Tunnel smooth surface effect prediction method with self-adaptive blast hole parameters

Abstract

The invention discloses a blast hole parameter adaptive tunnel smooth surface effect prediction method, which comprises the steps of constructing a smooth surface effect prediction model based on a graph neural network; predicting rock mass displacement data by using the smooth surface effect prediction model; and calculating a tunnel smooth surface effect in combination with a prediction result of the rock mass displacement data. According to the method, the adaptive blast hole parameter smooth surface effect rapid prediction model based on the depth map neural network is constructed, smooth surface blasting data are comprehensively extracted by using the map neural network, the accuracy of a deep learning-based calculation method is improved, and the method has important theoretical significance and application value for tunnel blasting construction.

Description

Technical Field

[0001] The invention relates to the field of tunnel smooth surface blasting construction, and is a tunnel smooth surface effect prediction method based on artificial intelligence algorithm and blasthole parameter self-adaptation. Background Art

[0002] The smooth surface effect is an important factor affecting the quality and safety of blasting projects. Accurate prediction and optimization can not only improve the efficiency of blasting operations and reduce construction costs, but also effectively reduce the safety hazards caused by poor smooth surface effects. In blasting operations, the selection of blasthole parameters (such as hole depth, hole spacing, charge, etc.) directly affects the quality of the smooth surface effect and the damage mode.

[0003] At present, the research on smooth blasting using traditional methods includes "intelligent optimization system for smooth blasting of tunnel drilling and blasting method based on digital twin", which proposes to use information modules to obtain surrounding rock parameters, tunnel parameters, hole layout parameters and charging parameters; then optimize the construction process parameters, and build explosive state models, rock mass state models and air and water medium state models based on the optimized parameters; collect images of gravel and rock wall after blasting, detect the gravel particle size, and obtain the degree of over-excavation and under-excavation and the degree of half-hole retention as reference indicators for blasting quality evaluation, and study the blasting effect. Traditional methods usually need to rely on experience or manual adjustment to optimize blasthole design. Although they can provide certain guidance for engineering, they lack dynamic adaptability when facing complex geological conditions, changing blasthole parameters and multiple nonlinear factors.

[0004] Unlike traditional rule-based prediction models, deep learning models have strong generalization capabilities and can handle complex and nonlinear data relationships. Through continuous optimization and updating of the model, it can adapt to different geological conditions and different types of blasting operations, and adjust itself according to new blasting data. At present, the research on smooth blasting using artificial intelligence methods includes "Prediction Method and System for Smooth Blasting of Highway Tunnels", which proposes to establish sample data of the PSO-GA-LSSVM model, train the PSO-GA-LSSVM model through sample data, and use the trained model to predict the effect of smooth blasting of highway tunnels. However, the existing artificial intelligence methods have strict requirements on the use conditions and cannot adapt to the changes in blasthole parameters; and when constructing sample data, due to the inevitable data deviation in the process of converting unstructured data to structured data, it brings unpredictable deviations to the prediction accuracy of the artificial intelligence model. Summary of the invention

[0005] In order to improve the prediction accuracy and real-time performance of smooth surface effects in blasting projects, the present invention provides a tunnel smooth surface effect prediction method based on artificial intelligence algorithm and adaptive blasthole parameter smooth surface effect prediction model based on deep graph neural network is constructed, and the graph neural network is used to comprehensively extract smooth surface blasting data, thereby improving the accuracy of deep learning calculation methods, which has important theoretical significance and application value for tunnel blasting construction.

[0006] The technical solution for achieving the purpose of the present invention is:

[0007] In a first aspect, a method for predicting a tunnel smooth surface effect with self-adaptive blasthole parameters is provided, comprising the following steps:

[0008] Based on graph neural network, a glossy surface effect prediction model is constructed;

[0009] The rock mass displacement data is predicted using the smooth surface effect prediction model;

[0010] The tunnel smooth surface effect is calculated based on the prediction results of rock mass displacement data.

[0011] Preferably, the step of constructing a light surface effect prediction model includes:

[0012] Construct a database of glossy blasting effect images including training data sets and validation data sets;

[0013] Build a neural network model based on graph neural network;

[0014] The neural network model is trained and optimized using the training data set and the validation data set to obtain a glossy surface effect prediction model.

[0015] Preferably, the step of constructing a database of smooth surface blasting effect diagrams including a training data set and a verification data set comprises:

[0016] Mesh the three-dimensional geometric model of the tunnel face to obtain mesh data of the tunnel face;

[0017] The blasthole three-dimensional position data, blasthole blasting depth data, blasthole blasting time series data, and blasthole blasting charge data are sequentially encoded and processed into vectors, and then the vectors are normalized and spliced ​​to obtain a blasthole parameter matrix;

[0018] The grids in the face grid data are used as graph data nodes, and the faces between the grids are used as edges between the graph data nodes to generate an adjacency matrix;

[0019] In the graph data node, the grid position data and the blasthole parameter matrix are stored as graph data samples, and the corresponding rock displacement data are stored as graph data labels;

[0020] Obtain rock displacement data under different blasting hole parameters, obtain several different graph data samples and graph data labels, and stack them to form a graph structure data set, where the blasting hole parameters include blasting charge, blasting sequence, blasting hole position, and blasting hole depth for smooth blasting scenarios;

[0021] Divide the graph structure data set, enter it into the library system, and complete the construction of the smooth blasting effect graph database.

[0022] Preferably, the modified finite element simulation model is used to obtain rock displacement data in smooth blasting scenarios with different blasting charges, blasting sequences, blasthole positions, and blasthole depths, wherein the modified finite element simulation model acquisition step includes:

[0023] The tunnel face is fully scanned by 3D laser scanning to obtain 3D data of the tunnel face. After preprocessing the 3D data, a 3D geometric model of the tunnel face is reconstructed.

[0024] Three non-overlapping points are set as reference points in the tunnel face, and the distances between the blastholes and the reference points are measured in sequence to obtain the three-dimensional position data of the blastholes;

[0025] Use a laser rangefinder to obtain blast hole depth data;

[0026] The blast holes are numbered according to the set blasting time sequence information to obtain the blasting time sequence data of the blast holes;

[0027] According to the set blasting equivalent of the blasthole, collect the blasting charge data of the blasthole;

[0028] Perform finite element simulation on smooth blasting to obtain smooth blasting simulation data including at least rock mass displacement data and over-break and under-break volume;

[0029] Conduct smooth blasting tests to obtain smooth blasting test data including at least rock displacement data and over-break and under-break data;

[0030] The smooth surface blasting simulation data and smooth surface blasting test data are used as the smooth surface effect correction data. The finite element structure constitutive model is corrected using the finite element model correction method based on SGMD and LWOA-ELM to obtain the corrected finite element simulation model.

[0031] Preferably, the neural network model includes a feature compression encoding layer, an asynchronous feature association layer and a glossy surface effect decoding layer;

[0032] In the feature compression coding layer, several stacked graph convolutional networks (GCNs) are used to compress the input data. Perform feature extraction to obtain compressed feature data , where each GCN is connected to a ReLU activation function, and n represents the number of blastholes;

[0033] In the asynchronous feature association layer, Input n transformer encoders for calculation respectively, and the calculation results of each transformer encoder are input into a gated recurrent unit GRU respectively, and the residual connection is used to connect Add the corresponding GRU output result to get the hidden state ;Will Together as The corresponding GRU input also uses residual connection to Add the corresponding GRU output result to get the hidden state ;Will and Together as The corresponding GRU input also uses residual connection to Add the corresponding GRU output result to get the hidden state ; and so on, to Together as The corresponding GRU input also uses residual connection to Add to the corresponding GRU output result to get the hidden state ;

[0034] Glossy effect decoding layer: using stacked GCN to hide the state Perform expansion reduction to obtain rock mass displacement map data , where each GCN is followed by a ReLU activation function.

[0035] In a second aspect, a computer-readable storage medium storing one or more programs is also provided. The one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform the method as described above.

[0036] According to a third aspect, an electronic device is also provided, comprising one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as described above.

[0037] Compared with the prior art, the present invention has the following significant advantages:

[0038] 1. Efficient calculation: Different from the layer-by-layer reasoning method of traditional numerical simulation, the tunnel surface effect prediction method adopts an end-to-end deep learning prediction model. Through complete training, the rock displacement data can be directly predicted in milliseconds, and then the surface effect data can be quickly calculated;

[0039] 2. Accurate prediction: Using graph neural network as the prediction model structure, it fully covers all grid data and avoids distortion and offset during grid data conversion;

[0040] 3. Good robustness: By establishing a smooth surface blasting effect graph database containing multiple blasting scenarios, the graph neural network model is effectively trained, and the generalization ability of smooth surface prediction of different blasting charges, blasting sequences, blasthole positions, and blasthole depths is achieved.

[0041] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is the main flow chart of the tunnel smooth surface effect prediction method with adaptive blasthole parameters.

[0043] Figure 2 It is a schematic diagram of the glossy surface effect prediction model framework. DETAILED DESCRIPTION

[0044] like Figure 1 As shown, the present invention provides a tunnel smooth surface effect prediction method with adaptive blasthole parameters, and the specific steps are as follows:

[0045] S10, light surface effect data collection: collect test data and simulation data, compare data differences, use test data to correct simulation model, and then use simulation model to generate a large amount of light surface effect data and collect the light surface effect data;

[0046] The light surface effect data collection step comprises:

[0047] S11. Tunnel face geometry data acquisition: The tunnel face is fully scanned by 3D laser scanning to obtain 3D data of the tunnel face. Then, the pre-processed 3D data is converted into a 3D geometry model of the tunnel face by 3D surface reconstruction technology.

[0048] S12. Collection of parameters of smooth blasting blastholes: three non-overlapping points are set as reference points in the face, and the distances between the blastholes and the reference points are measured in sequence, and the three-dimensional position data of the blastholes are obtained based on the distances; the depth data of the blastholes are obtained using a laser rangefinder; the blastholes are numbered according to the set blasting sequence information, and the blasting sequence data of the blastholes are obtained; and the blasting charge data of the blastholes are collected based on the set blasting equivalent of the blastholes.

[0049] S13. Correction data collection: perform finite element simulation on smooth surface blasting to obtain finite element simulation data, and carry out smooth surface blasting test to obtain smooth surface blasting test data, collect rock displacement data and over-excavation and under-excavation of simulation and test as smooth surface effect correction data.

[0050] S14. Finite element simulation model correction: Use the finite element model correction method based on SGMD and LWOA-ELM, and use the correction data to correct the finite element structural constitutive model.

[0051] S15. Generation of smooth surface effect data: Use Python to call the modified finite element simulation model to batch generate smooth surface effect data for different blasting hole parameters (i.e. different blasting charges, blasting sequences, hole positions, and hole depths) and collect the smooth surface effect data, where the smooth surface effect data includes rock displacement data and over-excavation and under-excavation calculated from the rock displacement data, that is, the difference between the actual excavated section and the designed excavation contour line.

[0052] S20, graph database production: mesh the face geometry, and then encode the blasthole parameters; use the OpenFoam source file parsing method to obtain graph structure data, and then batch store the blasthole parameter matrix and rock displacement data into the graph structure data to obtain a graph data sample set and a graph data label set, and divide the data set ratio to obtain a graph database.

[0053] The graph database preparation steps include:

[0054] S21. Geometry meshing of the tunnel face: meshing the three-dimensional geometry model of the tunnel face to obtain mesh data of the tunnel face.

[0055] S22, blasthole parameter coding processing: convert each item of coding in the blasthole parameter data into a vector, perform data normalization on the vector, and then concatenate the normalized vectors to obtain a blasthole parameter matrix that integrates multi-source data.

[0056] S221, blasthole parameter data encoding: blasthole three-dimensional position data, blasthole blasting hole depth data, blasthole blasting time series data, and blasthole blasting charge quantity data are encoded into vectors in sequence.

[0057] S222, blasthole parameter normalization processing: perform minimum-maximum normalization processing on the encoded vector.

[0058] S223, blasthole parameter vector stitching: splice the normalized vectors to obtain a blasthole parameter matrix that fuses multi-source data.

[0059] S23. Graph structure data generation: Graph structure data is produced using the idea of ​​using grids as graph nodes and surfaces between grids as edges between graph nodes, and the OpenFOAM source file parsing method is used to specifically implement the conversion from grid to graph structure data.

[0060] Introduction to basic principles: Corresponding node and edge information is extracted from the grid data. Nodes represent cells in the grid, while edges represent the connection relationship between these cells. Each node is given the corresponding characteristics of the grid it represents, including the coordinates, attributes or other relevant information of the grid. Secondly, considering the topological characteristics of the graph data structure, if two cells are adjacent in the grid, then in the graph data structure, the nodes corresponding to these two cells will be connected by an edge. The weight of the edge is calculated based on the numerical information in the two cells. For example, the average or difference of the values ​​in the two cells is taken as the weight of the edge. Finally, a new graph data structure is obtained in this way, which not only retains the spatial distribution characteristics of the original grid data, but also inherits the numerical information in the grid data.

[0061] S231, OpenFOAM source file analysis: The mesh files of the example are stored in the constant / polyMesh directory of the OpenFOAM example. The analysis content includes points and faces file information. The files are analyzed in turn to obtain mesh data.

[0062] S2311, points file analysis: This file records all the nodes of the grid. The coordinate information of each node is arranged in the order of (xyz), that is, the three-dimensional space coordinates of the analyzed node.

[0063] S2312. Faces file parsing: This file records all faces and the node numbers that make up each face, that is, the order of parsing all nodes in the points file.

[0064] S232, graph structure data generation: obtain node information and edge information according to the parsed points file and faces file.

[0065] S2321, graph node generation: traverse from the lower left corner of the calculation domain to the right, and when reaching the right boundary, go up one row in the positive direction of the vertical axis, and then traverse and parse in a loop; according to the node information in the parsed points file, store the node coordinate information in the array in sequence according to the order of the nodes, and obtain the node features of the graph data.

[0066] S2322, graph edge generation: According to the parsed faces file, the edge connection relationship of the nodes is obtained; according to these connection relationships, a graph structure representation of the node edges, that is, an adjacency list, is constructed, and the existing adjacency lists of all nodes are concatenated to obtain the adjacency matrix of the graph structure data.

[0067] For example, for a structured grid, each face contains 4 points. The representation of a face in the faces file is 4(1 6 31 26), where 4 means that the face is composed of four nodes, and the brackets represent the four node numbers that make up the first face. According to the node numbers, the edges that make up the face are: [1, 6], [6, 31], [31, 26], [26, 1].

[0068] S24. Graph data sample production: Produce graph structure data sequence samples for input into the prediction model, and store grid position data and blasthole parameter matrix in the graph structure data nodes.

[0069] The graph data sample preparation step comprises:

[0070] S241, graph structure sample generation: grids in the face grid data are used as graph data nodes; faces between grids are used as edges between graph data nodes, and an adjacency matrix is ​​generated.

[0071] S242, graph structure sample node storage: Assume that the number of blastholes is n, then a single sample input into the prediction model contains n graph structure data, which is a sequence of length n, namely ; Graph structure data input at the first time step In the example, each node stores the grid position data of the corresponding grid and the blasthole parameter matrix of the first blasting; the graph structure data input in the second time step In each node, the grid position data of the corresponding grid and the blasthole parameter matrix of the second blasting are stored; and so on. .

[0072] S25, graph data label production: produce graph structure label data, and store glossy effect data in graph structure data nodes.

[0073] The step of making the graph data label comprises:

[0074] S251, graph structure label generation: using the grids in the face grid data as graph data nodes and the faces between the grids as edges between the graph data nodes to generate an adjacency matrix.

[0075] S252, graph structure label node storage: correspond the rock mass displacement data to the tunnel face mesh data; store the rock mass displacement data in the graph data node corresponding to the tunnel face mesh to obtain the graph data label.

[0076] S26. Batch generation of samples and labels: Steps S21 to S25 are performed in sequence on the smooth surface blasting data with different blasting charges, blasting sequences, blasthole positions, and blasthole depths to obtain a large number of different graph data samples and graph data labels.

[0077] S27. Graph data set production: stack the results of all smooth surface blasting scenes to form a graph structure data set, divide all data into a training graph data set and a verification graph data set in a ratio of 7 to 3, and then enter them into the library system to establish a smooth surface blasting effect graph database.

[0078] S30, building a glossy surface effect prediction model framework: constructing a neural network model structure consisting of a feature compression coding layer, an asynchronous feature association layer, and a glossy surface effect decoding layer to obtain a glossy surface effect prediction model framework;

[0079] like Figure 2 As shown, the steps of building the light surface effect prediction model framework include:

[0080] S31, Feature compression coding layer: Use Graph Convolutional Network (GCN) as the basic module of the feature compression coding layer. The three-layer GCN performs Extract features, and connect ReLU activation function after each layer to improve the nonlinearity of the model. The input data is calculated by the feature compression coding layer to obtain compressed feature data .

[0081] S32, asynchronous feature association layer: Introduce transformer encoder and gated recurrent units (Gated Recurrent Units, GRU) as the basic units of the asynchronous feature association layer; the compressed feature data calculated by the transformer encoder , use GRU to calculate the output of the first time step and add the residual connection to calculate the hidden state ; Compressed feature data after transformer encoder calculation , use GRU to calculate the output of the second time step and add the residual connection to calculate the hidden state ; Skip connection is used between each layer of units to prevent the model gradient disappearance problem during the training phase, and the last layer output hidden state is obtained by analogy .

[0082] S33, glossy effect decoding layer: GCN is used as the basic module of the glossy effect decoding layer. The three-layer GCN is used to decode the input data. Dilation and restoration: ReLU activation function is connected after each layer to improve the nonlinearity of the model. The input data is calculated by dilation and restoration to obtain the rock mass displacement map data. .

[0083] S34, constructing a prediction model framework: based on the neural network model structure of the feature compression coding layer, the asynchronous feature association layer and the glossy surface effect decoding layer, a glossy surface effect prediction model framework is obtained.

[0084] S40, glossy surface effect prediction model training: using the training graph data set prediction model to train, to obtain a glossy surface effect prediction model, and using the verification graph data set to verify the general applicability of the model.

[0085] The light surface effect prediction model training step comprises:

[0086] S41, Model calculation: After a single sample data is input into the model, the graph data of all time steps After the feature compression coding layer, the compressed feature data is obtained ; In the asynchronous feature association layer, the graph data of all time steps According to the blasting order, the basic unit of the asynchronous feature association layer is input, and the image data after the blasting parameter compression feature of the first blast hole is stored. As the input of the first time step, the hidden state is calculated ; Store the second blast hole blasting parameter compressed feature map data As the input of the second time step, combined with the hidden state of the previous time step Calculate the hidden state of the second time step ; Store the compressed feature data of the third blast hole blasting parameters As the input of the third time step, combined with the hidden states of the first two time steps and Calculate the hidden state of the third time step ; And so on, enter the remaining , calculate the hidden state of the last time step ; The last hidden state The rock mass displacement map data is obtained by calculating the smooth surface effect decoding layer .

[0087] S42, parameter update: based on rock mass displacement map data Calculate the value of the loss function mean square error (MSE) and calculate the gradient of the model parameters relative to the loss value, and use the back propagation algorithm to update the model parameters.

[0088] S43, model optimization: input sample data in all training graph datasets in sequence, repeat S41 and S42, and obtain the optimized model.

[0089] S44, Model verification: Input the verification graph dataset data into the optimized model to verify the versatility of the model.

[0090] S50, application of smooth surface effect prediction model: determine blasting environment data, mesh the target surface, set blasting hole parameters, and make model input data to predict smooth surface effect;

[0091] The light surface effect prediction model application step comprises:

[0092] S51. Determination of blasting environment data: Use 3D laser scanning to perform full coverage scanning of the target surface, and convert the pre-processed 3D data into a 3D geometric model of the target surface through 3D surface reconstruction technology;

[0093] S52, target surface meshing: meshing the target surface three-dimensional geometric model to obtain target surface mesh data;

[0094] S53, smooth surface effect prediction: set the blasting hole parameters, use S23 step to make model input data; use the smooth surface effect prediction model to predict the displacement data of the grid rock mass corresponding to the tunnel face of the node, and calculate the smooth surface effect data such as over-excavation and under-excavation, smooth surface area, etc. by combining the displacement data.

[0095] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes the above-mentioned face prediction method.

[0096] Based on the same technical solution, the present invention also discloses a computing device, comprising one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned prediction method.

[0097] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0101] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A tunnel smooth surface effect prediction method with adaptive blasthole parameters, characterized in that: The steps include: Based on graph neural network, a glossy surface effect prediction model is constructed; The rock mass displacement data is predicted using the smooth surface effect prediction model; The tunnel smooth surface effect is calculated based on the prediction results of rock mass displacement data.

2. The method according to claim 1, characterized in that The steps of building the glossy surface effect prediction model include: Construct a database of glossy blasting effect images including training data sets and validation data sets; Build a neural network model based on graph neural network; The neural network model is trained and optimized using the training data set and the validation data set to obtain a glossy surface effect prediction model.

3. The method according to claim 2, characterized in that The steps for constructing the smooth surface blasting effect map database including the training data set and the validation data set include: Mesh the three-dimensional geometric model of the tunnel face to obtain mesh data of the tunnel face; The blasthole three-dimensional position data, blasthole blasting depth data, blasthole blasting time series data, and blasthole blasting charge data are sequentially encoded and processed into vectors, and then the vectors are normalized and spliced ​​to obtain a blasthole parameter matrix; The grids in the face grid data are used as graph data nodes, and the faces between the grids are used as edges between the graph data nodes to generate an adjacency matrix; In the graph data node, the grid position data and the blasthole parameter matrix are stored as graph data samples, and the corresponding rock displacement data are stored as graph data labels; Obtain rock displacement data under different blasting hole parameters, obtain several different graph data samples and graph data labels, and stack them to form a graph structure data set, where the blasting hole parameters include blasting charge, blasting sequence, blasting hole position, and blasting hole depth for smooth blasting scenarios; Divide the graph structure data set, enter it into the library system, and complete the construction of the smooth blasting effect graph database.

4. The method according to claim 2, characterized in that: The modified finite element simulation model is used to obtain rock displacement data under smooth blasting scenarios with different blasting charges, blasting sequences, blasthole positions, and blasthole depths. The modified finite element simulation model acquisition steps include: The tunnel face is fully scanned by 3D laser scanning to obtain 3D data of the tunnel face. After preprocessing the 3D data, a 3D geometric model of the tunnel face is reconstructed. Three non-overlapping points are set as reference points in the tunnel face, and the distances between the blastholes and the reference points are measured in sequence to obtain the three-dimensional position data of the blastholes; Use a laser rangefinder to obtain blast hole depth data; The blast holes are numbered according to the set blasting time sequence information to obtain the blasting time sequence data of the blast holes; According to the set blasting equivalent of the blasthole, collect the blasting charge data of the blasthole; Perform finite element simulation on smooth blasting to obtain smooth blasting simulation data including at least rock mass displacement data and over-break and under-break volume; Conduct smooth blasting tests to obtain smooth blasting test data including at least rock displacement data and over-break and under-break data; The smooth surface blasting simulation data and smooth surface blasting test data are used as the smooth surface effect correction data. The finite element structure constitutive model is corrected using the finite element model correction method based on SGMD and LWOA-ELM to obtain the corrected finite element simulation model.

5. The method according to claim 1, characterized in that The neural network model includes a feature compression encoding layer, an asynchronous feature association layer, and a glossy effect decoding layer; In the feature compression coding layer, several stacked graph convolutional networks (GCNs) are used to compress the input data. Perform feature extraction to obtain compressed feature data , where each GCN is connected to a ReLU activation function, and n represents the number of blastholes; In the asynchronous feature association layer, The calculation results of each transformer encoder are input into a gated recurrent unit GRU, and the residual connection is used to connect Add the corresponding GRU output result to get the hidden state ;Will Together as The corresponding GRU input also uses residual connection to Add the corresponding GRU output result to get the hidden state ;Will and Together as The corresponding GRU input also uses residual connection to Add the corresponding GRU output result to get the hidden state ; and so on, to Together as The corresponding GRU input also uses residual connection to Add to the corresponding GRU output result to get the hidden state ; Glossy effect decoding layer: using stacked GCN to hide the state Perform expansion reduction to obtain rock mass displacement map data , where each GCN is followed by a ReLU activation function.

6. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device is caused to perform the method according to any one of claims 1 to 5.

7. An electronic device, characterized in that: The method comprises one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as described in any one of claims 1 to 5.

Images