Quantitative evaluation method for area random slope instability influence range

Abstract

The invention discloses a quantitative evaluation method for a region random slope instability influence range, and relates to the technical field of geological disasters. The method comprises the following steps: acquiring historical slope geological disaster data and geographic space data of a research area to establish a database; constructing a refined grid unit and a slope unit for the research area; carrying out susceptibility evaluation on the research region by adopting a machine learning algorithm to obtain a source target region of the area random slope instability; setting a multi-factor and multi-level uniform design experiment for the numerical simulation parameters by adopting uniform design; constructing a surface domain random slope instability range pseudo sample set based on slope units; and constructing a training data set to train a machine learning agent model, and taking the optimized machine learning agent model as a random slope instability influence range machine learning agent model. According to the method provided by the invention, the selection precision of the area random slope instability object source target region is improved, and meanwhile, the quantitative evaluation of the area random slope instability influence range is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of geological disasters, and in particular relates to a quantitative evaluation method for the impact range of random slope instability in a surface area. Background Art

[0002] Slope instability is one of the important manifestations of geological disasters. Its occurrence poses a major threat to regional security and human activities, especially in mountainous and hilly areas. Landslides caused by slope instability have serious consequences for the safety of life and property, infrastructure construction and ecological environment. In recent years, with the intensification of global climate change and the rapid advancement of urbanization, the frequency and scale of slope instability disasters have increased significantly. Slope instability and the landslides it causes are affected by a variety of factors, including terrain conditions, rainfall intensity, soil type, vegetation coverage and engineering activities. The complex coupling between these factors makes the prediction of slope instability and the assessment of its impact range subject to great uncertainty. Therefore, it is crucial to develop a scientific, systematic and efficient quantitative evaluation method for the prediction and prevention of geological disasters.

[0003] The existing quantitative evaluation methods for the impact range of random slope instability mainly rely on statistical models and numerical simulations. In engineering, there are sporadic reliable data on the impact range of landslide instability, but there is no effective and satisfactory data set based on measured data for the impact range of landslide instability. Although these traditional methods have laid the foundation for early slope instability prediction, they still have limitations. First, the traditional statistical methods lack systematicity and comprehensiveness. By establishing a simple linear fitting relationship to predict the impact range of slope instability, they fail to fully consider multiple nonlinear factors such as terrain, rainfall, soil properties and vegetation coverage and their complex interactions, resulting in insufficient scientificity and reliability of the prediction results. Numerical simulation methods are mainly used for the inversion analysis of landslide processes, focusing on the movement process and impact range of a single slope. However, this method is highly dependent on expert experience for the selection of slope instability source target areas, and the inversion of the landslide impact range requires adjustment of slope rock and soil parameters. Since the parameters of random slopes are different, it is difficult to achieve the prediction of the impact range of random slope instability in the area by numerical simulation methods alone, and its reliability is difficult to guarantee, which is difficult to meet the actual needs of modern engineering construction, disaster prevention and emergency response for high-precision prediction.

[0004] In view of the above shortcomings, the present invention proposes a quantitative evaluation method for the influence range of random slope instability in the area. Based on the refined grid unit combined with the comprehensive analysis of multiple influencing factors and the dynamic evolution simulation, the accurate selection of the source target area of ​​random slope instability in the area and the quantitative prediction of the influence range are achieved. The method uses a machine learning algorithm to accurately evaluate the susceptibility of landslides, and selects high-prone areas as the source target areas of random slope instability in the area. On this basis, combined with numerical simulation technology, the evolution process and influence range of random slope instability in the area under the combination of physical and mechanical parameters are dynamically analyzed to construct a pseudo sample set of the influence range of random slope instability in the area. Redundant pseudo samples are eliminated by data mining methods, and samples in the pseudo sample set that meet the actual landslide range distribution are screened out as data sets for training machine learning proxy models. Then, the proxy model of the influence range of random slope instability in the area based on machine learning is trained through the screened sample set, and the robustness and prediction accuracy of the model are improved through hyperparameter optimization and cross-validation. This method not only comprehensively integrates multiple key influencing factors, but also can dynamically simulate the slope instability process and quantitatively evaluate its impact range, thereby greatly improving the scientificity and applicability of the prediction results. Summary of the invention

[0005] The purpose of the present invention is to provide a method for quantitatively evaluating the influence range of random slope instability in a surface area, so as to solve the problems of low prediction accuracy, factor redundancy, imperfect experimental design, etc. existing in the existing traditional methods proposed in the above background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention proposes a quantitative evaluation method for the influence range of random slope instability in a surface area, comprising the following steps:

[0008] S1. Obtain historical slope geological hazard data and geospatial data in the study area and establish a database;

[0009] S2. Construct fine grid units and slope units for the study area, extract geospatial data to corresponding grid units and slope units, and establish a data set for quantitative evaluation of the study area;

[0010] S3. Use machine learning algorithms to evaluate the susceptibility of the study area and obtain the distribution of high-susceptibility areas in the study area as the source target area for random slope instability in the area;

[0011] S4, using uniform design to set multi-factor and multi-level uniform design experiments for numerical simulation parameters;

[0012] S5. Based on the source target area of ​​random slope instability in the surface domain, the simulation results are obtained based on the uniform design experiment, the simulation results are extracted to the corresponding slope units, and a pseudo sample set of the range of random slope instability in the surface domain based on the slope units is constructed;

[0013] S6. Construct a training data set based on historical slope geological disaster data and the screening of pseudo sample sets of random slope instability in the area;

[0014] S7. Construct a machine learning proxy model, use the training data set to train and optimize the machine learning proxy model, and use the optimized machine learning proxy model as the random slope instability influence range machine learning proxy model. Based on the data set of quantitative evaluation of the study area, the random slope instability influence range machine learning proxy model is used to predict the influence range.

[0015] Preferably, the historical slope geological disaster data in S1 include landslide point data or surface data, and the geographic spatial data include topographic factors, geological structure factors, hydrological factors, vegetation factors, meteorological factors, and human activity factors.

[0016] Further, the topographic factors include, but are not limited to, elevation, slope, slope aspect, slope shape, slope position, slope variability, slope aspect variability, micro-topography, profile curvature, plane curvature, land use type, slope type, terrain relief, terrain roughness, surface cutting depth, elevation variation coefficient, terrain moisture index;

[0017] The geological structural factors include but are not limited to the distance from the fault and the age of the strata;

[0018] The hydrological factors include but are not limited to groundwater type, water flow dynamic index, sediment transport index, runoff modulus, distance from the river, and river network density;

[0019] The vegetation factor includes but is not limited to the normalized vegetation index;

[0020] The meteorological factors include but are not limited to multi-year average rainfall, 24-hour rainfall, and previous effective rainfall;

[0021] The human activity factors include but are not limited to POI kernel density, distance from roads, and distance from buildings.

[0022] Preferably, S2 is specifically as follows:

[0023] S201. Use ARCGIS software to construct refined grid units and extract various factors from geographic spatial data into grid units as the susceptibility evaluation factor dataset of the study area;

[0024] S202. Use r.slopeunits to construct slope units that are consistent with the evaluation scale and accuracy, and extract various factors in the geographic spatial data into slope units as the evaluation factor data set for the slope instability impact range of the study area.

[0025] The susceptibility evaluation factor dataset and the slope instability impact range evaluation factor dataset are used as datasets for quantitative evaluation of the study area.

[0026] Furthermore, when extracting various factors from the geospatial data into slope units, the discrete factors are extracted into slope units by mode and the continuous factors are extracted into slope units by mean. Continuous factors include elevation, multi-year average rainfall, etc., and discrete factors include lithology, slope aspect, etc. Both continuous factors and discrete factors are geospatial data.

[0027] The division of fine grid units is used for susceptibility assessment to obtain the target area of ​​random slope source in the area. The division of slope units is to extract the simulated landslide area to the corresponding slope unit, which is used to construct a pseudo sample set for the machine learning agent model. Both grid units and slope units are evaluation units.

[0028] Preferably, S3 is specifically as follows:

[0029] S301, using a machine learning algorithm to construct a susceptibility assessment model; using landslide point data as positive samples, randomly selecting an equal amount of non-landslide point data as negative samples, and constructing a training data set for the susceptibility assessment model for model training;

[0030] S302, performing factor screening based on a factor screening method to eliminate redundant factors, and optimizing hyperparameters of the model;

[0031] S303. Based on the susceptibility assessment factor data set, an optimized susceptibility assessment model is used to obtain a susceptibility map, and high-susceptibility areas are screened out as source target areas for random slope instability in the surface area.

[0032] Furthermore, the machine learning algorithms in S301 include but are not limited to decision trees, random forests, support vector machines, artificial neural networks, XGBoost, and LightGBM, and the susceptibility evaluation model is constructed using one or a combination of two or more of the machine learning algorithms;

[0033] The factor screening method in S302 includes but is not limited to principal component analysis, recursive feature elimination, and geographic factor detector, and the factor screening adopts one or a combination of two or more of the factor screening methods;

[0034] The optimization method in S302 includes but is not limited to Bayesian optimization, gradient descent algorithm, grid search, random search, and genetic algorithm, and the optimization adopts one or a combination of more than two of the optimization methods.

[0035] Preferably, the numerical simulation parameters in S4 include but are not limited to substrate friction coefficient, cohesion, density, pore water pressure coefficient, and a uniform design experiment with 3 factors and 10 levels is set based on the numerical simulation parameters.

[0036] Preferably, the S5 is specifically as follows:

[0037] S501. Based on the source target area of ​​random slope instability in the surface area, numerical simulations are carried out under different parameter combinations through uniform design experimental sampling to obtain simulation results;

[0038] S502, extracting the simulation results to the corresponding slope units, obtaining the influence range of each random slope instability in the surface domain under different inducing conditions, and using this as a pseudo sample set of the random slope instability range of the surface domain.

[0039] Furthermore, the numerical simulation in S501 is based on the finite difference method and is simulated using MASSFLOW software, and the simulation results are imported into ARCGIS software to be extracted to the corresponding slope units.

[0040] Preferably, the screening in S6 is to eliminate redundant pseudo samples by a data mining method, and screen out samples that are consistent with the actual landslide range distribution in the area random slope instability pseudo sample set;

[0041] The data mining methods include but are not limited to generating adversarial networks, classification, clustering, and regression, and one or a combination of two or more of the data mining methods is selected and adopted.

[0042] Preferably, the S7 is specifically as follows:

[0043] S701, training the machine learning agent model using the training data set, and optimizing the hyperparameters of the machine learning agent model;

[0044] S702: Verify the robustness of the model by dividing the sample sets into different groups or using cross-validation. Evaluate the model performance based on the discreteness of the evaluation indicators. The evaluation indicators include but are not limited to R 2 , MAE, RMSE;

[0045] S703. Based on the slope instability impact range evaluation factor data set, a random slope instability impact range machine learning agent model is used to predict the random slope instability impact range under different factors.

[0046] Furthermore, the basic model of the machine learning proxy model in S701 includes but is not limited to random forest, support vector machine, artificial neural network, and XGBoost, and the machine learning proxy model is constructed based on one or a combination of two or more of the basic models;

[0047] The optimization method in S701 includes but is not limited to Bayesian optimization, gradient descent algorithm, grid search, random search, genetic algorithm, and the optimization adopts one or a combination of more than two of the optimization methods.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) The method of the present invention combines geospatial data, data mining methods, machine learning algorithms, uniform design experiments and numerical simulation technology to comprehensively evaluate the impact range of random slope instability in the surface area, and has the advantages of high accuracy and strong scalability. It not only solves the problems of insufficient systematicity and comprehensiveness and low prediction accuracy in traditional methods, but also provides scientific and reliable technical support for the prediction and prevention of the impact range of random slope instability in the surface area. The quantitative evaluation method of the impact range of random slope instability in the surface area proposed by the present invention can provide important decision-making basis for disaster warning, engineering construction and emergency management, and minimize the social and economic losses caused by geological disasters.

[0050] (2) The method of the present invention is based on refined grid units, which significantly improves the accuracy of target area selection for random slope instability sources in the surface area, and at the same time realizes the quantitative evaluation of the impact range of random slope instability in the surface area.

[0051] (3) The method of the present invention collects historical slope geological disaster data and geospatial data, combines machine learning algorithms with numerical simulation technology, and uses uniform design numerical experimental methods to obtain pseudo-sample sets of random slope instability impact ranges under arbitrary parameter combinations and in the surface area, and predicts the impact range of multiple random slopes that may become unstable from a wide-area perspective. This method provides quantitative prediction results for the impact range of slope instability that has not been monitored in the region, which provides an important basis for formulating scientific and reasonable geological disaster prevention and control plans, and helps optimize resource allocation and efficiently implement prevention and control measures. In addition, this method can also provide scientific decision-making support for urban planning, land use, infrastructure construction and other fields, and reduce potential disaster risks caused by inappropriate construction activities in high-risk areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of the method for quantitatively evaluating the influence range of random slope instability in the surface area of ​​the present invention;

[0053] Figure 2 A schematic diagram of constructing a refined grid unit in the present invention;

[0054] Figure 3 A schematic diagram of dividing the slope units in the present invention;

[0055] Figure 4It is a schematic diagram of the source target area (part) of the surface random slope instability in the present invention;

[0056] Figure 5 It is a schematic diagram of the influence range of random slope instability in the numerical simulation surface area in the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Embodiment 1:

[0059] See also Figure 1 The present invention provides a method for quantitatively evaluating the influence range of random slope instability, comprising the following steps:

[0060] S1: Obtain historical slope geological hazard data and geospatial data in the study area and establish a database; the details are as follows:

[0061] The historical slope geological hazard data and geospatial data of the study area include but are not limited to: landslide point coordinate data, elevation data (DEM), slope, slope aspect, slope shape, slope position, slope variability, slope aspect variability, micro-topography, profile curvature, plane curvature, land use type, slope type, terrain undulation, terrain roughness, surface cutting depth, elevation variation coefficient, terrain moisture index, groundwater type, water flow dynamics index, sediment transport index, runoff modulus, distance from river, river network density, distance from fault, stratigraphic age, normalized difference vegetation index, multi-year average rainfall, POI kernel density, and distance from road.

[0062] S2: Construct fine grid cells and slope cells and extract evaluation factors; the details are as follows:

[0063] Grid cells of different resolutions (10m, 20m, 30m, etc.) were created in the study area using ARCGIS software. Figure 2 ; Use r.slopeunits to divide the slope units, such as Figure 3 . The geospatial data will be extracted into grid cells and landslide points, and a machine learning database will be constructed. Each grid cell will contain the corresponding geographic factor data to provide support for subsequent evaluation.

[0064] S3: Use machine learning algorithms to evaluate the susceptibility of the study area and obtain the distribution of high-susceptibility areas in the study area as the source target area for random slope instability in the area; the details are as follows:

[0065] The landslide point data were taken as positive samples, and the same amount of non-landslide point data were randomly selected as negative samples and exported to csv files to construct a training data set for machine learning. Then, the machine learning algorithm was used to evaluate the susceptibility of the study area. The study area was classified according to the natural breakpoint method, and the high-susceptibility areas were selected as the source target areas for random slope instability in the surface area, such as Figure 4 .

[0066] Machine learning algorithms include but are not limited to: decision tree, random forest, support vector machine, artificial neural network, XGBoost, LightGBM and other models.

[0067] Factor screening methods include, but are not limited to: principal component analysis, recursive feature elimination, and geographic factor detectors.

[0068] Methods for optimizing machine learning models include but are not limited to: Bayesian optimization, gradient descent algorithm, grid search, random search, genetic algorithm, etc.

[0069] S4: Set up numerical experiments, adopt uniform design, and set multiple factors and multiple levels for numerical simulation parameters; the details are as follows:

[0070] Parameters required for numerical simulation: density (1800~2500kg / m 3 ), cohesion (0-30 kPa), base friction coefficient (0.25-0.75), 3 factors and 10 levels A uniform design experiment was conducted. Monte Carlo random sampling was used for density, cohesion, and base friction coefficient. At the same time, the pore water pressure coefficient (0-1) was uniformly designed into 6 fixed levels according to different rainfall intensities (no rain, light rain, moderate rain, heavy rain, rainstorm, and heavy rainstorm), and the above design was repeated for each level.

[0071] S5: Based on the set experimental level, multi-factor and multi-level uniform design sampling is carried out to perform numerical simulation and construct a pseudo sample set of random slope instability range in the surface area based on slope units; the details are as follows:

[0072] According to the designed numerical experiment plan, numerical simulation is carried out for each parameter combination. Based on the finite difference method, MASSFLOW software is used for simulation, and the simulation results are imported into ARCGIS software, such as Figure 5 (Partial results), extracted to the corresponding slope units, and finally constructed a pseudo sample set of random slope instability range in the surface domain based on slope units.

[0073] S6: Based on the collected landslide impact range data and the constructed pseudo-sample set of random slope instability in the surface area, redundant pseudo-samples are eliminated through data mining methods, and samples in the pseudo-sample set that are consistent with the actual landslide range distribution are screened out as the data set for training the machine learning proxy model; the details are as follows:

[0074] Through the generative adversarial neural network (GAN), redundant pseudo samples in the pseudo sample set of random slope instability in the surface area are eliminated. By continuously optimizing the network structure, samples that meet the actual landslide range are screened out and used as the data set for training the machine learning proxy model.

[0075] S7: Based on the screened random slope instability sample set, train the random slope instability influence range machine learning proxy model; specifically take the following:

[0076] The screened sample set is divided into a training set and a test set, and the machine learning agent model is trained. The model performance is improved through hyperparameter optimization, and the model performance is improved through cross-validation based on evaluation indicators (such as R 2 , MAE, RMSE, etc.), evaluate the model performance and verify the robustness of the model.

[0077] Machine learning agent models include but are not limited to: random forest, artificial neural network, support vector machine, XGboost, etc.

[0078] Model optimization methods include but are not limited to gradient descent algorithm, grid search, random search, Bayesian optimization, genetic algorithm, etc.

[0079] In summary, this embodiment uses ARCGIS software to preprocess the geospatial data of the study area and establish a spatial database of slope geological disasters in the study area. Then, a refined grid susceptibility evaluation model is constructed and the source selection mechanism of high-prone areas is revealed. The susceptibility of the study area is evaluated using the machine learning algorithm random forest, and the high-prone areas are screened out as the target areas of random slope instability sources in the area. Then, a uniform design experiment with 3 factors and 10 levels is set for the parameters required for the MASSFLOW numerical simulation. At the same time, the pore number pressure coefficient is designed to 6 levels according to different rainfall intensities, and the above experiment is repeated at each level to construct a pseudo-sample set of the influence range of random slope instability in the area, and to obtain a pseudo-sample set of the influence range of random slope instability in the area under any parameter combination conditions using the uniform design numerical experimental method. Then, redundant samples are eliminated and samples that meet the actual landslide range are screened by adversarial neural networks. Based on the screened sample set, a machine learning agent model is trained, and the model is optimized through hyperparameter optimization, and the model is evaluated through cross-validation to verify the robustness of the model. This method provides quantitative prediction results, which provides an important basis for formulating scientific and reasonable geological disaster prevention and control plans, and helps to optimize resource allocation and efficiently implement prevention and control measures.

[0080] The above description is only used to help understand the method of the present invention and its core essence, but the protection scope of the present invention is not limited thereto. For those skilled in the art in the art, equivalent replacement or change according to the technical solution and inventive concept of the present invention within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A quantitative evaluation method for the influence range of random slope instability in a surface area, characterized in that: The following steps are involved: S1. Obtain historical slope geological hazard data and geospatial data in the study area and establish a database; S2. Construct fine grid units and slope units for the study area, extract geospatial data to corresponding grid units and slope units, and establish a data set for quantitative evaluation of the study area; S3. Use machine learning algorithms to evaluate the susceptibility of the study area and obtain the distribution of high-susceptibility areas in the study area as the source target area for random slope instability in the area; S4, using uniform design to set multi-factor and multi-level uniform design experiments for numerical simulation parameters; S5. Based on the source target area of ​​random slope instability in the surface domain, the simulation results are obtained based on the uniform design experiment, the simulation results are extracted to the corresponding slope units, and a pseudo sample set of the range of random slope instability in the surface domain based on the slope units is constructed; S6. Construct a training data set based on historical slope geological disaster data and the screening of pseudo sample sets of random slope instability in the area; S7. Construct a machine learning proxy model, use the training data set to train and optimize the machine learning proxy model, and use the optimized machine learning proxy model as the random slope instability influence range machine learning proxy model. Based on the data set of quantitative evaluation of the study area, the random slope instability influence range machine learning proxy model is used to predict the influence range.

2. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 1 is characterized in that: The historical slope geological disaster data in S1 include landslide point data or surface data, and the geographic spatial data include topographic factors, geological structure factors, hydrological factors, vegetation factors, meteorological factors, and human activity factors.

3. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 2 is characterized in that: The S2 is specifically as follows: S201. Use ARCGIS software to construct refined grid units and extract various factors from geographic spatial data into grid units as the susceptibility evaluation factor dataset of the study area; S202. Use r.slopeunits to construct slope units that are consistent with the evaluation scale and accuracy, and extract various factors in the geographic spatial data into slope units as the evaluation factor data set for the slope instability impact range of the study area.

4. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 3 is characterized in that: The S3 is as follows: S301, using a machine learning algorithm to construct a susceptibility assessment model; using landslide point data as positive samples, randomly selecting an equal amount of non-landslide point data as negative samples, and constructing a training data set for the susceptibility assessment model for model training; S302, performing factor screening based on a factor screening method to eliminate redundant factors, and optimizing hyperparameters of the model; S303. Based on the susceptibility assessment factor data set, an optimized susceptibility assessment model is used to obtain a susceptibility map, and high-susceptibility areas are screened out as source target areas for random slope instability in the surface area.

5. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 4 is characterized in that: The machine learning algorithms in S301 include but are not limited to decision trees, random forests, support vector machines, artificial neural networks, XGBoost, and LightGBM, and the susceptibility evaluation model is constructed using one or a combination of two or more of the machine learning algorithms; The factor screening method in S302 includes but is not limited to principal component analysis, recursive feature elimination, and geographic factor detector, and the factor screening adopts one or a combination of two or more of the factor screening methods; The optimization method in S302 includes but is not limited to Bayesian optimization, gradient descent algorithm, grid search, random search, and genetic algorithm, and the optimization adopts one or a combination of more than two of the optimization methods.

6. A method for quantitatively evaluating the influence range of random slope instability in a surface area according to any one of claims 1 to 5, characterized in that: The numerical simulation parameters in S4 include but are not limited to substrate friction coefficient, cohesion, density, and pore water pressure coefficient.

7. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 6 is characterized in that: The S5 is specifically as follows: S501. Based on the source target area of ​​random slope instability in the surface area, numerical simulations are carried out under different parameter combinations through uniform design experimental sampling to obtain simulation results; S502, extracting the simulation results to the corresponding slope units, obtaining the influence range of each random slope instability in the surface domain under different inducing conditions, and using this as a pseudo sample set of the random slope instability range of the surface domain.

8. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 7 is characterized in that: The screening in S6 is to eliminate redundant pseudo samples by data mining method, and screen out samples that are consistent with the actual landslide range distribution in the random slope instability pseudo sample set in the area domain; The data mining methods include but are not limited to generating adversarial networks, classification, clustering, and regression, and one or a combination of two or more of the data mining methods is selected and adopted.

9. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 8 is characterized in that: The S7 is specifically as follows: S701, training the machine learning agent model using the training data set, and optimizing the hyperparameters of the machine learning agent model; S702. Verify the robustness of the model by dividing into different sample sets or using cross-validation, and evaluate the model performance according to the discrete degree of the evaluation index; S703. Based on the slope instability impact range evaluation factor data set, a random slope instability impact range machine learning agent model is used to predict the random slope instability impact range under different factors.

10. The method for quantitatively evaluating the influence range of random slope instability in a surface area according to claim 9 is characterized in that: The basic model of the machine learning proxy model in S701 includes but is not limited to random forest, support vector machine, artificial neural network, and XGBoost, and the machine learning proxy model is constructed based on one or a combination of two or more of the basic models; The optimization method in S701 includes but is not limited to Bayesian optimization, gradient descent algorithm, grid search, random search, and genetic algorithm, and the optimization adopts one or a combination of two or more of the optimization methods.

Images