Method for determining microscopic parameters of multi-scale block discrete element geologic model

Abstract

The invention discloses a method for determining microscopic parameters of a multi-scale block discrete element geological model, which belongs to the field of geotechnical engineering, and comprises the following steps: S1, establishing a cylinder model for a compression model of geological rock, and carrying out block subdivision and gridding; s2, setting shear modulus, bulk modulus, normal rigidity, tangential rigidity, internal friction angle and joint roughness parameters for the established cylinder model; s3, the loading speed during compression is simulated for the cylinder model, and confining pressure is given; s4, carrying out compression simulation on the cylinder model until the cylinder is damaged; s5, obtaining a continuous yield curve according to a compression simulation condition, establishing a data set, and constructing a convolutional neural network model; s6, geologic model parameters are determined according to the convolutional neural network model; the discrete element microscopic parameters can be quickly and accurately calibrated, the influence of multiple scales on the discrete element microscopic parameters is considered, and the problem that the parameter calibration process of a conventional discrete element method is tedious and time-consuming is solved to a certain extent.

Description

Technical Field

[0001] The invention relates to the field of geotechnical engineering, and in particular to a method for determining microscopic parameters of a multi-scale block discrete element geological model. Background Art

[0002] In the field of geotechnical engineering and geomechanics, the mechanical behavior of reservoirs has important engineering significance for oil and gas extraction, geothermal energy development, and environmental engineering. In order to accurately predict and evaluate the response of reservoir rocks in engineering, effective numerical simulation methods are needed to simulate their mechanical behavior. At present, finite element or discrete element methods are generally used for simulation. However, for shales with natural weak planes and carbonate rocks with cemented natural fractures, it is difficult for finite elements to couple a large number of natural fractures. The discrete element method can couple a large number of random fractures, so the discrete element method is often used to simulate strong heterogeneous formations. As a numerical simulation technology based on Newton's second law, the discrete element method can accurately simulate the failure process of complex rocks such as shale. It needs to use microscopic parameters to characterize macroscopic characteristics, but due to the influence of the grid size division of multi-scale discrete element geological modeling, the mechanical microscopic parameters and macroscopic parameters have different mapping relationships, which makes the discrete element method micro parameter calibration process cumbersome, time-consuming, and has multiple solutions. In response to the difficulty of discrete element method parameter calibration, predecessors have proposed a variety of solutions. The trial and error method gradually adjusts the microscopic parameters to match the numerical simulation results with the experimental data. The method is simple and easy to operate, but the adjustment process relies on experience and is time-consuming. It is difficult to ensure global optimization and has low efficiency. The scale conversion method uses formulas to map macroscopic parameters to microscopic parameters. It is efficient but has limited applicability and is difficult to characterize complex reservoir conditions. At present, there is no good method to quickly and accurately determine the microscopic parameters of multi-scale discrete element geological modeling. Summary of the invention

[0003] In view of the above problems, the present invention provides a method for determining microscopic parameters of a multi-scale block discrete element geological model, which can solve the problem that the parameter determination process of the conventional discrete element method is cumbersome and time-consuming.

[0004] The technical solution of the present invention is:

[0005] A method for determining microscopic parameters of a multi-scale block discrete element geological model comprises the following steps:

[0006] S1. Establish a column model for the compression model of geological rocks and perform block subdivision and meshing;

[0007] S2. Setting shear modulus, bulk modulus, normal stiffness, tangential stiffness, internal friction angle, and joint roughness parameters for the established column model;

[0008] S3, simulating the loading speed of the column model during compression and assigning confining pressure;

[0009] S4, performing compression simulation on the column model until the column is destroyed;

[0010] S5. According to the compression simulation of step S4, a continuous yield curve is obtained, a data set is established, and a convolutional neural network model is constructed;

[0011] S6. Determine geological model parameters according to the convolutional neural network model obtained in step S5.

[0012] In the step S1, the column model is established by using Rhino software, and the block subdivision and meshing are processed by discrete element software.

[0013] In step S4, the compression simulation adopts a continuous yield model, and the compression simulation algorithm is as follows:

[0014] Δσ n =k n Δu n

[0015]

[0016] Where: Δσ n : Normal stress variation, k n : Normal stiffness, Δu n : Normal displacement change, σ n : Normal stress, a n and e n are model parameters;

[0017] For shear loading, an irreversible nonlinear behavior is exhibited starting from shear, and the shear stress increment is calculated as:

[0018] Δτ=Fk s Δu s

[0019]

[0020]

[0021] Where: Δτ: shear stress increment, F: correction factor, k s : Tangential stiffness, Δu s : Change in tangential displacement, a s and e s is the model parameter, T: current shear stress, T m : boundary strength, r: coefficient;

[0022] The boundary strength is given by:

[0023] τ m =σ m tanφ msgn(Δu s )

[0024]

[0025] Where: T m： Boundary strength, σ m : actual normal stress, Ф m : Effective internal friction angle, ΔФ m : Change in effective internal friction angle, R: joint roughness parameter, Ф: internal friction angle, Δu s p : Plastic cumulative tangential displacement change, Δu s : tangential displacement variation;

[0026] The incremental relationship is given by the following formula:

[0027]

[0028] Where: Ф m (i) : Initial internal friction angle, indicating the original state of the joint, u s p : Plastic accumulated tangential displacement;

[0029] The effective expansion angle i is given by the following formula:

[0030] i=tan -1 (τ| / σ n )-φ

[0031] Where: T: current shear stress, σ n : Normal stress, Ф: Internal friction angle.

[0032] In step S5, the data set is obtained by using a normalization method, and the formula is as follows:

[0033]

[0034] Where: x': normalized data, x: data to be normalized, min(x): minimum value of the same type of data before normalization, max(x): maximum value of the same type of data before normalization.

[0035] In step S5, the data set is randomly divided into a training set and a test set, and the convolutional neural network is trained using the training set.

[0036] In step S6, model performance evaluation needs to be performed before the geological model parameters are determined. The evaluation formula is as follows:

[0037]

[0038] Where: R2 is the coefficient of determination, y p is the predicted value, y t is the true value, y a is the average of the true values.

[0039] The beneficial effects of the present invention are:

[0040] The present invention discloses a multi-scale block discrete element geological model parameter calibration method based on convolutional neural network. The nonlinear mapping capability of convolutional neural network is utilized, the influence of multiple scales on the microscopic parameters of discrete elements is considered, and the microscopic parameters of discrete elements are rapidly calibrated. To a certain extent, the problem that the parameter calibration process of conventional discrete element method is complicated and time-consuming is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for determining microscopic parameters of a multi-scale block discrete element geological model according to an embodiment of the present invention;

[0042] Figure 2 is a typical stress-displacement curve under constant normal stress of a method for determining microscopic parameters of a multi-scale block discrete element geological model according to an embodiment of the present invention;

[0043] Figure 3 It is a schematic diagram of the model structure of a CNN of a method for determining microscopic parameters of a multi-scale block discrete element geological model according to an embodiment of the present invention;

[0044] Figure 4 It is a column model required by a compression model of a method for determining microscopic parameters of a multi-scale block discrete element geological model described in an embodiment of the present invention;

[0045] Figure 5 is the determination coefficient value of each parameter value of the method for determining microscopic parameters of a multi-scale block discrete element geological model according to an embodiment of the present invention;

[0046] Figure 6 is a true continuous yield curve of a method for determining microscopic parameters of a multi-scale block discrete element geological model according to an embodiment of the present invention;

[0047] Figure 7 It is a comparison diagram of the stress-displacement curve after the determined micro-parameters and scales of a multi-scale block discrete element geological model micro-parameter determination method described in an embodiment of the present invention are simulated by discrete elements and the continuous yield curve under the same experimental conditions. DETAILED DESCRIPTION

[0048] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0049] Example:

[0050] like Figure 1 - Figure 7 As shown, a method for determining microscopic parameters of a multi-scale block discrete element geological model comprises the following steps:

[0051] S1. Establish a column model for the compression model of geological rocks and perform block subdivision and meshing;

[0052] A set of experimental body heights and radii, subdivided polygon side lengths, and minimum grid lengths were set. The cylindrical model required for the compression model was established using Rhino software, and then imported into discrete element software and used Voronoi polygons for block subdivision and meshing.

[0053] S2. Setting shear modulus, bulk modulus, normal stiffness, tangential stiffness, internal friction angle, and joint roughness parameters for the established column model;

[0054] S3, simulating the loading speed of the column model during compression and assigning confining pressure;

[0055] The mesh nodes on the upper and lower end surfaces were assigned a downward and upward constant speed of 0.05 mm / min, respectively, to simulate the loading rate during compression, and a confining pressure of 15 MPa was assigned.

[0056] S4, performing compression simulation on the column model until the column is destroyed;

[0057] The above parameters are input into the discrete element simulation software through programming, the continuous yield model is called for calculation, the reaction forces of the upper and lower end faces are monitored, and compression simulation is performed until the column is destroyed.

[0058] The compression simulation adopts the continuous yield model, and the compression simulation algorithm is as follows:

[0059] Δσ n =k n Δu n

[0060]

[0061] Where: Δσ n : Normal stress variation, k n : Normal stiffness, Δu n : Normal displacement change, σ n : Normal stress, a n and e n are model parameters;

[0062] For shear loading, an irreversible nonlinear behavior is exhibited starting from shear, and the shear stress increment is calculated as:

[0063] Δτ=Fk s Δus

[0064]

[0065] Where: Δτ: shear stress increment, F: correction factor, k s : Tangential stiffness, Δu s : Change in tangential displacement, a s and e s is the model parameter, T: current shear stress, T m : boundary strength, r: coefficient;

[0066] The boundary strength is given by:

[0067] τ m =σ m tanφ m sgn(Δu s )

[0068]

[0069] Where: T m： Boundary strength, σ m : actual normal stress, Ф m : Effective internal friction angle, ΔФ m : Change in effective internal friction angle, R: joint roughness parameter, Ф: internal friction angle, Δu s p : Plastic cumulative tangential displacement change, Δu s : tangential displacement variation;

[0070] The incremental relationship is given by the following formula:

[0071]

[0072] Where: Ф m (i) : Initial internal friction angle, indicating the original state of the joint, u s p : Plastic accumulated tangential displacement;

[0073] The effective expansion angle i is given by the following formula:

[0074] i=tan -1 (τ| / σ n )-φ

[0075] Where: T: current shear stress, σ n : Normal stress, Ф: Internal friction angle.

[0076] S5. According to the compression simulation of step S4, a continuous yield curve is obtained, a data set is established, and a convolutional neural network model is constructed;

[0077] After calculation by the software, a set of continuous yield curves simulated by the parameters of step 1 is obtained. The parameters in steps 1 and 2 are arbitrarily changed within a reasonable range, and then the operations of steps 3-4 are performed to obtain n sets of continuous yield curves simulated by the parameters of steps 1 and 2. The parameter range is as follows:

[0078]

[0079] The n groups of data were normalized to obtain a data set characterized by continuous yield curve, height and radius of the test body, side length of subdivided polygon, minimum grid length, and labeled by shear modulus, bulk modulus, normal stiffness, tangential stiffness, internal friction angle, and joint roughness parameters;

[0080] Using the Min-Max normalization method, the image pixel values ​​and data are scaled to [0,1], the formula is as follows:

[0081]

[0082] Where: x': normalized data, x: data to be normalized, min(x): minimum value of the same type of data before normalization, max(x): maximum value of the same type of data before normalization.

[0083] Take the number of labels as the number of neurons in the last layer of the fully connected module of the convolutional neural network, construct a convolutional neural network model CNN, determine the model's learning rate l earning rate, model training times step, model loss function loss function, model optimization algorithm opt imize algor ithm, input the training set into the constructed CNN model, the model automatically grabs batch size data for training each time, the number of training times is step times, each training will get the predicted label value, use the loss function to calculate the error between the predicted value and the true value, and then optimize the loss function through the optimization algorithm. When the training is completed, save the trained CNN model.

[0084] Among them: the ratio of training set to test set is 8:2, and the batch size is 16;

[0085] The convolutional neural network consists of 4 convolutional layers and 3 fully connected layers. The convolutional layer finally has 128 channels. Each convolution operation contains a ReLU activation function to capture more local features while keeping the model training stable. Use 4 poolings to gradually reduce the resolution of the feature map. The pooled data is input into the flattening layer to convert the data into one-dimensional data, and the information is further integrated through multiple layers of full connection. The dimension of the fully connected layer is gradually reduced to output 6 final labels. The learning rate l earn i ngrate is 0.001, the number of model training steps is 8000 times, the loss function of the model is the mean square error loss function (MSELoss), and the optimization algorithm of the model is Adam;

[0086] The formula for the ReLU activation function is:

[0087] ReLU(X)=max(0,X)

[0088] Setting all negative values ​​to zero and retaining positive values ​​can effectively alleviate the gradient vanishing problem;

[0089] The MSELoss formula is:

[0090]

[0091] Where: y p : predicted value, y t : true value, n: number of samples.

[0092] S6. Determine geological model parameters according to the convolutional neural network model obtained in step S5.

[0093] The trained CNN model is used for data prediction, and the test set is input into the trained model to obtain the prediction results, namely, shear modulus, bulk modulus, normal stiffness, tangential stiffness, internal friction angle, and joint roughness parameters.

[0094] The coefficient of determination (R 2 )formula:

[0095]

[0096] Where: y p is the predicted value, y t is the true value, y a is the average value of the true value to evaluate the model performance, and the determination coefficient R 2 The average value is 0.905, which indicates good model performance;

[0097] The real continuous yield curve under the same experimental conditions (same confining pressure, compression loading rate), the height and radius of the test body, the side length of the subdivided polygon, and the minimum grid length are input into the CNN model as input parameters to obtain the shear modulus, bulk modulus, normal stiffness, tangential stiffness, internal friction angle, and joint roughness parameters. The specific parameters are as follows:

[0098]

[0099] It can be seen that the above-mentioned microscopic parameters after the model is determined and the continuous yield curve after discrete element simulation has a good fit with the continuous yield curve under the same experimental conditions, which shows that the model has a strong ability to determine the microscopic parameters and is suitable for the determination of microscopic parameters in multi-scale block discrete element geological modeling.

[0100] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for determining microscopic parameters of a multi-scale block discrete element geological model, characterized in that: The steps include: S1. Establish a column model for the compression model of geological rocks and perform block subdivision and meshing; S2. Setting shear modulus, bulk modulus, normal stiffness, tangential stiffness, internal friction angle, and joint roughness parameters for the established column model; S3, simulating the loading speed of the column model during compression and assigning confining pressure; S4, performing compression simulation on the column model until the column is destroyed; S5. According to the compression simulation of step S4, a continuous yield curve is obtained, a data set is established, and a convolutional neural network model is constructed; S6. Determine geological model parameters according to the convolutional neural network model obtained in step S5.

2. The method for determining microscopic parameters of a multi-scale block discrete element geological model according to claim 1, characterized in that: In the step S1, the column model is established by using Rhino software, and the block subdivision and meshing are processed by discrete element software.

3. The method for determining microscopic parameters of a multi-scale block discrete element geological model according to claim 1, characterized in that: In step S4, the compression simulation adopts a continuous yield model, and the compression simulation algorithm is as follows: Board n =k n Thu n Where: Δσ n : Normal stress variation, k n : Normal stiffness, Δu n : Normal displacement change, σ n : Normal stress, a n and e n are model parameters; For shear loading, an irreversible nonlinear behavior is exhibited starting from shear, and the shear stress increment is calculated as: Δτ=Fk s Thu s Where: Δτ: shear stress increment, F: correction factor, k s : Tangential stiffness, Δu s : Change in tangential displacement, a s and e s is the model parameter, T: current shear stress, T m : boundary strength, r: coefficient; The boundary strength is given by: t m =s m tanf m sgn(Δu s ) Where: T m： Boundary strength, σ m : actual normal stress, Ф m : Effective internal friction angle, ΔФ m : Change in effective internal friction angle, R: joint roughness parameter, Ф: internal friction angle, Δu s p : Plastic cumulative tangential displacement change, Δu s : tangential displacement variation; The incremental relationship is given by the following formula: Where: Ф m (i) : Initial internal friction angle, indicating the original state of the joint, u s p : Plastic accumulated tangential displacement; The effective expansion angle i is given by the following formula: i=tan -1 (τ| / σ n )-φ Where: T: current shear stress, σ n : Normal stress, Ф: Internal friction angle.

4. The method for determining microscopic parameters of a multi-scale block discrete element geological model according to claim 1, characterized in that: In step S5, the data set is obtained by using a normalization method, and the formula is as follows: Where: x': normalized data, x: data to be normalized, min(x): minimum value of the same type of data before normalization, max(x): maximum value of the same type of data before normalization.

5. A method for determining microscopic parameters of a multi-scale block discrete element geological model according to claim 1 or 4, characterized in that: In step S5, the data set is randomly divided into a training set and a test set, and the convolutional neural network is trained using the training set.

6. The method for determining microscopic parameters of a multi-scale block discrete element geological model according to claim 1, characterized in that: In step S6, model performance evaluation needs to be performed before the geological model parameters are determined. The evaluation formula is as follows: Where: R 2 is the coefficient of determination, y p is the predicted value, y t is the true value, y a is the average of the true values.

Images