A method for analyzing tunnel surrounding rock stress field based on physical information neural network

Abstract

The present invention proposes a tunnel surrounding rock stress field analysis method based on physical information neural network, which belongs to the field of tunnel engineering technology. The physical information neural network can effectively combine known information data with physical knowledge to realize the prediction of the tunnel surrounding rock elastic-plastic stress field, including: step one, constructing an artificial neural network ANN framework; step two, generating physical driving information based on the Mohr-Coulomb model; step three, generating data-driven information based on adaptive strain sampling points; step four, constructing a multi-objective loss function for training the physical information neural network; step five, training the constructed physical information neural network to solve the tunnel surrounding rock stress field. The analysis method of the present application does not require grid division, avoids problems such as grid distortion, and adopts a strain adaptive sampling strategy with low computational cost, high accuracy, and small error. It can accurately predict the complex elastic-plastic stress field of the tunnel surrounding rock and flexibly adjust the constitutive equation to match the characteristics of complex surrounding rock materials.

Description

Technical Field

[0001] The invention belongs to the technical field of tunnel engineering, and in particular relates to a tunnel surrounding rock stress field analysis method based on physical information neural network. Background Art

[0002] At present, the analysis and research on the elastic-plastic stress field of tunnel surrounding rock mainly adopts methods such as numerical simulation and model test. The numerical simulation method mainly uses finite element or finite difference software to perform numerical modeling and calculation on the tunnel surrounding rock to obtain the elastic-plastic stress field of the tunnel. The model test method is to build a physical model with appropriate proportions and materials for experimental research based on the principle of similarity.

[0003] The inventors of the present application discovered during their research that existing numerical simulation analysis methods incur high costs in terms of software learning, numerical modeling, and computational efficiency, and that model testing methods are limited by factors such as difficulty in controlling errors, high testing costs, and long cycles. Summary of the invention

[0004] In view of this, the present invention provides a tunnel surrounding rock stress field analysis method based on physical information neural network to solve the problems in the prior art that the numerical simulation analysis method has high costs in software learning, numerical modeling, and calculation efficiency, and the model test method is limited by factors such as difficult error control, high test cost, and long cycle.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A tunnel surrounding rock stress field analysis method based on physical information neural network, comprising:

[0007] Step 1: Use the artificial neural network ANN framework to build the neural network (NN) part of the physical information neural network (PINN), and prepare the input unit and output unit of the neural network;

[0008] In step 1, the input unit of the neural network includes the real space coordinates of the strain sampling points in the tunnel area (including the stratum) to be studied. , the real displacement required by data drive , true strain tensor and the true stress tensor ,in x is the abscissa in the Cartesian coordinate system, y is the vertical axis, for x Direction real displacement, for y Direction real displacement.

[0009] The output unit of the neural network includes a displacement prediction value output by neural network training ( , ), stress prediction value ( , , and ),in,( and ) respectively represent x and y The predicted displacement field in the direction, ( , and ) respectively represent x , y and z The predicted normal stress field in the direction, Represents the predicted shear stress field. Based on this, the multi-layer artificial neural network model can be expressed as:

[0010] ;

[0011] in , , , , and Respectively represent the output , , , , ,and Parameters of multi-layer artificial neural network ANN model.

[0012] At the same time, it is necessary to use the parameters directly output by the above neural network and the automatic differentiation function of the neural network to obtain the predicted strain field, which specifically includes the following steps:

[0013] Step A1: Calculate the strain field. The strain field is the derivative of the displacement field and is obtained by differentiation, as shown in the following formula:

[0014]

[0015] In the formula, It is expressed by the tensor index method x , y , z , and It is a shorthand for finding the derivative of displacement;

[0016] Step A2: Calculate the total volumetric strain , which is expressed as follows:

[0017]

[0018] In the formula, etc. represent the strain in the x direction, etc. represent the strain in the y direction, etc. represent the strain in the z direction;

[0019] Step A3: Calculate elastic volume strain , expressed by the predicted stress field output by the artificial neural network ANN:

[0020]

[0021] in is the total mean stress; is the bulk modulus;

[0022] Step A4: Calculate volumetric plastic strain ,pass and calculate:

[0023] ;

[0024] Step 2: Use machine learning language to express the physical part of the Mohr-Coulomb constitutive model to constrain the physical information neural network (PINN) and generate physical drive information.

[0025] In step 2, the machine learning language model construction of the Mohr-Coulomb constitutive model specifically includes the following steps:

[0026] Step B1: Establish the Mohr-Coulomb yield criterion:

[0027]

[0028] in is the yield function ( F <0 means the material is in the elastic stage. F = 0 means the material has just reached yield. F ＞ 0 means the material enters the plastic stage). is the cohesive force of the material, is the material friction angle, is the first stress invariant, is the second deviatoric stress invariant, is the stress Lode angle;

[0029] Step B2: When F＞ When 0, the plastic strain is calculated using the neural network as shown in the following formula:

[0030]

[0031] in is the plastic strain vector, is the plasticity multiplier, is the stress vector, is the plastic potential function, which is similar to the yield function when the relevant flow rule is adopted Equal; stress gradient of plastic potential function Determined by the following equation:

[0032]

[0033] in and is the deviatoric stress, represents the Kronecker symbol (i.e. ), is the second deviatoric stress invariant, and About Lord's Point Parameters; , , and is the normal stress or shear stress predicted by the neural network;

[0034] Step B3: Calculation of volumetric plastic strain , volumetric plastic strain Only with the first stress invariant The plastic potential is related to the plastic potential, so the plastic multiplier As shown below:

[0035]

[0036] in, represents the first stress invariant, is the material friction angle, is the yield function, represents volumetric plastic strain;

[0037] Step B4: Calculate the deviatoric strain tensor and the plastic strain tensor The component is expressed as:

[0038]

[0039] in is the strain field, is the plastic strain vector, represents the Kronecker symbol (i.e. ), represents the sum of positive strains, represents the sum of the plastic normal strains, , , Respectively x, y, z The strain in the axial direction, , , Respectively x , y , z Plastic strain in the axial direction;

[0040] Step B5: Calculate the stress tensor as shown below:

[0041]

[0042] in and is the Lamé constant.

[0043] Step 3: Generate multiple strain sampling points based on the adaptive method. The training coordinate data sets such as stress and displacement obtained at the strain sampling points are the information part of the physical information neural network, generating data-driven information;

[0044] The number of strain adaptive sampling points is configured to be 538; the roulette wheel selection method is used to select the sampling points for PINN training, and the individual fitness is defined as the Euclidean norm with respect to the strain vector and is expressed as:

[0045]

[0046] In the formula, , Indicates that the sample point is x, y , the positive strain value on represents the shear strain value, Indicates different sample points;

[0047] Used to assign a probability of selection to each data point , and is determined by the following formula:

[0048]

[0049] Step 4: Based on the physical driving information of step 2 and the data driving information of step 3, a physical information multi-objective loss function is constructed to optimize the performance of the physical information neural network PINN;

[0050] In step 4, the multi-objective loss function is as follows:

[0051]

[0052] in is the data-driven loss term; , , and is a loss term based on the physical equation.

[0053] The multi-objective loss function is constructed in a configuration form, where the data-driven loss term is calculated through different variables, as shown below:

[0054]

[0055] in is the total number of given sampling points; For different sample points; , , and is the parameter predicted by the network; , , and Represents the given data parameters of the finite element simulation;

[0056] Constitutive loss In configuration form it is constructed as:

[0057]

[0058] In the formula, deviatoric stress , represents the Kronecker symbol (i.e. );

[0059] Karush-Kuhn-Tucker (KKT) condition, that is ,KKT loss The configuration format is as follows:

[0060] ;

[0061] in, represents the plasticity multiplier, and F represents the yield function;

[0062] Loss term in the balance equation The configuration is written as:

[0063]

[0064] in is the physical force per unit mass;

[0065] Boundary Condition Loss Term Corresponding to the Dirichlet boundary and Neumann boundary, the configuration form is expressed as:

[0066]

[0067] in is the number of sampling points on the boundary; and represents the traction and displacement on the corresponding boundary; represents the unit external normal vector on the corresponding boundary.

[0068] Step 5: Combine the training coordinate data set generated in step 3 with the physical information loss function in step 4 to train the physical information neural network and solve the tunnel surrounding rock stress field.

[0069] A multi-layer artificial neural network ANN is used to establish the network part of the physical information neural network PINN. The weights, biases and activation functions in adjacent neural network layers are closely connected, and the mathematical expression is:

[0070]

[0071] in is the output of each layer of the neural network model; is the number of layers of the neural network; when or hour, and Represent the input and output of the model respectively; and yes The weights and biases of the neurons in the previous layer; represents the activation function for all layers except the last one.

[0072] , , , , and The artificial neural networks (ANNs) are constructed. Each ANN has 5 hidden layers, each layer has 100 neurons, and there are 40801 trainable parameters in total. The hyperbolic tangent function is selected as the activation function, and the neural network is trained by the L-BFGS optimizer.

[0073] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0074] 1. In the present invention, the complex elastic-plastic stress-strain field of the tunnel surrounding rock can be accurately predicted. When analyzing the elastic-plastic stress field of the tunnel surrounding rock, the traditional numerical method is highly dependent on grid division, and the test conditions of the model test method are difficult to control, and both require high costs and time resources. This application proposes a calculation framework that combines a physical information neural network calculation method. This method does not require grid division, avoids problems such as grid distortion, and adopts a strain adaptive sampling strategy. It has the advantages of low calculation cost, high accuracy and small error, and has shown excellent performance in predicting the elastic-plastic stress field of the tunnel surrounding rock;

[0075] 2. In the present invention, the physical information-based neural network efficiently utilizes the existing data information and physical laws, has a strong generalization ability, can flexibly adjust the constitutive equation to match the complex and changeable surrounding rock material properties, and opens up a new intelligent analysis approach for tunnel surrounding rock stress field analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0077] Figure 1 A schematic diagram of a physical information neural network PINN computing framework provided by an embodiment of the present invention;

[0078] Figure 2 A schematic diagram of a shallow buried tunnel provided by an embodiment of the present invention;

[0079] Figure 3 A schematic diagram of sample point distribution provided by an embodiment of the present invention, wherein: (a) random sample point distribution, (b) adaptive sample point distribution;

[0080] Figure 4 This is an error diagram of the physical information neural network prediction results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0082] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0084] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0085] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0086] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0087] Example 1

[0088] like Figure 1-Figure 4 As shown, an embodiment of the present invention discloses a tunnel surrounding rock stress field analysis method based on a physical information neural network, comprising:

[0089] Step 1: Use the artificial neural network ANN framework to build the neural network (NN) part of the physical information neural network (PINN), and prepare the input unit and output unit of the neural network;

[0090] In step 1, the input unit of the neural network includes the tunnel area (including the stratum) to be studied.

[0091] The real space coordinate vector of the variable sampling point , the real displacement required by data drive , true strain tensor and the true stress tensor ,in x is the abscissa in the Cartesian coordinate system, y is the vertical axis, for x Direction real displacement, for y Direction real displacement.

[0092] The output unit of the neural network includes a displacement prediction value output by neural network training ( , ), stress prediction value ( , , and ),in,( and ) respectively represent x and y The predicted displacement field in the direction, ( , and ) respectively represent x , y and z The predicted normal stress field in the direction, Represents the predicted shear stress field. Based on this, the multi-layer artificial neural network model can be expressed as:

[0093]

[0094] in , , , , and They represent the output predicted displacement field ( , ), predict the stress field ( , , ,and ) multi-layer artificial neural network ANN model.

[0095] According to the predicted displacement field output by the multi-layer artificial neural network ANN, the predicted strain field can be obtained, which is characterized by being realized by automatic differentiation of the neural network, and specifically includes the following steps:

[0096] Step A1: Calculate the strain field. The strain field is the derivative of the displacement field and is obtained by differentiation, as shown in the following formula:

[0097]

[0098] In the formula, It is expressed by the tensor index method x , y , z , and It is a shorthand for finding the derivative of displacement;

[0099] Step A2: Calculate the total volumetric strain , which is expressed as follows:

[0100]

[0101] In the formula, etc. indicates positive strain;

[0102] Step A3: Calculate volumetric elastic strain , represented by the stress field output by the artificial neural network ANN:

[0103]

[0104] in is the total mean stress; is the bulk modulus;

[0105] Step A4: Calculate volumetric plastic strain ,pass and calculate:

[0106]

[0107] Step 2: Use machine learning language to express the Mohr-Coulomb constitutive model and construct physical driving information for neural network training;

[0108] In step 2, the machine learning language model construction of the Mohr-Coulomb constitutive model specifically includes the following steps:

[0109] Step B1: Establishing the Mohr-Coulomb Yield Criterion

[0110]

[0111] in is the yield function ( F <0 means the material is in the elastic stage. F = 0 means the material has just reached yield. F >0 means the material enters the plastic stage). is the cohesive force of the material, is the material friction angle, is the first stress invariant, is the second deviatoric stress invariant, is the stress Lode angle;

[0112] Step B2: When F When >0, the plastic strain is calculated using a neural network, as shown in the following formula:

[0113]

[0114] in is the plastic strain vector, is the plasticity multiplier, is the stress vector, is the plastic potential function, which, when the relevant flow rule is adopted, is regarded as ; Stress gradient of plastic potential function Determined by the following equation:

[0115]

[0116] in and is the deviatoric stress, represents the Kronecker symbol (i.e. ), is the second deviatoric stress invariant, and About The parameters are shown in Table 1; , , and is the correct normal stress or shear stress predicted by the neural network;

[0117]

[0118] Step B3: Calculate the plasticity multiplier , volumetric plastic strain Only with the first stress invariant The plastic potential is related to the plastic potential, so the plastic multiplier As shown below:

[0119]

[0120] in, represents the first stress invariant, is the material friction angle, is the yield function, represents volumetric plastic strain;

[0121] Step B4: Calculate the deviatoric strain tensor and the plastic strain tensor The component is expressed as:

[0122]

[0123] in is the strain field, is the plastic strain vector, represents the Kronecker symbol, represents the sum of positive strains, represents the sum of the plastic normal strains, , , Respectively indicate x, y, z The strain in the axial direction, , , Respectively x , y , z Plastic strain in the axial direction;

[0124] Step B5: Calculate the stress tensor as shown below:

[0125]

[0126] in and is the Lamé constant.

[0127] Step 3: Generate multiple strain sampling points based on the adaptive method. The training coordinate data sets such as stress and displacement obtained at the strain sampling points are the information part of the physical information neural network, generating data-driven information;

[0128] The number of strain adaptive sampling points is configured to be 538; the roulette wheel selection method is used to select the sampling points for PINN training, and the individual fitness is defined as the Euclidean norm with respect to the strain vector and is expressed as:

[0129]

[0130] In the formula, etc. represent the strain value of the sample point, Indicates different sample points;

[0131] Used to assign a probability of selection to each data point , and is determined by the following formula:

[0132]

[0133] Step 4: Construct a physical information multi-objective loss function to optimize the performance of the physical information neural network PINN;

[0134] In step 4, the multi-objective loss function is as follows:

[0135]

[0136] in is the data-driven loss term; , , and is a loss term based on the physical equation.

[0137] The multi-objective loss function is constructed in a configuration form, where the data-driven loss term is calculated through different variables, as shown below:

[0138]

[0139] in is the total number of given sampling points; For different sample points; , , and is the parameter predicted by the network; , , and Represents the given data parameters of finite element simulation;

[0140] In the formula, deviatoric stress , represents the Kronecker symbol (i.e. );

[0141] Karush-Kuhn-Tucker (KKT) condition, that is ,KKT loss The configuration format is as follows:

[0142]

[0143] Loss term in the balance equation The configuration is written as:

[0144]

[0145] in is the physical force per unit mass;

[0146] Boundary Condition Loss Term Corresponding to the Dirichlet boundary and Neumann boundary, the configuration form is expressed as:

[0147]

[0148] in is the number of sampling points on the boundary; and represents the traction and displacement on the corresponding boundary; represents the unit external normal vector on the corresponding boundary.

[0149] Step 5: Combine the training coordinate data set generated in step 3 with the physical information multi-objective loss function in step 4 to train the physical information neural network and solve the tunnel surrounding rock stress field.

[0150] In the physical information neural network PINN network part established by multi-layer artificial neural network ANN, the weights, biases and activation functions in adjacent neural network layers are closely connected, and its mathematical expression is:

[0151]

[0152] in is the output of each layer of the neural network model; is the number of layers of the neural network; when or hour, and Represent the input and output of the model respectively; and yes The weights and biases of the neurons in the previous layer; represents the activation function for all layers except the last one.

[0153] , , , , and The artificial neural networks (ANNs) are constructed. Each ANN has 5 hidden layers, each layer has 100 neurons, and there are 40801 trainable parameters in total. The hyperbolic tangent function is selected as the activation function, and the neural network is trained by the L-BFGS optimizer.

[0154] Example 2

[0155] This embodiment proposes another specific embodiment based on embodiment 1, such as Figure 1 and Figure 2 As shown, Figure 2 This is a schematic diagram of a tunnel surrounding rock calculation case under plane strain conditions. Due to geometric symmetry, only one quarter of the domain is modeled. The side length of the stratum is 10 meters. There is a tunnel with a radius of 2 meters in the lower left corner of the stratum and a burial depth of 8 meters. Distributed force applied to the top side of the formation; the bottom side of the plate is fixed (i.e. ); and the left and right edges are vertical and have no traction (i.e. ,in Indicates the boundary xy direction), there is no external force at the lower left quarter (i.e. ,in Indicates the direction of the parallel external normal n The normal stress, Indicates the direction of the vertical external normal n The surrounding rock parameters of the case are shown in Table 2, including Young's elastic modulus ( ), Poisson's ratio ( )、Cohesion( )、Internal friction angle( ), soil unit weight ( ) and the overburden load distribution force ( ).

[0156]

[0157] Figure 3 Schematic diagram of sample point distribution of the embodiment, where (a) is a distribution diagram of equally spaced sampling points. The strain prediction result obtained by pre-training the neural network with random sampling points, and then the improved sampling points can be generated by the strain adaptive sampling strategy in step 4, such as Figure 3 As shown in (b), the use of adaptive sampling points to train the neural network can improve the accuracy of the neural network prediction results. The prediction results of the neural network in this case are compared with the high-fidelity finite element simulation results using a mesh of 20,000 triangular elements, as shown in Figure 4 As shown, this is a relative error diagram of the prediction results of the physical information neural network. The relative errors of more than 99% of the prediction points are within 0.1, indicating that the neural network performs well.

[0158] The circuits, electronic components and modules involved are all prior art and can be fully implemented by those skilled in the art. Needless to say, the content protected by the present invention does not involve improvements to software and methods.

[0159] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0160] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A tunnel surrounding rock stress field analysis method based on physical information neural network, characterized in that: include: Step 1: Use the artificial neural network ANN framework to build the neural network NN part of the physical information neural network PINN, and prepare the input unit and output unit of the neural network; Step 2: Use machine learning language to express the physical part of the Mohr-Coulomb constitutive model to constrain the physical information neural network PINN and generate physical driving information; Step 3: Generate multiple strain sampling points based on the adaptive method. The stress, displacement and training coordinate data sets obtained at the strain sampling points are the information part of the physical information neural network PINN, generating data-driven information; Step 4: Based on the physical driving information of step 2 and the data driving information of step 3, a physical information multi-objective loss function is constructed to optimize the performance of the physical information neural network PINN; Step 5: Combine the training coordinate data set generated in step 3 and the physical information multi-objective loss function in step 4 to train the physical information neural network PINN to solve the tunnel surrounding rock stress field; In step 1, the input unit of the neural network includes the real space coordinate vector of the strain sampling point in the tunnel area to be studied , the real displacement required by data drive , true strain tensor and the true stress tensor ,in x is the abscissa in the Cartesian coordinate system, y is the vertical axis, for x Directional real displacement, for y Directional real displacement; The output unit includes a displacement prediction value output by neural network training ( , ), stress prediction value ( , , and ),in, and Respectively x and y The predicted displacement field in the direction, , and Respectively x , y and z The predicted normal stress field in the direction, Represents the predicted shear stress field, and the multi-layer artificial neural network ANN can be expressed as: ; in( , )and( , , and ) represent the predicted displacement field ( , ) and stress field ( , , ,and )’s multi-layer artificial neural network ANN model; According to the predicted displacement field output by the multi-layer artificial neural network ANN, the predicted strain field can be obtained, which is realized by the automatic differentiation function of the neural network, specifically including the following steps: Step A1: Calculate the tunnel surrounding rock strain field. The strain field is the derivative of the displacement field and is obtained by differentiation, as shown in the following formula: ; In the formula, is the Cartesian coordinate expressed by the tensor index method x , y , z , and This is a shorthand for finding the differential of displacement; Step A2: Calculate the total volume strain of the tunnel surrounding rock , which is expressed as follows: ， In the formula, represents the strain in the x direction, represents the strain in the y direction, represents the strain in the z direction; Step A3: Calculate the elastic volume strain of the tunnel surrounding rock , expressed by the predicted stress field output by the artificial neural network ANN: ; Where P is the total mean stress; K is the bulk modulus; Step A4: Calculate the plastic volume strain of the tunnel surrounding rock ,pass and calculate: ； In step 2, the physical part of the Mohr-Coulomb constitutive model constraint physical information neural network PINN expressed in machine learning language specifically includes the following steps: Step B1: Establish the Mohr-Coulomb yield criterion as shown below: ； where F is the yield function, F <0 means the material is in the elastic stage. F = 0 means the material has just reached yield. F＞ 0 means the material enters the plastic stage. is the cohesive force of the material, is the material friction angle, is the first stress invariant, is the second deviatoric stress invariant, is the stress Lode angle; Step B2: When F＞ 0 Calculate the plastic strain of the tunnel surrounding rock as shown in the following formula: ; in is the plastic strain vector, is the plasticity multiplier, is the stress vector, is the plastic potential function. When the relevant flow rule is adopted, With yield function equal; Step B3: Calculate the plasticity multiplier , as shown below: ， in, represents the first stress invariant, is the material friction angle, is the yield function, represents volumetric plastic strain; Step B4: Calculate the deviatoric strain tensor of the tunnel surrounding rock and the plastic strain tensor The component is expressed as: ; in is the strain field, is the plastic strain vector, , represents the sum of positive strains, represents the sum of the plastic normal strains, , , Respectively x, y, z The total strain in the axial direction, , , Respectively x , y , z Plastic strain in the axial direction; Step B5: Calculate the tunnel surrounding rock stress tensor as shown below: ; in and is the Lamé constant.

2. The method for analyzing tunnel surrounding rock stress field based on physical information neural network according to claim 1 is characterized in that: In step 3, in which multiple strain sampling points are generated based on the adaptive method, the probability of each strain sampling point being selected as a training point is determined according to the strain size, and the implementation method is as follows: The number of strain adaptive sampling points is set to 538, and the roulette wheel selection method is used to select the sampling points for physical information neural network PINN training. It is defined as the Euclidean norm of the strain vector, which is expressed as: ; In the formula, , Indicates that the sample point is x, y , positive strain value, represents the shear strain at the sample point, Indicates different sample points; Used to assign a probability of selection to each data point , and is determined by the following formula: 。 3. The tunnel surrounding rock stress field analysis method based on physical information neural network according to claim 1 is characterized in that: In step 4, the multi-objective loss function is as follows: ; in is the data-driven loss term, , , and is a loss term based on the physical equation.

Images