Tunnel structure response prediction and protection optimization method based on mechanical response model

Abstract

The invention discloses a tunnel structure response prediction and protection optimization method based on a mechanical response model, and the method comprises the steps: carrying out the training and optimization of an explosion structure response neural network through an explosion structure mechanical response prediction database, and obtaining a tunnel explosion structure mechanical response model; based on a tunnel explosion structure mechanical response model, tunnel structure responses under different tunnel anti-explosion geometric protection means are obtained, and a geometric protection damage data set is constructed by calculating tunnel physical damage and functional damage; training and optimizing the geometric protection damage neural network by using the geometric protection damage data set to obtain a geometric protection damage model; and based on the geometric protection damage model, geometric protection means optimization is realized according to a damage minimization principle. The method is high in efficiency, strong in generalization ability and high in precision.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular to a tunnel structure response prediction and protection optimization method based on a mechanical response model. Background Art

[0002] Tunnel explosion refers to an explosion that occurs inside or around a tunnel. It is affected by the explosion equivalent, the distance between the explosive and the tunnel structure, and the tunnel depth. The shock wave generated by the explosion propagates at an extremely high speed, exerting instantaneous high pressure on the tunnel structure, causing the structure to respond dynamically.

[0003] Structural mechanics response refers to the changes in physical quantities such as displacement, velocity, acceleration, stress, and strain of the tunnel structure under the action of external loads (such as explosion shock waves), including the displacement, pressure, stress, strain, acceleration, velocity, etc. of the tunnel structure.

[0004] The digital intelligence model refers to a model that uses a combination of numerical simulation and intelligent algorithms to model and analyze the mechanical response of tunnel explosion structures. It overcomes the shortcomings of high experimental costs and high risks, solves the simulation problems of complex factors that are difficult to consider with theoretical analytical methods, and can more comprehensively and accurately predict the response of tunnel structures under explosions. Research on the digital intelligence model of the mechanical response of tunnel explosion structures can help design safer tunnel single buildings and protective structures under explosions to resist possible explosion threats and provide a scientific basis for engineering design. In addition, studying the transient response of structural mechanics under explosion-driven impact can better predict and evaluate the performance of tunnel structures in explosion events, thereby taking effective protective measures. It is the basis for evaluating the functional loss of the mechanical response of tunnel explosion structures and is of great significance.

[0005] Protection optimization is an application that optimizes geometric protection measures by designing a geometric protection damage neural network based on the mechanical response model of tunnel explosion structures. This application aims to optimize the geometric protection strategy in the mechanical response model of tunnel explosion structures.

[0006] At present, the research on the digital intelligence model of tunnel explosion structure mechanical response is as follows: "Mechanical response and toughness evolution characteristics of tunnel structure under additional load" (Liu Liying, Yang Chunshan, Yu Zhiyuan, etc., Modern Tunnel Technology, 2024, 61(5): 167-173, 233. DOI: 10.13807 / j.cnki.mtt. 2024.05.018.) discloses a numerical simulation analysis method for the mechanical response of a tunnel structure under an additional load, including: simulating the mechanical response of a tunnel structure under different working conditions by establishing a three-dimensional calculation model; analyzing the stress-deformation characteristics of the tunnel structure under each working condition by setting multiple calculation conditions, including initial ground stress analysis; extracting the internal force (such as bending moment) and displacement data of key parts such as the top plate, bottom plate, and side wall of the tunnel structure, analyzing their changing laws under the action of additional loads, and completing the internal force and displacement analysis; structural damage analysis: by calculating the maximum crack width of the tunnel structure, analyzing the relationship between the structural crack width and the additional load and the influence of structural damage on toughness, and evaluating the degree of damage to the structure; proposing a tunnel structure toughness evaluation method based on redundancy factor, calculating the redundancy factor and dividing the damage level, comprehensively considering indicators such as damage level, stability, crack width, redundancy, dividing different toughness levels, and completing toughness evaluation. However, the above methods rely on numerical simulation models that simplify the interactions and cannot accurately reflect the true mechanical behavior of materials under complex loads; the load conditions are not considered comprehensively, the additional load types are single, and the action time is not fully considered; the toughness evaluation mainly relies on the redundancy factor as an indicator, which can only partially reflect the bearing capacity of the structure after damage, and does not consider other aspects such as the deformation capacity, energy absorption capacity, and repair capacity of the structure, and cannot comprehensively evaluate the toughness of the tunnel structure; it is mainly based on numerical simulation analysis, lacks long-term monitoring data support for actual projects, and cannot verify the long-term accuracy of the model prediction results. It is also difficult to optimize and adjust the evaluation method based on feedback from actual projects, and the evaluation accuracy is poor.

[0007] For example, the paper "The influence of the shape of the suspended tunnel body on the structural response under underwater explosion" (Luo Gang, Yang Yunsheng, Yan Li, et al. Tunnel Construction, 2024, 44(6): 1233-1240. DOI: 10.3973 / j.issn.2096-4498.2024.06.010.) discloses a method for evaluating the structural response of a suspended tunnel body under underwater explosion, including: establishing a numerical model of the suspended tunnel by using LS-DYNA software; applying constraints and contact to simulate the interface between the water area and the pipe body by using the fluid-solid coupling algorithm; calculating the peak pressure of the underwater explosion shock wave by using the Cole empirical formula, and comparing it with the numerical calculation results to verify the reliability of the model; using the results of numerical simulation, further evaluating the explosion resistance of pipes with different shapes by comparing the dynamic responses of suspended tunnels with different cross-sectional shapes under underwater explosion loads, and providing a scientific basis for the cross-sectional optimization design of suspended tunnels.

[0008] However, the above methods rely on numerical simulation and cannot fully capture all dynamic behaviors in the actual explosion process, such as nonlinear behavior of materials, local damage of structural mechanics, etc. Moreover, they rely on simple self-compiled programs. When there are large differences in multi-scale structural mechanics responses, the model convergence is poor and the accuracy of the evaluation will decrease.

[0009] For example, the paper "Study on the Safety of Ellipticity of Metro Shield Tunnels Based on Refined Numerical Simulation" (Wang Yizhao, Chen Junsheng, Min Xing, et al. Railway Standard Design, 2024, 68(9): 140-148, 157. DOI: 10.13238 / j. issn.1004-2954.202212160003.) discloses a method for analyzing the structural safety status of shield tunnels using refined numerical simulation, including: by establishing a three-dimensional finite element model, using the actual monitored tunnel deformation data as the basis for numerical simulation, simulating the structural mechanical response and damage mechanical behavior of the shield tunnel under different load conditions. By defining the stress-strain curve and damage factor of concrete under uniaxial tension and compression conditions, the behavior of concrete under different stress states can be accurately simulated. By utilizing load-structure and considering various loads on the tunnel, including vertical earth pressure, lateral earth pressure, foundation reaction, and segment deadweight, the shield tunnel is analyzed and the load parameters are adjusted according to the actual situation to match the actual deformation. The simulation results under different ovality are compared, and the changing trends of indicators such as segment damage, bolt stress, and joint opening are analyzed to complete the structural safety analysis of the shield tunnel.

[0010] However, the above method uses a refined numerical simulation method and requires the establishment of a highly detailed three-dimensional finite element model, including the geometric details of the pipe segment, the contact relationship between the bolt and the hand hole, the spring constraint of the surrounding rock, etc. The model construction and calibration process is complicated, the resources and time required for calculation are greatly increased, the calculation cost is high, and the rapid evaluation and real-time monitoring of the safety status of the tunnel structure are not efficient enough. Although the accuracy of the model is verified by comparing with the actual monitoring data, the limited verification data affects the universality and reliability of the model. The complex situations in actual engineering may not be fully covered by limited verification data, and the generalization ability is not strong enough.

[0011] In summary, the problems existing in the existing technology are: the method for determining the mechanical response of the tunnel building explosion structure has weak generalization ability and low accuracy, and the structural mechanical response varies greatly at multiple scales, resulting in decreased model convergence and prediction accuracy. Summary of the invention

[0012] The purpose of the present invention is to provide a tunnel structure response prediction and protection optimization method based on a mechanical response model, which has high efficiency, strong generalization ability and high precision.

[0013] The technical solution to achieve the purpose of the present invention is:

[0014] Firstly, a tunnel structure response prediction method based on a mechanical response model is provided, including:

[0015] The explosion structure mechanical response prediction database is used to train and optimize the explosion structure response neural network to obtain the tunnel explosion structure mechanical response model;

[0016] Based on the mechanical response model of tunnel explosion structure, the tunnel structure response under different tunnel anti-blast geometric protection measures is obtained.

[0017] Furthermore, the explosion structure response neural network includes an encoder, a processor and a decoder:

[0018] The encoder is composed of multiple single encoders connected in parallel. In each single encoder, a node encoder and an edge encoder for encoding node features and encoding edge features are respectively constructed based on the first feedforward neural network. A hierarchical coding strategy is used to firstly perform edge feature coding, and then perform node feature coding with the help of edge feature coding, so as to realize the combination of node encoder and edge encoder to form a single encoder.

[0019] In the processor, a node feature processor and an edge feature processor for processing node features and edge features are respectively constructed based on a multi-level star-shaped neural network. A hierarchical processing strategy is used to first perform edge feature processing, and then perform node feature processing with the help of edge feature processing, so as to realize the combination of the node feature processor and the edge feature processor to form a processor; wherein the multi-level star-shaped neural network includes a plurality of feedforward neural network parallel networks, each feedforward neural network parallel network is formed by weighted fusion of the second and third feedforward neural networks, and at the same time, a weight is given to each feedforward neural network parallel network through the fourth feedforward neural network, and a plurality of feedforward neural network parallel networks are weighted fused to form a multi-level star-shaped neural network;

[0020] The decoder is composed of multiple fifth feedforward neural networks connected in series.

[0021] Furthermore, the steps of constructing the explosion structure mechanical response prediction database include:

[0022] Based on the digital twin of the real tunnel area and combined with the design specifications of tunnel geometric protection means, the type, center position, and geometric parameter information of the tunnel anti-blast geometric protection means are obtained to construct a tunnel area dataset;

[0023] Construct a tunnel area neural network, substitute the tunnel area data set into the tunnel area neural network, perform data training and optimization, and construct a tunnel area prediction model;

[0024] Design a simulated tunnel area prediction data set, combine it with the tunnel area prediction model, and obtain the simulated tunnel area point cloud data;

[0025] By simulating the point cloud data of the tunnel area and using the point cloud reconstruction method, the spatial entity of the tunnel area is obtained;

[0026] The explosion center and explosion equivalent are randomly designed to obtain the explosion information of each tunnel area case. The tunnel area point cloud of the same tunnel area case has the same explosion information. The tunnel area explosion event is constructed by combining the simulated tunnel area point cloud data and the tunnel area space entity.

[0027] Using the meshing method, the mesh nodes of the tunnel area are obtained. Using the numerical simulation method, combined with the explosion event in the tunnel area, the working condition design before the explosion is combined with the mechanical response of the tunnel structure according to the explosion event in the tunnel area and the node number of the tunnel area, and the explosion simulation data of the mechanical response of the tunnel structure is obtained.

[0028] Obtain historical and experimental tunnel area prediction data sets, combine with tunnel area prediction models, obtain historical and experimental tunnel area point cloud data, and then obtain tunnel structure mechanical response explosion history and experimental data;

[0029] Combining the explosion simulation data of tunnel structure mechanical response with historical and experimental data, an explosion structure mechanical response database is established through data conversion and data combination.

[0030] Furthermore, the node data features in the tunnel area dataset are [tunnel area number, area area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there are tunnel anti-explosion geometric protection measures, tunnel anti-explosion geometric protection measures type, tunnel area node number, tunnel area node X-axis position, tunnel area node Y-axis position, tunnel area node Z-axis position].

[0031] Furthermore, the tunnel area neural network includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a third convolutional layer, a second pooling layer, a fully connected layer and an output layer, which are connected in sequence, and the output layer includes three fully connected layers connected in series.

[0032] Furthermore, the steps of designing the simulation tunnel area prediction data set include:

[0033] Through the random number method, combined with the tunnel design specifications, the random area area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there is tunnel anti-explosion geometric protection means, and tunnel anti-explosion geometric protection means type are designed;

[0034] According to the length, width, height and area of ​​the tunnel, obtain the range of the number of nodes in the tunnel area;

[0035] Obtain the number of nodes in the tunnel area through the random number method;

[0036] Obtain the tunnel area node number, combine data according to the tunnel area node number, obtain tunnel area cases, combine tunnel area cases according to the tunnel area number, and obtain a simulation tunnel area prediction data set.

[0037] Furthermore, the node data characteristics of the explosion structure mechanics response database are [tunnel number, whether the tunnel has anti-explosion geometric protection measures, tunnel node number, tunnel node X-axis position, tunnel node Y-axis position, tunnel node Z-axis position, explosion center X-axis position, explosion center Y-axis position, explosion center Z-axis position, explosion equivalent, tunnel node stress, tunnel node strain, tunnel node displacement].

[0038] Furthermore, the steps of constructing the explosion structure mechanical response prediction database also include:

[0039] According to the order of magnitude of the tunnel structure mechanical unit response, the damage level of the tunnel node is obtained, and the obtained tunnel node damage level is used as the newly added node data feature of the explosion structure mechanical response prediction database.

[0040] Secondly, a protection optimization method based on a mechanical response model is also provided, including:

[0041] Based on the tunnel structure response prediction method described above, the tunnel structure response under different tunnel anti-blast geometric protection measures is obtained, and the geometric protection damage data set is constructed by calculating the tunnel physical damage and functional damage;

[0042] The geometric protection damage data set is used to train and optimize the geometric protection damage neural network to obtain the geometric protection damage model;

[0043] Based on the geometric protection damage model, the optimization of geometric protection measures is achieved according to the principle of minimum damage.

[0044] Furthermore, the node data features of the geometric protection damage dataset are [tunnel number, presence or absence of tunnel anti-blast geometric protection means, type of tunnel anti-blast geometric protection means, length of tunnel anti-blast geometric protection means, width of tunnel anti-blast geometric protection means, height of tunnel anti-blast geometric protection means, geometric protection distance, distance from explosion center, explosion equivalent, functional damage].

[0045] Furthermore, the geometric protection damage neural network is composed of a sixth feedforward neural network, and the sixth feedforward neural network is composed of 1 input layer, 3 hidden layers, and 1 output layer.

[0046] Compared with the prior art, the present invention has the following significant advantages:

[0047] 1. Stronger generalization ability: The present invention establishes a tunnel area prediction model and constructs a tunnel explosion structure response database with tunnel area information, thereby ensuring the diversity of the database and the robustness of the neural network. On the other hand, a geometric protection means optimization method is designed to enhance the engineering application ability of the method proposed in this paper and improve the generalization ability of the structural mechanical response and its application under the adaptive explosion condition.

[0048] 2. Higher accuracy: The present invention establishes a tunnel area prediction model, constructs a tunnel explosion structure response database with tunnel area information, considers the influence of tunnel environment on the prediction of explosion structure response, designs a parallel encoder and a multi-level star-shaped neural network processor, and combines the solid mechanics physical information embedding method to construct a tunnel explosion structure response neural network. It can not only ensure the model's learning ability for the physical problems of the mechanical response of the explosion structure of a single building in the tunnel, but also ensure the model's learning ability for three-dimensional non-structural mechanics data, thereby achieving high convergence of the three-dimensional model and high-accuracy prediction of the mechanical response of the tunnel explosion structure.

[0049] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is the main flow chart of the protection optimization method based on the mechanical response model of the present invention.

[0051] Figure 2 yes Figure 1 Flowchart of the steps for building the prediction model for the tunnel area.

[0052] Figure 3 This is a schematic diagram of the neural network architecture in the tunnel area.

[0053] Figure 4 yes Figure 1 Flowchart of the steps for establishing a database for prediction of mechanical responses of explosion structures.

[0054] Figure 5 yes Figure 1 Flowchart of the steps for constructing the structural mechanical response model of the tunnel explosion.

[0055] Figure 6 This is the architecture diagram of the explosion structure response neural network encoder.

[0056] Figure 7 This is the architecture diagram of the explosion structure response neural network processor.

[0057] Figure 8 This is the architecture diagram of the explosion structure response neural network decoder.

[0058] Fig. 9 yes Figure 1 Flowchart of the steps for optimizing the geometric protection measures. DETAILED DESCRIPTION

[0059] like Figure 1 As shown in FIG. 1 , the steps of the protection optimization method based on the mechanical response model include:

[0060] S10. Construction of tunnel area prediction model: Through the digital twin method, the digital twin of the real tunnel and its surrounding environment is obtained. Through the analysis of tunnel type and regional building configuration combined with the existing tunnel geometric protection means design specifications, a tunnel area dataset is constructed. Through the three-dimensional convolutional neural network, a tunnel area neural network is constructed. The tunnel area dataset is substituted into the tunnel area neural network for data training and optimization to construct a tunnel area prediction model.

[0061] like Figure 2 As shown, the tunnel area prediction model building steps include:

[0062] S11. Construction of building area dataset: Using the digital twin method and combining it with geographic information data sources, we can obtain the digital twin of the real tunnel area. Using image recognition methods, we can analyze the tunnel type and regional building configuration. Combined with the existing tunnel geometric protection means design specifications, we can obtain the type, center position, and geometric parameter information of the tunnel anti-explosion geometric protection means. Through point cloud data preprocessing, we can construct the building area dataset.

[0063] 1. Digital twin of tunnel area: The obtained digital twin of tunnel area is represented in three-dimensional formats such as stl and vtk containing node information and surface information. The position information contained in the node information represents the node position information of tunnel area. The node position information is composed of the X-axis, Y-axis and Z-axis positions in the world coordinate system with the center of the area as the origin.

[0064] 2. Tunnel area data set: Node data features of the tunnel area data set [tunnel area number, area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there are tunnel anti-explosion geometric protection measures, tunnel anti-explosion geometric protection measures type, tunnel area node number, tunnel area node X-axis position, tunnel area node Y-axis position, tunnel area node Z-axis position], where tunnel area number, area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there are tunnel anti-explosion geometric protection measures, tunnel anti-explosion geometric protection measures type, tunnel area node number are the input data features of the tunnel area data set, and the number of input data features of the tunnel area data set is 12; tunnel area node X-axis position, tunnel area node Y-axis position, tunnel area node Z-axis position are the output data features of the tunnel area data set, and the number of output data features of the tunnel area data set is 3. The tunnel area number and the tunnel area node number respectively represent the numbers of the tunnel area and the tunnel area indicating the source of this data located in the tunnel area; the unit of the area is square kilometer; different numbers in the environment type represent different types, such as 0 for mountain, 1 for underwater, and 2 for city; the entire area takes the center of the area as the origin, obtains the X-axis, Y-axis, and Z-axis according to the world coordinate system, and thereby obtains the tunnel center position and area node position information; the unit of tunnel length, width, and height is meter; when the value of whether there is a tunnel anti-explosion geometric protection means is 0, it means that there is no tunnel anti-explosion geometric protection means at this place, and when the value of whether there is a tunnel anti-explosion geometric protection means is 1, it means that there is a tunnel anti-explosion geometric protection means at this place; different numbers in the tunnel anti-explosion geometric protection means type represent different types, such as 0 for explosion-proof wall and 1 for protection net.

[0065] For example, [1, 10, 0, 0, 0, 5000, 7.5, 5, 1, 1, 50010, 10, 20, 10] means that the tunnel area is numbered 1. In the area of ​​10 square kilometers, the environment type is mountainous. With the center of the area as the origin, in the area under the world coordinate system, the tunnel is located at the center of the area (0m, 0m, 0m). The length, width and height of the tunnel are 5000m, 7.5m and 5m respectively. The tunnel has protection measures, which are protective nets. The tunnel space node is numbered 50010, and the space node is located at (10m, 20m, 10m).

[0066] 3. Point cloud information of tunnel area building types and configurations: The data is combined according to the regional space nodes. Each piece of data represents a node. During training, a single tunnel area space case consisting of a tunnel area space node group containing millions of nodes is used as the data unit for training. The single tunnel area space consisting of millions of nodes contains the tunnel area building types and configurations.

[0067] 4. Point cloud data preprocessing: Through data point cloud format conversion, the point cloud data is converted from one format to a format that is convenient for subsequent processing and analysis. Using the point cloud standardization method, the point cloud data is converted into a relatively unified coordinate system or distribution to complete data preprocessing.

[0068] S12. Neural network construction: By constructing the input layer, alternately constructing multiple convolutional layers and pooling layers, constructing the fully connected layer and the output layer, and constructing a three-dimensional convolutional neural network, the tunnel area neural network construction is realized. The schematic diagram of the tunnel area neural network architecture is shown in the figure. Figure 3 shown.

[0069] The neural network construction steps include:

[0070] S121. Input layer construction: The input layer consists of a simple data receiving layer, which does not perform any complex calculations or transformations and is only used to receive and organize input data features.

[0071] The specific construction method of the input layer is as follows, assuming that the data set is The input is in the form of is the batch size, is the number of channels (obtained from the node data features of the tunnel area dataset), It is the three-dimensional spatial dimension of the point cloud data and constructs the input layer.

[0072] S122, alternately constructing multiple convolutional layers and pooling layers: by constructing convolutional layer 1, pooling layer 1, convolutional layer 2, convolutional layer 3, and pooling layer 2, alternately constructing multiple convolutional layers and pooling layers;

[0073] The step of alternately constructing the multi-layer convolutional layer and the pooling layer in S122 includes:

[0074] S1221, Convolutional layer 1 construction: Convolutional layer 1 contains 32 3D convolution kernels with a size of , the step size is 1, the padding is 1, the spatial dimension remains unchanged, and the ReLU activation function is used for activation to complete the construction of convolutional layer 1.

[0075] S1222, pooling layer 1 construction: using the maximum pooling strategy, the size of pooling layer 1 is , with a step size of 2, to realize the spatial dimension of the data and complete the construction of pooling layer 1.

[0076] S1223, Convolutional layer 2 construction: Convolutional layer 2 contains 64 3D convolution kernels with a size of , the step size is 1, the padding is 1, the activation function is ReLU, and the construction of convolutional layer 2 is completed.

[0077] S1224, Convolutional layer 3 construction: Convolutional layer 3 contains 128 3D convolution kernels with a size of , the step size is 1, the padding is 1, the activation function is ReLU, and the construction of convolutional layer 3 is completed.

[0078] S1225, pooling layer 2 construction: By adopting the maximum pooling strategy, the size of pooling layer 2 is , with a step size of 2, further reducing the spatial dimension of the data and completing the construction of pooling layer 2.

[0079] S123, fully connected layer construction: Flatten the pooled data and connect them in series through multiple identical fully connected layers. The activation function is ReLU to complete the fully connected layer construction.

[0080] S124, output layer construction: The output layer includes three fully connected layers in series, which process the physical information of the buildings around the three-dimensional tunnel, realize point cloud data prediction, and complete the output layer construction.

[0081] S125. Neural network construction: Through multiple layers of convolutional layers, pooling layers, and fully connected layers, a three-dimensional convolutional neural network architecture is implemented to complete the neural network construction.

[0082] S13, prediction model construction: the constructed building area data set is combined with the S11 building area data set and the S12 neural network is constructed to construct a tunnel area neural network, and a tunnel area prediction model is obtained through training and optimization.

[0083] 1. Training and optimization: The mean square error is used as the loss function to realize data training. By using the Adams optimizer, the model parameters are updated, and the model is trained in a supervised manner to complete the iterative optimization of the neural network data and realize training and optimization; Loss function: By using the mean square error as the loss function, the difference between the predicted point cloud and the real point cloud is obtained; Optimizer: Using the Adam optimizer, the adaptive mechanism is used to adjust the learning rate, realize the complexity processing of point cloud data, and complete data optimization.

[0084] S20. Establishment of a prediction database for the mechanical response of explosion structures: Design a tunnel area prediction data set, combine it with a tunnel area prediction model, obtain the tunnel area point cloud data, use point cloud reconstruction to obtain the tunnel area spatial entity, design random explosion information, construct tunnel explosion events, use numerical simulation methods to obtain tunnel structure mechanical response simulation data, and obtain an explosion structure mechanical response prediction database through data processing of historical and experimental data, and through conversion and combination of historical and experimental data and simulation data.

[0085] like Figure 4 As shown, the steps of establishing the explosion structure mechanical response prediction database include:

[0086] S21. Acquisition of simulation data of mechanical response of tunnel structure: By designing a prediction data set for the simulated tunnel area and combining it with the prediction model for the tunnel area, point cloud data of the simulated tunnel area is acquired. The spatial entity construction of the tunnel area is acquired using the data reconstruction method. The random number method is used to acquire random explosion information and construct explosion events. The numerical simulation method is used to acquire tunnel explosion simulation data.

[0087] The step of acquiring the tunnel area structural mechanical response simulation data comprises:

[0088] S211. Design of simulation tunnel area prediction data set: through random number method combined with tunnel design specifications, design random area area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there are tunnel anti-explosion geometric protection measures, tunnel anti-explosion geometric protection measures type and other input data features, and according to the tunnel length, width, height and area, obtain the range of tunnel area node quantity, obtain the number of tunnel area nodes through random number method, and further obtain the tunnel area node number, combine data according to the tunnel area node number, obtain tunnel area cases, combine tunnel area cases according to the tunnel area number, and obtain the tunnel area prediction data set.

[0089] S212, acquiring point cloud data of simulated tunnel area: acquiring the X-axis, Y-axis, and Z-axis position information of each tunnel area node under each tunnel area case through the tunnel area prediction data set and in combination with the tunnel area prediction model, and obtaining the tunnel area point cloud data.

[0090] S213, tunnel area spatial entity construction: obtain the tunnel area spatial entity through the tunnel area point cloud data and the point cloud reconstruction method.

[0091] S214. Construction of tunnel area explosion events: By randomly designing the explosion center and explosion equivalent, the explosion information of each tunnel area case is obtained. The tunnel area point cloud of the same tunnel area case has the same explosion information. The tunnel area explosion event is constructed by combining the tunnel area point cloud data and the tunnel area spatial entity.

[0092] S215, acquisition of tunnel structure mechanical response simulation data: using a meshing method to obtain mesh nodes in the tunnel area, using a numerical simulation method, combined with the explosion event in the tunnel area, the working condition design before the explosion is combined with the mechanical response of the tunnel structure according to the explosion event in the tunnel area and the node number in the tunnel area, to obtain the explosion simulation data of the mechanical response of the tunnel structure;

[0093] The tunnel nodes in the simulation data are obtained by judging whether the mesh nodes in the tunnel area belong to the tunnel. The tunnel node position information refers to the X-axis, Y-axis, and Z-axis position information of the tunnel node in the world coordinate system with the center of the area as the origin.

[0094] S22. Data processing of historical and experimental data: Collect historical and experimental data, obtain historical and experimental tunnel area prediction data sets, combine with tunnel area prediction models, obtain historical and experimental tunnel area point cloud data, delete data outside the area that does not conform to geometric laws through data screening, and supplement missing data through data reconstruction based on historical and tunnel node location information to complete historical and experimental data processing.

[0095] The historical and experimental data processing steps include:

[0096] S221. Data collection: Collect test data on mechanical response of tunnel building explosion structure by designing experiments and consulting literature, and collect historical data on mechanical response of tunnel building explosion structure by consulting literature, databases and accident investigation reports;

[0097] The data collected include explosion conditions and structural mechanical responses of historical and test data, among which the tunnel area number, area, environment type, X-axis position of tunnel center, Y-axis position of tunnel center, Z-axis position of tunnel center, tunnel length, tunnel width, tunnel height, whether there are tunnel anti-explosion geometric protection measures, type of tunnel anti-explosion geometric protection measures, and number of tunnel nodes are explosion condition data.

[0098] S222. Acquisition of historical and test tunnel area point cloud data: Design tunnel node numbering according to the number of tunnel nodes collected in the explosion conditions, combine data according to the tunnel area number and tunnel node number, obtain historical and test area prediction data sets, and combine historical and test prediction models to obtain historical and test tunnel area point cloud data.

[0099] The node data features of the historical and experimental area prediction data sets include [tunnel area number, area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there is a tunnel anti-explosion geometric protection method, tunnel anti-explosion geometric protection method type, tunnel node number]. The predicted tunnel area point cloud data consists of tunnel area node location information. The tunnel node location information is obtained by judging whether the tunnel area node belongs to the tunnel.

[0100] S222. Data screening: By cleaning the historical and test explosion data, the erroneous, incomplete or inconsistent data in the data are deleted, and the explosion structure mechanical response data of the tunnel building that meets the requirements are extracted.

[0101] S223, data reconstruction: interpolate and supplement the data deleted during data screening to obtain correct data.

[0102] S23. Database construction: Combine historical and experimental explosion data with numerical simulation data to obtain data sources. Through data conversion and data combination, multiple sets of explosion impact input data characteristics and structural mechanics response data are obtained to establish an explosion structural mechanics response database.

[0103] 1. The node data features of the explosion structure mechanics response database include [tunnel number, whether the tunnel has anti-explosion geometric protection measures, tunnel node number, tunnel node X-axis position, tunnel node Y-axis position, tunnel node Z-axis position, explosion center X-axis position, explosion center Y-axis position, explosion center Z-axis position, explosion equivalent, tunnel node stress, tunnel node strain, tunnel node displacement], among which, tunnel number, whether the tunnel has anti-explosion geometric protection measures, tunnel node number, tunnel node X-axis position, tunnel node Y-axis position, tunnel node Z-axis position, explosion center X-axis position, explosion center Y-axis position, explosion center Z-axis position, explosion equivalent are the input data features of the explosion structure mechanics response database, and the number of input data features of the explosion structure mechanics response database is 10, and tunnel node stress, tunnel node strain, and tunnel node displacement are structural mechanics response data. Among them, the tunnel number, whether the tunnel has anti-explosion geometric protection measures, and the tunnel node number are derived from the historical and experimental tunnel area prediction data sets and the simulation tunnel area prediction data sets. The tunnel node X-axis position, tunnel node Y-axis position, and tunnel node Z-axis position are obtained from the tunnel node position information. The explosion center X-axis position, explosion center Y-axis position, and explosion center Z-axis position are derived from the explosion center position. The explosion equivalent and the explosion center position constitute the explosion information. The tunnel node stress, tunnel node strain, and tunnel node displacement belong to the structural mechanics response data.

[0104] For example, [0, 0, 1, 0.2, 0.03, 0.2, 5000, 20, 10, 200, 100, 8, 5] means that the data belongs to the first node of the 0th tunnel. The tunnel does not have anti-explosion geometric protection measures. The node is located at (0.2m, 0.03m, 0.2m). The center of the explosion in this explosion event is located at (5000m, 20m, 10m). The explosion equivalent is 200kg. The stress of this node in this explosion event is 100Mpa, the strain is 8, and the displacement is 5mm.

[0105] 2. Data conversion and data combination: Format and standardize the data, and classify and combine the standardized data according to the tunnel numbers and tunnel node numbers to achieve data conversion and combination.

[0106] S30. Construction of mechanical response model of tunnel explosion structure: The tunnel node damage is obtained through damage judgment of tunnel node unit response. The explosion structure response neural network is constructed by designing encoders, processors, and decoders, combining solid mechanics physical information to embed equations. Combined with the explosion structure mechanical response prediction database data, the mechanical response model of the tunnel explosion structure is obtained through data training and optimization.

[0107] like Figure 5 As shown, the step of constructing the tunnel explosion structure mechanical response model in S30 includes:

[0108] S31. Tunnel node damage acquisition: According to the order of magnitude of the tunnel building structure mechanics unit response, the tunnel node damage level is acquired, and the acquired tunnel node damage level is used as the node data feature newly added to the explosion structure mechanics response database to realize tunnel node damage acquisition.

[0109] 1. The order of magnitude of structural mechanics unit response includes: slight damage, the deformation is usually at the millimeter level or even smaller, and the stress is within the lower proportion of the allowable stress of the material; moderate damage, for metal structural mechanics, it may reach the order of several millimeters to more than ten millimeters, and concrete structural mechanics may have cracks, the crack width is 0.5 mm to 2 mm or even larger, and the stress is about 30%-60% of the allowable stress of the material; severe damage, the deformation of metal structural mechanics may be tens of millimeters or even larger, and concrete structural mechanics may have large-scale cracks and crushing, and the stress may reach more than 80% of its yield strength or even completely yield; complete damage, the structural mechanics unit completely loses its original shape and structure, and serious distortion, fracture, collapse, etc. occur, and the deformation can no longer be measured by conventional size concepts. The stress far exceeds the ultimate strength of the material, the material is completely destroyed, and can no longer carry any load. The stress state of the structural mechanics unit is no longer of practical significance because the structure has completely failed.

[0110] 2. Damage level of tunnel nodes: The damage levels of tunnel nodes include slight damage, moderate damage, severe damage, and complete damage, which are represented by 0, 1, 2, and 3 respectively in the data set.

[0111] 3. The node data features of the data set after adding tunnel node damage are: [tunnel number, whether the tunnel has anti-explosion geometric protection measures, tunnel node number, tunnel node X-axis position, tunnel node Y-axis position, tunnel node Z-axis position, explosion center X-axis position, explosion center Y-axis position, explosion center Z-axis position, explosion equivalent, tunnel node stress, tunnel node strain, tunnel node displacement, tunnel node damage].

[0112] S32. Construction of explosion structure response neural network: With the help of feedforward neural network, encoder, processor and decoder are designed, and the explosion structure response neural network is constructed by combining the physical information of solid mechanics and embedding equations.

[0113] S321, encoder design: encoder design as follows Figure 6 As shown, a feedforward neural network 1 is designed, and a node encoder and an edge encoder are constructed for encoding node features and encoding edge features respectively. The node encoder and the edge encoder are combined to generate a single encoder. With the help of node damage, multiple single encoders are connected in parallel to realize the design of a neural network encoder.

[0114] 1. Feedforward neural network 1: Feedforward neural network 1 consists of 1 input layer, 1 hidden layer and 1 output layer. The number of neurons in the input layer is obtained by the number of node feature inputs of the node encoder and the number of edge feature inputs of the edge encoder. The number of neurons in the hidden layer is the same and is designed to be a multiple of 2 greater than 64. The number of neurons in the output layer is designed to be the same as the number of neurons in the input layer.

[0115] The number of node feature inputs refers to the number of input data features used as nodes in the number of node features, and the number of edge feature inputs refers to the number of input data features used as edge features in the number of edge features. The input data features of edge features are obtained based on the input data features of aggregated neighboring nodes.

[0116] 2. Edge features: Edge features are obtained through the neighbor node feature aggregation strategy. The neighbor node feature aggregation strategy is a method of generating edge features using node features. By combining point features and edge features, graph data is obtained.

[0117] 3. Single encoder: Combined with the hierarchical coding strategy, edge feature encoding is performed first, and then point feature encoding is performed with the help of edge feature encoding to realize the combination of node encoder and edge encoder to construct a single encoder.

[0118] 4. Single encoder parallel connection: Node damage is divided into four levels: slight damage, moderate damage, severe damage, and complete damage. By connecting four single encoders in parallel, the data of four different node damages are trained and predicted respectively to realize single encoder parallel connection.

[0119] S322, Processor design: Processor design such as Figure 7 As shown, by connecting feedforward neural network 2 and feedforward neural network 3 in parallel, feedforward neural network parallel network 1 is constructed, and feedforward neural network parallel network 1, feedforward neural network parallel network 2, feedforward neural network parallel network 3 and feedforward neural network 4 are connected in parallel to construct a multi-level star-shaped neural network. Point feature processors and edge feature processors are constructed by the multi-level star-shaped neural network to realize processor design.

[0120] 1. Feedforward neural network 2: Feedforward neural network 2 consists of 1 input layer, 1 hidden layer and 1 output layer. The number of neurons in the input layer of feedforward neural network 2 corresponding to the point feature processor is obtained by the input number, the number of neurons in the input layer of feedforward neural network 2 corresponding to the edge feature processor is obtained by the edge feature input number, the number of neurons in the hidden layer is obtained by 256 or more multiples of 2, and the number of neurons in the output layer and the hidden layer are the same.

[0121] 2. Feedforward neural network 3, feedforward neural network 4: The architecture of feedforward neural network 3 and feedforward neural network 4 is consistent with feedforward neural network 2.

[0122] 3. Feedforward neural network parallel network 1: Feedforward neural network 2 and feedforward neural network 3 are connected in parallel, and a weighted fusion strategy is adopted to obtain high-dimensional graph data of feedforward neural network parallel network 1, and construct feedforward neural network parallel network 1.

[0123] 4. Feedforward neural network parallel network 2, feedforward neural network parallel network 3: The neural network architecture of feedforward neural network parallel network 2 and feedforward neural network parallel network 3 is consistent with feedforward neural network parallel network 1.

[0124] 5. Multi-level star-shaped neural network: Feedforward neural network parallel network 1, feedforward neural network parallel network 2, feedforward neural network parallel network 3 and feedforward neural network 4 are connected in parallel, and a multi-level star-shaped neural network is constructed using a data weighted fusion strategy.

[0125] 6. Processor: Based on the number of point feature inputs and edge feature inputs, a point feature processor and an edge feature processor are constructed in combination with a multi-level star-shaped neural network. A hierarchical processing strategy is used to first perform edge feature processing, and then point feature processing is performed with the help of edge feature processing to construct a processor.

[0126] S323, decoder design: decoder design as follows Figure 8 As shown, multiple feedforward neural networks 5 are connected in series to restore node feature data and construct a decoder.

[0127] 1. Feedforward neural network 5: Feedforward neural network 1 is constructed by 1 input layer, 3 hidden layers and 1 output layer. The number of neurons in the input layer is obtained by the number of neurons in the output layers of feedforward neural network 2, feedforward neural network 3 and feedforward neural network 4 in point feature processing. The number of neurons in the hidden layer is a multiple of 2 greater than 256. The number of neurons in the output layer is obtained by the number of node feature outputs; the number of node feature outputs refers to the number of output features in the number of node features that serve as node data, the number of edge feature outputs refers to the number of output features in the number of edge features that serve as edge data, and the output features of the edge data are obtained based on the data output data features of the aggregated neighboring nodes.

[0128] 2. Decoder: The purpose of the decoder is to restore high-dimensional graph data to node data features.

[0129] S324, physical information embedding: by constructing the solid mechanics physical information equation, constructing the physical information constraint equation, using the MSE method to obtain the data constraint term, adding the two together to calculate the data loss, and realizing physical information embedding;

[0130] S325, neural network construction: construct a neural network by connecting an encoder, a processor, and a decoder in series and embedding physical information in the loss term;

[0131] S33. Acquisition of tunnel explosion structure mechanical response model: Utilizing the data in the tunnel explosion structure mechanical response database, combined with the explosion structure response neural network, with the help of the Adam algorithm and supervised training, and with the help of hyperparameter tuning strategy optimization, the tunnel explosion structure mechanical response model is obtained.

[0132] S331, model training: with the help of the tunnel explosion structure mechanical response database, input data into the explosion structure response neural network, use the Adam algorithm to update the model parameters, and train the model in a supervised manner.

[0133] The model parameters are randomly generated at the beginning of training. The back propagation algorithm is needed to backpropagate the value of the loss function and calculate the gradient of the model parameters relative to the loss value to update the model parameters. The selected back propagation algorithm is the Adam algorithm, which combines the characteristics of gradient descent and momentum optimization and has good performance and convergence speed.

[0134] S332. Model construction: By tuning the hyperparameters of the explosion structure response neural network and optimizing the model, a tunnel explosion structure mechanical response model is obtained.

[0135] When performing hyperparameter tuning, you need to pay attention to a series of key hyperparameters, including the number of iterations, batch size, learning rate, and adjustment of model structural mechanics. The key points of adjustment are as follows:

[0136] 1. The number of iterations is between 50 and 300 epochs;

[0137] 2. The batch size can be adjusted between 2 and 128;

[0138] 3. The learning rate is usually adjusted between 0.0001 and 0.01;

[0139] 4. Adjustment of the model’s structural mechanics is also a key step in optimization. According to the gradient calculation formula, the design of the structural mechanics affects the gradient update direction. Therefore, the model’s structural mechanics needs to be modified based on the model’s prediction results. This includes adjusting the number of layers in the model, the number of units in each layer, and selecting the appropriate graph network layer configuration.

[0140] S40, Geometric protection means optimization: Design node data features of different anti-explosion geometric protection means in the tunnel area, combine with the tunnel area prediction model, reconstruct through point cloud, design random explosion information, combine with tunnel explosion structure mechanical response model, obtain tunnel structure response under different anti-explosion geometric protection means, calculate tunnel physical damage and functional damage, construct geometric protection damage data set, design feedforward neural network 2, construct geometric protection damage neural network, construct geometric protection damage model through training and optimization, and realize geometric protection means optimization according to the principle of minimum damage;

[0141] like Fig. 9 As shown, the geometric protection means preferably comprises the following steps:

[0142] S41. Acquisition of tunnel explosion structure response: According to the number of tunnel anti-blast geometric protection means corresponding to the tunnel, the node data characteristics of different tunnel anti-blast geometric protection means in the tunnel area are designed. Combined with the tunnel area prediction model, the tunnel area point cloud data of the same tunnel under different tunnel anti-blast geometric protection means are obtained. Through point cloud reconstruction, random explosion information is designed, and the tunnel explosion structure mechanical response model is used to obtain the tunnel explosion structure response of the same tunnel under different tunnel anti-blast geometric protection means.

[0143] S42. Construction of geometric protection damage data set: According to the structural response of the tunnel explosion, the functional damage of the tunnel is obtained through the existing building physical damage and functional damage calculation formula. Through image recognition, the parameters of the tunnel anti-blast geometric protection measures are obtained to construct a geometric protection damage data set.

[0144] 1. Parameters of geometric protection means: The parameters of geometric protection means include the size of the tunnel anti-explosion geometric protection means and the geometric protection distance is obtained through the length, width and height of the tunnel anti-explosion geometric protection means. The geometric protection distance is the distance between the center of the tunnel anti-explosion geometric protection means and the nearest tunnel surface.

[0145] 2. Node data features of the geometric protection and damage data set: [tunnel number, whether there is a tunnel anti-blast geometric protection means, type of tunnel anti-blast geometric protection means, length of tunnel anti-blast geometric protection means, width of tunnel anti-blast geometric protection means, height of tunnel anti-blast geometric protection means, geometric protection distance, distance from explosion center, explosion equivalent, functional damage], among which, tunnel number, whether there is a tunnel anti-blast geometric protection means, type of tunnel anti-blast geometric protection means, length of tunnel anti-blast geometric protection means, width of tunnel anti-blast geometric protection means, height of tunnel anti-blast geometric protection means, geometric protection distance, distance from explosion center, explosion equivalent are input data, and functional damage is output data; distance from explosion center is the distance between the explosion center and the center of the tunnel anti-blast geometric protection means.

[0146] For example, [0, 1, 0, 250000, 1500, 2000, 5000, 5000, 200, 0.1] means a tunnel numbered 0, with a length, width and height of 250000mm, 1500mm and 2000mm respectively, a protective wall with a protection distance of 5000mm, a blast center distance of 5000mm and an explosion equivalent of 200kg, and the functional damage is 0.1.

[0147] S43. Construction of geometric protection damage neural network: By designing the number of hidden layers, the number of neurons in the input layer, hidden layer, and output layer, a feedforward neural network 6 is constructed to construct a geometric protection damage neural network.

[0148] Feedforward neural network 6: The feedforward neural network 2 consists of 1 input layer, 3 hidden layers, and 1 output layer. The number of neurons in the input layer is obtained by the number of input data features in the geometric protection damage dataset, the number of neurons in the hidden layer is obtained according to a multiple of 2 greater than 256, and the number of neurons in the output layer is obtained by the number of output data features in the geometric protection damage dataset.

[0149] S44. Acquisition of geometric protection damage model: Substitute the geometric protection damage data set into the geometric protection damage neural network, and construct the geometric protection damage model through training and optimization.

[0150] S45. Optimization of geometric protection means: obtain the input data features of the geometric protection damage data set of the optimization evaluation case of the geometric protection means, obtain the functional damage of the optimization evaluation case of the geometric protection means with the help of the geometric protection damage model, select the tunnel anti-blast protection means with the smallest functional damage as the optimal geometric protection means, and realize the optimization of geometric protection means.

[0151] 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 structure response prediction method based on a mechanical response model, characterized in that: include: The explosion structure mechanical response prediction database is used to train and optimize the explosion structure response neural network to obtain the tunnel explosion structure mechanical response model; Based on the mechanical response model of tunnel explosion structure, the tunnel structure response under different tunnel anti-blast geometric protection measures is obtained.

2. The tunnel structure response prediction method according to claim 1, characterized in that: The explosion structure response neural network consists of an encoder, a processor, and a decoder: The encoder is composed of multiple single encoders connected in parallel. In each single encoder, a node encoder and an edge encoder for encoding node features and encoding edge features are respectively constructed based on the first feedforward neural network. A hierarchical coding strategy is used to firstly perform edge feature coding, and then perform node feature coding with the help of edge feature coding, so as to realize the combination of node encoder and edge encoder to form a single encoder. In the processor, a node feature processor and an edge feature processor for processing node features and edge features are respectively constructed based on a multi-level star-shaped neural network. A hierarchical processing strategy is used to first perform edge feature processing, and then perform node feature processing with the help of edge feature processing, so as to realize the combination of the node feature processor and the edge feature processor to form a processor; wherein the multi-level star-shaped neural network includes a plurality of feedforward neural network parallel networks, each feedforward neural network parallel network is formed by weighted fusion of the second and third feedforward neural networks, and at the same time, a weight is given to each feedforward neural network parallel network through the fourth feedforward neural network, and a plurality of feedforward neural network parallel networks are weighted fused to form a multi-level star-shaped neural network; The decoder is composed of multiple fifth feedforward neural networks connected in series.

3. The tunnel structure response prediction method according to claim 1, characterized in that: The steps for constructing the explosion structure mechanical response prediction database include: Based on the digital twin of the real tunnel area and combined with the design specifications of tunnel geometric protection means, the type, center position, and geometric parameter information of the tunnel anti-blast geometric protection means are obtained to construct a tunnel area dataset; Construct a tunnel area neural network, substitute the tunnel area data set into the tunnel area neural network, perform data training and optimization, and construct a tunnel area prediction model; Design a simulated tunnel area prediction data set, combine it with the tunnel area prediction model, and obtain the simulated tunnel area point cloud data; By simulating the point cloud data of the tunnel area and using the point cloud reconstruction method, the spatial entity of the tunnel area is obtained; The explosion center and explosion equivalent are randomly designed to obtain the explosion information of each tunnel area case. The tunnel area point cloud of the same tunnel area case has the same explosion information. The tunnel area explosion event is constructed by combining the simulated tunnel area point cloud data and the tunnel area space entity. Using the meshing method, the mesh nodes of the tunnel area are obtained. Using the numerical simulation method, combined with the explosion event in the tunnel area, the working condition design before the explosion is combined with the mechanical response of the tunnel structure according to the explosion event in the tunnel area and the node number of the tunnel area, and the explosion simulation data of the mechanical response of the tunnel structure is obtained. Obtain historical and experimental tunnel area prediction data sets, combine with tunnel area prediction models, obtain historical and experimental tunnel area point cloud data, and then obtain tunnel structure mechanical response explosion history and experimental data; Combining the explosion simulation data of tunnel structure mechanical response with historical and experimental data, an explosion structure mechanical response database is established through data conversion and data combination.

4. The tunnel structure response prediction method according to claim 3, characterized in that: The tunnel area neural network includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a third convolutional layer, a second pooling layer, a fully connected layer and an output layer, which are connected in sequence. The output layer includes three fully connected layers in series.

5. The tunnel structure response prediction method according to claim 3, characterized in that: The design steps of the simulation tunnel area prediction dataset include: Through the random number method, combined with the tunnel design specifications, the random area area, environment type, tunnel center X-axis position, tunnel center Y-axis position, tunnel center Z-axis position, tunnel length, tunnel width, tunnel height, whether there is tunnel anti-explosion geometric protection means, and tunnel anti-explosion geometric protection means type are designed; According to the length, width, height and area of ​​the tunnel, obtain the range of the number of nodes in the tunnel area; Obtain the number of nodes in the tunnel area through the random number method; Obtain the tunnel area node number, combine data according to the tunnel area node number, obtain tunnel area cases, combine tunnel area cases according to the tunnel area number, and obtain a simulation tunnel area prediction data set.

6. The tunnel structure response prediction method according to claim 3, characterized in that: The node data features in the tunnel area dataset are: tunnel area number, area, environment type, X-axis position of tunnel center, Y-axis position of tunnel center, Z-axis position of tunnel center, tunnel length, tunnel width, tunnel height, whether there are tunnel anti-explosion geometric protection measures, type of tunnel anti-explosion geometric protection measures, tunnel area node number, tunnel area node X-axis position, tunnel area node Y-axis position, tunnel area node Z-axis position; the node data features of the explosion structure mechanics response database are: tunnel number, whether there are tunnel anti-explosion geometric protection measures, tunnel node number, tunnel node X-axis position, tunnel node Y-axis position, tunnel node Z-axis position, explosion center X-axis position, explosion center Y-axis position, explosion center Z-axis position, explosion equivalent, tunnel node stress, tunnel node strain, tunnel node displacement.

7. The tunnel structure response prediction method according to claim 3, characterized in that: The steps of building the explosion structure mechanical response prediction database also include: According to the order of magnitude of the tunnel structure mechanical unit response, the damage level of the tunnel node is obtained, and the obtained tunnel node damage level is used as the newly added node data feature of the explosion structure mechanical response prediction database.

8. A protection optimization method based on a mechanical response model, characterized in that: include: Based on the tunnel structure response prediction method as described in any one of claims 1 to 7, the tunnel structure response under different tunnel anti-blast geometric protection means is obtained, and a geometric protection damage data set is constructed by calculating the tunnel physical damage and functional damage; The geometric protection damage data set is used to train and optimize the geometric protection damage neural network to obtain the geometric protection damage model; Based on the geometric protection damage model, the geometric protection means selection is realized according to the principle of minimum damage.

9. The protection optimization method according to claim 8, characterized in that: The node data features of the geometric protection damage dataset are: tunnel number, presence or absence of tunnel anti-blast geometric protection means, type of tunnel anti-blast geometric protection means, length of tunnel anti-blast geometric protection means, width of tunnel anti-blast geometric protection means, height of tunnel anti-blast geometric protection means, geometric protection distance, distance from explosion center, explosion equivalent, and functional damage.

10. The protection optimization method according to claim 8, characterized in that: The geometric protection damage neural network is composed of the sixth feedforward neural network, which consists of 1 input layer, 3 hidden layers and 1 output layer.

Images