A grouting risk area assessment method and system based on multimodal attribute constraints

Abstract

The present invention discloses a multimodal attribute-constrained grouting risk area assessment method and system, comprising: constructing a three-dimensional tunnel geological model of the area to be tested, obtaining the three-dimensional distribution of the tunnel face and underground cracks, performing crack modeling in the tunnel geological model, and adaptively refining the grid density according to the density of the cracks; based on the adjusted grid, spatially assigning crack density, porosity and water content attributes to the refined grid units of the tunnel geological model; and realizing the attribute transition between the refined grid units and the coarse grid units through the weighted averaging method; inputting the crack density, porosity and water content attributes of each grid unit into a trained deep learning geological model fusion network to obtain the permeability prediction results of each grid unit, and then evaluating the grouting risk area; the present invention significantly improves the prediction accuracy of the model in complex geological environments, and provides a scientific basis for the accurate division of areas requiring grouting.

Description

Technical Field

[0001] The present invention relates to the technical field of underground engineering grouting pre-control analysis, and in particular to a multi-modal attribute constrained grouting risk area assessment method and system. 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] In engineering geology and underground structure construction, grouting technology is widely used to improve the strength, stability and permeability of soil or rock mass, especially in engineering projects such as tunnels, underground caverns, and dams. The effect of grouting and the quality of construction directly affect the safety and durability of the project. Accurately predicting the area requiring grouting and its effect is of great significance for optimizing grouting design, improving construction efficiency and reducing costs. Traditional grouting area prediction methods mostly rely on geological exploration data, physical models or numerical simulations, but these methods often face problems such as insufficient accuracy and long prediction time when dealing with complex geological conditions, non-uniform porous media and multiple influencing factors. Therefore, how to accurately predict the area requiring grouting with the support of multimodal data has become a research hotspot in the current field of geotechnical engineering.

[0004] With the development of artificial intelligence and deep learning technologies, data-driven prediction methods have gradually become a mainstream research direction. Especially in complex geological conditions, the key to solving the problem of grouting area prediction is to fuse multi-source heterogeneous data (such as geological exploration data, acoustic waves, seismic waves, resistivity, radar signals, etc.) and effectively extract the potential information from different modal data. However, how to effectively fuse different modal data and improve prediction accuracy through multimodal fusion remains a major challenge in current research. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a grouting risk area assessment method and system with multimodal attribute constraints. By deeply exploring the correlation between continuous and discontinuous geological attributes, a multi-source attribute fusion model is constructed using the Deep Learning Geology Fusion Network (DL-GFN) to achieve dynamic and intelligent prediction of the geological conditions of the tunnel face.

[0006] In some embodiments, the following technical solutions are adopted:

[0007] A multimodal attribute-constrained grouting risk area assessment method includes:

[0008] Construct a three-dimensional tunnel geological model of the area to be measured, divide the tunnel geological model into initial coarse grid units and assign values;

[0009] Obtain the three-dimensional distribution of tunnel faces and underground cracks, model the cracks in the tunnel geological model, and adaptively refine the grid density based on the density of the cracks;

[0010] Based on the adjusted grid, the fracture density, porosity, and water content attributes are spatially assigned to the fine grid cells of the tunnel geological model. The attribute transition between the fine grid cells and the coarse grid cells is achieved through the weighted averaging method.

[0011] The fracture density, porosity and water content properties of each grid cell are input into the trained deep learning geological model fusion network to obtain the permeability prediction results of each grid cell, and then evaluate the grouting risk area; wherein, the deep learning geological model fusion network is based on the coupling relationship between fracture density, porosity and water content properties and permeability, and constructs a multi-physical field coupled permeability prediction model for each grid cell to realize the permeability prediction of each grid cell.

[0012] Furthermore, the three-dimensional distribution of the tunnel face and underground cracks is obtained. The specific process is as follows:

[0013] Generating two-dimensional plane distribution data of cracks from the tunnel face crack image, extracting three-dimensional geometric morphology of cracks from the crack point cloud data; combining the two-dimensional plane distribution data and the three-dimensional geometric morphology to obtain the three-dimensional distribution of cracks on the tunnel face;

[0014] The crack edges are extracted from the borehole crack images, and the three-dimensional distribution of underground cracks is extracted by combining the borehole position coordinates.

[0015] Furthermore, the fracture density, porosity and water content attributes are spatially assigned to the refined grid cells of the tunnel geological model, specifically:

[0016]

[0017] Among them, a f (x) is the attribute value of the refined grid unit x, a c is the property value of the initial coarse grid cell, x c is the fracture center grid cell, Δa is the local increment, and σ is the attribute influence range; the attribute represents fracture density, porosity, or water content.

[0018] Furthermore, the attribute transition between the refined grid and the surrounding coarse grid is achieved through the weighted averaging method, specifically:

[0019] At the junction of the fine grid cell and the surrounding coarse grid cells, a coupling region is defined. The fine grid cell closest to the current coarse grid cell in the coupling region is determined. The attribute values ​​of the fine grid cell and the current coarse grid cell are weighted averaged to obtain the fused attribute value of the current coarse grid cell:

[0020] The weight of the attribute value of the refined grid unit is determined according to the distance between the current coarse grid unit and the refined grid unit.

[0021] Furthermore, the weight w of the attribute value of the refined grid unit 细化 Specifically:

[0022]

[0023] The weight w of the current coarsening grid cell attribute value 周围 Specifically:

[0024] w 周围 =1-w 细化 ;

[0025] The fusion attribute values ​​of the current coarse grid cells are as follows:

[0026] Fusion attribute value = w 裂隙 ×Refine the grid cell attribute value + w 周围 ×Current coarse grid cell attribute value;

[0027] Wherein, d represents the distance between the current coarse grid unit and the refined grid unit, and α is a parameter for adjusting the influence range of the crack.

[0028] Furthermore, based on the coupling relationship between fracture density, porosity, water content and permeability, a multi-physics field coupled permeability prediction model is constructed for each grid cell, specifically:

[0029] The coupling relationship between fracture density and permeability is as follows:

[0030]

[0031] Among them, k f (x) is the permeability at the grid cell position x away from the center of the fracture, c is the center grid unit of the fracture, Δk is the increment of the fracture on the local permeability, σ is the influence range of the fracture, k avg is the average permeability;

[0032] The coupling relationship between porosity and permeability is specifically:

[0033]

[0034] Among them, k pore (x) is the permeability of the grid cell in the pore area, φ f (x) is the spatial distribution function of porosity, S is the specific surface area;

[0035] The coupling relationship between water content and permeability is as follows:

[0036] k sat (x) = k dry ·S w (x) n ;

[0037] Among them, k sat (x) is the permeability of the grid cell in the water-bearing area, k dry is the permeability under completely dry conditions, S w is the saturation, i.e. the ratio of water content to total pore space, and n is an empirical index reflecting the nonlinear effect of water saturation on permeability;

[0038] Construct a multi-physics field coupled permeability prediction model for each grid cell, specifically:

[0039] k total (x) = k f (x)+k pore (x)+k sat (x);

[0040] Among them, k total (x) is the coupled permeability of each grid cell.

[0041] Furthermore, based on the permeability prediction results, the grouting risk areas are evaluated, specifically:

[0042] If the predicted permeability value is greater than the set first threshold, it is determined to be a high-risk grouting area;

[0043] If the predicted permeability value is between the critical value and the first threshold, it is determined to be a risky area for grouting;

[0044] If the predicted permeability value is lower than the critical value, it is determined to be a low-risk area for grouting; the critical value is obtained based on actual engineering practice.

[0045] Furthermore, the deep learning geological model fusion network needs to ensure that the generated permeability field complies with the flow conservation law:

[0046]

[0047] Among them, Q is the total flow, k i Indicates the permeability of a certain area, A i represents the area of ​​the region, ΔP represents the pressure difference at both ends of the flow path, and L represents the length of the flow path.

[0048] In other embodiments, the following technical solutions are adopted:

[0049] A multimodal attribute-constrained grouting risk area assessment system, comprising:

[0050] The model building module is used to build a three-dimensional tunnel geological model of the area to be measured, divide the tunnel geological model into initial coarse grid units and assign values;

[0051] The mesh refinement module is used to obtain the three-dimensional distribution of the tunnel face and underground cracks, model the cracks in the tunnel geological model, and adaptively refine the mesh density according to the density of the cracks;

[0052] The attribute assignment module is used to perform spatial assignment of fracture density, porosity, and water content attributes to the fine grid cells of the tunnel geological model based on the adjusted grid; and to achieve attribute transition between fine grid cells and coarse grid cells through weighted averaging method;

[0053] The grouting risk assessment module is used to input the fracture density, porosity and water content properties of each grid unit into the trained deep learning geological model fusion network to obtain the permeability prediction results of each grid unit, and then evaluate the grouting risk area; wherein, the deep learning geological model fusion network is based on the coupling relationship between fracture density, porosity and water content properties and permeability, and constructs a multi-physical field coupled permeability prediction model for each grid unit to realize the permeability prediction of each grid unit.

[0054] In other embodiments, the following technical solutions are adopted:

[0055] 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 multi-modal attribute constrained grouting risk area assessment method.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) Improving the model's adaptability and prediction accuracy for complex geological conditions: This paper constructs a deep learning-based geological model fusion network to uniformly model continuous and discontinuous geological attributes such as fracture density, porosity, and permeability. It also combines dynamic fuzzy neural networks and message passing neural networks to implement multi-attribute nonlinear correlation mining and feature aggregation. This method can fully explore the potential relationships between multimodal data and dynamically adjust prediction parameters, significantly improving the model's prediction accuracy in complex geological environments and providing a scientific basis for the accurate demarcation of areas requiring grouting.

[0058] (2) The present invention adopts adaptive meshing technology to dynamically adjust the coarsening and refining grids of the tunnel geological model according to attribute values ​​such as crack density, porosity, permeability and groundwater flow rate. A smooth attribute transition mechanism between the coarsening and refining grids is constructed to avoid attribute mutations or discontinuities caused by differences in grid scales. This method not only ensures the accurate description of complex geometric features, but also reduces the consumption of computing resources, realizes the seamless connection and efficient modeling of attributes in multi-level and multi-scale grid structures, and significantly improves the computational efficiency and operability of the model.

[0059] (3) This paper proposes a coupled modeling method for fracture attributes and permeability attributes, comprehensively considering the local impact of multiple attributes on permeability and combining flow conservation conditions to achieve a reasonable allocation of attributes in the refined grid. This ensures the physical consistency of attribute assignment. The above mechanism enables the model to more accurately describe the permeability characteristics in complex geological structures, providing a scientific basis for grouting design and optimization in underground engineering.

[0060] 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 practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Flowchart of a grouting risk area assessment method with multimodal attribute constraints in an embodiment of the present invention;

[0062] Figure 2 Schematic diagram of the generator structure of the deep learning geological model fusion network in an embodiment of the present invention;

[0063] Figure 3 Schematic diagram of the discriminator structure of the deep learning geological model fusion network in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] It should be noted that the following detailed description is illustrative and is 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 meaning as commonly understood by those skilled in the art to which the present application belongs.

[0065] 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, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0066] Example 1

[0067] In one or more embodiments, a multimodal attribute-constrained grouting risk area assessment method is disclosed, combining Figure 1 , specifically including the following process:

[0068] S101: Construct a three-dimensional tunnel geological model of the area to be measured, and perform initial coarse mesh division on the tunnel geological model.

[0069] In this embodiment, the three-dimensional tunnel geological model is constructed according to the tunnel design dimensions or converted through a dxf format file.

[0070] Constructing a 3D tunnel geological model can be achieved in a variety of ways. One approach involves using the tunnel design dimensions, which provide the tunnel's geometry and structural dimensions, such as length, cross-sectional dimensions, and curve radius. Based on this, and incorporating the actual requirements of the tunnel project, a high-precision 3D geological model is constructed to describe the geological features and surrounding environment along the tunnel. Another common approach is to convert a 2D tunnel design plan (e.g., a DXF file) into a 3D model using CAD software, and then further integrate it with geological data for spatial modeling.

[0071] First, an initial coarse grid is established according to the set grid density to cover the entire 3D tunnel geological model.

[0072] In this embodiment, the attributes of the initial coarse grid, such as crack distribution, water content, and fragmentation, can be assigned by using the difference method based on the attribute values ​​obtained through borehole measurement to obtain the corresponding attribute value of each coarse grid unit.

[0073] S102: Obtain the three-dimensional distribution of the tunnel face and underground cracks, perform crack modeling in the tunnel geological model, and adaptively refine the grid density according to the density of the cracks.

[0074] In the process of fracture attribute modeling, this embodiment utilizes multi-source data sources, including face fracture images and fracture identification results obtained through point cloud data. The fracture image extracts the geometric features of the fracture through image processing technology (such as edge detection, deep learning fracture identification algorithm) to generate two-dimensional plane distribution data of the fracture. Point cloud fracture identification obtains high-precision point cloud data of the rock surface through three-dimensional scanning technology, and automatically extracts the three-dimensional geometric shape of the fracture in combination with the fracture identification algorithm. The point cloud fracture identification results usually contain information such as the spatial position of the fracture, the fracture surface normal vector, the inclination angle, etc., which can provide important basic data for the three-dimensional spatial modeling of the fracture. By jointly analyzing the fracture image and the point cloud fracture identification results, a more comprehensive and accurate fracture network is constructed. In this process, the borehole fracture image provides direct observation data of the fractures inside the borehole, so that the fracture attribute modeling can cover the fracture characteristics of the underground space. The borehole fracture image can automatically extract the fracture edge through the fracture identification algorithm, and form a three-dimensional distribution model of the fracture in combination with the borehole position coordinates. Finally, the fracture network is composed of the geometric information of surface and underground fractures, which enables the complete modeling of the fracture distribution around the tunnel.

[0075] Based on the constructed fracture network, this example employs an adaptive fine-meshing method to refine the fracture mesh. Taking into account the irregular morphology and multi-scale characteristics of fractures within the mesh cells, a finer mesh is used in areas with dense fractures or large variations in fracture parameters to improve the model's ability to depict complex fracture structures.

[0076] In this embodiment, based on the density of cracks in each grid unit (such as the length, width, or inclination of the cracks), when a certain attribute value of the cracks is greater than a set threshold, a refinement algorithm based on local error estimation is used to refine the corresponding grid unit.

[0077] The refinement algorithm based on local error estimation is as follows:

[0078]

[0079] Among them, E local (x,y,z) represents the error estimate of the local area, Indicates the optimization, reduction or minimization of the current quantity or target. is the gradient of the point, Ω is the spatial region where G continues to generate, and G is a control parameter related to grid division, which represents the fineness or density of the grid and is used to guide the generation and adaptive division of the grid.

[0080] S103: Based on the adjusted grid, spatially assign fracture density, porosity, and water content attributes to the fine grid cells of the tunnel geological model; and implement attribute transition between the fine grid cells and the coarse grid cells using a weighted averaging method;

[0081] In this embodiment, the spatial assignment of fracture density, porosity, and water content properties is performed on the refined grid cells of the tunnel geological model, specifically as follows:

[0082]

[0083] Among them, a f (x) is the attribute value of the refined grid unit, a c is the attribute value of the initial coarsened mesh cell, Δa is the local increment, and σ is the attribute influence range. The specific values ​​of Δa and σ are usually obtained through experimentation or historical data fitting to make the attribute distribution more realistic. The attribute here refers to any of fracture density, porosity, or water content.

[0084] Because the fracture area has a finer mesh, while the surrounding area has a relatively coarser mesh, this example implements a transition mechanism to smooth out the differences in properties between the finer fracture mesh and the coarser surrounding mesh. The core of this method is to gradually transition fracture property values ​​based on the distance from the fracture to the surrounding medium, ensuring that the physical properties are spatially continuous and smooth.

[0085] Specifically, in areas close to the crack, the mesh is finer, and the crack attribute values ​​dominate. As the distance from the crack increases, the influence of the crack attribute values ​​gradually decreases, and the attribute values ​​of the surrounding coarser mesh are more used. In areas farther away, the attribute values ​​of the coarser mesh are fully used.

[0086] To achieve attribute fusion, this embodiment adopts a weighted average method based on weights. The specific process is as follows:

[0087] At the junction of the crack fine mesh and the surrounding coarse mesh, a coupling region is defined. The fine mesh closest to the current coarse mesh in the coupling region is determined. The attribute values ​​of the fine mesh and the current coarse mesh are weighted averaged to obtain the fused attribute value of the current coarse mesh:

[0088] Among them, the weight of the refined grid attribute value w 裂隙 Specifically:

[0089]

[0090] The weight w of the current coarsened grid attribute value 周围 Specifically:

[0091] w 周围 =1-w 裂隙 ;

[0092] The fusion attribute values ​​of the current coarse grid are as follows:

[0093] Fusion attribute value = w 裂隙 ×Refine mesh attribute value + w 周围 ×Current coarse mesh attribute value;

[0094] Where d represents the distance between the current coarse grid and the fine grid, and α is a parameter for adjusting the influence range of the crack.

[0095] S104: Input the fracture density, porosity, and water content attributes of each grid cell into the trained deep learning geological model fusion network (DL-GFN) to obtain the permeability prediction results of each grid cell, and then evaluate the grouting risk area.

[0096] In this embodiment, the deep learning geological model fusion network (DL-GFN) constructs a multi-physics field coupled permeability prediction model for each grid cell based on the coupling relationship between fracture density, porosity, water content and permeability, so as to realize the permeability prediction of each grid cell.

[0097] As a specific implementation method, the multi-physics field coupled permeability prediction model for each grid unit is specifically as follows:

[0098] k total (x) = k f (x)+k pore (x)+k sat (x);

[0099] Among them, k total (x) is the coupled permeability of each grid cell; k f (x), k pore (x) and k sat (x) are the permeabilities obtained based on the coupling of fracture density, porosity and water content properties.

[0100] Specifically, the coupling relationship between fracture density and permeability is:

[0101]

[0102] Among them, k f (x) is the permeability at the grid cell position x away from the center of the fracture, c is the center grid unit of the fracture, Δk is the increment of the fracture on the local permeability, σ is the influence range of the fracture, k avg is the average permeability, which can be estimated from the initial permeability based on the fracture geometry and porosity.

[0103] The coupling relationship between porosity and permeability is specifically:

[0104]

[0105] Among them, k pore (x) is the permeability of the grid cell in the pore area, φ f (x) is the spatial distribution function of porosity, and S is the specific surface area.

[0106] The coupling relationship between water content and permeability is as follows:

[0107] k sat (x) = k dry ·S w (x) n ;

[0108] Among them, k sat (x) is the permeability of the grid cell in the water-bearing area, k dry is the permeability under completely dry conditions, S w is the saturation, that is, the proportion of water content to total pore space, and n is an empirical index that reflects the nonlinear effect of water saturation on permeability.

[0109] The Deep Learning Geological Model Fusion Network (DL-GFN) of this embodiment typically consists of two parts: a generator and a discriminator. The generator is responsible for generating a coupled permeability field for each grid cell based on input geological attributes (such as fracture length, width, and porosity) using the aforementioned multi-physics coupled permeability prediction model. Its core is to learn the complex mapping relationship between input geological attributes and permeability.

[0110] The network structure of the generator can use convolutional neural networks (CNNs), and its structure example is as follows Figure 2As shown in the figure, preliminary spatial features are extracted through the convolution layer (Conv Layer 1), and downsampled through the pooling layer (Pooling Layer 1) to reduce the feature size and extract global information; the convolution layer (Conv Layer 2) deepens the feature extraction and identifies more advanced spatial features; the pooling layer (Pooling Layer 2) and the residual block (Residual Block 1) further downsample and enhance the feature representation capability to capture complex spatial features. The feature map is amplified through the upsampling layer (Upsampling Layer 1) to restore the spatial resolution; the upsampled features are fused with the early encoded features through the concatenation layer (Concatenate Layer 1); the feature map is refined through the convolution layer (Conv Layer 3); then the global and local features are further fused through the upsampling layer (Upsampling Layer 2) and the concatenation layer (Concatenate Layer 2), and finally the permeability field prediction result is output through the convolution layer (Conv Layer 4). The generator of this embodiment uses multiple upsampling layers and convolutional layers to gradually generate a high-resolution permeability field; the network output dimension needs to match the spatial distribution of the entire grid (such as a two-dimensional or three-dimensional permeability field).

[0111] The goal of the discriminator is to determine whether the generated permeability field complies with actual physical constraints (such as flow conservation, rationality of permeability, etc.). The design of the discriminator needs to focus on the following aspects:

[0112] (1) Flow conservation constraint: The discriminator needs to ensure that the generated permeability field complies with the flow conservation law, that is, the local flow and the overall flow should be consistent.

[0113] In this embodiment, under the condition of flow conservation, the fusion permeability needs to meet the following conditions:

[0114]

[0115] Among them, Q is the total flow, k i Indicates the permeability of a certain area, A i represents the area of ​​the region, ΔP represents the pressure difference at both ends of the flow path, and L represents the length of the flow path.

[0116] (2) Geological constraints: The discriminator needs to verify whether the coupling relationship between fractures, pores and permeability is consistent with geological common sense, such as the greater the porosity, the greater the permeability.

[0117] An example of the discriminator network structure is as follows Figure 3As shown, it is a downsampling network. The discriminator receives the permeability field generated by the generator as input, extracts global and local features through the convolution layer (Conv Layer D1) and the pooling layer (PoolingLayer D1), and performs downsampling to compress the feature dimension. The discriminant feature extraction is deepened through the convolution layer (Conv LayerD2) and the pooling layer (Pooling Layer D2), and downsampled again; the features are mapped to a high-dimensional space through the fully connected layer (FullyConnected Layer) for classification; and finally it is judged whether the generated permeability field meets the physical and geological constraints. The discriminator of this embodiment uses the convolution layer to gradually reduce the dimension of the input data and extract global features; the output is a single scalar or a classification score, which is used to indicate whether the generated result meets the physical constraints or the real distribution.

[0118] In this embodiment, the loss function of the deep learning geological model fusion network (DL-GFN) should include the following parts when designed:

[0119] (1) Physical losses: Ensure that the permeability field satisfies the law of conservation of flow.

[0120] (2) Spatial distribution loss: The generated permeability is constrained by a Gaussian attenuation function (describing the spatial effect of cracks on permeability).

[0121] (3) Porosity and permeability coupling penalty: penalizes the deviation between the generated permeability and porosity.

[0122] (4) Flow conservation and physical constraint loss: In order to ensure that the permeability field generated by the model complies with the flow conservation law, flow conservation can be used as part of the loss function. The flow conservation law usually requires:

[0123]

[0124] Where p(x) is the pressure field, k(x) is the permeability field, and Ω is the flow area. This constraint can be implemented using physical losses or a physical guidance network to force the model to generate a permeability field that satisfies flow conservation.

[0125] Finally, the Deep Learning Geological Model Fusion Network (DL-GFN) model obtains the permeability prediction results for each grid cell and evaluates the grouting risk areas based on the permeability prediction results. Specifically:

[0126] If the predicted permeability value is greater than the set first threshold, it is determined to be a high-risk grouting area;

[0127] If the predicted permeability value is between the critical value and the first threshold, it is determined to be a risky area for grouting;

[0128] If the predicted permeability value is lower than the critical value, it is determined to be a low-risk area for grouting.

[0129] Among them, the critical value can be obtained through engineering practice.

[0130] A grouting risk distribution map (Risk Map) is generated based on the grouting risk levels in different areas, with different risk levels represented by color gradients. The permeability changes during the grouting process are also dynamically displayed (if time data is available).

[0131] This method comprehensively considers the local impact of multiple attributes on permeability and, in conjunction with flow conservation conditions, achieves a rational allocation of attributes within the refined grid. This ensures the physical consistency of attribute assignments. This mechanism enables the model to more accurately describe permeability characteristics in complex geological structures, providing a scientific basis for grouting design and optimization in underground engineering projects.

[0132] Example 2

[0133] In one or more embodiments, a multimodal attribute-constrained grouting risk area assessment system is disclosed, comprising:

[0134] The model building module is used to build a three-dimensional tunnel geological model of the area to be measured, divide the tunnel geological model into initial coarse grid units and assign values;

[0135] The mesh refinement module is used to obtain the three-dimensional distribution of the tunnel face and underground cracks, model the cracks in the tunnel geological model, and adaptively refine the mesh density according to the density of the cracks;

[0136] The attribute assignment module is used to perform spatial assignment of fracture density, porosity, and water content attributes to the fine grid cells of the tunnel geological model based on the adjusted grid; and to achieve attribute transition between fine grid cells and coarse grid cells through weighted averaging method;

[0137] The grouting risk assessment module is used to input the fracture density, porosity and water content properties of each grid unit into the trained deep learning geological model fusion network to obtain the permeability prediction results of each grid unit, and then evaluate the grouting risk area; wherein, the deep learning geological model fusion network is based on the coupling relationship between fracture density, porosity and water content properties and permeability, and constructs a multi-physical field coupled permeability prediction model for each grid unit to realize the permeability prediction of each grid unit.

[0138] It should be noted that the specific implementation of the above modules is the same as that in Example 1 and will not be described in detail.

[0139] Example 3

[0140] In one or more embodiments, a terminal device is disclosed, which 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 multi-modal attribute constrained grouting risk area assessment method described in Example 1.

[0141] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-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 any conventional processor, etc.

[0142] The memory may include a read-only memory and a random access memory, and provides 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.

[0143] During implementation, each step of the above method may be completed by an integrated logic circuit of hardware in a processor or by instructions in the form of software.

[0144] Although the above describes the specific embodiments 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 any creative work are still within the scope of protection of the present invention.

Claims

1. A multimodal attribute-constrained grouting risk area assessment method, characterized in that: include: Construct a three-dimensional tunnel geological model of the area to be measured, divide the tunnel geological model into initial coarse grid units and assign values; Obtain the three-dimensional distribution of tunnel faces and underground cracks, model the cracks in the tunnel geological model, and adaptively refine the grid density based on the density of the cracks; Based on the adjusted grid, the fracture density, porosity, and water content attributes are spatially assigned to the fine grid cells of the tunnel geological model. The attribute transition between the fine grid cells and the coarse grid cells is achieved through the weighted averaging method. The fracture density, porosity, and water content attributes of each grid cell are input into a trained deep learning geological model fusion network to obtain the permeability prediction results of each grid cell, thereby evaluating the grouting risk area. The deep learning geological model fusion network constructs a multi-physics field coupled permeability prediction model for each grid cell based on the coupling relationship between fracture density, porosity, and water content attributes and permeability, thereby realizing the permeability prediction of each grid cell. The attribute transition between the refined grid and the surrounding coarse grid is achieved by a weighted averaging method, specifically: a coupling region is defined at the junction of the refined grid cell and the surrounding coarse grid cells, the refined grid cell closest to the current coarse grid cell in the coupling region is determined, and the attribute values ​​of the refined grid cell and the current coarse grid cell are weighted averaged to obtain a fused attribute value of the current coarse grid cell: wherein the weight of the refined grid cell attribute value is determined according to the distance between the current coarse grid cell and the refined grid cell; The weight of the attribute value of the refined grid cell Specifically: ； The weight of the current coarse mesh cell attribute value Specifically: ； The fusion attribute values ​​of the current coarse grid cells are as follows: Fusion attribute values Refine the grid cell property value + Current coarse grid cell property values; in, represents the distance between the current coarse grid cell and the refined grid cell, It is a parameter to adjust the scope of influence of the crack.

2. A multimodal attribute-constrained grouting risk area assessment method according to claim 1, characterized in that: Obtain the three-dimensional distribution of the tunnel face and underground cracks. The specific process is as follows: Generating two-dimensional plane distribution data of cracks from the tunnel face crack image, extracting three-dimensional geometric morphology of cracks from the crack point cloud data; combining the two-dimensional plane distribution data and the three-dimensional geometric morphology to obtain the three-dimensional distribution of cracks on the tunnel face; The crack edges are extracted from the borehole crack images, and the three-dimensional distribution of underground cracks is extracted by combining the borehole position coordinates.

3. The multimodal attribute-constrained grouting risk area assessment method according to claim 1, characterized in that: The fracture density, porosity, and water content properties are spatially assigned to the refined grid cells of the tunnel geological model, specifically: ； in, To refine the mesh cells The attribute value of is the property value of the initial coarse grid cell, is the crack center grid unit, is a local increment, It is the attribute influence range; the attribute represents fracture density, porosity or water content.

4. The multimodal attribute-constrained grouting risk area assessment method according to claim 1, characterized in that: Based on the coupling relationship between fracture density, porosity, water content and permeability, a multi-physics field coupled permeability prediction model is constructed for each grid cell. Specifically: The coupling relationship between fracture density and permeability is as follows: ； in, is the distance from the crack center grid unit The permeability at the location, is the crack center grid unit, is the increment of local permeability due to cracks, is the crack impact range, is the average permeability; The coupling relationship between porosity and permeability is specifically: ； in, is the permeability of the grid cell in the pore area, is the spatial distribution function of porosity, is the specific surface area; The coupling relationship between water content and permeability is as follows: ； in, is the permeability of the grid cell in the water-bearing area, is the permeability under completely dry conditions, is the saturation, that is, the proportion of water content to the total pore space, is an empirical index reflecting the nonlinear effect of water saturation on permeability; Construct a multi-physics field coupled permeability prediction model for each grid cell, specifically: ； in, is the coupled permeability for each grid cell.

5. The multimodal attribute-constrained grouting risk area assessment method according to claim 1, characterized in that: Based on the permeability prediction results, the grouting risk areas are assessed, specifically: If the predicted permeability value is greater than the set first threshold, it is determined to be a high-risk grouting area; If the predicted permeability value is between the critical value and the first threshold, it is determined to be a risky area for grouting; If the predicted permeability value is lower than the critical value, it is determined to be a low-risk area for grouting; The critical value is obtained based on actual engineering practice.

6. The multimodal attribute-constrained grouting risk area assessment method according to claim 1, characterized in that: The deep learning geological model fusion network needs to ensure that the generated permeability field complies with the flow conservation law: ； in, is the total flow, represents the permeability of a certain area, represents the area of ​​the region, It represents the pressure difference between the two ends of the flow path. Indicates the flow path length.

7. A multi-modal attribute constrained grouting risk area assessment system, used to implement the multi-modal attribute constrained grouting risk area assessment method according to any one of claims 1 to 6, characterized in that: include: The model building module is used to build a three-dimensional tunnel geological model of the area to be measured, divide the tunnel geological model into initial coarse grid units and assign values; The mesh refinement module is used to obtain the three-dimensional distribution of the tunnel face and underground cracks, model the cracks in the tunnel geological model, and adaptively refine the mesh density according to the density of the cracks; The attribute assignment module is used to perform spatial assignment of fracture density, porosity, and water content attributes to the fine grid cells of the tunnel geological model based on the adjusted grid; and to achieve attribute transition between fine grid cells and coarse grid cells through weighted averaging method; The grouting risk assessment module is used to input the fracture density, porosity and water content properties of each grid unit into the trained deep learning geological model fusion network to obtain the permeability prediction results of each grid unit, and then evaluate the grouting risk area; wherein, the deep learning geological model fusion network is based on the coupling relationship between fracture density, porosity and water content properties and permeability, and constructs a multi-physical field coupled permeability prediction model for each grid unit to realize the permeability prediction of each grid unit.

8. A terminal device comprising a processor and a memory, wherein the processor is used to implement instructions; the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the multi-modal attribute constrained grouting risk area assessment method described in any one of claims 1-6.

Images