordinate system, and a multi-dimensional spatiotemporal cube feature space is constructed by extending along the time axis. The calculation module is used to calculate the rate of change of the volume of the projection trajectory of the spatial geometric vector in the feature space of the multidimensional spatiotemporal cube and the centroid offset distance, so as to obtain the comprehensive safety state coupling index. The training module is used to input the comprehensive safety status coupling index and the current real-time construction control parameters of the box culvert jacking into the pre-trained settlement trend prediction model to obtain the track settlement prediction trajectory. The comparison module is used to dynamically compare the predicted track settlement trajectory with the preset track geometry safety threshold to obtain the settlement early warning judgment result; The determination module is used to calculate the support deformation compensation requirements of the current support system based on the deviation of the track settlement prediction trajectory and the comprehensive safety status coupling index when the settlement early warning determination result is a triggered early warning, and obtain adaptive monitoring and control instructions.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images