A Prediction Method for Alteration Zone Extension Distribution Based on Geostress Inversion and Machine Learning

Abstract

This invention discloses a method for predicting the extension and distribution of alteration zones based on geostress inversion and machine learning. First, based on geological survey data of the engineering site, historical tectonic deformation data, geostress test data, rock mechanics parameters, and spatial coordinates of the engineering site are collected. Then, a geostress field correlation model and a three-dimensional geological numerical model of the engineering site are constructed, and the evolution process of the geostress field in the engineering site is obtained through inversion. Next, a database of the relationship between alteration zones and stress fields is constructed to establish the mechanical relationship between the geometric characteristic parameters of alteration zones and the geostress state. Finally, an alteration zone extension and distribution prediction model is constructed to predict the extension and distribution of alteration zones. This invention reveals the geometric characteristics and extension patterns of alteration zones under tectonic deformation, providing guidance for the accurate exploration and spatial distribution prediction of alteration zones in hydropower projects.

Description

Technical Field

[0001] This invention relates to the field of hydropower engineering technology, specifically a method for predicting the extension and distribution of alteration zones based on geostress inversion and machine learning. Background Technology

[0002] Under tectonic stress, engineering rock masses undergo varying degrees of deformation, damage, fragmentation, or dynamic recrystallization, altering their mineral composition and microstructure. This results in compromised rock integrity, increased porosity and fissures, and a significant reduction in strength and hardness, characterized by short self-stabilization time, rapid stress adjustment, large deformation, and long duration. Most altered rocks exhibit fractured structures, low strength, and extremely poor stability. Without effective measures during tunnel excavation, they are highly susceptible to large deformations, collapses, and instability of the surrounding rock. Furthermore, the numerous pores and fissures within the altered rock mass provide seepage channels for groundwater. Altered rocks possess strong water-softening, swelling, and disintegration properties; increased permeability after tunnel excavation easily induces water inrush and mudslide disasters. For example, the two water diversion tunnels of the Yingliangbao Hydropower Station experienced water inrush and mudslides due to encountering altered rock, forcing work stoppages and route rerouting, causing delays and posing significant engineering geological safety challenges to the project.

[0003] Tectonic alteration zones are mostly located in tectonically active areas. The formation process of altered rocks is complex, and their extension and development span large spatial and temporal scales, with intricate mechanisms. Due to the lack of effective methods for predicting the spatial distribution of tectonic alteration zones, the accuracy of current tectonic alteration zone surveys is insufficient. Therefore, there is an urgent need to develop methods for predicting the extension and distribution of tectonic alteration zones in hydropower projects, providing guidance for the structural design and safe construction of underground engineering projects within tectonic alteration zones. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method for predicting the extension and distribution of alteration zones based on geostress inversion and machine learning, which reveals the geometric characteristics and extension law of alteration zones under tectonic deformation, and provides guidance for the accurate exploration and spatial distribution prediction of alteration zones in hydropower projects.

[0005] The technical solution of this invention is as follows: The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning includes the following steps: (1) Based on the geological survey data of the engineering site, collect historical data on structural deformation, geostress test data, rock mechanics parameters and spatial coordinates of the engineering site; (2) Based on the historical data of structural deformation and the geostress test data of the engineering site, construct the geostress field correlation model of the engineering site. That is, based on the historical data of structural deformation, establish a stratigraphic geometric model containing the corresponding fault zone level, and then combine the geostress test data to obtain the quantitative relationship between structural deformation and geostress field evolution. (3) Based on the geological survey data of the engineering site, a three-dimensional geological numerical model of the engineering site is constructed, and the GPS velocity field of the engineering site is used as the boundary condition for structural deformation to invert the evolution process of the geostress field of the engineering site. (4) Based on the alteration zone survey data and geostress test data of the entire engineering process, construct a database of the relationship between alteration zone and stress field, thereby establishing the mechanical relationship between the geometric characteristic parameters of alteration zone and geostress state; (5) Construct an alteration zone extension distribution prediction model. Input the geostress field correlation model, the geostress field evolution process of the engineering site, the mechanical relationship between the geometric characteristic parameters of the alteration zone and the geostress state, as well as the rock mass mechanical parameters and spatial coordinates into the alteration zone extension distribution prediction model to predict the degree of alteration at each discrete point under the spatial coordinates of the engineering site, that is, the alteration category of the rock mass at each discrete point.

[0006] The geological survey data of the project site includes geological environment, hydrogeology and geostress conditions of the active fault area of ​​the project site. The geostress field correlation model was constructed using the finite element analysis software Abaqus. The three-dimensional geological numerical model was constructed using three-dimensional modeling software. Then, the evolution process of the geostress field in the engineering site was obtained by using the GPS velocity field of the engineering site as the boundary condition for structural deformation in the finite element analysis software Abaqus.

[0007] The geological geometry model includes three fault levels: Level I, Level II, and Level III. Level I faults refer to faults with a total surface fracture length exceeding 100 kilometers and a fracture depth exceeding 10 kilometers. Level II faults refer to faults with a total surface fracture length of tens of kilometers and a fracture depth of several kilometers. Level III faults refer to faults with a total surface fracture length of kilometers and a fracture depth of less than one kilometer. A numerical model at the tectonic scale of 100 kilometers is established for Level I faults, a numerical model at the scale of 30 kilometers is established for Level II faults, and a numerical model at the scale of the engineering site of 5 kilometers is established for Level III faults.

[0008] The quantitative relationship between tectonic deformation and geostress field characteristics is obtained by combining geostress test data, specifically as follows: (a) Set the tangential friction properties of the fault zone geometry and analyze the influence trend of the continuous slip dislocation of the strike-slip fault on the geostress field; (b) Set the normal contact properties and cross-sectional dip angle of the fault zone geometry, and analyze the trend of the influence of continuous thrusting of the compression fault on the geostress field; (c) Summarize the stress field distribution patterns under different compression and strike-slip modes, analyze the mapping relationship between tectonic deformation and geostress field evolution, and establish a quantitative relationship between tectonic deformation and geostress field characteristics.

[0009] The three-dimensional geological numerical model selects appropriate constitutive models and parameter assignments based on the characteristics of the rock mass: the tectonic scale of the shallow upper crustal rock mass is mainly brittle failure, so a linear elastic constitutive model is adopted, and the strength characteristics are described by the Mohr-Coulomb criterion; a viscoelastic model is adopted for the middle and lower crust; fault zones and tectonic alteration zones are set as weak zones or regional zones, and the Young's modulus parameter of the overall medium is multiplied by a weakening coefficient of 1 / 10 to 1 / 3.

[0010] During the evolution of the geostress field in the engineering site, the local geostress data obtained from the hydraulic fracturing test of the borehole in the site is used as a constraint to verify the accuracy of the simulation results of the three-dimensional geological numerical model within the point range. The calculation is repeated multiple times until the error between the geostress results obtained by the numerical simulation of the three-dimensional geological numerical model and the measured data is within the set allowable range. The distribution characteristics of the initial geostress field are restored, so as to carry out the long-term geostress extrapolation simulation in the future.

[0011] The alteration zone and stress field relationship database includes spatial location, type, number and development characteristics of alteration zone rocks at the corresponding location, and geostress parameters at the corresponding location.

[0012] The alteration zone extension distribution prediction model is based on an artificial neural network model, which includes an input layer, a hidden layer, an output layer, and a Softmax activation function. The number of neurons in the input layer is equal to the feature dimension of the input data. The hidden layer includes two fully connected layers and a ReLU activation function. The number of neurons in the output layer is equal to the total number of rock categories in the alteration zone. The Softmax activation function converts the output categories into a probability distribution, i.e., the probability of rock of the k-th alteration zone category.

[0013] The alteration zone extension distribution prediction model is trained using alteration zone data confirmed in the early stage of the project, verified using alteration zone data confirmed in the middle stage of the project construction, and tested using alteration zone data confirmed after excavation in the later stage of the project construction.

[0014] After the alteration degree prediction of each discrete point under the spatial coordinates of the engineering site is completed, the prediction results of all discrete points are imported into a geographic information system or 3D modeling software. Kriging interpolation is used to generate a continuous distribution map, and visualization processing is performed to generate an alteration zone extension distribution prediction map, including a plan view, a cross view and a 3D model, indicating the spatial distribution of rock types in different alteration zones. Finally, the alteration zone extension distribution prediction map and the corresponding confidence distribution map are output.

[0015] Advantages of this invention: (1) The geostress field correlation model constructed in this invention breaks through the problem that traditional geomechanical models cannot accurately reproduce the influence of 10-kilometer-level fault deformation on the long-term evolution of geostress field. By using the finite element analysis software Abaqus to establish a stress field correlation model from the regional tectonic scale to the field scale, the mapping relationship between the deformation of the compression and strike-slip composite fault zone and the evolution of geostress field is quantitatively revealed. The dynamic background and characteristic elements of the alteration zone are accurately analyzed, providing a dynamic perspective for the formation of the alteration zone.

[0016] (2) This invention constructs a three-dimensional geological numerical model of the engineering site, inverts the evolution process of the geostress field, explores the stress accumulation process of the full-scale geological model under millions of years of geological action, and reveals the current regional distribution law of tectonic stress.

[0017] (3) By integrating information such as geological exploration and geostress field inversion, this invention constructs an alteration zone extension distribution prediction model based on artificial neural network (ANN), which realizes in-depth exploration of the complex nonlinear relationship between tectonic deformation characteristics, geostress state and alteration zone distribution, making the alteration zone extension development conformity reach 50%, significantly improving the prediction accuracy of alteration zone spatial distribution, and providing reliable technical support for underground engineering layout and disaster prevention and control. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention.

[0019] Figure 2 This is a schematic diagram of the three-dimensional geological numerical model of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] See Figure 1 The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning includes the following steps: (1) Based on the geological survey data of the engineering site, collect historical data on tectonic deformation, geostress test data, rock mechanics parameters and spatial coordinates of the engineering site; the geological survey data of the engineering site includes geological environment, hydrogeological and geostress condition survey data of active fault areas in the engineering site. The project site is located in the transition zone from the eastern edge of the Qinghai-Tibet Plateau to the Sichuan Basin, at the intersection of the Sichuan-Yunnan SN-trending tectonic belt, the NE-trending Longmenshan tectonic belt, and the NW-trending Xianshuihe tectonic belt. The main active fault zones within the site are F1, F2, F4, and F5. The region is basically in a modern tectonic stress field dominated by the NW-SE-EY horizontal principal compressive stress and the NNE-SSW horizontal principal stress. The maximum principal stress σ1 in the site area is 10.83MPa~24.73MPa, the minimum principal stress σ3 is 3.16MPa~6.74MPa, and the intermediate stress σ2 is 6.32MPa~12.92MPa. The ratio of σ1:σ2:σ3 is 1:0.6:0.2. The direction of the maximum principal stress is N60°W-S80.9°W, which is classified as medium to high ground stress. (2) Based on the historical tectonic deformation data and geostress test data of the engineering site, a geostress field correlation model of the engineering site is constructed using the finite element analysis software Abaqus. That is, based on the historical tectonic deformation data, a stratigraphic geometric model containing the corresponding fault zone level is established, and then combined with the geostress test data, the quantitative relationship between tectonic deformation and geostress field evolution is obtained, specifically: S21. For the active fault zones F1, F2, F4, and F5 mainly distributed in the engineering site area, establish fault zone geometry based on fault strike and deep dip, and establish stratigraphic geometric models containing three fault levels respectively. Based on the structural deformation characteristics of the fault zones, the fault levels are divided as follows: Level I faults (total surface rupture length exceeding 100 kilometers and fault depth exceeding 10 kilometers) establish a 100-kilometer-scale structural numerical model; Level II faults (total surface rupture length of tens of kilometers and fault depth of several kilometers) establish a 30-kilometer-scale numerical model; and Level III faults (total surface rupture length of kilometers and fault depth of less than one kilometer) establish a 5-kilometer-scale numerical model for the engineering site area. S22. Based on the geostress test data, obtain the quantitative relationship between tectonic deformation and geostress field characteristics, specifically: (a) Set the tangential friction properties of the fault zone geometry and analyze the influence trend of the continuous slip dislocation of the strike-slip fault on the geostress field; (b) Set the normal contact properties and cross-sectional dip angle of the fault zone geometry, and analyze the trend of the influence of continuous thrusting of the compression fault on the geostress field; (c) Summarize the stress field distribution patterns under different compression and strike-slip modes, analyze the mapping relationship between tectonic deformation and geostress field evolution, and establish a quantitative relationship between tectonic deformation and geostress field characteristics; focus on analyzing the control effect of fault F5 on geostress field in the Yingliangbao site area, and determine the main controlling factors of tectonic deformation on geostress field evolution. (3) Based on the geological survey data of the engineering site, a three-dimensional geological numerical model of the engineering site was constructed using the three-dimensional modeling software Solidworks. Then, the evolution process of the geostress field of the engineering site was obtained by using the GPS velocity field of the engineering site as the boundary condition for structural deformation in the finite element analysis software Abaqus. S31, Three-dimensional geological numerical model (see...) Figure 2 According to the characteristics of the rock mass, select the appropriate constitutive model and parameter assignment: the tectonic scale of the shallow upper crust rock mass is mainly brittle failure, so adopt the linear elastic constitutive model and describe the strength characteristics by the Mohr-Coulomb criterion; the middle and lower crust adopt the viscoelastic model; fault zones and tectonic alteration zones are set as weak zones or regional zones, and the Young's modulus parameter of the whole medium is multiplied by the weakening coefficient 1 / 10 to 1 / 3; S32. During the evolution of the geostress field in the engineering site, the local geostress data obtained from the hydraulic fracturing test of the borehole in the site is used as a constraint to verify the accuracy of the simulation results of the three-dimensional geological numerical model within the point range. The calculation is repeated until the error between the geostress results obtained by the numerical simulation of the three-dimensional geological numerical model and the measured data is within the set allowable range. The distribution characteristics of the initial geostress field are restored, so as to carry out the long-term geostress extrapolation simulation in the future. (4) Based on the alteration zone survey data and geostress test data of the entire engineering process, construct a database of the relationship between alteration zone and stress field, thereby establishing the mechanical relationship between the geometric characteristic parameters of alteration zone and geostress state; The alteration zone and stress field relationship database includes spatial location, the type, number, and development characteristics of the alteration zone rocks at the corresponding location, as well as the geostress parameters at the corresponding location, as detailed in Table 1 below: Table 1

[0022] By collecting the structural alteration zone survey data of the entire phase of the Yingliangbao Hydropower Station project, it was found that the complex geological tectonic process created the widely distributed diabase alteration zone in the project area. The altered diabase can be divided into six categories: micro-altered diabase, fractured diabase, fractured diabase, diabase breccia, foliated diabase, and argillaceous diabase. Geometric characteristic parameters such as alteration zone attitude, thickness, extension length, and spatial curvature are extracted using geostatistical methods. The mechanical relationship between these parameters and the geostress state is then established. The specific technical process is as follows: First, the database of the correspondence between alteration zone development characteristics and tectonic stress fields was registered using spatial coordinates. Sampling points were then placed at fixed intervals along the strike of each alteration zone, and the stress values ​​and geometric parameters at each point were extracted to form sample pairs. Then, the thickness of the alteration zone was used as the basis for further analysis. For example, let's construct a multiple linear regression model: ,in, This represents the thickness of the alteration zone. , These represent the maximum and minimum principal stresses, respectively. Represents the maximum shear stress. Represents the lateral pressure coefficient. For constant terms, , , and As regression coefficients, stress components that play a dominant role in the development of alteration zones are screened through significance tests; then, based on the Coulomb-Mohr failure criterion, the mechanical relationship between alteration zone thickness and geostress state is established: ,in, Represents an empirical coefficient. The normal stress representing the alteration zone. The internal friction angle representing the alteration zone. The initial thickness of the alteration zone is represented by this mechanical relationship, which shows that the greater the shear stress and the lower the rock mass strength, the greater the thickness of the alteration zone. This reflects the control and quantitative relationship between tectonic stress and the development of the alteration zone. (5) Construct an alteration zone extension distribution prediction model, input the geostress field correlation model, the geostress field evolution process of the engineering site, the mechanical relationship between the geometric characteristic parameters of the alteration zone and the geostress state, as well as the rock mass mechanical parameters and spatial coordinates into the alteration zone extension distribution prediction model, and predict the degree of alteration at each discrete point under the spatial coordinates of the engineering site, that is, the alteration category of the rock mass at each discrete point. Before inputting the geostress field correlation model, the geostress field evolution process of the engineering site, the mechanical relationship between the geometric characteristic parameters of the alteration zone and the geostress state, and the rock mass mechanical parameters and spatial coordinates into the alteration zone extension distribution prediction model, data preprocessing is performed. Samples with too many missing values ​​are removed, and K-nearest neighbor interpolation is used for the remaining missing values. Outliers are identified and processed based on box plots. Z-score standardization is used to eliminate the influence of dimensions, transforming each feature into a distribution with a mean of 0 and a standard deviation of 1. The standardized feature vector and the corresponding rock type of the alteration zone (micro-altered diabase, fractured diabase, fractured diabase, diabase breccia, foliated diabase, argillaceous diabase) are output. The alteration zone extension distribution prediction model is based on an artificial neural network model, which includes an input layer, a hidden layer, an output layer, and a Softmax activation function. The number of neurons in the input layer is equal to the feature dimension of the input data. The hidden layer includes two fully connected layers and a ReLU activation function. The number of neurons in the output layer is equal to the total number of rock categories in the alteration zone. The Softmax activation function converts the output categories into a probability distribution, i.e., the probability of rock of the k-th alteration zone category. The alteration zone extension distribution prediction model was trained using alteration zone data confirmed in the early stage of the project, validated using alteration zone data confirmed in the middle stage of the project construction, and tested using alteration zone data confirmed after excavation in the later stage of the project construction. The loss function used for training the alteration zone extension distribution prediction model employs cross-entropy loss as the optimization objective: Where N represents the batch size. Represents the actual label (the actual rock type of the alteration zone). To predict probabilities, This represents the total number of rock types in the alteration zone; the Adam optimizer is used to update network weights and biases via gradient descent, minimizing the cross-entropy loss. (6) After the alteration degree prediction of each discrete point under the spatial coordinates of the engineering site is completed, the prediction results of all discrete points are imported into the Geographic Information System (GIS) or 3D modeling software. Kriging interpolation is used to generate a continuous distribution map and perform visualization processing to generate an alteration zone extension distribution prediction map, including a plan view, a profile view and a 3D model, indicating the spatial distribution of different alteration zone rock types. Finally, the alteration zone extension distribution prediction map and the corresponding confidence distribution map are output.

[0023] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning, characterized in that: It includes the following steps: (1) Based on the geological survey data of the engineering site, collect historical data on structural deformation, geostress test data, rock mechanics parameters and spatial coordinates of the engineering site; (2) Based on the historical data of structural deformation and the geostress test data of the engineering site, construct the geostress field correlation model of the engineering site. That is, based on the historical data of structural deformation, establish a stratigraphic geometric model containing the corresponding fault zone level, and then combine the geostress test data to obtain the quantitative relationship between structural deformation and geostress field evolution. (3) Based on the geological survey data of the engineering site, a three-dimensional geological numerical model of the engineering site is constructed, and the GPS velocity field of the engineering site is used as the boundary condition for structural deformation to invert the evolution process of the geostress field of the engineering site. (4) Based on the alteration zone survey data and geostress test data of the entire engineering process, construct a database of the relationship between alteration zone and stress field, thereby establishing the mechanical relationship between the geometric characteristic parameters of alteration zone and geostress state; (5) Construct an alteration zone extension distribution prediction model. Input the geostress field correlation model, the geostress field evolution process of the engineering site, the mechanical relationship between the geometric characteristic parameters of the alteration zone and the geostress state, as well as the rock mass mechanical parameters and spatial coordinates into the alteration zone extension distribution prediction model to predict the degree of alteration at each discrete point under the spatial coordinates of the engineering site, that is, the alteration category of the rock mass at each discrete point.

2. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: The geological survey data of the project site includes geological environment, hydrogeology and geostress conditions of the active fault area of ​​the project site. The geostress field correlation model was constructed using the finite element analysis software Abaqus. The three-dimensional geological numerical model was constructed using three-dimensional modeling software. Then, the evolution process of the geostress field in the engineering site was obtained by using the GPS velocity field of the engineering site as the boundary condition for structural deformation in the finite element analysis software Abaqus.

3. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: The geological geometry model includes three fault levels: Level I, Level II, and Level III. Level I faults refer to faults with a total surface fracture length exceeding 100 kilometers and a fracture depth exceeding 10 kilometers. Level II faults refer to faults with a total surface fracture length of tens of kilometers and a fracture depth of several kilometers. Level III faults refer to faults with a total surface fracture length of kilometers and a fracture depth of less than one kilometer. A numerical model at the tectonic scale of 100 kilometers is established for Level I faults, a numerical model at the scale of 30 kilometers is established for Level II faults, and a numerical model at the scale of the engineering site of 5 kilometers is established for Level III faults.

4. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 3, characterized in that: The quantitative relationship between tectonic deformation and geostress field characteristics is obtained by combining geostress test data, specifically as follows: (a) Set the tangential friction properties of the fault zone geometry and analyze the influence trend of the continuous slip dislocation of the strike-slip fault on the geostress field; (b) Set the normal contact properties and cross-sectional dip angle of the fault zone geometry, and analyze the trend of the influence of continuous thrusting of the compression fault on the geostress field; (c) Summarize the stress field distribution patterns under different compression and strike-slip modes, analyze the mapping relationship between tectonic deformation and geostress field evolution, and establish a quantitative relationship between tectonic deformation and geostress field characteristics.

5. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: The three-dimensional geological numerical model selects appropriate constitutive models and parameter assignments based on the characteristics of the rock mass: the tectonic scale of the shallow upper crustal rock mass is mainly brittle failure, so a linear elastic constitutive model is adopted, and the strength characteristics are described by the Mohr-Coulomb criterion; a viscoelastic model is adopted for the middle and lower crust; fault zones and tectonic alteration zones are set as weak zones or regional zones, and the Young's modulus parameter of the overall medium is multiplied by a weakening coefficient of 1 / 10 to 1 / 3.

6. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: During the evolution of the geostress field in the engineering site, the local geostress data obtained from the hydraulic fracturing test of the borehole in the site is used as a constraint to verify the accuracy of the simulation results of the three-dimensional geological numerical model within the point range. The calculation is repeated multiple times until the error between the geostress results obtained by the numerical simulation of the three-dimensional geological numerical model and the measured data is within the set allowable range. The distribution characteristics of the initial geostress field are restored, so as to carry out the long-term geostress extrapolation simulation in the future.

7. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: The alteration zone and stress field relationship database includes spatial location, type, number and development characteristics of alteration zone rocks at the corresponding location, and geostress parameters at the corresponding location.

8. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: The alteration zone extension distribution prediction model is based on an artificial neural network model, which includes an input layer, a hidden layer, an output layer, and a Softmax activation function. The number of neurons in the input layer is equal to the feature dimension of the input data. The hidden layer includes two fully connected layers and a ReLU activation function. The number of neurons in the output layer is equal to the total number of rock categories in the alteration zone. The Softmax activation function converts the output categories into a probability distribution, i.e., the probability of rock of the k-th alteration zone category.

9. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: The alteration zone extension distribution prediction model is trained using alteration zone data confirmed in the early stage of the project, verified using alteration zone data confirmed in the middle stage of the project construction, and tested using alteration zone data confirmed after excavation in the later stage of the project construction.

10. The method for predicting the extension distribution of alteration zones based on geostress inversion and machine learning according to claim 1, characterized in that: After the alteration degree prediction of each discrete point under the spatial coordinates of the engineering site is completed, the prediction results of all discrete points are imported into a geographic information system or 3D modeling software. Kriging interpolation is used to generate a continuous distribution map, and visualization processing is performed to generate an alteration zone extension distribution prediction map, including a plan view, a cross view and a 3D model, indicating the spatial distribution of rock types in different alteration zones. Finally, the alteration zone extension distribution prediction map and the corresponding confidence distribution map are output.

Images