A tunnel unfavorable geological body three-dimensional modeling method based on advanced prediction data

Abstract

The application discloses a tunnel bad geological body three-dimensional modeling method based on advanced prediction data, which comprises the following steps: firstly, inputting multi-source advanced prediction data, and establishing prediction rules of different bad geological bodies according to the data; then, establishing initial bad geological points and giving the geological points attribute values according to a comprehensive prediction system; according to the SVM bad geological classifier trained according to the excavated data, the initial bad geological points are processed to obtain geological points of different categories; according to the geometric range of the points, an implicit surface model is established, local adaptive updating of the model is realized, and an updated three-dimensional bad geological body model is formed. The three-dimensional model of the bad geological body is dynamically constructed by combining the modeling method of global radial basis function and implicit Boolean calculation and through local adaptive updating, which not only brings intuitive judgment to advanced support, but also facilitates subsequent related geological analysis, and can be loaded into a front-end platform for unified management, so that the whole process of tunnel construction is more complete.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic spatial information systems, and in particular to a tunnel adverse geological body three-dimensional modeling method based on advanced prediction data. BACKGROUND

[0002] Before tunnel excavation, advanced prediction is usually performed to detect possible adverse geological bodies in front. If the judgment is not accurate and appropriate excavation means and support measures are not selected, these hidden dangers may cause serious tunnel geological disasters, including collapse, water and mud inrush, gas, rock burst, etc. Therefore, accurately obtaining the position and geometry of the adverse geological body in front of the tunnel face is a key step for safe construction. In recent years, various advanced prediction technologies have been developed to predict the position and geometry of adverse geology in front of the tunnel face, as well as the surrounding rock grade and its stability, etc. Mainly include geophysical exploration methods, including seismic wave reflection method, electromagnetic wave method, high direct current method, etc. These prediction methods currently obtain detection results through professional instrument equipment and supporting interpretation software, and present the distribution of single adverse geological body (such as faults, fracture zones, water-rich zones, weak rock masses, etc.). It is difficult to fully utilize multiple data sources collected by different devices, resulting in a large error between the prediction results and the actual excavation situation. At present, there have been studies on comprehensive analysis of adverse geology, but the accuracy of this method is not high, and a large number of samples are required. For a newly excavated tunnel, there is little advanced prediction data, and the tunnel excavation verification length is short, so the actual sample quantity is small.

[0003] In addition, the current advanced prediction results are usually presented in the form of two-dimensional charts, which have poor interpretability. Only professional personnel can interpret them to form advanced geological prediction comprehensive analysis achievement reports. The report content is often complex and numerous, which is not convenient for construction personnel to check and is not convenient for dynamic updating. With people's deeper exploration of underground space, two-dimensional visualization can no longer better explain the distribution and structure of underground space. Three-dimensional visualization technology combined with deep learning, GPU parallel computing, computer graphics, etc. is increasingly used in many fields of underground geological engineering, and three-dimensional geological modeling has thus developed rapidly.

[0004] In the existing geological modeling methods, for different data or combinations of different data, due to different types of data structures and available information, many corresponding modeling methods have been generated. However, no method can be applied to different data types, so there is an urgent need for a geological modeling method that can be applied to multi-source advanced prediction data. SUMMARY

[0005] In order to solve the problems in the prior art, the application provides a tunnel adverse geological body three-dimensional modeling method based on advanced prediction data, which can quickly build a model by combining global radial basis functions and implicit Boolean calculation, and the model effect is smooth and can reflect the actual adverse geological boundary, and the self-adaptive local updating method is used to improve the solving efficiency and dynamically build a three-dimensional model of the adverse geological body, so that the problems in the above background art are solved.

[0006] In order to achieve the above object, the application provides the following technical scheme: a tunnel adverse geological body three-dimensional modeling method based on advanced prediction data, comprising the following steps:

[0007] S1, establishing prediction rules of different adverse geological bodies according to multi-source advanced prediction data;

[0008] S2, constructing a three-dimensional experimental range, dividing the three-dimensional experimental range into a plurality of continuous voxel units, and taking the midpoint of each voxel unit as an initial adverse geological point;

[0009] S3, giving the initial adverse geological point the attribute value of the multi-source advanced prediction data according to the prediction rule, to obtain initial adverse geological point data with attribute values;

[0010] S4, dividing the advanced prediction data of the excavated tunnel section into a training set and a test set, performing SVM training, and establishing an SVM classifier of different adverse geological categories;

[0011] S5, passing each initial adverse geological point with attribute values in the experimental range through the SVM classifier to obtain the adverse geological category to which the initial point belongs;

[0012] S6, according to the distribution of the identified adverse geological points of the same category, surrounding the geological points with similar distribution with a convex hull, constructing a preliminary geological body three-dimensional geometric range, and randomly generating a large number of discrete points on each face of the convex hull;

[0013] S7, establishing a global radial basis implicit function according to the discrete points, and using a Marching Cube algorithm to visualize the implicit function to generate a corresponding implicit surface model;

[0014] S8, locally dynamically and adaptively updating to obtain an updated implicit surface model;

[0015] S9, putting the updated implicit surface model into step S7 to form an updated three-dimensional adverse geological body model.

[0016] Preferably, the multi-source advanced prediction data comprises transient electromagnetic data, seismic wave reflection data, geological radar data and advanced horizontal borehole data.

[0017] Preferably, the different adverse geological bodies specifically include: faults, fracture zones, water-rich areas and weak rock masses.

[0018] Preferably, the step S4 specifically includes the following: selecting sample data of the excavated tunnel section, dividing into a training set and a test set, training the data by using the svm.SVC() function of the SVM training package in the sklearn library of python, and selecting a penalty parameter C by using a cross-validation method to obtain an SVM training model with high accuracy on the training set and the test set.

[0019] Preferably, the step S6 specifically includes the following:

[0020] S61, a convex hull of a range of the same type of adverse geological point is constructed by using a pcl library, a point cloud.ply file of the adverse geological point is loaded by calling a loadPLYFile function, and the convex hull is visualized by using a pcl::ConvexHull function;

[0021] S62, a large number of dense point clouds are randomly generated on each face of the generated convex hull by using a pcl::PointXYZ function;

[0022] S63, a voxel filter is created by using a pcl::VoxelGrid function, the dense point cloud of the convex hull is input, voxel filtering is performed on the dense point cloud, and a sparse point cloud of the filtered convex hull is output;

[0023] S64, a new normal point of the sparse point cloud of the convex hull is generated by using a pcl::Normal function, the normal direction is set to outward, and a normal point cloud corresponding to the normal direction is generated by using a pcl::PointXYZ function.

[0024] Preferably, the step S7 specifically includes the following:

[0025] S71, according to the sparse point cloud and the normal point cloud, the number of sparse point clouds on the convex hull is set to N, a constraint point is selected as a center, and the distance of the remaining points to the point is set to wherein the point is represented by r i , and the corresponding three-dimensional coordinates are (x i , y i , z i ), the remaining points are represented by r, and the corresponding three-dimensional coordinates are (x, y, z). In order to ensure that the interpolation matrix is invertible, a linearly independent polynomial function p(x) = k1x + k2y + k3z + k4 is added, and k1-k4 are corresponding coefficients;

[0026] At this time, the global radial basis function expression is wherein, the function For the radial basis function, the known condition of the function F(r) is the three-dimensional coordinates of N discrete points, and the to-be-solved parameters are λ1,..., λ N And k1, k2, k3, k4 are N+4, in order to ensure the smoothness of the implicit surface, the radial basis function must belong to the Beppo-Levi space, and the constraint condition needs to be met Finally, the N+4-dimensional linear equation group to be solved is as follows:

[0027]

[0028] The above linear equation group is simply written as A·x=y;

[0029] S72, first, in python, the sub-matrix of the four parts of the matrix A is obtained, and the four sub-matrices are spliced to obtain the complete matrix A by using the np.hstack function and the np.vstack function, and the column vector y is obtained in the same way;

[0030] S73, the Intel core mathematical library is used to solve the linear equation group, and the solving function is LAPACKE_dsysv, the matrix A and the column vector y of step S72 are input, and the coefficient vector x is obtained.

[0031] S74, the Marching Cube algorithm is used to visualize the obtained implicit function, first, a space region is divided into many equal-sized voxels, the intersection points of the isosurface and the voxels are calculated, and the intersection points are connected into a triangular mesh to form a visual three-dimensional adverse geological implicit surface model.

[0032] Preferably, in step S8, the following steps are specifically included:

[0033] S81, if the updated data is newly added advanced prediction data, first, the SVM is used to automatically identify the adverse geological category of the added data;

[0034] S82, if the expert knowledge is added, the category of the adverse geological point can be directly determined without SVM identification;

[0035] S83, if the two types of data are updated in a large range, the new data and the previous data are considered together to reconstruct the implicit surface model of the discrete adverse geological points;

[0036] S84, if the update range is small, the implicit function is constructed for the updated part of the data, and then the implicit function of the updated part and the original implicit function are calculated to realize local updating.

[0037] The beneficial effects of the present application are:

[0038] 1) The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0039] 2) The present application combines global radial basis function and Boolean implicit calculation algorithm to dynamically construct a three-dimensional geometric model of adverse geological bodies. This method can quickly build a model, and the model effect is smooth, which can reflect the actual adverse geological boundary. Moreover, since the advanced prediction data is constantly updated during construction, some geological experts' identification may also correct the existing model. If full re-modeling is performed, a large amount of time will be wasted. Therefore, a self-adaptive local updating method is proposed. In the case of only a small part of data changing, the radial basis function modeling is directly performed on this part, then the obtained function and the previous model function are re-calculated to obtain a new function expression. In this way, the time spent on re-solution of large radial basis function is saved. If the change is large, the re-computation efficiency will be higher. The dynamic construction of the three-dimensional model of adverse geological bodies not only brings intuitive judgment to the advanced support, but also facilitates subsequent related geological analysis, and can be loaded into the front-end platform for unified management, making the whole process of tunnel construction more complete. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0041] Figure 2 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0042] Figure 3 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0043] Figure 4 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0044] Figure 5 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0045] Figure 6 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped.

[0046] Figure 7 The present application is directed to the characteristics of multiple source advanced prediction data, different lengths of prediction distance, different accuracy, and different advanced prediction data suitable for interpretation of different adverse geological bodies. The SVM automatic identification method for comprehensive prediction of adverse geological bodies is redeveloped. DETAILED DESCRIPTION

[0047] With reference to the accompanying drawings: clear and complete description of the technical solutions in the embodiments of the present application will be described below, apparently, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor belong to the scope of protection of the present application.

[0048] At present, three-dimensional geological modeling methods have display modeling method, that is, a geological interface is established according to drilling data manually, which can cause serious topological inconsistency at the junction, because it is widely used due to its convenient and simple characteristics, but its shortcomings are also obvious, for a large range of regional geology, it is time-consuming and laborious and easy to make mistakes.

[0049] Implicit modeling obtains a mathematical function by interpolating data, and then reconstructs the model surface according to the function, and its advantage is that it can quickly model and update. The reconstruction algorithm of implicit surface includes moving least square method, Shepard method and radial basis function method. Radial basis function can stably and accurately construct three-dimensional modeling of discrete point cloud data, so it has a wide application in civil engineering, medical field, film and game field. The radial basis function method is generally divided into two categories, namely global support radial basis function method and local support radial basis function method. The former takes each discrete point as the center, establishes the distance function from the rest of the points to the point, and adds all the functions to obtain the fitting function of all discrete points, which is very suitable for the case of data missing, and has good extrapolation extension ability, but because all points are used at the same time, it is very difficult to solve the function, and the corresponding is the local support radial basis function method, which establishes local radial basis function by selecting appropriate support radius, so the solution is faster and more suitable for uniformly distributed point cloud data.

[0050] The data source in the present application is advanced prediction data, which is different from any kind of geological modeling data source in the past, because here not only the actual stratum structure model and the geological structure model are established, but also a three-dimensional model of a certain range is established according to the physical parameters of rock (such as Poisson's ratio, Young's modulus, longitudinal and transverse wave velocity), and the rock failure degree and water content difference in the range are divided into different types of adverse geological bodies. Multi-source advanced prediction data includes transient electromagnetic data, seismic wave reflection data, geological radar data and advanced horizontal drilling data.

[0051] The application firstly analyzes the characteristics of four kinds of advanced prediction data, and summarizes the relationship between each kind of advanced prediction data and adverse geological body, and establishes the comprehensive prediction rules of four kinds of adverse geological bodies (fault, fracture zone, water-rich area, and weak rock mass). Then, the tunnel section to be predicted is divided into continuous voxels, and a point of the voxel unit is referred to as a geological point to represent the properties of the voxel. The input data is in the form of adverse geological points as sample data, and four kinds of SVM adverse geological body classifiers are established. The classification results may be discontinuous or missing some parts. In view of the continuity of the geological body in geography, the convex hull structure of these points is constructed, and the geological points are randomly generated on the convex hull as new classification results, that is, adverse geological points of a certain class. Finally, the global radial basis function is used to establish an implicit model of the adverse geological body, and the model is dynamically updated by combining implicit Boolean calculation.

[0052] Referring to Figures 1-7 The application provides a technical solution: a tunnel adverse geological body three-dimensional modeling method based on advanced prediction data, including the following steps:

[0053] S1, establishing prediction rules of different adverse geological bodies according to multi-source advanced prediction data.

[0054] S2, constructing a three-dimensional experimental range, and dividing the three-dimensional experimental range into multiple continuous voxel units, and taking the midpoint of each voxel unit as an initial adverse geological point.

[0055] The cross-sectional range of the tunnel face is obtained as the width and height of the experimental range, and the prediction distance of 100 m of the first seismic wave reflection method is selected as the length of the experimental range.

[0056] The three-dimensional experimental range is divided into 1m×1m×1m continuous voxel units, and the midpoint of each voxel unit is taken as an initial adverse geological point. The initial adverse geological point has no attribute value.

[0057] S3, according to the prediction rule, the initial adverse geological point is given the attribute value of the multi-source advanced prediction data, and the initial adverse geological point data with the attribute value is obtained.

[0058] By establishing the comprehensive prediction rules of different advanced prediction data to corresponding adverse geological bodies, and according to the rules, the attribute value of the initial adverse geological point is given. The final result is that each geological point has multiple attribute values converted from the advanced prediction data, which lays the foundation for subsequent judgment of the category of the geological point (fault point, fracture zone point, water-rich area point, and weak rock mass point).

[0059] S4, the advanced prediction data of the excavated tunnel section is divided into a training set and a test set, and SVM training is performed to establish SVM classifiers of different adverse geological categories.

[0060] S5, each initial adverse geological point with attribute value in the experimental range is classified by the SVM classifier to obtain the adverse geological category to which the initial point belongs.

[0061] S6, the convex hull is used to surround the geological points with similar distribution according to the distribution of the identified adverse geological points in the same category to construct a preliminary three-dimensional geometric range of the geological body, and a large number of discrete points are randomly generated on each face of the convex hull for subsequent modeling.

[0062] S7, a global radial basis hidden function is established according to the discrete points, and the Marching Cube algorithm is used to visualize the hidden function to generate a corresponding implicit surface model.

[0063] S8, local dynamic adaptive updating is performed to obtain an updated implicit surface model.

[0064] When new advanced prediction data is generated, the step S5 is returned to determine the category of the new geological point, if the expert knowledge is added, the category of the adverse geological point is directly determined without returning to the step S5, if the two types of data are updated in a large range, the new data and the previous data are considered together to reconstruct the implicit surface of the discrete adverse geological point, if the update range is small, the implicit function is constructed for the changed data, and then the Boolean calculation is performed using the two implicit functions to realize the local adaptive updating.

[0065] S9, the updated implicit surface model is substituted into the step S7 to form an updated three-dimensional adverse geological body model.

[0066] The present application is directed to the characteristics of multiple source advanced prediction data, the prediction distance is long and short, the accuracy is different, and different advanced prediction data is suitable for the interpretation of different adverse geological bodies, so for a specific adverse geological body, the interpretation accuracy is low, and the multi-source advanced prediction data cannot be fully utilized, on the basis of the existing comprehensive identification method, four kinds of adverse geology such as fault, fracture zone, water-rich area and weak rock mass are taken as examples, the SVM automatic identification method for comprehensive prediction of adverse geological bodies is re-integrated and formulated, the advanced prediction range geographical space is first subdivided into a large number of voxel units, each voxel center is an initial geological point, then the interpretation value of each advanced prediction data is assigned to the geological point, then the obtained SVM classifier is applied to each geological point, and a two-class result is output, that is, 1 represents a certain kind of adverse geological body, and 0 represents that the adverse geological body is not of this kind, so the geometric range of the four kinds of adverse geological bodies is obtained. According to the geometric range of the adverse geological points, the present application combines the global radial basis function and the Boolean implicit calculation algorithm to dynamically construct a three-dimensional geometric model of the adverse geological body. The implicit function expression of the three-dimensional surface model is first calculated quickly, and then the Marching Cube algorithm is used to reconstruct the implicit surface of the adverse geological body. In the actual tunnel engineering environment, the advanced prediction data is continuously acquired or locally updated, and some expert knowledge identification may also be used to locally modify the existing model. Due to the different sizes of the update range, the speed of the overall model reconstruction is different. When the update range is small, the Boolean calculation of the implicit surface can update the model more quickly, because the small range of geological points is faster in solving linear equations or visualizing with the MC algorithm. When the update range is large, recalculating the linear equations of the updated part will take a lot of time. Therefore, at this time, the previous geological points and the updated geological points are merged to generate new convex hull points, and then the implicit function of the merged points is calculated, which is more efficient. Dynamically constructing the three-dimensional model of the adverse geological body not only provides a direct judgment for the advanced support, but also facilitates subsequent related geological analysis, and can be loaded into the front-end platform for unified management, making the whole process of tunnel construction more complete.

[0067] Embodiment 2

[0068] The overall step flow of the method of the present application is shown in Figure 1 The multi-source advanced prediction data is first input, and the prediction rules of different adverse geological bodies are established according to the data. Then the initial adverse geological points are established and the attribute values of the geological points are given according to the comprehensive prediction system. The SVM adverse geological classifier trained according to the excavated data is applied to these initial adverse geological points to obtain geological points of different categories, the implicit surface model is established according to the geometric range of the points, and the local adaptive update of the model is realized. The implementation process is realized by computer automation, including the following detailed steps:

[0069] Step 1: Open the transient electromagnetic data. Based on the characteristic that lower resistivity in transient electromagnetic data indicates higher water content, this is reflected on the transient electromagnetic contour map by different colors for low and high resistivity. Image recognition is used to determine the low-resistivity range of the input electromagnetic data and output a series of coordinate values ​​for this range. These values ​​are then used to assign attribute values ​​to geological points. Similarly, physical property information is extracted from seismic wave reflection data, ground-penetrating radar data, and borehole data in a similar manner. For example... Figure 2 As shown, these extracted attribute values ​​are assigned to the unfavorable geological points within the experimental range. The unfavorable geological points containing these attributes are also called initial unfavorable geological points. The specific definitions are as follows.

[0070] Step 2, as follows Figure 3 As shown, the area for constructing the adverse geological model is divided according to the detection range of a single seismic wave reflection method, namely 50m×50m×100m. In order to make full use of the data source, this range is divided into many voxels, each with a length, width and height of 1m. Then, the center of the voxel is taken as the representative of the voxel unit and called an adverse geological point. This invention studies four types of adverse geological bodies, and therefore corresponds to four types of adverse geological points.

[0071] Step 3: Select 30 sets of sample data from the excavated tunnel section and train the data using the `svm.SVC()` function from the `sklearn` library in Python. Figure 4 As shown, the penalty parameter C was selected using cross-validation for parameter tuning. A linear kernel function was chosen, and the optimal value for C was 4. At this point, the accuracy on the training set was 95.55%, and the accuracy on the test set was 87.5%. This trained model has high accuracy and can be used for prediction of other unknown segments. The initial adverse geological points with multi-source physical property parameters generated in step one were used as input data. After passing through an SVM classifier, a binary classification result was output, indicating whether the area was water-rich or not. The output binary classification of adverse geological points was in .csv format. This was converted into a .json file storing the coordinates of each point in Visual Studio 2107, and then further converted into a point cloud format .ply file.

[0072] Step 4: Based on the extent of the unfavorable geological points, calculate the radial basis functions surrounding these points, such as... Figure 5 As shown, the specific steps are as follows:

[0073] Step 4.1: Use the pcl library to construct the convex hull for the area of ​​the same type of unfavorable geological points, call the loadPLYFile function to load the point cloud .ply file of the unfavorable geological points, and then use a series of functions of the pcl::ConvexHull class to visualize the convex hull.

[0074] Step 4.2, a large number of dense point clouds are randomly generated on each face of the generated convex hull by using the pcl:PointXYZ function, for subsequent establishment of a radial basis function.

[0075] Step 4.3, in order to shorten the time for solving the linear equation set, unnecessary redundant points need to be reduced, and here the point cloud on the convex hull is filtered, the purpose is to reduce the size of the matrix when solving the radial basis function without changing the shape of the convex hull, and to increase the calculation efficiency. By creating a voxel filter with a length, width and height of 1 cm using the pcl:VoxelGrid function, the dense point cloud of the convex hull is input, and the filtered sparse point cloud of the convex hull is output.

[0076] Step 4.4, in the radial basis function for expressing the visual implicit surface, in order to prevent trivial solutions, new points are generated in the direction of the convex hull sparse point cloud, the normal point is generated by using the pcl:Normal function, the normal direction is set to outward, and the corresponding point cloud in the normal direction is generated by using the pcl:PointXYZ function, and the distance between the normal point cloud and the corresponding original point is set to 1 m.

[0077] Step 5, according to the convex hull sparse point cloud and the normal point cloud, the radial basis implicit function expression of the implicit surface corresponding to the convex hull is obtained. First, the number of sparse point clouds on the convex hull is set to N, a constraint point is selected as the center, and the distance from the remaining points to the point is set to In order to ensure that the interpolation matrix is invertible, a linearly independent polynomial function p(x)=k1x+k2y+k3z+k4 is added, and then the global radial basis function expression is Where the function is the radial basis function. The known condition of this function F(r) is the three-dimensional coordinates of the N discrete points, and the to-be-solved parameters are λ1,…,λ N N+4, and another condition is to ensure that the implicit surface is smooth, and the radial basis function must belong to the Beppo-Levi space, so the constraint condition The N+4-dimensional linear equation set to be solved finally is as follows:

[0078]

[0079] The above linear equation set is simply written as A·x=y.

[0080] Step 6, the flow of solving the linear equation set A·x=y and visualizing it of the present application is as shown in Figure 6 The specific steps are as follows:

[0081] Step 6.1, first in python, get the sub-matrix of the four parts of matrix A, use np.hstack function and np.vstack function to splice the four sub-matrices to get the complete matrix A, and get the column vector y in the same way.

[0082] Step 6.2, because the matrix is large, use Intel core mathematical library (MathKernel Library, MKL) to solve this large linear equation set in visual studio 2017, the solving function is LAPACKE_dsysv, input matrix A and column vector y of step 6.1, and solve to get the coefficient vector x.

[0083] Step 6.3, use Marching Cube algorithm, MC for short, to visualize the implicit function obtained. It first divides a space region into many equal-sized voxels, calculates the intersection of the isosurface and the voxel, and connects it into a triangular mesh to achieve the visualization effect.

[0084] Step 6.4, set the moving cube side length to 0.1m, make the cube move from the lower left corner of the test range, and generate an array of coordinates of the lower left corner of each cube. The number of cube voxels is huge, and cuda parallel computing is used to calculate the intersection of each cube with the point cloud range. First, get the visualization result of the rich water area and poor geology.

[0085] Step 6.5, loop steps 6.1 to 6.4 to get other poor geology 3D visualization results, including fracture zone, fault, and weak rock mass.

[0086] Step 7, the model local dynamic adaptive update process is shown in Figure 7 , which specifically includes the following:

[0087] Step 7.1, if the updated data is new advanced prediction data, first use SVM to automatically identify the poor geology type of the added data.

[0088] Step 7.2, when adding expert knowledge, the type of unknown points can be directly determined without SVM identification.

[0089] Step 7.3, if both types of data are updated in a large range, consider the new data together with the previous data to reconstruct the implicit surface model of the discrete poor geology points.

[0090] Step 7.4, if the update range is small, construct an implicit function for the updated part of the data, then do Boolean calculation between the implicit function of the updated part and the original implicit function to realize local update.

[0091] The present application analyzes and processes the characteristics of four kinds of advanced geological prediction data, including seismic wave reflection data, geological radar data, transient electromagnetic data and advanced horizontal drilling data; establishes the corresponding relationship between these data and adverse geological bodies (faults, fracture zones, water-rich areas, weak rock bodies), that is, the comprehensive prediction rule of adverse geology; the prediction range of the first seismic wave reflection is taken as the research range of the experiment, a voxel unit with a continuous distribution of 1m accuracy is established in the range, and the midpoint is selected as the initial adverse geological point; the measured advanced prediction data interpretation information is given to the initial adverse geological points in the experimental range according to the comprehensive prediction rule, as the attribute information of the geological points; the initial adverse geological points of the excavated tunnel section and the corresponding revealed adverse geological category result data are selected as the training set and the verification set, and the SVM classifier of each type of adverse geology is trained; the classifier is applied to the adverse geological points in the experimental range, and the category of each adverse geological point is automatically identified; the geological points of a certain adverse geological category identified are constructed into a convex hull; a dense point cloud is randomly generated on the convex hull, and a radial basis function is solved therefrom; a three-dimensional implicit surface model is established according to the solved radial basis function; when the data is updated, a large-scale update is directly recalculated for the radial basis function, and a small-scale update is established for the implicit function of the new updated data, and then a Boolean calculation is performed with the initial implicit function to obtain the updated implicit function, thereby improving the calculation efficiency and establishing an updated three-dimensional model.

[0092] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to part of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A tunnel adverse geological body three-dimensional modeling method based on advanced forecast data, characterized in that, The method comprises the following steps: S1, establishing a prediction rule of different adverse geological bodies according to multi-source advanced prediction data; S2, constructing a three-dimensional experimental range, and dividing the three-dimensional experimental range into a plurality of continuous voxel units, and taking the midpoint of each voxel unit as an initial adverse geological point; S3, assigning the initial adverse geological point with an attribute value of the multi-source advanced prediction data according to the prediction rule, to obtain initial adverse geological point data with an attribute value; S4, dividing the advanced prediction data of the excavated tunnel section into a training set and a test set, performing SVM training, and establishing an SVM classifier of different adverse geological categories; S5, passing each initial adverse geological point with an attribute value in the experimental range through the SVM classifier to obtain the adverse geological category to which the initial point belongs; S6, according to the distribution of the identified adverse geological points of the same category, surrounding the geological points with similar distribution with a convex hull, constructing a preliminary three-dimensional geometric range of the geological body, and randomly generating a large number of discrete points on each face of the convex hull; S7, establishing a global radial basis hidden function according to the discrete points, and using a Marching Cube algorithm to visualize the hidden function to generate a corresponding implicit surface model; specifically comprising the following: S71、According to the sparse point cloud and the normal point cloud, the number of the sparse point cloud on the convex hull is set as N, a constraint point is selected as the center, and the distance from the remaining points to the point is set as In order to ensure that the interpolation matrix is reversible, a linearly independent polynomial function is added ; The global radial basis function expression is ; wherein the function is a radial basis function, and the function is a known condition of three-dimensional coordinates of N discrete points, and the parameters to be solved are and , , , There are N+4 in total, and in order to ensure that the implicit surface is smooth, the radial basis function must belong to the Beppo-Levi space, and the constraint condition needs to be met The N+4-dimensional linear equation group to be solved finally is as follows: ； The above linear equations can be written simply as A·x = y ; S72. First, in Python, obtain the matrix. A The four sub-matrices are concatenated using the np.hstack and np.vstack functions to obtain the complete matrix. A Similarly, the column vectors are obtained. y ; S73, solve the linear equations by using Intel Core Math Library, the solving function is LAPACKE_dsysv, input the matrix in step S72 A and column vector y , and get the coefficient vector x ; S74, using the Marching Cube algorithm to visualize the obtained hidden function, first dividing a space region into a plurality of equal-sized voxels, calculating the intersection points of the isosurface and the voxels, and connecting them into a triangular mesh to form a visual three-dimensional adverse geological implicit surface model; S8, locally dynamically and adaptively updating to obtain an updated implicit surface model; S9, substituting the updated implicit surface model into step S7 to form an updated three-dimensional adverse geological body model.

2. The method for three-dimensional modeling of adverse geological bodies in tunnels based on advanced forecast data according to claim 1, characterized in that: The multi-source advanced prediction data comprises transient electromagnetic data, seismic wave reflection data, geological radar data, and advanced horizontal borehole data.

3. The method for three-dimensional modeling of adverse geological bodies in tunnels based on advanced forecast data according to claim 1, characterized in that: The different adverse geological bodies specifically include faults, fracture zones, water-rich areas, and weak rock bodies.

4. The method for three-dimensional modeling of adverse geological bodies in tunnels based on advanced forecast data according to claim 1, characterized in that: The step S4 specifically comprises the following: selecting sample data of the excavated tunnel section, dividing the sample data into a training set and a test set, using the svm.SVC() function of the SVM training package in the sklearn library of python to train the data, using a cross-validation method to select a penalty parameter C, and obtaining an SVM training model with high accuracy on the training set and the test set.

5. The method for three-dimensional modeling of adverse geological bodies in tunnels based on advanced forecast data according to claim 1, characterized in that: The step S6 specifically comprises the following: S61, using the pcl library to construct a convex hull of the range of the adverse geological points of the same type, calling the loadPLYFile function to load the point cloud.ply file of the adverse geological points, and then using the pcl::ConvexHull function to visualize the convex hull; S62, randomly generating a large number of dense point clouds on each face of the generated convex hull by using the pcl::PointXYZ function; S63, creating a voxel filter by using the pcl::VoxelGrid function, inputting the dense point cloud of the convex hull, voxel filtering the dense point cloud, and outputting the sparse point cloud filtered by the convex hull; S64, use the pcl: Normal function to generate new normal points for the convex hull sparse point cloud, set the normal direction outward, and then use the pcl: PointXYZ function to generate a normal point cloud corresponding to the normal direction.

6. The method for three-dimensional modeling of adverse geological bodies in tunnels based on advanced forecast data according to claim 1, characterized in that: The step S8 specifically comprises the following steps: S81, if the updated data is newly added advanced prediction data, first use SVM to automatically identify the bad geological category of the added data; S82, if the added data is expert knowledge, the category of the bad geological point can be directly determined without SVM identification; S83, if both types of data are updated in a large range, the new data and the previous data are considered together to reconstruct the implicit surface model of the discrete bad geological points; S84, if the update range is small, construct an implicit function for the updated part of the data, and then perform Boolean calculation on the implicit function of the updated part and the original implicit function to realize local update.

Images