Three-dimensional unit body attribute modeling method and system based on multi-source heterogeneous information fusion

Through the three-dimensional unit body attribute modeling method of multi-source heterogeneous information fusion, combined with survey, geophysical exploration and surrounding rock image data, GS-RBF and convolutional neural network are used to solve the problem of comprehensive utilization of geological data in traditional tunnel construction, and realize the Efficient three-dimensional geological model construction and construction decision support.

CN118378319BActive Publication Date: 2024-11-22SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410260860.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-07
Publication Date
2024-11-22
Estimated Expiration
2044-03-07

AI Technical Summary

Technical Problem

It is difficult to comprehensively utilize a variety of geological data in traditional tunnel construction, resulting in low visualization and inversion accuracy of geological three-dimensional models, which reduces construction efficiency.

Method used

Using a three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion, three-dimensional is constructed by combining survey data, geophysical exploration data and surrounding rock images, using GS-RBF implicit surface generation algorithm and convolutional neural network model. The unit body attribute data structure realizes the fine representation of underground space attributes and three-dimensional visualization of structure.

Benefits of technology

The comprehensive utilization of geological data is realized, the three-dimensional visualization capabilities of underground space attribute distribution and structure are improved, the decision-making efficiency of construction workers is enhanced, and the real-time and accuracy of the model are improved through the adaptive update mechanism.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118378319B_ABST
    Figure CN118378319B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of three-dimensional modeling, and specifically discloses a three-dimensional unit body attribute modeling method and system based on multi-source heterogeneous information fusion, the method comprising: generating a mean stratigraphic body to obtain a poor geological body; embedding the poor geological body into the mean stratigraphic body to obtain a geological body structure model of the area to be tested; constructing a three-dimensional unit body attribute data structure; based on surrounding rock images, drilling perception and in-situ test data information, using a convolutional neural network model to identify cracks, lithology distribution and broken areas in the surrounding rock image; according to the discreteness of the data, simulating and obtaining the three-dimensional unit body attribute information value of the entire area to be tested; based on the three-dimensional unit body attribute data structure and the three-dimensional unit body attribute information value, obtaining the geological body attribute model of the area to be tested. The present invention can adaptively update implicit functions based on the size of the update range of geophysical data, taking into account the relative speed of solving implicit functions and Boolean calculations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling of tunnel engineering, and in particular to a three-dimensional unit body attribute modeling method and system based on multi-source heterogeneous information fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The geological 3D model is key information for optimizing design and risk control during tunnel construction, and helps tunnel construction workers adopt correct advanced support and excavation methods.

[0004] Traditional modeling methods make it difficult to comprehensively utilize the massive multi-source information in the entire process of tunnel construction, such as survey data, drilling information, geophysical data, and surrounding rock image information.

[0005] Because various geological data information has the characteristics of independence, dispersion, different formats, and continuous changes, various commercial processing software can usually only perform two-dimensional or three-dimensional visualization of a single physical parameter. For example, in the seismic wave reflection method, the TSP303Plus instrument is usually used for data acquisition, and then the supporting AmbergTSPPlus software is used for data processing to obtain intermediate results such as the time section, depth migration section, rock reflection layer, physical and mechanical parameters, and energy size of each reflection layer in front of the excavation working face, which are used to predict and invert the geological conditions such as karst, weak rock mass, fault, fracture zone, and water-rich area in front of the excavation working face. Various commercial software work independently of each other and cannot directly visualize the three-dimensional model of the site geology.

[0006] In addition, due to the incompleteness of a single data source, the inversion results are not accurate and the reported results are discrete, which often reduces the work efficiency of tunnel workers. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a three-dimensional unit body attribute modeling method and system based on multi-source heterogeneous information fusion, which makes full use of various data sources collected by different equipment, combines geological data inside and outside the tunnel with drilling, geophysical exploration, and surrounding rock image data, and realizes three-dimensional visualization of underground space attribute distribution and structure.

[0008] In some embodiments, the following technical solutions are adopted:

[0009] A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion, comprising:

[0010] Based on the geological mapping and drilling logging information in the survey data, the stratigraphic boundary and groundwater level are obtained, and the stratigraphic plane is further fitted to generate a mean stratigraphic volume;

[0011] Based on the acquired geophysical data, inversion is performed to obtain the type and scale information of the adverse geology, and then the adverse geological body is obtained based on the implicit surface generation algorithm of GS-RBF;

[0012] Embedding the bad geological body into the mean stratum body to obtain the geological body structure model of the area to be tested; performing three-dimensional unit body segmentation on the geological structure model to construct a three-dimensional unit body attribute data structure; the three-dimensional unit body attribute data structure includes spatial information and attribute information;

[0013] Based on the surrounding rock image, drilling perception and in-situ test data information, the convolutional neural network model is used to identify the cracks, lithology distribution and broken areas in the surrounding rock image; according to the discreteness of the data, the three-dimensional unit body attribute information value of the entire test area is simulated;

[0014] Based on the three-dimensional unit body attribute data structure and the three-dimensional unit body attribute information value, a geological body attribute model of the area to be measured is obtained.

[0015] If the geophysical data is updated and the updated data is less than the set threshold, an implicit function is constructed for the updated geophysical data separately; the constructed implicit function is Boolean-calculated with the previous implicit function to obtain a new implicit function, and then the updated poor geological body is obtained.

[0016] If the geophysical data is updated and the updated data is greater than the set threshold, the implicit function is reconstructed based on the updated geophysical data.

[0017] The process of identifying the crack density in the surrounding rock image is as follows:

[0018] The DexiNed edge detection algorithm is used to identify surrounding rock traces in surrounding rock images;

[0019] Gabor filtering is used to enhance the image of surrounding rock trace recognition results;

[0020] The DBSCAN clustering algorithm is used to divide the pixels into different density connected regions and perform cluster analysis on the identification traces;

[0021] The surrounding rock image is divided into evenly sized square cells according to the custom standard, and the number, length and inclination information of all the cracks involved in the square cells are counted;

[0022] For each grid cell, the length of each crack multiplied by the weight of its inclination angle is calculated, and then these values ​​are added to obtain the total weighted length of the crack; the unit crack density is obtained based on the ratio of the total weighted length of the crack to the area of ​​each grid cell.

[0023] In other embodiments, the following technical solutions are adopted:

[0024] A three-dimensional unit body attribute modeling system based on multi-source heterogeneous information fusion, comprising:

[0025] The mean stratigraphic volume generation module is used to obtain stratigraphic boundaries and groundwater levels based on geological mapping and borehole cataloging information in the survey data, and further fit the stratigraphic surface to generate a mean stratigraphic volume;

[0026] The module for acquiring unfavorable geological bodies is used to perform inversion based on the acquired geophysical data to obtain the type and scale information of unfavorable geology, and then obtain the unfavorable geological bodies based on the implicit surface generation algorithm of GS-RBF;

[0027] A three-dimensional unit body attribute data structure construction module is used to embed the bad geological body into the mean stratum body to obtain the geological body structure model of the area to be tested; perform three-dimensional unit body segmentation on the geological structure model to construct a three-dimensional unit body attribute data structure; the three-dimensional unit body attribute data structure includes spatial information and attribute information;

[0028] The three-dimensional unit body attribute information value calculation module is used to identify the cracks, lithology distribution and broken areas in the surrounding rock image based on the surrounding rock image, drilling perception and in-situ test data information using the convolutional neural network model; according to the discreteness of the data, the three-dimensional unit body attribute information value of the entire area to be tested is simulated;

[0029] The geological body attribute model building module is used to obtain the geological body attribute model of the area to be measured based on the three-dimensional unit body attribute data structure and the three-dimensional unit body attribute information value.

[0030] In other embodiments, the following technical solutions are adopted:

[0031] A terminal device includes a processor and a memory, wherein the processor is used to implement instructions; the memory is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the above-mentioned three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion.

[0032] In other embodiments, the following technical solutions are adopted:

[0033] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the above-mentioned three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) The three-dimensional unit volume attribute modeling method of the present invention combines the geological data inside and outside the tunnel with the data of drilling, geophysical exploration, and surrounding rock image data. It can comprehensively utilize various geological data sources to realize the three-dimensional visualization of the attribute distribution and structure of the underground space, which is convenient for construction personnel to view and refer to. It solves the limitation that a single geological data source is presented in the form of text or two-dimensional charts and the results can only be interpreted by professionals.

[0036] (2) The present invention obtains poor geological bodies based on the implicit surface generation algorithm of GS-RBF, which usually have complex boundary shapes due to their irregularity and complexity. The implicit surface generation algorithm of GS-RBF can provide a fine representation of the boundaries of these complex geological bodies, so that the model can more accurately reflect the actual geological conditions.

[0037] At the same time, the implicit surface generation algorithm can generate a continuous geological body model throughout the entire space. By considering the boundaries of the bad geological body, the algorithm can better capture the transition and changes of the geological structure, making the model more consistent and realistic in space.

[0038] Furthermore, the present invention can adaptively update the implicit function based on the size of the update range of the geophysical exploration data, taking into account the relative speed of solving the implicit function and the Boolean calculation.

[0039] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of a three-dimensional unit body attribute modeling method for multi-source heterogeneous information fusion in an embodiment of the present invention;

[0041] Figure 2 A flowchart of modeling and updating of bad geological bodies in an embodiment of the present invention;

[0042] Figure 3 This is the identification and quantification process of the unit crack density in the embodiment of the present invention;

[0043] Figure 4 This is a flow chart of the DexiNed edge detection algorithm in an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of an upsampling block in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0046] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0047] Embodiment 1

[0048] In one or more embodiments, a three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion is disclosed, combined with Figure 1 , specifically including the following process:

[0049] (1) Based on the geological mapping and borehole logging information in the survey data, the stratigraphic boundary and groundwater level are obtained, and the stratigraphic surface is further fitted to generate a mean stratigraphic volume;

[0050] In this embodiment, geological mapping and borehole logging information in the survey data are used to obtain stratigraphic boundary information, including stratigraphic boundary lines in geological profiles, stratigraphic change points in borehole logging, and groundwater levels, and a surface fitting algorithm is used to generate stratigraphic interfaces.

[0051] The specific surface fitting method is implemented by existing technologies, including but not limited to the least square method, spline surface method, RBF interpolation, RANSAC (random sampling consensus) algorithm, B-spline surface, NURBS surface, T-spline surface, etc. According to the characteristics of the survey data of different sites, the appropriate surface fitting algorithm is selected, and factors such as the number of data points, accuracy requirements, and calculation speed are usually considered; based on the fitting formation surface, a homogeneous formation body is generated.

[0052] (2) Inversion is performed based on the acquired geophysical data to obtain the type and scale information of the adverse geology, and then the adverse geological body is obtained based on the implicit surface generation algorithm of GS-RBF (Generalized Shepard Radial Basis Function); the type and scale information of the adverse geological body obtained by inversion is actually some discrete spatial data points. The construction of the adverse geological body is actually to fit the boundary surface of the adverse geological body with these discrete points. The boundary surface of the adverse geological body obtained by fitting is the entity filled with the adverse geological body.

[0053] In this embodiment, the type and location information of the adverse geology are obtained based on the inversion analysis of the geophysical data; the geophysical data can be data obtained by means of seismic wave reflection, electrical and electromagnetic methods; using the acquired underground physical field data, applying scattering theory and other methods, the reflection, refraction and other phenomena in the data are explained, and the existence and distribution of adverse geological bodies (such as water-rich, karst, faults, etc.) are obtained.

[0054] Based on the interpreted geological information, appropriate RBF is selected to represent the boundary of the bad geological surface, and an implicit surface body based on GS-RBF, i.e., the bad geological body, is constructed. Common RBFs (Radial Basis Functions) include Gaussian functions, multiple Gaussian functions, etc.

[0055] In the RBF model, it is necessary to select center points, which will be used as the reference points of the RBF. The selection of center points should be reasonably selected according to the distribution of the poor geological body. Some parameters also need to be determined in the RBF model, such as the shape parameters and amplitude of the RBF. These parameters can be determined by methods such as data fitting. Using the selected RBF and parameters, implicit geological surface bodies are constructed. These surface bodies can represent the interfaces of different underground geological units. The implicit surface fitting goal of discrete data points is to find N constraint points (x i ,y i ,z i )’s function F:

[0056] F(x i ,y i ,z i )=0,i=1,…,N

[0057] Such a surface is called an implicit surface or zero isosurface. Solving this function directly will result in a trivial solution (i.e., zero solution). To solve this problem, at each constraint point, move a small distance inward or outward along the normal direction to obtain a new offset point and assign a non-zero value to satisfy the formula:

[0058]

[0059] where r i =(x i ,y i ,z i ) represents discrete bad geological points. In order to make the implicit surface smooth enough, we adopt the following general solution (that is, all possible solutions of the surface function):

[0060]

[0061] Among them, the function is the radial basis function, Represents the distance between two constraint points. In this embodiment, the Gaussian kernel is selected as the radial basis kernel function, that is, the form is The probability distribution function of i represents the distance between any two points, and c is the shape function.

[0062] Before solving the general solution, in order to ensure the reversibility of interpolation, a linearly independent polynomial function needs to be added. The present invention chooses to add a first-order polynomial: p(x) = k1x + k2y + k3z + k4. In this way, the general solution is equivalent to The parameters that need to be solved are λ1,…,λ n There are N+4 k1, k2, k3, k4 in total. In addition to ensuring the reversibility of the interpolation, it is also necessary to ensure the smoothness of the implicit surface.

[0063] This requires that the radial basis function must belong to the Beppo Levi space, that is, satisfy the following constraints:

[0064]

[0065] make The above problem can be transformed into solving the following linear equations:

[0066]

[0067] Among them, the function is the radial basis function, Represents the distance between two constraint points; r i 、r j Represent two constraint points respectively; K, M, O are all abbreviation symbols, k1, k2, k3, k4, λ i They are all unknown parameters in the first-order linear equation constructed during the solution process and have no specific meanings; K, M, and O are all ellipsis symbols in the linear equation system, K represents column ellipsis, M represents row ellipsis, and O represents intersection ellipsis of rows and columns.

[0068] Finally, the implicit function F is solved, which is a closed surface, and this closed surface is used to represent the poor geological body.

[0069] It should be noted that if new geophysical data is added later, expert experience is needed to judge the range of updated data. If the updated geophysical data changes within a large range (the range of change exceeds the set threshold), it is necessary to comprehensively consider the new data and the old data to reconstruct the implicit surface of discrete poor geological points. The size of the update range is determined based on expert opinion. If only a small part is updated (the range of change is less than the set threshold), the implicit function of this part must be constructed separately. Then, a Boolean calculation is performed on the implicit function and the previous implicit function to obtain a new implicit function. Finally, the updated poor geological body is obtained. The poor geological body modeling and update process is as follows: Figure 2 The adaptive update method makes full use of the fast Boolean calculation of implicit surfaces and takes into account the relative speed of solving implicit functions and Boolean calculations.

[0070] (3) Embed the unfavorable geological body into the mean stratigraphic body to obtain the geological body structure model of the area to be tested; divide the geological structure model into three-dimensional units and construct a three-dimensional unit attribute data structure; the three-dimensional unit attribute data structure includes spatial information and attribute information, wherein the spatial information includes the three-dimensional spatial coordinates of the unit, and the attribute information includes water content attributes, fracture density attributes, element content attributes, rock strength attributes, ground stress attributes, lithology distribution attributes, etc.

[0071] In this embodiment, each adverse geological body (water-rich, karst cavity, fault) obtained by Boolean calculation is embedded into a homogeneous stratum body as a whole to obtain a geological body structure model of the site.

[0072] The geological structure model of the site is divided into closely arranged three-dimensional units (three-dimensional units are cubes of customizable size, but at special locations such as site boundaries and stratigraphic boundaries, units can be divided into irregular cubes by boundary constraints to better adapt to the model structure. The unit size can be set based on the overall geological model size and attribute density of the site), and the center of each unit is a point, called a geological point. Then, a geological point attribute data structure is constructed, and each type of unfavorable geology is stored as an attribute in the geological point (that is, when the unit is located in an unfavorable geological body such as water-rich, karst, or fault, the attribute of the unit is assigned a value of 1).

[0073] The three-dimensional unit body attribute data structure includes: water content attribute, crack density attribute, element content attribute, rock mass strength attribute, ground stress attribute and lithology distribution attribute;

[0074] (4) Based on the surrounding rock image, drilling perception and in-situ test data information, the convolutional neural network model is used to identify the cracks, lithology distribution and broken areas in the surrounding rock image; according to the discreteness of the data, the three-dimensional unit volume attribute information value of the entire test area is simulated;

[0075] In this embodiment, the collected multi-source data is analyzed. For relatively special image data, a pre-processing step is required first, using convolutional neural network algorithms based on deep learning such as semantic segmentation and edge detection to identify the cracks, lithology, and broken areas in the surrounding rock image, and based on the recognition results and zoning and grading methods, the information such as the occurrence of cracks, lithology distribution, and rock mass integrity is coarsened and extracted.

[0076] Among them, the fracture-related statistical data are obtained through the fracture occurrence information, and then the fracture line / surface density is obtained. Figure 3 Taking cracks as an example, the identification and quantification process of unit crack density is demonstrated; the details are as follows:

[0077] (4-1) The DexiNed (Dense Extreme Inception) edge detection algorithm is used to identify the surrounding rock traces. Figure 4 A specific structure example of DexiNed is given.

[0078] The DexiNed edge detection algorithm can predict edge mapping from a given image, just like traditional methods based on pattern recognition and image analysis, and to a certain extent overcomes the problems of low efficiency and accuracy in feature extraction in traditional machine learning, with better prediction performance. It does not have the problem of relying on a large number of training samples that most deep learning methods have, and can be applied to any edge detection task without pre-training or fine-tuning.

[0079] In this embodiment, the ultimate goal of DexiNed design is to generate thin edges to enhance the accuracy of the predicted trace map. One of the key components of DexiNed for edge thinning is the Upsampling Block (UB), such as Figure 5 As shown; Figure 4Each output (I-VI) from the Dexi block in is fed to UB. UB consists of conditionally stacked sub-blocks. Each sub-block has two layers, one is convolution (Conv) and the other is deconvolution (Deconv). There are two types of sub-blocks. The first seed block (sub-block1) is fed from Dexi or the second seed block (sub-block2), which is only used when the scale difference between the feature map and the ground truth (GroundTruth, GT) is equal to 2; when the difference is greater than 2, the second seed block (sub-block2) will be enabled to iterate until the scale of the feature map relative to the ground truth (GT) reaches 2. The first seed block (sub-block1) is set as follows: the kernel size of the convolution layer (Conv) is 1×1; followed by the activation function ReLU; the kernel size of the deconvolution layer (DeConv) or the transposed convolution (transpose convolution) is s×s, where the feature at scale s is the input feature map scale level; both layers return a filter, and the last layer gives a feature map of the same size as the ground truth (GT). The last layer Conv has no activation function. The second seed block (sub-block2) is set similarly to the first seed block (sub-block1), except that the number of filters is 16. For example, the output feature map (feature) of the main block VI from Dexi map) has a scale of 16, and three iterations are performed in sub-block2 before feeding it to sub-block1. The upsampling process starting from the second layer of the sub-block can be completed by bilinear interpolation, sub-pixel convolution, and transposed convolution.

[0080] (4-2) Gabor filtering is used to perform image enhancement processing on the surrounding rock trace recognition results to improve the recognition accuracy of the trace morphology;

[0081] (4-3) The DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm is used to perform cluster analysis on the identified traces, and statistical information on the shape, center position, trace length, inclination and other parameters of a single trace is used for subsequent calculation of unit fracture density.

[0082] DBSCAN is a density-based clustering algorithm that can be used for cluster analysis of pixels. It identifies clusters by dividing pixels into different density-connected regions and marks isolated outliers as noise. The clustering principle of DBSCAN is as follows: For a given pixel, if its neighborhood contains a specified minimum number of samples (MinPts) of pixels, it is called a core point. MinPts is a predefined threshold used to determine the minimum number of samples that should be included in the neighborhood of a core point. For a core point, if any pixel in its neighborhood is also a core point, then these pixels are directly density reachable. For any pixel, if there is a sequence of core points such that two adjacent core points are directly density reachable, then these pixels are density reachable. If two pixels are density reachable or density connected, then they belong to the same cluster.

[0083] The steps of the DBSCAN algorithm are as follows:

[0084] ①Randomly select an unvisited pixel.

[0085] ② Determine whether the pixel is a core point. If so, create a new cluster and add it to the cluster.

[0086] ③ According to the neighborhood of the core point, recursively add all directly density-reachable pixels to the current cluster.

[0087] ④ Repeat steps ①-③ until all pixels are visited.

[0088] The advantage of DBSCAN is that it can effectively handle noise and outliers without the need to pre-specify the number of clusters. It can automatically identify clusters of different shapes, sizes, and densities, and is relatively less sensitive to the choice of parameters.

[0089] The line segment fitting adopts the least squares method. When the least squares method is used to fit the line segment of the pixel points, the main idea is to find a straight line so that the sum of the squares of the distances between the straight line and the discrete pixel points is minimized. The following is the specific principle of using the least squares method for line segment fitting: the fitting model of the line segment is defined according to the pixel point data set obtained by cluster analysis (these pixel points should correspond to the expected line segment shape). In a simple case, the straight line can be represented by its slope and intercept, that is, a suitable loss function is selected to measure the distance from the observation point to the fitting line segment. The square of the Euclidean distance is selected in this method, that is, the square of the perpendicular distance from the observation point to the fitting line segment. This can be achieved by calculating the perpendicular distance from each pixel point to the fitting line segment and then summing the squares of all distances. The best fitting line segment is determined by minimizing the loss function. This can be achieved by solving the parameters of the fitting model (such as the slope and intercept) to minimize the sum of the squares of the distances from the observation point to the fitting line segment. Use an iterative optimization algorithm (such as the gradient descent method) for solution. By solving the optimization problem of minimizing the loss function, the optimal model parameters of the fitting segment, such as the slope, the coordinates of the segment center, and the segment length, are obtained. These parameters can be used to represent the equation or parameterized form of the trace fitting segment.

[0090] The calculation of unit crack density first divides the surrounding rock image into square cells of uniform size according to the custom standard, and counts the number, length and inclination information of all the cracks involved in the square cells. For each square cell, the length of each crack is calculated multiplied by the weight of its inclination, and then these values ​​are added to obtain the total weighted length of the crack. The specific formula is as follows:

[0091] Total weighted length of crack = ∑ i Crack length i ×Fissure inclination i

[0092] Where i represents each crack in the grid cell. Then calculate the cell crack density, the formula is as follows:

[0093]

[0094] The unit lithology and integrity identification and quantification process is similar to the above-mentioned unit fracture density identification and quantification process, except that the identification algorithms and quantification methods used are different, which will not be repeated here.

[0095] The pre-processed surrounding rock image information, the directly acquired while-drilling information and the in-situ test information are uniformly converted to the coordinate system of the geological body structure model according to the coordinate conversion formula and the reference point coordinate information;

[0096] By calculating the distance between each test information and the coordinates of the geological point, the multi-source data information is assigned to the unit body using the minimum distance principle (that is, each attribute is assigned to the unit body to which the geological point closest to it belongs). Specifically, the surrounding rock image exists in the form of an image. The above means can be used to obtain the crack density within the grid unit and the position information of the grid, and calculate the information of the grid and the geological point. The information while drilling is generally the lithology and crack information revealed by the borehole. By aligning the two control points at the start and end of the borehole with the geological space, the coordinate information of any position on the borehole and the related attribute information revealed can be obtained, and then the distance between the location of these attributes and the geological point can be calculated. In-situ test data refers to the actual location of the field test, and the corresponding information generally includes ground stress and water inflow.

[0097] The attribute information assigned to each geological point may come from different data sources, so there may be contradictions and conflicts. In this embodiment, the attribute information assigned to each unit body is digitized using a support vector machine for subsequent interpolation calculations. The support vector machine is a pre-trained model. By conducting model experiments in the laboratory, obtaining test data of different detection methods and data sets of actual attributes of the unit body, the support vector machine is trained to obtain a trained support vector machine.

[0098] Since the various attribute information obtained by detection is assigned to limited geological points, the attribute information of many undetected data points is blank, so it needs to be obtained through interpolation calculation. According to the discrete characteristics of the data of the attribute information assigned to each unit body, this embodiment uses sequential Gaussian simulation and sequential indicator simulation to perform interpolation calculations in the entire field area to obtain a multi-attribute geological body model of the entire field area. It should be noted that in the interpolation calculation process, the karst poor geological body obtained by geophysical inversion is special in nature. When the attribute passes through the area where it is located, the change law does not follow the general interpolation rules. Therefore, it is treated as an independent individual and is not included in the field area involved in the attribute interpolation calculation, that is, it does not participate in the interpolation calculation process of the attribute information; but for the two poor geological bodies of rich water and fault, when performing the interpolation calculation of water content attribute and fracture density attribute, the water content and fracture density attribute values ​​of the unit body contained in its area are set to 1 respectively, and when performing other attribute interpolation calculations, no interpolation calculation is performed.

[0099] (5) Based on the three-dimensional unit body attribute data structure and the three-dimensional unit body attribute information value, a geological body attribute model of the area to be measured is obtained.

[0100] In this embodiment, the geological body attribute model includes: a water content attribute model, a fracture density attribute model, an element content attribute model, a rock strength attribute model, a ground stress attribute model and a lithology distribution attribute model; the attribute data of different attribute models are normalized, and different weights are assigned to different attribute models according to their importance to obtain a fused geological attribute model.

[0101] For the acquired multi-attribute geological body model, its attribute distribution status can be checked separately, and different attributes can be assigned different weights according to their importance according to engineering needs to obtain a fused geological attribute model. Before establishing a fused geological attribute model, each attribute data needs to be normalized, that is, the dimension differences between various attribute values ​​are eliminated and mapped to the dimensionless [0, 1) interval. The fused geological attribute model can simultaneously reflect the comprehensive impact of different attributes on the geological field, which has certain engineering significance.

[0102] Embodiment 2

[0103] In one or more embodiments, a three-dimensional unit body attribute modeling system based on multi-source heterogeneous information fusion is disclosed, comprising:

[0104] The mean stratigraphic volume generation module is used to obtain stratigraphic boundaries and groundwater levels based on geological mapping and borehole cataloging information in the survey data, and further fit the stratigraphic surface to generate a mean stratigraphic volume;

[0105] The module for acquiring unfavorable geological bodies is used to perform inversion based on the acquired geophysical data to obtain the type and scale information of unfavorable geology, and then obtain the unfavorable geological bodies based on the implicit surface generation algorithm of GS-RBF;

[0106] A three-dimensional unit body attribute data structure construction module is used to embed the bad geological body into the mean stratum body to obtain the geological body structure model of the area to be tested; perform three-dimensional unit body segmentation on the geological structure model to construct a three-dimensional unit body attribute data structure; the three-dimensional unit body attribute data structure includes spatial information and attribute information;

[0107] The three-dimensional unit body attribute information value calculation module is used to identify the cracks, lithology distribution and broken areas in the surrounding rock image based on the surrounding rock image, drilling perception and in-situ test data information using the convolutional neural network model; according to the discreteness of the data, the three-dimensional unit body attribute information value of the entire area to be tested is simulated;

[0108] The geological body attribute model building module is used to obtain the geological body attribute model of the area to be measured based on the three-dimensional unit body attribute data structure and the three-dimensional unit body attribute information value.

[0109] The specific implementation methods of the above modules have been described in detail in Example 1 and will not be described in detail here.

[0110] Embodiment 3

[0111] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion in Example 1 when executing the program. For the sake of brevity, it will not be described here.

[0112] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0113] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0114] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0115] Embodiment 4

[0116] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion described in Example 1.

[0117] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion, characterized in that: include: Based on the geological mapping and drilling logging information in the survey data, the stratigraphic boundary and groundwater level are obtained, and the stratigraphic plane is further fitted to generate a mean stratigraphic volume; Based on the acquired geophysical data, inversion is performed to obtain the type and scale information of the adverse geology, and then the adverse geological body is obtained based on the implicit surface generation algorithm of GS-RBF; Embedding the bad geological body into the mean stratum body to obtain the geological body structure model of the area to be tested; performing three-dimensional unit body segmentation on the geological body structure model to construct a three-dimensional unit body attribute data structure; the three-dimensional unit body attribute data structure includes spatial information and attribute information; Based on the surrounding rock image, drilling perception and in-situ test data information, the convolutional neural network model is used to identify the cracks, lithology distribution and broken areas in the surrounding rock image; according to the discreteness of the data, the three-dimensional unit body attribute information value of the entire test area is simulated; Based on the three-dimensional unit body attribute data structure and the three-dimensional unit body attribute information value, a geological body attribute model of the area to be measured is obtained; The process of identifying the crack density in the surrounding rock image is as follows: The DexiNed edge detection algorithm is used to identify surrounding rock traces in surrounding rock images; Gabor filtering is used to enhance the image of surrounding rock trace recognition results; The DBSCAN clustering algorithm is used to divide the pixels into different density connected regions and perform cluster analysis on the identification traces; The line segment fitting adopts the least square method to determine the best fitting line segment by minimizing the loss function and using an iterative optimization algorithm to solve it; The surrounding rock image is divided into evenly sized square cells according to the custom standard, and the number, length and inclination information of all the cracks involved in the square cells are counted; For each grid cell, the length of each crack multiplied by the weight of its inclination angle is calculated, and then these values ​​are added to obtain the total weighted length of the crack; the unit crack density is obtained based on the ratio of the total weighted length of the crack to the area of ​​each grid cell.

2. A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as claimed in claim 1, characterized in that: The poor geological body is obtained based on the implicit surface generation algorithm of GS-RBF, specifically: Using the selected RBF and parameters, implicit geological surfaces are constructed; these surfaces can represent the interfaces of different underground geological units; the implicit surface fitting goal of discrete data points is to find N constraints in three-dimensional space. The implicit function F is: Constructing a general solution , ; make , convert the solution function F into solving the following linear equations: Among them, the function is the radial basis function, , represents the distance between two constraint points; , Represent two constraint points respectively; , , All represent ellipsis symbols. , , , , These are the parameters to be solved.

3. A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as claimed in claim 1, characterized in that: If the geophysical data is updated and the updated data is less than the set threshold, an implicit function is constructed for the updated geophysical data separately; the constructed implicit function is Boolean-calculated with the previous implicit function to obtain a new implicit function, and then the updated poor geological body is obtained.

4. A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as claimed in claim 1, characterized in that: If the geophysical data is updated and the updated data is greater than the set threshold, the implicit function is reconstructed based on the updated geophysical data.

5. The three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion according to claim 1, characterized in that: The geological structure model is divided into three-dimensional units to construct a three-dimensional unit attribute data structure, specifically: The geological structure model of the area to be tested is divided into closely arranged three-dimensional units, and the center of each unit is a point, called a geological point; a geological point attribute data structure is constructed, and each type of unfavorable geology is stored as an attribute in the geological point.

6. A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as claimed in claim 1, characterized in that: The surrounding rock images, drilling perception and in-situ test data information are converted into the coordinate system of the geological structure model. By calculating the distance between each test data and the coordinates of the geological point, the multi-source data information is assigned to the three-dimensional unit body using the minimum distance principle, and the support vector machine is used to digitize the attribute information assigned to each unit body.

7. A three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as claimed in claim 1, characterized in that: According to the discreteness of the data, the three-dimensional unit body attribute information value of the entire test area is simulated, specifically: According to whether the attribute information assigned to each three-dimensional unit body is continuous data or discrete data, sequential Gaussian simulation and sequential indicator simulation are used to perform interpolation calculations in the entire area to be measured to obtain the three-dimensional unit body attribute information values ​​of the entire area to be measured; among them, no interpolation calculation is performed for karst poor geological bodies, water-rich and fault poor geological bodies.

8. The three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion according to claim 1, characterized in that: The geological body attribute model includes: a water content attribute model, a fracture density attribute model, an element content attribute model, a rock mass strength attribute model, a ground stress attribute model and a lithology distribution attribute model; the attribute data of different attribute models are normalized, and different attribute models are assigned different weights according to their importance to obtain a fused geological attribute model.

9. A three-dimensional unit body attribute modeling system based on multi-source heterogeneous information fusion, characterized in that: include: The mean stratigraphic volume generation module is used to obtain stratigraphic boundaries and groundwater levels based on geological mapping and borehole cataloging information in the survey data, and further fit the stratigraphic surface to generate a mean stratigraphic volume; The module for acquiring unfavorable geological bodies is used to perform inversion based on the acquired geophysical data to obtain the type and scale information of unfavorable geology, and then obtain the unfavorable geological bodies based on the implicit surface generation algorithm of GS-RBF; A three-dimensional unit body attribute data structure construction module is used to embed the bad geological body into the mean stratum body to obtain the geological body structure model of the area to be tested; perform three-dimensional unit body segmentation on the geological body structure model to construct a three-dimensional unit body attribute data structure; the three-dimensional unit body attribute data structure includes spatial information and attribute information; The three-dimensional unit body attribute information value calculation module is used to identify the cracks, lithology distribution and broken areas in the surrounding rock image based on the surrounding rock image, drilling perception and in-situ test data information using the convolutional neural network model; according to the discreteness of the data, the three-dimensional unit body attribute information value of the entire area to be tested is simulated; A geological body attribute model building module is used to obtain a geological body attribute model of the area to be measured based on a three-dimensional unit body attribute data structure and a three-dimensional unit body attribute information value; The process of identifying the crack density in the surrounding rock image is as follows: The DexiNed edge detection algorithm is used to identify surrounding rock traces in surrounding rock images; Gabor filtering is used to enhance the image of surrounding rock trace recognition results; The DBSCAN clustering algorithm is used to divide the pixels into different density connected regions and perform cluster analysis on the identification traces; The line segment fitting adopts the least square method to determine the best fitting line segment by minimizing the loss function and using an iterative optimization algorithm to solve it; The surrounding rock image is divided into evenly sized square cells according to the custom standard, and the number, length and inclination information of all the cracks involved in the square cells are counted; For each grid cell, the length of each crack multiplied by the weight of its inclination angle is calculated, and then these values ​​are added to obtain the total weighted length of the crack; the unit crack density is obtained based on the ratio of the total weighted length of the crack to the area of ​​each grid cell.

10. A terminal device, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as described in any one of claims 1-8.

11. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the three-dimensional unit body attribute modeling method based on multi-source heterogeneous information fusion as described in any one of claims 1-8.