Method and apparatus for dynamic prediction of stress characteristics of surrounding rock in large tunnels

Abstract

This invention provides a method and apparatus for dynamically predicting the stress characteristics of surrounding rock in large tunnels, relating to the field of underground engineering surrounding rock stability prediction technology. The method includes: establishing a loss function for a dynamic multi-physics coupled neural network framework by combining data fitting residuals, temporal continuity regularization, and a loss function for a mixed physical constraint of surrounding rock in large tunnels based on layered temporal parameters, multi-source physical field parameters, and geological parameters; extracting the layered temporal features of the layered excavation of the surrounding rock in the large tunnel and dynamically determining the sampling data; then optimizing the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock in the large tunnel; and obtaining the trained dynamic multi-physics coupled neural network framework based on the optimized loss function, thereby generating dynamic prediction results of the stress characteristics of the surrounding rock in the large tunnel. This invention can effectively solve the problems of embedding multi-field coupling equations, capturing temporal dynamic features, and optimizing data in high-risk areas.

Description

Technical Field

[0001] This invention relates to the field of underground engineering surrounding rock stability prediction technology, and in particular to a method and device for dynamic prediction of the stress characteristics of surrounding rock in large tunnels. Background Technology

[0002] In underground engineering projects such as water conservancy and hydropower, and mining, the stability of the surrounding rock in large tunnels is directly related to the safety of the project. During the layered excavation process, the surrounding rock undergoes complex elastoplastic deformation, seepage pressure changes, and damage accumulation, and its stress characteristics exhibit multi-physics field coupling effects and temporal dynamic evolution features. Traditional numerical simulation methods, such as the finite element method, can partially describe the mechanical behavior of surrounding rock, but they have the following limitations: First, existing methods simplify the coupling process of elastoplastic, seepage, and damage fields. For example, they ignore the dynamic influence of damage on seepage channels or use a step-by-step solution strategy, leading to error accumulation and making it difficult to truly reflect the interaction of multiple fields. Second, the deformation and damage of surrounding rock during layered excavation have strong temporal dependence, while traditional methods rely on static parameters or piecewise assumptions and cannot capture the dynamic correlation between excavation stages. Third, machine learning models based on pure data lack physical constraints, and the prediction results are prone to deviating from actual physical laws when training data is insufficient or the working conditions are complex. Existing physical information neural network frameworks are mostly designed for single physical fields and do not effectively integrate multi-field coupling equations and temporal characteristics. Finally, traditional sampling strategies are difficult to optimize data for high-risk areas, resulting in low model training efficiency and insufficient prediction accuracy in key areas. At the same time, changes in geological conditions and the dynamic process of layered excavation require the model to have adaptive weight adjustment capabilities, while existing methods usually use fixed weights, which are difficult to adapt to complex working conditions. In recent years, some studies have attempted to introduce physical information neural networks into the field of geotechnical engineering, but their application is still limited to a single physical field or static working conditions. They have failed to effectively solve key problems such as neural network embedding of multi-field coupled control equations, dynamic fusion of layered excavation time series data and physical constraints, and adaptive sampling and model optimization for high-risk areas. Summary of the Invention

[0003] To overcome the problems existing in related technologies, this invention provides a training method and apparatus for a neural network framework with dynamic multiphysics coupling.

[0004] According to a first aspect of the present invention, a method for training a dynamically multi-physics coupled neural network framework is provided, the method comprising:

[0005] Obtain the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel;

[0006] Based on the aforementioned hierarchical time-series parameters, multi-source physical field parameters, and geological parameters, a hybrid physical constraint loss function for the surrounding rock of the large chamber is constructed.

[0007] By combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term, a loss function for a neural network framework with dynamic multi-physics coupling is established.

[0008] Extract the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage from the real-time data of stratified excavation of the large tunnel;

[0009] Based on the aforementioned hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data;

[0010] Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function.

[0011] Based on the optimized loss function, a trained neural network framework with dynamic multiphysics coupling is obtained.

[0012] In some exemplary embodiments of the present invention, based on the foregoing scheme, the layering timing parameter includes: the current excavation layer number. , No. Layer height of excavation layer , No. Span of excavation layer , No. Continuous excavation time of each layer and the total number of excavated layers ;

[0013] The multi-source physical field parameters include: the first Displacement field of the previous layer of excavation layer , No. Stress field of the layer preceding the excavation layer , No. seepage pressure field of the layer before the excavation layer and the Damage factor of the previous layer of excavation layer ;

[0014] The geological parameters include: permeability coefficient Damage threshold, critical osmotic pressure gradient .

[0015] In some exemplary embodiments of the present invention, based on the foregoing scheme, the mixed physical constraint loss function of the surrounding rock of the large chamber includes:

[0016] Based on the elastoplastic constitutive relation, the residual terms of the stress constraint equation are established;

[0017] Based on Darcy's law and the damage-seepage coupling relationship, a constraint residual term for seepage field formation is constructed.

[0018] Based on damage evolution theory and excavation sequence, residual terms of constraint equations are constructed;

[0019] By fitting the residual terms of the stress constraint equation, the constraint residual terms of the seepage field formation, and the residual terms of the constraint equation, the mixed physical constraint loss function of the surrounding rock of the large chamber is obtained.

[0020] In some exemplary embodiments of the present invention, based on the foregoing scheme, and combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the temporal continuity regularization term, the loss function of the dynamically multi-physics coupled neural network framework includes:

[0021] Based on the predicted displacement value, measured displacement value, predicted damage value and measured damage value of each layer in all excavated layers, calculate the data fitting term and residual term for all excavated layers;

[0022] Based on the derivatives of the continuous excavation time and the rate of change of the predicted displacement value for each of all excavated layers, the temporal continuity regularization term for all excavated layers is calculated.

[0023] The first balance result is obtained by balancing the mixed physical constraint loss function of the surrounding rock of the large chamber using the first weighting coefficient;

[0024] The second balance result is obtained by balancing the time series continuity regularization term using the second weighting coefficient.

[0025] The loss function of the dynamically multi-physics coupled neural network framework is obtained by summing the first equilibrium result, the second equilibrium result, and the residual term of the data fitting term.

[0026] In some exemplary embodiments of the present invention, based on the foregoing scheme, the hierarchical temporal characteristics include strain energy density and damage factor. Based on the hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampling data, including:

[0027] Calculate the probability of the sampling point in the region based on the strain energy density and the damage factor;

[0028] Based on the probability of the sampling points in the region, the sampling priority is determined;

[0029] The sampling points are dynamically adjusted based on the sampling priority.

[0030] Collect sampling data from the sampling points.

[0031] In some exemplary embodiments of the present invention, based on the foregoing scheme, calculating the probability of regional sampling points according to the strain energy density and the damage factor includes:

[0032] Calculate the first The exponential function value of the strain energy density and the first preset sensitivity coefficient at each sampling point;

[0033] Calculate the first The Gaussian error function values ​​of the damage factor and the second preset sensitivity coefficient at each sampling point;

[0034] The summation is calculated by multiplying the exponential function value and the Gaussian error function value at all sampling points;

[0035] According to the The sum of the product of the exponential function value and the Gaussian error function value at the nth sampling point is used to calculate the nth... The probability of each sampling point.

[0036] According to a second aspect of the present invention, a method for dynamically predicting the stress characteristics of surrounding rock in a large tunnel is provided, comprising:

[0037] Obtain the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel;

[0038] Based on the aforementioned hierarchical time-series parameters, multi-source physical field parameters, and geological parameters, a hybrid physical constraint loss function for the surrounding rock of the large chamber is constructed.

[0039] By combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term, a loss function for a neural network framework with dynamic multi-physics coupling is established.

[0040] Extract the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage from the real-time data of stratified excavation of the large tunnel;

[0041] Based on the aforementioned hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data;

[0042] Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function.

[0043] Based on the optimized loss function, a trained neural network framework with dynamic multiphysics coupling is obtained.

[0044] Input the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel. Using the trained dynamic multi-physics coupled neural network framework, generate dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel. The dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel include predicted displacement field, predicted stress field, predicted seepage field, and predicted damage factor.

[0045] In some exemplary embodiments of the present invention, based on the foregoing scheme, the dynamically multi-physics coupled neural network framework includes:

[0046] The module consists of an input module, a residual module, a time series module, a feature fusion module, and an output module.

[0047] The input module is used to input the layered time-series parameters, the multi-source physical field parameters, and the geological parameters. The residual module is built on the basis of a residual network. Based on the layered time-series parameters, the multi-source physical field parameters, and the geological parameters, it uses the loss function of a dynamic multi-physics coupled neural network framework established by the constraint equation residuals based on elastoplastic constitutive modeling, the constraint residuals based on Darcy's law, and the constraint equation residuals based on damage evolution to generate the difference between the displacement field, stress field, seepage field, and damage factor. The time-series module is used to input the layered time-series parameters, the multi-source physical field parameters, and the geological parameters. The sequence parameters, the multi-source physical field parameters, and the geological parameters generate preliminary predictions of the displacement field, stress field, seepage field, and damage factor over time. The feature fusion module introduces data fitting terms, residual terms, and time series continuity regularization terms to superimpose and fuse the preliminary predictions of the displacement field, stress field, seepage field, and damage factor over time output by the time series module with the differences calculated by the residual module to obtain a fusion result. The output module is used to generate predicted displacement field, predicted stress field, predicted seepage field, and predicted damage factor based on the fusion result.

[0048] According to a third aspect of the present invention, a dynamic prediction device for the stress characteristics of surrounding rock in a large tunnel is provided, comprising:

[0049] The data acquisition module is used to acquire the stratification time-series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber;

[0050] The first function construction module is used to construct a mixed physical constraint loss function for the surrounding rock of the large chamber based on the layered time series parameters, multi-source physical field parameters, and geological parameters.

[0051] The second function construction module is used to combine the loss function of the mixed physical constraint loss function of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term to establish the loss function of the neural network framework of dynamic multi-physics coupling.

[0052] The feature extraction module is used to extract the stratified temporal features of surrounding rock deformation, stress, seepage and damage from the real-time data of the layered excavation of the surrounding rock in the large chamber;

[0053] The sampling data determination module is used to dynamically determine the sampling data based on the hierarchical time series characteristics and using a dual-threshold dynamic sampling strategy.

[0054] The function optimization module is used to optimize the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber based on the sampled data using an adaptive weight adjustment algorithm, so as to obtain the optimized loss function.

[0055] The network determination module is used to obtain the trained dynamic multiphysics coupled neural network framework based on the optimized loss function;

[0056] The prediction module is used to input the stratification time parameters, multi-source physical field parameters and geological parameters of the surrounding rock of the large chamber, and to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber using the trained dynamic multi-physics field coupled neural network framework. The dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber include predicted displacement field, predicted stress field, predicted seepage field and predicted damage factor.

[0057] According to a fourth aspect of the present invention, an electronic device is provided, comprising: a processor; and a memory, wherein computer-readable instructions are stored thereon, and when executed by the processor, the computer-readable instructions implement the dynamic prediction method for the stress characteristics of the surrounding rock of a large tunnel in the second aspect.

[0058] According to a fifth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the dynamic prediction method for the stress characteristics of the surrounding rock of a large tunnel in the second aspect.

[0059] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0060] This invention constructs a hybrid physical constraint loss function that integrates elastoplasticity, seepage, and damage fields. By combining temporal continuity regularization and dynamic sampling strategies, it can effectively solve the problems of embedding multi-field coupled equations, capturing temporal dynamic features, and optimizing data in high-risk areas. It has the advantages of improving the accuracy of surrounding rock stress prediction and the model generalization ability under complex geological conditions.

[0061] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0062] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the specification, serve to explain the principles of the invention.

[0063] Figure 1 This diagram illustrates a system architecture of an exemplary application environment for a training method and apparatus for a dynamically multi-physics coupled neural network framework, which can be applied according to embodiments of the present invention.

[0064] Figure 2 The schematic diagram illustrates a flowchart of a training method for a dynamically multi-physics coupled neural network framework according to some embodiments of the present invention.

[0065] Figure 3 The schematic diagram illustrates a method for dynamically predicting the stress characteristics of the surrounding rock in a large tunnel according to some embodiments of the present invention;

[0066] Figure 4 The schematic diagram illustrates the structure of a neural network framework for dynamic multiphysics coupling according to some embodiments of the present invention;

[0067] Figure 5 A schematic diagram of a dynamic prediction device for the stress characteristics of the surrounding rock of a large tunnel according to some embodiments of the present invention is shown.

[0068] Figure 6 The schematic diagram illustrates the structure of a computer system of an electronic device according to some embodiments of the present invention;

[0069] Figure 7 A schematic diagram of a computer-readable storage medium according to some embodiments of the present invention is shown. Detailed Implementation

[0070] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0071] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0072] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0073] Figure 1 The diagram illustrates a system architecture of an exemplary application environment for a method and apparatus for dynamically predicting the stress characteristics of surrounding rock in a large tunnel, which can be applied according to embodiments of the present invention.

[0074] like Figure 1 As shown, system architecture 100 may include one or more terminal devices such as desktop computer 101, portable computer 102, and smartphone 103, network 104, and server 105. Network 104 is used as a medium to provide a communication link between the terminal devices and server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc. Terminal devices may be various electronic devices with data processing capabilities, which have a display screen for displaying the dynamic prediction process and / or results of the stress characteristics of the surrounding rock in the large tunnel to the user, including but not limited to the aforementioned desktop computer, portable computer, smartphone, etc. It should be understood that... Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of terminal devices, networks, and servers. For example, server 105 could be a sub-server cluster composed of multiple sub-servers.

[0075] The dynamic prediction method for the stress characteristics of surrounding rock in large tunnels provided in this embodiment of the invention can generally be executed by a terminal device, and correspondingly, the dynamic prediction device for the stress characteristics of surrounding rock in large tunnels is generally installed in the terminal device. However, it is readily understood by those skilled in the art that the dynamic prediction method for the stress characteristics of surrounding rock in large tunnels provided in this embodiment of the invention can also be executed by a server 105, and correspondingly, the dynamic prediction device for the stress characteristics of surrounding rock in large tunnels can also be installed in the server 105. This exemplary embodiment does not impose any special limitations on this.

[0076] Furthermore, it should be understood that the dynamic prediction method for the stress characteristics of surrounding rock in large tunnels according to embodiments of the present invention can be configured as a software module. In some implementation scenarios, the dynamic prediction scheme for the stress characteristics of surrounding rock in large tunnels according to the present invention can be deployed independently to achieve dynamic prediction of the stress characteristics of surrounding rock in large tunnels in different regions. In other implementation scenarios, the dynamic prediction scheme for the stress characteristics of surrounding rock in large tunnels according to the present invention can be deployed in other software as a functional module of that software, such as in software for analyzing surrounding rock in large tunnels. The present invention does not impose any particular restrictions on the application of the dynamic prediction method for the stress characteristics of surrounding rock in large tunnels.

[0077] In traditional underground engineering rock stability analysis, the finite element method (FEM) often employs a step-by-step solution strategy to handle the coupling problem between elasto-plastic deformation and seepage field. For example, the damage evolution equation is simplified to static parameter input, leading to the neglect of the feedback effect of permeability changes on seepage channels. Regarding time-series dynamic modeling, existing numerical simulations rely on fixed time steps for layered excavation calculations, failing to effectively capture the dynamic correlation between rock damage accumulation and excavation procedures. When using a purely data-driven neural network model, the strain energy density data obtained based on a uniform sampling strategy exhibits an unbalanced distribution, with insufficient sample density in high-gradient stress regions, causing a conflict between the predicted damage factor and the physical laws of the elasto-plastic constitutive equations.

[0078] For example, during the layered excavation of the underground powerhouse of a hydropower station, the finite element model set the permeability coefficient to a fixed value, without considering the layered excavation. The increase in damage factors after layer excavation leads to a change in permeability coefficient The increased actual situation resulted in seepage pressure prediction errors exceeding the baseline value. The lack of time-series modeling led to... When excavating in layers, the stress field data from the previous layer is directly inherited, which fails to reflect the influence of the residual stress generated by the excavation of the previous five layers on the expansion of the plastic zone in the current layer. When using a conventional convolutional neural network to predict the deformation of the surrounding rock, the sampling coverage of high strain energy density areas (exceeding 1.5 kJ / m³) is insufficient, causing the predicted stress concentration values ​​in key areas to deviate from the actual physical laws.

[0079] If the above problems are not addressed, multi-field coupling modeling errors will cause the design parameters of the support structure to deviate from actual working conditions, potentially leading to progressive instability of the surrounding rock. The lack of temporal dynamic characteristics will reduce the reliability of excavation process optimization decisions and increase the risk of sudden water inrush accidents. The disconnect between data sampling strategies and physical constraints will exacerbate the model's prediction inaccuracies under complex geological conditions, especially in areas with weak interlayers or well-developed joints; insufficient model generalization ability will directly affect the scientific validity of engineering safety assessment conclusions.

[0080] To address the aforementioned problems, this invention first considers how to effectively embed the elastoplastic constitutive equation, Darcy's law, and damage evolution equation into a neural network training framework to overcome the shortcomings of multi-field coupled modeling. While traditional step-by-step solution strategies simplify the computation process, they neglect the dynamic feedback of damage to seepage channels; for example, the permeability variation coefficient is not dynamically adjusted with the damage factor. To address this, this invention attempts to simultaneously introduce stress field residuals, seepage field residuals, and damage evolution residuals into the loss function, forming a hybrid physical constraint. However, a fixed allocation of weights for different physical fields is insufficient to adapt to the dynamic changes of the dominant physical field during layered excavation. For instance, the first five excavation layers are dominated by elastoplastic deformation, while the seepage field weight needs to be increased in the latter three layers due to damage accumulation.

[0081] To address the lack of temporal dynamic characteristics, this invention explores adding a temporal continuity regularization term in addition to the data fitting term. Traditional static parameter inputs cannot reflect the impact of excavation processes on residual stress; for example, the expansion of the plastic zone generated by the previous excavation layer will change the stress distribution of the current layer. By designing a regularization term based on the displacement change rate, the neural network is forced to maintain physical continuity between the prediction results of adjacent excavation layers, avoiding the dynamic correlation break caused by time step simplification.

[0082] To address the disconnect between data-driven approaches and physical constraints, this invention analyzes the shortcomings of uniform sampling strategies in covering high-risk areas. For example, in regions with high strain energy density, sparse samples cause predicted stress to deviate from the actual trend. A dual-threshold dynamic sampling strategy is attempted, calculating regional priorities based on strain energy density and damage factor, and dynamically adjusting the distribution of sampling points. Simultaneously, an adaptive weight adjustment algorithm is combined to optimize the weight coefficients of each physical constraint term during training based on real-time sampling data. For instance, when the seepage pressure gradient exceeds a critical value, the weight of the seepage field residual term is automatically increased.

[0083] In this regard, such as Figure 2 As shown, Figure 2 This is a flowchart illustrating a training method for a dynamically multi-physics coupled neural network framework according to an exemplary embodiment of the present invention, comprising the following steps:

[0084] S210: Obtain the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel;

[0085] S220: Based on the aforementioned layered time series parameters, multi-source physical field parameters, and geological parameters, construct a mixed physical constraint loss function for the surrounding rock of the large chamber;

[0086] S230: Combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term, establish the loss function of the neural network framework of dynamic multi-physics coupling;

[0087] S240: Extract the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage from the real-time data of stratified excavation of the surrounding rock in the large chamber;

[0088] S250: Based on the aforementioned hierarchical time-series characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data;

[0089] S260: Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function;

[0090] S270: Based on the optimized loss function, the trained dynamic multiphysics coupled neural network framework is obtained.

[0091] The embodiments of the present invention will now be described in detail.

[0092] In S210, the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel are obtained.

[0093] Layered time-series parameters refer to the geometric and temporal parameters that describe the changes over time during the layered excavation of surrounding rock. These parameters can be implemented using the current excavation layer number, layer height, span, continuous excavation time, and total number of layers. They are used to capture the temporal dependence of surrounding rock deformation and damage during excavation. They include: the current excavation layer number... , No. Layer height of excavation layer , No. Span of excavation layer , No. Continuous excavation time of each layer and the total number of excavated layers .

[0094] Multi-source physical field parameters refer to physical quantities that include displacement field, stress field, seepage pressure field, and damage factor. These can be realized using variables related to the elastoplastic constitutive equation, Darcy's law, and damage evolution equation, and are used to establish physical constraints for multi-field coupling. They include: the... Displacement field of the previous layer of excavation layer , No. Stress field of the layer preceding the excavation layer , No. seepage pressure field of the layer before the excavation layer and the Damage factor of the previous layer of excavation layer .

[0095] Geological parameters include: permeability coefficient Damage threshold, critical osmotic pressure gradient .

[0096] Current number of excavation layers It can be set to 1 to 5 layers, for example, when excavating the third layer. =3; Floor height It can be 3 to 5 meters, span The depth can range from 8 to 12 meters, with a continuous excavation time. Available for 24 to 72 hours 。 Previous layer displacement field It can include displacements along the X, Y, and Z axes, i.e. ,For example =[0.12m, 0.08m, -0.05m] represents the displacement state after the second layer of excavation, and the stress field of the previous layer. The principal stress components can be recorded in tensor form, for example... =[15MPa, 10MPa, 8MPa]. Seepage pressure field It can be decomposed into pore water pressure components in three directions, for example, before the third layer was excavated. =2.3kPa =1.8kPa = 4.1 kPa. Damage factors The value can range from 0 to 1, for example =0.35 indicates the degree of damage after the second layer of excavation. The permeability coefficient is a geological parameter. Can be cm / s to cm / s, critical osmotic pressure gradient It can range from 0.5 kPa / m to 2.0 kPa / m.

[0097] Specifically, the current number of excavation layers Compared with the total number of excavated layers Constructing a time series benchmark, for example when =5 and When the value is 3, it indicates that the current construction phase is the third layer of the five-layer excavation plan. (Layer height) and span Together they define the geometric dimensions of the excavated space, for example =4.2 meters and =10.5 meters combined to form the spatial characteristics of the third layer of excavation section. Continuous excavation time Form time-related constraints with the physical field parameters of the previous layer, for example when At 48 hours, the model passes through the previous layer's displacement field. Calculate the displacement increment within 48 hours of excavation for this layer, using the parameters of the current layer. Previous layer stress field. With the current layer damage factor Establishing cross-layer mechanical relationships, for example when =12MPa and When the coefficient of stress is 0.28, the model calculates the stress redistribution caused by the third layer of excavation based on the elastoplastic constitutive equation. The previous layer's seepage pressure field... With permeability coefficient Conditions for forming the seepage field iteration, such as after the second layer of excavation. =2.1kPa and = The calculation of the third-layer seepage velocity constrained by cm / s. Damage threshold and critical osmotic pressure gradient. Joint control of the damage evolution process, for example when When the pressure is 1.2 kPa / m, the model automatically triggers the seepage-damage coupling calculation module. The combined application of these parameters enables the model to derive the multi-field coupling effect of the current layer from the physical field state of the previous layer, avoiding error propagation caused by step-by-step solutions. At the same time, the geological parameters limit the boundary of rock mass characteristics, ensuring the applicability of the physical constraint equations under complex geological conditions.

[0098] Therefore, this invention enables the precise definition of parameters for dynamic multiphysics coupling models. Layered temporal parameters provide the model with a spatiotemporal dynamic benchmark for layered excavation, allowing the model to accurately track temporal and spatial scale changes during the excavation phase. Multi-source physical field parameters, by introducing the physical field states of adjacent excavation layers, force the model to learn the recursive relationship of multi-field coupling, avoiding the error accumulation of step-by-step solutions in traditional methods. Geological parameters are directly related to the seepage characteristics and damage evolution laws of the rock mass, providing necessary geological boundary conditions for the residual terms of the seepage field and damage field in the physical constraint equations, ensuring the model's generalization ability under complex geological conditions. This parameter definition method solves the key problem caused by parameter ambiguity in dynamic multiphysics coupling models, improves the accuracy of the model in capturing the temporal dynamic characteristics of layered excavation during dynamic multiphysics coupling, and achieves reliable embedding of multi-field coupling equations.

[0099] Traditional loss function designs often employ only a single physical field constraint equation or oversimplify multi-field coupling equations, resulting in models that fail to accurately capture the dynamic response of the surrounding rock under multi-field interactions. To address this, in S220, this invention constructs a hybrid physical constraint loss function for the surrounding rock of a large tunnel based on the aforementioned layered time-series parameters, multi-source physical field parameters, and geological parameters.

[0100] The hybrid physical constraint loss function refers to the loss term of the residuals of the control equations of stress field, seepage field and damage field. It can be realized by weighted combination of elastoplastic constitutive residuals, Darcy's law residuals and damage evolution residuals, and is used to embed multi-field coupled physical laws into the neural network training process.

[0101] The elastoplastic deformation constraint term is constructed based on the improved elastoplastic constitutive equation, and its expression is the sum of squared residuals between the predicted stress and the theoretically calculated stress. A damage factor, such as the elastic modulus, is introduced as a material stiffness reduction factor in the theoretical stress calculation. The value range can be from 10 GPa to 50 GPa, damage factor The value of can range from 0 to 0.9, with the specific value depending on the surrounding rock type. The seepage field constraint term is implemented by modifying Darcy's law equation, adding an adjustment term linearly related to the damage factor to the permeability coefficient expression, where the permeability variation coefficient... The value of can range from 0.1 to 5.0, used to characterize the amplification effect of different damage levels on permeability. The damage evolution constraint term adopts a cumulative time-series residual design, based on excavation time. Compared with theoretical damage increment The product term establishes the correlation between damage accumulation and construction procedures, where the excavation time unit can be hours or days, depending on the project schedule requirements. Weighting coefficients. An adaptive adjustment mechanism is adopted, for example, settings can be configured in the initial stage. =5:3:2, and the ratio is dynamically adjusted during training based on the convergence speed of the residuals of each physical field.

[0102] Specifically, during neural network training, the elastoplastic deformation constraint term forces the model to learn the material weakening effect caused by damage by embedding the damage factor into the stress calculation process. For example, when the damage factor... When the elastic strain tensor reaches 0.5, the material stiffness will decrease to 50% of its initial value. The calculations need to be adjusted accordingly. The seepage field constraint term is obtained through (1+ This section establishes a quantitative relationship between damage and permeability. For every 0.1 increase in the damage factor, the permeability coefficient can be amplified by 1% to 50%, with the specific amplification factor determined by the permeability change coefficient. The damage evolution constraint term captures the temporal cumulative effect of damage during layered construction by accumulating the damage increment residuals at each excavation stage. For example, a 10% increase in excavation time for each layer will lead to an 8%-15% increase in damage increment. The weighting coefficients employ a dynamic adjustment algorithm based on the residual change rate. When the residual decrease rate of a certain physical field falls below a threshold, its corresponding weighting coefficient can be increased by 10%-30%, ensuring the balance of multi-field coupling constraints. This combined loss function, in conjunction with a dual-threshold dynamic sampling strategy, prioritizes the collection of sample data from high-damage areas, improving the model's prediction accuracy in key areas by 15%-25%.

[0103] As one implementation method, a mixed physical constraint loss function for the surrounding rock of the large tunnel can be designed. include:

[0104]

[0105] in, The residual terms of the constraint equations formed based on elastoplastic constitutive modeling are expressed as follows: , Indicates the total number of samples. Indicates the first One sample, Indicates the first Predicted stress for each sample, Indicates the first Damage factors for each sample Indicates the first The elastic modulus of a sample Indicates the first The elastic strain tensor of a sample Indicates the first The additional stress caused by plastic deformation in each sample This represents the constraint residual term for the seepage field formation based on Darcy's law, and , Indicates the first The permeability change coefficient of each sample Indicates the first The predicted pore water pressure tensor for each sample. Indicates the first The permeability tensor of a sample Indicates the first One sample in Coordinates in direction Indicates the first One sample in Coordinates in direction Indicates the first One sample in seepage velocity in the direction, Indicates to Differentiate, Represents the residual terms of the constraint equations based on damage evolution, and , Indicates the first The number of layers in the excavation layer where each sample is located. Indicates the current excavation layer number. For the first The first sample was taken during the excavation. Predicted loss factor after layer completion Indicates the excavation layer and =1,2…; For the first The first sample was taken during the excavation. The prediction loss factor after the layer ends; For the excavation of the first The continuous excavation time of the layer, For the excavation of the first Theoretical loss increment per unit time for each layer , and They are respectively , and The weight.

[0106] In other words, The calculation method is as follows: First, for Iterate through each sample, and for each sample... Calculate and predict stress Compared with considering damage factors Stress after influence The squared Euclidean distance between each sample is calculated. Then, the sum of the squared distances over all samples is divided by the total number of samples. ,get The value of .

[0107] The calculation method for the item is as follows: Iterate through each sample, and for each sample... ,calculate The square of the Euclidean norm. The permeability variation coefficient, As a damage factor, Let be the permeability tensor. Let be the pore water pressure. Then, sum the calculation results for all samples and divide by the total number of samples. ,get The value of .

[0108] The calculation method for the item is as follows: Iterate through each sample, and for each sample... From the excavation layer where the sample is located up to the current excavation layer Calculate the difference in damage factors between every two layers. Compared with theoretical damage increment The absolute value of the difference between the layers. Then sum the differences for all layers, and sum the results for all samples and divide by the total number of samples. ,get The value of .

[0109] Finally, , and Multiply by the corresponding weighting coefficients respectively , and Then, by adding the three together, the final physical constraint loss function of the surrounding rock of the main chamber is obtained. .

[0110] Therefore, this invention constructs a hybrid physical constraint loss function that integrates multi-physics coupling effects. This loss function can effectively balance the coupling relationship between elasto-plastic deformation, dynamic changes in the seepage field, and damage evolution. Specifically, The term introduces damage factors It can achieve a direct correlation between stress calculation and material damage; Item passed (1+ It can capture the dynamic impact of damage on the permeability coefficient; The residual design employs an additive form, which can reflect the temporal cumulative effect of damage evolution during layered excavation. Therefore, this loss function can accurately capture the dynamic response law of the surrounding rock under multi-field interactions, overcoming the problem of oversimplification of multi-field coupling equations in traditional methods. Furthermore, by introducing weighting coefficients... , and This invention allows for dynamic adjustment of constraint strength during training based on the importance of the residuals of each physical field, thus avoiding the weakening of the constraint effect of some physical fields.

[0111] Because relying solely on the physical constraint loss function in dynamic multiphysics coupling scenarios may lead to an over-reliance on theoretical equations and neglect of the fit to measured data, and because the temporal dynamic characteristics of layered excavation are not explicitly modeled, the prediction results may exhibit jumps or discontinuities in the time dimension. Therefore, in S230, a loss function for the neural network framework of dynamic multiphysics coupling is established by combining the mixed physical constraint loss function of the surrounding rock of the large tunnel, the residual term of the data fitting term, and the temporal continuity regularization term.

[0112] The data fitting residual term ensures consistency between the model output and actual observation data by comparing the mean square error of predicted displacement with that of measured displacement, and the mean square error of predicted damage value with that of measured damage value, layer by layer. The time series continuity regularization term forces the model to learn the continuous evolution law of the displacement field by constraining the matching of the displacement change rate of adjacent excavation layers with the time interval. Weighting coefficients and It is set as a dynamically adjustable parameter, and the contribution of the physical constraint term and the time continuity term is optimized through an adaptive weight adjustment algorithm. It can be appropriately increased in regions with high strain energy density. The value enhances the physical constraint effect, and can increase the value in regions sensitive to sudden displacement changes. Value-enhanced temporal continuity constraints.

[0113] Specifically, the calculation of the data fitting residual term involves comparing the norm of the predicted displacement with the measured displacement for each excavation layer. For example, if measured displacement data for a certain excavation layer is missing, the system automatically reduces the weight of the data term for that layer. The time-series continuity regularization term establishes the correlation between the rate of change of the displacement field and the actual excavation time through mathematical derivatives. For example, when the time interval between adjacent excavation layers is 2 days, the model automatically calculates the reasonable gradient range of displacement change within that period. Weighting coefficients and An exponential decay strategy is employed for dynamic adjustment. Initially set to 0.8 and 0.5 during the early training phase to enhance data fitting, the values ​​are gradually adjusted to 0.3 and 0.7 as the number of iterations increases to improve physical regularity. Through this composite loss function structure, the model can both adhere to the constitutive equations of elastoplastic mechanics and maintain consistency with measured data, while ensuring the continuous evolution of prediction results over time.

[0114] As one implementation method, a loss function can be designed that combines the loss function of the mixed physical constraints of the surrounding rock in the large tunnel, the data fitting term, the residual term, and the temporal continuity regularization term to establish a loss function for a dynamic multi-physics coupled neural network framework. include:

[0115]

[0116] in, Represents the data fitting term and the residual term. , This indicates the total number of layers that have been excavated. Indicates the first Predicted displacement values ​​for each excavated layer. Indicates the first Measured displacement values ​​of each excavated layer. Indicates the first Predicted damage values ​​for each excavated layer. Indicates the first Measured damage values ​​of the excavated layer Represents the time series continuity regularization term and , For the first The duration of continuous excavation of each excavation layer. For the first The duration of continuous excavation of each excavation layer. Indicates the first The time variable sign for the continuous excavation time of each layer, i.e. This represents the rate of change of displacement when the derivative is taken with respect to the displacement. This represents the loss function for the mixed physical constraints of the surrounding rock in the main tunnel. and They are respectively and The weighting coefficients are the first weighting coefficient and the second weighting coefficient.

[0117] Through the above technical solutions, this invention achieves a multi-dimensional fusion of data-driven, physical constraints, and temporal dynamics, solving the problem of synergistic optimization of model generalization and physical regularity in multi-field coupled prediction. The data fitting residual term directly constrains the deviation between the model output and the measured engineering data, avoiding overfitting caused by purely physical constraints. The temporal continuity regularization term forces the model to capture the continuous and gradual characteristics of surrounding rock deformation during excavation, eliminating the predictive jumps in the time dimension inherent in traditional static models. The introduction of weight coefficients allows the model to adaptively adjust the weight ratio of physical constraint terms and temporal continuity terms based on the differences in the importance of data quality and physical regularity under different working conditions. This multi-objective optimization method ensures that the neural network framework adheres to fundamental theories such as elastoplastic mechanics and seepage mechanics, while also conforming to actual monitoring data, guaranteeing the physical rationality of the prediction results on the time axis, thus improving the accuracy and reliability of dynamic prediction of the stress characteristics of surrounding rock in large tunnels.

[0118] In S240, the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage are extracted from the real-time data of the layered excavation of the surrounding rock in the large chamber.

[0119] This invention does not impose specific limitations on the methods for extracting layered temporal features. In some embodiments, layered temporal convolution modules or spatiotemporal attention mechanisms can be designed to extract layered temporal features related to rock deformation, stress, seepage, and damage from real-time data of layered excavation of the surrounding rock in a large tunnel.

[0120] In this embodiment of the invention, a time series module is designed based on a bidirectional LSTM layer or a temporal convolutional layer, and the sampling time points are arranged in an orderly manner to extract the stratified temporal features of surrounding rock deformation, stress, seepage and damage from the real-time data of the stratified excavation of the surrounding rock in the large tunnel.

[0121] In S250, based on the hierarchical time-series characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data.

[0122] The dual-threshold dynamic sampling strategy refers to a method for calculating regional sampling priority based on strain energy density and damage factor. It can be implemented by calculating the probability distribution through a combination of exponential function and error function, and is used to prioritize the collection of data in high-risk areas to improve the prediction accuracy of key areas.

[0123] Traditional uniform sampling strategies cannot effectively capture key areas with high strain energy density and high damage factor, resulting in insufficient prediction accuracy for high-risk areas during model training. At the same time, static sampling mechanisms are difficult to adapt to the dynamic evolution of the surrounding rock physical field during layered excavation, causing the dual problems of data redundancy and loss of key features.

[0124] In this embodiment of the invention, the layered temporal characteristics include strain energy density and damage factor. Based on these layered temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data, including:

[0125] Calculate the region sampling priority based on the strain energy density and the damage factor;

[0126] Based on the sampling priority of the region, the sampling points are dynamically adjusted;

[0127] Collect sampling data from the sampling points.

[0128] Strain energy density, by quantifying the degree of energy accumulation during the elastoplastic deformation of surrounding rock, has become a key indicator for identifying high-energy accumulation areas. For example, the measured strain energy density at the edge of the excavated layer can reach the order of 3.5-5.2 MJ / m³. The damage factor is calculated using a continuous damage mechanics model, reflecting the degree of fracture propagation within the rock mass. When the damage factor exceeds 0.65, it is identified as a high-risk area. The calculation of regional sampling priority combines an exponential function and an error function; the specific expression is as follows:

[0129]

[0130] in, Indicates the first The probability of each sampling point This represents the first preset sensitivity coefficient. This represents the second preset sensitivity coefficient. Indicates the first Strain energy density at each sampling point Indicates the first Damage factor at each sampling point Indicates the total number of sampling points. Indicates the first One sampling point, Represents an exponential function. This represents the Gaussian error function.

[0131] Specifically, during the layered excavation process, strain energy density and damage factor field distribution data for each layer are acquired synchronously after excavation. A dual-threshold dynamic sampling strategy is employed. First, the entire grid cells are prioritized, with areas exhibiting strain energy densities higher than 3.2 MJ / m³ and damage factors exceeding 0.6 automatically designated as primary sampling zones. Subsequently, based on priority calculations, the sampling point density in primary sampling zones is increased to 2-3 times that of conventional zones, while the spacing between sampling points in low-priority zones is expanded to 1.5 times the original spacing. During data acquisition, the evolution of the physical field is monitored in real-time. When the strain energy density growth rate in a region exceeds 25% after excavation of two adjacent layers, a denser sampling mechanism for that region is automatically triggered. The acquired sampling data is processed by an adaptive weight adjustment algorithm to optimize the weight allocation of each physical field in the hybrid physical constraint loss function. This dynamic sampling mechanism improves data acquisition efficiency by over 40% in the transition zone between excavation layers, and reduces stress prediction errors in key areas to within 5.8%.

[0132] Exponential function Configured to nonlinearly amplify the high strain energy density region, with a first preset sensitivity coefficient. The value can be adjusted between 0.5 and 3.0. For example, in areas with complex geological conditions, it can be adjusted. Set to 2.5 to enhance strain energy monitoring. Error function Configured to probabilistically process damage factors, with a second preset sensitivity coefficient. The value can be adjusted between 0.3 and 1.2. For example, in areas with significant seepage, it can be adjusted. Set to 1.0 to emphasize damage assessment. Normalization of the denominator ensures that the sum of the priorities of each region is 1, forming a dynamic weight allocation mechanism.

[0133] When strain energy density When the threshold value of 1.5 MJ / m³ is exceeded, the nonlinear amplification effect of the exponential function can increase the priority of the corresponding region by more than three times, effectively capturing energy-concentrated areas. Error function damage factors Mapped to the interval between 0 and 1, when When the damage threshold of 0.8 is reached, the error function output value can reach above 0.9, significantly distinguishing the degree of damage. This can be achieved through dynamic adjustment. and The combination of parameters allows sampling priority allocation to adapt to the needs of different engineering stages, such as setting it in the early stages of excavation. : The sampling ratio was initially 2:1, focusing on energy monitoring, and later adjusted to 1:2, focusing on damage assessment. The dynamic adjustment of sampling points was based on real-time calculated priority parameters, prioritizing the selection of... Regions with a probability value higher than 0.15 are sampled intensively, while Regions with a probability value below 0.05 will have their sampling frequency reduced or even be removed. This dynamic allocation mechanism can increase the sampling density of critical regions by 2-3 times compared to uniform sampling, while reducing the proportion of redundant sampling by more than 40%.

[0134] Based on this, the present invention not only ensures that the spatiotemporal distribution of the neural network training dataset matches the dynamic process of multi-field coupling in the surrounding rock, improving the model's prediction sensitivity and training efficiency for key hazardous areas, but also effectively overcomes the shortcomings of traditional uniform sampling in complex working conditions, such as data redundancy and insufficient feature capture, capturing key areas with high strain energy density and high damage factor, and improving the model's prediction accuracy for high-risk areas.

[0135] In S260, based on the sampled data, an adaptive weight adjustment algorithm is used to optimize the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber, thereby obtaining the optimized loss function.

[0136] The adaptive weight adjustment algorithm refers to the mechanism of dynamically optimizing the weight coefficients of different physical fields based on real-time sampled data. It can be implemented by using an adjustment strategy that correlates gradient backpropagation error with residual contribution, and is used to balance the weight distribution of multi-field coupling equations in the loss function.

[0137] This invention does not limit the specific content of the adaptive adjustment algorithm. For example, in some embodiments, the adaptive adjustment algorithm may be a particle swarm optimization algorithm, a genetic algorithm, a simulated annealing algorithm, etc.

[0138] Subsequently, the sampled data was used as input data for the adaptive stiffness adjustment algorithm to optimize and adjust the loss function of the mixed physical constraint of the surrounding rock in the large chamber, resulting in the optimized loss function.

[0139] In S270, a trained neural network framework with dynamic multiphysics coupling is obtained based on the optimized loss function.

[0140] In other words, the optimized loss function is used as the loss function of the dynamically coupled multiphysics neural network framework, thereby obtaining the trained dynamically coupled multiphysics neural network framework.

[0141] According to a second aspect of the present invention, a method for dynamically predicting the stress characteristics of surrounding rock in a large tunnel is also provided, comprising:

[0142] S310: Obtain the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber;

[0143] S320: Based on the aforementioned layered time series parameters, multi-source physical field parameters, and geological parameters, construct a mixed physical constraint loss function for the surrounding rock of the large chamber;

[0144] S330: Combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term, establish the loss function of the neural network framework of dynamic multi-physics coupling;

[0145] S340: Extract the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage from the real-time data of layered excavation of the surrounding rock in the large chamber;

[0146] S350: Based on the aforementioned hierarchical time-series characteristics, the sampling data is dynamically determined using a dual-threshold dynamic sampling strategy;

[0147] S360: Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function;

[0148] S370: Based on the optimized loss function, the trained dynamic multiphysics coupled neural network framework is obtained;

[0149] S380: Input the stratification time series parameters, multi-source physical field parameters and geological parameters of the surrounding rock of the large tunnel, and use the trained dynamic multi-physics field coupled neural network framework to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel. The dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel include predicted displacement field, predicted stress field, predicted seepage field and predicted damage factor.

[0150] In S310, the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel are obtained.

[0151] This part has been described in detail in S210, and will not be repeated here.

[0152] In S320, a mixed physical constraint loss function for the surrounding rock of the large tunnel is constructed based on the layered time series parameters, multi-source physical field parameters, and geological parameters.

[0153] This part has been described in detail in S220, and will not be repeated here.

[0154] In S330, a loss function for a dynamic multi-physics coupled neural network framework is established by combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the residual term of the data fitting term, and the time series continuity regularization term.

[0155] This part has been described in detail in S230, and will not be repeated here.

[0156] In S340, the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage are extracted from the real-time data of the layered excavation of the surrounding rock in the large chamber.

[0157] This part has been described in detail in S240, and will not be repeated here.

[0158] In S350, based on the hierarchical time-series characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data.

[0159] This part has been described in detail in S250, and will not be repeated here.

[0160] In S360, based on the sampled data, an adaptive weight adjustment algorithm is used to optimize the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber, thereby obtaining the optimized loss function.

[0161] This part has been described in detail in S260, and will not be repeated here.

[0162] In S370, a trained neural network framework with dynamic multiphysics coupling is obtained based on the optimized loss function.

[0163] This part has been described in detail in S270, and will not be repeated here.

[0164] S380: Input the stratification time series parameters, multi-source physical field parameters and geological parameters of the surrounding rock of the large tunnel, and use the trained dynamic multi-physics field coupled neural network framework to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel. The dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel include predicted displacement field, predicted stress field, predicted seepage field and predicted damage factor.

[0165] In some implementations, reference Figure 4 As shown, the neural network framework for dynamic multiphysics coupling includes:

[0166] The module consists of an input module, a residual module, a time series module, a feature fusion module, and an output module.

[0167] The input module is used to input the layered time-series parameters, the multi-source physical field parameters, and the geological parameters. The residual module is built on the basis of a residual network. Based on the layered time-series parameters, the multi-source physical field parameters, and the geological parameters, it uses the loss function of a dynamic multi-physics coupled neural network framework established by the constraint equation residuals based on elastoplastic constitutive modeling, the constraint residuals based on Darcy's law, and the constraint equation residuals based on damage evolution to generate the difference between the displacement field, stress field, seepage field, and damage factor. The time-series module is used to input the layered time-series parameters, the multi-source physical field parameters, and the geological parameters. The sequence parameters, the multi-source physical field parameters, and the geological parameters generate preliminary predictions of the displacement field, stress field, seepage field, and damage factor over time. The feature fusion module introduces data fitting terms, residual terms, and time series continuity regularization terms to superimpose and fuse the preliminary predictions of the displacement field, stress field, seepage field, and damage factor over time output by the time series module with the differences calculated by the residual module to obtain a fusion result. The output module is used to generate predicted displacement field, predicted stress field, predicted seepage field, and predicted damage factor based on the fusion result.

[0168] The skip connection structure of the residual module can be configured with three residual blocks, each containing two convolutional layers and a ReLU activation function, mitigating the gradient vanishing problem through cross-layer connections. The time step of the time series module corresponds to the number of layers being mined; for example, when processing the [number of layers]... When excavating layer data, the time series module can store the previous data. The historical state of the layer, and at the same time, combined with time series forecasting methods to perform the first layer Layer prediction. The stitching dimension of the feature fusion module can be dynamically adjusted according to the size of the input feature map. For example, on a 256×256 feature map, the 64 channels of the static feature and the 64 channels of the dynamic feature are stitched together along the channel dimension to form a 128-channel system.

[0169] Specifically, the stratified temporal parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber are input into a dynamic multi-physics coupled neural network framework through the input module. The residual module and time series module call relevant data for calculation: the constraint equations based on elastoplastic constitutive modeling, the constraints based on Darcy's law for seepage field, and the constraint equations based on damage evolution introduced by the residual module are used to calculate the residual terms of the relevant equations. Specifically, these are the residual values ​​between the input displacement field, stress field, seepage field, and damage factor and the displacement field, stress field, seepage field, and damage factor calculated through the physical constraint equations. These different types of residual values ​​form multiple residual blocks in the residual network. The time series module builds a time series prediction method, such as introducing a bidirectional LSTM layer or a time convolutional layer, to obtain preliminary predicted values ​​of the displacement field, stress field, seepage field, and damage factor over time. Through the feature fusion module, data fitting terms, residual terms, and time series continuity regularization terms are introduced. The displacement field, stress field, seepage field, and damage factor values ​​initially predicted by the time series module are superimposed and fused with the residuals calculated by the residual module to form a complete dynamic multi-physics coupled neural network, which can further predict the output displacement field, stress field, seepage field, and damage factor.

[0170] Therefore, this invention effectively separates the encoding processes of static field data and dynamic temporal features, avoiding prediction bias caused by feature confusion. The residual network structure retains shallow geometric information through skip connections, improving the stability of deep feature extraction. The temporal module uses a bidirectional loop structure to capture the dependencies between different stages of the excavation process, enhancing the modeling accuracy of damage accumulation effects. The feature fusion module achieves spatiotemporal feature complementarity through channel splicing, strengthening the joint representation of multi-physics coupling relationships. This framework optimizes the interaction mechanism between multi-field coupling equations and temporal dynamic features through modular design, significantly improving the prediction robustness of layered excavation processes under complex geological conditions.

[0171] According to a third aspect of the present invention, a dynamic prediction device for the stress characteristics of surrounding rock in a large tunnel is also provided, with reference to... Figure 5 As shown, it includes:

[0172] The data acquisition module 510 is used to acquire the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber;

[0173] The first function construction module 520 is used to construct a mixed physical constraint loss function for the surrounding rock of the large chamber based on the layered time series parameters, multi-source physical field parameters and geological parameters.

[0174] The second function construction module 530 is used to combine the loss function of the mixed physical constraint loss function of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term to establish the loss function of the neural network framework of dynamic multi-physics coupling.

[0175] The feature extraction module 540 is used to extract the stratified temporal features of surrounding rock deformation, stress, seepage and damage from the real-time data of the layered excavation of the surrounding rock in the large chamber;

[0176] The sampling data determination module 550 is used to dynamically determine the sampling data based on the hierarchical time series characteristics and using a dual-threshold dynamic sampling strategy.

[0177] The function optimization module 560 is used to optimize the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber according to the sampled data using an adaptive weight adjustment algorithm, so as to obtain the optimized loss function.

[0178] The network determination module 570 is used to obtain the trained dynamic multiphysics coupled neural network framework based on the optimized loss function;

[0179] The prediction module 580 is used to input the stratification time-series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber, and to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber using the trained dynamic multi-physics field coupled neural network framework. The dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber include predicted displacement field, predicted stress field, predicted seepage field, and predicted damage factor.

[0180] It should be noted that although several modules of the dynamic prediction device for the stress characteristics of the surrounding rock in the large tunnel have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules described above can be embodied in a single module or unit. Conversely, the features and functions of a single module described above can be further divided into multiple modules or sub-modules for embodiment.

[0181] Furthermore, in an exemplary embodiment of the present invention, an electronic device capable of implementing the above-described method for dynamic prediction of the stress characteristics of the surrounding rock in a large tunnel is also provided.

[0182] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented as entirely hardware embodiments, entirely software embodiments (including firmware, microcode, etc.), or embodiments combining hardware and software aspects, collectively referred to herein as “circuit,” “module,” or “system.”

[0183] The following reference Figure 6 To describe an electronic device 600 according to such an embodiment of the present invention. Figure 6 The electronic device 600 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0184] like Figure 6 As shown, the electronic device 600 is manifested in the form of a general-purpose computing device. The components of the electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different system components (including storage unit 620 and processing unit 610), and a display unit 640.

[0185] The storage unit stores program code that can be executed by the processing unit 610, causing the processing unit 610 to perform the steps described in the "Exemplary Method" section above, based on various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 3 S310: Obtain the layered temporal parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel; S320: Based on the layered temporal parameters, multi-source physical field parameters, and geological parameters, construct a mixed physical constraint loss function for the surrounding rock of the large tunnel; S330: Combine the mixed physical constraint loss function of the surrounding rock of the large tunnel, the data fitting term, the residual term, and the temporal continuity regularization term to establish the loss function of the dynamic multi-physics coupling neural network framework; S340: Extract the layered temporal features of surrounding rock deformation, stress, seepage, and damage from the real-time data of layered excavation of the surrounding rock of the large tunnel; S350: Based on the layered temporal features, dynamically determine the... S360: Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function; S370: Based on the optimized loss function, a trained dynamic multi-physics coupling neural network framework is obtained; S380: The layered time-series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber are input, and the trained dynamic multi-physics coupling neural network framework is used to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber. The dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber include predicted displacement field, predicted stress field, predicted seepage field, and predicted damage factor.

[0186] Storage unit 620 may include readable media in the form of volatile storage units, such as random access memory (RAM) 621 and / or cache memory 622, and may further include read-only memory (ROM) 623.

[0187] Storage unit 620 may also include a program / utility 624 having a set (at least one) of program modules 625, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0188] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0189] Electronic device 600 can also communicate with one or more external devices 670 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 660. As shown, network adapter 660 communicates with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0190] Through the description of the above embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions of the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of the present invention.

[0191] In exemplary embodiments of the present invention, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of the present invention may also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the present invention described in the "Exemplary Methods" section above.

[0192] refer to Figure 7As shown, a program product 700 for implementing the above-described dynamic prediction method for the stress characteristics of surrounding rock in a large tunnel, according to an embodiment of the present invention, is described. It can be a portable compact disc read-only memory (CD-ROM) and includes program code, and can run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In the present invention, the readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0193] The program product may employ any combination of one or more readable storage media. Readable storage media may be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0194] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0195] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0196] Through the description of the above embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions of the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of the present invention.

[0197] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. The invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0198] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A training method for a dynamically multi-physics coupled neural network framework, characterized in that, include: Obtain the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel; Based on the aforementioned layered time-series parameters, multi-source physical field parameters, and geological parameters, a mixed physical constraint loss function for the surrounding rock of the large excavation chamber is constructed; the layered time-series parameters include: the current number of excavation layers. , No. Layer height of excavation layer , No. Span of excavation layer , No. Continuous excavation time of each layer and the total number of excavated layers ; By combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term, a loss function for a neural network framework with dynamic multi-physics coupling is established. Extract the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage from the real-time data of stratified excavation of the large tunnel; Based on the aforementioned hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data; Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function. Based on the optimized loss function, a trained dynamic multiphysics coupled neural network framework is obtained; the mixed physical constraint loss function of the surrounding rock of the large chamber includes: Based on the elastoplastic constitutive relation, the residual terms of the stress constraint equation are established; Based on Darcy's law and the damage-seepage coupling relationship, a constraint residual term for seepage field formation is constructed. Based on damage evolution theory and excavation sequence, residual terms of constraint equations are constructed; By fitting the residual terms of the stress constraint equation, the constraint residual terms formed by the seepage field, and the residual terms of the constraint equation, the mixed physical constraint loss function of the surrounding rock of the large chamber is obtained. The hierarchical temporal characteristics include strain energy density and damage factor. Based on these hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data, including: Calculate the probability of the sampling point in the region based on the strain energy density and the damage factor; Based on the probability of the sampling points in the region, the sampling priority is determined; The sampling points are dynamically adjusted based on the sampling priority. Collect sampling data from the sampling points.

2. The training method for the dynamically multi-physics coupled neural network framework according to claim 1, characterized in that, The multi-source physical field parameters include: the first Displacement field of the previous layer of excavation layer , No. Stress field of the layer preceding the excavation layer , No. seepage pressure field of the layer before the excavation layer and the Damage factor of the previous layer of excavation layer ; The geological parameters include: permeability coefficient Damage threshold and critical osmotic pressure gradient .

3. The training method for the dynamically multi-physics coupled neural network framework according to claim 1, characterized in that, Combining the loss function of the mixed physical constraints of the surrounding rock in the large tunnel, the data fitting term, the residual term, and the temporal continuity regularization term, the loss function of the neural network framework for dynamic multi-physics coupling includes: Based on the predicted displacement value, measured displacement value, predicted damage value and measured damage value of each layer in all excavated layers, calculate the data fitting term and residual term for all excavated layers; Based on the derivatives of the continuous excavation time and the rate of change of the predicted displacement value for each of all excavated layers, the temporal continuity regularization term for all excavated layers is calculated. The first balance result is obtained by balancing the mixed physical constraint loss function of the surrounding rock of the large chamber using the first weighting coefficient; The second balance result is obtained by balancing the time series continuity regularization term using the second weighting coefficient. The loss function of the dynamically multi-physics coupled neural network framework is obtained by summing the first equilibrium result, the second equilibrium result, and the residual term of the data fitting term.

4. The training method for the dynamically multi-physics coupled neural network framework according to claim 1, characterized in that, The probability of a sampling point in the region is calculated based on the strain energy density and the damage factor, including: Calculate the first The exponential function value of the strain energy density and the first preset sensitivity coefficient at each sampling point; Calculate the first The Gaussian error function values ​​of the damage factor and the second preset sensitivity coefficient at each sampling point; The summation is calculated by multiplying the exponential function value and the Gaussian error function value at all sampling points; According to the The sum of the product of the exponential function value and the Gaussian error function value at the nth sampling point is used to calculate the nth... The probability of each sampling point.

5. A method for dynamic prediction of the stress characteristics of surrounding rock in a large tunnel, characterized in that, include: Obtain the stratification time series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large tunnel; The layering timing parameters include: the current excavation layer number. , No. Layer height of excavation layer , No. Span of excavation layer , No. Continuous excavation time of each layer and the total number of excavated layers ; Based on the aforementioned hierarchical time-series parameters, multi-source physical field parameters, and geological parameters, a hybrid physical constraint loss function for the surrounding rock of the large chamber is constructed. By combining the loss function of the mixed physical constraint of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term, a loss function for a neural network framework with dynamic multi-physics coupling is established. Extract the stratified temporal characteristics of surrounding rock deformation, stress, seepage and damage from the real-time data of stratified excavation of the large tunnel; Based on the aforementioned hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data; Based on the sampled data, the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber are optimized using an adaptive weight adjustment algorithm to obtain the optimized loss function. Based on the optimized loss function, a trained neural network framework with dynamic multiphysics coupling is obtained; Input the stratification time series parameters, multi-source physical field parameters and geological parameters of the surrounding rock of the large tunnel, and use the trained dynamic multi-physics field coupled neural network framework to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel. The dynamic prediction results of the stress characteristics of the surrounding rock of the large tunnel include predicted displacement field, predicted stress field, predicted seepage field and predicted damage factor. The mixed physical constraint loss function of the surrounding rock of the main chamber includes: Based on the elastoplastic constitutive relation, the residual terms of the stress constraint equation are established; Based on Darcy's law and the damage-seepage coupling relationship, a constraint residual term for seepage field formation is constructed. Based on damage evolution theory and excavation sequence, residual terms of constraint equations are constructed; By fitting the residual terms of the stress constraint equation, the constraint residual terms formed by the seepage field, and the residual terms of the constraint equation, the mixed physical constraint loss function of the surrounding rock of the large chamber is obtained. The hierarchical temporal characteristics include strain energy density and damage factor. Based on these hierarchical temporal characteristics, a dual-threshold dynamic sampling strategy is used to dynamically determine the sampled data, including: Calculate the probability of the sampling point in the region based on the strain energy density and the damage factor; Based on the probability of the sampling points in the region, the sampling priority is determined; The sampling points are dynamically adjusted based on the sampling priority. Collect sampling data from the sampling points.

6. The method for dynamic prediction of the stress characteristics of the surrounding rock in a large tunnel according to claim 5, characterized in that, The neural network framework for dynamic multiphysics coupling includes: The module consists of an input module, a residual module, a time series module, a feature fusion module, and an output module. The input module is used to input the layered time-series parameters, the multi-source physical field parameters, and the geological parameters. The residual module is built on the basis of a residual network. Based on the layered time-series parameters, the multi-source physical field parameters, and the geological parameters, it uses the loss function of a dynamic multi-physics coupled neural network framework established by the constraint equation residuals based on elastoplastic constitutive modeling, the constraint residuals based on Darcy's law, and the constraint equation residuals based on damage evolution to generate the difference between the displacement field, stress field, seepage field, and damage factor. The time-series module is used to input the layered time-series parameters, the multi-source physical field parameters, and the geological parameters. The sequence parameters, the multi-source physical field parameters, and the geological parameters generate preliminary predictions of the displacement field, stress field, seepage field, and damage factor over time. The feature fusion module introduces a data fitting term, a residual term, and a time series continuity regularization term to superimpose and fuse the preliminary predictions of the displacement field, stress field, seepage field, and damage factor over time output by the time series module with the differences calculated by the residual module to obtain a fusion result. The output module is used to generate predicted displacement field, predicted stress field, predicted seepage field, and predicted damage factor based on the fusion result.

7. A dynamic prediction device for the stress characteristics of surrounding rock in a large tunnel, characterized in that, The method for dynamic prediction of the stress characteristics of the surrounding rock of a large chamber according to claim 5 includes: The data acquisition module is used to acquire the stratification time-series parameters, multi-source physical field parameters, and geological parameters of the surrounding rock of the large chamber; The first function construction module is used to construct a mixed physical constraint loss function for the surrounding rock of the large chamber based on the layered time series parameters, multi-source physical field parameters, and geological parameters. The second function construction module is used to combine the loss function of the mixed physical constraint loss function of the surrounding rock of the large chamber, the data fitting term, the residual term, and the time series continuity regularization term to establish the loss function of the neural network framework of dynamic multi-physics coupling. The feature extraction module is used to extract the stratified temporal features of surrounding rock deformation, stress, seepage and damage from the real-time data of the layered excavation of the surrounding rock in the large chamber; The sampling data determination module is used to dynamically determine the sampling data based on the hierarchical time series characteristics and using a dual-threshold dynamic sampling strategy. The function optimization module is used to optimize the weights of different physical fields in the mixed physical constraint loss function of the surrounding rock of the large chamber based on the sampled data using an adaptive weight adjustment algorithm, so as to obtain the optimized loss function. The network determination module is used to obtain the trained dynamic multiphysics coupled neural network framework based on the optimized loss function; The prediction module is used to input the stratification time parameters, multi-source physical field parameters and geological parameters of the surrounding rock of the large chamber, and to generate dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber using the trained dynamic multi-physics field coupled neural network framework. The dynamic prediction results of the stress characteristics of the surrounding rock of the large chamber include predicted displacement field, predicted stress field, predicted seepage field and predicted damage factor.

8. An electronic device, characterized in that, include: processor; as well as The memory stores computer-readable instructions, which, when executed by the processor, implement the dynamic prediction method for the stress characteristics of the surrounding rock of the large chamber as described in claim 5 or 6.

Images