A method for constructing a digital twin monitoring model for slopes based on an agent model

Abstract

The present invention relates to the technical field of slope early warning, and discloses a method for constructing a digital twin monitoring model of a slope based on an agent model, comprising the following steps: step S101, constructing a slope geometric model composed of M triangular facets by means of three-dimensional reconstruction software; step S102, constructing a first training data set for pre-training a pre-agent model by means of the slope geometric model; step S103, acquiring monitoring data of the M triangular facets in real time, and constructing a second training data set for training the agent model; step S104, generating a slope early warning cloud map; the present invention obtains simulation data under multi-physical field conditions by means of a slope geometric model, and constructs a training data set for pre-training a pre-agent model according to the simulation data, thereby greatly reducing the number of training samples of the agent model; in addition, multiple base learners in the agent model can reduce the calculation deviation of a single time series analysis model, thereby improving the prediction accuracy of the agent model.

Description

Technical Field

[0001] The present invention relates to the field of slope early warning technology, and more specifically, to a method for constructing a slope digital twin monitoring model based on an agent model. Background Art

[0002] Slopes are widely found in mountain roads, along railways, and in mining areas. They are affected by multiple factors such as geological conditions, hydrological environment, and climate change. They are prone to geological disasters such as landslides and collapses, and their stability is directly related to the safety of people's lives and property.

[0003] The existing slope warning system mainly relies on monitoring data collected by various sensors distributed on the slope, such as displacement meters, inclinometers, water level meters, rain gauges, etc., which can collect key parameters such as slope displacement, inclination angle, groundwater level, rainfall, etc. in real time, and generate warning information when it exceeds the critical threshold; however, static threshold settings may not be able to adapt to dynamic environmental changes, which may lead to false alarms or missed alarms. With the development of big data technology, slope warning is currently achieved through time series analysis models, such as LSTM (Long Short-Term Memory Neural Network Model) or GRU (Gated Neural Network Model). Time series analysis models extract the laws of historical sensor data to predict the future trends of slope physical properties, such as displacement and soil pressure; however, a single time series analysis model may lack long-term prediction capabilities or rely too much on raw data, resulting in low prediction accuracy.

[0004] Therefore, a slope early warning model is urgently needed to solve the above problems. Summary of the invention

[0005] The present invention provides a method for constructing a digital twin monitoring model of a slope based on an agent model, which solves the technical problems in the above-mentioned background technology.

[0006] The present invention provides a method for constructing a slope digital twin monitoring model based on an agent model, comprising the following steps:

[0007] Step S101, using the drone oblique photography technology to collect the elevation information of the slope, and constructing a slope geometric model consisting of M triangular facets through 3D reconstruction software;

[0008] Step S102, using the slope geometry model to perform a simulation experiment under a multi-physical field condition, and using the simulation data and simulation results to construct a first training data set for pre-training the pre-proxy model;

[0009] Multi-physics field conditions include: rainfall conditions, earthquake conditions and mining conditions;

[0010] The simulation data includes: the center coordinate points of M triangular facets at the current time and test data. The center coordinate points are represented by horizontal coordinate values ​​X, vertical coordinate values ​​Y and elevation coordinate values ​​Z. The test data include: temperature, rainfall, soil moisture, groundwater level, wind speed, wind direction slope deviation angle and vegetation density.

[0011] The simulation results are represented by the structural response values ​​of M triangular patches at the current time. The structural response values ​​include: pore water pressure, displacement and soil pressure;

[0012] The input of the pre-proxy model is the pre-processed simulation data, and the output is the simulation result;

[0013] The first training data set includes A training samples, each training sample includes a sample data and a sample label, the sample data corresponds to the input of the pre-proxy model, and the sample label corresponds to the output of the pre-proxy model;

[0014] Step S103, migrating the pre-trained pre-proxy model to the proxy model, and acquiring monitoring data of M triangle patches in real time to construct a second training data set for training the proxy model;

[0015] The monitoring data of M triangle patches are represented in the same way as the simulation data;

[0016] The input of the surrogate model is the same as the input of the pre-surrogate model, and the output is the warning value at the future time point G, which is generated according to the structural response value at the future time point G;

[0017] The second training data set includes B training samples, the sample data corresponds to the input of the proxy model, and the sample label corresponds to the output of the proxy model;

[0018] Step S104, inputting the monitoring data of the M triangular facets acquired in real time into the trained proxy model, and generating a slope warning cloud map according to the output of the proxy model.

[0019] Furthermore, the number M of triangular facets of the slope geometric model, the number A of training samples of the first training data set, the number B of training samples of the second training data set, and the future time point G are all custom parameters.

[0020] Furthermore, the center coordinate points of the M triangular facets and the test data are preprocessed, including the following steps:

[0021] Step S201, converting the wind direction slope deviation angle and vegetation density in the test data into numerical representation;

[0022] Wind direction slope deviation angle of the mth triangle patch The calculation formula is as follows:

[0023] ;

[0024] Where 1≤m≤M, Indicates the wind direction of the mth triangle patch, with due north as 0° and clockwise as the positive direction, ranging from 0° to 360°. Indicates the slope of the mth triangle patch and the angle between the triangle patch and the ground plane, ranging from 0° to 90°;

[0025] Vegetation density of the mth triangle patch The calculation formula is as follows:

[0026] ;

[0027] in , and They represent the reflectivity of the near-infrared band, the reflectivity of the red light band, and the reflectivity of the blue light band of the m-th triangular facet respectively;

[0028] Step S202, filling the missing values ​​by taking the average value of the center coordinate points of two adjacent triangle facets and the test data;

[0029] Step S203, normalizing the center coordinate points of the M triangle patches and the test data;

[0030] The calculation formula for normalization is as follows:

[0031] ;

[0032] Where 1≤m≤M, and Respectively represent the center coordinate point and test data of the mth triangle patch after and before preprocessing, and Respectively represent the center coordinate points of the M triangle patches and the minimum and maximum values ​​of the test data.

[0033] Further, the pre-proxy model includes M parallel modules, all parallel modules share weight parameters and bias parameters, and each parallel module includes: a first hidden layer, a concatenation layer, and a second hidden layer;

[0034] The first hidden layer includes: a first unit, a second unit, a third unit, a fourth unit and a fifth unit;

[0035] The first unit, the second unit, the third unit, the fourth unit and the fifth unit of the first hidden layer of the mth parallel module all input the center coordinate point and the test data of the preprocessed mth triangular face patch, and all output the structural response value of the mth triangular face patch;

[0036] The concatenation layer of the mth parallel module is used to concatenate the structural response values ​​of the mth triangular patch output by the first unit, the second unit, the third unit, the fourth unit and the fifth unit of the first hidden layer of the mth parallel module, and output a combination vector, the size of which is 1×15;

[0037] The second hidden layer of the mth parallel module inputs the combined vector output by the splicing layer of the mth parallel module, and outputs the structural response value of the mth triangular face;

[0038] The surrogate model includes a pre-surrogate model, and the output of the second hidden layer of the surrogate model is represented as the structural response value of G at the future time point;

[0039] The agent model also includes a linear layer, which is used to generate an early warning value based on the structural response value of G at a future time point.

[0040] Furthermore, the first unit is constructed based on a gradient boosting decision tree, the second unit is constructed based on a support vector machine of a radial basis function, the third unit is constructed based on a support vector machine of a polynomial kernel function, the fourth unit is constructed based on a random forest, and the fifth unit is constructed based on a K-nearest neighbor algorithm.

[0041] Furthermore, the second hidden layer is constructed based on a lightweight gradient boosting decision tree.

[0042] Furthermore, the proxy model was trained by 5-fold cross validation, and the second training data set was divided into 5 data sets of the same size. In each training round, only one data set was used as a validation set, and the remaining four data sets were used as training sets.

[0043] Furthermore, the loss function of the pre-proxy model is The calculation formula is as follows:

[0044] ;

[0045] in Represents the output value of the sample data of the ath training sample of the first training data set input to the pre-proxy model, Represents the sample label of the ath training sample;

[0046] Loss function of the surrogate model The calculation formula is as follows:

[0047] ;

[0048] Where 1≤b≤B, Represents the output value of the sample data of the bth training sample of the second training data set to the proxy model, represents the sample label of the bth training sample, Indicates a custom parameter.

[0049] Furthermore, the warning value of the mth triangle patch The calculation formula is as follows:

[0050] ;

[0051] Where 1≤m≤M, , and represent the pore water pressure, displacement and earth pressure of the mth triangular patch respectively, , and represent the first weight coefficient, the second weight coefficient and the third weight coefficient respectively, , and are custom parameters, and , and The sum of is 1, and e represents a natural constant.

[0052] Furthermore, the warning value is rendered to the corresponding triangle patch through the unity rendering engine to generate a slope warning cloud map.

[0053] The beneficial effects of the present invention are as follows: the present invention obtains simulation data under multi-physical field conditions through a slope geometry model, and constructs a training data set for a pre-trained pre-proxy model based on the simulation data, which greatly reduces the number of training samples of the proxy model. In addition, the multiple base learners in the proxy model can reduce the computational deviation of a single time series analysis model, thereby improving the prediction accuracy of the proxy model. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of a method for constructing a digital twin monitoring model of a slope based on an agent model of the present invention;

[0055] Figure 2 It is a flow chart of preprocessing the center coordinate points and test data of M triangular facets of the present invention;

[0056] Figure 3 It is a schematic diagram of the slope based on the oblique photography of the drone of the present invention;

[0057] Figure 4 is a schematic diagram of the slope geometric model of the present invention;

[0058] Figure 5 It is a schematic diagram of 5-fold cross validation of the present invention;

[0059] Figure 6 is a schematic diagram of a parallel module of the present invention;

[0060] Figure 7 It is a schematic diagram of the slope early warning cloud map of the present invention. DETAILED DESCRIPTION

[0061] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0062] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0063] like Figure 1 to Figure 7 As shown, a method for constructing a slope digital twin monitoring model based on an agent model includes the following steps:

[0064] Step S101, using the drone oblique photography technology to collect the elevation information of the slope, and constructing a slope geometric model consisting of M triangular facets through 3D reconstruction software;

[0065] Step S102, using the slope geometry model to perform a simulation experiment under a multi-physical field condition, and using the simulation data and simulation results to construct a first training data set for pre-training the pre-proxy model;

[0066] Multi-physics field conditions include: rainfall conditions, earthquake conditions and mining conditions;

[0067] The simulation data includes: the center coordinate points of M triangular facets at the current time and test data. The center coordinate points are represented by horizontal coordinate values ​​X, vertical coordinate values ​​Y and elevation coordinate values ​​Z. The test data include: temperature, rainfall, soil moisture, groundwater level, wind speed, wind direction slope deviation angle and vegetation density.

[0068] The simulation results are represented by the structural response values ​​of M triangular patches at the current time. The structural response values ​​include: pore water pressure, displacement and soil pressure;

[0069] The input of the pre-proxy model is the pre-processed simulation data, and the output is the simulation result;

[0070] The first training data set includes A training samples, each training sample includes a sample data and a sample label, the sample data corresponds to the input of the pre-proxy model, and the sample label corresponds to the output of the pre-proxy model;

[0071] Step S103, migrating the pre-trained pre-proxy model to the proxy model, and acquiring monitoring data of M triangle patches in real time to construct a second training data set for training the proxy model;

[0072] The monitoring data of M triangle patches are represented in the same way as the simulation data;

[0073] The input of the surrogate model is the same as the input of the pre-surrogate model, and the output is the warning value at the future time point G, which is generated according to the structural response value at the future time point G;

[0074] The second training data set includes B training samples, the sample data corresponds to the input of the proxy model, and the sample label corresponds to the output of the proxy model;

[0075] Step S104, inputting the monitoring data of the M triangular facets acquired in real time into the trained proxy model, and generating a slope warning cloud map according to the output of the proxy model.

[0076] It should be noted that the drone carries a high-definition camera to shoot the slope from multiple angles, including vertically downward (providing plane information of the surface, which is the basis for constructing an orthophoto map), tilting 30° to the left (for capturing terrain features on the left), tilting 30° to the right (for capturing terrain features on the right), tilting 30° forward and tilting 30° backward to ensure that the entire slope is covered as much as possible; the multi-angle slope images are imported into the 3D reconstruction software, which will generate a dense point cloud model, each point representing a 3D coordinate of the slope surface, and then connect adjacent points in the point cloud model to form M triangular patches, wherein the point cloud model can also be subjected to point cloud denoising and point cloud thinning to improve the quality of the slope geometry model. The 3D reconstruction software can be Agisoft Metashape or Pix4Dmapper, etc. Constructing the slope geometry model through 3D reconstruction software is a conventional technical means and will not be elaborated here.

[0077] In one embodiment of the present invention, the number M of triangular facets of the slope geometric model, the number A of training samples of the first training data set, the number B of training samples of the second training data set, and the future time point G are all custom parameters, wherein M is set by a geological expert according to the volume size of the slope, the default value of M is set to 2000, the default value of A is set to 2000, the default value of B is set to 500, and the default value of G is set to 1 hour.

[0078] In one embodiment of the present invention, Figure 2 As shown, the center coordinate points of M triangular facets and the test data are preprocessed, including the following steps:

[0079] Step S201, converting the wind direction slope deviation angle and vegetation density in the test data into numerical representation;

[0080] Wind direction slope deviation angle of the mth triangle patch The calculation formula is as follows:

[0081] ;

[0082] Where 1≤m≤M, Indicates the wind direction of the mth triangle patch, with due north as 0° and clockwise as the positive direction, ranging from 0° to 360°. Indicates the slope of the mth triangle patch and the angle between the triangle patch and the ground plane, ranging from 0° to 90°;

[0083] Vegetation density of the mth triangle patch The calculation formula is as follows:

[0084] ;

[0085] in , and They represent the reflectivity of the near-infrared band, the reflectivity of the red light band, and the reflectivity of the blue light band of the m-th triangular facet respectively;

[0086] Step S202, filling the missing values ​​by taking the average value of the center coordinate points of two adjacent triangle facets and the test data;

[0087] Step S203, normalizing the center coordinate points of the M triangle patches and the test data;

[0088] The calculation formula for normalization is as follows:

[0089] ;

[0090] Where 1≤m≤M, and Respectively represent the center coordinate point and test data of the mth triangle patch after and before preprocessing, and Respectively represent the center coordinate points of the M triangle patches and the minimum and maximum values ​​of the test data.

[0091] It should be noted that the reflectivity of the near-infrared band, the reflectivity of the red light band, and the reflectivity of the blue light band can be obtained by using a hyperspectral camera mounted on a drone; in addition, filling in missing values ​​can ensure the integrity of the input data, normalization processing can prevent deviations caused by large differences in the orders of magnitude between the input data, and normalization processing can speed up the training of the proxy model and improve the generalization ability and robustness of the proxy model.

[0092] In one embodiment of the present invention, the pre-proxy model includes M parallel modules, all parallel modules share weight parameters and bias parameters, and each parallel module includes: a first hidden layer, a concatenation layer, and a second hidden layer;

[0093] The first hidden layer includes: a first unit, a second unit, a third unit, a fourth unit and a fifth unit;

[0094] The first unit, the second unit, the third unit, the fourth unit and the fifth unit of the first hidden layer of the mth parallel module all input the center coordinate point and the test data of the preprocessed mth triangular face patch, and all output the structural response value of the mth triangular face patch;

[0095] The concatenation layer of the mth parallel module is used to concatenate the structural response values ​​of the mth triangular patch output by the first unit, the second unit, the third unit, the fourth unit and the fifth unit of the first hidden layer of the mth parallel module, and output a combination vector, the size of which is 1×15;

[0096] The second hidden layer of the mth parallel module inputs the combined vector output by the splicing layer of the mth parallel module, and outputs the structural response value of the mth triangular face;

[0097] The surrogate model includes a pre-surrogate model, and the output of the second hidden layer of the surrogate model is represented as the structural response value of G at the future time point;

[0098] The agent model also includes a linear layer, which is used to generate an early warning value based on the structural response value of G at a future time point.

[0099] It should be noted that a first training data set of a large number of training samples can be obtained through simulation experiments. Pre-training of the first training data set allows the pre-proxy model to learn the data characteristics of the slope under multi-physical field conditions. After migrating the pre-trained pre-proxy model to the proxy model, the training speed of the proxy model can be greatly accelerated, thereby reducing the number of training samples of the second training data set, because the cost of simulation data is lower than that of real-time monitoring data, and the training of the second training data set enables the proxy model to have the ability to predict changes in slope structure response values, so as to generate slope warning cloud maps. In addition, the number of parallel modules is the same as the number of triangular patches of the slope geometric model. The design of parallel modules can speed up the calculation speed of the proxy model.

[0100] In one embodiment of the present invention, the first unit is constructed based on XGBoost (gradient boosting decision tree), the second unit is constructed based on RBF (radial basis function support vector machine), the third unit is constructed based on SVM (polynomial kernel function support vector machine), the fourth unit is constructed based on RF (random forest), and the fifth unit is constructed based on KNN (K nearest neighbor algorithm).

[0101] In one embodiment of the present invention, the second hidden layer is constructed based on LightGBM (Lightweight Gradient Boosting Decision Tree).

[0102] It should be noted that XGBoost gradually reduces the error of the previous decision tree by adding new decision trees; LightGBM selects the current optimal leaf node for splitting each time, instead of splitting in hierarchical order like XGBoost. This method can reduce the loss function faster, so it usually converges faster; RF consists of multiple decision trees, each of which is trained independently, and the prediction value of random forest is the average of the prediction values ​​of all decision trees; RBF is similar to SVM, but the type of kernel function is different.

[0103] In one embodiment of the present invention, since the number of training samples of the second training data set is limited, the proxy model is trained through 5-fold cross-validation, and the second training data set is divided into 5 groups of data sets of the same size. Only one group of data sets is used as a validation set for each training time, and the remaining four groups of data sets are used as training sets.

[0104] In one embodiment of the present invention, the loss function of the pre-proxy model is The calculation formula is as follows:

[0105] ;

[0106] in Represents the output value of the sample data of the ath training sample of the first training data set input to the pre-proxy model, Represents the sample label of the ath training sample;

[0107] Loss function of the surrogate model The calculation formula is as follows:

[0108] ;

[0109] Where 1≤b≤B, Represents the output value of the sample data of the bth training sample of the second training data set to the proxy model, represents the sample label of the bth training sample, Indicates custom parameters, preferably, Set to 1.5.

[0110] It should be noted that the gradient information of the weight parameters and bias parameters in the pre-proxy model and the proxy model is first calculated according to the chain rule, and the weight parameters and bias parameters in the pre-proxy model and the proxy model are reversely updated through the gradient descent algorithm according to the gradient information to minimize the loss function.

[0111] In one embodiment of the present invention, the warning value of the mth triangle patch The calculation formula is as follows:

[0112] ;

[0113] Where 1≤m≤M, , and represent the pore water pressure, displacement and earth pressure of the mth triangular patch respectively, , and represent the first weight coefficient, the second weight coefficient and the third weight coefficient respectively, , and are custom parameters, and , and The sum of is 1, and e represents a natural constant.

[0114] It should be noted that the larger the warning value, the greater the danger of the slope. According to the calculation formula of the warning value, it can be known that the value range of the warning value is between 0 and 255. Therefore, a grayscale image can be generated according to the warning value of each triangular face, or the warning value can be divided into three ranges, corresponding to the red channel value, the green channel value and the blue channel value, respectively. In addition, the pore water pressure, displacement and soil pressure of the triangular face can also be scaled to between 0 and 255 to generate an RGB image.

[0115] In one embodiment of the present invention, the warning value is rendered to the corresponding triangular facet through the unity rendering engine to generate a slope warning cloud map to display the status of the slope in real time. In addition, warning measures can also be generated by setting thresholds. For example, if the warning value is greater than or equal to 80, the monitoring frequency is increased; if the warning value is greater than or equal to 160, staff are arranged to go to the site for investigation, etc., which will not be elaborated here.

[0116] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A method for constructing a slope digital twin monitoring model based on an agent model, characterized in that: The following steps are involved: Step S101, using the drone oblique photography technology to collect the elevation information of the slope, and constructing a slope geometric model composed of M triangular facets through 3D reconstruction software; Step S102, using the slope geometry model to perform a simulation experiment under multi-physical field conditions, and using the simulation data and simulation results to construct a first training data set for pre-training the pre-proxy model; Multi-physics field conditions include: rainfall conditions, earthquake conditions and mining conditions; The simulation data includes: the center coordinate points of M triangular facets at the current time and test data. The center coordinate points are represented by horizontal coordinate values ​​X, vertical coordinate values ​​Y and elevation coordinate values ​​Z. The test data include: temperature, rainfall, soil moisture, groundwater level, wind speed, wind direction slope deviation angle and vegetation density. The simulation results are represented by the structural response values ​​of M triangular patches at the current time. The structural response values ​​include: pore water pressure, displacement and soil pressure; The input of the pre-proxy model is the pre-processed simulation data, and the output is the simulation result; The first training data set includes A training samples, each training sample includes a sample data and a sample label, the sample data corresponds to the input of the pre-proxy model, and the sample label corresponds to the output of the pre-proxy model; Step S103, migrating the pre-trained pre-proxy model to the proxy model, and acquiring monitoring data of M triangle patches in real time to construct a second training data set for training the proxy model; The monitoring data of M triangle patches are represented in the same way as the simulation data; The input of the surrogate model is the same as the input of the pre-surrogate model, and the output is the warning value at the future time point G, which is generated according to the structural response value at the future time point G; The second training data set includes B training samples, the sample data corresponds to the input of the proxy model, and the sample label corresponds to the output of the proxy model; Step S104, inputting the monitoring data of the M triangular facets acquired in real time into the trained proxy model, and generating a slope warning cloud map according to the output of the proxy model.

2. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: The number M of triangular patches of the slope geometric model, the number A of training samples of the first training data set, the number B of training samples of the second training data set, and the future time point G are all custom parameters.

3. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: Preprocessing the center coordinate points of the M triangular patches and the test data includes the following steps: Step S201, converting the wind direction slope deviation angle and vegetation density in the test data into numerical representation; Wind direction slope deviation angle of the mth triangle patch The calculation formula is as follows: ； Where 1≤m≤M, Indicates the wind direction of the mth triangle patch, with due north as 0° and clockwise as the positive direction, ranging from 0° to 360°. Indicates the slope of the mth triangle patch and the angle between the triangle patch and the ground plane, ranging from 0° to 90°; Vegetation density of the mth triangle patch The calculation formula is as follows: ； in , and They represent the reflectivity of the near-infrared band, the reflectivity of the red light band, and the reflectivity of the blue light band of the m-th triangular facet respectively; Step S202, filling the missing values ​​by taking the average value of the center coordinate points of two adjacent triangle facets and the test data; Step S203, normalizing the center coordinate points of the M triangle patches and the test data; The calculation formula for normalization is as follows: ； Where 1≤m≤M, and Respectively represent the center coordinate point and test data of the mth triangle patch after and before preprocessing, and Respectively represent the center coordinate points of the M triangle patches and the minimum and maximum values ​​of the test data.

4. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: The pre-proxy model includes M parallel modules, all parallel modules share weight parameters and bias parameters, and each parallel module includes: a first hidden layer, a concatenation layer, and a second hidden layer; The first hidden layer includes: a first unit, a second unit, a third unit, a fourth unit and a fifth unit; The first unit, the second unit, the third unit, the fourth unit and the fifth unit of the first hidden layer of the mth parallel module all input the center coordinate point and the test data of the preprocessed mth triangular face patch, and all output the structural response value of the mth triangular face patch; The concatenation layer of the mth parallel module is used to concatenate the structural response values ​​of the mth triangular patch output by the first unit, the second unit, the third unit, the fourth unit and the fifth unit of the first hidden layer of the mth parallel module, and output a combination vector, the size of which is 1×15; The second hidden layer of the mth parallel module inputs the combined vector output by the splicing layer of the mth parallel module, and outputs the structural response value of the mth triangular face; The surrogate model includes a pre-surrogate model, and the output of the second hidden layer of the surrogate model is represented as the structural response value of G at the future time point; The agent model also includes a linear layer, which is used to generate an early warning value based on the structural response value of G at a future time point.

5. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 4, characterized in that: The first unit is built based on the gradient boosting decision tree, the second unit is built based on the support vector machine of the radial basis function, the third unit is built based on the support vector machine of the polynomial kernel function, the fourth unit is built based on the random forest, and the fifth unit is built based on the K nearest neighbor algorithm.

6. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 4, characterized in that: The second hidden layer is built based on a lightweight gradient boosted decision tree.

7. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: The proxy model was trained by 5-fold cross validation, and the second training data set was divided into 5 groups of data sets of the same size. In each training round, only one data set was used as the validation set, and the remaining 4 data sets were used as the training sets.

8. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: Loss function of the pre-surrogate model The calculation formula is as follows: ； in Represents the output value of the sample data of the ath training sample of the first training data set input to the pre-proxy model, Represents the sample label of the ath training sample; Loss function of the surrogate model The calculation formula is as follows: ； Where 1≤b≤B, Represents the output value of the sample data of the bth training sample of the second training data set to the proxy model, represents the sample label of the bth training sample, Indicates a custom parameter.

9. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: Warning value of the mth triangle patch The calculation formula is as follows: ； Where 1≤m≤M, , and represent the pore water pressure, displacement and earth pressure of the mth triangular patch respectively, , and represent the first weight coefficient, the second weight coefficient and the third weight coefficient respectively, , and are custom parameters, and , and The sum of is 1, and e represents a natural constant.

10. The method for constructing a slope digital twin monitoring model based on an agent model according to claim 1, characterized in that: The warning value is rendered to the corresponding triangle patch through the unity rendering engine to generate a slope warning cloud map.

Images