Construction method and system of rock-soil shear strength parameter prediction model

Abstract

The invention provides a construction method and system of a rock-soil shear strength parameter prediction model, and relates to the technical field of model construction.The construction method comprises the steps that CT scanning data of a real rock-soil sample is obtained and preprocessed, and an irregular three-dimensional particle set for geometric modeling is obtained; performing particle surface roughness modeling analysis on the three-dimensional particle body set to obtain a surface roughness coefficient set of the three-dimensional particle body set; performing particle size cumulative distribution function segmentation processing and fractal dimension control on the surface roughness coefficient set to obtain a structure grading sample set; constructing a THMC multi-field coupling model based on the structure grading sample set to obtain a working condition particulate matter physical response parameter set; and performing a simulation experiment on the working condition particle physical response parameter set, and performing deep neural network modeling processing based on an experiment result to obtain a strength response model capable of predicting the cohesive force and the internal friction angle of the rock-soil body under any working condition input. According to the invention, the prediction efficiency and accuracy are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of model building technology, and more specifically, to a method and system for constructing a model for predicting the shear strength parameters of soil and rock. Background Technology

[0002] Currently, obtaining shear strength parameters (such as cohesion and internal friction angle) of geotechnical materials in geotechnical engineering still mainly relies on numerous laboratory tests and field tests. While these traditional testing methods can provide relatively reliable strength indicators, they are cumbersome, time-consuming, and costly, and are often limited by sample representativeness and test repeatability, making it difficult to meet the rapid prediction needs of large-scale projects or complex working conditions. In recent years, some studies have attempted to estimate shear strength using empirical formulas, statistical regression, or shallow machine learning methods. However, these methods generally suffer from weak modeling capabilities and insufficient feature representation, especially when dealing with the effects of multi-physics coupling or complex morphological parameters at the particle scale, where prediction accuracy and generalization ability are significantly reduced. Furthermore, existing modeling methods often employ the assumption of regular particles and idealized gradation simulation, neglecting the true geometric morphology of geotechnical materials and the deep coupling mechanism between them and physical responses. This makes it impossible to accurately predict shear strength under arbitrary microscopic working conditions, limiting their further application in constitutive model inversion, structural stability analysis, and engineering numerical simulation.

[0003] Therefore, there is an urgent need for a method and system for constructing a prediction model for soil and rock shear strength parameters to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for constructing a prediction model for soil and rock shear strength parameters, thereby improving the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a method for constructing a prediction model for soil and rock shear strength parameters, including:

[0006] CT scan data of real soil and rock samples were acquired and preprocessed to obtain an irregular three-dimensional particle set for geometric modeling.

[0007] The surface roughness of the three-dimensional particle set is modeled and analyzed to obtain the surface roughness coefficient set of the three-dimensional particle set.

[0008] The surface roughness coefficient set is processed by segmentation of the cumulative particle size distribution function and fractal dimension control to obtain the structural gradation sample set;

[0009] Based on the aforementioned structural gradation sample set, a THMC multi-field coupling model was constructed, and the multi-physics response equations were solved under different heat conduction, pore flow and chemical corrosion conditions to obtain the set of particle physical response parameters under working conditions.

[0010] The set of particle physical response parameters under the aforementioned working conditions was simulated, and deep neural network modeling was performed based on the experimental results to obtain a strength response model that can predict the cohesion and internal friction angle of soil and rock under any working condition input.

[0011] Secondly, this application also provides a system for constructing a prediction model for soil and rock shear strength parameters, including:

[0012] The acquisition unit is used to acquire CT scan data of real soil and rock samples and preprocess them to obtain an irregular three-dimensional particle set for geometric modeling.

[0013] Analysis unit is used to perform particle surface roughness modeling and analysis on the three-dimensional particle body set to obtain the surface roughness coefficient set of the three-dimensional particle body set;

[0014] The processing unit is used to perform piecewise processing of the surface roughness coefficient set by the particle size cumulative distribution function and fractal dimension control to obtain the structural gradation sample set;

[0015] The computing unit is used to construct a THMC multi-field coupling model based on the structure gradation sample set, and solve the multi-physics response equation set under different heat conduction, pore flow and chemical corrosion conditions to obtain the set of particle physical response parameters under working conditions.

[0016] The modeling unit is used to simulate the set of particle physical response parameters under the working conditions, and to perform deep neural network modeling based on the experimental results to obtain a strength response model that can predict the cohesion and internal friction angle of the soil and rock under any working condition input.

[0017] The beneficial effects of this invention are as follows:

[0018] As is understood, this invention first reconstructs an irregular three-dimensional particle mass set from real soil and rock CT scan data, and extracts roughness and particle size distribution information based on fractal modeling to construct a structural gradation sample set. Then, by establishing a multiphysics coupling model, it simulates the particle response behavior under different heat conduction, pore flow, and chemical corrosion conditions. Next, it conducts triaxial compression numerical simulation experiments on a PFC platform to obtain shear strength parameters. Finally, it utilizes a deep neural network constructed with multilayer nonlinear activation functions to achieve a high-precision mapping between soil and rock micro-parameters and shear strength. This method not only significantly improves prediction efficiency and accuracy but also enables rapid response under complex working conditions.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the construction method of the soil and rock shear strength parameter prediction model described in the embodiments of the present invention;

[0022] Figure 2 This is a schematic diagram of the system structure for constructing the geotechnical shear strength parameter prediction model as described in this embodiment of the invention.

[0023] In the diagram: 701, acquisition unit; 702, analysis unit; 703, processing unit; 704, calculation unit; 705, modeling unit. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. 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 claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a method for constructing a prediction model for soil and rock shear strength parameters.

[0028] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0029] Step S1: Obtain CT scan data of real soil and rock samples and preprocess them to obtain an irregular three-dimensional particle set for geometric modeling;

[0030] Understandably, this step aims to reconstruct the true structure of soil and rock materials at the microscopic particle scale by acquiring high-resolution CT scan data of real soil and rock samples. This step overcomes the limitations of traditional modeling using approximate regular spherical or ellipsoidal particles, ensuring the authenticity and representativeness of the input data for the prediction model, and fundamentally improving the accuracy and reliability of subsequent mechanical response simulation and neural network training. In this step, step S1 includes steps S11, S12, and S13.

[0031] Step S11: Data acquisition and processing are performed based on the CT scan results of real soil and rock samples. The three-dimensional point cloud coordinate set is obtained by reconstructing the sample layer by layer with a resolution of 1μm.

[0032] It is understandable that the formula for generating the 3D coordinate set in this step is as follows:

[0033]

[0034] Where P is the set of three-dimensional point cloud coordinates, representing the actual particle surface coordinate data generated after CT scanning, x i ,y i ,z i Let be the 3D coordinates of the i-th point, and N be the total number of points in the point cloud data.

[0035] Step S12: Perform noise removal processing on the three-dimensional point cloud coordinate set according to the median filtering algorithm to obtain high-quality point cloud data after noise reduction;

[0036] It is understandable that this step uses a median filtering algorithm to remove outliers, eliminate isolated outliers, and fill in missing structures, thereby obtaining point cloud data with clear shapes and complete boundaries. The specific formula is shown below:

[0037] z filtered (x,y)=median({z|x+k,y+l|k,l∈[-w,w]})

[0038] Among them, z filtered(x,y) represents the height data after median filtering, w is the radius of the filtering window, k,l are the offsets relative to the center point (x,y) within the filtering window, where x is the x-coordinate of the center point, y is the y-coordinate of the center point, and z is the unprocessed height data.

[0039] Step S13: Perform surface reconstruction processing on the denoised point cloud data. Specifically, a local smooth surface is constructed by using the moving least squares method and Delaunay triangulation is performed globally to obtain an irregular three-dimensional particle set for geometric modeling.

[0040] Understandably, this step first constructs a smooth surface S(x,y), and then minimizes the local weighted squared error, as shown in the following formula:

[0041]

[0042] Where a is the coefficient vector to be optimized, w i (x,y) represents the weight function value at the i-th point, φ(x) i ,y i Let z be the basis function vector. i Let T be the original height value of the i-th point, and T be the transpose symbol.

[0043] The basis functions are shown below:

[0044] φ(x,y)=[1,x,y,x 2 ,xy,y 2 ] T

[0045] Where φ(x,y) is the basis function vector, x is the abscissa of a point on the surface, y is the ordinate of a point on the surface, and T is the transpose sign.

[0046] The weighting function is shown below:

[0047]

[0048] Among them, w i (x, y) represents the weight function value at the i-th point, h is the smoothing bandwidth, y is the ordinate of the point on the surface, and x is the abscissa of the point on the surface. i The x-coordinate of the center, y i The ordinate of the center point.

[0049] This step utilizes Delaunay triangulation to convert the smoothed point cloud into irregular polyhedral particles, each composed of 4-8 triangular pieces, achieving high-fidelity reconstruction of irregular geometric shapes. This processing not only preserves the original geometric complexity of the soil particles but also provides physically realistic and structurally complete input data for subsequent particle roughness modeling and multi-field coupled response simulation.

[0050] Step S2: Perform particle surface roughness modeling and analysis on the three-dimensional particle body set to obtain the surface roughness coefficient set of the three-dimensional particle body set;

[0051] It is understandable that this step accurately captures and quantifies the true roughness characteristics of the particle surface, making up for the error caused by the assumption of ideal smoothness of particles in the traditional model, so that the subsequent modeling is closer to the actual mechanical behavior of rock and soil materials. In this step, step S2 includes steps S21, S22 and S23.

[0052] Step S21: Extract the surface height of the irregular three-dimensional particle set to obtain a particle surface height dataset;

[0053] Understandably, this step first extracts the surface height data of each three-dimensional particle through a geometric processing algorithm. This step is based on the spatial normal vector to expand the local mesh or construct the normal projection profile, thereby obtaining a discrete dataset of the height of each point on the particle surface relative to the reference surface.

[0054] Step S22: Perform fractal function modeling on the particle surface height dataset. By introducing the Weierstrass-Mandelbrot function to superimpose multi-scale spatial frequency terms and setting the number of fractal cones and random phase angle, the discrete expression of the height corresponding to each particle is obtained.

[0055] Understandably, this step introduces the Weierstrass-Mandelbrot fractal function from fractal theory to perform multi-scale modeling of surface height data. This function allows for the superposition of multiple frequency terms to simulate the multi-level irregularities of the particle surface. Simultaneously, setting different fractal cone numbers and phase angles ensures the model's ability to fit particle diversity and natural variability. The discrete expression for the height of each particle is shown below:

[0056]

[0057] in, (A is the amplitude coefficient, β = 2.5 is the fractal cone number), f k For spatial frequency, f k =f min ·2 k / N ,φk ,ψ k ∈[0,2π) represents a random phase angle, N represents the total number of particles, and k represents the kth particle.

[0058] Step S23: Perform roughness statistical analysis based on the height function expression corresponding to each particle, and use the root mean square value of the height distribution on the surface of each particle as the set of surface roughness coefficients corresponding to each three-dimensional particle.

[0059] It is understandable that this step will use the surface roughness coefficient R q Defined as the root mean square value of the height distribution, and after scanning a sufficient number of samples, the surface roughness distribution is statistically analyzed to classify particles of a preset number of roughnesses.

[0060] The formula is as follows:

[0061]

[0062] Among them, R q This is the root mean square value of the surface roughness. The average of all height values, where N is the total number of particles, and z is the average of all height values. i This is the original height value of the i-th point.

[0063] Step S3: Perform piecewise processing of the particle size cumulative distribution function and fractal dimension control on the surface roughness coefficient set to obtain the structural gradation sample set;

[0064] Understandably, this step achieves an organic mapping and fractal control from microscopic roughness properties to macroscopic structural gradation, significantly improving the representativeness and distribution diversity of the simulation samples, and ensuring that subsequent modeling can cover the impact of structural changes in soil and rock materials on mechanical properties under complex working conditions. In this step, step S3 includes steps S31, S32, and S33.

[0065] Step S31: Classify the surface roughness coefficient set by roughness distribution, divide the distribution range of the root mean square value of roughness by statistical analysis, divide the surface roughness level into a preset number of types, and construct the mapping relationship between particle surface properties and particle size range to obtain the initial particle size range division result corresponding to the roughness level.

[0066] Understandably, this step involves statistically analyzing the root mean square roughness values ​​of all particles to construct their probability density or cumulative distribution curves and identify key distribution intervals and extreme points. Subsequently, based on the natural discrete characteristics of particles in surface roughness, methods such as equal-frequency grouping, K-means clustering, or Gaussian mixture models are used to classify the roughness values ​​into a predetermined number of roughness levels. Next, combining domain experience and experimental data, the system further analyzes the correlation between rough particles and their actual particle size—generally, rougher particles have larger particle sizes, and vice versa. Therefore, within each roughness level, a multi-level mapping model is constructed by jointly analyzing roughness and particle size corresponding samples from a historical soil and rock test library, achieving a preliminary correlation from surface properties to particle size ranges. This process not only preserves the detailed information of the original roughness in particle-level modeling but also provides particle size range boundary conditions for subsequent fractal control.

[0067] Step S32: Based on the fractal cumulative distribution function, the initial interval division result of particle size is segmented to obtain a multi-level particle size combination sample set that meets the target fractal dimension control requirements;

[0068] It is understandable that the cumulative particle size distribution function in this step is:

[0069]

[0070] Where P(d) is the cumulative particle size distribution function, d max For the maximum particle size, D f is the fractal dimension, between 2.2 and 2.8, and d is the calculated particle size.

[0071] In this step, the particles are discretized into multiple particle size ranges, and the number of particles in the i-th range is:

[0072] N i =N total ·[P(d i )-P(d i-1 )]

[0073] Where, N i N represents the number of particles in the i-th particle size range. total P(d) represents the total number of particles. i Let P(d) be the cumulative particle size distribution function for the i-th interval. i-1 Let d be the cumulative particle size distribution function for the (i-1)th interval. i Let d be the upper limit of the i-th particle size range. i-1 This is the upper limit value of the (i-1)th particle size range.

[0074] Step S33: Generate multiple proportioning schemes based on a multi-level particle size combination sample set with different particle size and roughness levels to obtain a structural gradation sample set.

[0075] It is understandable that in order to improve sample diversity and cover a wider range of working conditions, this step will generate multiple formulation schemes, in which each level is mapped to multiple particle sizes, thereby generating multiple schemes.

[0076] Step S4: Construct a THMC multi-field coupling model based on the structural gradation sample set, and solve the multi-physics response equation set under different heat conduction, pore flow and chemical corrosion conditions to obtain the set of particle physical response parameters under working conditions;

[0077] It is understandable that this step, by constructing a highly coupled and multi-field interconnected physical model, achieves a full-process prediction of the response characteristics of soil particles under complex environments. This provides representative and physically reasonable input data for the subsequent construction of an intensity prediction model, greatly enhancing the model's practicality and theoretical support. In this step, step S4 includes steps S41, S42, and S43.

[0078] Step S41: Perform multi-field coupling model initialization processing on the structural gradation sample set to obtain a multi-field input configuration set under coupled state;

[0079] Understandably, during the initialization process of this step, the system also needs to invoke the coupling coordination mechanism to ensure that the boundary conditions of each physical field are interoperable and consistent under a unified time domain and spatial resolution. This step, by systematically constructing a multi-field input configuration set under coupled conditions, not only significantly improves the coverage and field coupling accuracy of subsequent simulation conditions, but also ensures the repeatability of simulation results and the physical rationality of boundary conditions, laying a crucial foundation for obtaining a highly reliable set of particle response parameters.

[0080] Step S42: Solve the multi-physics response equation system for the multi-field input configuration set. The multi-physics response equation system is solved by using preset working conditions to obtain the state dataset of each physical field.

[0081] It is understandable that the equations in this step include the heat conduction equation:

[0082]

[0083] Q fric =μ·σ n ·|v slip |

[0084] Where ρ is the material density, C P Let Q be the specific heat capacity, T be the temperature field, k be the thermal conductivity, and Q be the thermal conductivity. fricThe frictional heat generation power is μ, the coefficient of friction is σ n For normal stress, v slip Let t be the interparticle slip velocity, and t be time.

[0085] The formula for dynamically adjusting particle size through the thermal expansion effect is as follows:

[0086] d e =d0[1+α T (T-T0)]

[0087] Where, d e d0 is the particle size after thermal expansion, T0 is the reference temperature, T is the actual temperature, and α is the particle size after thermal expansion. T is the coefficient of thermal expansion of the material.

[0088] It also includes Darcy's fluid pressure equation:

[0089]

[0090] Where φ is porosity, ρ f v is the fluid density. f Let be the fluid velocity, k1 be the permeability, μ1 be the fluid dynamic viscosity, p be the fluid pressure, and t be the time.

[0091] The particle-fluid interaction force can be solved using the following formula:

[0092] F drag =β(v f -v p )

[0093]

[0094] Among them, F drag β1 is the drag force of the fluid on the particles, and β1 is the drag coefficient, which comprehensively reflects the interaction between the fluid and the particles. f v is the fluid velocity. p Let d be the particle velocity. p Where φ is the particle diameter, φ is the porosity, and ρ is the particle diameter. f ρ is the fluid density, and μ1 is the fluid dynamic viscosity.

[0095] The kinetic equation for the cement-dissolution bond in this step is as follows:

[0096]

[0097] σ c (t)=σ c0 exp(-k c pH·t)

[0098] Where, σc (t) represents the bond strength at time t, where t is time and k is the bond strength. c σ is the cementation-dissolution rate constant, pH is the acidity / alkalinity, and σ is the cementation-dissolution rate constant. c0 This represents the initial bond strength.

[0099] The ion migration model formula is shown below:

[0100]

[0101] Among them, c i Let D be the concentration of the i-th ion, t be time, and D be the concentration of the ion. i Z is the diffusion coefficient. i The ionic charge number, μ i φ1 represents the ion mobility and φ1 represents the electric potential.

[0102] Step S43: Extract particle physical response parameters from the state datasets of all physical fields to obtain a set of physical response parameters with working condition discrimination.

[0103] Understandably, this step extracts a set of physical response parameters from the large-scale state data output from various physical field simulations. These parameters accurately reflect the response characteristics of soil particles under specific working conditions. The resulting parameter set not only carries strong physical coupling characteristics but also possesses clear working condition labels, providing a well-structured and clearly labeled data foundation for subsequent deep learning prediction models. Specifically, by combining the differences in particle size and roughness levels in structural gradation, a spatial partitioning mapping method is used to establish a correlation between local responses and particle properties, thereby extracting response parameters representative of subsequent changes in mechanical properties, such as the rate of change of thermal expansion coefficient, effective stress tensor field, slope of porosity evolution curve, and cumulative local shear strain.

[0104] Step S5: Conduct simulation experiments on the set of particle physical response parameters under the working conditions, and perform deep neural network modeling based on the experimental results to obtain a strength response model that can predict the cohesion and internal friction angle of soil and rock under any working condition input.

[0105] It is understood that in this step, step S5 includes steps S51, S52, S53 and S54.

[0106] Step S51: Perform sample modeling processing on the set of particle physical response parameters under the above working conditions to obtain a set of numerical samples that can be used for triaxial compression simulation.

[0107] Understandably, this step reconstructs the particle aggregation geometry and contact network structure using a parametric generation algorithm based on information such as particle size distribution, roughness level, contact stiffness, porosity, and thermal expansion coefficient in each set of physical response parameters. Subsequently, the interparticle interaction model is set to a nonlinear constitutive relationship that includes temperature dependence, pore pressure influence, and corrosion degradation effects through PFC (Particle Flow Code), ensuring that the particle behavior during the simulation can realistically reflect the extracted coupled response characteristics. The resulting numerical specimen not only inherits the original characteristics of the structural gradation and physical parameters but also possesses clear working condition labels and loadability, enabling it to be directly used for subsequent mechanical response simulations such as triaxial compression.

[0108] Step S52: Apply load conditions to the numerical sample set using PFC software to obtain the peak principal stress and corresponding confining pressure dataset under the load conditions.

[0109] It is understood that in this step, the numerical sample set is assigned to the cementation model (parallel cementation model). Two confining pressures are designed for each test and called separately to obtain stress-strain curves under different confining pressures. Finally, the shear strength parameters of the soil and rock mass under this working condition are calculated based on the relationship between the stress peak and the confining pressure.

[0110] Step S53: Calculate the shear strength parameters of soil and rock under all working conditions based on Python programming to obtain the shear strength parameters of soil and rock under each group of working conditions. The shear strength parameters of soil and rock include cohesion and internal friction angle.

[0111] Understandably, this step first reads the peak principal stress and confining pressure datasets derived from numerical simulations (such as PFC), and then builds a data processing flow in the Python environment to obtain the cohesion and internal friction angle corresponding to each set of data.

[0112] Step S54: Based on the shear strength parameters of soil and rock under each working condition, perform deep neural network modeling and processing, and perform model training and parameter optimization based on the model structure obtained by modeling to obtain a strength response model that can predict the cohesion and internal friction angle of soil and rock under any working condition input.

[0113] Understandably, in this step, a deep feedforward neural network trained based on the backpropagation algorithm is established: the input layer contains 10 key elements, namely particle size, effective modulus, friction coefficient, normal bonding strength, tangential bonding strength, bonding friction angle, heat transfer, seepage, and chemical reaction. The hidden layer adopts a 7-layer hidden layer architecture, and the number of neurons in each layer decreases exponentially, from 512 to 8, in order to achieve deep feature extraction and complex relationship mining of the data.

[0114] First hidden layer (512 neurons): GELU activation function is used, the formula is:

[0115] GELU(x)=xΦ(x)

[0116] Where GELU(x) is the activation function value with x as the neuron input value, x is the neuron input value, and Φ(x) is the cumulative distribution function of the standard normal distribution.

[0117] The cumulative distribution function of the standard normal distribution is shown below:

[0118]

[0119] Where Φ(x) is the cumulative distribution function of the standard normal distribution, t is the integral dummy variable with no actual physical meaning, e is the natural constant, and π is pi.

[0120] This function adaptively adjusts the activation level of neurons based on the probability distribution of the input data, which can efficiently capture complex features in the data and enhance the nonlinear expressive power of the network.

[0121] The second hidden layer (256 neurons) uses the Swish-Max activation function, as shown below.

[0122] Swish-Max(x)=max(0,x·σ(x))

[0123] Where Swish-Max(x) is the Swish-Max activation function, σ(x) is the Sigmoid function, and x is the neuron input value.

[0124] in:

[0125]

[0126] Where σ(x) is the Sigmoid function, x is the neuron input value, and e is the natural constant.

[0127] This function combines the adaptability of the Swish function with the non-linear filtering characteristics of the ReLU function, which can further enhance the network's ability to filter and combine important features.

[0128] The third hidden layer (128 neurons): uses the Mish-Leaky activation function, which is shown below:

[0129]

[0130] Where x is the neuron input value, α is the negative half-axis slope coefficient, tanh() is the hyperbolic tangent function, ln() is the natural logarithm, and e is the natural constant.

[0131] In this step, α = 0.01. The Mish-Leaky activation function combines the smoothness of the Mish function with the LeakyReLU function's ability to handle negative data, effectively avoiding the "death" problem of neurons and improving the network's learning effect on data features of different polarities.

[0132] The fourth hidden layer (64 neurons): uses the ELU (Exponential Linear Unit) activation function, which is shown below:

[0133]

[0134] Where x is the neuron input value, α1 is the negative half-axis scaling factor, and e is the natural constant.

[0135] Typically, a = 1 is chosen. This function enables the network to have a non-zero gradient on the negative half-axis, accelerating training convergence, while maintaining linearity on the positive half-axis to ensure effective transfer of feature information.

[0136] Fifth hidden layer (32 neurons): Introduces the Softplus activation function, as shown below:

[0137]

[0138] Where x is the neuron input value, β is the smoothness parameter, ln() is the natural logarithm, and e is the natural constant.

[0139] In this step, β = 1 is usually chosen to smoothly map the input data to the positive range. This method performs well when dealing with data features that have non-negative characteristics and approximates the ReLU function near the origin, which helps to alleviate the gradient vanishing problem.

[0140] The sixth hidden layer (16 neurons) uses the SiLU (Sigmoid-weighted Linear Unit) activation function, which is shown below:

[0141]

[0142] Where x is the neuron input value and e is a natural constant.

[0143] In this step, the function uses the Sigmoid function to weight the input, enhancing the network's nonlinear transformation capability and improving the model's fitting effect on complex data.

[0144] The seventh hidden layer (8 neurons): uses the ReLU6 activation function, which is shown below:

[0145] ReLU6(x) = min(max(0,x),6)

[0146] Where x is the neuron input value, max() is the maximum value operator, and min() is the minimum value operator.

[0147] In this step, the function can effectively suppress the overactivation of neurons, thereby improving the stability and generalization ability of the network.

[0148] Output layer: The output layer uses the internal friction angle and cohesion of the soil and rock mass as output indicators, and the number of neurons is set to 2.

[0149] In network training optimization, an Elastic-Net Loss Function combining L1 and L2 regularization and the Adafactor optimization algorithm are adopted.

[0150] The formula for the loss function of an elastic network combining L1 and L2 regularization is:

[0151]

[0152] Where L is the total loss value, n is the sample size, and y i It is the actual value. These are the predicted values, where p is the number of network parameters, and w is the number of network parameters. j It is the wth j There are 3 parameters, where λ1 and λ2 are the weight coefficients for L1 and L2 regularization, respectively.

[0153] In this step, the Adafactor optimization algorithm is used for updating. The Adafactor optimization algorithm update process is as follows:

[0154] Calculate the gradient:

[0155] Where θ represents the network parameters, and J(θ) is the loss function.

[0156] Calculate the second moment estimate (variance):

[0157] ρ is the attenuation rate, which is usually taken as 0.95.

[0158] Calculate the adaptive learning rate:

[0159]

[0160] η tThe adaptive learning rate is η, the initial learning rate is ∈, a small constant to prevent the denominator from being zero is ρ, the decay rate is t, and the number of steps is t.

[0161] Update parameters:

[0162]

[0163] θ t+1 For the parameter vector of the next step, θ t Let η be the parameter vector for this operation. t For adaptive learning rate, To calculate the gradient.

[0164] Example 2:

[0165] like Figure 2 As shown, this embodiment provides a system for constructing a prediction model for soil and rock shear strength parameters. See [link to documentation]. Figure 2 The system includes an acquisition unit 701, an analysis unit 702, a processing unit 703, a calculation unit 704, and a modeling unit 705.

[0166] The acquisition unit 701 is used to acquire CT scan data of real soil and rock samples and preprocess them to obtain an irregular three-dimensional particle set for geometric modeling.

[0167] Analysis unit 702 is used to perform particle surface roughness modeling analysis on the three-dimensional particle body set to obtain the surface roughness coefficient set of the three-dimensional particle body set;

[0168] Processing unit 703 is used to perform piecewise processing of the surface roughness coefficient set by particle size cumulative distribution function and fractal dimension control to obtain a structural gradation sample set;

[0169] The computing unit 704 is used to construct a THMC multi-field coupling model based on the structure gradation sample set, and solve the multi-physics response equation set under different heat conduction, pore flow and chemical corrosion conditions to obtain the set of particle physical response parameters under working conditions.

[0170] Modeling unit 705 is used to simulate the set of particle physical response parameters under the working conditions, and perform deep neural network modeling based on the experimental results to obtain a strength response model that can predict the cohesion and internal friction angle of soil and rock under any working condition input.

[0171] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0172] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0173] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included 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 constructing a predictive model for shear strength parameters of soil and rock, characterized in that, include: CT scan data of real soil and rock samples were acquired and preprocessed to obtain an irregular three-dimensional particle set for geometric modeling. The surface roughness of the three-dimensional particle set is modeled and analyzed to obtain the surface roughness coefficient set of the three-dimensional particle set. The surface roughness coefficient set is processed by segmentation of the cumulative particle size distribution function and fractal dimension control to obtain the structural gradation sample set; Based on the structural gradation sample set, a THMC multi-field coupling model was constructed, and the multi-physics response equations were solved under different heat conduction, pore flow and chemical corrosion conditions to obtain the set of particle physical response parameters under working conditions. The set of particle physical response parameters under the aforementioned working conditions was simulated, and deep neural network modeling was performed based on the experimental results to obtain a strength response model that can predict the cohesion and internal friction angle of soil and rock under any working condition input.

2. The method for constructing a prediction model for soil and rock shear strength parameters according to claim 1, characterized in that... Obtain CT scan data of real soil and rock samples and preprocess them, including: Data acquisition and processing were performed based on the CT scan results of real soil and rock samples. The three-dimensional point cloud coordinate set was obtained by reconstructing the samples layer by layer with a resolution of 1μm. The noise removal process is performed on the three-dimensional point cloud coordinate set according to the median filtering algorithm to obtain high-quality point cloud data after noise reduction. The denoised point cloud data is then subjected to surface reconstruction. This involves constructing a locally smooth surface using the moving least squares method and performing Delaunay triangulation globally to obtain a set of irregular 3D particles for geometric modeling.

3. The method for constructing a prediction model for soil and rock shear strength parameters according to claim 1, characterized in that... The surface roughness of the three-dimensional particle set is modeled and analyzed to obtain a set of surface roughness coefficients for the three-dimensional particle set, including: The surface height of the irregular three-dimensional particle set is extracted to obtain a particle surface height dataset. The particle surface height dataset is modeled using fractal functions. By introducing the Weierstrass-Mandelbrot function and superimposing multi-scale spatial frequency terms, and setting the number of fractal cones and random phase angles, a discrete expression for the height of each particle is obtained. Roughness statistical analysis is performed based on the height function expression corresponding to each particle. The root mean square value of the height distribution on the surface of each particle is calculated as the set of surface roughness coefficients corresponding to each three-dimensional particle.

4. The method for constructing a prediction model for soil and rock shear strength parameters according to claim 1, characterized in that... The surface roughness coefficient set is subjected to piecewise processing of the cumulative particle size distribution function and fractal dimension control, including: The surface roughness coefficient set is classified according to its roughness distribution. By statistically analyzing the distribution range of the root mean square value of roughness, a preset number of surface roughness levels are divided, and a mapping relationship between particle surface properties and particle size range is constructed to obtain the initial particle size range division result corresponding to the roughness level. The initial interval division result of particle size is segmented based on the fractal cumulative distribution function to obtain a multi-level particle size combination sample set that meets the target fractal dimension control requirements. Multiple proportioning schemes are generated based on a multi-level particle size combination sample set with different particle size and roughness levels, resulting in a structural gradation sample set.

5. The method for constructing a prediction model for soil and rock shear strength parameters according to claim 1, characterized in that... Based on the aforementioned structural gradation sample set, a THMC multi-field coupled model was constructed, and the multiphysics response equations were solved under different conditions of thermal conduction, pore flow, and chemical corrosion, including: The structural gradation sample set is subjected to multi-field coupling model initialization processing to obtain a multi-field input configuration set under coupled state; The multi-physics response equations are solved using a set of multiple input configurations. Specifically, the multi-physics response equations are solved using preset operating parameters to obtain the state dataset for each physics field. The state datasets of all physical fields are processed to extract particle physical response parameters, resulting in a set of physical response parameters that are distinguishable by operating conditions.

6. A system for constructing a predictive model for soil and rock shear strength parameters, characterized in that, include: The acquisition unit is used to acquire CT scan data of real soil and rock samples and preprocess them to obtain an irregular three-dimensional particle set for geometric modeling. Analysis unit is used to perform particle surface roughness modeling and analysis on the three-dimensional particle body set to obtain the surface roughness coefficient set of the three-dimensional particle body set; The processing unit is used to perform piecewise processing of the surface roughness coefficient set by the particle size cumulative distribution function and fractal dimension control to obtain the structural gradation sample set; The computing unit is used to construct a THMC multi-field coupling model based on the structure gradation sample set, and solve the multi-physics response equation set under different heat conduction, pore flow and chemical corrosion conditions to obtain the set of particle physical response parameters under working conditions. The modeling unit is used to simulate the set of particle physical response parameters under the working conditions, and to perform deep neural network modeling based on the experimental results to obtain a strength response model that can predict the cohesion and internal friction angle of the soil and rock under any working condition input.

7. The system for constructing a prediction model for soil and rock shear strength parameters according to claim 6, characterized in that, The acquisition unit includes: The first acquisition subunit is used to acquire and process data based on the CT scan results of real soil and rock samples. By reconstructing the sample layer by layer with a resolution of 1μm, a three-dimensional point cloud coordinate set is obtained. The second acquisition subunit is used to perform noise removal processing on the three-dimensional point cloud coordinate set according to the median filtering algorithm to obtain high-quality point cloud data after noise reduction. The third acquisition subunit is used to perform surface reconstruction processing on the denoised point cloud data. Specifically, by using the moving least squares method to construct a local smooth surface and implementing Delaunay triangulation in the global scope, an irregular three-dimensional particle set for geometric modeling is obtained.

8. The system for constructing a prediction model for soil and rock shear strength parameters according to claim 6, characterized in that, The analysis unit includes: The first analysis subunit is used to perform surface height extraction processing on the irregular three-dimensional particle body set to obtain a particle surface height dataset. The second analysis subunit is used to perform fractal function modeling on the particle surface height dataset. By introducing the Weierstrass-Mandelbrot function to superimpose multi-scale spatial frequency terms and setting the number of fractal cones and random phase angle, the discrete expression of the height corresponding to each particle is obtained. The third analysis subunit is used to perform roughness statistical analysis based on the height function expression corresponding to each particle. The root mean square value of the height distribution on the surface of each particle is calculated as the set of surface roughness coefficients corresponding to each three-dimensional particle.

9. The system for constructing a prediction model for soil and rock shear strength parameters according to claim 6, characterized in that, The processing unit includes: The first processing subunit is used to classify the surface roughness coefficient set by roughness distribution, divide the distribution range of the root mean square value of roughness by statistical analysis, divide the surface roughness level into a preset number of types, and construct the mapping relationship between particle surface properties and particle size range to obtain the initial particle size range division result corresponding to the roughness level. The second processing subunit is used to segment the initial interval division result of particle size based on the fractal cumulative distribution function to obtain a multi-level particle size combination sample set that meets the target fractal dimension control requirements. The third processing subunit is used to generate multiple proportioning schemes based on a multi-level particle size combination sample set with different particle size and roughness levels, thereby obtaining a structural gradation sample set.

10. The system for constructing a prediction model for soil and rock shear strength parameters according to claim 6, characterized in that, The computing unit includes: The first computational subunit is used to perform multi-field coupling model initialization processing on the structural gradation sample set to obtain a multi-field input configuration set under coupled state. The second calculation subunit is used to solve the multi-physics response equation system of the multi-field input configuration set. The multi-physics response equation system is solved by using preset working conditions to obtain the state dataset of each physical field. The third computational subunit is used to extract particle physical response parameters from the state dataset of all physical fields to obtain a set of physical response parameters with working condition discrimination.

Images