Method for judging reinforcement failure of reinforced retaining wall based on apparent displacement monitoring

Abstract

The invention discloses a reinforced retaining wall reinforcement failure judgment method based on apparent displacement monitoring, which comprises the following steps: S1, determining the apparent displacement of a site reinforced retaining wall monitoring point, and carrying out site investigation; s2, performing screening, classification, interpolation and integration by adopting a Kriging method; s3, a theoretical model of the reinforced retaining wall is constructed, and the theoretical distribution range of the elastic modulus of the rib material is determined; s4, performing numerical simulation analysis on the theoretical model of the reinforced retaining wall to obtain a corresponding analysis result of the target displacement; s5, establishing a neural network model, and learning a mapping relation between the elastic modulus of the layered rib material and the displacement of the surface monitoring point; s6, performing inversion through the neural network model to obtain the elastic modulus of the layered rib material; and S7, analyzing the relative relationship between the elastic modulus of the rib material and the elastic modulus of the soil body, and predicting whether pull-out failure occurs or not. The actual working state of the retaining wall can be accurately captured, and comprehensive evaluation and failure prediction of the stability of the reinforced retaining wall are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of civil engineering operation and maintenance, and in particular to a method for determining reinforcement failure of a reinforced retaining wall based on apparent displacement monitoring. Background Art

[0002] As an economical and efficient retaining structure, reinforced retaining walls are widely used in the construction of infrastructure such as roads, railways, and water conservancy. The structural safety and stability evaluation of reinforced retaining walls during their service life is of great significance for the reinforcement and maintenance of embankment slopes at this stage and for ensuring the safety of road operations. The failure forms of reinforced retaining walls mainly include external stability failure and internal stability failure. External failure generally manifests itself as overall structural instability and overturning; internal failure mainly manifests itself as tensile failure and pull-out failure of the reinforcement materials, as well as connection failure due to the connection between the reinforcement and the panel. In actual engineering, the failure of reinforced retaining walls often presents itself as comprehensive failure, with multiple failure effects occurring together and coupled with each other.

[0003] At present, the methods for analyzing and predicting the stability of reinforced retaining walls are mainly divided into two categories: one is the method based on the solution of physical and mechanical models, and the other is the method using monitoring data feedback analysis. The methods based on the solution of physical and mechanical models mainly include the limit equilibrium method and the finite element method. The limit equilibrium method is to analyze the strength of the reinforcement material required for the internal and external stability of the overall structure of the wall, that is, the stress distribution. For flexible reinforced retaining walls, the most commonly used method is the anchor wedge method. Compared with the limit equilibrium method, the finite element method can calculate the displacement, stress, and strain levels of each point in the soil, and can simulate the construction process and consider complex boundary conditions, so as to more accurately evaluate the stability of reinforced earth retaining walls and provide a scientific basis for engineering design and construction operation and maintenance. However, the parameters of the finite element method need to be determined through complex experiments, and it is difficult to make quantitative judgments on the damage, which limits its applicability in engineering.

[0004] The starting point of the monitoring data feedback analysis method is opposite to that of the physical mechanics model solution method. This method mainly conducts model learning to analyze and predict the structural stability of the reinforced retaining wall. However, the traditional neural network model has certain limitations. For example, the BP neural network uses static modeling, which is less practical for highly nonlinear and time-correlated data. Summary of the invention

[0005] In order to solve the problem of analyzing and predicting the stability of reinforced retaining walls by existing methods based on physical and mechanical model solutions and monitoring data feedback analysis methods, the present invention proposes a reinforcement failure determination method for reinforced retaining walls based on apparent displacement monitoring to solve the above problems.

[0006] The present application discloses a method for determining reinforcement failure of a reinforced retaining wall based on apparent displacement monitoring, comprising the following steps: S1. Determine the apparent displacement of the monitoring point of the on-site reinforced retaining wall, and conduct on-site survey of the location of the reinforced retaining wall to obtain the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure; S2. Based on the physical and mechanical parameters of the surrounding soil and geological data obtained from the on-site survey, the Kriging method is used for screening, classification, interpolation and integration; S3. According to the arrangement form and distribution law of the reinforcement in the retaining wall structure system, the theoretical distribution range of the reinforcement elastic modulus is determined, and the elastic modulus estimation of the reinforcement is generated by random sampling in uniform distribution, and the characteristic distribution is completed according to its arrangement form and distribution law in the reinforced retaining wall; S4. Based on the on-site geological characteristics of the reinforced retaining wall after interpolation, the physical parameters of the reinforcement material generated by random sampling are used as variation characteristics to construct a numerical simulation model of the retaining wall and perform numerical simulation analysis to obtain the corresponding analysis results of the target displacement; S5. Establish a neural network model, analyze the results of several monitoring points in the actual working point corresponding to the numerical simulation model of the retaining wall, and learn the mapping relationship between the elastic modulus of the layered reinforcement and the displacement of the surface monitoring point, until the numerical simulation model of the retaining wall passes the test; S6. Importing the surface displacement data of the reinforced retaining wall measured on site into the tested neural network model, and obtaining the elastic modulus of the layered reinforcement material through model inversion; S7. Feedback the reinforcement elastic modulus result to the actual measurement point of the project, and analyze the relative relationship between the reinforcement elastic modulus and the soil elastic modulus in combination with the actual distribution of the reinforcement, so as to predict whether the geogrid will fail due to pull-out.

[0007] Preferably, the S1 comprises the following steps: S11. Determine the on-site monitoring points of the reinforced retaining wall, deploy equipment including earth pressure cells and flexible displacement meters, and mark the monitoring points; S12. Carry out geological survey on the slope soil that affects the reinforced retaining wall, select multiple survey points, and obtain the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure, including the elastic modulus of the reinforcement, soil density, soil elastic modulus, soil Poisson's ratio, interface cohesion between soil and reinforcement, and interface friction angle between soil and reinforcement.

[0008] Preferably, S2 comprises the following steps: S21. Integrate the physical and mechanical parameter values ​​of the surrounding soil, including soil density and elastic properties, and mark the coordinates of the corresponding survey points; S22. According to the first law of geography, select points within the effective influence range of Kriging interpolation as points to be interpolated, and mark the coordinates of the points to be interpolated; S23, determine the mathematical form of Kriging interpolation, set 0 as the point to be interpolated, is the surrounding survey point, then the estimated value of a regionalized material parameter at 0 is:

[0009] in, Indicates material parameters in The estimated value at Indicates material parameters in The observed value at is the weight coefficient, represents the regionalized material parameters for the calculation, is the Lagrange constant, For materials located at The higher-order trend function at is the correlation coefficient of the higher-order trend function, is the number of higher-order trend functions; Minimize the difference between the expected and variance of the estimated value and the true value of the material parameter at the point to be interpolated, that is:

[0010]

[0011] in, is the true value of the material parameter at the point to be inserted, is the estimated value of the material parameter at the point to be inserted; S24. Calculation The estimated value of the distance between the material parameters and the corresponding strain difference is constructed based on a certain type of material parameter of a single point to be interpolated. The Kriging coefficient matrix is ​​in the form of:

[0012] The Kriging interpolation expansion matrix of the material to be interpolated is constructed in the form of:

[0013] in, To be inserted At known points The variation value between the material parameters at is the covariance matrix of the variable itself, is the covariance matrix between different variables; S25. After calculating the material parameter distance and the corresponding strain difference estimation results, the estimated value of the variogram is calculated by the following formula:

[0014] in, The spacing is The total number of all point pairs, For Point deviation The measured value of the variable; S26, will The distances and corresponding strain differences between the data points are divided into several groups, each containing distance values, and calculate the average distance of each group of distance values and the mean variation , select the Gaussian model as the variogram theoretical model, fit the variogram and regress each parameter of the Gaussian model, so as to obtain the fitted variogram model; S27, the distance value between the survey point and the point to be inserted is brought into the fitted variogram to calculate the variogram value of the point to be inserted , and put all the variance values ​​into the formula In the equation, find the weight coefficient of the point to be inserted , then according to the formula Obtain the true value of the material parameter of the point to be inserted; S28, repeat S25 to S27 until the material parameter values ​​of all selected points to be inserted are determined in sequence; S29, repeat S25 to S28 until the material parameters of all selected categories are determined in sequence.

[0015] Preferably, S3 comprises the following steps: S31. Determine the arrangement and distribution pattern of reinforcement in the retaining wall structure system, including quantity, horizontal and vertical spacing, and anchorage length; S32. Make the overall sample of the given reinforcement elastic modulus obey uniform distribution , the sample parameters of the elastic modulus of the reinforcement after integration and interpolation are ; S33. The uniform distribution form of the elastic modulus of the reinforcement is solved by using the Gauss-Cauchy mixed density estimation method. The kernel density estimation is expressed as:

[0016] in, Indicate point The estimated elastic modulus at is the number of samples of tendon elastic modulus, For the The elastic modulus of the tendon sample is is the bandwidth of the Gaussian kernel, is the bandwidth of the Cauchy kernel, is the weight coefficient of Gaussian kernel density estimation, is the weight coefficient of the Cauchy kernel density estimate; bandwidth The optimal value is calculated according to the Silverman rule, namely:

[0017] in, is the standard deviation of the tendon elastic modulus samples; S34. Analyze the results of kernel density estimation. By drawing and observing the kernel density estimation graph, identify the area where the density curve is significantly higher than zero, determine the support interval of the data, that is, the effective range of the data distribution, and take the lower limit of this interval as the uniform distribution parameter of the tendon body. The upper limit of this interval is taken as the uniform distribution parameter of the tendon The value of S35. Determine the uniform distribution of the elastic modulus of the reinforcement based on the solved parameters, and take random sampling of the uniform distribution by the sine-linear composite congruential method, as shown in the following formula:

[0018] in, The random number sequence to be determined in the next step values, Indicates the currently determined A random number, is the multiplier, is the increment, is the modulus, is the weight coefficient of the sine function on the random number sequence, is the frequency parameter of the sine function; Determine the upper and lower boundaries of the distribution, and map the normalized sampled random numbers to the distribution interval to determine the elastic modulus of the reinforcement; S36. Arrange the elastic modulus of the randomly selected reinforcement material according to its arrangement form and distribution law in the reinforced retaining wall to complete the characteristic distribution.

[0019] Preferably, S4 comprises the following steps: S41. Based on the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure obtained from the on-site survey in S1, a three-dimensional model of the retaining wall is established in sequence, the analysis steps for calculating the retaining wall simulation model are determined, the contact conditions of the surrounding soil, the retaining wall and the reinforcement are defined, loads and constraints are created and meshing is performed, the preliminary establishment of the numerical simulation model of the retaining wall is completed, the results are submitted and an inp file is generated; S42, constructing an Abaqus random field model according to various soil material parameters after interpolation and the elastic modulus parameters of the reinforcement obtained by random extraction, assigning the soil material parameters to the grid cells, and assigning the elastic modulus of the reinforcement to the reinforcement component material; S43, write a programming script, generate inp files in batches according to the results of the random field and import them into the Abaqus software, set a batch calculation task to run all the generated inp files, and obtain the target displacement analysis results; S44. Arrange the reinforcement elastic modulus parameter values ​​and the corresponding target displacement results in all inp files, and determine whether the reinforced retaining wall has failed based on the displacement.

[0020] Preferably, S5 comprises the following steps: S51, using the KNN weighted proximity algorithm to remove noise and clean the abnormal values ​​in the elastic modulus parameter value and the corresponding retaining wall target displacement value caused by calculation errors in S4; S52. Calculate local weighted anomaly factor :

[0021] in, is the normalized elastic modulus corresponding to the apparent displacement data point The average nearest neighbor distance, is the normalized elastic modulus corresponding to the apparent displacement data point The average distance between data points, is the comprehensive weighting factor;

[0022] in, is the weighting factor bias coefficient, is the distance-based weight factor, is the weighting factor for the correlation between the elastic modulus and the apparent displacement; S53, integrate the local weighted anomaly factor and standardized data to define the outlier discrimination index :

[0023] in, is the weight parameter, is the weighted average of the normalized apparent displacement distance, is the median absolute deviation of the normalized apparent displacement distance; Assuming the anomaly threshold is three times the median absolute deviation, if there is , then it is believed that If a point is an outlier, find the non-outlier value closest to the point to replace it; S54, forming a data set based on the apparent displacement data of the elastic modulus after cleaning, dividing the data set into a training set and a test set, establishing a deep support vector machine regression model, and using a radial basis function as a kernel function; Selecting the best support vector regression parameters using Bayesian optimization: Regularization parameter , and error tolerance , determine the value ranges of the three types of parameters in turn; Use the negative mean square error of the apparent displacement observations as the optimization objective:

[0024] in, is the true value of the apparent displacement in the test set, is the predicted value of the apparent displacement in the test set, is the number of test set samples; S55. Select Gaussian process EI as the proxy model and use the acquisition function to select the next optimization point:

[0025] Repeat the optimization process until the optimal parameters with the lowest mean square error are found; S56. Construct a deep support vector regression model, using a multi-layer support vector regression structure to capture the data pattern of elastic modulus corresponding to apparent displacement. The model output is expressed as:

[0026] in, is the number of layers in the deep support vector regression model, is the radial basis function kernel function, is the weight of each layer, is the bias term; S57. The deep support vector regression model is trained using the training set. The optimal parameters determined by Bayesian optimization are used to train the deep support vector regression model. A regularization term is introduced to prevent overfitting. The objective function is shown in the following formula:

[0027] in, is the weight vector, and is the slack variable, is a hyperparameter that controls the strength of regularization. is the regularization term; The objective function optimization is constrained by the following conditions:

[0028] in, is the kernel function mapping, As input, a randomly generated set of reinforcement elastic moduli is For the set of monitoring displacement observation results outputted, the upper and lower limit constraints of the prediction results are specified, and the deep support vector regression model is fitted to obtain the objective function; S58, apply the model trained in S57 to the test set for prediction. A test set of samples was created to evaluate the accuracy of the support vector regression model for determining the depth of reinforced soil displacement observations by using mean square error and mean absolute error. S59. Analyze the results of the mean square error and the mean absolute error to evaluate the prediction performance of the deep support vector regression model. If one of the results of the mean square error and the mean absolute error is higher than 10, modify the parameter value range and the Bayesian optimization proxy model through the methods of S54-S55 to further correct the parameters of the deep support vector regression model until the deep support vector regression model has good prediction accuracy.

[0029] Preferably, S6 comprises the following steps: S61, obtaining displacement data of the soil measured on site by means of a flexible displacement meter, and checking the integrity and continuity of the data; S62, for abnormal displacement data, the KNN weighted proximity algorithm of S51-S53 is used to clean the data, the limit of the abnormal value is calculated according to the degree of variability, the abnormal value is identified, and the abnormal value is replaced; S63. Construct a feature matrix based on the cleaned displacement data, use a deep support vector regression model to predict the input displacement data, and output the corresponding reinforcement elastic modulus prediction value.

[0030] Preferably, the S7 comprises the following steps: If the elastic modulus of the reinforcement and the elastic modulus of the soil satisfy any of the following formulas, the prediction results indicate that the geocell will fail by pull-out:

[0031]

[0032]

[0033] in, is the effective reinforcement system number, is the elastic modulus of the tendon, is the elastic modulus of the soil, is the number of reinforcements in the horizontal direction of the retaining wall, is the number of reinforcements in the longitudinal direction of the retaining wall, It is a reinforcement object at any horizontal row of the retaining wall. It is a reinforcement object at any longitudinal position of the retaining wall.

[0034] Effective reinforcement system number Satisfies the following formula:

[0035] in, is the actual length of the tendon, is the effective length of the tendon.

[0036] Beneficial effects of the present invention: (1) The present invention starts from on-site monitoring and geological survey conditions, collects apparent displacement data from on-site monitoring points, and combines the first-hand soil physical and mechanical parameters obtained from geological surveys to accurately capture the actual working status of the retaining wall and provide effective data support for prediction and analysis.

[0037] (2) The present invention uses Kriging interpolation to perform spatial interpolation of geological data, effectively expanding local soil properties to the overall area that affects the stability of the retaining wall, thereby improving the continuity and integrity of the spatial distribution of parameters.

[0038] (3) The present invention simulates the actual distribution of the elastic modulus of the reinforcement material through a random sampling method and constructs a numerical simulation model of the retaining wall, so that the reinforcement failure is more in line with the laws of physics.

[0039] (4) The present invention introduces a neural network model based on SVR to achieve accurate inversion of the elastic modulus of the reinforcement, thereby predicting the stability of the retaining wall.

[0040] (5) The invention realizes a comprehensive assessment of the stability and failure prediction of reinforced retaining walls, and provides a new technical means for stability assessment and failure warning of reinforced retaining walls. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flow chart of a method for determining reinforcement failure of a reinforced retaining wall based on apparent displacement monitoring according to an embodiment of the present invention; Figure 2 It is a layout diagram of geogrid monitoring point instruments according to an embodiment of the present invention; Figure 3 A schematic diagram of the actual reinforcement layout of a reinforced retaining wall according to an embodiment of the present invention; Figure 4 This is a flowchart for implementing the use of a deep support vector regression model to predict elastic modulus according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and examples.

[0043] The embodiment of the present invention discloses a method for determining reinforcement failure of a reinforced retaining wall based on apparent displacement monitoring, the process of which is as follows: Figure 1 As shown, the following steps are included: S1. Determine the apparent displacement of the monitoring point of the on-site reinforced retaining wall, and conduct on-site investigation of the location of the reinforced retaining wall. Based on physical exploration, drilling and other means, obtain the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure.

[0044] S11, determine the monitoring points of the reinforced retaining wall on-site engineering, deploy equipment including earth pressure cells and flexible displacement meters, and mark the monitoring points. Figure 2 As shown, a total of Layer, number of layers Determined by the actual project, a number of flexible displacement meters (>3) are set on each layer, and the distance between layers is staggered. The earth pressure box is set just above the monitoring point of each layer and close to the back side of the reinforced retaining wall, totaling The marking of the monitoring point includes the geometric position of the point, the apparent displacement of the point and the specific value of the earth pressure.

[0045] S12. Conduct geological surveys on the slope soil that affects the reinforced retaining wall, and select multiple survey points. In this embodiment, the survey points are randomly selected above the soil on the back of the reinforced retaining wall, and the plane distance between the points is not less than 20 cm. The survey mainly includes the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure, including but not limited to the elastic modulus of the reinforcement, the soil density, the elastic modulus of the soil, the Poisson's ratio of the soil, the interface cohesion between the soil and the reinforcement, and the interface friction angle between the soil and the reinforcement.

[0046] S2. Based on the physical and mechanical parameters of the surrounding soil and geological data obtained from the on-site survey, the Kriging method is used for screening, classification, interpolation and integration. The Kriging method is used for interpolation of geological data, and the material properties of the local soil of the embankment slope are interpolated and extended to the overall part that affects the stability of the retaining wall.

[0047] S21. Integrate the physical and mechanical parameter values ​​of the surrounding soil, including soil density and elastic properties, and mark the coordinates of the corresponding survey points.

[0048] S22. According to the first law of geography, select points within the effective influence range of Kriging interpolation as points to be interpolated, and mark the coordinates of the points to be interpolated.

[0049] S23, determine the mathematical form of Kriging interpolation, set 0 as the point to be interpolated, is the surrounding survey point, then the estimated value of a regionalized material parameter at 0 is:

[0050] in, Indicates material parameters in The estimated value at Indicates material parameters in The observed value at is the weight coefficient, represents the regionalized material parameters for the calculation, is the Lagrange constant, For materials located at The higher-order trend function at is the correlation coefficient of the higher-order trend function, is the number of higher-order trend functions; To ensure unbiased estimation, the difference between the expected value and the true value of the material parameter at the point to be interpolated is minimized, that is, it satisfies:

[0051]

[0052] in, is the true value of the material parameter at the point to be inserted, is the estimated value of the material parameter at the point to be interpolated.

[0053] S24. Calculation The estimated value of the distance between the material parameters and the corresponding strain difference is constructed based on a certain type of material parameter of a single point to be interpolated. The Kriging coefficient matrix is ​​in the form of:

[0054] Based on this matrix, the cross-correlation between multiple material variables is considered, and the Kriging interpolation expansion matrix of the material to be interpolated point is constructed in the form of:

[0055] in, To be inserted At known points The variation value between the material parameters at is the covariance matrix of the variable itself, is the covariance matrix between different variables.

[0056] S25. After calculating the material parameter distance and the corresponding strain difference estimation results, the estimated value of the variogram is calculated by the following formula:

[0057] in, For distance The variation value of The spacing is The total number of all point pairs, For Point deviation The measured value of the variable.

[0058] S26, will The distances and corresponding strain differences between the data points are divided into several groups, each containing distance values, and calculate the average distance of each group of distance values and the mean variation , select the Gaussian model as the variogram theoretical model, and the expression of the variogram is:

[0059] in, is the error value caused by data variability within a short distance, is the theoretical maximum value of the variogram, is the range of the variogram, To adjust the curve smoothness parameter.

[0060] The variogram is fitted and the parameters of the Gaussian model are regressed to obtain the fitted variogram model.

[0061] S27, the distance value between the survey point and the point to be inserted is brought into the fitted variogram to calculate the variogram value of the point to be inserted , and put all the variance values ​​into the formula In the equation, find the weight coefficient of the point to be inserted , then according to the formula Obtain the true value of the material parameter of the point to be inserted.

[0062] S28. Repeat S25 to S27 until the material parameter values ​​of all selected points to be inserted are determined in sequence.

[0063] S29, repeat S25 to S28 until the material parameters of all selected categories are determined in sequence.

[0064] S3. Determine the theoretical distribution range of the elastic modulus of the reinforcement based on the arrangement and distribution pattern of the reinforcement in the retaining wall structure system, generate an estimate of the elastic modulus of the reinforcement using random sampling in uniform distribution, and complete the characteristic distribution based on its arrangement and distribution pattern in the reinforced retaining wall.

[0065] S31, such as Figure 3As shown in the figure, determine the arrangement and distribution pattern of reinforcement in the retaining wall structure system, including quantity, horizontal and vertical spacing, anchorage length, etc. The anchorage length and the horizontal and vertical spacing are determined by the actual situation on site. The reinforcement is arranged equidistantly in the horizontal direction and unequally in the vertical direction.

[0066] S32. Make the overall sample of the given reinforcement elastic modulus obey uniform distribution , the sample parameters of the elastic modulus of the reinforcement after integration and interpolation are .

[0067] S33. The uniform distribution form of the elastic modulus of the reinforcement is solved by using the Gauss-Cauchy mixed density estimation method. The kernel density estimation is expressed as:

[0068] in, Indicate point The estimated elastic modulus at is the number of samples of tendon elastic modulus, For the The elastic modulus of the tendon sample is is the bandwidth of the Gaussian kernel, is the bandwidth of the Cauchy kernel, is the weight coefficient of Gaussian kernel density estimation, is the weight coefficient of the Cauchy kernel density estimate.

[0069] bandwidth The optimal value is calculated according to the Silverman rule, namely:

[0070] in, is the standard deviation of the tendon elastic modulus samples.

[0071] S34. Analyze the results of kernel density estimation. By drawing and observing the kernel density estimation graph, identify the area where the density curve is significantly higher than zero, determine the support interval of the data, that is, the effective range of the data distribution, and take the lower limit of this interval as the uniform distribution parameter of the tendon body. The upper limit of this interval is taken as the uniform distribution parameter of the tendon The value of .

[0072] S35. Determine the uniform distribution of the elastic modulus of the reinforcement based on the solved parameters, and take random sampling of the uniform distribution by the sine-linear composite congruential method, as shown in the following formula:

[0073] in, The random number sequence to be determined in the next step values, Indicates the currently determined A random number, is the multiplier, is the increment, and Using Park-Miller constants, the established model is a high-parameter system. is the modulus, in this embodiment The value is or , is the weight coefficient of the sine function on the random number sequence, is the frequency parameter of the sine function.

[0074] The upper and lower boundaries of the distribution are determined, and the random numbers obtained by sampling after normalization are mapped to the distribution interval to determine the elastic modulus of the reinforcement.

[0075] S36. Arrange the elastic modulus of the randomly selected reinforcement material according to its arrangement form and distribution law in the reinforced retaining wall to complete the characteristic distribution.

[0076] S4. Based on the on-site geological characteristics of the reinforced retaining wall after interpolation, the physical parameters of the reinforcement material generated by random sampling are used as the variation characteristics. A numerical simulation model of the retaining wall is constructed and numerical simulation analysis is performed to obtain the corresponding analysis results of the target displacement.

[0077] S41. Based on the physical and mechanical parameters of the surrounding soil and the structural mechanical parameter characteristics of the retaining wall obtained from the on-site survey in S1, a three-dimensional model of the retaining wall is established in sequence, the analysis steps for calculating the retaining wall simulation model are determined, the contact conditions of the surrounding soil, retaining wall and reinforcement are defined, loads and constraints are created and meshing is performed, the preliminary establishment of the numerical simulation model of the retaining wall is completed, the results are submitted and an inp file is generated.

[0078] S42. An Abaqus random field model is constructed based on the interpolated soil material parameters and the randomly extracted reinforcement elastic modulus parameters, the soil material parameters are assigned to the grid cells, and the reinforcement elastic modulus is assigned to the reinforcement component material.

[0079] S43. Write a programming script to batch generate inp files according to the results of the random field and import them into the Abaqus software. Set a batch calculation task to run all the generated inp files to obtain the target displacement analysis results.

[0080] S44. Arrange the reinforcement elastic modulus parameter values ​​and the corresponding target displacement results in all inp files, refer to the reinforced earth retaining wall specifications, and determine whether the reinforced retaining wall has failed based on the displacement.

[0081] S5, such as Figure 4As shown in the figure, a neural network model is established to analyze the results of several monitoring points in the actual working point corresponding to the numerical simulation model of the retaining wall, and learn the mapping relationship between the elastic modulus of the layered reinforcement and the displacement of the surface monitoring points until the numerical simulation model of the retaining wall passes the test.

[0082] S51, the elastic modulus parameter value in S4 and the corresponding retaining wall target displacement value are de-noised by using the KNN weighted proximity algorithm. And the apparent displacement data Standardization is performed, as shown in formula (13), and the standardized elastic modulus data is obtained And the apparent displacement data . Assume that the two sets of data correspond one to one to form two-dimensional coordinate points ( )、( ),……( ), traverse any two points in two-dimensional space ( )and( ), perform weighted KNN distance calculation:

[0083]

[0084] in, is the coordinate of any point in two-dimensional space, is the corresponding coordinate of any other point in the two-dimensional space, is the Euclidean distance between any two points in two-dimensional space, is the distance-based weighting factor.

[0085]

[0086] in, is the normalized elastic modulus, is the normalized apparent displacement data, is the mean of the elastic modulus and the apparent displacement, is the standard deviation of the elastic modulus and apparent displacement.

[0087] The similarity measure between each point is calculated using cosine similarity to obtain a weighting factor for the correlation between elastic modulus and apparent displacement:

[0088] in, is the average value of the standardized elastic modulus sample data, The average value of the normalized apparent displacement data.

[0089] S52. Calculate local weighted anomaly factor :

[0090] in, is the normalized elastic modulus corresponding to the apparent displacement data point The average nearest neighbor distance, is the normalized elastic modulus corresponding to the apparent displacement data point The average distance between data points, is the comprehensive weighting factor.

[0091]

[0092] in, It is the weighting factor, and its value range is between 0 and 1.

[0093] S53, integrate the local weighted anomaly factor and standardized data to define the outlier discrimination index :

[0094] in, is a weight parameter, ranging from 0 to 1. is the weighted average of the normalized apparent displacement distance, is the median absolute deviation of the normalized apparent displacement distance.

[0095] Assuming the anomaly threshold is three times the median absolute deviation, if there is , then it is believed that If a point is an outlier, find the non-outlier value closest to it to replace it.

[0096] S54, forming a data set based on the apparent displacement data corresponding to the elastic modulus after cleaning, dividing the data set into a training set and a test set, establishing a deep support vector machine regression model, and using a radial basis function (RBF) as a kernel function.

[0097] Using Bayesian optimization to select the best support vector regression (SVR) parameters: Regularization parameter , and error tolerance , determine the value range of the three types of parameters in turn: .

[0098] Use the negative mean square error of the apparent displacement observations as the optimization objective:

[0099] in, is the true value of the apparent displacement in the test set, is the predicted value of the apparent displacement in the test set, is the number of samples in the test set.

[0100] S55. Select Gaussian process EI as the proxy model and use the acquisition function to select the next optimization point:

[0101] The optimization process is repeated until the optimal parameters with the lowest mean square error are found.

[0102] S56. Construct a deep support vector regression model, using a multi-layer support vector regression structure to capture the data pattern of elastic modulus corresponding to apparent displacement. The model output is expressed as:

[0103] in, is the number of layers in the deep support vector regression model, is the radial basis function kernel function, is the weight of each layer, is the bias term.

[0104] S57. Use the training set to train the deep support vector regression model. Use the optimal parameters determined by Bayesian optimization to train the deep support vector regression model. Introduce a regularization term to prevent the model from overfitting. The objective function is shown in the following formula:

[0105] in, is the weight vector, and is the slack variable, is a hyperparameter that controls the strength of regularization. is the regularization term.

[0106] The objective function optimization is constrained by the following conditions:

[0107] in, is the kernel function mapping, As input, a randomly generated set of reinforcement elastic moduli is is the set of output monitoring displacement observation results. The upper and lower limits of the prediction results are specified, that is, the maximum and minimum values ​​of the monitoring displacement historical data are taken as the preliminary upper and lower limits. Finally, the deep support vector regression model is fitted to obtain the objective function.

[0108] S58, apply the model trained in S57 to the test set for prediction. A test set samples were selected and the accuracy of the support vector regression model for determining the depth of reinforced soil displacement observation was evaluated by the mean square error (MSE) and mean absolute error (MAE).

[0109]

[0110]

[0111] in, For the The actual displacement observation results of samples are For the The predicted displacement observation results of samples.

[0112] S59. Analyze the results of mean square error and mean absolute error to evaluate the prediction performance of the deep support vector regression model. If one of the results of MSE and MAE is higher than 10, modify the parameter value range and Bayesian optimization proxy model through the methods of S54-S55, and further correct the parameters of the deep support vector regression model until the MSE and MAE values ​​after the last fitting are no more than 10% different from the MSE and MAE values ​​of the deep support vector regression model after the previous fitting, indicating that the deep support vector regression model has good prediction accuracy.

[0113] S6, such as Figure 4 As shown in the figure, the surface displacement data of the reinforced retaining wall measured on site is imported into the tested neural network model, and the elastic modulus of the layered reinforcement is obtained through model inversion.

[0114] S61. Obtain the displacement data of the soil measured on site by means of a flexible displacement meter, and check the integrity and continuity of the data.

[0115] S62. For abnormal displacement data, the KNN weighted proximity algorithm of S51-S53 is used to clean the data, the limit of the outlier is calculated according to the degree of variability, the outlier is identified, and the outlier is replaced.

[0116] S63. Construct a feature matrix based on the cleaned displacement data, use a deep support vector regression model to predict the input displacement data, and output the corresponding reinforcement elastic modulus prediction value.

[0117] S7. Feedback the reinforcement elastic modulus result to the actual measurement point of the project, and analyze the relative relationship between the reinforcement elastic modulus and the soil elastic modulus in combination with the actual distribution of the reinforcement, so as to predict whether the geogrid will fail due to pull-out.

[0118] If the elastic modulus of the reinforcement and the elastic modulus of the soil satisfy any of the following formulas, the prediction results indicate that the geocell will fail by pull-out:

[0119]

[0120]

[0121] in, is the effective reinforcement system number, is the elastic modulus of the tendon, is the elastic modulus of the soil, is the number of reinforcements in the horizontal direction of the retaining wall, is the number of reinforcements in the longitudinal direction of the retaining wall, It is a reinforcement object at any horizontal row of the retaining wall. It is a reinforcement object at any longitudinal position of the retaining wall.

[0122] Effective reinforcement system number Satisfies the following formula:

[0123] in, is the actual length of the tendon, is the effective length of the tendon.

[0124] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring, characterized in that: The following steps are involved: S1. Determine the apparent displacement of the monitoring point of the on-site reinforced retaining wall, and conduct on-site survey of the location of the reinforced retaining wall to obtain the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure; S2. Based on the physical and mechanical parameters of the surrounding soil and geological data obtained from the on-site survey, the Kriging method is used for screening, classification, interpolation and integration; S3. According to the arrangement form and distribution law of the reinforcement in the retaining wall structure system, the theoretical distribution range of the reinforcement elastic modulus is determined, and the elastic modulus estimation of the reinforcement is generated by random sampling in uniform distribution, and the characteristic distribution is completed according to its arrangement form and distribution law in the reinforced retaining wall; S4. Based on the on-site geological characteristics of the reinforced retaining wall after interpolation, the physical parameters of the reinforcement material generated by random sampling are used as variation characteristics to construct a numerical simulation model of the retaining wall and perform numerical simulation analysis to obtain the corresponding analysis results of the target displacement; S5. Establish a neural network model, analyze the results of several monitoring points in the actual working point corresponding to the numerical simulation model of the retaining wall, and learn the mapping relationship between the elastic modulus of the layered reinforcement and the displacement of the surface monitoring point, until the numerical simulation model of the retaining wall passes the test; S6. Importing the surface displacement data of the reinforced retaining wall measured on site into the tested neural network model, and obtaining the elastic modulus of the layered reinforcement material through model inversion; S7. Feedback the reinforcement elastic modulus result to the actual measurement point of the project, and analyze the relative relationship between the reinforcement elastic modulus and the soil elastic modulus in combination with the actual distribution of the reinforcement, so as to predict whether the geogrid will fail due to pull-out.

2. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 1 is characterized in that: The S1 comprises the following steps: S11. Determine the on-site monitoring points of the reinforced retaining wall, deploy equipment including earth pressure cells and flexible displacement meters, and mark the monitoring points; S12. Carry out geological survey on the slope soil that affects the reinforced retaining wall, select multiple survey points, and obtain the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure, including the elastic modulus of the reinforcement, soil density, soil elastic modulus, soil Poisson's ratio, interface cohesion between soil and reinforcement, and interface friction angle between soil and reinforcement.

3. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 2 is characterized in that: The S2 comprises the following steps: S21. Integrate the physical and mechanical parameter values ​​of the surrounding soil, including soil density and elastic properties, and mark the coordinates of the corresponding survey points; S22. According to the first law of geography, select points within the effective influence range of Kriging interpolation as points to be interpolated, and mark the coordinates of the points to be interpolated; S23, determine the mathematical form of Kriging interpolation, set 0 as the point to be interpolated, is the surrounding survey point, then the estimated value of a regionalized material parameter at 0 is: in, Indicates material parameters in The estimated value at Indicates material parameters in The observed value at is the weight coefficient, represents the regionalized material parameters for the calculation, is the Lagrange constant, For materials located at The higher-order trend function at is the correlation coefficient of the higher-order trend function, is the number of higher-order trend functions; Minimize the difference between the expected and variance of the estimated value and the true value of the material parameter at the point to be interpolated, that is: in, is the true value of the material parameter at the point to be inserted, is the estimated value of the material parameter at the point to be inserted; S24. Calculation The estimated value of the distance between the material parameters and the corresponding strain difference is constructed based on a certain type of material parameter of a single point to be interpolated. The Kriging coefficient matrix is ​​in the form of: The Kriging interpolation expansion matrix of the material to be interpolated is constructed in the form of: in, To be inserted At known points The variation value between the material parameters at is the covariance matrix of the variable itself, is the covariance matrix between different variables; S25. After calculating the material parameter distance and the corresponding strain difference estimation results, the estimated value of the variogram is calculated by the following formula: in, The spacing is The total number of all point pairs, For Point deviation The measured value of the variable; S26, will The distances and corresponding strain differences between the data points are divided into several groups, each containing distance values, and calculate the average distance of each group of distance values and the mean variation , select the Gaussian model as the variogram theoretical model, fit the variogram and regress each parameter of the Gaussian model, so as to obtain the fitted variogram model; S27, the distance value between the survey point and the point to be inserted is brought into the fitted variogram to calculate the variogram value of the point to be inserted , and put all the variance values ​​into the formula In the equation, find the weight coefficient of the point to be inserted , then according to the formula Obtain the true value of the material parameter of the point to be inserted; S28, repeat S25 to S27 until the material parameter values ​​of all selected points to be inserted are determined in sequence; S29, repeat S25 to S28 until the material parameters of all selected categories are determined in sequence.

4. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 3 is characterized in that: The S3 comprises the following steps: S31. Determine the arrangement and distribution pattern of reinforcement in the retaining wall structure system, including quantity, horizontal and vertical spacing and anchorage length; S32. Make the overall sample of the given reinforcement elastic modulus obey uniform distribution , the sample parameters of the elastic modulus of the reinforcement after integration and interpolation are ; S33. The uniform distribution form of the elastic modulus of the reinforcement is solved by using the Gauss-Cauchy mixed density estimation method. The kernel density estimation is expressed as: in, Indicate point The estimated elastic modulus at is the number of samples of tendon elastic modulus, For the The elastic modulus of the tendon sample is is the bandwidth of the Gaussian kernel, is the bandwidth of the Cauchy kernel, is the weight coefficient of Gaussian kernel density estimation, is the weight coefficient of the Cauchy kernel density estimate; bandwidth The optimal value is calculated according to the Silverman rule, namely: in, is the standard deviation of the tendon elastic modulus samples; S34. Analyze the results of kernel density estimation, determine the support interval of the data, and take the upper limit of the interval as the uniform distribution parameter of the tendon The value of S35. Determine the uniform distribution of the elastic modulus of the reinforcement based on the solved parameters, and take random sampling of the uniform distribution by the sine-linear composite congruential method, as shown in the following formula: in, The random number sequence to be determined in the next step values, Indicates the currently determined A random number, is the multiplier, is the increment, is the modulus, is the weight coefficient of the sine function on the random number sequence, is the frequency parameter of the sine function; Determine the upper and lower boundaries of the distribution, and map the normalized random numbers sampled to the distribution interval to determine the elastic modulus of the reinforcement; S36. Arrange the elastic modulus of the randomly selected reinforcement material according to its arrangement form and distribution law in the reinforced retaining wall to complete the characteristic distribution.

5. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 4 is characterized in that: The S4 comprises the following steps: S41. Based on the physical and mechanical parameters of the surrounding soil and the mechanical parameter characteristics of the retaining wall structure obtained from the on-site survey in S1, a three-dimensional model of the retaining wall is established in sequence, the analysis steps for calculating the retaining wall simulation model are determined, the contact conditions of the surrounding soil, the retaining wall and the reinforcement are defined, loads and constraints are created and meshing is performed, the preliminary establishment of the numerical simulation model of the retaining wall is completed, the results are submitted and an inp file is generated; S42, constructing an Abaqus random field model according to various soil material parameters after interpolation and the elastic modulus parameters of the reinforcement obtained by random extraction, assigning the soil material parameters to the grid cells, and assigning the elastic modulus of the reinforcement to the reinforcement component material; S43, write a programming script, generate inp files in batches according to the results of the random field and import them into the Abaqus software, set a batch calculation task to run all the generated inp files, and obtain the target displacement analysis results; S44. Arrange the reinforcement elastic modulus parameter values ​​and the corresponding target displacement results in all inp files, and determine whether the reinforced retaining wall has failed based on the displacement.

6. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 5 is characterized in that: The S5 comprises the following steps: S51, using the KNN weighted proximity algorithm to remove noise and clean the abnormal values ​​in the elastic modulus parameter value and the corresponding retaining wall target displacement value caused by calculation errors in S4; S52. Calculate local weighted anomaly factor : in, is the normalized elastic modulus corresponding to the apparent displacement data point The average nearest neighbor distance, is the normalized elastic modulus corresponding to the apparent displacement data point The average distance between data points, is the comprehensive weighting factor; in, is the weighting factor bias coefficient, is the distance-based weight factor, is the weighting factor for the correlation between the elastic modulus and the apparent displacement; S53, integrate the local weighted anomaly factor and standardized data to define the outlier discrimination index : in, is the weight parameter, is the weighted average of the normalized apparent displacement distance, is the median absolute deviation of the normalized apparent displacement distance; Assuming the anomaly threshold is three times the median absolute deviation, if there is , then it is believed that If a point is an outlier, find the non-outlier value closest to the point to replace it; S54, forming a data set based on the apparent displacement data of the elastic modulus after cleaning, dividing the data set into a training set and a test set, establishing a deep support vector machine regression model, and using a radial basis function as a kernel function; Selecting the best support vector regression parameters using Bayesian optimization: Regularization parameter , and error tolerance , determine the value ranges of the three types of parameters in turn; Use the negative mean square error of the apparent displacement observations as the optimization objective: in, is the true value of the apparent displacement in the test set, is the predicted value of the apparent displacement in the test set, is the number of test set samples; S55. Select Gaussian process EI as the proxy model and use the acquisition function to select the next optimization point: Repeat the optimization process until the optimal parameters with the lowest mean square error are found; S56. Construct a deep support vector regression model, using a multi-layer support vector regression structure to capture the data pattern of elastic modulus corresponding to apparent displacement. The model output is expressed as: in, is the number of layers in the deep support vector regression model, is the radial basis function kernel function, is the weight of each layer, is the bias term; S57. The deep support vector regression model is trained using the training set. The optimal parameters determined by Bayesian optimization are used to train the deep support vector regression model. A regularization term is introduced to prevent overfitting. The objective function is shown in the following formula: in, is the weight vector, and is the slack variable, is a hyperparameter that controls the strength of regularization. is the regularization term; The objective function optimization is constrained by the following conditions: in, is the kernel function mapping, As input, a randomly generated set of reinforcement elastic moduli is For the set of monitoring displacement observation results outputted, the upper and lower limit constraints of the prediction results are specified, and the deep support vector regression model is fitted to obtain the objective function; S58, apply the model trained in S57 to the test set for prediction. A test set of samples was created to evaluate the accuracy of the support vector regression model for determining the depth of reinforced soil displacement observations by using mean square error and mean absolute error. S59. Analyze the results of mean square error and mean absolute error to evaluate the prediction performance of the deep support vector regression model. The mean square error and mean absolute error modify the parameter value range and Bayesian optimization proxy model through the methods of S54-S55 to further correct the parameters of the deep support vector regression model until the deep support vector regression model has good prediction accuracy.

7. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 6 is characterized in that: The S6 comprises the following steps: S61, obtaining displacement data of the soil measured on site by means of a flexible displacement meter, and checking the integrity and continuity of the data; S62, for abnormal displacement data, the KNN weighted proximity algorithm of S51-S53 is used to clean the data, the limit of the abnormal value is calculated according to the degree of variability, the abnormal value is identified, and the abnormal value is replaced; S63. Construct a feature matrix based on the cleaned displacement data, use a deep support vector regression model to predict the input displacement data, and output the corresponding reinforcement elastic modulus prediction value.

8. The method for determining reinforcement failure of reinforced retaining wall based on apparent displacement monitoring according to claim 7 is characterized in that: The S7 comprises the following steps: If the elastic modulus of the reinforcement and the elastic modulus of the soil satisfy any of the following formulas, the prediction results indicate that the geocell will fail by pull-out: in, is the effective reinforcement system number, is the elastic modulus of the tendon, is the elastic modulus of the soil, is the number of reinforcements in the horizontal direction of the retaining wall, is the number of reinforcements in the longitudinal direction of the retaining wall, It is a reinforcement object at any horizontal row of the retaining wall. It is a reinforcement object at any column position in the longitudinal direction of the retaining wall; Effective reinforcement system number Satisfies the following formula: in, is the actual length of the tendon, is the effective length of the tendon.

Images