Urban rail transit deformation monitoring method and system based on three-dimensional laser scanning

Through spectral-enhanced laser scanner and fractional-order differential spectral decomposition technology, combined with tensor representation and topological data analysis, the problem of noise processing and nonlinear deformation recognition in three-dimensional laser scanning technology in monitoring tiny and dynamic deformation and complex environments is solved. It realizes high-precision and high-reliability deformation monitoring of urban rail transit.

CN118500281BActive Publication Date: 2024-12-03JIANGSU TESTING CENT FOR QUALITY OF CONSTR ENG
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410960113.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-12-03
Estimated Expiration
2044-07-17

AI Technical Summary

Technical Problem

Urban rail transit deformation monitoring technology based on three-dimensional laser scanning is difficult to capture instantaneous deformation caused by high-speed running trains in real time, and there are accuracy limitations and noise influences in data processing in complex environments, making it difficult to capture complex nonlinear deformation processes. .

Method used

The spectral-enhanced laser scanner is used to combine fractional differential transformation, singular spectral decomposition and Hankel matrix reconstruction to convert it into a fourth-order tensor for tensor decomposition and feature extraction, and data cleaning is carried out by combining multi-scale morphological filtering and geodesic filtering. , multi-scale deformation warning is performed using fractal dynamic prediction model and topological data analysis.

Benefits of technology

Accurate measurement of tiny and dynamic deformations is achieved, forward-looking monitoring and system sensitivity are improved, potential structural problems and abnormal patterns can be identified early, and the accuracy and reliability of deformation monitoring are significantly improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118500281B_ABST
    Figure CN118500281B_ABST
Patent Text Reader

Abstract

The present invention discloses a deformation monitoring method and system for urban rail transit based on three-dimensional laser scanning, including: obtaining laser data of urban rail transit based on a preconfigured laser scanner to construct an initial data set; calling a preconfigured data cleaning model to process the laser data to obtain a cleaned laser data set; obtaining the cleaned laser data set, and sequentially performing fractional differential transformation, singular spectrum decomposition and Hankel matrix reconstruction to obtain deformation data with an accuracy higher than a preset value; converting the deformation data into a fourth-order tensor form, and performing tensor decomposition and feature extraction to obtain a deformation severity index and abnormal probability distribution data; calling a pre-constructed fractal dimensional dynamics prediction model, and using the deformation severity index and abnormal probability distribution data as input to obtain multi-scale deformation warning data. By combining laser point cloud with spectral materials, deformation caused by small deformations and dynamic factors can be accurately measured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a building deformation monitoring technology, in particular to a method and system for monitoring deformation of urban rail transit based on three-dimensional laser scanning. Background Art

[0002] The deformation monitoring method of urban rail transit based on 3D laser scanning is one of the key technologies to ensure the safe operation of urban rail transit. With the acceleration of urbanization and the rapid expansion of rail transit networks, higher requirements are placed on the accuracy, real-time and reliability of deformation monitoring technology. Traditional measurement methods, such as leveling and total station measurement, are highly accurate but inefficient and cannot meet the needs of large-scale and high-frequency monitoring. 3D laser scanning technology provides a new technical path for rail transit deformation monitoring due to its fast, non-contact and high-density data collection characteristics.

[0003] At present, research on rail transit deformation monitoring based on 3D laser scanning has made some progress. Researchers have developed a variety of data processing algorithms, such as point cloud registration, surface reconstruction and deformation analysis, to improve monitoring accuracy and reliability. Some scholars have proposed combining multi-sensor fusion technology, such as combining laser scanning with inertial measurement units (IMUs), to improve dynamic monitoring capabilities. In terms of data analysis, machine learning and deep learning technologies have been introduced into deformation pattern recognition and prediction, improving the intelligence level of the system. In addition, some studies have attempted to combine 3D laser scanning with BIM (Building Information Modeling) technology to achieve visualization and full life cycle management of deformation data.

[0004] However, despite these advances, rail transit deformation monitoring technology based on 3D laser scanning still faces some key challenges. In terms of dynamic deformation monitoring, existing technologies have difficulty in real-time capture of instantaneous deformations caused by high-speed trains, which limits the accurate assessment of the dynamic response of the track. In terms of small deformation identification, current technologies have accuracy limitations when detecting millimeter-level or even sub-millimeter-level deformations, especially for some key structures such as tunnel linings. Such small deformations may indicate potential safety hazards. In terms of data processing in complex environments, factors such as multiple reflections, dust and vibration in closed spaces such as subways seriously affect the quality of laser scanning data. How to effectively filter out these noises and extract real deformation information remains a challenge. Existing methods often find it difficult to capture complex nonlinear deformation processes, limiting the system's early warning capabilities.

[0005] Hence the need for research and innovation. Summary of the invention

[0006] The purpose of the invention is to provide a method and system for urban rail transit deformation monitoring based on three-dimensional laser scanning, in order to solve at least some of the problems existing in the prior art.

[0007] Technical solution, according to one aspect of the present application, provides a method for monitoring deformation of urban rail transit based on three-dimensional laser scanning, comprising the following steps:

[0008] Step S1, acquiring laser data of urban rail transit based on a preconfigured laser scanner to construct an initial data set, wherein the laser data at least includes stimulated characteristic spectra of some key areas;

[0009] Step S2, calling a preconfigured data cleaning model to process the laser data to obtain a cleaned laser data set;

[0010] Step S3, obtaining the cleaned laser data set, and sequentially performing fractional differential transformation, singular spectrum decomposition and Hankel matrix reconstruction to obtain deformation data with an accuracy higher than a preset value;

[0011] Step S4, converting the deformation data into a fourth-order tensor, and performing tensor decomposition and feature extraction to obtain a deformation severity index and abnormal probability distribution data;

[0012] Step S5: calling a pre-built fractal dynamics prediction model, and using the deformation severity index and abnormal probability distribution data as input to obtain multi-scale deformation warning data.

[0013] According to one aspect of the present application, step S1 specifically comprises:

[0014] Step S11, calling pre-configured spectrum enhanced laser scanner parameters, setting the scanning range, resolution and spectrum acquisition configuration;

[0015] Step S12, controlling the spectrum enhanced laser scanner to emit a predefined wide spectrum laser to illuminate a key monitoring area of ​​the rail transit system;

[0016] Step S13, collecting spectral signals returned from the surface of the monitored object, the spectral signals including light intensity and wavelength distribution data; the spectral signals including stimulated characteristic spectra from some key areas;

[0017] Step S14, associating the collected spectral signal with the corresponding three-dimensional space coordinates to form point cloud data of the associated spectral signal;

[0018] Step S15, preprocessing the point cloud data of the associated spectral signal to form an initial data set; wherein the preprocessing includes removing outliers, performing spatial registration and coordinate system conversion on the point cloud.

[0019] According to one aspect of the present application, step S2 further comprises:

[0020] Step S21, loading a preconfigured multi-scale morphological filtering model, wherein the multi-scale morphological filtering model includes structural elements and morphological operation sequences for different scales;

[0021] Step S22, applying a multi-scale morphological filtering model to perform multi-scale morphological filtering on the initial data set to obtain filtered data;

[0022] Step S23, performing nonlinear manifold mapping on the filtered data to obtain laser point cloud data in a low-dimensional manifold space;

[0023] Step S24: performing geodesic filtering and cleaning on the laser point cloud data in the low-dimensional manifold space in the manifold space, and outputting the cleaned laser data set.

[0024] According to one aspect of the present application, step S3 further comprises:

[0025] Step S31, converting the cleaned laser data set into a time series matrix:

[0026] Step S32: Perform fractional differential transformation on the time series matrix to obtain a fractional differential matrix:

[0027] Step S33, perform singular spectrum decomposition and feature extraction on the fractional differential matrix to obtain a feature matrix:

[0028] Step S34: Reconstruct the Hankel matrix to obtain accurate deformation data:

[0029] Step S35: verify whether the accuracy of the precise deformation data is higher than a preset value.

[0030] According to one aspect of the present application, step S4 is further:

[0031] Step S41, constructing a fourth-order tensor;

[0032] Step S42, performing mixed tensor decomposition on the fourth-order tensor to obtain a decomposition result;

[0033] Step S43: extract features from the decomposition results and perform anomaly detection to obtain a deformation severity index:

[0034] Step S44, estimating the abnormal probability distribution of the deformation severity index, and outputting the deformation severity index and abnormal probability distribution data.

[0035] According to one aspect of the present application, step S5 is further:

[0036] Step S51, loading a pre-built fractal kinetic prediction model:

[0037] Step S52: Input the deformation severity index and abnormal probability distribution data into the fractal dynamics prediction model, perform multi-step fractal numerical solution, and obtain the state vector evolution:

[0038] Step S53: In the iterative solution process, the prediction window is adaptively adjusted to obtain and output multi-scale deformation warning data.

[0039] According to one aspect of the present application, the method further includes step S6, constructing an abnormal deformation prediction model based on topological data analysis and continuous coherence, analyzing multi-scale deformation warning data, and identifying potential abnormal deformation patterns.

[0040] According to one aspect of the present application, step S22 is specifically:

[0041] Step S221, using wavelet transform to decompose the initial data set into multiple scale levels;

[0042] Step S222: for each scale level, apply the corresponding structural element and morphological operation sequence, and use an adaptive weight function to merge the filtering results of different scales;

[0043] Step S23 is specifically as follows:

[0044] Step S231, construct a k-nearest neighbor graph, and use an adaptive k-value algorithm to determine the neighborhood of each point;

[0045] Step S232, calculating the tangent space of each local neighborhood, and using an improved principal component analysis method to calculate the local coordinates of the data points in the tangent space for each tangent space;

[0046] Step S233, constructing a global alignment function, and taking minimizing the inconsistency of the local tangent space as the optimization goal;

[0047] Step S234, obtaining a nonlinear mapping function from high dimension to low dimension by iteratively optimizing the alignment function;

[0048] Step S24 is specifically as follows:

[0049] Step S241, constructing a discrete geodesic network on the manifold;

[0050] Step S242, constructing a geodesic kernel function, taking into account the local curvature of the manifold;

[0051] Step S243: for each data point in the laser point cloud data obtained in the low-dimensional manifold space, apply a geodesic kernel function to perform local smoothing on a discrete geodesic network;

[0052] Step S244, iteratively apply geodesic filtering using an adaptive step size algorithm until convergence conditions are reached, and output the cleaned laser data set;

[0053] Step S31 is specifically as follows:

[0054] Step S311, construct an adaptive time window algorithm to determine the optimal time series length L;

[0055] Step S312, reorganizing the cleaned laser data set into multiple overlapping time series, each with a length of L;

[0056] Step S313, constructing a Hankel matrix H, arranging the time series into columns of the matrix;

[0057] Step S32 is specifically as follows:

[0058] Step S321, constructing a fractional-order differential operator Dα, where α is a non-integer order;

[0059] Step S322, using an adaptive algorithm to determine the optimal α value based on the local characteristics of the signal;

[0060] Step S323, applying a fractional differential operator to each column of the time series matrix to obtain a fractional differential matrix;

[0061] Step S33 is specifically as follows:

[0062] Step S331, applying an improved fast SVD algorithm to decompose the fractional differential matrix;

[0063] Step S332: construct a feature importance evaluation function; and adaptively select main feature vectors based on the feature importance evaluation function to construct a feature matrix;

[0064] Step S34 is specifically as follows:

[0065] Step S341, analyzing and constructing a matrix completion algorithm according to the structural characteristics of the Hankel matrix;

[0066] Step S342, using a matrix completion algorithm to expand the characteristic matrix into a complete Hankel matrix;

[0067] Step S343, applying the diagonal averaging method to extract the time series from the complete Hankel matrix;

[0068] Step S344, converting the time series back to the original data format to obtain accurate deformation data;

[0069] According to one aspect of the present application, step S41 is specifically:

[0070] Step S411, constructing an adaptive discretization algorithm to determine the optimal resolution of space, time, frequency and scale;

[0071] Step S412, reorganizing the precise deformation data into a four-dimensional array corresponding to space, time, frequency and scale;

[0072] Step S413, applying a tensorization algorithm to convert the four-dimensional array into a fourth-order tensor;

[0073] Step S42 is specifically as follows:

[0074] Step S421, constructing a hybrid Tucker-CP decomposition algorithm;

[0075] Step S422, initializing the core tensor C factor matrix;

[0076] Step S423, iteratively optimizing the core tensor C factor matrix with the goal of minimizing the reconstruction error; during the iteration, using an adaptive stopping criterion to strike a balance between accuracy and computational efficiency;

[0077] Step S43 is specifically as follows:

[0078] Step S431, constructing a tensor feature mapping function to extract key deformation features;

[0079] Step S432, using a tensor eigenmapping function on the factor matrix of the core tensor to obtain an eigenvector;

[0080] Step S433: construct an anomaly detector based on Mahalanobis distance;

[0081] Step S434, using an anomaly detector to calculate the anomaly score of the feature vector to obtain a deformation severity index;

[0082] Step S44 is specifically as follows:

[0083] Step S441, constructing a kernel density estimation algorithm K to adapt to the non-Gaussian characteristics of the deformed data;

[0084] Step S442, using a kernel density estimation algorithm and historical data to estimate a probability density function of a deformation severity index;

[0085] Step S443, calculating the cumulative distribution function by numerical integration; outputting the deformation severity index and abnormal probability distribution data, including the probability density function of the deformation severity index and the cumulative distribution function;

[0086] Step S51 is specifically as follows:

[0087] Step S511, read model parameters, including fractional dimension operators and nonlinear dynamic equations; verify model version and compatibility, and update;

[0088] Step S52 is specifically as follows:

[0089] Step S521, converting the deformation severity index and abnormal probability distribution data into initial conditions and boundary conditions of the characteristic matrix;

[0090] Step S522, applying the fractal dimension numerical method to perform multi-step iterative solution to obtain the evolution data of the state vector;

[0091] Step S53 is specifically as follows:

[0092] Step S531, construct a prediction error evaluation function E(Δt), where Δt is the prediction time window; use the gradient descent method to optimize the prediction time window Δt and minimize E(Δt); use an adaptive step size algorithm to adjust the prediction time window Δt in each iteration;

[0093] Step S532: construct a multi-scale threshold function and divide the prediction results into three time scales: immediate, short-term and medium-term; calculate the warning level for each scale; and output multi-scale deformation warning data.

[0094] According to one aspect of the present application, step S61 is specifically:

[0095] Step S611, reading multi-scale deformation warning data;

[0096] Step S612, constructing an adaptive point cloud sampling algorithm module, including: determining a sampling density function based on the warning level in the multi-scale deformation warning data; using the Poisson disk sampling method to generate a representative point set P according to the sampling density function; performing spatial normalization on the representative point set to ensure that the point set is distributed in a unit hypercube;

[0097] Step S613, constructing a Vietoris-Rips complex family {SCε}, including: constructing a number of radius parameters, and for each radius parameter, constructing a corresponding Vietoris-Rips complex SCεi; recording the simplex of each Vietoris-Rips complex and its inclusion relationship;

[0098] Step S614, using a complex simplification algorithm to reduce computational complexity, including: calculating the persistence measure of each simplex; and removing simplexes with persistence below the threshold according to a preset threshold; re-adjusting the connection relationship of the remaining simplexes; and outputting a simplified multi-scale simplicial complex SC1;

[0099] Step S62, calculating continuous homology and extracting topological features, including:

[0100] Step S621, inputting the simplified multi-scale simplicial complex;

[0101] Step S622, initializing a boundary matrix, which represents a mapping from a k-dimensional simplex to a (k-1)-dimensional simplex; performing a column echelon transformation on the boundary matrix to obtain a simplified boundary matrix; extracting persistent homology information from the simplified boundary matrix, including birth and death times;

[0102] Step S623: for each homology class, draw a horizontal line segment from birth to death; sort the barcodes by dimension and duration to obtain a continuous homology barcode map;

[0103] Step S624, representing each homology class as a point (b, d) on a plane, where b is the birth time and d is the death time; calculating the statistical characteristics of the duration graph, including the average duration and the duration distribution;

[0104] Step S625, outputting the continuous coherent barcode graph and the continuous graph.

[0105] According to another aspect of the present application, there is also provided an urban rail transit deformation monitoring system based on three-dimensional laser scanning, comprising:

[0106] at least one processor; and,

[0107] a memory communicatively connected to at least one of the processors; wherein,

[0108] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the urban rail transit deformation monitoring method based on three-dimensional laser scanning as described in any one of the above technical solutions.

[0109] Beneficial effect: By combining laser point cloud with spectral materials, it is possible to accurately measure deformation caused by small deformations and dynamic factors. In particular, deformations caused by vibrations when vehicles pass by are difficult to monitor in static conditions, which improves the foresight of monitoring. The relevant advantages will be described below in conjunction with the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] Figure 1 It is a flow chart of the present invention.

[0111] Figure 2 It is a flow chart of step S1 of the present invention.

[0112] Figure 3 It is a flow chart of step S2 of the present invention.

[0113] Figure 4 It is a flow chart of step S3 of the present invention.

[0114] Figure 5 It is a flow chart of step S4 of the present invention.

[0115] Figure 6It is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0116] like Figure 1 As shown, a method for monitoring deformation of urban rail transit based on three-dimensional laser scanning is provided, comprising the following steps:

[0117] Step S1, based on the pre-configured laser scanner, the laser data of the urban rail transit is acquired to construct an initial data set, wherein the laser data at least includes the stimulated characteristic spectra of some key areas; specifically, Figure 2 As shown, including:

[0118] S11, calling pre-configured spectrum enhanced laser scanner (SELS) parameters, setting the scanning range, resolution and spectrum acquisition configuration;

[0119] S12, controlling SELS to emit a predefined wide spectrum laser to illuminate the key monitoring area of ​​the rail transit system;

[0120] S13, collecting spectral signals returned from the surface of the monitored object, including light intensity and wavelength distribution data;

[0121] S14, associating the collected spectral signal with the corresponding three-dimensional space coordinates to form point cloud data;

[0122] S15, preprocessing the point cloud data to form an initial data set; including:

[0123] S151, remove obvious outliers and noise data;

[0124] S152, perform spatial registration on the point cloud to ensure the consistency of multiple scanning data;

[0125] S153, applying a predefined coordinate transformation to convert the point cloud data into a unified reference coordinate system;

[0126] S154, integrating the processed point cloud data and the associated spectral information to construct an initial data set D1.

[0127] By adopting a spectrally enhanced laser scanner (SELS), this method can simultaneously acquire high-resolution three-dimensional spatial data and rich spectral information. SELS emits a predefined wide-spectrum laser to excite the special spectral material on the surface of the monitored object, so that even small, instantaneous, and dynamic deformations can be reflected in the returned spectral signal. This innovative data acquisition method not only improves the accuracy of spatial measurement, but also provides additional spectral dimension information for subsequent deformation analysis. By associating point cloud data with spectral information and performing preliminary spatial registration and coordinate transformation, a multi-dimensional, high-precision initial data set is constructed. This method is suitable for detecting small deformations that are difficult to capture with traditional technologies, such as millimeter-level changes in tunnel linings, or deformations when trains pass by. In addition, the non-contact measurement characteristics of SELS improve the efficiency and safety of data acquisition, allowing large-scale scanning to be completed quickly even in complex rail transit environments. This step lays a solid foundation for subsequent data processing and analysis, and significantly improves the starting point accuracy and data quality of deformation monitoring.

[0128] It should be noted that in the field of quantum dot materials, there are some materials that can emit fluorescence after being excited, and their fluorescence intensity and wavelength will change with slight deformation. Cadmium selenide (CdSe) quantum dots, zinc sulfide (ZnS) quantum dots, cadmium sulfide (CdS) quantum dots, silicon quantum dots (Si QDs), carbon quantum dots (Carbon Quantum Dots, CQDs), copper sulfide (CuS) quantum dots, etc.

[0129] like Figure 3 As shown, according to one aspect of the present application,

[0130] Step S2: calling a pre-configured data cleaning model to process the laser data to obtain a cleaned laser data set; specifically including:

[0131] S21, loading a preconfigured multi-scale morphological filtering model M1, which includes structural elements and morphological operation sequences for different scales;

[0132] S22, applying the multi-scale morphological filtering model M1 to perform multi-scale morphological filtering on the initial data set D1, the process is as follows:

[0133] S221, decompose D1 into multiple scale levels {L1, L2, ..., Ln} using wavelet transform;

[0134] S222, for each level Li, apply the corresponding structure element SEi and morphological operation sequence Oi;

[0135] S223, using an adaptive weight function W(Li) to merge filtering results of different scales;

[0136] S23, performing nonlinear manifold mapping on the filtered data to obtain laser point cloud data in a low-dimensional manifold space; the steps are as follows:

[0137] S231, constructing a k-nearest neighbor graph, and using an adaptive k-value algorithm to determine the neighborhood of each point;

[0138] S232, calculating the tangent space Ti of each local neighborhood using an improved principal component analysis (PCA) method;

[0139] S2321. For each data point p, consider the local data set Lp formed by its k nearest neighbors;

[0140] S2322. Centralize the local data set Lp to obtain centered_Lp;

[0141] S2323. Calculate the covariance matrix C of centered_Lp;

[0142] S2324. Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ1 ≥ λ2 ≥ ... ≥ λd and corresponding eigenvectors v1, v2, ..., vd;

[0143] S2325. Adaptively select the number of principal components m using the improved variance contribution rate threshold method: calculate the cumulative variance contribution rate Ri = Σ(j=1 to i)λj / Σ(j=1 to d)λj; select the minimum m value that satisfies Rm > threshold and Rm+1 -Rm < ε;

[0144] S2326. Use the first m eigenvectors [v1, v2, ..., vm] to construct the local tangent space Ti;

[0145] S2327. Calculate the projection of the data point p onto the tangent space Ti to obtain a low-dimensional representation pi_low;

[0146] S233, constructing a global alignment function F to minimize the inconsistency of the local tangent space;

[0147] S234, obtaining a nonlinear mapping function M from high dimension to low dimension by iteratively optimizing the global alignment function F;

[0148] S235, applying a nonlinear mapping function M to map the data to a low-dimensional manifold space to obtain a data set D3;

[0149] S24, perform geodesic filtering and cleaning in the manifold space, and output the cleaned laser data set. Specifically:

[0150] S241. Constructing discrete geodesic networks GN on manifolds;

[0151] S242, constructing a geodesic kernel function K, taking into account the local curvature of the manifold;

[0152] S243, for each point p in the data set D3, apply the geodesic kernel function K to perform local smoothing on GN;

[0153] S244, iteratively applying geodesic filtering using an adaptive step size algorithm until a convergence condition is reached;

[0154] S245 , output the cleaned laser data set D4.

[0155] In this embodiment, the problem of data noise processing in complex environments is effectively solved by integrating multi-scale morphological filtering, nonlinear manifold mapping and geodesic filtering technology. The multi-scale morphological filtering model M1 can adaptively process noise of different scales, and is particularly suitable for processing complex noise caused by multiple reflections, dust and vibration in closed spaces such as subways. By applying specific structural elements and morphological operation sequences to each scale level, the method can effectively remove noise while retaining important structural information. The introduction of nonlinear manifold mapping technology solves the limitations of traditional linear dimensionality reduction methods in processing high-dimensional nonlinear data. By constructing a k-nearest neighbor graph and calculating the local tangent space, the method can better retain the intrinsic structure of the data and improve the accuracy of subsequent analysis. The application of geodesic filtering further enhances the consistency of the data, especially when processing deformation data of curved structures (such as tunnel inner walls). This multi-level data cleaning strategy not only significantly improves the data quality, but also reduces the computing resources required for subsequent processing, laying the foundation for real-time monitoring and analysis.

[0156] Step S3, obtaining the cleaned laser data set, and sequentially performing fractional differential transformation, singular spectrum decomposition and Hankel matrix reconstruction to obtain deformation data with an accuracy higher than a preset value; specifically, Figure 4 As shown, including:

[0157] S31. Convert the cleaned data set D4 into a time series matrix T1:

[0158] S311, construct an adaptive time window algorithm to determine the optimal time series length L;

[0159] S312, reorganizing the cleaned data set D4 into multiple overlapping time series, each with a length of L;

[0160] S313, constructing a Hankel matrix H, arranging the time series into columns of the matrix;

[0161] S32. Perform fractional differential transformation on the time series matrix T1 to obtain the fractional differential matrix F1:

[0162] S321. Construct a fractional-order differential operator Dα, where α is a non-integer order;

[0163] S322, using an adaptive algorithm to determine an optimal α value based on local characteristics of the signal;

[0164] S323, apply Dα to each column of T1 to obtain the fractional differential matrix F1;

[0165] S33, perform singular spectrum decomposition and feature extraction on the fractional differential matrix F1 to obtain the feature matrix E1:

[0166] S331. Apply the improved fast SVD algorithm to decompose the fractional differential matrix F1: F1 = USV^T;

[0167] S332, constructing a feature importance evaluation function I(σi), where σi is a singular value;

[0168] S333, adaptively selecting the main feature vectors based on the feature importance evaluation function I(σi) to construct the feature matrix E1;

[0169] S34, perform Hankel matrix reconstruction to obtain accurate deformation data D5:

[0170] S341, construct a matrix completion algorithm A, considering the structural characteristics of the Hankel matrix;

[0171] S342, using matrix completion algorithm A to expand the characteristic matrix E1 into a complete Hankel matrix H';

[0172] S343, extracting the time series from H' using the diagonal averaging method;

[0173] S344, converting the time series back to the original data format to obtain accurate deformation data D5;

[0174] S35 , verify whether the accuracy of the precise deformation data D5 is higher than a preset value, if not, return to S31 to adjust the parameters.

[0175] In this embodiment, fractional differential transform, singular spectrum decomposition and Hankel matrix reconstruction technology are combined to achieve accurate identification of small deformations. The introduction of fractional differential transform is a major innovation of this step, which can capture subtle changes that are difficult to identify with traditional integer differentials. The optimal fractional order α is determined by an adaptive algorithm. This method can be flexibly adjusted according to the local characteristics of the signal, improving the adaptability to different types of deformations. The application of singular spectrum decomposition further enhances the ability to extract key deformation features. By constructing a feature importance evaluation function, this method can adaptively select the most representative eigenvectors, effectively reducing the dimension of the data while retaining key information. The use of Hankel matrix reconstruction technology solves the problem of reconstructing high-quality time series, especially when dealing with discontinuous or missing data. This matrix completion-based method can effectively restore the hidden structure in the data and improve the accuracy of the reconstructed signal. This step can accurately extract small deformation information from noisy data, and can even identify sub-millimeter deformations that are difficult to capture with traditional methods. This is of great significance for early detection of potential structural problems, such as small cracks in tunnel linings or small deformations of tracks, and improves the sensitivity and reliability of the monitoring system.

[0176] Step S4, converting the deformation data into a fourth-order tensor, and performing tensor decomposition and feature extraction to obtain deformation severity index and abnormal probability distribution data; specifically, Figure 5 As shown, including:

[0177] S41. Construct the fourth-order tensor T2:

[0178] S411. Construct an adaptive discretization algorithm to determine the optimal resolution in space, time, frequency and scale;

[0179] S412, reorganizing the precise deformation data D5 into a four-dimensional array corresponding to space-time-frequency-scale;

[0180] S413, applying a tensorization algorithm to convert the four-dimensional array into a fourth-order tensor T2;

[0181] S42. Perform mixed tensor decomposition on the fourth-order tensor T2 to obtain the decomposition result:

[0182] S421, construct the Hankel matrix of the hybrid Tucker-CP decomposition algorithm;

[0183] S422, initialize the core tensor C and factor matrix {U1, U2, U3, U4};

[0184] S423, iteratively optimize the core tensor C and factor matrix {U1, U2, U3, U4} to minimize the reconstruction error;

[0185] S424, applying an adaptive stopping criterion to balance accuracy and computational efficiency;

[0186] S43. Extract features from the decomposition results and perform anomaly detection to obtain a deformation severity index:

[0187] S431, constructing a tensor feature mapping function F to extract key deformation features;

[0188] S432, apply the tensor feature mapping function F to the core tensor C and the factor matrix {U1, U2, U3, U4} to obtain the feature vector v;

[0189] S433, construct an anomaly detector D based on Mahalanobis distance;

[0190] S434, using the anomaly detector D to calculate the anomaly score s of the feature vector v, and obtain the deformation severity index SI =s;

[0191] S44, estimate the abnormal probability distribution, and output the deformation severity index SI and the abnormal probability distribution data P1 = {pdf, cdf}. The specific steps of estimating the abnormal probability distribution are:

[0192] S441, constructing a kernel density estimation algorithm K to adapt to the non-Gaussian characteristics of the deformed data;

[0193] S442, estimating the probability density function pdf of the deformation severity index SI using a kernel density estimation algorithm K and historical data;

[0194] S443. Calculate the cumulative distribution function cdf by numerical integration.

[0195] In this embodiment, by introducing the fourth-order tensor representation and hybrid Tucker-CP decomposition technology, a comprehensive analysis of deformation data is achieved. The construction of the fourth-order tensor makes it possible to simultaneously consider the deformation characteristics of four dimensions: space, time, frequency, and scale. This multidimensional representation method enhances the ability to describe complex deformation patterns. The application of the adaptive discretization algorithm ensures that the optimal resolution can be selected in each dimension, improving the efficiency and accuracy of data representation. The construction of the hybrid Tucker-CP decomposition algorithm combines the advantages of Tucker decomposition and CP decomposition, and can better capture the multi-scale structure and local features in the data. By iteratively optimizing the core tensor and factor matrix, this method can significantly reduce the dimension of the data while retaining key information. The introduction of the anomaly detector based on the Mahalanobis distance further enhances the system's ability to identify abnormal deformations. Considering the correlation between features, it can more accurately identify abnormal points in multidimensional space than the traditional Euclidean distance. By calculating the deformation severity index and the abnormal probability distribution, this step can not only quantitatively evaluate the degree of deformation, but also provide a statistical basis for deformation abnormality. This comprehensive deformation feature extraction and anomaly detection method greatly improves the sensitivity and accuracy of the monitoring system, especially in identifying complex nonlinear deformation patterns and early signs of abnormalities, providing technical support for timely detection of potential safety hazards.

[0196] Step S5, calling the pre-built fractal dynamics prediction model, and taking the deformation severity index and abnormal probability distribution data as input, to obtain multi-scale deformation warning data; specifically, Figure 6 As shown, including:

[0197] S51, load the pre-built fractal kinetic prediction model M2:

[0198] S511, reading model parameters, including fractional dimension operator ∇α and nonlinear dynamic equation group E1;

[0199] S512. Verify the model version and compatibility, and update if necessary;

[0200] S52, input the deformation severity index SI and the abnormal probability distribution data P1 into the model M2, perform multi-step fractal numerical solution, and obtain the state vector evolution:

[0201] S521, converting the deformation severity index SI and the abnormal probability distribution data P1 into initial conditions and boundary conditions of the nonlinear dynamic equation group E1;

[0202] S522, applying the fractal numerical method N to perform multi-step iterative solution to obtain the evolution of the state vector X(t);

[0203] S53, performing adaptive prediction window adjustment, obtaining multi-scale deformation warning data and outputting it;

[0204] S531, constructing a prediction error evaluation function E(Δt), where Δt is a prediction time window;

[0205] S532, using the gradient descent method to optimize the prediction time window Δt and minimize the error evaluation function E(Δt);

[0206] S533, applying an adaptive step size algorithm to adjust the prediction time window Δt in each iteration;

[0207] S534, constructing a multi-scale threshold function T(s), where s represents a time scale;

[0208] S535, dividing the prediction results into three time scales: immediate (s1), short-term (s2) and medium-term (s3);

[0209] S536. For each scale si, calculate the warning level Li = X(si) / T(si);

[0210] S537. Output multi-scale deformation warning data W1 = {L1, L2, L3}.

[0211] In this embodiment, a fractal dimension dynamic prediction model is introduced, and combined with the adaptive prediction window adjustment technology, an accurate prediction of rail transit deformation is achieved. The use of the fractal dimension dynamic model is the core innovation of this step. By introducing the concept of fractional dimension, it can more accurately describe the complex nonlinear deformation process. It is particularly suitable for dealing with multi-scale, nonlinear deformation problems commonly found in rail transit systems, such as elastic deformation of tracks and long-term settlement of tunnels. The construction of the fractional dimension operator ∇α enables the model to capture the rate of change between integer-order derivatives, improving the sensitivity to small gradual changes. The application of multi-step fractal dimension numerical solution methods further enhances the stability and accuracy of the prediction, especially when dealing with long-term series predictions. The introduction of the adaptive prediction window adjustment technology solves the problem of selecting the prediction time scale. By constructing the prediction error evaluation function E(Δt) and using the gradient descent method to optimize the time window Δt, the method can find the best balance between prediction accuracy and computational efficiency. This adaptive mechanism enables the system to dynamically adjust the prediction strategy according to the characteristics of the deformation data, which can capture both short-term rapid changes and long-term slow trends. The generation of multi-scale warning signals further enhances the practicality of the system. By dividing the prediction results into three time scales: immediate, short-term and medium-term, it provides targeted information support for decision-making at different levels. This comprehensive and accurate prediction capability improves the foresight of rail transit deformation monitoring, allowing managers to take preventive measures in advance and effectively reduce safety hazards.

[0212] S61. Construct a multi-scale simplicial complex to transform deformation warning data into a topological structure:

[0213] S611, input multi-scale deformation warning data W1 = {L1, L2, L3};

[0214] S612. Constructing adaptive point cloud sampling algorithm A:

[0215] S6121. Determine a sampling density function d(x) based on the warning level in W1.

[0216] S6122. Use the Poisson disk sampling method to generate a representative point set P based on d(x);

[0217] S6123, perform spatial normalization on P to ensure that the point set is distributed in a unit hypercube;

[0218] S613. Construct the Vietoris-Rips complex family {SCε}:

[0219] S6131. ​​Set a series of radius parameters {ε1, ε2, ..., εn};

[0220] S6132. For each εi, construct the corresponding Vietoris-Rips complex SCεi;

[0221] S6133, record the simplex and its inclusion relationship of each SCεi;

[0222] S614, applying complex shape simplification algorithm S to reduce computational complexity:

[0223] S6141. Calculate the persistence measure m of each simplex;

[0224] S6142, according to a preset threshold τ, removing simplexes whose persistence is lower than τ;

[0225] S6143, readjust the connection relationship of the remaining simplexes;

[0226] S615, outputting the simplified multi-scale simplicial complex SC1;

[0227] S62, calculate continuous homology and extract topological features:

[0228] S621, input the simplified multi-scale simplicial complex SC1;

[0229] S622. Build and apply a fast continuous synchronization algorithm:

[0230] S6221, initialize the boundary matrix ∂k, representing the mapping from the k-dimensional simplex to the (k-1)-dimensional simplex;

[0231] S6222, performing a column echelon transformation on ∂k to obtain a simplified boundary matrix R;

[0232] S6223, extracting continuous homology information from R, including birth and death times;

[0233] S623, generating a continuous coherent barcode diagram:

[0234] S6231. For each homology class, draw a horizontal line segment from birth to death;

[0235] S6232, sort barcodes by dimension and duration;

[0236] S624, build a persistence graph:

[0237] S6241. Represent each homology class as a point (b, d) on the plane, where b is the birth time and d is the death time;

[0238] S6242. Calculate statistical characteristics of the duration graph, including average duration and duration distribution;

[0239] S625, outputting a continuous coherent barcode graph and a continuous graph;

[0240] S63. Extract key topological features:

[0241] S631, input persistence diagram;

[0242] S632. Construct discrete Morse function M:

[0243] S6321. Define the height function h(x) = d(x) - b(x), where d(x) and b(x) are the death and birth times of point x, respectively;

[0244] S6322, calculating the discrete gradient field of h(x);

[0245] S6323, Identify critical points (local minima, saddle points, and local maxima);

[0246] S633. Extract key topological features of the persistence graph:

[0247] S6331. For each dimension k, find the m homology classes with the longest duration;

[0248] S6332. Determine the representative cycles corresponding to these homology classes;

[0249] S6333. Calculate the geometric center and range of each representative cycle;

[0250] S634, build topological skeleton:

[0251] S6341, using key topological features as nodes of the topological skeleton;

[0252] S6342. Add edges between nodes based on the structure of the persistence graph;

[0253] S6343, assign a weight to each edge to reflect the similarity of the connected features;

[0254] S635. Calculate the persistence measure of topological features:

[0255] S6351. For each feature f, calculate its persistence p(f) = (d(f) - b(f)) / max(d - b);

[0256] S6352, Normalize the persistence measure so that its sum is 1;

[0257] S636, outputting a key topological feature set T3, including geometric information, topological skeleton and continuity measure of the feature;

[0258] S64. Anomaly detection based on topological features:

[0259] S641, inputting the key topological feature set T3 and the new deformation warning data point p;

[0260] S642. Construct a topological anomaly score function S(p, T3):

[0261] S6421, map p to the persistence graph space;

[0262] S6422, calculate the distance d(p, f) from p to each feature in T3;

[0263] S6423. Calculate the weighted anomaly score based on the distance and persistence of the feature: S(p, T3) = Σ(w(f) * d(p, f)) / Σw(f), where w(f) is the persistence measure of feature f;

[0264] S643. Constructing topological anomaly detector TAD1:

[0265] S6431. Set an abnormal threshold θ based on the distribution of historical data;

[0266] S6432. If S(p, T3) > θ, then mark p as a potential anomaly;

[0267] S644. Apply TAD1 to p and calculate its anomaly score:

[0268] S6441. Convert p to a representation compatible with T3;

[0269] S6442, calculate S(p, T3);

[0270] S6443, compare S(p, T3) with θ;

[0271] S645. If an abnormality is detected, an abnormality alarm is generated:

[0272] S6451. Record detailed information of the abnormal point p, including timestamp, location and abnormal score;

[0273] S6452, determine the topological features most relevant to p;

[0274] S6453, analyze the similarity of p with historical anomaly patterns;

[0275] S646, output abnormal deformation mode report R1:

[0276] S6461. Summarize all detected abnormal points;

[0277] S6462. Cluster the outliers and identify potential anomaly patterns;

[0278] S6463, generating a feature description of each abnormal pattern, including topological features and geometric features;

[0279] S6464. Provide an assessment of the severity of the abnormality and an analysis of possible causes.

[0280] In summary, in this embodiment, a comprehensive analysis of the potential structure of the deformation data is achieved. The construction of multi-scale simple complexes is the basis of this step. It converts the deformation warning data into a topological structure, so that the internal relationship of the data can be analyzed from a global perspective. The application of the adaptive point cloud sampling algorithm ensures that the computational complexity is effectively reduced while maintaining the key features of the data. The introduction of the continuous homology algorithm can capture the topological features of the data at different scales, which is particularly suitable for analyzing the common multi-scale deformation phenomena in rail transit systems. By generating continuous homology barcodes and continuous graphs, the method provides an intuitive visualization of the topological structure of the deformation data, which helps to identify key deformation patterns and evolution processes. The construction of discrete Morse functions further enhances the system's ability to extract key topological features, and can effectively identify critical points in the data, such as local extremes and saddle points, which often correspond to important deformation events or turning points. The construction of a topological skeleton provides a new perspective for the classification and comparison of deformation patterns, especially when analyzing complex nonlinear deformation processes. The anomaly detection method based on topological features is another important innovation of this step. It can evaluate the degree of abnormality of new data points from a topological perspective by constructing a topological anomaly score function. It can not only identify single-point anomalies, but also capture abnormal patterns caused by overall structural changes, which improves the system's sensitivity to potential dangers. By generating detailed abnormal deformation pattern reports, this step provides decision makers with comprehensive and in-depth risk assessment information, effectively supporting the safety management and maintenance strategy formulation of rail transit systems.

[0281] The present invention specifically addresses the challenges faced by traditional methods, from high-precision data collection to noise processing in complex environments, to accurate identification of tiny deformations, multi-dimensional feature extraction, accurate prediction, and deep topological analysis. It significantly improves the accuracy, real-time performance, and reliability of deformation monitoring, can detect potential structural problems early, accurately predict deformation trends, and identify abnormal patterns in a timely manner, thereby improving safety and operational efficiency.

[0282] Specifically, this problem is solved by using spectrally enhanced laser scanners and fractional order differential decomposition (FOSD) technology. SELS uses a wide spectrum laser to excite special spectral materials pre-coated on the surface of the monitored object. These materials are very sensitive to tiny deformations, and even instantaneous deformations caused by high-speed trains can be reflected in the returned spectral signals. The high-frequency sampling capability of SELS (up to hundreds of hertz) enables it to capture millisecond-level deformations.

[0283] In terms of data processing, FOSD technology excels at processing rapidly changing signals. The fractional-order differential operator Dα can capture the rate of change between integer-order derivatives, which enables the system to identify fast and small deformations that are difficult to capture with traditional methods. By determining the optimal α value through an adaptive algorithm, the system can flexibly adjust according to the local characteristics of the signal to improve its sensitivity to high-speed dynamic deformations. The fourth-order tensor representation takes the time dimension into account, allowing the system to comprehensively analyze the spatiotemporal characteristics of dynamic deformations. This multi-dimensional analysis method improves the accuracy of the assessment of the dynamic response of the track.

[0284] The problem of identifying small deformations is solved by SELS technology, FOSD technology and high-order tensor network deformation analysis (HOTNDA). FOSD technology enhances the system's sensitivity to small changes through fractional differential transformation. Fractional differentials can amplify subtle changes that are difficult to identify with traditional integer differentials, allowing the system to detect sub-millimeter deformations. HOTNDA technology converts deformation data into a fourth-order tensor, while considering the four dimensions of space, time, frequency and scale. This multidimensional representation method can capture small deformation patterns that may be overlooked by traditional methods. Through hybrid Tucker-CP decomposition, the system can extract key small deformation features from massive data. Finally, the topological data analysis method provides a new perspective for the identification of small deformations. By analyzing the topological structure of the deformation data, the system can identify small deformations that are not obvious on a global scale but significant in the local topological structure.

[0285] The multiscale morphological filtering and manifold learning (MSMFL) technology is used to solve the data processing problem in complex environments. The multiscale morphological filtering model can adaptively process noise of different scales. By constructing specific structural elements and morphological operation sequences for each scale level, this method can effectively remove complex noise caused by factors such as multiple reflections, dust and vibration, while retaining important structural information. The nonlinear manifold mapping technology can better preserve the intrinsic structure of the data by constructing a k-nearest neighbor graph and calculating the local tangent space. It is particularly suitable for processing nonlinear data collected in complex environments. The application of geodesic filtering further enhances the consistency of the data, especially when processing deformation data of curved structures (such as the inner wall of a tunnel). It can effectively remove local noise while maintaining global structural features. Singular spectrum decomposition technology also helps in noise filtering. By constructing a feature importance evaluation function, the system can adaptively select the most representative feature vectors to effectively separate signals and noise.

[0286] The MSMFL technology significantly reduces the amount of data through multi-scale processing and dimensionality reduction, laying the foundation for subsequent real-time analysis. The FOSD technology effectively compresses the data while retaining key information through fractional differential transformation and singular spectrum decomposition, reducing the computing resources required for subsequent processing. Although the HOTNDA technology introduces high-dimensional tensor representation, it achieves efficient feature extraction through hybrid Tucker-CP decomposition, enabling the system to quickly process large-scale data. The fractal dimension dynamics prediction model (FDDPM) uses adaptive prediction window adjustment technology, which can dynamically adjust the computational complexity according to data characteristics, improving processing speed while ensuring accuracy. Through topological data analysis methods, especially complex simplification algorithms, the amount of data that needs to be processed can be greatly reduced while retaining key topological features.

[0287] After obtaining the digital model of urban rail, online laser devices can be deployed to conduct online tracking of specific areas and key areas to obtain deformation data during rail operation, thereby improving the frequency and accuracy of monitoring and reducing the impact of detection work on rail transit operations.

[0288] According to another aspect of the present application, there is also provided an urban rail transit deformation monitoring system based on three-dimensional laser scanning, comprising:

[0289] at least one processor; and,

[0290] a memory communicatively connected to at least one of the processors; wherein,

[0291] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the urban rail transit deformation monitoring method based on three-dimensional laser scanning as described in any one of the above technical solutions.

[0292] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A method for monitoring deformation of urban rail transit based on three-dimensional laser scanning, characterized in that: The steps include: Step S1, acquiring laser data of urban rail transit based on a preconfigured laser scanner to construct an initial data set, wherein the laser data at least includes stimulated characteristic spectra of some key areas; Step S2, calling a preconfigured data cleaning model to process the laser data to obtain a cleaned laser data set; Step S3, obtaining the cleaned laser data set, and sequentially performing fractional differential transformation, singular spectrum decomposition and Hankel matrix reconstruction to obtain deformation data with an accuracy higher than a preset value; Step S4, converting the deformation data into a fourth-order tensor, and performing tensor decomposition and feature extraction to obtain a deformation severity index and abnormal probability distribution data; Step S5, calling a pre-built fractal dynamics prediction model, and using the deformation severity index and abnormal probability distribution data as input to obtain multi-scale deformation warning data; The step S5 is further as follows: Step S51, loading a pre-built fractal kinetics prediction model; Step S52: Input the deformation severity index and abnormal probability distribution data into the fractal dynamics prediction model, perform multi-step fractal numerical solution, and obtain the state vector evolution: Step S53: in the iterative solution process, adaptively adjust the prediction window to obtain and output multi-scale deformation warning data; Step S51 is specifically as follows: Step S511, read model parameters, including fractional dimension operators and nonlinear dynamic equations; verify model version and compatibility, and update; Step S52 is specifically as follows: Step S521, converting the deformation severity index and abnormal probability distribution data into initial conditions and boundary conditions of the characteristic matrix; Step S522, applying the fractal dimension numerical method to perform multi-step iterative solution to obtain the evolution data of the state vector; Step S53 is specifically as follows: Step S531, constructing a prediction error evaluation function E(Δt), where Δt is a prediction time window; using a gradient descent method to optimize the prediction time window Δt, minimizing the prediction error evaluation function E(Δt); using an adaptive step size algorithm, adjusting the prediction time window Δt in each iteration; Step S532: construct a multi-scale threshold function and divide the prediction results into three time scales: immediate, short-term and medium-term; For each scale, calculate the warning level; Output multi-scale deformation warning data.

2. The urban rail transit deformation monitoring method based on three-dimensional laser scanning according to claim 1, characterized in that: The step S1 is specifically as follows: Step S11, calling pre-configured spectrum enhanced laser scanner parameters, setting the scanning range, resolution and spectrum acquisition configuration; Step S12, controlling the spectrum enhanced laser scanner to emit a predefined wide spectrum laser to illuminate a key monitoring area of ​​the rail transit system; Step S13, collecting spectral signals returned from the surface of the monitored object, the spectral signals including light intensity and wavelength distribution data; the spectral signals include stimulated characteristic spectra of some key areas; Step S14, associating the collected spectral signal with the corresponding three-dimensional space coordinates to form point cloud data of the associated spectral signal; Step S15, preprocessing the point cloud data of the associated spectral signal to form an initial data set; The preprocessing includes removing outliers, spatially registering the point cloud, and converting the coordinate system.

3. The urban rail transit deformation monitoring method based on three-dimensional laser scanning as claimed in claim 2, characterized in that: The step S2 is further as follows: Step S21, loading a preconfigured multi-scale morphological filtering model, wherein the multi-scale morphological filtering model includes structural elements and morphological operation sequences for different scales; Step S22, applying a multi-scale morphological filtering model to perform multi-scale morphological filtering on the initial data set to obtain filtered data; Step S23, performing nonlinear manifold mapping on the filtered data to obtain laser point cloud data in a low-dimensional manifold space; Step S24: performing geodesic filtering and cleaning on the laser point cloud data in the low-dimensional manifold space in the manifold space, and outputting the cleaned laser data set.

4. The urban rail transit deformation monitoring method based on three-dimensional laser scanning as claimed in claim 3, characterized in that: The step S3 is further as follows: Step S31, converting the cleaned laser data set into a time series matrix: Step S32: Perform fractional differential transformation on the time series matrix to obtain a fractional differential matrix: Step S33, perform singular spectrum decomposition and feature extraction on the fractional differential matrix to obtain a feature matrix: Step S34: Reconstruct the Hankel matrix to obtain accurate deformation data: Step S35: verify whether the accuracy of the precise deformation data is higher than a preset value.

5. The urban rail transit deformation monitoring method based on three-dimensional laser scanning according to claim 4, characterized in that: The step S4 is further as follows: Step S41, constructing a fourth-order tensor; Step S42, performing mixed tensor decomposition on the fourth-order tensor to obtain a decomposition result; Step S43: extract features from the decomposition results and perform anomaly detection to obtain a deformation severity index: Step S44, estimating the abnormal probability distribution of the deformation severity index, and outputting the deformation severity index and abnormal probability distribution data.

6. The urban rail transit deformation monitoring method based on three-dimensional laser scanning according to claim 5, characterized in that: The method further includes step S6, constructing an abnormal deformation prediction model based on topological data analysis and continuous coherence, analyzing multi-scale deformation warning data, and identifying potential abnormal deformation patterns.

7. The urban rail transit deformation monitoring method based on three-dimensional laser scanning according to claim 6, characterized in that: Step S22 is specifically as follows: Step S221, using wavelet transform to decompose the initial data set into multiple scale levels; Step S222: for each scale level, apply the corresponding structural element and morphological operation sequence, and use an adaptive weight function to merge the filtering results of different scales; Step S23 is specifically as follows: Step S231, construct a k-nearest neighbor graph, and use an adaptive k-value algorithm to determine the neighborhood of each point; Step S232, calculating the tangent space of each local neighborhood, and using an improved principal component analysis method to calculate the local coordinates of the data points in the tangent space for each tangent space; Step S233, constructing a global alignment function, and taking minimizing the inconsistency of the local tangent space as the optimization goal; Step S234, obtaining a nonlinear mapping function from high dimension to low dimension by iteratively optimizing the alignment function; Step S24 is specifically as follows: Step S241, constructing a discrete geodesic network on the manifold; Step S242, constructing a geodesic kernel function, taking into account the local curvature of the manifold; Step S243: for each data point in the laser point cloud data obtained in the low-dimensional manifold space, apply a geodesic kernel function to perform local smoothing on a discrete geodesic network; Step S244, iteratively apply geodesic filtering using an adaptive step size algorithm until convergence conditions are reached, and output the cleaned laser data set; Step S31 is specifically as follows: Step S311, construct an adaptive time window algorithm to determine the optimal time series length L; Step S312, reorganizing the cleaned laser data set into multiple overlapping time series, each with a length of L; Step S313, constructing a Hankel matrix H, arranging the time series into columns of the matrix; Step S32 is specifically as follows: Step S321, constructing a fractional-order differential operator Dα, where α is a non-integer order; Step S322, using an adaptive algorithm to determine the optimal α value based on the local characteristics of the signal; Step S323, applying a fractional differential operator to each column of the time series matrix to obtain a fractional differential matrix; Step S33 is specifically as follows: Step S331, applying an improved fast SVD algorithm to decompose the fractional differential matrix; Step S332: construct a feature importance evaluation function; and adaptively select main feature vectors based on the feature importance evaluation function to construct a feature matrix; Step S34 is specifically as follows: Step S341, analyzing and constructing a matrix completion algorithm according to the structural characteristics of the Hankel matrix; Step S342, using a matrix completion algorithm to expand the characteristic matrix into a complete Hankel matrix; Step S343, applying the diagonal averaging method to extract the time series from the complete Hankel matrix; Step S344: convert the time series back to the original data format to obtain accurate deformation data.

8. The urban rail transit deformation monitoring method based on three-dimensional laser scanning according to claim 6, characterized in that: Step S41 is specifically as follows: Step S411, constructing an adaptive discretization algorithm to determine the optimal resolution of space, time, frequency and scale; Step S412, reorganizing the precise deformation data into a four-dimensional array corresponding to space, time, frequency and scale; Step S413, applying a tensorization algorithm to convert the four-dimensional array into a fourth-order tensor; Step S42 is specifically as follows: Step S421, constructing a hybrid Tucker-CP decomposition algorithm; Step S422, initializing the core tensor C factor matrix; Step S423, iteratively optimizing the core tensor C factor matrix with the goal of minimizing the reconstruction error; during the iteration, using an adaptive stopping criterion to strike a balance between accuracy and computational efficiency; Step S43 is specifically as follows: Step S431, constructing a tensor feature mapping function to extract key deformation features; Step S432, using a tensor eigenmapping function on the factor matrix of the core tensor to obtain an eigenvector; Step S433: construct an anomaly detector based on Mahalanobis distance; Step S434, using an anomaly detector to calculate the anomaly score of the feature vector to obtain a deformation severity index; Step S44 is specifically as follows: Step S441, constructing a kernel density estimation algorithm K to adapt to the non-Gaussian characteristics of the deformed data; Step S442, using a kernel density estimation algorithm and historical data to estimate a probability density function of a deformation severity index; Step S443, calculating the cumulative distribution function by numerical integration; outputting the deformation severity index and abnormal probability distribution data, including the probability density function of the deformation severity index and the cumulative distribution function.

9. The urban rail transit deformation monitoring method based on three-dimensional laser scanning according to claim 6, characterized in that: Step S61 is specifically as follows: Step S611, reading multi-scale deformation warning data; Step S612, constructing an adaptive point cloud sampling algorithm module, including: determining a sampling density function based on the warning level in the multi-scale deformation warning data; using the Poisson disk sampling method to generate a representative point set P according to the sampling density function; performing spatial normalization on the representative point set to ensure that the point set is distributed in a unit hypercube; Step S613, constructing a Vietoris-Rips complex family {SCε}, including: constructing a number of radius parameters, and for each radius parameter, constructing a corresponding Vietoris-Rips complex SCεi; recording the simplex of each Vietoris-Rips complex and its inclusion relationship; Step S614, using a complex simplification algorithm to reduce computational complexity, including: calculating the persistence measure of each simplex; and removing simplexes with persistence below the threshold according to a preset threshold; re-adjusting the connection relationship of the remaining simplexes; and outputting a simplified multi-scale simplicial complex SC1; Step S62, calculating continuous homology and extracting topological features, including: Step S621, inputting the simplified multi-scale simplicial complex; Step S622, initializing a boundary matrix, which represents a mapping from a k-dimensional simplex to a (k-1)-dimensional simplex; performing a column echelon transformation on the boundary matrix to obtain a simplified boundary matrix; extracting persistent homology information from the simplified boundary matrix, including birth and death times; Step S623: for each homology class, draw a horizontal line segment from birth to death; sort the barcodes by dimension and duration to obtain a continuous homology barcode map; Step S624, representing each homology class as a point (b, d) on a plane, where b is the birth time and d is the death time; calculating the statistical characteristics of the duration graph, including the average duration and the duration distribution; Step S625, outputting the continuous coherent barcode graph and the continuous graph.

10. An urban rail transit deformation monitoring system based on three-dimensional laser scanning, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the urban rail transit deformation monitoring method based on three-dimensional laser scanning as described in any one of claims 1 to 9.

Citation Information

Patent Citations

  • Subway tunnel disease identification and detection device and method

    CN116476888A

  • Intelligent monitoring and early warning method for deformation of underground roadway

    CN117189254A