PINN-based slope soil shear strength parameter inversion method and equipment

Abstract

The invention discloses a PINN-based slope soil shear strength parameter inversion method, which comprises the following steps: determining a geometric model of a slope, and arranging a plurality of monitoring points in the slope; uniformly collecting a plurality of coordinate points in a domain and a boundary defined by the geometric model, wherein the collected coordinate points comprise all monitoring points; constructing a physical information neural network taking coordinates of the coordinate points as input and displacement prediction data of the coordinate points as output; the method comprises the following steps: initializing a shear strength parameter of a slope soil body, taking the shear strength parameter as a hyper-parameter of a physical information neural network, constructing a comprehensive loss function, and training the constructed physical information neural network by taking minimization of the comprehensive loss function as a target; and when a preset convergence condition is reached, outputting the shear strength parameter of the side slope soil body. According to the method, the data preparation cost is saved, the displacement field predicted by the mandatory neural network and the stress field derived from the displacement field must meet the yield criterion and the force balance condition of the material, and it is ensured that the shear strength parameter obtained through final inversion is mechanically self-consistent.

Description

Technical Field

[0001] This invention relates to the field of physical and mechanical parameter identification technology for slope soil in geotechnical engineering, specifically to a method and equipment for inverting slope soil shear strength parameters based on PINN (Physical Information Neural Network). Background Technology

[0002] Shear strength parameters of slope soil: cohesion and internal friction angle These parameters are core parameters for evaluating slope stability and for landslide disaster early warning and prevention. However, obtaining these parameters directly through laboratory or field tests is often costly and time-consuming, and is affected by sample disturbance and scale effects, making it difficult to accurately reflect the true mechanical state of the slope soil. In engineering practice, test data alone is generally not directly applicable to design work. Parameter inversion analysis is an effective way to solve this problem and occupies an important position in the research of shear strength parameter identification of slope soil.

[0003] Inversion methods for slope soil shear strength parameters can be divided into inversion methods based on slope stability coefficients and inversion methods based on slope observation responses. Before the development of data-driven methods, inversion was mostly based on slope stability coefficients. Inversion based on stability coefficients is relatively mature, but it is too subjective. The shear strength parameters obtained through inversion are equivalent mechanical calculation parameters for the entire slip surface, representing a comprehensive reflection of the slip surface strength parameters. This method inevitably has some drawbacks: First, it emphasizes the importance of judging the slope's "state," and the empirical judgment of slope stability coefficients is often highly empirical; different slope stability coefficients can lead to significant differences in inversion results. Second, inversion from multiple cross-sections may result in non-unique intersection points. If the parameter values ​​differ greatly between different intersection points, it is necessary to verify the stability coefficients of each cross-section.

[0004] With the development of monitoring and computer technologies, data-driven methods have been introduced into the study of slope soil shear strength parameter inversion, leading to the development of inversion methods based on slope observation responses. The core idea of ​​this type of inversion method can be summarized as follows: by establishing a data-driven meta-model, fitting the relationship between a readily observable variable and the undetermined parameter, and then determining the value of the undetermined parameter. In slope engineering, slope monitoring displacement is one of the most commonly used observation responses for inversion. Inversion based on monitoring displacement can be divided into research that directly or indirectly uses machine learning methods.

[0005] Displacement inversion using machine learning methods directly typically employs backpropagation (BP) neural networks, support vector machines, and other machine learning techniques as tools to fit the linear or nonlinear relationship between displacement and soil shear strength parameters, establishing a corresponding meta-model to invert the soil shear strength parameters. The soil shear strength parameters serve as the output, and the monitored displacement is the input. Displacement inversion using machine learning methods indirectly reverses the input and output of the aforementioned meta-models. In this case, it is generally necessary to construct an objective function representing the difference between the calculated displacement and the actual displacement, and then optimize and solve it using a metaheuristic algorithm. Therefore, regardless of the approach taken in displacement inversion, establishing a data-driven meta-model with strong fitting performance is an essential step. Data-driven methods provide a more objective basis for identifying the physical and mechanical parameters of slope soil and rock. However, this type of method completely ignores the physical and mechanical mechanisms, focusing only on the relationship between soil shear strength parameters and observed responses. The performance of the model depends entirely on the quantity and quality of the training samples. To train a meta-model of the relationship between parameters and observed responses (such as slope displacement), it is usually necessary to construct training samples through multiple finite element analyses. When considering the overall performance of the model, the actual efficiency of the model is significantly reduced if the time spent constructing training samples is included. In addition, the quality of the samples is often difficult to control, and imbalanced training samples can lead to poor model adaptability and failure to reflect the real situation.

[0006] In summary, existing methods for inverting slope soil shear strength parameters have significant limitations: firstly, traditional inversion methods lack sufficient objective evidence; secondly, traditional data-driven methods rely entirely on massive training samples generated through numerical simulations such as finite element methods, making the inversion process a "black box" lacking any physical constraints, leading to mathematically optimal but physically unreasonable results. Therefore, there is an urgent need for an efficient inversion method that can significantly reduce dependence on sample size and ensure that the inversion process conforms to mechanical mechanisms. Summary of the Invention

[0007] This invention addresses the shortcomings of existing technologies by providing a method and equipment for inverting slope soil shear strength parameters based on PINN, thereby solving the problems of insufficient objectivity, high sample dependence, and unclear physical mechanisms in existing technologies.

[0008] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions: The inversion method for slope soil shear strength parameters based on PINN includes the following steps: Step 1: Determine the geometric model of the slope, arrange several monitoring points in the slope, and obtain the monitoring point coordinates and displacement monitoring data of each monitoring point; Step 2: Collect multiple coordinate points evenly within the domain and boundary defined by the geometric model to obtain the coordinates of the coordinate points. The collected coordinate points include all monitoring points. Step 3: Construct a physical information neural network that takes the coordinates of the coordinate points as input and the displacement prediction data of the coordinate points as output; Step 4: Initialize the slope soil shear strength parameters and use them as hyperparameters for the physical information neural network to construct a comprehensive loss function. Comprehensive loss function For data loss items Physical rule loss item and boundary condition loss terms The weighted sum; Step 5: Minimize the overall loss function The constructed physical information neural network is trained with the goal of [target]. Step 6: When the training process reaches the preset convergence condition, the optimized slope soil shear strength parameters are output as the final inversion result.

[0009] As described above, the physical information neural network consists of an input layer, multiple fully connected hidden layers, and an output layer.

[0010] As mentioned above, data loss items Based on the following formula: ; in, and These are the number and serial number of monitoring points, respectively. Indicates the first monitoring points Displacement prediction data With displacement monitoring data The difference.

[0011] As described above, the physical rule loss term Based on the following formula: ; in, and These represent the number and index of coordinate points within the domain and boundary defined by the geometric model, respectively. The divergence of the stress tensor at a single coordinate point. For a single coordinate point, the force vector. For the first coordinate points stress tensor divergence and body force vector The sum of.

[0012] As described above, stress tensor Calculated based on the following steps: Calculate the displacement prediction data for each coordinate point The corresponding displacement gradient tensor , Further calculation of strain tensor for: ; in, Let be the displacement gradient tensor. It is the transpose operator. Further calculation of elastic test stress for: ; in, Let be the elastic stiffness tensor. This is the double dot product operator. Elastic stress test at each coordinate point Perform principal stress decomposition to obtain the corresponding principal stresses. and , Calculate elastic test stress Corresponding constitutive relation and critical yield condition ,like Greater than 0 or If it is greater than 0, then the stress tensor ;like and If all are less than or equal to 0, then the stress tensor , in, Let be the plastic potential function. For plastic multipliers, the plastic potential function Based on the following formula: ; Among them, parameters = , It is the expansion angle.

[0013] As described above, constitutive relations Based on the following formula: ; The critical yield condition Based on the following formula: ; Among them, parameters , and These are the shear strength parameters of the slope soil: cohesion and angle of internal friction. This refers to the tensile strength of the slope soil.

[0014] As described above, the boundary condition loss term Based on the following formula: ; in, and These represent the number and sequence number of coordinate points on the boundary. For the first coordinate points Displacement prediction data output by the physical information neural network and displacement boundary conditions The difference, For the first Stress tensor calculated from displacement prediction data output by a physics-based neural network at each coordinate point. The difference between the stress boundary conditions and the stress boundary conditions.

[0015] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described PINN-based method for inverting slope soil shear strength parameters.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described PINN-based method for inverting slope soil shear strength parameters.

[0017] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described PINN-based method for inverting slope soil shear strength parameters.

[0018] Compared with the prior art, the present invention achieves at least the following beneficial effects: 1) This invention directly embeds the basic physical laws controlling slope deformation into the comprehensive loss function as a strong constraint, enabling the physical information neural network to learn directly from the displacement monitoring data of limited, real field monitoring points. This greatly reduces the dependence on a large number of high-quality simulation samples, saves data preparation costs, and enhances the universality of the method in scenarios lacking complete numerical models.

[0019] 2) This invention employs a physical information neural network for one-time end-to-end training. During training, the soil shear strength parameter is optimized as a hyperparameter along with the network weights. After training, the optimal soil shear strength parameter and network weights can be directly output. This method avoids repeated forward calculations, transforming the complex inversion problem into an optimization problem, thus improving computational efficiency. It is particularly suitable for engineering scenarios requiring rapid evaluation.

[0020] 3) Compared to data-driven inversion methods, this invention introduces the Mohr-Coulomb criterion, the core physical mechanism for slope stability analysis, as a hard constraint into the comprehensive loss function. This forces the displacement field and the derived stress field of the physical information neural network to satisfy the material's yield criterion and force equilibrium conditions. This ensures that the soil shear strength parameters obtained from the final inversion are mechanically self-consistent, and the corresponding stress state can reasonably explain the actual deformation and failure modes of the slope. This fundamentally guarantees the physical credibility and engineering reliability of the results, avoiding the irrationality caused by subjective judgment in inversion methods based on stability coefficients.

[0021] 4) In this invention, the physical information neural network is a complex nonlinear function approximator. Its optimization process is carried out under the strong constraint of physical laws. This guidance helps the search process to jump out of local minima and converge to the global optimal solution that satisfies physical laws, thereby improving the stability and accuracy of the inversion results.

[0022] 5) The method of this invention has no special restrictions on the geometry of the slope and is, in principle, applicable to various complex slope models, such as homogeneous / heterogeneous, two-dimensional / three-dimensional, etc. By adjusting the definition of the boundary condition term in the loss function, various complex boundary constraint problems can be easily handled. In addition, this method relies only on displacement monitoring data and is compatible with data obtained from various on-site monitoring methods (such as GPS, inclinometers, remote sensing measurements, etc.), demonstrating strong engineering applicability potential. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the physical information network structure of the method of the present invention; Figure 3 This is a schematic diagram of the slope geometric model, monitoring points, and sampling coordinates in an embodiment of the present invention, wherein the monitoring points are red dots and the sampling points are black dots; Figure 4 In Embodiment 2 of this invention, the shear strength parameter cohesion of slope soil based on a physical information neural network is... The inversion results diagram; Figure 5 The internal friction angle is a parameter in the slope soil shear strength based on a physical information neural network in Embodiment 2 of this invention. The inversion result diagram. Detailed Implementation

[0024] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these examples do not limit the scope of the invention.

[0025] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way.

[0026] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0027] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0028] Example 1: See Figure 1 As shown, this invention provides a method for inverting the shear strength parameters of slope soil based on PINN, including: Step 1: Determine the geometric model of the slope, arrange several monitoring points in the slope, and obtain the monitoring point coordinates and displacement monitoring data of each monitoring point; Step 2: Collect multiple coordinate points evenly within the domain and boundary defined by the geometric model to obtain the coordinates of the coordinate points. The collected coordinate points must include all monitoring points. Step 3: Construct a physical information neural network that takes the coordinates of the coordinate points as input and the displacement prediction data of the coordinate points as output. The physical information neural network consists of an input layer, multiple fully connected hidden layers and an output layer. Step 4: Initialize the shear strength parameters of the slope soil to be inverted, use them as trainable hyperparameters of the physical information neural network, and construct a comprehensive loss function. The comprehensive loss function include: (1) Data loss items The difference between the displacement prediction data and the displacement monitoring data of each monitoring point predicted by the physical information neural network is used to constrain the difference, and the calculation method is as follows: ; in, and These are the number and serial number of monitoring points, respectively. and These are the predicted displacement data and the monitored displacement data of the monitoring points, respectively. For the first The coordinates of each monitoring point Indicates the first The difference between the predicted displacement data and the monitored displacement data at each monitoring point. The closer the predicted displacement value is to the monitored value, the better. The closer the value is to 0.

[0029] (2) Physical rule loss item Based on the equilibrium equations and the constitutive relations based on the Mohr-Coulomb criterion, a constitutive relation is constructed to constrain the displacement field predicted by the physical information neural network to satisfy the equilibrium equations and the mechanical constitutive relations of the slope soil. The calculation method is as follows: ; in, and These represent the number and index of coordinate points within the domain and boundary defined by the geometric model, respectively. The divergence of the stress tensor at a single coordinate point. This represents the body force vector at a single coordinate point, typically gravity, but can also include seismic forces used in quasi-static analysis. For the first The coordinates of each point For the first coordinate points stress tensor divergence and body force vector The sum. When the equilibrium equations are completely satisfied. The value decreased to 0. It is an L2 norm.

[0030] Stress tensor The calculations must satisfy the constitutive relations based on the Mohr-Coulomb criterion. for: ; in, and Principal stress, Its magnitude is determined by the elastic test stress. Eigenvalue decomposition is performed to obtain the parameters. , and These are the shear strength parameters of the slope soil: cohesion and angle of internal friction; when At that time, shear yielding will occur at a certain point inside the slope.

[0031] Because soil and rock masses have low tensile strength, tensile failure often occurs at the trailing edge of soil slopes. Therefore, in addition to the above formula, the critical yield condition for tensile yielding should also be considered. Critical yield condition The calculation method is as follows: ; in, For the tensile strength of the slope soil, when At this time, the rock and soil mass will undergo tensile yielding; the above constitutive relations assume positive tension and negative compressive force, and consider the non-associated flow law.

[0032] Specifically, stress tensor The calculation uses a return mapping algorithm, and its calculation steps are as follows: For the displacement prediction data of each coordinate point... The displacement gradient tensor is calculated using the self-differentiation technique. The strain tensor is obtained. for: ; in, Let be the displacement gradient tensor. As a transpose operator, the elastic test stress is further calculated based on this. for: ; in, The elastic stiffness tensor is determined by the given elastic modulus and Poisson's ratio. This is a double dot product operator. Then, the elastic stress at each coordinate point is calculated. Perform principal stress decomposition to obtain the corresponding principal stresses. and Then calculate the elastic test stress. Corresponding constitutive relation and critical yield condition ,like or If either of these values ​​is greater than 0, it indicates that the stress state at that coordinate point exceeds the yield surface, and the stress tensor needs to be adjusted. Perform plastic correction; if and If all values ​​are less than or equal to 0, it indicates that the stress tensor is... The stress tensor needs to be calculated. The coordinate points for plastic correction are corrected using the following method: ; in, Let be the plastic potential function, its form is similar to The same, only the formula is the same. Replace with the expansion angle of the slope soil That's all. The plasticity multiplier is determined by satisfying the consistency condition in plasticity theory, specifically: ; Among them, parameters = , It is the expansion angle.

[0033] (3) Boundary condition loss term The displacement field predicted by the physical information neural network is constrained to meet the boundary conditions set by the geometric model, and the calculation method is as follows: ; in, and These represent the number and sequence number of coordinate points on the boundary. The displacement prediction data output by the neural network for the physical information of a single coordinate point on the boundary. The stress tensor is obtained by solving the neural network outputting displacement prediction data based on the physical information of a single coordinate point on the boundary using the return mapping algorithm. and These are the displacement boundary conditions and stress boundary conditions for a single coordinate point on the preset boundary, respectively. For the first on the boundary The coordinates of each point For the first coordinate points Displacement prediction data output by the physical information neural network and displacement boundary conditions The difference, For the first Stress tensor calculated from displacement prediction data output by a physics-based neural network at each coordinate point. The difference between the stress boundary conditions and the boundary conditions. When the boundary conditions are fully satisfied, The value decreased to 0.

[0034] Comprehensive loss function This is expressed as a weighted sum of all loss items: ; in, For data loss items, For physical rule loss terms, This refers to the boundary condition loss term; , , represents the corresponding non-negative weighting coefficient.

[0035] Step 5: Minimize the comprehensive loss function constructed in Step 3. The constructed physical information neural network is trained with the goal of simultaneously optimizing two types of parameters: one type is the internal hyperparameters of the neural network (weights and biases) to ensure that it can flexibly approximate complex displacement fields; the other type is the shear strength parameter, which is also a hyperparameter: the shear strength parameter of the slope soil, cohesion. and internal friction angle Its optimization process is strongly constrained by each loss term, that is... and The value of must ensure that the predicted displacement and stress fields satisfy the constitutive relations, equilibrium equations, and boundary conditions based on the Mohr-Coulomb criterion as much as possible, and the predicted displacement must also match the monitored displacement. During training, the weights, bias parameters, and slope soil shear strength parameters (as hyperparameters) of the physical information neural network are optimized simultaneously, and the gradient of the loss function is calculated using automatic differentiation.

[0036] Step 6: Output Results: When the training process reaches the preset convergence condition, stop the training and output the optimized slope soil shear strength parameters as the final inversion result.

[0037] The convergence criterion for step 6 is: the value of the comprehensive loss function changes less than a threshold in a predetermined number of consecutive iterations, or the number of training iterations reaches a preset upper limit.

[0038] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0039] Example 2: This embodiment, based on the PINN-based method for inverting slope soil shear strength parameters provided in Embodiment 1, provides a specific application example, which is as follows: Step 1: Determine the geometric model of the slope and acquire monitoring data. The geometric model of the slope is determined based on engineering geological survey data. For example... Figure 3 As shown, the location of monitoring points is determined within the domain of the geometric model (including the slope surface, interior, and boundary), and the coordinates and displacement monitoring data of the monitoring points are obtained.

[0040] To verify the accuracy of the PINN-based slope soil shear strength parameter inversion method proposed in this embodiment, finite element simulation can optionally be used to generate the parameters. Figure 3 The actual monitored displacement of the monitoring point corresponding to the geometric model shown.

[0041] Step 2: Uniformly collect multiple coordinate points within the domain (including the boundary) defined by the geometric model to calculate the data loss term, physical rule loss term, and boundary condition loss term. In this embodiment, the sampling interval of the coordinate points is 0.5m×0.5m, that is, one coordinate point is sampled within a region of 0.5m×0.5m.

[0042] Step 3: Construct the Physical Information Neural Network. The input layer of this neural network has two nodes, corresponding to the horizontal and vertical coordinates respectively. The fully connected hidden layers are set to 5 layers, each containing 28, 56, 128, 56, and 28 neurons respectively. The hyperbolic tangent function is used as the activation function. The output layer consists of two nodes, each outputting the predicted displacement data for its corresponding coordinate point, including the predicted horizontal displacement. and vertical predicted displacement .

[0043] Step 4: Initialize parameters and construct the comprehensive loss function. The shear strength parameter of the slope soil to be inverted, i.e., cohesion. and internal friction angle Initialized to cohesion and internal friction angle 100 reasonable guesses (set here) , This is defined as a trainable hyperparameter in a physical information neural network. The purpose of setting 100 sets of reasonable guesses here is to verify the robustness of the proposed method. Optionally, 100 sets of cohesion values ​​are randomly generated using a uniform distribution. and internal friction angle The guessed value. In this embodiment, the body force vectors at each coordinate point only consider gravity. For boundary conditions, the bottom boundary is set to have a horizontal and vertical displacement of 0, the two side boundaries are set to have a horizontal displacement of 0, and the upper boundary is set to have a stress of 0. A data loss term is constructed based on the aforementioned formula. Physical rule loss item and boundary condition loss terms Comprehensive loss function The weighting coefficient is set to... .

[0044] Step 5: Train the physical information neural network and inversion parameters. Here, the Adam optimizer is chosen, taking the coordinates of all collected coordinate points (intra-domain points and boundary points, which include all monitoring points) as input, to minimize the comprehensive loss function. The algorithm is trained with an initial learning rate of 0.001. During training, the optimization algorithm synchronously updates the weights, biases, and shear strength parameter cohesion of the neural network. and internal friction angle The value of is used to calculate the gradient of the comprehensive loss function using automatic differentiation.

[0045] Step 6: Output the inversion results. Optionally, when the number of training iterations reaches a preset maximum of 10,000 steps, the training is considered converged. At this point, the optimized shear strength parameter cohesion is... and internal friction angle The value is output as the final inversion result.

[0046] like Figure 4 and Figure 5 As shown, Figure 4 and Figure 5 These are the shear strength parameters, specifically the cohesion, among the slope soil shear strength parameters based on a physical information neural network in this embodiment of the invention. and internal friction angle The inversion results show that the shear strength parameter cohesion... The mean of the inversion results is 11.97 kPa, and the variance is 0.05 kPa. The absolute value of the relative error between the mean of the inversion results and the theoretical value of 12 kPa is only 0.25%. The shear strength parameter, cohesion... The mean of the inversion results is 20.05° and the variance is 0.08°. The absolute value of the relative error between the mean of the inversion results and the theoretical value of 20° is only 0.4%. Therefore, the inversion results show that the present invention is accurate and effective.

[0047] Example 3: In this embodiment, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0048] Example 4: In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0049] Example 5: In this embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0050] The above description is merely a partial embodiment of the present invention and does not limit the scope of protection of the present invention. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for inverting slope soil shear strength parameters based on PINN, characterized in that, Includes the following steps: Step 1: Determine the geometric model of the slope, arrange several monitoring points in the slope, and obtain the monitoring point coordinates and displacement monitoring data of each monitoring point; Step 2: Collect multiple coordinate points evenly within the domain and boundary defined by the geometric model to obtain the coordinates of the coordinate points. The collected coordinate points include all monitoring points. Step 3: Construct a physical information neural network that takes the coordinates of the coordinate points as input and the displacement prediction data of the coordinate points as output; Step 4: Initialize the slope soil shear strength parameters and use them as hyperparameters for the physical information neural network to construct a comprehensive loss function. Comprehensive loss function For data loss items Physical rule loss item and boundary condition loss terms The weighted sum; Step 5: Minimize the overall loss function The constructed physical information neural network is trained with the goal of [target]. Step 6: When the training process reaches the preset convergence condition, the optimized slope soil shear strength parameters are output as the final inversion result.

2. The method for inverting slope soil shear strength parameters based on PINN according to claim 1, characterized in that, The physical information neural network consists of an input layer, multiple fully connected hidden layers, and an output layer.

3. The method for inverting slope soil shear strength parameters based on PINN according to claim 1, characterized in that, The data loss item Based on the following formula: ； in, and These are the number and serial number of monitoring points, respectively. Indicates the first monitoring points Displacement prediction data With displacement monitoring data The difference.

4. The method for inverting slope soil shear strength parameters based on PINN according to claim 1, characterized in that, The physical rule loss item Based on the following formula: ； in, and These represent the number and index of coordinate points within the domain and boundary defined by the geometric model, respectively. The divergence of the stress tensor at a single coordinate point. For a single coordinate point, the force vector. For the first coordinate points stress tensor divergence and body force vector The sum of.

5. The method for inverting slope soil shear strength parameters based on PINN according to claim 4, characterized in that, The stress tensor Calculated based on the following steps: Calculate the displacement prediction data for each coordinate point The corresponding displacement gradient tensor , Further calculation of strain tensor for: ； in, Let be the displacement gradient tensor. It is the transpose operator. Further calculation of elastic test stress for: ； in, Let be the elastic stiffness tensor. This is the double dot product operator. Elastic stress test at each coordinate point Perform principal stress decomposition to obtain the corresponding principal stresses. and , Calculate elastic test stress Corresponding constitutive relation and critical yield condition ,like Greater than 0 or If it is greater than 0, then the stress tensor ;like and If all are less than or equal to 0, then the stress tensor , in, Let be the plastic potential function. For plastic multipliers, the plastic potential function Based on the following formula: ； Among them, parameters = , It is the expansion angle.

6. The method for inverting slope soil shear strength parameters based on PINN according to claim 5, characterized in that, The constitutive relation Based on the following formula: ； The critical yield condition Based on the following formula: ； Among them, parameters , and These are the shear strength parameters of the slope soil: cohesion and angle of internal friction. This refers to the tensile strength of the slope soil.

7. The method for inverting slope soil shear strength parameters based on PINN according to claim 1, characterized in that, The boundary condition loss term Based on the following formula: ； in, and These represent the number and sequence number of coordinate points on the boundary. For the first coordinate points Displacement prediction data output by the physical information neural network and displacement boundary conditions The difference, For the first Stress tensor calculated from displacement prediction data output by a physics-based neural network at each coordinate point. The difference between the stress boundary conditions and the stress boundary conditions.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for inverting slope soil shear strength parameters based on PINN as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for inverting the shear strength parameters of slope soil based on PINN as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the PINN-based method for inverting slope soil shear strength parameters as described in any one of claims 1 to 7.

Images