Subway line deformation risk monitoring method based on InSAR and multi-source data

Abstract

In the embodiment of the present invention, a subway line deformation risk monitoring method based on InSAR and multi-source data is provided, which belongs to the field of measurement technology, and specifically includes: after obtaining time series deformation data and performing two-dimensional deformation decomposition on it, using the logistic model to invert the time deformation characteristics, and using the Peck model to invert the spatial deformation characteristics, and monitoring the current deformation of the target area according to the inverted time deformation characteristics and spatial deformation characteristics; using the wavelet transform method to qualitatively and quantitatively explore the coupling relationship between the influencing factors and the deformation data along the subway line, and using the random forest method to calculate the characteristic proportion of each influencing factor; inputting the time series deformation data into the GA-BP neural network to obtain the deformation value of the next moment in the target area, and calculating the risk level corresponding to the target area according to the characteristic proportion. Through the scheme of the present invention, the monitoring efficiency and accuracy are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of measurement technology, and in particular to a subway line deformation risk monitoring method based on InSAR and multi-source data. Background Art

[0002] Geodetic methods such as leveling are effective technologies for monitoring the displacement of underground tunnels. However, this method requires setting deformation monitoring benchmarks at a relatively stable location at a certain distance from the tunnel construction area, building a benchmark network, and conducting regular observations across time periods, which is time-consuming and labor-intensive, and the validity of the data decreases as the monitoring time interval increases. Interferometric synthetic aperture radar (InSAR) technology can achieve high-precision, high-resolution, and wide-area deformation monitoring, and has significant advantages for conducting long-distance, continuous deformation monitoring of subway lines.

[0003] At present, many scholars at home and abroad have carried out research on subway deformation monitoring, including qualitative research on factors affecting subway deformation and underground structure deformation monitoring, but there is still no complete and comprehensive research process.

[0004] It can be seen that there is an urgent need for a subway line deformation risk monitoring method based on InSAR and multi-source data with high monitoring efficiency and accuracy. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a subway line deformation risk monitoring method based on InSAR and multi-source data, which at least partially solves the problem of poor monitoring efficiency and accuracy in the prior art.

[0006] The embodiment of the present invention provides a subway line deformation risk monitoring method based on InSAR and multi-source data, comprising:

[0007] Step 1: Use the differential interferometry short baseline set time series analysis technology to process the SAR data in the target area, obtain the time series deformation data and perform two-dimensional deformation decomposition on it, then use the logistic model to invert the time deformation characteristics, and use the Peck model to invert the spatial deformation characteristics, and monitor the current deformation of the target area based on the inverted time deformation characteristics and spatial deformation characteristics;

[0008] Step 2: Extract the deformation data along the subway line from the decomposed time series deformation data, and use the wavelet transform method to qualitatively and quantitatively explore the coupling relationship between the influencing factors and the deformation data along the subway line, and use the random forest method to calculate the characteristic proportion of each influencing factor;

[0009] Step 3: Input the time series deformation data into the GA-BP neural network to obtain the deformation value of the target area at the next moment, and calculate the risk level corresponding to the target area according to the feature proportion.

[0010] According to a specific implementation of the embodiment of the present invention, the expression of the logistic model is:

[0011]

[0012] in, represents the maximum settlement value, a and b are the parameters of the model, and t represents the time series.

[0013] According to a specific implementation of an embodiment of the present invention, the step of inverting the time deformation characteristics using a logistic model includes:

[0014] The temporal deformation data after two-dimensional deformation decomposition is divided into three segments, among which the first segment is , The second paragraph is The third paragraph is ;

[0015] make is the reciprocal sum of the total settlement in the three sections, that is:

[0016]

[0017] Substituting it into the formula and performing linear transformation, we get:

[0018]

[0019] Use the least squares method to solve the parameters a and b and transform the above formula into

[0020]

[0021] make , , , ,in, represents the number of observations, represents the observed value, represents the observation vector, represents the coefficient matrix, Indicates the parameters to be requested. Indicates accidental error;

[0022] Written in matrix form:

[0023]

[0024] By least squares have to,

[0025]

[0026] ;

[0027] in, Indicates transpose.

[0028] According to a specific implementation of the embodiment of the present invention, the expression of the Peck model is:

[0029]

[0030]

[0031] in, represents the ground settlement, x represents the horizontal distance from the tunnel axis, represents the formation loss rate, represents the sedimentation tank width coefficient, is the tunnel excavation radius.

[0032] According to a specific implementation of an embodiment of the present invention, the step of inverting the spatial deformation characteristics using the Peck model includes:

[0033] Acquire surface deformation data of target area through InSAR monitoring technology;

[0034] The width of the sedimentation trough and the formation loss rate are calculated using the preset formula of the Peck model.

[0035] According to a specific implementation of an embodiment of the present invention, the step of calculating the feature proportion of each influencing factor using a random forest method includes:

[0036] Perform bag sampling on the influencing factor set X=(x, y) to obtain a subset , train decision trees on subsets one by one , out-of-bag samples in the current decision tree Apply the decision tree to get the number of samples classified correctly , then for the feature , randomly shuffle the out-of-bag samples The d-th feature of constitutes D new out-of-bag samples ;

[0037] Make decision tree predictions on D new out-of-bag samples, and the correct number is recorded as , each sample subset performs the above calculations in turn, and finally calculates the importance score of the dth feature as the feature proportion, where the expression of the feature proportion is

[0038]

[0039] in, represents the proportion of the dth feature, M is the total number of test samples, m=1, 2,…,M.

[0040] According to a specific implementation of the embodiment of the present invention, before step 3, the method further includes:

[0041] Determine the GA-BP neural network structure and obtain the initial population of the genetic algorithm by encoding its weights and thresholds;

[0042] The initial weights and thresholds of the GA-BP neural network are continuously updated according to the evolution of the individuals. The GA-BP neural network is trained with measured data and then the output is predicted. The mean square error function shown in the formula is used as the fitness function to calculate the fitness of each individual and determine whether it meets the optimization criteria. If it does, proceed to the next step. If it does not, select, crossover, and mutate the individuals according to their fitness to generate new individuals and re-judge their fitness until they meet the criteria.

[0043]

[0044] Where M is the total number of test samples, m=1, 2, …, M; is the true value of the test sample, Predict values ​​for the test sample;

[0045] The optimal individuals obtained above are split in order and used as the initial weights and thresholds of the BP neural network;

[0046] Perform forward propagation and back propagation of the GA-BP neural network, update the weights and thresholds, and then re-perform forward propagation and error judgment until the preset requirements are met to obtain the optimal weights and thresholds.

[0047] According to a specific implementation of an embodiment of the present invention, the step of inputting the time series deformation data into a GA-BP neural network to obtain the deformation value of the target area at the next moment includes:

[0048] The deformation values ​​of the last preset number of time nodes of the time series deformation data are selected as the input vector, input into the trained model, calculate the corresponding activation values ​​in each layer of neurons in turn, and use the output of the last layer of neurons as the deformation value at the next moment.

[0049] According to a specific implementation of an embodiment of the present invention, the step of calculating the risk value corresponding to the target area according to the feature proportion includes:

[0050] All influencing factors are divided into different preset types, and the susceptibility (U) of each influencing factor is graded by the natural break point classification method. Then, the risk value of each preset type is calculated based on the characteristic proportion of the influencing factors to the deformation obtained by the random forest model as the weight. The preset types include topographic factors, hydrological environmental factors, and human activity factors. The expression of the risk value is:

[0051]

[0052] in, represents the score value of each category, n is the number of influencing factors of each category, i represents the number of preset categories, i=1, 2, 3, j represents the number of influencing factors of each category, j=1, 2, ..., n, and They represent the characteristic proportion and susceptibility level of the jth factor of the i-th category respectively;

[0053] A three-dimensional XOY coordinate system is established, where XYZ represents the risk values ​​of geological and geomorphological factors, hydrological and environmental factors, and human activity factors, respectively. The corresponding score values ​​are divided into multiple intervals (J, H, G) through the three axes. The risk importance level obtained by the comprehensive evaluation of the three is O, and the relationship between the three is expressed as O=F(x, y, z);

[0054] According to the geological and geomorphological factors X ( ), hydrological environmental factor Y ( ), geological and geomorphological factors Z ( ) to construct a The O value at the intersection of rows and columns is the final risk level.

[0055] The subway line deformation risk monitoring scheme based on InSAR and multi-source data in the embodiment of the present invention includes: step 1, using differential interferometry short baseline set time series analysis technology to process SAR data in the target area, obtaining time series deformation data and performing two-dimensional deformation decomposition on it, using logistic model to invert time deformation characteristics, and using Peck model to invert spatial deformation characteristics, and monitoring the current deformation of the target area according to the inverted time deformation characteristics and spatial deformation characteristics; step 2, extracting subway line deformation data from the decomposed time series deformation data, and using wavelet transform method to qualitatively and quantitatively explore the coupling relationship between influencing factors and subway line deformation data, and using random forest method to calculate the characteristic proportion of each influencing factor; step 3, inputting the time series deformation data into GA-BP neural network to obtain the deformation value of the target area at the next moment, and calculating the risk level corresponding to the target area according to the characteristic proportion.

[0056] The beneficial effects of the embodiments of the present invention are as follows: through the scheme of the present invention, a method for realizing wide-area, real-time subway deformation monitoring using InSAR technology is proposed, which makes up for the shortcomings of traditional leveling monitoring methods, such as long observation time, long time span and low spatial resolution; the least squares method is used to solve the Logistic regression model, thereby improving the fitting accuracy; a subway deformation risk assessment method based on a hazard matrix is ​​proposed, which considers multiple influencing factors and conducts assessment by classification, thereby avoiding the interference of a single factor on the overall evaluation and improving the effectiveness and reference significance of risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 A schematic flow chart of a subway line deformation risk monitoring method based on InSAR and multi-source data provided by an embodiment of the present invention;

[0059] Figure 2 A schematic diagram of a specific implementation process of a subway line deformation risk monitoring method based on InSAR and multi-source data provided by an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of time-sequential deformation along a subway line provided by an embodiment of the present invention;

[0061] Figure 4 A schematic diagram of a different method for solving a Logistic regression model provided by an embodiment of the present invention;

[0062] Figure 5 A schematic diagram of a random forest calculation of contribution values ​​of influencing factors provided by an embodiment of the present invention;

[0063] Figure 6 A schematic diagram of deformation prediction along a subway line using a different method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0065] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0066] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present invention, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0067] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0068] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0069] An embodiment of the present invention provides a subway line deformation risk monitoring method based on InSAR and multi-source data, and the method can be applied to subway line monitoring and early warning processes in geodetic scenes.

[0070] See also Figure 1 , is a flow chart of a subway line deformation risk monitoring method based on InSAR and multi-source data provided by an embodiment of the present invention. Figure 1 and Figure 2 As shown, the method mainly comprises the following steps:

[0071] Step 1: Use the differential interferometry short baseline set time series analysis technology to process the SAR data in the target area, obtain the time series deformation data and perform two-dimensional deformation decomposition on it, then use the logistic model to invert the time deformation characteristics, and use the Peck model to invert the spatial deformation characteristics, and monitor the current deformation of the target area based on the inverted time deformation characteristics and spatial deformation characteristics;

[0072] Furthermore, the expression of the logistic model is

[0073]

[0074] in, represents the maximum settlement value, a and b are the parameters of the model, and t represents the time series.

[0075] Furthermore, the step of inverting the time deformation characteristics using the logistic model includes:

[0076] The temporal deformation data after two-dimensional deformation decomposition is divided into three segments, among which the first segment is , The second paragraph is The third paragraph is ;

[0077] make is the reciprocal sum of the total settlement in the three sections, that is:

[0078]

[0079] Substituting it into the formula and performing linear transformation, we get:

[0080]

[0081] Use the least squares method to solve the parameters a and b and transform the above formula into

[0082]

[0083] make , , , ,in, represents the number of observations, represents the observed value, represents the observation vector, represents the coefficient matrix, Indicates the parameters to be requested. Indicates accidental error;

[0084] Written in matrix form:

[0085]

[0086] By least squares have to,

[0087]

[0088] ;

[0089] in, Indicates transpose.

[0090] Furthermore, the expression of the Peck model is:

[0091]

[0092]

[0093] in, represents the ground settlement, x represents the horizontal distance from the tunnel axis, represents the formation loss rate, represents the sedimentation tank width coefficient, is the tunnel excavation radius.

[0094] Furthermore, the step of inverting the spatial deformation characteristics using the Peck model includes:

[0095] Acquire surface deformation data of target area through InSAR monitoring technology;

[0096] The width of the sedimentation trough and the formation loss rate are calculated using the preset formula of the Peck model.

[0097] In the specific implementation, the differential interferometry short baseline set time series analysis technology SBAS-InSAR is used to process the SAR data in the study area, obtain the time series deformation data and perform two-dimensional deformation decomposition. The decomposition principle and steps are as follows:

[0098] Due to the principle of SAR image sensor oblique imaging, the displacement obtained by InSAR technology monitoring is the upward projection of the three-dimensional deformation of the surface point on the one-dimensional line of sight LOS, that is, the LOS displacement is the vector superposition of the three-dimensional deformation in the vertical direction (U), north-south direction (N), and east-west direction (E). There is the following geometric relationship between them:

[0099] ;

[0100] In the above formula, is the upward displacement of LOS, is the displacement in azimuth, θ is the radar incident angle, and α is the flight azimuth. We ignore the north-south deformation and project the line-of-sight deformation to the vertical direction:

[0101]

[0102] The specific steps of inverting the time deformation characteristics using the Logistic regression model can be shown as follows:

[0103] The Logistic time function model is used to invert the process of subway station deformation occurrence, development, maturity and saturation. The specific function model is expressed as:

[0104] ;

[0105] In the formula, represents the maximum settlement value, a and b are the parameters of the model, and t represents the time series.

[0106] This process first uses the three-stage method to solve the maximum settlement of the parameters: the data is divided into n / 3=r segments, the first segment is The second paragraph is ,... The third paragraph is .make is the reciprocal sum of the total settlement in the three sections, that is:

[0107] ;

[0108] Substituting it into the formula and performing linear transformation, we can get:

[0109] ;

[0110] Parameters a and b are solved using the least squares method, and the function model can be transformed into

[0111]

[0112] make ,in, represents the number of observations, represents the observed value, represents the observation vector, represents the coefficient matrix, Indicates the parameters to be requested. Indicates accidental error;

[0113] Written in matrix form:

[0114]

[0115] By least squares have to,

[0116]

[0117] ;

[0118] in, Indicates transpose.

[0119] The process of inverting spatial deformation characteristics using the Peck model can be shown as follows:

[0120] The Peck model is proposed based on the analysis of a large amount of tunnel excavation settlement data. It can be used to analyze the deformation characteristics of subway stations in space. Its curve formula is:

[0121]

[0122]

[0123] in: represents the ground settlement, x represents the horizontal distance from the tunnel axis, represents the formation loss rate, Sedimentation tank width coefficient, is the tunnel excavation radius.

[0124] The calculation steps of the Peck formula can be divided into three steps: obtaining the surface deformation data of the study area through InSAR monitoring technology; calculating the width of the sedimentation trough and the formation loss rate through the above formula, and performing regression calculation:

[0125] ;

[0126] Least squares estimates in a univariate linear model:

[0127]

[0128] The width k of the sedimentation tank can be calculated from the ratio b:

[0129] ;

[0130] The formation loss rate can also be calculated And the sedimentation tank width coefficient v:

[0131] ;

[0132] Step 2: Extract the deformation data along the subway line from the decomposed time series deformation data, and use the wavelet transform method to qualitatively and quantitatively explore the coupling relationship between the influencing factors and the deformation data along the subway line, and use the random forest method to calculate the characteristic proportion of each influencing factor;

[0133] Based on the above embodiment, the step of calculating the feature proportion of each influencing factor using the random forest method includes:

[0134] Perform bag sampling on the influencing factor set X=(x, y) to obtain a subset , train decision trees on subsets one by one , out-of-bag samples in the current decision tree Apply the decision tree to get the number of samples classified correctly , then for the feature , randomly shuffle the out-of-bag samples The d-th feature of constitutes D new out-of-bag samples ;

[0135] Make decision tree predictions on D new out-of-bag samples, and the correct number is recorded as , each sample subset performs the above calculations in turn, and finally calculates the importance score of the dth feature as the feature proportion, where the expression of the feature proportion is

[0136]

[0137] in, represents the proportion of the dth feature, M is the total number of test samples, m=1, 2, ..., M. In specific implementation, the method of the present invention can use the wavelet transform method to qualitatively and quantitatively explore the coupling relationship between the influencing factors and subway deformation, and use the random forest method to calculate the characteristic proportion of each factor.

[0138] a) Wavelet Transform

[0139] The continuous wavelet transform is used to calculate the period of the influencing factor itself, which will have equal time steps The discrete time series And the wavelet function after scaling and translation Perform convolution to obtain the continuous wavelet transform:

[0140] ;

[0141] In the formula is the wavelet transform coefficient, s is the scale expansion factor, is the time shift factor.

[0142] Cross wavelet transform can quantitatively evaluate the correlation and time lag relationship between the two. and First perform continuous wavelet transform and get the result and , then the cross wavelet spectrum is defined as:

[0143] ;

[0144] In the formula, and Sequential signals and The continuous wavelet transform result is: is the complex conjugate. is the power of the cross wavelet transform. Represents the phase delay between two timing signals, through the resonant cycle between the two , the phase can be converted into the amount of time delay between the two:

[0145]

[0146] b) Random Forest calculates feature importance

[0147] The importance of a feature indicates the degree of influence of different feature variables on the prediction results. The importance of a feature in random forest regression is the average of the importance of the feature in all internal decision trees. If the importance of a feature is greater, it means that the feature has a greater impact on the prediction results. The smaller the importance value, the smaller the impact on the prediction results. The sum of the importance values ​​of the regularized features is 1. Random forests usually use the average descent precision method to measure the importance of features:

[0148] Perform bag sampling on the original training data set X=(x, y) to obtain a sample subset , train decision trees on subsets one by one , out-of-bag samples in the current decision tree Apply the decision tree to get the number of samples classified correctly , then for the feature , randomly shuffle the out-of-bag samples The d-th feature of constitutes D new out-of-bag samples . Make decision tree predictions on D new out-of-bag samples, and the correct number is recorded as , each sample subset performs the above calculations in turn, and finally calculates the importance score of the d-th feature:

[0149]

[0150] Step 3: Input the time series deformation data into the GA-BP neural network to obtain the deformation value of the target area at the next moment, and calculate the risk level corresponding to the target area according to the feature proportion.

[0151] Furthermore, before step 3, the method further includes:

[0152] Determine the GA-BP neural network structure and obtain the initial population of the genetic algorithm by encoding its weights and thresholds;

[0153] The initial weights and thresholds of the GA-BP neural network are continuously updated according to the evolution of the individuals. The GA-BP neural network is trained with measured data and then the output is predicted. The mean square error function shown in the formula is used as the fitness function to calculate the fitness of each individual and determine whether it meets the optimization criteria. If it does, proceed to the next step. If it does not, select, crossover, and mutate the individuals according to their fitness to generate new individuals and re-judge their fitness until they meet the criteria.

[0154]

[0155] Where M is the total number of test samples, m=1, 2, …, M; is the true value of the test sample, Predict values ​​for the test sample;

[0156] The optimal individuals obtained above are split in order and used as the initial weights and thresholds of the BP neural network;

[0157] Perform forward propagation and back propagation of the GA-BP neural network, update the weights and thresholds, and then re-perform forward propagation and error judgment until the preset requirements are met to obtain the optimal weights and thresholds.

[0158] Furthermore, the step of inputting the time series deformation data into the GA-BP neural network to obtain the deformation value of the target area at the next moment includes:

[0159] The deformation values ​​of the last preset number of time nodes of the time series deformation data are selected as the input vector, input into the trained model, calculate the corresponding activation values ​​in each layer of neurons in turn, and use the output of the last layer of neurons as the deformation value at the next moment.

[0160] Furthermore, the step of calculating the risk value corresponding to the target area according to the feature proportion includes:

[0161] All influencing factors are divided into different preset types, and the susceptibility of each influencing factor is graded by the natural break point classification method. Then, the risk value of each preset type is calculated based on the characteristic proportion of the influencing factor to the deformation obtained by the random forest model as the weight. The preset types include topographic factors, hydrological environmental factors, and human activity factors. The expression of the risk value is:

[0162]

[0163] in, represents the score value of each category, n is the number of influencing factors of each category, i represents the number of preset categories, i=1, 2, 3, j represents the number of influencing factors of each category, j=1, 2, ..., n, and They represent the characteristic proportion and susceptibility level of the jth factor of the i-th category respectively;

[0164] A three-dimensional XOY coordinate system is established, where XYZ represents the risk values ​​of geological and geomorphological factors, hydrological and environmental factors, and human activity factors, respectively. The corresponding score values ​​are divided into multiple intervals (J, H, G) through the three axes. The risk importance level obtained by the comprehensive evaluation of the three is O, and the relationship between the three is expressed as O=F(x, y, z);

[0165] According to the geological and geomorphological factors X ( ), hydrological environmental factor Y ( ), geological and geomorphological factors Z ( ) to construct a The O value at the intersection of rows and columns is the final risk level.

[0166] In specific implementation, the GA-BP neural network method can be used to predict the deformation value of the subway node at the next moment.

[0167] a) Here, it is necessary to train the GA-BP neural network model. The specific process can be as follows:

[0168] ① Population initialization, that is: first determine the BP neural network structure, and obtain the initial population of the genetic algorithm by encoding its weights and thresholds.

[0169] ②According to the evolution of individuals, the initial weights and thresholds of the BP neural network are continuously updated. The output is predicted after the BP neural network is trained with measured data. The mean square error function shown in the formula is used as the fitness function to calculate the fitness of each individual and determine whether it meets the optimization criteria. If it does, turn to ③. If it does not, select, crossover, and mutate the individuals according to their fitness to generate new individuals, and re-judge the fitness until it meets the criteria.

[0170]

[0171] In the MSE formula, M is the total number of test samples; m=1, 2, ..., M; is the true value of the test sample, Predicted value for the test sample.

[0172] ③ Split the optimal individuals obtained above in order as the initial weights and thresholds of the BP neural network

[0173] ④Perform forward propagation and back propagation of the BP neural network, update the weights and thresholds, and then re-perform forward propagation and error judgment until they meet the requirements.

[0174] b) Deformation prediction

[0175] Through the above model training, the optimal weights are obtained and threshold Then select the deformation values ​​of the last few time nodes of the time series data as the input vector , input into the trained model, in each layer of neurons ( , ..., ) to calculate the corresponding activation values ​​in sequence:

[0176] , ,...

[0177]

[0178] is the final output value of the model, that is, the deformation value at the next moment predicted by the corresponding model.

[0179] When assessing risk, the process can be as follows:

[0180] The factors affecting subway deformation are divided into three categories: topographic factors, hydrological environmental factors, and human activity factors. Each factor is first divided into five categories using the natural break point classification method, corresponding to susceptibility 1 to 5 (U). Then, the important characteristic values ​​of the influencing factors on deformation obtained by the random forest model ( ) is used as the weight to calculate the risk score of each category.

[0181]

[0182] in, Represents the score value of each category, They represent the important eigenvalue and susceptibility value of the jth factor of the ith category respectively.

[0183] First, a three-dimensional XOY coordinate system is established, where XYZ represents the score values ​​of geological and geomorphological factors, hydrological and environmental factors, and human activity factors, respectively. The three axes can be combined with the actual situation to divide the corresponding score values ​​into several intervals, such as (J, H, G). The risk importance level obtained by the comprehensive evaluation of the three is O. The relationship between the three can be expressed by the function relationship O=F(x, y, z), which can be expressed in matrix form:

[0184] According to the geological and geomorphological factors X ( ), hydrological environmental factor Y ( ), geological and geomorphological factors Z ( ) to construct a The O value at the intersection of rows and columns is the final risk level.

[0185] The subway line deformation risk monitoring method based on InSAR and multi-source data provided in this embodiment proposes a wide-area, real-time subway deformation monitoring method using InSAR technology, which makes up for the shortcomings of traditional level monitoring methods such as long observation time, long time span and low spatial resolution; the least squares method is used to solve the Logistic regression model to improve the fitting accuracy; a subway deformation risk assessment method based on a hazard matrix is ​​proposed, which takes into account multiple influencing factors and conducts assessment by classification, avoids the interference of a single factor on the overall evaluation, and improves the effectiveness and reference significance of risk assessment.

[0186] The method of the present invention will be further described below in conjunction with a specific embodiment. The present invention uses Sentinel-1 to cover the SAR image of a subway line for comprehensive testing, which verifies the effectiveness and practicality of our proposed research process in subway deformation monitoring, and can provide a method reference for subway network security monitoring work in provinces and cities across the country.

[0187] Step 1: Inversion of space-time deformation

[0188] 1) Deformation field acquisition and two-dimensional decomposition

[0189] This example uses Sentinel-1 data collected in the study area from August 2016 to December 2023. The time-series deformation field obtained by SBAS-InSAR is used to extract the cumulative deformation values ​​along the subway line and decompose them into two dimensions to obtain the vertical deformation results. It can be seen that the spatial difference changes in settlement and the time-series evolution of the settlement funnel are as follows. Taking August 2016 as the starting time, the time-series cumulative deformation results for 2017, 2019 and 2023 are shown as follows: Figure 3 shown.

[0190] 2) Logistic model inversion of time series deformation characteristics

[0191] The subway stations were selected to extract the single-point time series, and the logistic regression model was used to perform time series inversion to obtain the life cycle of the subway stations, and three periods were obtained: the rapid deformation zone, the deformation slowing period, and the stable period. Figure 4As shown in the figure, (a) represents the least square method to solve the logistic regression model, and (b) represents the three-stage method to solve the logistic regression model. The corresponding RMSEs of the least square method and the three-stage method to solve the logistic regression model are 2.51 and 3.66, respectively, and the corresponding correlation coefficients are 0.9676 and 0.9306, respectively, as shown in Table 1. It can be seen that the least square method curve has a higher fitting accuracy with the actual monitored deformation. It can be found that the settlement of the subway station increased in the early stage, and the settlement continued to increase. In March 2018, the deformation rate began to decrease, and the settlement occurred slowly. In the summer of the same year, the deformation rate continued to decrease, and the settlement gradually stabilized.

[0192] Table 1

[0193]

[0194] Step 2: Evaluate the importance of influencing factors

[0195] 1) Wavelet transform to explore the periodic and phase relationship between site deformation and various factors

[0196] This application takes the deformation and rainfall data of subway stations A, B and C as examples, and performs wavelet continuous transform and cross transform respectively. The transformation results show that the rainfall at the three stations has a cycle of 4 to 6 months and 12 months. The time series deformation has a cycle of 3 to 6 months and 12 months. From the results of the cross wavelet transform, it can be seen that most of the phases of rainfall and time series deformation are consistent. At the same time, there is also a rainfall deformation of about 1 / 12 cycle in advance, which corresponds to about 1 month. This shows that there is a certain correlation between rainfall and deformation, and the overall deformation lags behind rainfall by 1 month.

[0197] 2) Random forest calculates the importance of influencing factors

[0198] This application uses temperature, population, groundwater and rainfall as sample data, and uses random forest to calculate the contribution of each factor to surface subsidence. First, the data is resampled to a consistent spatial resolution in Arcmap, and the target points with a subsidence greater than 60 mm in the study area are selected as dependent variables through a threshold. The random forest model is used to calculate the influence of influencing factors on ground subsidence. Figure 5 The results show that groundwater is of the highest importance, which is consistent with the conclusions of previous literature on the study area.

[0199] Step 3: Deformation prediction and risk assessment

[0200] Using different models to fit and predict single-point time series, by comparing the RMSE, MAE, and MAPE of the prediction results of BP, LSTM, PSO-BP, RF, and GA-BP models, it can be seen that GA-BP has the best accuracy, as shown in Table 2.

[0201] Table 2

[0202]

[0203] The model was applied to the deformation prediction along the subway line. The model was trained with the time series data of the first 50 scenes and short-term time series prediction was performed. The prediction results were consistent with the actual data trend, and the fitting accuracy reached 99%. Figure 6 As shown in the figure, (a) shows the comparison between the deformation prediction of the subway line at the next moment using GA-BP and the actual monitoring value, and (b) shows the fitting accuracy of the GA-BP prediction value and the InSAR monitoring value.

[0204] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof.

[0205] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A subway line deformation risk monitoring method based on InSAR and multi-source data, characterized in that: include: Step 1: Use the differential interferometry short baseline set time series analysis technology to process the SAR data in the target area, obtain the time series deformation data and perform two-dimensional deformation decomposition on it, then use the logistic model to invert the time deformation characteristics, and use the Peck model to invert the spatial deformation characteristics, and monitor the current deformation of the target area based on the inverted time deformation characteristics and spatial deformation characteristics; Step 2: Extract the deformation data along the subway line from the decomposed time series deformation data, and use the wavelet transform method to qualitatively and quantitatively explore the coupling relationship between the influencing factors and the deformation data along the subway line, and use the random forest method to calculate the characteristic proportion of each influencing factor; The step of using the wavelet transform method to qualitatively and quantitatively explore the coupling relationship between the influencing factors and the deformation data along the subway line includes: The influencing factors And deformation data along the subway line Perform continuous wavelet transform, get the result and define the cross wavelet spectrum based on it, through the resonance period between the two , converting the phase into the amount of time delay between the two; The step of using the random forest method to calculate the characteristic proportion of each influencing factor includes: The influencing factor set Perform bag sampling to obtain a subset , train decision trees on subsets one by one , out-of-bag samples in the current decision tree Apply the decision tree to get the number of samples classified correctly , then for the feature , randomly shuffle the out-of-bag samples The d-th feature of constitutes D new out-of-bag samples ; Make decision tree predictions on D new out-of-bag samples, and the correct number is recorded as , each sample subset performs the above calculations in turn, and finally calculates the importance score of the dth feature as the feature proportion, where the expression of the feature proportion is ； in, represents the proportion of the dth feature, M is the total number of test samples, m=1, 2, …, M; Step 3: Input the decomposed time series deformation data into the GA-BP neural network to obtain the deformation value of the target area at the next moment, and calculate the risk level corresponding to the target area according to the feature proportion.

2. The method according to claim 1, characterized in that , the expression of the logistic model is in, represents the maximum settlement value, a and b are the parameters of the model, and t represents the time series.

3. The method according to claim 2, characterized in that The step of inverting the time deformation characteristics using the logistic model includes: The temporal deformation data after two-dimensional deformation decomposition is divided into three segments, among which the first segment is , The second paragraph is ,... The third paragraph is ; make is the reciprocal sum of the total settlement in the three sections, that is: Substituting it into the formula and performing linear transformation, we get: Use the least squares method to solve the parameters a and b and transform the above formula into make , , , ,in, represents the number of observations, represents the observed value, represents the observation vector, represents the coefficient matrix, Indicates the parameters to be requested. Indicates accidental error; Written in matrix form: By least squares have to, ； in, Indicates transpose.

4. The method according to claim 3, characterized in that , the expression of the Peck model is in, represents the ground settlement, x represents the horizontal distance from the tunnel axis, represents the formation loss rate, represents the sedimentation tank width coefficient, is the tunnel excavation radius.

5. The method according to claim 4, characterized in that The step of inverting the spatial deformation characteristics using the Peck model includes: Acquire surface deformation data of target area through InSAR monitoring technology; The width of the sedimentation trough and the formation loss rate are calculated using the preset formula of the Peck model.

6. The method according to claim 5, characterized in that , before step 3, the method further includes: Determine the GA-BP neural network structure and obtain the initial population of the genetic algorithm by encoding its weights and thresholds; The initial weights and thresholds of the GA-BP neural network are continuously updated according to the evolution of the individuals. The output is predicted after the GA-BP neural network is trained with measured data. The mean square error function shown in the formula is used as the fitness function to calculate the fitness of each individual and determine whether it meets the optimization criteria. If it meets the criteria, it proceeds to the next step. If it does not meet the criteria, the individuals are selected, crossed, and mutated according to the fitness to generate new individuals, and the fitness is re-determined until it meets the criteria. Where M is the total number of test samples, m=1, 2, ..., M; is the true value of the test sample, Predict values ​​for the test sample; The optimal individuals obtained above are split in order and used as the initial weights and thresholds of the BP neural network; Perform forward propagation and back propagation of the GA-BP neural network, update the weights and thresholds, and then re-perform forward propagation and error judgment until the preset requirements are met to obtain the optimal weights and thresholds.

7. The method according to claim 6, characterized in that The step of inputting the time series deformation data into the GA-BP neural network to obtain the deformation value of the target area at the next moment includes: The deformation values ​​of the last preset number of time nodes of the time series deformation data are selected as the input vector, input into the trained model, calculate the corresponding activation values ​​in each layer of neurons in turn, and use the output of the last layer of neurons as the deformation value at the next moment.

8. The method according to claim 7, characterized in that ,The step of calculating the risk value corresponding to the target area according to the feature proportion includes: All influencing factors are divided into different preset types, and the susceptibility of each influencing factor is graded by the natural break point classification method. Then, the risk value of each preset type is calculated based on the characteristic proportion of the influencing factor to the deformation obtained by the random forest model as the weight. The preset types include topographic factors, hydrological environmental factors, and human activity factors. The expression of the risk value is: in, represents the score value of each category, n is the number of influencing factors of each category, i represents the number of preset categories, i=1,2,3, j represents the number of influencing factors of each category, j=1,2,...,n, and They represent the characteristic proportion and susceptibility level of the jth factor of the i-th category respectively; A three-dimensional XOY coordinate system is established, where XYZ represents the risk values ​​of geological and geomorphological factors, hydrological and environmental factors, and human activity factors, respectively. The corresponding score values ​​are divided into multiple intervals (J, H, G) through the three axes. The risk importance level obtained by the comprehensive evaluation of the three is O, and the relationship between the three is expressed as O=F(x, y, z); According to the geological and geomorphological factors X ( ), hydrological environmental factor Y ( ), geological and geomorphological factors Z ( ) to construct a The O value at the intersection of rows and columns is the final risk level.

Images