Ground stress inversion method and system for optimizing BP neural network based on genetic algorithm

Abstract

The invention provides a ground stress inversion method and system for optimizing a BP neural network based on a genetic algorithm. The method comprises the steps that modeling software FLAC3D is adopted to establish an experimental model; carrying out stress solution on the model; the solved data are imported into MATLAB, and the BP neural network is optimized based on the genetic algorithm in the MATLAB; and predicting the stress value of each point of the model. The technical problems that the initial crustal stress influence factors are not fully considered, the underground environment reduction effect is poor, the detection accuracy of the crustal stress field distribution rule is low, and the inversion operation is prone to interference due to the fact that the measuring point positions are unreasonably arranged are solved.

Description

Technical Field

[0001] The invention relates to the field of address and mining survey data processing, and in particular to a geostress inversion method and system based on genetic algorithm-optimized BP neural network. Background Art

[0002] Geostress is the fundamental driving force for deformation and destruction of surrounding rocks in underground projects such as coal mining. Stress field research is one of the most basic and important tasks in rock mechanics and mining engineering. However, the causes of geostress field are complex. In addition, there are problems such as complex geology and high costs in the actual measurement process, which makes the measured geostress discrete, low in precision, and can only reflect local stress fields. Therefore, how to accurately obtain the distribution law of geostress field is of great significance to the construction of underground projects and the prevention and control of deep geological disasters.

[0003] Commonly used methods for studying stress fields include multivariate linear regression and boundary load adjustment. Multivariate linear regression is used in practical engineering for inversion of geostress because of its simplicity, good fitting effect and high inversion accuracy. For example, the existing invention patent application document "A three-dimensional stress field inversion method and system considering geostress zoning characteristics" with publication number CN116796606A includes: obtaining the measured stress values ​​and formation mechanics parameter information of measuring points in different spatial parts of three-dimensional space; analyzing the measured stress values ​​by mathematical statistics, and dividing the engineering area into several geostress zones; establishing a finite element numerical analysis model based on the formation mechanics parameter information, and determining the calculated stress component values ​​of the measuring points at each stress zone under the corresponding load; establishing a regression equation considering the characteristics of geostress zoning and performing inversion analysis to obtain the first stress regression coefficient, and determining the comprehensive calculated stress value at the measuring point in combination with the calculated stress component value. However, the aforementioned existing method ignores the interaction between the various factors affecting the initial geostress due to its linear assumption, and the relationship between the initial geostress in actual conditions and the various factors affecting the initial geostress is difficult to express with a linear function. Artificial neural networks in intelligent algorithms are used in the inversion research of geostress fields because of their global search ability, generalization ability, and nonlinear mapping ability.

[0004] For example, the existing invention patent application document "Prediction Method of Geostress Field of Underground Cavern Groups in Deep River Valley Areas" with publication number CN114820949A includes the following steps: establishing a three-dimensional numerical calculation model of the underground cavern group, and using the measured geostress data of the underground cavern group to invert the three-dimensional geostress field, the inversion result is a boundary displacement combination of the three-dimensional numerical calculation model, the boundary displacement combination is applied to the three-dimensional numerical calculation model, and the initial geostress field of the cavern group is obtained by using the three-dimensional numerical calculation model; according to the initial geostress field of the cavern group, the excavation plan of the first layer of the underground cavern group is designed, and the excavation simulation of the first layer of the cavern group is performed using the three-dimensional numerical calculation model to predict the deformation and failure characteristics of the surrounding rock after the excavation of the first layer of the cavern group is completed; excavation is carried out according to the excavation plan of the first layer of the underground cavern group, and after the excavation is completed, observation is made. The deformation and failure characteristics of the surrounding rock. If the observed deformation and failure characteristics of the surrounding rock are within the preset error range with the predicted deformation and failure characteristics of the surrounding rock of the cavern after the first layer of the cavern group is excavated, the geostress field of the underground cavern group is the initial geostress field of the cavern group. Otherwise, the initial geostress field of the cavern group is secondary inversion. The secondary inversion process is as follows: based on the orthogonal experimental design, different boundary displacement combination samples are constructed, different boundary displacement samples are applied to the three-dimensional numerical calculation model, and the excavation process of the first layer of the cavern group is simulated to obtain the deformation and failure characteristics of the surrounding rock corresponding to each displacement combination sample, and a neural network of the boundary displacement combination and the deformation and failure characteristics of the surrounding rock is constructed. The observed deformation and failure characteristics of the surrounding rock are used as the inversion target to obtain a new boundary displacement combination, and then the new boundary displacement combination is applied to the three-dimensional numerical calculation model to obtain the geostress field of the cavern group area. However, when using the above existing methods, the established model cannot restore the complex underground environment well, and at the same time, the distribution law of the geostress field cannot be obtained more accurately.

[0005] In summary, the existing technology has technical problems such as the influencing factors of initial geostress are not fully considered, the underground environment restoration effect is poor, the detection accuracy of the distribution law of geostress field is low, and the unreasonable setting of measuring point positions makes the inversion operation susceptible to interference. Summary of the invention

[0006] The technical problem to be solved by the present invention is: how to solve the technical problems in the prior art that the influencing factors of initial geostress are not fully considered, the underground environment restoration effect is poor, the detection accuracy of the geostress field distribution law is low, and the unreasonable setting of measuring point positions makes the inversion operation susceptible to interference.

[0007] The present invention adopts the following technical solutions to solve the above technical problems: the inversion method of geostress based on genetic algorithm optimization BP neural network includes:

[0008] S1. Determine the coordinates of no less than 2 measuring points, collect the coal mine measurement parameters, and use the preset modeling tools to model the experimental model, assign parameters to the experimental model, and perform generalized zoning operations on the coal mine;

[0009] S2. Decompose the complex ground stress into at least two different boundary stress forms by using the method of stress field superposition, and apply normal stress load and shear stress load to the mine area model by using the preset modeling tool to solve the stress of the model and obtain the solution data of different working conditions;

[0010] S3. Import the data obtained from the difference working conditions into the preset modeling tool to build and train the BP neural network optimized based on the genetic algorithm, and use the BP neural network to perform inversion of deep mine ground stress to obtain the data of the points to be predicted;

[0011] S4. Import the data of the points to be predicted into the preset modeling tool, and predict the stress values ​​of each point.

[0012] The present invention calculates the error between the output value and the expected value, and uses back propagation to continuously adjust the weight and threshold to make the error between the output value and the expected value smaller. The model established by the present invention can better restore the complex underground environment, and can also more accurately obtain the distribution law of the ground stress field.

[0013] In a more specific technical solution, step S1 includes:

[0014] S11. Measure the preset coal mine to obtain the actual ground stress position as the measuring point coordinates;

[0015] S12. Measure and obtain coal mine measurement parameters based on the coordinates of the measuring points.

[0016] In a more specific technical solution, in step S11, the coordinates of the measuring point are in the non-top plate, bottom plate, and boundary area of ​​the model.

[0017] When the present invention establishes a model, the measuring point is located in the middle of the model, not close to the top plate, the bottom plate and the boundary of the established model, which is suitable for inversion calculation and avoids interference.

[0018] In a more specific technical solution, step S12 includes:

[0019] S121, selecting the area where the measuring point coordinates are located;

[0020] S122, obtaining mechanical property information and lithology interlayer regularity information of each rock layer in the area where the measuring point coordinates are located;

[0021] S123. Based on the mechanical property information and the lithology interlayer regularity information, generalized partitioning operations are performed on the area where the measuring point coordinates are located.

[0022] In a more specific technical solution, step S2 includes:

[0023] S21. According to the distribution of the measuring point coordinates, determine the stress increase and decrease conditions in the x, y, and z directions of the mine area model;

[0024] S22, using the preset modeling tool FLAC3D to apply different types of normal stress loads to each surface of the mine area model;

[0025] S23, applying shear stress loads to each side of the plane shear tectonic movement of the mine area model using shear boundary displacement;

[0026] S24, controlling and applying displacement load according to preset velocity boundary conditions in the form of differential boundary stress;

[0027] S25. Perform stress solving through the preset modeling tool FLAC3D to obtain solution data for no less than 2 different working conditions.

[0028] In a more specific technical solution, in step S3, the topological structure of the BP neural network includes: an input layer, a hidden layer and an output layer.

[0029] In a more specific technical solution, the threshold of the hidden layer is defined using the following logic:

[0030]

[0031] Use the following logic to define the weights from the nodes in the hidden layer to the nodes in the output layer:

[0032]

[0033] In a more specific technical solution, the activation function of the hidden layer neuron is set to the tansig function using the following logic:

[0034]

[0035] Using the following logic, the activation function of the output layer is set to purelin:

[0036] y=χ.

[0037] The model of the present invention is a multi-condition fitting, using uniform load, triangular load, trapezoidal load, shear displacement load application mode, a total of 27 working conditions, using BP neural network algorithm to achieve mine ground stress inversion, can more realistically restore the diversity and complexity of mine area ground stress. And using MATLAB to optimize the BP neural network algorithm based on genetic algorithm, the distribution law of the ground stress field can be accurately obtained.

[0038] In a more specific technical solution, in step S3, the mean square error function of the BP neural network model is defined using the following formula:

[0039]

[0040] In a more specific technical solution, the in-situ stress inversion system based on genetic algorithm optimization of BP neural network includes:

[0041] Modeling complex generalized partitioning module, used to determine the coordinates of no less than 2 measuring points, collect coal mine measurement parameters, and use preset modeling tools to model, assign parameters to the experimental model, and perform generalized partitioning operations on the coal mine;

[0042] The differential working condition solving module is used to decompose the complex ground stress into at least two differential boundary stress forms by using the method of stress field superposition, and then use the preset modeling tool to apply normal stress load and shear stress load to the mine area model, solve the stress of the model to obtain differential working condition solution data. The differential working condition solving module is connected with the complex generalized partitioning module of modeling;

[0043] The BP neural network builds an optimization module, which is used to import the differential working condition solution data into the preset modeling tool, so as to build and train the BP neural network based on the genetic algorithm optimization, and use the BP neural network to perform deep mine ground stress inversion, thereby obtaining the data of the points to be predicted. The BP neural network builds an optimization module and is connected with the differential working condition solution module;

[0044] The stress value prediction module is used to import the data of the point to be predicted into the preset modeling tool, and predict the stress value of each point. The stress value prediction module is connected with the BP neural network optimization module.

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] The present invention calculates the error between the output value and the expected value, and uses back propagation to continuously adjust the weight and threshold to make the error between the output value and the expected value smaller. The model established by the present invention can better restore the complex underground environment, and can also more accurately obtain the distribution law of the ground stress field.

[0047] When the present invention establishes a model, the measuring point is located in the middle of the model, not close to the top plate, the bottom plate and the boundary of the established model, which is suitable for inversion calculation and avoids interference.

[0048] The model of the present invention is a multi-condition fitting, using uniform load, triangular load, trapezoidal load, shear displacement load application mode, a total of 27 working conditions, using BP neural network algorithm to achieve mine ground stress inversion, can more realistically restore the diversity and complexity of mine area ground stress. And using MATLAB to optimize the BP neural network algorithm based on genetic algorithm, the distribution law of the ground stress field can be accurately obtained.

[0049] The present invention solves the technical problems existing in the prior art that the influencing factors of initial geostress are not fully considered, the underground environment restoration effect is poor, the detection accuracy of the geostress field distribution law is low, and the unreasonable setting of the measuring point position makes the inversion operation susceptible to interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a schematic diagram of the basic steps of the in-situ stress inversion method based on genetic algorithm optimization of BP neural network according to Example 1 of the present invention;

[0051] Figure 2 This is a schematic diagram of a BP neural network model optimized based on a genetic algorithm according to Embodiment 1 of the present invention;

[0052] Figure 3 This is a schematic diagram of specific steps for optimizing a BP neural network based on a genetic algorithm according to Embodiment 1 of the present invention;

[0053] Figure 4 A curve diagram of the actual value and the predicted value of the training set of Example 2 of the present invention;

[0054] Figure 5 This is a curve chart of the true value and predicted value of the test set of Example 2 of the present invention. DETAILED DESCRIPTION

[0058] Example 1

[0059] like Figure 1 As shown, the in-situ stress inversion method based on genetic algorithm optimization of BP neural network provided by the present invention includes the following basic steps:

[0060] S1. Use modeling software FLAC3D to establish the experimental model;

[0061] In this embodiment, according to the location of the actual ground stress provided by the Xieqiao Coal Mine, the coordinates of the measuring points and the important parameters obtained by the measurement are determined, and the modeling software FLAC3D is used to simulate the area and assign relevant parameters to the model.

[0062] This simulation selected an area where five measuring points are located. The measuring area of ​​Xieqiao Mine was divided into generalized areas according to the mechanical properties of each rock layer, the rules of lithological interlayers, etc. According to the rules of inversion after multiple model building, when building the model, the positions of the five measuring points should be located in the middle of the model, and should not be close to the roof, floor and the boundary of the built model, so as to be suitable for inversion calculation and avoid interference.

[0063] like Figure 2 As shown, in this embodiment, the width of the model mining area is 7000m, the length is 3000m, and the depth is 600m. The coordinates of the five measuring points are shown in Table 1 below:

[0064] Table 1 Location coordinates of measured points in Xieqiao Mine

[0065]

[0066] S2, solving stress of the model;

[0067] In this embodiment, the geostress of the deep well is complex and the underground space is almost completely closed. Therefore, the method of superposition of finger stress field is adopted to decompose the complex geostress into 27 different boundary stress forms, that is, 27 working conditions. In this embodiment, according to the distribution of 5 measuring points, the working conditions of reducing or increasing in the x, y, and z directions are judged; FLAC3D is used to apply uniform loads to the z surface of the mine area model, triangular and trapezoidal loads along the x direction, and trapezoidal loads along the y direction, and positive stress; uniform loads are applied to the x surface, and positive stress is applied to the trapezoidal loads along the y and z directions; uniform loads are applied to the y surface, and positive stress is applied to the trapezoidal loads along the x and z directions.

[0068] In this embodiment, shear stress is applied on each side of the shear structural motion in the xy, xz, and yz planes by, for example, a shear boundary displacement of 600 mm, and the displacement load applied is controlled according to the velocity boundary conditions of uniform, triangular, and trapezoidal forms, the boundary velocity is 3e-4, and the running step is set to, for example, 2000 steps. Thus, 27 working conditions are obtained. In this embodiment, stress is solved by FLAC3D, and finally all data under the 27 working conditions are obtained.

[0069] S3, import the solved data into MATLAB, and optimize the BP neural network based on the genetic algorithm in MATLAB;

[0070] In this embodiment, the in-situ stress inversion method based on genetic algorithm optimization of BP neural network is applied to MATLAB to build and train the genetic algorithm optimization BP neural network model;

[0071] In this embodiment, the data set is divided into training samples and test samples in a ratio of 4:1. The number of hidden layer nodes is 7, the population size is 5, and the learning rate is 0.01;

[0072] like Figure 3 As shown, in this embodiment, the topological structure of the BP neural network model includes but is not limited to: an input layer, a hidden layer and an output layer.

[0073] In this embodiment, the related functions of the aforementioned BP neural network model include but are not limited to:

[0074] In this embodiment, the threshold of the hidden layer is defined by the following formula:

[0075]

[0076] In this embodiment, the weights from the nodes of the hidden layer to the nodes of the output layer are defined by the following formula:

[0077]

[0078] In the selection process of this embodiment, geometric planning sorting selection is selected to perform individual optimization; in the crossover process of this embodiment, arithmetic mean is used for crossover; in the mutation process of this embodiment, non-uniform mutation is used;

[0079] In this embodiment, the activation function of the hidden layer neuron is set to the tansig function using the following formula, where the expression of tansig is:

[0080]

[0081] In this embodiment, the activation function of the output layer is set to purelin, where the expression of purelin is:

[0082] y=χ

[0083] In this embodiment, the mean square error function of the BP neural network model is defined by the following formula:

[0084]

[0085] S4. Predict the stress values ​​at each point of the model.

[0086] In this embodiment, the data of the points to be predicted are imported into MATLAB, so that the stress value can be predicted.

[0087] Example 2

[0088] In this embodiment, an experiment is conducted on the geostress inversion method based on genetic algorithm optimization of BP neural network provided by the present invention, and the experimental results are obtained and analyzed.

[0089] like Figure 4 and Figure 5 As shown, in this embodiment, experiments are conducted on the training samples and test samples in the above-mentioned embodiment 1, and the true value and predicted value curve of the training set are shown in Figure 4 , the true value and predicted value curve of the test set can be found in Figure 5 .

[0090] In this embodiment, the absolute errors between the principal stresses of the measured values ​​of the in-situ stress at the five measuring points and the principal stresses of the predicted values ​​obtained by optimizing the BP neural network based on the genetic algorithm are shown in Table 2, indicating that the model training effect is good:

[0091] Table 2. Comparison of the errors between the principal stresses of the measured values ​​of geostress and the principal stresses predicted by the BP algorithm optimized by the genetic algorithm

[0092]

[0093] The present invention calculates the error between the output value and the expected value, and uses back propagation to continuously adjust the weight and threshold to make the error between the output value and the expected value smaller. The model established by the present invention can better restore the complex underground environment, and can also more accurately obtain the distribution law of the ground stress field.

[0094] When the present invention is building a model, the measuring point should be located in the middle of the model, not close to the top plate, the bottom plate and the boundary of the built model, so that it is suitable for inversion calculation and avoids interference.

[0095] The model of the present invention is a multi-condition fitting, using uniform load, triangular load, trapezoidal load, shear displacement load application mode, a total of 27 working conditions, using BP neural network algorithm to achieve mine ground stress inversion, can more realistically restore the diversity and complexity of mine area ground stress. And using MATLAB to optimize the BP neural network algorithm based on genetic algorithm, the distribution law of the ground stress field can be accurately obtained.

[0096] The present invention solves the technical problems existing in the prior art that the influencing factors of initial geostress are not fully considered, the underground environment restoration effect is poor, the detection accuracy of the geostress field distribution law is low, and the unreasonable setting of the measuring point position makes the inversion operation susceptible to interference.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The in-situ stress inversion method based on genetic algorithm optimized BP neural network is characterized by: The method comprises: S1. Determine the coordinates of no less than 2 measuring points, collect the coal mine measurement parameters, and use the preset modeling tools to model the experimental model, assign parameters to the experimental model, and perform generalized zoning operations on the coal mine; S2. Decomposing the complex ground stress into at least two different boundary stress forms by using the method of stress field superposition, applying normal stress load and shear stress load to the mine area model by using the preset modeling tool, solving the stress of the model to obtain solution data of different working conditions; S3, importing the difference working condition solution data into the preset modeling tool to build and train a BP neural network based on genetic algorithm optimization, and using the BP neural network to perform deep mine ground stress inversion to obtain the point data to be predicted; S4, importing the data of the points to be predicted into the preset modeling tool, and predicting the stress values ​​of each point.

2. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 1 is characterized in that: The step S1 comprises: S11, measuring the preset coal mine to obtain the actual ground stress position as the measuring point coordinates; S12. Measure and obtain the coal mine measurement parameters according to the measurement point coordinates.

3. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 2 is characterized in that: In the step S11, the coordinates of the measuring points are located in the non-top plate, non-bottom plate, and non-boundary areas of the model.

4. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 2 is characterized in that: The step S12 comprises: S121, selecting the area where the measuring point coordinates are located; S122, obtaining mechanical property information and lithology interlayer regularity information of each rock layer in the area where the measuring point coordinates are located; S123: performing the generalized partitioning operation on the area where the measuring point coordinates are located according to the mechanical property information and the lithology interlayer regularity information.

5. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 1 is characterized in that: The step S2 comprises: S21, judging the stress increase and decrease conditions in the x, y, and z directions of the mine area model according to the distribution of the measuring point coordinates; S22, using the preset modeling tool FLAC3D to apply different types of normal stress loads to each surface of the mine area model; S23, applying the shear stress load to each side of the plane shear tectonic movement of the mine area model by using shear boundary displacement; S24, controlling and applying displacement load according to the differential boundary stress form and preset velocity boundary conditions; S25. Perform stress solving through the preset modeling tool FLAC3D to obtain solution data for no less than 2 different working conditions.

6. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 1, characterized in that: In step S3, the topological structure of the BP neural network includes: an input layer, a hidden layer and an output layer.

7. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 6 is characterized in that: The threshold of the hidden layer is defined using the following logic: The weights from the nodes of the hidden layer to the nodes of the output layer are defined using the following logic:

8. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 6 is characterized in that: Using the following logic, the activation function of the hidden layer neurons is set to the tansig function: Using the following logic, the activation function of the output layer is set to purelin: y=χ.

9. The in-situ stress inversion method based on genetic algorithm optimized BP neural network according to claim 1, characterized in that: In step S3, the mean square error function of the BP neural network model is defined by the following formula:

10. The in-situ stress inversion system based on genetic algorithm optimized BP neural network is characterized by: The system comprises: Modeling complex generalized partitioning module, used to determine the coordinates of no less than 2 measuring points, collect coal mine measurement parameters, and use preset modeling tools to model, assign parameters to the experimental model, and perform generalized partitioning operations on the coal mine; A differential working condition solving module is used to decompose the complex ground stress into at least two differential boundary stress forms by using the method of stress field superposition, and to apply normal stress load and shear stress load to the mine area model by using the preset modeling tool, and to solve the stress of the model to obtain differential working condition solving data, and the differential working condition solving module is connected to the modeling complex generalized partitioning module; A BP neural network construction optimization module is used to import the difference working condition solution data into the preset modeling tool to build and train a BP neural network based on a genetic algorithm optimization, and use the BP neural network to perform deep mine ground stress inversion to obtain the point data to be predicted. The BP neural network construction optimization module is connected to the difference working condition solution module; The stress value prediction module is used to import the data of the point to be predicted into the preset modeling tool to predict the stress value of each point. The stress value prediction module is connected to the BP neural network construction optimization module.

Images