A ground stress prediction method based on an alternative model acceleration optimization algorithm

Abstract

The application discloses a ground stress prediction method based on a substitute model acceleration optimization algorithm, and the method comprises the following steps: on the basis of determining the form of boundary conditions required for ground stress prediction in a research area, a three-dimensional geological model is established through geological data, N groups of stress boundary conditions are randomly selected, learning samples are obtained by solving stress by using finite difference software, then an optimal boundary condition is predicted by using a neural network substitute model and a random search algorithm, and finally, stress is solved by using finite difference software to obtain a ground stress field. The acceleration optimization algorithm of the application concentrates the advantages of the substitute model and the random search algorithm. The random search algorithm can continuously generate random points in a defined optimization interval, so that the optimal interval can be quickly sought. The substitute model can quickly seek a local optimal solution in the optimal interval, and the two are iterated with each other, so that the global optimal solution is obtained by acceleration.

Description

Technical Field

[0001] This invention relates to a geostress prediction method based on an accelerated optimization algorithm using an alternative model, which can be applied to obtaining the geostress field distribution characteristics in various geotechnical engineering projects such as mineral resource development, underground space construction, water conservancy and hydropower, military, and transportation. Background Technology

[0002] Geostress is caused by the weight of the rock mass, geological tectonic movements, stratigraphic structure, gravitational force, and external loads. For geotechnical engineering, the main factors to consider are self-weight stress, tectonic stress, and stratigraphic stress, which together constitute the initial stress field of the rock mass.

[0003] Ground stress is the fundamental force that causes deformation and damage in mining, water conservancy and hydropower, civil engineering, railway, highway, military and other underground or open-pit rock and soil (mass) excavation projects. It is a necessary prerequisite for determining the mechanical properties of rock and soil (mass) and realizing the scientific design and decision-making of rock and soil (mass) engineering excavation.

[0004] In-situ stress prediction is a crucial aspect of integrated geological and engineering research. Therefore, in-situ stress field prediction is a scientifically sound, economical, and effective method for obtaining in-situ stress data, and it has gained widespread acceptance. However, due to the complex geological structure and strong heterogeneity of underground rock masses, the in-situ stress field is complex and variable. The results obtained from in-situ in-situ stress measurements can only reflect the local stress state at that point. Furthermore, in-situ in-situ stress testing of underground rock masses has many limitations, including high requirements for rock mass quality, low test success rate, high measurement costs, and large result dispersion. Limited in-situ in-situ in-situ stress measurements alone are far from sufficient to provide scientific and effective guidance for underground engineering construction.

[0005] Existing geostress measurement methods are classified in various ways. Some scholars classify them based on the characteristics of the measured data, into two aspects: absolute stress measurement and relative stress measurement. Absolute stress measurement can be further divided into direct measurement methods and indirect measurement methods. Direct measurement methods include hydraulic fracturing, acoustic emission (AE) method, geological mapping method, etc., while indirect measurement methods include core stress relief method, stress recovery method, X-ray method, geological structure information method, viscoelastic strain recovery method, etc. ("Research Status and Progress of Geostress Prediction Technology", Yin Xingyao et al., Petroleum Geophysical Exploration, July 2018).

[0006] Some scholars have proposed introducing the BP neural network method into the study of geostress field, selecting six parameters as the main indicators for geostress prediction research: depth, core density (natural density), core elastic modulus, core triaxial compressive strength (10MPa confining pressure), core acoustic emission geostress measurement, and core fracture rate (“Research on Geostress Prediction of Deeply Buried Tunnels Based on BP Neural Network”, Sun Weifeng et al., Journal of Geomechanics, September 2007.

[0007] In addition, some scholars have proposed to construct a shale rock physics equivalent model by analyzing the mineral, pore, fluid and anisotropic characteristics of the shale strata and equating it with a transversely isotropic medium with a vertical axis of symmetry. Then, empirical formulas for P-wave and S-wave velocities of shale formations were established, and both the empirical formulas and the rock physics equivalent model were applied to the prediction of S-wave velocities in actual shale working areas ("Research on the in-situ stress prediction method based on the shale rock physics equivalent model", Zhang Guangzhi et al., Chinese Journal of Geophysics, June 2015).

[0008] However, existing geostress prediction methods are difficult to accelerate and optimize, and alternative models are prone to getting trapped in local optima. In addition, the large amount of computation required for numerical analysis leads to high requirements for computer performance, which increases the difficulty of geostress prediction. Summary of the Invention

[0009] The purpose of this invention is to propose a geostress prediction method based on an accelerated optimization algorithm using an alternative model, in order to overcome the shortcomings of existing technologies.

[0010] The technical solution of the present invention is as follows:

[0011] This method mainly involves establishing a three-dimensional geological model based on geological data, combining N randomly selected stress boundary conditions, using finite difference software to solve for stress and obtain learning samples, then using a neural network to replace the model and a random search algorithm to predict the optimal boundary conditions, and finally using finite difference software to solve for stress and obtain the geostress field.

[0012] Specifically, the complete steps for implementing the technical solution of the present invention should include:

[0013] (1) Multiple sets of measured ground stress data in the study area were obtained through ground stress testing, providing data support for subsequent ground stress prediction models.

[0014] (2) Collect existing measured data of geostress around the study area, analyze the statistical data through fitting analysis, obtain the distribution characteristics of geostress with burial depth in the study area, and determine the form of boundary load applied by the geomechanical model in the process of geostress prediction.

[0015] (3) Through field investigation of the study area, geological data such as topographic and geological maps, exploration line profiles, and borehole information were obtained, and a three-dimensional geological model was established. Rock mechanics parameters were obtained through rock mechanics tests, and the three-dimensional geological model and rock mechanics parameters were imported into numerical analysis software to obtain a three-dimensional geomechanical model.

[0016] (4) Based on the obtained measured data of geostress, boundary condition forms and geomechanical models, a geostress prediction model with an alternative model is established by self-written program to accelerate the optimization algorithm.

[0017] (5) The optimal boundary conditions predicted by the geostress prediction model are written into the numerical analysis software through a programming language for calculation, thereby obtaining the prediction results of the geostress field of the underground rock mass.

[0018] This invention obtains multiple sets of measured geostress data in the study area through geostress testing methods. Based on determining the boundary conditions required for geostress prediction in the study area, a three-dimensional geological model is established using geological data. A geostress prediction model for the study area is then established by combining numerical analysis and machine learning. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the technical process of the present invention.

[0020] Figure 2 This demonstrates the distribution characteristics of the maximum horizontal principal stress with burial depth in the study area in one embodiment of the present invention;

[0021] Figure 3 This demonstrates the distribution characteristics of the minimum horizontal principal stress with burial depth in the study area in one embodiment of the present invention;

[0022] Figure 4 This illustrates how model boundary conditions are applied in one embodiment of the invention;

[0023] Figure 5 A schematic diagram of a neural network structure is shown in one embodiment of the present invention;

[0024] Figure 6 The training and testing results (training set performance) of the model in one embodiment of the present invention are shown.

[0025] Figure 7 The training and testing results (test set performance) of the model in one embodiment of the present invention are shown;

[0026] Figure 8 The study area stress prediction results (maximum principal stress) are shown in one embodiment of the present invention;

[0027] Figure 9 The results of geostress prediction (intermediate principal stress) in the study area are shown in one embodiment of the present invention;

[0028] Figure 10 The study area stress prediction results (minimum principal stress) are shown in one embodiment of the present invention;

[0029] Figure 11The results of geostress prediction (vertical stress) in the study area are shown in one embodiment of the present invention;

[0030] Figure 12 The results of geostress prediction (X-direction stress) in the study area are shown in one embodiment of the present invention;

[0031] Figure 13 The results of ground stress prediction (Y-direction stress) in the study area are shown in one embodiment of the present invention. Detailed Implementation

[0032] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so as to better understand the purpose, features and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are only for illustrating the essential spirit of the technical solution of the present invention.

[0033] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0034] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0035] Example 1:

[0036] like Figure 1 As shown, Figure 1 A flowchart of one embodiment of the present invention is shown, in which the prediction of ground stress includes the following steps:

[0037] (1) First, the in-situ stress test of the study area was carried out using the hole stress relief method, and the indoor test was carried out using the acoustic emission Kaiser effect stress test method. The stress test data of multiple measuring points in the study area were obtained, which provided data support for the subsequent stress prediction model.

[0038] (2) The key to successful geostress prediction lies in the rationality of the boundary conditions applied. For areas with large burial depths, complex topography, and intense geological processes, simply using lateral pressure coefficients or linear boundary conditions during rock mass geostress field inversion has significant biases and limitations. This invention proposes a regional geostress statistical analysis method to determine the boundary load form. Through fitting analysis, it is found that ( Figure 2-3 For complex geological regions, the geostress field exhibits obvious nonlinear distribution characteristics, and using nonlinear boundary conditions for geostress prediction is more in line with engineering practice.

[0039] (3) Through field investigation of the study area, geological features such as topography, strata lithology, and geological structure were obtained. Following the basic principles of geometric simulation and constitutive simulation, contour maps containing elevation information were drawn using AutoCAD software. These contour maps were then imported into Midas GTS NX software to obtain a geometric model of the study area containing the actual surface. After obtaining the geometric model, stratigraphic division was performed based on geological data such as the mine exploration line profile and borehole columnar section. The entity segmentation function of the software was used to segment the geometric model, thereby simulating geological structures such as faults. To enable computational capabilities, the created three-dimensional geological geometric model was meshed, with a mixed tetrahedral and hexahedral mesh type. Rock mechanics parameters were obtained through rock mechanics experiments and imported into numerical analysis software to obtain a three-dimensional geomechanical model.

[0040] (4) Based on the above steps, this invention further proposes an efficient global optimization algorithm that combines a neural network (BP) substitution model with a genetic algorithm (GA) random search. It is called an accelerated optimization algorithm because it combines the advantages of both the substitution model and the random search algorithm. The random search algorithm can continuously generate random points within a defined optimization interval, thus quickly finding the optimal interval; the substitution model can quickly find local optima within the optimal interval. The two iterate with each other, accelerating the attainment of the global optimal solution.

[0041] This method avoids the drawback of alternative models being prone to getting trapped in local optima, and also solves the problems of large computational load and high computer performance requirements in numerical analysis.

[0042] The specific implementation steps are as follows: First, randomly select N sets of boundary condition parameter sets, and generate N sets of samples corresponding to the calculated values ​​of boundary loads and stresses at measuring points based on numerical simulation software. The correspondence between boundary loads and measuring points is as follows: Figure 4As shown. After obtaining the learning samples, each set of boundary condition parameters is used as the input value of the neural network self-learning model, and the corresponding geostress data at the measuring points is used as the output value of the neural network self-learning model. A neural network self-learning model of the nonlinear mapping relationship between the set of boundary condition parameters and the stress values ​​at the measuring points is constructed using MATLAB software. Figure 5 An optimal model was trained as a replacement model, enabling it to output the calculated stress values ​​at the corresponding measuring points for each set of input boundary load parameters. The prediction accuracy of the model was verified using numerical analysis results. The model training effect is as follows: Figure 6-7 As shown. Then, the GA algorithm is used to perform global optimization of the parameters. The range of the parameters is defined by expanding the upper limit of the interval and narrowing the lower limit of the interval, and parameter sets can be randomly generated. The optimal parameter set is obtained by iteratively constructing the optimal approximation objective function (Equation 1) between the measured values ​​and the calculated values ​​of the measurement points.

[0043]

[0044] In the formula, i represents the measuring point number, n represents the total number of measuring points, and j represents the three principal stresses. The calculated value of the geostress vector at the measuring point for the random boundary condition parameter set x. This represents the measured value of ground stress.

[0045] (5) The optimal boundary conditions predicted by the geostress prediction model are written into numerical analysis software using a programming language for calculation, thereby obtaining the predicted results of the three-dimensional geostress field of the underground rock mass, such as... Figure 8-13 As shown in Table 1, the corresponding elements of the geostress measuring points in the study area are located in the model using coordinate positioning. The calculated geostress values ​​of the measuring point elements are extracted as prediction results and compared with the measured geostress values ​​in the study area.

[0046] Table 1 Comparison of Predicted and Actual Values

[0047]

[0048] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A geostress prediction method based on an accelerated optimization algorithm using an alternative model, characterized in that: The method includes: based on determining the form of boundary conditions required for predicting geostress in the study area, establishing a three-dimensional geological model through geological data, combining N randomly selected stress boundary conditions, using finite difference software to solve for stress to obtain learning samples, then using a neural network to replace the model and a random search algorithm to predict the optimal boundary conditions, and finally using finite difference software to solve for stress to obtain the geostress field. The stress solution to obtain the learning samples includes: randomly selecting N sets of boundary condition parameter groups, and generating N sets of samples corresponding to the boundary load and the stress calculation values ​​at the measuring points based on numerical simulation software. The method of using a neural network replacement model and a random search algorithm to predict optimal boundary conditions includes: after obtaining learning samples, using each set of boundary condition parameters as input values ​​to a neural network self-learning model, and using the corresponding geostress data at the measuring points as output values; constructing a neural network self-learning model with a nonlinear mapping relationship between the set of boundary condition parameters and the stress values ​​at the measuring points using MATLAB software; training the optimal model as a replacement model, enabling the model to output the corresponding stress calculation values ​​at the measuring points for each set of boundary load parameters input, and verifying the prediction accuracy of the model using numerical analysis results; then, using the GA algorithm to perform global optimization of the parameters, defining the range of parameter values ​​by expanding the upper limit of the interval and narrowing the lower limit of the interval, and randomly generating parameter sets; and continuously iterating by constructing the optimal approximation objective function between the measured values ​​and the calculated values ​​at the measuring points, thereby obtaining the optimal parameter set.

2. The geostress prediction method based on the accelerated optimization algorithm using an alternative model as described in claim 1, characterized in that: The required boundary condition form for predicting ground stress in the study area was determined by analyzing statistical data using a fitting analysis method.

3. The geostress prediction method based on the accelerated optimization algorithm using an alternative model according to claim 2, characterized in that: The statistical data is obtained through geological surveys, regional geostress statistical analysis, and geostress testing, or a combination thereof.

4. The geostress prediction method based on the accelerated optimization algorithm using an alternative model according to claim 3, characterized in that: The geostress test includes one or a combination of the following two methods: 1) conducting in-situ geostress tests in the study area using the borehole stress relief method; 2) conducting indoor tests using the acoustic emission Kaiser effect geostress test method.

5. The geostress prediction method based on the accelerated optimization algorithm using an alternative model according to claim 1, characterized in that: The process of establishing a three-dimensional geological model using geological data includes importing a drawn contour map into geotechnical analysis software to obtain a geometric model capable of stratigraphic division, and then using the geotechnical analysis software to segment the geometric model to simulate geological structures, including faults.

6. The geostress prediction method based on the accelerated optimization algorithm using an alternative model according to claim 5, characterized in that: The contour map was drawn using software including AutoCAD; the geotechnical analysis software included Midas GTSNX software.

7. The geostress prediction method based on the accelerated optimization algorithm using an alternative model according to claim 6, characterized in that: The process of establishing a three-dimensional geological model using geological data also includes: dividing the created three-dimensional geological geometric model into a mesh; obtaining rock mechanics parameters through rock mechanics tests and importing them into numerical analysis software to obtain a three-dimensional geomechanical model.

8. The geostress prediction method based on the accelerated optimization algorithm using an alternative model according to claim 1, characterized in that: The objective function formula is: (Equation 1) In the formula, i represents the measuring point number, n represents the total number of measuring points, and j represents the three principal stresses. The calculated value of the geostress vector at the measuring point for the random boundary condition parameter set x. This represents the measured value of ground stress.

Images