A Method for Calibrating Mesoscopic Parameters of UDEC Based on GA-BP Neural Network

Abstract

The present invention relates to the technical field of rock mechanics calculation, and specifically discloses a calibration method for UDEC mesoscopic parameters based on a GA-BP neural network, including the following steps: establishing a UDEC model; generating a mesoscopic parameter data set; calculating macroscopic parameters; performing Pearson correlation analysis; constructing and training a GA-BP neural network; and using the trained optimal model to output UDEC mesoscopic parameters. The present invention adopts a GA-BP model. The genetic algorithm optimized BP neural network is a prediction method that combines the genetic algorithm and the BP neural network. The genetic algorithm is used to optimize the weights and thresholds of the BP neural network, improving the prediction accuracy and generalization ability of the BP neural network. The advantages of the genetic algorithm optimized BP neural network are that it can overcome problems such as the BP neural network being prone to falling into local optimal solutions, having a slow convergence speed, and difficulty in parameter selection, and can improve the fitting effect.

Description

Technical Field

[0001] The present invention relates to the field of rock mechanics calculation technology, and in particular to a UDEC microscopic parameter calibration method based on a GA-BP neural network. Background Art

[0002] With the rapid development of computer technology, a growing number of numerical analysis methods are being applied to rock mechanics calculations. Finite element method, finite difference method, and discrete element method are widely used by many scholars. However, natural rock masses contain numerous discontinuities such as joints, which significantly affect their mechanical properties. Finite element method and finite difference method, based on continuum theory, divide objects into a finite number of units, connecting the nodes with mathematical equations. They are more suitable for homogeneous materials. Therefore, for natural rock masses containing discontinuities, finite element software calculation results have large errors. Discrete element method, on the other hand, approximates rock by stacking and cementing units with specific mechanical properties, and can naturally characterize the various microstructures and mechanical properties of rock masses.

[0003] Currently, discrete element models such as PFC2D, PFC3D, 3DEC, and UDEC are commonly used in tunneling. In PFC, the model primarily represents the material using particles plus connecting bonds. Boundary walls determine the model's size and shape, and mechanical properties are assigned through stiffness, strength, and friction parameters. Because the units are round particles and the deformation capacity of crystals is ignored, PFC is more suitable for exploring the microscale simulation and research of granular materials such as sand. UDEC discretizes the material into multiple independent blocks and simulates the overall behavior of the material through the interactions between these blocks. Because the blocks are in full contact, arbitrary polygons provide a better mosaic effect and more realistically represent the rock's microstructure. Therefore, block-based discrete element models offer significant advantages over particle models.

[0004] When performing discrete element calculations, different element and contact parameters (microscopic parameters) need to be set to simulate actual rock tests. However, the parameters obtained from macroscopic rock test structures cannot be directly corresponded to microscopic parameters. Therefore, in order to obtain a macroscopic response consistent with the results of laboratory tests, it is necessary to calibrate the microscopic parameters of the model. There are many domestic studies on the calibration of PFC microscopic parameters. In PFC, to improve the calibration efficiency, some scholars use methods such as orthogonal test method and response surface method to analyze the relationship between PFC microscopic parameters and macroscopic parameters, and establish a function expression of macro-microscopic parameters. Natural rock masses contain a large number of discontinuity surfaces such as joints, and these discontinuity surfaces significantly affect the mechanical properties of rock masses. The discrete element method approximates the simulation of rocks by stacking and cementing elements with specific mechanical properties, and can better characterize various microscopic structures and mechanical properties of rock masses. However, the microscopic parameters input during discrete element calculations cannot be corresponded to the macroscopic parameters obtained from tests. Therefore, calibration is needed. The existing calibration methods are generally trial-and-error methods, which are complex, cumbersome, computationally intensive, and inefficient. Therefore, it is necessary to seek new methods to achieve rapid calibration of macro-microscopic parameters. Summary of the Invention

[0005] To solve the problems existing in the prior art, the present invention provides a UDEC microscopic parameter calibration method based on GA-BP neural network, which fits the relationship between macro-microscopic parameters with GA-BP neural network to achieve rapid calibration, and solves the problems mentioned in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: A UDEC microscopic parameter calibration method based on GA-BP neural network, comprising the following steps:

[0007] S1. Establish a UDEC model;

[0008] S2. Generate a microscopic parameter data set;

[0009] S3. Calculate and obtain macroscopic parameters;

[0010] S4. Pearson correlation analysis;

[0011] S5. Construct a GA-BP neural network and train it;

[0012] S6. Use the trained optimal model to output UDEC microscopic parameters.

[0013] Preferably, in step S1, the UDEC model includes the UDEC uniaxial compression test, triaxial compression test, and Brazilian splitting test models; the model divides the rock specimen into independent block elements through Thiessen cutting. The elements adopt the elastic block model, the contact adopts the mohr-coulomb plastic model, and the seeds are fixed. Rigid plates are established respectively above and below. The lower plate restricts the displacements in the x and y directions; the upper plate restricts the displacement in the x direction and applies a downward velocity-controlled loading.

[0014] Preferably, in step S2, it specifically includes: determining the range of mesoscopic parameters and randomly generating mesoscopic parameters; the mesoscopic parameters include: block elastic modulus, block Poisson's ratio, contact normal stiffness, contact stiffness ratio, contact cohesion, contact friction angle, and contact tensile strength.

[0015] Preferably, the block elastic modulus, Poisson's ratio, and contact normal stiffness have the following relationship:

[0016]

[0017] E = 2G(1 + ν)

[0018] where E is the block elastic modulus; G is the block shear modulus; ΔZ min is the minimum width of the area adjacent to the contact point in the normal direction; K n is the normal stiffness; ν is Poisson's ratio, and n represents the correction coefficient, with a value range of 1 to 10.

[0019] Preferably, in step S3, it specifically includes: performing UDEC simulation calculations for three tests for each set of mesoscopic parameters to obtain the corresponding macroscopic parameters; the macroscopic parameters include uniaxial compressive strength, elastic modulus, tensile strength, friction angle, and cohesion.

[0020] Preferably, in step S4, Pearson product-moment correlation coefficient is used for correlation analysis to determine the input and output parameters for subsequent GA-BP neural network fitting; the Pearson product-moment correlation coefficient is used to measure the correlation between two variables X and Y, and its value ranges from -1 to 1, describing the strength of the linear correlation between the two variables. The larger the absolute value, the stronger the correlation; the calculation formula for the Pearson correlation coefficient is:

[0021]

[0022] where: X i and Y i are the i-th observed values of the two variables X and Y; and is the average value of variables X and Y; r is the Pearson correlation coefficient; r reflects the strength of the linear correlation between two variables. The larger the absolute value of r, the stronger the correlation;

[0023] The value range of r is from -1 to 1. When r > 0, it indicates that the two variables are positively correlated, that is, the larger the value of one variable, the larger the value of the other variable; when r < 0, it indicates that the two variables are negatively correlated, that is, the larger the value of one variable, the smaller the value of the other variable; when r = 0, it indicates that the two variables are not linearly correlated.

[0024] Preferably, after Pearson correlation analysis, the input parameters for the subsequent GA-BP neural network fitting are determined as the uniaxial compressive strength of the specimen, the elastic modulus of the specimen, the tensile strength of the specimen, the friction angle of the specimen, and the cohesion of the specimen, while the output parameters are the elastic modulus of the block, the contact cohesion, the contact friction angle, and the contact tensile strength.

[0025] Preferably, in step S5, it specifically includes the following:

[0026] a. Data preprocessing: Delete the outliers and duplicate values of the data, and perform min-max normalization on the data using MinMaxScaler, that is, linearly scale all data to 0-1;

[0027] b. Define the model: Use Sequential to create a BP neural network with three hidden layers;

[0028] c. Determine the hyperparameter range: Determine the value ranges of the number of neurons in the hidden layer, the learning rate, the batch size, and the number of training epochs;

[0029] d. Initialize the population and define the fitness function: Initialize the neural network weights and determine the fitness function, through which the genetic algorithm evaluates each individual;

[0030] e. Execution of the genetic algorithm: First, calculate the fitness of the initial individuals in the population, and then select the excellent individuals to exchange some of their information to generate new individuals. The probability of an individual being selected depends on its fitness. The higher the fitness, the greater the probability of being selected. There is a small probability that the newly generated individuals will mutate, so as to increase the input space of the sample; repeat the above operations for the newly generated individuals until the maximum number of iterations;

[0031] f. Retrain and test: Retrain the model using the optimized best parameters, use the test set for prediction, inverse normalization, and error evaluation to obtain the optimal GA-BP neural network after training.

[0032] Preferably, in the error evaluation, the following indicators are used to evaluate the accuracy of the model fitting, specifically as follows:

[0033]

[0034] Among them, RMSE represents the root mean square error, MAE represents the mean absolute error, and R 2 represents the coefficient of determination.

[0035] The beneficial effects of the present invention are as follows:

[0036] 1) The present invention uses a GA-BP neural network to fit the relationship between macro and micro parameters when UDEC simulates rock masses, greatly improving the calibration efficiency and filling the gap in related field research. The present invention uses Pearson correlation analysis to determine the correlation between macro and micro parameters, which can improve the subsequent fitting effect and efficiency.

[0037] 2) The present invention adopts a GA-BP model. The genetic algorithm optimized BP neural network is a prediction method that combines the genetic algorithm and the BP neural network. The genetic algorithm is used to optimize the weights and thresholds of the BP neural network, improving the prediction accuracy and generalization ability of the BP neural network. The advantage of the genetic algorithm optimized BP neural network is that it can overcome problems such as the BP neural network being prone to falling into local optimal solutions, slow convergence speed, and difficult parameter selection, and can improve the fitting effect.

[0038] 3) For the test results of the present invention, the RMSE of the model of the test data is 3.3, the MAE is 2.52, and R 2 = 0.94, indicating that the predicted value is close to the actual value and the prediction result is good. The method of using the GA-BP neural network model to fit macro and micro parameters proposed by the present invention is simple and has good accuracy, and can provide a fast calibration method for UDEC rock mass discrete element calculation, with broad application scenarios. Description of the Drawings

[0039] Figure 1 is a schematic diagram of the UDEC uniaxial compression model in the embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of the UDEC triaxial compression model in the embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of the UDEC Brazilian splitting model in the embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of the Pearson correlation analysis in the embodiment of the present invention;

[0043] Figure 5 is a flow chart of the GA-BP neural network in the embodiment of the present invention;

[0044] Figure 6 is a scatter plot of the prediction result and the actual result in the embodiment of the present invention;

[0045] Figure 7 This is the visualization loss change curve in the embodiments of the present invention. Detailed implementation manners

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] Compared with traditional data processing methods, the BP neural network can handle complex non-linear relationships, automatically extract features, and has strong learning ability. Therefore, the present invention uses the BP neural network to explore the relationships among the macroscopic tensile strength, uniaxial compressive strength, elastic modulus of rocks and the mesoscopic parameters of UDEC. In the BP neural network algorithm, the modification of weights is achieved through error backpropagation and gradient descent. Such a mechanism is prone to falling into local minima and unable to obtain the global minimum. On the contrary, the genetic algorithm (GA) is good at global search. Using the genetic algorithm to optimize the initial weights of the BP neural network is beneficial to solving the problem that the BP algorithm is prone to falling into local minima.

[0048] Embodiment 1

[0049] A method for calibrating mesoscopic parameters of UDEC based on GA-BP neural network, the steps include:

[0050] (1) Establish a UDEC model

[0051] Establish UDEC uniaxial compression test, triaxial compression test, and Brazilian splitting test models as Figure 1 、 Figure 2 and Figure 3 shown. The model divides the rock specimen into independent block units through Thiessen cutting. The unit adopts an elastic block model, the contact adopts a mohr-coulomb plastic model, and the seeds are fixed. Rigid plates are established at the top and bottom respectively. The lower plate restricts the displacements in the x and y directions; the upper plate restricts the displacement in the x direction and applies a downward velocity control loading. In the uniaxial compression model and the triaxial compression model, the rock specimen is 80 cm high and 40 cm long, and the maximum side length of the Thiessen polygon is 2 cm. During the triaxial compression test, it is necessary to pre-press the model first, that is, apply the confining pressure σ3 until equilibrium is reached, and then load. In the Brazilian splitting model, the rock specimen has a diameter of 30 cm and the maximum side length of the Thiessen polygon is 0.75 cm.

[0052] (2) Generate a mesoscopic parameter data set

[0053] Determine the mesoscopic parameter range based on the surveyed literature, and randomly generate mesoscopic parameters, including: block elastic modulus, block Poisson's ratio, contact normal stiffness, contact stiffness ratio, contact cohesion, contact friction angle, and contact tensile strength. Among them, the elastic modulus, Poisson's ratio, and contact normal stiffness of the block have the following relationship:

[0054]

[0055] E = 2G(1 + ν)

[0056] where E is the block elastic modulus; G is the block shear modulus; ΔZ min is the minimum width of the area adjacent to the contact point in the normal direction; K n is the normal stiffness; ν is Poisson's ratio.

[0057] (3) Calculate the macroscopic parameters

[0058] For each set of mesoscopic parameters, perform UDEC simulation calculations for the above three tests to obtain the corresponding macroscopic parameters: uniaxial compressive strength, elastic modulus, tensile strength, friction angle, and cohesion.

[0059] (4) Pearson correlation analysis

[0060] Perform correlation analysis through the Pearson product-moment correlation coefficient to determine the input and output parameters for subsequent GA-BP neural network fitting, as Figure 4 shown. In statistics, the Pearson product-moment correlation coefficient (PPMCC) can be used to measure the correlation between two variables X and Y, and its value ranges from -1 to 1, describing the strength of the linear correlation between the two variables. The larger the absolute value, the stronger the correlation. The calculation formula for the Pearson correlation coefficient is:

[0061]

[0062] where: r is the Pearson correlation coefficient; X i and Y i are the i-th observed values of the two variables X and Y. and is the average value of variables X and Y. r reflects the strength of the linear correlation between the two variables. The larger the absolute value of r, the stronger the correlation. The value range of r is from -1 to 1. When r > 0, it indicates that the two variables are positively correlated, that is, the larger the value of one variable, the larger the value of the other variable; when r < 0, it indicates that the two variables are negatively correlated, that is, the larger the value of one variable, the smaller the value of the other variable; when r = 0, it indicates that the two variables are not linearly correlated (note that it is only non-linearly correlated), but there may be other ways of correlation (such as in a curve form); when r = 1 and -1, it means that the two variables X and Y can be well described by a linear equation, and all sample points fall well on a straight line. Usually, when |r| ≥ 0.8, the two variables can be considered highly correlated; when 0.5 ≤ |r| < 0.8, they can be considered moderately correlated; when 0.3 ≤ |r| < 0.5, they can be considered lowly correlated; when |r| < 0.3, the two variables can be considered basically uncorrelated. Since the bulk modulus and the contact normal stiffness in (2) are calculated by a formula, the correlation coefficient between the two is 1. Only the bulk modulus will be discussed later and the contact normal stiffness will not be discussed.

[0063] (5) Construct a GA-BP neural network, and the network process is as Figure 5 shown, specifically including the following:

[0064] a. Data preprocessing: Manually delete the outliers and duplicate values in the data. Use MinMaxScaler to perform min-max normalization on the data, that is, linearly scale all data to 0 - 1;

[0065] b. Define the model: Use Sequential to create a BP neural network with three hidden layers.

[0066] c. Determine the range of hyperparameters: Determine the value ranges of the number of neurons in the hidden layer (hidden_neurons), learning rate (learning_rate), batch size (batch_size), and number of training epochs (epochs).

[0067] d. Initialize the population and define the fitness function: Initialize the neural network weights and determine the fitness function, through which the genetic algorithm evaluates each individual.

[0068] e. Execution of the genetic algorithm: Each individual in the population represents a combination of weights, that is, a solution. We expect to find a solution that makes the model perform best. The basis for judging the quality of a solution is the error between the model output and the real data. Therefore, we first need to calculate the error of the model under each solution (i.e., the fitness function described in d). First, calculate the fitness of the initial individuals in the population, and then select excellent individuals to exchange some of their information to generate new individuals. The probability of an individual being selected depends on its fitness. The higher the fitness, the greater the probability of being selected. The newly generated individuals have a small probability of mutation to increase the input space of the sample. Repeat the above operations on the newly generated individuals until the maximum number of iterations.

[0069] f. Retraining and testing: Retrain the model using the optimized best parameters, and use the test set for prediction, denormalization, and error evaluation.

[0070] (6) Error analysis

[0071] Use the following metrics to evaluate the accuracy of the model fitting: RMSE represents the root mean square error, MAE represents the mean absolute error, and R 2 represents the coefficient of determination, and the expressions are as follows:

[0072]

[0073] The smaller the RMSE, the smaller the difference between the predicted results of the model and the actual observed values. Different from RMSE, MAE does not consider the square of the error, so it is not sensitive to outliers. The smaller the value of MAE, the smaller the average absolute difference between the predicted results of the model and the actual observed values. R 2 represents the proportion of the variance explained by the model. When the predicted values of the model are exactly the same as the actual observed values, R 2 is 1; when the difference between the predicted values of the model and the actual observed values is equal to the difference between the mean of the target variable and the actual observed values, R 2 is 0; if the performance of the model is worse than directly using the mean of the target variable as the predicted value, R 2 is negative. Use the trained optimal model to output the UDEC mesoscopic parameters.

[0074] Example 2

[0075] A method for calibrating UDEC mesoscopic parameters based on the GA-BP neural network, the steps include:

[0076] (1) Establish a UDEC model

[0077] Establish UDEC uniaxial compression test, triaxial compression test, and Brazilian splitting test models.

[0078] (2) Generate the mesoscopic parameter dataset

[0079] Determine the mesoscopic parameter range according to the investigated literature and randomly generate 200 groups of mesoscopic parameters.

[0080] (3) Calculate the macroscopic parameters

[0081] Perform UDEC simulation calculations for the above three tests for each group of mesoscopic parameters to obtain the corresponding macroscopic parameters: uniaxial compressive strength, elastic modulus, tensile strength, friction angle, and cohesion.

[0082] (4) Pearson correlation analysis

[0083] Conduct correlation analysis through the Pearson product-moment correlation coefficient to determine the relevant parameters of the input and output for the subsequent GA-BP neural network fitting, as Figure 4 shown. It can be seen from the figure that the uniaxial compressive strength of the rock is mainly affected by the contact cohesion and contact friction angle in the mesoscopic parameters, with r being 0.53 and 0.56 respectively; the tensile strength of the rock is mainly affected by the contact tensile strength, with r being 0.68; the elastic modulus of the specimen is affected by the block elastic modulus, with r being 0.75; the friction angle of the specimen is affected by the contact friction angle, with r being 0.84; the cohesion of the specimen is affected by the contact cohesion, with r being 0.63; the correlation coefficients between the block Poisson's ratio and contact stiffness ratio in the mesoscopic parameters and the test macroscopic parameters are all less than 0.2. Therefore, it is judged that they are basically not related to the test macroscopic parameters. In summary, the input parameters for the subsequent neural network fitting are the uniaxial compressive strength of the specimen, the elastic modulus of the specimen, the tensile strength of the specimen, the friction angle of the specimen, and the cohesion of the specimen, while the output parameters are the block elastic modulus, contact cohesion, contact friction angle, and contact tensile strength.

[0084] (5) Construct the GA-BP neural network

[0085] a. Data preprocessing: Manually delete the outliers and duplicate values in the 200 groups of data. Use

[0086] MinMaxScaler to perform min-max normalization on the data, that is, linearly scale all data to 0-1;

[0087] b. Define the model: Use Sequential to create a BP neural network with three hidden layers. The activation function is tanh, and the output layer is linear.

[0088] c. Determine the GA parameter range: Determine the value ranges of the number of hidden layer neurons (hidden_neurons), learning rate (learning_rate), batch size (batch_size), and number of training epochs (epochs) as shown in Table 1. d. Initialize the population and define the fitness function: Initialize the neural network weights and determine to use the mean squared error (MSE) as the fitness scoring criterion.

[0089] e. Execution of the genetic algorithm: Use the DEAP library to optimize the genetic algorithm, and the optimal parameters are shown in Table 1.

[0090] f. Retrain and test: Retrain the model using the optimized best parameters, and use the test set for prediction, denormalization, and error evaluation.

[0091] Table 1 Parameter space optimized by the genetic algorithm and selection of optimal parameters

[0092]

[0093] (6) Error analysis

[0094] The final results are as Figure 6 、 Figure 7 shown. The predicted values and the actual values are close, indicating a good fitting effect. The fitting of the model converges normally, without overfitting or underfitting. The RMSE of the model is 3.3, MAE is 2.52, and R 2 = 0.94, indicating that the predicted values and the actual values are relatively close and the prediction results are good.

[0095] In several embodiments provided by the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0096] In addition, in each embodiment of the present invention, the functional modules can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0097] If the above-mentioned functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that makes contributions to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs. It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without more limitations, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0098] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the", and "said" used in the embodiments of the present invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0099] It should be understood that the term " / and" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0100] Depending on the context, as used herein, the word "if" can be interpreted as "when" or "while" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0101] The "first / second" mentioned in the embodiments is only used to distinguish similar objects and does not represent a specific order for the objects. Understandably, the "first / second" can be interchanged in their specific order or sequence when allowed. It should be understood that the objects distinguished by the "first / second" can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0102] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for calibrating mesoscopic parameters of UDEC based on GA-BP neural network, characterized in that, It includes the following steps: S1. Establish a UDEC model; S2. Generate a mesoscopic parameter dataset; S3. Calculate the macroscopic parameters; S4. Pearson correlation analysis; perform correlation analysis through the Pearson product-moment correlation coefficient to determine the input and output parameters for subsequent GA-BP neural network fitting; the Pearson product-moment correlation coefficient is used to measure the correlation between two variables X and Y, and its value ranges from -1 to 1, describing the strength of the linear correlation between the two variables. The greater the absolute value, the stronger the correlation; the calculation formula for the Pearson correlation coefficient is: Where: X i and Y i are the i-th observations of two variables X and Y; and are the average values of variables X and Y; r is the Pearson correlation coefficient; r reflects the strength of the linear correlation between the two variables. The larger the absolute value of r, the stronger the correlation; The value range of r is from -1 to 1. When r > 0, it indicates that the two variables are positively correlated, that is, the larger the value of one variable, the larger the value of the other variable; when r < 0, it indicates that the two variables are negatively correlated, that is, the larger the value of one variable, the smaller the value of the other variable; when r = 0, it indicates that the two variables are not linearly correlated; S5. Construct and train a GA-BP neural network; specifically including the following: a. Data preprocessing: Delete the outliers and duplicate values of the data, and use MinMaxScaler to perform min-max normalization on the data, that is, linearly scale all data to 0 - 1; b. Define the model: Use Sequential to create a BP neural network with three hidden layers; c. Determine the range of hyperparameters: Determine the value ranges of the number of neurons in the hidden layer, learning rate, batch size, and number of training epochs; d. Initialize the population and define the fitness function: Initialize the neural network weights and determine the fitness function, through which the genetic algorithm evaluates each individual; e. Execution of the genetic algorithm: First calculate the fitness of the initial individuals in the population, then select excellent individuals to exchange some of their information to generate new individuals. The probability of a certain individual being selected depends on its fitness. The higher the fitness, the greater the probability of being selected. There is a small probability of mutation for the newly generated individuals to increase the input space of the sample; repeat the above operations for the newly generated individuals until the maximum number of iterations; f. Retrain and test: Retrain the model using the optimized best parameters, use the test set for prediction, inverse normalization, and error evaluation to obtain the optimal GA-BP neural network after training; S6. Use the trained optimal model to output the UDEC mesoscopic parameters.

2. The UDEC mesoscopic parameter calibration method based on the GA-BP neural network according to claim 1, characterized in that: In step S1, the UDEC model includes UDEC uniaxial compression test, triaxial compression test, and Brazilian splitting test models; the model divides the rock specimen into independent block units through Thiessen polygonization. The units adopt elastic block models, and the contacts adopt mohr-coulomb plastic models, and the seeds are fixed. Rigid plates are established above and below respectively. The lower plate restricts the displacements in the x and y directions; the upper plate restricts the displacement in the x direction and applies a downward velocity-controlled load.

3. The UDEC mesoscopic parameter calibration method based on the GA-BP neural network according to claim 1, characterized in that: In step S2, it specifically includes: determining the range of mesoscopic parameters and randomly generating mesoscopic parameters; the mesoscopic parameters include: block elastic modulus, block Poisson's ratio, contact normal stiffness, contact stiffness ratio, contact cohesion, contact friction angle, contact tensile strength.

4. The UDEC mesoscopic parameter calibration method based on the GA-BP neural network according to claim 3, characterized in that: The relationship between the elastic modulus, Poisson's ratio and contact normal stiffness of the block is as follows: E = 2G(1 + ν) where E is the bulk elastic modulus; G is the bulk shear modulus; ΔZ min is the minimum width of the region adjacent to the contact in the normal direction; K n is the normal stiffness; ν is Poisson's ratio, and n represents the correction coefficient, with a value range of 1 to 10.

5. The UDEC mesoscopic parameter calibration method based on the GA-BP neural network according to claim 1, characterized in that: In step S3, it specifically includes: UDEC simulation calculations of three tests are carried out for each group of mesoscopic parameters to obtain the corresponding macroscopic parameters; the macroscopic parameters include uniaxial compressive strength, elastic modulus, tensile strength, friction angle, and cohesion.

6. The UDEC mesoscopic parameter calibration method based on the GA-BP neural network according to claim 1, wherein: After Pearson correlation analysis, the input parameters for subsequent GA-BP neural network fitting are determined as the uniaxial compressive strength of the specimen, elastic modulus of the specimen, tensile strength of the specimen, friction angle of the specimen, and cohesion of the specimen, while the output parameters are the elastic modulus of the block, contact cohesion, contact friction angle, and contact tensile strength.

7. The UDEC mesoscopic parameter calibration method based on the GA-BP neural network according to claim 1, characterized in that: In the error evaluation described above, the following indicators are used to evaluate the accuracy of model fitting, specifically as follows: Among them, RMSE represents the root mean square error, MAE represents the mean absolute error, and R 2 represents the coefficient of determination.

Images