An open-pit mine landslide susceptibility evaluation method considering mining influence

Abstract

The application discloses a kind of open-pit mine landslide proneness evaluation method considering mining influence, belong to open-pit mine slope disaster prevention and control field.This method extracts surface subsidence and dynamic mining factors such as horizontal displacement towards the direction of slope foot platform by numerical simulation, and static factors such as topography, geomorphology, geological conditions, hydrological factors and surface coverage are jointly modeled;On this basis, a machine learning model is constructed for prediction, and finally the open-pit mine landslide proneness zoning result is formed, and the SHAP method is used to realize the main control factor identification and machine interpretation.The application realizes the accurate identification of landslide high-risk area and the clear analysis of disaster-causing mechanism, provides direct, reliable and scientific support for the fine management and control of mine slope risk, the optimization of monitoring network layout and safety production decision-making.

Description

Technical Field

[0001] This invention relates to the field of open-pit mine slope disaster prevention and control technology, specifically to a method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining activities. Background Technology

[0002] Open-pit mines are a crucial form of mineral resource development, and the long-term stability of their steep slopes directly impacts mine safety, equipment operation, and personnel safety. As mineral resources are developed at greater depths, the combined open-pit and underground mining model is becoming increasingly common in iron, coal, and non-ferrous metal mines. While this method can improve resource recovery rates, the goaf created by underground mining disrupts the original stress balance, inducing subsidence, horizontal displacement, fissure propagation, and localized slippage in the overlying strata. This subsidence is then transmitted to the open-pit slopes, placing them in a state of long-term disturbance.

[0003] Existing landslide susceptibility assessment methods typically rely on static factors such as elevation, slope, aspect, lithology, faults, drainage systems, and roads. While these factors can characterize the basic environmental conditions for landslide occurrence, they struggle to directly quantify the dynamic disturbances caused by underground mining. They often rely on substitute indicators such as distance from the mining area and mining intensity for indirect characterization, failing to accurately reflect the impact of stress redistribution and displacement field evolution on slope instability. Therefore, in scenarios involving combined open-pit and underground mining, existing methods tend to underestimate the impact of mining, leading to discrepancies between susceptibility zoning and actual high-risk areas.

[0004] On the other hand, while machine learning models have high accuracy in landslide spatial prediction, most models suffer from a black box problem, making it difficult to explain "what factors, in what direction, and within what range" drive the risk increase. The lack of interpretability in prediction results hinders the translation of model conclusions into engineering decisions such as monitoring deployment, mining scheme optimization, and remediation design.

[0005] Therefore, it is necessary to propose a landslide susceptibility assessment method for open-pit mines that can simultaneously achieve quantitative expression of mining impacts, robust model prediction, and interpretable identification of main control factors, in order to meet the needs of refined risk management in open-pit and underground mining. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention provides a method for evaluating the susceptibility of landslides in open-pit mines, which balances accurate quantification of mining effects, robustness of prediction models, and interpretability of decision-making processes. This method extracts dynamic mining factors such as surface subsidence (SV) and horizontal displacement (HD-SFP) towards the toe platform through numerical simulation, and jointly models these factors with static factors such as topography, geology, hydrology, and engineering disturbances. Based on this, a machine learning model is constructed for prediction, ultimately generating landslide susceptibility zoning results for open-pit mines. The SHAP method is then used to identify the main controlling factors and explain the mechanisms. This invention achieves accurate identification of high-risk landslide areas and clear analysis of disaster-causing mechanisms, providing direct and reliable scientific support for refined slope risk management, optimized monitoring network layout, and safe production decision-making in mines.

[0007] The technical solution of this invention is as follows:

[0008] A method for assessing the susceptibility of landslides in open-pit mines, taking into account the impact of mining activities, includes the following steps:

[0009] Step S1: Construct a comprehensive evaluation factor system that includes static evaluation factors and dynamic evaluation factors;

[0010] Step S2: Based on the comprehensive evaluation factor system, extract the static evaluation factor data of each mining area and perform standardized preprocessing to obtain a static factor raster database;

[0011] Step S3: Based on the comprehensive evaluation factor system, extract the dynamic evaluation factors of the research mining area and preprocess them to obtain a dynamic factor raster database;

[0012] Step S4: Construct a comprehensive evaluation factor library based on the static factor raster database and the dynamic factor raster database;

[0013] Step S5: Insert several random points into the research mining area, construct a comprehensive evaluation factor attribute table composed of raster data from all random points, and filter the evaluation factors in the comprehensive evaluation factor attribute table to obtain an optimized factor set;

[0014] Step S6: Collect evaluation factor data corresponding to the evaluation factor types in the optimization factor set for historical landslide bodies and stable areas in the research mining area, and construct a sample dataset;

[0015] Step S7: Construct a machine learning model and train the model using the sample dataset; extract the evaluation factor observations corresponding to the evaluation factor types in the optimized factor set from the comprehensive evaluation factor library constructed in Step S4, and input them into the trained machine learning model to obtain the landslide susceptibility prediction results.

[0016] Step S8: Based on the SHAP value, determine the contribution value of each evaluation factor to the landslide susceptibility prediction result output by the machine learning model in step S7, and identify the main controlling factors affecting landslide susceptibility.

[0017] Furthermore, in step S1, the static evaluation factors include topographic and geomorphological evaluation factors, geological condition evaluation factors, hydrological factor evaluation factors, and land cover evaluation factors; the dynamic evaluation factors include mining disturbance evaluation factors.

[0018] The topographic and geomorphological evaluation factors include elevation (DEM), slope (SA), aspect (AS), plane curvature (PC), and profile curvature (PrC).

[0019] The geological condition evaluation factors include lithology (LI) and distance from fault (DtF).

[0020] The hydrological factor evaluation factors include distance from the river (DtR) and hydrogeological unit (HM).

[0021] The evaluation factors for land cover include distance from the road (DtRd).

[0022] The evaluation factors for mining disturbance include surface subsidence (SV) and slope aspect displacement toward the toe platform (HD-SFP).

[0023] Furthermore, in step S2, the construction process of the static factor raster database includes:

[0024] S2.1: The research mining area is rasterized, and the research area is divided into a series of grid units of equal size and regular shape, i.e., raster units, and these raster units are used as the basic units for evaluation.

[0025] S2.2: Extract topographic and geomorphological evaluation factor data such as elevation, slope, aspect, plane curvature, and profile curvature of each grid point from the digital elevation model data obtained by comprehensively scanning the surface of the mining area.

[0026] S2.3: Based on the geological survey report and engineering exploration data of the mining area, extract evaluation factor data of geological conditions such as lithology, fault distribution and distance from fault for each grid point;

[0027] S2.4: Based on hydrogeological maps and river distribution data, extract hydrogeological unit data and hydrological factor data such as distance from river for each grid point;

[0028] S2.5: Based on road distribution and surface engineering disturbance vector data, extract surface cover evaluation factor data such as the distance of each grid point from the road;

[0029] S2.6: Import the static evaluation factor data of each grid point into the geographic information processing platform ArcGIS Pro. For all grid points, static evaluation factor data from different sources and in different formats are processed sequentially using spatial coordinate system I, study area clipping, vector to raster conversion, and resampling. This process unifies all preprocessed static evaluation factor data into raster data with the same spatial range, the same raster boundary, and consistent spatial resolution, forming a static factor raster database. Each raster data includes all static evaluation factors at the corresponding grid point.

[0030] Furthermore, in step S3, the construction process of the dynamic factor raster database includes:

[0031] S3.1: Using the Rhino modeling platform, a three-dimensional refined engineering geological model of the research mining area is established by combining exploration profile data, borehole data and surface point cloud data. This model comprehensively reflects the spatial distribution relationship of the ore body and the main lithological interfaces. The Griddle plugin is used to divide the three-dimensional refined engineering geological model into a grid model that meets the requirements of numerical simulation.

[0032] S3.2: Assign the Mohr-Coulomb constitutive relation to the obtained mesh model to form a mining numerical simulation model; and determine the mechanical parameters of the ore body and surrounding rock based on the results of indoor mechanical tests and strength criteria inversion; for the rock mass in the mining disturbance zone, reduce its mechanical parameters by 60% to 70% to characterize the rock mass weakening effect caused by underground excavation unloading.

[0033] The boundary conditions of the mining numerical simulation model are set as bottom fixed and surrounding normal constraints. The ore body is excavated in stages and stress release is simulated according to the actual underground mining sequence, so as to obtain the deformation response of the slope rock mass under the action of underground mining.

[0034] S3.3: After the numerical simulation of mining is completed, the displacement field data of each grid point on the surface of the open-pit mine slope divided in S2 is extracted. The vertical displacement is used to characterize the surface subsidence, and the horizontal displacement is used to characterize the slope displacement towards the slope toe platform.

[0035] S3.4: After importing the above displacement field data into the ArcGIS Pro platform, firstly, the displacement point layer is clipped according to the spatial coordinate system I and the study area, removing outliers, duplicate points, and invalid points outside the study area; then, using SV and HD-SFP as interpolation fields, the inverse distance weighting method is used to spatially interpolate the displacement field data, generating a continuously distributed surface settlement interpolation surface and a slope displacement interpolation surface facing the slope toe platform; through the clipping of the spatial coordinate system I and the study area, dynamic evaluation factor raster data with the same spatial range, the same raster boundary, and consistent spatial resolution are generated, forming a dynamic factor raster database.

[0036] Furthermore, in step S4, the construction process of the comprehensive evaluation factor library is as follows:

[0037] The static factor raster database obtained in step S2 and the dynamic factor raster database obtained in step S3 are uniformly imported into the same evaluation database. Null value checks, outlier handling, and standardized coding are performed on each data. Continuous factors are normalized, and categorical factors are coded to obtain a comprehensive evaluation factor library containing all evaluation factor raster data. Among them, the comprehensive evaluation factors are divided into continuous evaluation factors and categorical evaluation factors according to data attributes. The continuous evaluation factors take continuous numerical values, including elevation (DEM), slope (SA), plane curvature (PC), profile curvature (PrC), distance from fault (DtF), distance from river (DtR), distance from road (DtRd), surface subsidence (SV), and horizontal displacement towards the slope toe platform (HD-SFP). The categorical evaluation factors take discrete categories, including slope aspect (AS), lithology (LI), and hydrogeological unit (HM).

[0038] Furthermore, in step S5, the set of optimization factors is obtained based on the following method:

[0039] S5.1: Using the random point insertion function, a certain number of random points are inserted into the research mining area; the raster data corresponding to the location of each random point is assigned to the corresponding random point to obtain the random point attribute table; all random point attribute tables are combined to form the comprehensive evaluation factor attribute table;

[0040] S5.2: Conduct correlation analysis on each continuous evaluation factor in the comprehensive evaluation factor attribute table, using the Pearson correlation coefficient as the characterization. The calculation formula is as follows:

[0041]

[0042] in, Represents two continuous evaluation factors and The Pearson correlation coefficient between them ranges from [-1, 1]; , They represent the first Continuous evaluation factors in a random point attribute table and Specific observations; , These represent all continuous evaluation factors in the comprehensive evaluation factor attribute table. The average and The average value; To study the total number of random points within the mining area;

[0043] When | |≥0.7, then and For highly correlated continuous evaluation factors, for highly correlated factor pairs, retain the one with clearer business significance or higher correlation with landslide, and delete the other;

[0044] S5.3: Based on step S5.2, calculate the variance inflation factor of each evaluation factor in the comprehensive evaluation factor attribute table. and tolerance The formula is as follows:

[0045]

[0046] in, For the first The multiple determination coefficient of each evaluation factor;

[0047] Eliminate or The evaluation factors, and the evaluation factor data of all random points that are finally retained in the comprehensive evaluation factor attribute table, constitute the optimization factor set.

[0048] Furthermore, in step S6, the process of constructing the sample dataset is as follows:

[0049] S6.1: Based on the optimization factor set, collect the evaluation factor data corresponding to the evaluation factor types in the landslide area of ​​the historical landslide boundary or landslide hazard point in the research mining area, and form a landslide sample as a positive sample. Define the random points in the random points inserted in S5 that are located within the landslide area as landslide points, and define the category label as 1. Collect the evaluation factor data corresponding to the evaluation factor types in the optimization factor set in the relatively stable area far away from known landslide bodies, obvious deformation anomaly areas, and mining high impact anomaly areas, and form a non-landslide sample as a negative sample. Define the random points in the random points inserted in S5 that are located within the relatively stable area as non-landslide points, and define the category label as 0. All positive and negative samples constitute the sample dataset.

[0050] S6.2: Divide the sample dataset into a training set and a test set according to a preset ratio. The training set is used for subsequent learning and construction of machine learning model parameters, and the test set is used for independent verification of machine learning models and evaluation of prediction accuracy. To avoid class imbalance caused by sample division, stratified random sampling is performed in the positive and negative samples in the same proportion to make the ratio of positive and negative samples in the training set and the test set basically consistent, thereby ensuring the reliability and stability of model training and testing results.

[0051] Furthermore, the specific process of step S7 includes:

[0052] S7.1: Construct a machine learning model and train the machine learning model using the training set in step S6; wherein, the evaluation factor data corresponding to each training sample in the training set are arranged into a multi-dimensional optimized factor feature vector and used as input features, and the category label is the output category (1 for landslide and 0 for non-landslide).

[0053] S7.2: Each grid point in the research area is considered as an evaluation unit. Referring to the evaluation factor types in the optimization factor set, the evaluation factor observation values ​​corresponding to the evaluation factor types in the optimization factor set for each evaluation unit are extracted from the comprehensive evaluation factor library constructed in step S4. An optimization factor feature vector with the same arrangement order as in the training stage of step S7.1 is constructed and input into the trained machine learning model to obtain the probability value of each evaluation unit belonging to the landslide category.

[0054] S7.3: Write back the landslide probability values ​​of each evaluation unit to the corresponding evaluation unit according to their spatial location, generate a landslide susceptibility probability raster layer, and export the susceptibility evaluation result tif file;

[0055] S7.4: Import the tif file into ArcGIS Pro, and use the natural breakpoint method (0~1) to divide the research mining area into five levels of landslide susceptibility zones: extremely low, low, medium, high and extremely high, to form a susceptibility zone map. At the same time, the area of ​​the extremely high and high risk zones is calculated and a column chart is generated.

[0056] Furthermore, step S8 is detailed as follows:

[0057] S8.1: For any evaluation unit Suppose it contains If there are several evaluation factors, then the machine learning model will evaluate the units based on the SHAP value. The predicted output is represented as:

[0058]

[0059] in, For machine learning models to evaluate units The predicted output; The baseline value represents the average output of the model on the background sample set; For the first The SHAP value of each evaluation factor represents its contribution to the prediction result of the current evaluation unit. The calculation formula is as follows:

[0060]

[0061] in, A set of indices representing all evaluation factors; The total number of evaluation factors; Not including the first Any subset of evaluation factors;

[0062] For subset The number of evaluation factors; These are the Shapley weighting coefficients, used to ensure a fair distribution of contributions from subsets of different sizes; To use only subsets When considering the evaluation factors, the model obtains the predicted values; To join the Model output after evaluating factors; For only subsets The input sample consists of the actual values ​​of those evaluation factors;

[0063] The direction and degree of influence of each evaluation factor on the landslide susceptibility prediction results are determined based on the SHAP value of each evaluation factor. When, it indicates the first Several evaluation factors have a promoting effect on landslide occurrence; when When, it indicates the first Each evaluation factor has an inhibitory effect on landslide occurrence; The larger the value, the higher the value. The more significant the impact of each evaluation factor on the prediction results;

[0064] S8.2: For any evaluation unit obtained from the calculation All evaluation factors Sort the results by the smallest to largest to obtain the contribution results of the evaluation factors for each evaluation unit;

[0065] S8.3: Take the absolute value of the SHAP value of each evaluation factor in all evaluation units and calculate the average to obtain the overall contribution of each evaluation factor; specifically, suppose the study area has a total of The evaluation unit, the first The evaluation factor is at the _ The SHAP value on each evaluation unit is [value]. Then the first Overall contribution of each evaluation factor Defined as its average absolute SHAP value across all evaluation units:

[0066]

[0067] For all evaluation factors The overall contribution ranking of each evaluation factor is obtained by sorting them from largest to smallest. SHAP bee colony diagrams and feature dependency scatter plots are then drawn to interpret the prediction results of the machine learning model and identify the main controlling factors affecting landslide susceptibility. The larger the value, the more significant the overall impact of the evaluation factor on the landslide susceptibility prediction results in the study area, and the main controlling factor can be identified accordingly.

[0068] Furthermore, the machine learning model includes a random forest model, an extreme gradient boosting model, a support vector machine model, or a logistic regression model.

[0069] The beneficial effects of this invention are:

[0070] 1) This invention improves the accuracy of the characterization of mining impact by extracting surface subsidence and horizontal displacement toward the slope toe platform through numerical simulation.

[0071] 2) Based on static factors such as topography, geology, and hydrology, this invention introduces dynamic factors related to mining, making the evaluation system more applicable to open-pit and underground mining conditions.

[0072] 3) This invention combines three-dimensional geological modeling, numerical simulation and GIS processing to realize the extraction, rasterization and modeling application of dynamic factors.

[0073] 4) This invention outputs landslide susceptibility probability maps, zoning maps, and statistical results of high-risk areas through machine learning models, which can directly serve mine safety management. Attached Figure Description

[0074] Figure 1 The geographical location of the study area in this embodiment of the invention is shown in (a) topographic map of the study area, (b) remote sensing image of the subsidence pit and open-pit mine, and (c) enlarged view of the local structure of the mine.

[0075] Figure 2 The following is a raster of static evaluation factors for landslide hazards in the study area in this embodiment of the invention; wherein (a) is a raster of elevation factors, (b) is a raster of slope factors, (c) is a raster of aspect factors, (d) is a raster of planar curvature factors, (e) is a raster of profile curvature factors, (f) is a raster of lithology factors, (g) is a raster of distance from fault factors, (h) is a raster of distance from river factors, (i) is a raster of hydrogeological unit factors, and (j) is a raster of distance from road factors; in (f): 1 represents garnet-ambium-chlorite schist, 2 represents metamorphic gabbro-diabase, 3 represents metamorphic sodic pyroxene, 4 represents amphibole-bearing metamorphic sodic gabbro, 5 represents biotite-columnar tremolite dolomitic marble, 6 represents amphibole-biotite dolomitic marble, and 7 represents mudstone and carbonaceous (rock / layer).

[0076] Figure 3The following is a spatial distribution map of the Dahongshan Iron Mine in an embodiment of the present invention; wherein (a) is a planar distribution map and (b) is a spatial distribution map.

[0077] Figure 4 This is a 3D modeling and mesh drawing diagram of the Dahongshan mining area in an embodiment of the present invention; wherein (a) the surface of the study area, (b) Boolean operation cutting of the solid model based on the surface, (c) schematic diagram of the 3D solid model, (d) schematic diagram of the lithological cutting line, (e) schematic diagram of generating the cutting surface based on the cutting line, (f) lithological division process based on the cutting surface, (g) final 3D geological model, and (h) numerical calculation mesh model.

[0078] Figure 5 This is a cloud map showing the overall rock movement in the Dahongshan mining area in an embodiment of the present invention.

[0079] Figure 6 The following are the settlement and aspect displacement grids in the embodiments of the present invention; wherein (a) is a surface settlement factor grid and (b) is an aspect displacement factor grid.

[0080] Figure 7 This is the Pearson coefficient matrix in an embodiment of the present invention.

[0081] Figure 8 The following is a landslide susceptibility zoning map in an embodiment of the present invention, wherein (a) is the RF model, (b) is the XGBoost model, (c) is the SVM model, and (d) is the LR model.

[0082] Figure 9 The diagram shows the area percentage of the risk area in this embodiment of the invention, where (a) is an overall schematic diagram and (b) is a magnified view of a portion of (a).

[0083] Figure 10 This is a ROC curve diagram from an embodiment of the present invention.

[0084] Figure 11 This is a bar chart showing the importance of SHAP in the RF model in this embodiment of the invention.

[0085] Figure 12 The above are the SHAP beehive graph and feature dependency scatter plot of the RF model in this embodiment of the invention; wherein (a) is the SHAP feature importance ranking graph, (b) is the HD-SFP feature dependency graph, and (c) is the AS feature dependency graph. Detailed Implementation

[0086] The specific embodiments of the present invention are described in detail below with reference to the technical solutions and accompanying drawings.

[0087] This application takes the Dahongshan Iron Mine open-pit mine as an example. The open-pit and underground mining areas overlap significantly on the horizontal projection plane, with an overlap ratio as high as 42%. Furthermore, the minimum vertical distance between the underground mine roof and the toe of the open-pit slope is only 85 meters. Under these complex engineering conditions of close collaboration between open-pit and underground mining, the rock strata movement and stress redistribution induced by underground mining activities have a significant disturbance effect on the stability of the open-pit slope, and its potential impact cannot be ignored. This application details a method for evaluating the susceptibility of open-pit mine landslides, taking into account the impact of mining activities, specifically for this working condition.

[0088] S1: Construct a comprehensive evaluation factor system that includes static evaluation factors and dynamic evaluation factors; the static evaluation factors include topographic and geomorphological evaluation factors, geological condition evaluation factors, hydrological factor evaluation factors, and land cover evaluation factors; the dynamic evaluation factors include mining disturbance evaluation factors.

[0089] The topographic and geomorphological evaluation factors include elevation (DEM), slope (SA), aspect (AS), plane curvature (PC), and profile curvature (PrC).

[0090] The geological condition evaluation factors include lithology (LI) and distance from fault (DtF).

[0091] The hydrological factor evaluation factors include distance from the river (DtR) and hydrogeological unit (HM).

[0092] The evaluation factors for land cover include distance from the road (DtRd).

[0093] The evaluation factors for mining disturbance include surface subsidence (SV) and slope aspect displacement toward the toe platform (HD-SFP).

[0094] S2: Based on the comprehensive evaluation factor system, extract the static evaluation factor data of the research mining area and perform standardized preprocessing to obtain a static factor raster database; the specific process includes:

[0095] S2.1: The Dahongshan open-pit mine is located in a mountainous area in southwest China, situated on the eastern side of a mountain range and the eastern bank of a major regional river. There are no large-scale man-made structures such as water conservancy, communication, or public facilities in the surrounding area. The geographical location of the study area is as follows: Figure 1As shown. The surface elevation of the mining area ranges from 600m to 1850m, with the maximum mining height in the open-pit mine reaching 435m. The surface elevation ranges from 730m to 1165m, characterized by eroded and denuded mountainous terrain, deep incision, large undulations, and well-developed network of gullies. River flow is controlled by rainfall and varies significantly. The vegetation in the mining area is mainly shrubs and weeds, with no large forests. The climate is hot and rainy in summer and autumn, and mild and dry in winter and spring. Landslides are the main geological hazard in the study area, with three potential landslide areas identified. For ease of statistical analysis, the study area was divided into 1m × 1m grid units, totaling 133,823 units.

[0096] S2.2: A 1m resolution digital elevation model (DEM) obtained by UAV oblique photography is used to extract topographic factors such as elevation (DEM), slope (SA), aspect (AS), plane curvature (PC), and profile curvature (PrC).

[0097] S2.3: Obtain information such as stratigraphic lithology (LI) and distance from fault (DtF) through a 1:2000 geological vector map to clarify the geological background of the landslide.

[0098] S2.4: Extract the distance data (DtRd) between the study area and the road using OpenStreetMap to quantify the disturbance of road excavation on the slope stress state.

[0099] S2.5: Different evaluation factors have different pixel sizes when generating raster data. To correctly assign the value of each evaluation factor, the resampling tool in ArcGIS Pro was used to adjust it to a 1m resolution raster to unify the number of raster cells and ensure the accuracy of raster value assignment. A total of 133,823 valid evaluation units were generated, which then formed a static factor raster database, providing a standardized data foundation for subsequent model training. Figure 2 This is a grid for static evaluation factors of landslide hazards in the study area.

[0100] Step S3: Based on the comprehensive evaluation factor system, extract the dynamic evaluation factors of the research mining area and preprocess them to obtain a dynamic factor raster database; the specific process includes:

[0101] S3.1: The open-pit mine is located between exploration lines A28 and A39, with an extraction elevation of 680–1012 m. The underlying main underground ore body is the II1 ore group, distributed along lines A25–A43′, with an elevation range of 25.72–945.00 m. Ore bodies III and IV are located along lines A29–A36, with an elevation range of 320–520 m, distributed to the west and north of the main II1 ore body. The spatial distribution of the Dahongshan iron ore mine is as follows: Figure 3Overall, the open-pit mine and the underground II1 ore group have a high degree of overlap in planar area, and its northwest side is also affected by the disturbance caused by the mining of the III-IV ore body, forming a spatial pattern of open-pit and underground mining combined with mining: "upper open-pit – underlying II1 – northwest III-IV". In particular, the minimum vertical distance between the bottom of the open-pit slope and the underground mining area is only 85 m, indicating that the overlying rock movement, stress redistribution and mining deformation caused by underground mining can be easily transmitted to the overlying slope, and have a direct controlling effect on the deformation response of the slope bottom and the rock mass above it, thus making the stability of the open-pit slope show obvious mining sensitivity.

[0102] S3.2: To accurately simulate rock strata movement caused by mining, this method, based on the Rhino modeling platform, combined with 18 exploration profiles (A28~A45) spaced 50 meters apart and surface point cloud data, established a refined three-dimensional engineering geological model with dimensions of 500m × 350m × 480m (length × width × height). The three-dimensional modeling and meshing of the Dahongshan mining area are shown below. Figure 4 As shown, the model fully considers the spatial distribution of different lithologies such as dolomite marble, gabbro-diabase, and iron ore bodies, and uses the Griddle plugin for mesh generation with a minimum side length of 10m to ensure calculation accuracy.

[0103] S3.3: The Mohr-Coulomb criterion was used as the constitutive model for the rock mass. The mechanical parameters were determined through field sampling tests and the Hoek-Brown strength criterion, as detailed in Table 1. The rock mass parameters in the mining disturbance area took into account the excavation unloading effect. The cohesion and internal friction angle were reduced to 65% of their natural values. The simulation process followed the principle of step-by-step excavation and stress release. The boundary conditions were set as fixed at the bottom and normal constraints on all four sides.

[0104] Table 1 Rock mass parameters

[0105]

[0106] S3.4: The results of the simulated overall rock movement cloud map of the Dahongshan mining area are shown below. Figure 5 Underground mining causes significant displacement of the open-pit mine slopes. By cutting the overall ore body model and refining the open-pit mine slope displacement cloud map, the patterns of settlement and slope aspect displacement can be observed. Figure 5 The cloud map shown indicates that the maximum settlement value reaches 0.41m, and the maximum slope displacement reaches 0.14m. The closer to the area affected by the open-pit mine, the larger the values. By extracting displacement data from the open-pit mine surface and using the inverse distance weighting method for spatial interpolation, the following data was finally generated. Figure 6 The surface subsidence (SV) raster map and the slope aspect displacement (HD-SFP) raster map shown are used for susceptibility assessment, which together form a dynamic factor raster database.

[0107] Step S4: Import the static factor raster database obtained in Step S2 and the dynamic factor raster database obtained in Step S3 into the same evaluation database. Perform null value checks, outlier handling, and standardized coding on each data. Normalize continuous factors and encode categorical factors to obtain a comprehensive evaluation factor library containing all evaluation factor raster data.

[0108] Step S5: Insert several random points into the research mining area, construct a comprehensive evaluation factor attribute table composed of raster data from all random points, and filter the evaluation factors in the comprehensive evaluation factor attribute table to obtain an optimized factor set; specifically including:

[0109] S5.1: Using the random point insertion function, a certain number of random points are inserted into the research mining area; the raster data corresponding to the location of each random point is assigned to the corresponding random point to obtain the random point attribute table; all random point attribute tables are combined to form the comprehensive evaluation factor attribute table.

[0110] S5.2: First, perform Pearson correlation coefficient analysis on each continuous evaluation factor in the comprehensive evaluation factor attribute table. The Pearson coefficient matrix calculated using the evaluation factor data is shown below. Figure 7 The matrix shows that the absolute value of the correlation coefficient between most evaluation factors is less than 0.3, only a very few evaluation factors show moderate correlation, and no extremely strong correlation is found.

[0111] S5.3: To further quantify multicollinearity among evaluation factors, this embodiment calculates the variance inflation factor (VIF) and tolerance (TOL) for each evaluation factor. It is generally considered that... or Severe multicollinearity exists. As shown in Table 2 of the multicollinearity analysis, the VIF of all evaluation factors is less than 4, with the highest HD-SFP at 3.73 and the lowest TOL at 0.2679, both far exceeding the critical criterion. This fully demonstrates that there is no significant multicollinearity problem among the 12 evaluation factors selected in this embodiment, meeting the basic requirements of machine learning models for input data and ensuring the stability and reliability of subsequent model training results. Finally, the 12 evaluation factor data from all random points retained in the evaluation factor attribute table constitute the optimization factor set.

[0112] Table 2 Multicollinearity Analysis Table

[0113]

[0114] Step S6: Collect evaluation factor data corresponding to the evaluation factor types within the optimization factor set for historical landslides and stable areas in the research mining area, and construct a sample dataset; the specific process is as follows:

[0115] S6.1: Based on the optimization factor set, collect the evaluation factor data corresponding to the evaluation factor types in the landslide area of ​​the historical landslide boundary or landslide hazard point in the research mining area, and form a landslide sample as a positive sample. Define the random points in the random points inserted in S5 that are located in the landslide area as landslide points, and define the category label as 1. Collect the evaluation factor data corresponding to the evaluation factor types in the optimization factor set in the relatively stable area far away from known landslide bodies, obvious deformation anomaly areas and mining high impact anomaly areas, and form a non-landslide sample as a negative sample. Define the random points in the random points inserted in S5 that are located in the relatively stable area as non-landslide points, and define the category label as 0. All positive and negative samples form a sample dataset.

[0116] S6.2: Divide the sample dataset into a training set and a test set according to a preset ratio. The training set is used for subsequent learning and construction of machine learning model parameters, and the test set is used for independent verification of machine learning models and evaluation of prediction accuracy. To avoid class imbalance caused by sample division, stratified random sampling is performed in the positive and negative samples in the same proportion to make the ratio of positive and negative samples in the training set and the test set basically consistent, thereby ensuring the reliability and stability of model training and testing results.

[0117] Step S7: Construct a machine learning model and train it using the sample dataset; extract the evaluation factor observations corresponding to the evaluation factor types within the optimized factor set from the comprehensive evaluation factor library constructed in Step S4, and input them into the trained machine learning model to obtain the landslide susceptibility prediction results; the specific process includes:

[0118] S7.1: This embodiment constructs four machine learning models: RF model, XGBoost model, SVM model, and LR model, and trains each model using the training set obtained in step S6. Specifically, the TreeBagger function is used to construct the RF model, and out-of-bag (OOB) error is used to dynamically evaluate model performance to determine the optimal parameter combination. Through training subsets, when the number of decision trees (Ntree) reaches 150, the OOB error tends to stabilize, indicating that further increasing the number of trees has limited effect on improving model accuracy. Considering both model stability and computational efficiency, the optimal parameters are finally determined to be Ntree = 400 and mtry = 2, at which point the optimal OOB accuracy reaches 0.9064. This setting helps to improve the model's ability to identify complex factor relationships while controlling variance.

[0119] The XGBoost model described above uses grid search combined with 5-fold cross-validation to optimize its hyperparameters. The optimal parameters are as follows: learning rate (eta) of 0.01, maximum tree depth of 3, regularization term (lambda) of 5, and number of estimators (n estimators) of 300. These parameter settings indicate that the XGBoost model in this embodiment tends to adopt a training strategy of "small step size, shallow trees, and strong regularization" to improve the model's ability to identify spatial differences in landslide susceptibility while ensuring stability.

[0120] The SVM model used was a radial basis function (RBF). The optimal parameters were determined through grid search and 5-fold cross-validation: penalty coefficient C = 100 and kernel parameter gamma = 0.01. At this setting, the F1 score on the validation set reached 0.9057.

[0121] To mitigate the impact of parameter fluctuations and redundant information under high-dimensional input conditions, this embodiment of the invention introduces L2 regularization into the LR model and employs regularized path analysis combined with 5-fold cross-validation to determine the optimal parameters. The optimal regularization strength is ultimately found to be C=0.5, with an average accuracy of 0.8989 on the model validation set. Although the LR model struggles to fully characterize complex nonlinear relationships, it serves as a control model, providing a clear reference for subsequent discussions of the results.

[0122] S7.2: Each grid point in the research area is considered as an evaluation unit. Referring to the 12 evaluation factors in the optimization factor set, the observation values ​​of the 12 evaluation factors for each evaluation unit are extracted from the comprehensive evaluation factor library constructed in step S4, and input into the trained machine learning model to obtain the probability value of each evaluation unit belonging to the landslide category.

[0123] S7.3: Write back the landslide probability values ​​of each evaluation unit to the corresponding evaluation unit according to their spatial location, generate a landslide susceptibility probability raster layer, and export the susceptibility evaluation results tif file.

[0124] S7.4: Import the aforementioned TIFF file into ArcGIS Pro. Using the natural breakpoint method (0~1), divide the study mining area into five levels of landslide susceptibility zones: extremely low, low, medium, high, and extremely high, forming a susceptibility zoning map. Simultaneously, calculate the area of ​​the extremely high and high-risk zones and generate a bar chart. The landslide susceptibility zoning obtained through four machine learning models is as follows: Figure 8 As shown in the figure. A comparison reveals that both high-risk and extremely high-risk landslide areas are concentrated closer to the mining area's impact zone, specifically above the 1000m bench in the southeast of the mining site. The areas of medium, extremely high, and high-risk zones were also statistically analyzed and a bar chart was generated, as shown below. Figure 9 .

[0125] S7.5: Based on the test set obtained in step S6, the correspondence between the predicted categories and the true categories of each machine learning model is statistically analyzed, a confusion matrix is ​​constructed, and the accuracy, precision, recall and F1 score are calculated. The results are shown in Table 3.

[0126] Table 3 Performance Evaluation of Machine Learning Models

[0127]

[0128] Overall, all four models demonstrated good predictive ability for landslide susceptibility. In terms of sensitivity (Se), the RF, SVM, and LR models all had a Se value of 1.0000, indicating that all landslide samples in the test set were correctly identified. The XGBoost model had a Se value of 0.9932, with only one landslide sample being misclassified, also demonstrating strong landslide identification ability. In terms of specificity (Sp), all four models had Sp values ​​above 0.79, with the XGBoost model having the highest at 0.8503, followed by the RF model at 0.8435, indicating that the XGBoost and RF models had relatively better ability to distinguish non-landslide samples. In terms of overall classification accuracy, the RF and XGBoost models had the highest Ac values ​​of 0.9215, while the LR model had an Ac value of 0.9147, slightly lower than the other two. The SVM model had an Ac value of 0.8976, which was relatively low. These results indicate that all four models can distinguish landslides effectively.

[0129] S7.6: Based on the landslide probability values ​​output by each machine learning model, calculate the true positive rate (TPR) and false positive rate (FPR) at different discrimination thresholds, and plot the receiver operating characteristic (ROC) curve accordingly. The calculation formula is as follows:

[0130] TPR = TP / (TP + FN)

[0131] FPR = FP / (FP + TN)

[0132] like Figure 10 The ROC curves shown indicate that the ROC curves of all models are significantly higher than the random classification baseline (AUC = 0.5), demonstrating their strong binary classification ability. Among them, the RF model has the highest AUC value of 0.9357, indicating its best overall discriminative performance. The SVM and LR models have AUC values ​​of 0.9311 and 0.9307, respectively, which are close to those of the RF model. The XGBoost model has the lowest AUC value, but it still reaches 0.9148. Overall, the AUC values ​​of all four models are higher than 0.91, indicating that they all have good discriminative ability in landslide susceptibility prediction.

[0133] Step S8: Based on the SHAP value, determine the contribution value of each evaluation factor to the landslide susceptibility prediction result output by the RF model in Step S7 and identify the main controlling factors affecting landslide susceptibility; details are as follows:

[0134] S8.1: Suppose the random forest model includes If there are 10 trees, then the overall output can be represented as:

[0135]

[0136] in, Indicates the first Decision trees for samples The output value. Since SHAP is additivity, the SHAP value of the overall random forest model can be decomposed into the sum of the contributions of each decision tree, i.e.:

[0137]

[0138] in, Indicates the first The first decision tree The contribution value of each evaluation factor. The baseline value also satisfies:

[0139]

[0140] S8.2: For any evaluation unit obtained from the calculation, all evaluation factors... The evaluation factor contribution results for each individual evaluation unit are obtained by sorting them from smallest to largest.

[0141] S8.3: Take the absolute value of the SHAP value of each evaluation factor in all evaluation units and calculate the average to obtain the overall contribution of each evaluation factor; calculate the overall contribution of all evaluation factors. Sort the evaluation factors from largest to smallest to obtain the overall contribution ranking, such as... Figure 11 As shown; and plot the SHAP bee colony diagram and feature dependency scatter plot, as shown. Figure 12 As shown in the figure, the results indicate that both the slope aspect and the direction of the slope towards the toe platform are key predictor variables. The slope aspect towards the toe platform has a stable positive effect; its increase significantly enhances landslide susceptibility, characterizing the enhanced effect of mining disturbance. The direction of the slope has a negative contribution; smaller direction values ​​correspond to higher susceptibility. Elevation, surface subsidence, and distance from the road have limited contributions. Lithological and hydrogeological units exhibit hierarchical clustering characteristics. Landslide susceptibility in the study area is synergistically controlled by topography, geology, and mining deformation, with the slope aspect towards the toe platform being the most stable positive predictor.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for evaluating the susceptibility of landslides in open-pit mines considering the impact of mining activities, characterized in that, include: Step S1: Construct a comprehensive evaluation factor system that includes static evaluation factors and dynamic evaluation factors; The dynamic evaluation factors include mining disturbance evaluation factors, which include surface subsidence and slope displacement toward the toe platform. Step S2: Based on the comprehensive evaluation factor system, extract the static evaluation factor data of each mining area and perform standardized preprocessing to obtain a static factor raster database; Step S3: Based on the comprehensive evaluation factor system, extract the dynamic evaluation factors of the research mining area and preprocess them to obtain a dynamic factor raster database; The specific construction process includes: S3.1: Using the Rhino modeling platform, a three-dimensional refined engineering geological model of the research mining area is established by combining exploration profile data, borehole data and surface point cloud data; the Griddle plugin is used to mesh the three-dimensional refined engineering geological model to form a mesh model that meets the requirements of numerical simulation. S3.2: Assign the Mohr-Coulomb constitutive relation to the obtained mesh model to form a mining numerical simulation model; and determine the mechanical parameters of the ore body and surrounding rock based on the results of indoor mechanical tests and strength criteria inversion; for the rock mass in the mining disturbance zone, reduce its mechanical parameters by 60% to 70% to characterize the rock mass weakening effect caused by underground excavation unloading. The boundary conditions of the mining numerical simulation model are set as bottom fixed and surrounding normal constraints. The ore body is excavated in stages and stress release is simulated according to the actual underground mining sequence, so as to obtain the deformation response of the slope rock mass under the action of underground mining. S3.3: After the numerical simulation of mining is completed, the displacement field data of each grid point on the surface of the open-pit mine slope divided in S2 is extracted. The vertical displacement is used to characterize the surface subsidence, and the horizontal displacement is used to characterize the slope displacement towards the slope toe platform. S3.4: After importing the above displacement field data into the ArcGIS Pro platform, firstly, the displacement point layer is clipped according to the spatial coordinate system and the study area, removing outliers, duplicate points, and invalid points outside the study area; then, using surface settlement and slope displacement towards the toe platform as interpolation fields, the inverse distance weighting method is used to spatially interpolate the displacement field data, generating a continuously distributed surface settlement interpolation surface and a slope displacement interpolation surface towards the toe platform; through clipping according to the spatial coordinate system and the study area, dynamic evaluation factor raster data with the same spatial range, the same raster boundary, and consistent spatial resolution are generated, forming a dynamic factor raster database; Step S4: Construct a comprehensive evaluation factor library based on the static factor raster database and the dynamic factor raster database; Step S5: Insert several random points into the research mining area, construct a comprehensive evaluation factor attribute table composed of raster data from all random points, and filter the evaluation factors in the comprehensive evaluation factor attribute table to obtain an optimized factor set; Step S6: Collect evaluation factor data corresponding to the evaluation factor types in the optimization factor set for historical landslide bodies and stable areas in the research mining area, and construct a sample dataset; Step S7: Construct a machine learning model and train the model using the sample dataset; extract the evaluation factor observations corresponding to the evaluation factor types in the optimized factor set from the comprehensive evaluation factor library constructed in Step S4, and input them into the trained machine learning model to obtain the landslide susceptibility prediction results. Step S8: Based on the SHAP value, determine the contribution value of each evaluation factor to the landslide susceptibility prediction result output by the machine learning model in step S7, and identify the main controlling factors affecting landslide susceptibility.

2. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 1, is characterized in that... In step S1, the static evaluation factors include topographic and geomorphological evaluation factors, geological condition evaluation factors, hydrological factor evaluation factors, and land cover evaluation factors. The topographic and geomorphological evaluation factors include elevation, slope, aspect, plan curvature, and profile curvature. The geological condition evaluation factors include lithology and distance from faults; The hydrological factor evaluation factors include distance from the river and hydrogeological units; The evaluation factors for land cover include distance from roads.

3. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 2, is characterized in that... In step S2, the construction process of the static factor raster database includes: S2.1: The research mining area is rasterized, and the research area is divided into grid units of equal size and regular shape. S2.2: Extract topographic and geomorphological evaluation factor data for each grid point from the digital elevation model data obtained by a comprehensive scan of the surface of the mining area, including elevation, slope, aspect, plane curvature, and profile curvature. S2.3: Based on the geological survey report and engineering exploration data of the mining area, extract the geological condition evaluation factor data for each grid point, including lithology, fault distribution and distance from faults; S2.4: Based on hydrogeological maps and river distribution data, extract hydrological factor data for each grid point, including hydrogeological units and distances from rivers; S2.5: Based on road distribution and surface engineering disturbance vector data, extract surface cover evaluation factor data for each grid point, including distance from the road; S2.6: Import the static evaluation factor data of each grid point into the geographic information processing platform ArcGIS Pro. For all grid points, static evaluation factor data from different sources and in different formats are processed sequentially using spatial coordinate system I, study area clipping, vector to raster conversion, and resampling. This process unifies all preprocessed static evaluation factor data into raster data with the same spatial range, the same raster boundary, and consistent spatial resolution, forming a static factor raster database. Each raster data includes all static evaluation factors at the corresponding grid point.

4. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 3, is characterized in that... In step S4, the comprehensive evaluation factor library is constructed as follows: The static factor raster database and the dynamic factor raster database are uniformly imported into the same evaluation database. Null value checks, outlier handling, and standardized coding are performed on each data. Continuous factors are normalized, and categorical factors are coded to obtain a comprehensive evaluation factor library containing all evaluation factor raster data. The comprehensive evaluation factors are divided into continuous evaluation factors and categorical evaluation factors according to data attributes. Continuous evaluation factors take continuous values, and categorical evaluation factors take discrete categories.

5. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 4, is characterized in that... In step S5, the set of optimization factors is obtained based on the following method: S5.1: Using the random point insertion function, a certain number of random points are inserted into the research mining area; the raster data corresponding to the location of each random point is assigned to the corresponding random point to obtain the random point attribute table; all random point attribute tables are combined to form the comprehensive evaluation factor attribute table; S5.2: Conduct correlation analysis on each continuous evaluation factor in the comprehensive evaluation factor attribute table, using the Pearson correlation coefficient as the characterization. The calculation formula is as follows: in, Represents two continuous evaluation factors and The Pearson correlation coefficient between them ranges from [-1, 1]; , They represent the first Continuous evaluation factors in a random point attribute table and Specific observations; , These represent all continuous evaluation factors in the comprehensive evaluation factor attribute table. The average and The average value; To study the total number of random points within the mining area; When | |≥0.7, then and For highly correlated continuous evaluation factors, retain and Delete the option that has a clearer business significance or a higher correlation with landslides; S5.3: Based on step S5.2, calculate the variance inflation factor of each evaluation factor in the comprehensive evaluation factor attribute table. and tolerance : in, For the first The multiple determination coefficient of each evaluation factor; Eliminate or The evaluation factors, and the evaluation factor data of all random points that are finally retained in the comprehensive evaluation factor attribute table, constitute the optimization factor set.

6. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 5, is characterized in that... In step S6, the process of constructing the sample dataset is as follows: Data on each evaluation factor corresponding to the evaluation factor type within the optimization factor set is collected from the historical landslide boundaries or landslide hazard points in the research mining area. This data forms a landslide sample and serves as a positive sample. Random points within the landslide area from the random points inserted in S5 are defined as landslide points, and the category label is defined as 1. Data on each evaluation factor corresponding to the evaluation factor type within the optimization factor set is collected from relatively stable areas far from known landslide bodies, areas with obvious deformation anomalies, and areas with high mining impact anomalies. This data forms a non-landslide sample and serves as a negative sample. Random points within the relatively stable area from the random points inserted in S5 are defined as non-landslide points, and the category label is defined as 0. All positive and negative samples form a sample dataset, and the sample dataset is divided into a training set and a test set according to a preset ratio.

7. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 6, is characterized in that... Step S7 specifically includes: S7.1: Construct a machine learning model, and train the machine learning model using the training set in step S6; wherein, the evaluation factor data corresponding to each training sample in the training set are arranged into a multi-dimensional optimized factor feature vector and used as input features, and the category label is the output category; S7.2: Each grid point in the research area is considered as an evaluation unit. The evaluation factor observation values ​​corresponding to the evaluation factor types in the optimization factor set are extracted from the comprehensive evaluation factor library constructed in step S4. The optimization factor feature vector is constructed in the same order as that in the training stage of step S7.

1. It is then input into the trained machine learning model to obtain the probability value of each evaluation unit belonging to the landslide category. S7.3: Write back the landslide probability values ​​of each evaluation unit to the corresponding evaluation unit according to their spatial location, generate a landslide susceptibility probability raster layer, and export the susceptibility evaluation result tif file; S7.4: Import the tif file into ArcGIS Pro, and use the natural breakpoint method to divide the research mining area into five levels of landslide susceptibility zones: extremely low, low, medium, high, and extremely high, forming a susceptibility zone map. At the same time, calculate the area of ​​the extremely high and high risk zones and generate a column chart.

8. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 7, is characterized in that... Step S8 specifically includes: S8.1: For any evaluation unit Suppose it contains If there are several evaluation factors, then the machine learning model will evaluate the units based on the SHAP value. The predicted output is represented as: in, Baseline value; For the first The SHAP value of each evaluation factor represents its contribution to the prediction result of the current evaluation unit. The calculation formula is as follows: in, A set of indices representing all evaluation factors; The total number of evaluation factors; Not including the first Any subset of evaluation factors; For subset The number of evaluation factors; Shapley weighting coefficients; To use only subsets When considering the evaluation factors, the model obtains the predicted values; To join the Model output after evaluating factors; For only subsets The input sample consists of the actual values ​​of those evaluation factors; when When, it indicates the first Several evaluation factors have a promoting effect on landslide occurrence; when When, it indicates the first Each evaluation factor has an inhibitory effect on landslide occurrence; The larger the value, the higher the value. The more significant the impact of each evaluation factor on the prediction results; S8.2: For any evaluation unit obtained from the calculation All evaluation factors Sort the results by the smallest to largest to obtain the contribution results of the evaluation factors for each evaluation unit; S8.3: Take the absolute value of the SHAP value of each evaluation factor in all evaluation units and calculate the average to obtain the overall contribution of each evaluation factor; specifically, suppose the study area has a total of The evaluation unit, the first The evaluation factor is at the _ The SHAP value on each evaluation unit is [value]. Then the first Overall contribution of each evaluation factor Defined as its average absolute SHAP value across all evaluation units: For all evaluation factors The overall contribution ranking of each evaluation factor is obtained by sorting them from largest to smallest. SHAP bee colony diagrams and feature dependency scatter plots are then drawn to interpret the prediction results of the machine learning model and identify the main controlling factors affecting landslide susceptibility.

9. The method for evaluating the susceptibility of open-pit mine landslides considering the impact of mining, as described in claim 1, is characterized in that... The machine learning models include random forest models, extreme gradient boosting models, support vector machine models, or logistic regression models.

Images