Intelligent early warning method for karst ground collapse combining fusion mechanism with multi-source data

Abstract

This invention discloses an intelligent early warning method for karst landslides that integrates mechanistic mechanisms and multi-source data, belonging to the field of karst geological disaster monitoring and early warning technology. The method first identifies key monitoring areas based on the hazard assessment results of surface karst landslides and constructs a database of roof stability under different geometric structures and physical and mechanical conditions through multi-condition mechanistic simulation. Subsequently, multi-source monitoring data from typical landslide hazard points are collected, and combined with historical landslide cases, surface hazard indicators, roof stability characteristics, and monitoring data are fused and modeled to obtain the probability of landslides at the hazard points. Furthermore, interpretable analysis methods are used to extract early warning thresholds under multiple indicator combinations, forming a four-level (red, orange, yellow, and blue) landslide early warning rule. This method achieves deep integration of mechanistic information and multi-source monitoring data, enabling real-time and interpretable intelligent early warning of karst landslide hazard points, and has significant engineering application value.

Description

Technical Field

[0001] This invention relates to the field of karst geological disaster monitoring and early warning technology, specifically to an intelligent early warning method for karst ground collapse that integrates mechanisms and multi-source data. It can be used for real-time risk identification, early warning level classification, and trend analysis of karst ground collapse hazard points, and belongs to the interdisciplinary technology direction of intelligent monitoring and early warning of geological disasters. Background Technology

[0002] Karst ground subsidence is one of the most typical and sudden geological hazards in karst regions of southern my country, characterized by its high degree of concealment, complex causes, significant destructiveness, and difficulty in early warning. Its formation process is typically controlled by the coupled effects of multiple factors, including underground cavity structures, the mechanical properties of the overlying rock and soil, changes in groundwater dynamics, rainfall infiltration, and surface load disturbances, exhibiting obvious nonlinear, multi-field coupling, and spatiotemporal abrupt change characteristics. With the acceleration of urbanization, the frequency of ground subsidence in karst areas is on the rise, posing a serious threat to urban infrastructure and the safety of residents' lives and property.

[0003] Existing karst collapse early warning methods mainly include the following categories:

[0004] (1) Early warning method based on empirical indicators. It makes empirical judgments based on single or a few indicators such as rainfall, groundwater level changes, and surface cracks, but lacks the characterization of underground structure and mechanical processes, and the accuracy of early warning is limited.

[0005] (2) Early warning methods based on surface deformation monitoring. GNSS, tiltmeters, InSAR and other instruments are used to monitor changes in surface displacement, but surface deformation often lags behind the destruction of underground structures, making it difficult to achieve early warning.

[0006] (3) Early warning methods based on statistical models or machine learning. Risk models are constructed using historical collapse data, but they fail to effectively integrate information on underground structural mechanisms, resulting in insufficient model generalization ability.

[0007] (4) Stability analysis method based on numerical simulation. The stability of the roof is analyzed by seepage-stress coupling simulation, but it is usually used for single-point analysis, which is difficult to link with real-time monitoring data and difficult to form a dynamic early warning system.

[0008] In summary, existing technologies generally have the following shortcomings: (1) The lack of in-depth integration of mechanism information and monitoring data makes it difficult to reflect the coupling effect between underground structural changes and external disturbances; (2) The lack of systematic stability analysis of multiple working conditions and multiple factors makes it impossible to fully characterize the roof failure mode under different geological conditions; (3) The lack of an interpretable early warning threshold system makes it difficult to form transparent and traceable early warning rules; (4) The lack of a unified four-level early warning system makes it difficult to meet the needs of engineering management departments for graded early warning; (5) Lacking real-time intelligent early warning capabilities for potential hazards, it is difficult to achieve dynamic identification and trend analysis of collapse risks.

[0009] In recent years, industry guidelines have proposed building an intelligent early warning system based on "mechanism-data fusion." This system would achieve four levels of early warning for karst ground subsidence: red, orange, yellow, and blue, through multi-indicator fusion, seepage-stress coupling mechanism analysis, dynamic monitoring data processing, anomaly identification, and the extraction of interpretable early warning thresholds. However, current technologies lack a systematic method that can simultaneously integrate area hazard, mechanism simulation results, multi-source monitoring data, and historical subsidence information.

[0010] Therefore, there is an urgent need for an intelligent early warning method for karst land collapse that integrates mechanisms and multi-source data, so as to achieve real-time, interpretable, and quantifiable early warning and identification of potential collapse points and improve the early warning capability for karst land collapse. Summary of the Invention

[0011] The purpose of this invention is to provide an intelligent early warning method for karst land collapse that integrates mechanism and multi-source data to solve the problems mentioned in the background art, such as the separation of mechanism information and monitoring data, the inability to interpret the early warning threshold, reliance on a single monitoring indicator, and the difficulty in achieving real-time dynamic early warning.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent early warning of karst landslides that integrates mechanisms and multi-source data includes: S1. Determine key monitoring areas based on the risk assessment results of karst surface collapse in the area; S2. Construct seepage-stress coupling numerical simulation models under different geometric structures, physical and mechanical parameters and hydrological conditions to obtain the stability characteristics of the roof under multiple working conditions. S3. Collect multi-source monitoring data of potential collapse sites; S4. The area hazard index, roof stability characteristics and multi-source monitoring data are integrated and modeled to obtain the probability of collapse of the hidden danger point; S5. Based on interpretable analysis methods, extract early warning thresholds under multiple index combinations to form a four-level collapse early warning rule; S6. Output the collapse warning level of the potential hazard point according to the warning rules.

[0013] Preferably, the seepage-stress coupling numerical simulation in S2 includes the joint solution of the pore water pressure field and stress field, and outputs at least one of the following: top plate displacement, stress distribution, plastic zone range and stability index.

[0014] Preferably, the seepage-stress coupled numerical simulation model in S2 generates a roof stability database by changing the roof thickness, cavity span, soil and rock parameters, groundwater level, and rainfall infiltration conditions.

[0015] Preferably, the multi-source monitoring data mentioned in S3 includes at least one of rainfall, groundwater level, air pressure, vertical displacement of the ground surface, and other monitoring data that can reflect changes in the surface or underground environment; the monitoring data can be obtained through automatic rain gauges, water level gauges, barometers, InSAR inversion, crack gauges, surface load monitoring systems, etc., and the specific monitoring equipment is not limited.

[0016] Preferably, the fusion modeling described in S4 establishes a collapse probability prediction model through machine learning methods:

[0017] in, The probability of collapse at the potential hazard point. As an indicator of area hazard, Characteristics of roof stability This is multi-source monitoring data; The probability mapping function is used to map the surface hazard index, the roof stability characteristics obtained from seepage-stress coupling numerical simulation, and multi-source monitoring data into the probability of imminent collapse. It can be trained through supervised learning, semi-supervised learning, or transfer learning. The model training can use historical collapse cases as positive samples and non-collapse periods as negative samples. The expressive power of the model can be enhanced through feature engineering, time window construction, and feature interaction.

[0018] Preferably, the fusion modeling further includes using historical collapse cases to map probability functions. To conduct training or calibration.

[0019] Preferably, the interpretable analysis method in S5 includes any one of feature contribution analysis, sensitivity analysis, disturbance analysis, or local interpretation methods, used to calculate the influence of each input feature on the probability of collapse, obtain a feature contribution vector, and determine the warning threshold under the combination of multiple indicators based on the feature contribution vector. The calculation formula for the feature contribution vector is as follows:

[0020] in, This indicates the impact of the area hazard index on the probability of collapse; This indicates the influence of the roof stability characteristics; This indicates the impact of the monitoring data.

[0021] Preferably, a threshold surface under a combination of multiple indicators is constructed based on the changing trend of the feature contribution vector:

[0022] in, These are any three types of variables from monitoring data, roof stability characteristics, and area hazard indicators; It is a three-dimensional threshold mapping function used to describe the joint collapse relationship among monitoring data, mechanism stability characteristics and surface hazard indicators; the threshold surface can be constructed by segmentation of feature contribution, probability distribution analysis, fitting of equal risk curves, etc., without limiting the specific method.

[0023] Preferably, the four-level collapse early warning rule in S5 is based on the probability of collapse. With threshold function Based on this relationship, potential hazard points are divided into four warning levels: red, orange, yellow, and blue. This is a combined variable of monitoring data and stability characteristics; The warning level can be determined in the following ways: when P ≥ T red The alert level was red. when T orange ≤ P < T red The current alert level is orange. when T yellow ≤ P < T orange The current alert level is yellow. when P < T yellow The alert level is currently blue. in, T red , T orange , T yellow For a multi-index combined threshold derived from the threshold surface, specific values ​​can be determined using methods such as experience, equal division, and natural breakpoints.

[0024] Preferably, the output in S6 includes the probability of collapse, the warning level, key influencing factors and their contribution, the risk change trend, and disposal suggestions; wherein the disposal suggestions may include engineering measures such as increased monitoring, drainage measures, surface load control, and on-site inspections, without limiting the specific content.

[0025] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention proposes a fusion framework of “mechanism simulation + multi-source monitoring data”, which integrates seepage-stress coupling mechanism information and real-time monitoring data into the early warning model, significantly improving the accuracy and foresight of the identification of the risk of collapse.

[0026] (2) The present invention constructs a multi-condition roof stability database, which breaks through the limitations of traditional reliance on a single condition or experience-based judgment, and can systematically characterize the roof failure mode and stability changes under different geological conditions.

[0027] (3) This invention introduces an interpretable analysis method to construct early warning thresholds under multiple index combinations, thereby realizing the transparency, interpretability and traceability of early warning rules, and solving the problem that traditional "black box" early warning models are difficult to interpret.

[0028] (4) This invention integrates area hazard, mechanism simulation results, monitoring data and historical collapse information to form a complete collapse probability prediction system, which can realize real-time dynamic early warning and trend analysis of hidden danger points.

[0029] (5) The method of the present invention is applicable to various scenarios such as urban construction, infrastructure operation and geological disaster monitoring in karst areas, and has good engineering applicability and promotion value. Attached Figure Description

[0030] Figure 1 A flowchart of an intelligent early warning method for karst landslides that integrates mechanisms and multi-source data; Figure 2 This is a data flow diagram in the present invention, illustrating the surface hazard index, the roof stability characteristics obtained from seepage-stress coupling numerical simulation, and the joint input of multi-source monitoring data into the collapse probability prediction model. Figure 3 This is a schematic diagram of the interpretable analysis method in this invention; Figure 4 This is a schematic diagram of the four-level collapse early warning output in this invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1: Please see Figure 1 This invention proposes an intelligent early warning method for karst landslides that integrates mechanisms and multi-source data, comprising the following steps: S1. Determine key monitoring areas based on the results of the karst surface collapse risk assessment: Multi-source basic data, including karst geology, hydrology, topography, and human activities, were collected to construct a machine learning model for assessing the risk of surface karst ground collapse. The surface risk index of each spatial unit in the study area was obtained, and based on this, extremely high to high risk areas were identified as key monitoring areas.

[0033] S2. Construct a multi-condition seepage-stress coupling mechanism simulation model to obtain roof stability characteristics: Based on different geometric structures, physical and mechanical parameters, and hydrological conditions, a seepage-stress coupled numerical simulation model was established. By changing the roof thickness, void span, soil and rock parameters, groundwater level, and rainfall infiltration conditions, multiple simulation scenarios were generated. The model outputs roof stability characteristics such as roof displacement, stress distribution, plastic zone range, and stability indices, forming a roof stability database. The seepage-stress coupled numerical simulation specifically includes: Based on the karst geological structure of key monitoring areas, two-dimensional or three-dimensional seepage-stress coupling models are constructed. By solving the groundwater seepage control equation and consolidation equilibrium equation, the stability characteristics of the roof under different working conditions are obtained.

[0034] The working conditions mentioned include, but are not limited to: changes in roof thickness, changes in cavity span, changes in soil and rock strength parameters, changes in groundwater level, and changes in rainfall infiltration intensity.

[0035] The stability characteristics of the simulated output may include the vertical displacement field of the top plate. w ( x , y , z Principal stress distribution σ 1, σ 3. Volume of the plastic zone V p Safety factor F s Or failure mode type, etc., are used to characterize the stability state of the top plate structure under different working conditions.

[0036] S3. Collect multi-source monitoring data from typical potential collapse sites: Monitoring data includes R ( t ), groundwater level W ( t air pressure P ( t InSAR vertical displacement D v ( t ), Changes in the width of surface cracks C ( t ), changes in surface load L (t The monitoring data should be obtained from at least one of the following time series data, combined with historical subsidence cases, to form a monitoring-event dataset. The monitoring data can be obtained through automatic rain gauges, water level gauges, barometers, InSAR inversion, crack gauges, surface load monitoring systems, etc., and the specific monitoring equipment is not limited.

[0037] S4. Construct a collapse probability prediction model based on area hazard, roof stability characteristics, and multi-source monitoring data: A collapse probability prediction model is established by fusing area hazard indicators, roof stability characteristics obtained from seepage-stress coupling numerical simulation, and multi-source monitoring data to create a fusion model. This model maps input features to collapse probabilities and outputs the collapse probability of potential hazard points. The collapse probability prediction model is constructed using statistical learning models, machine learning models, deep learning models, or combinations thereof, to achieve the following mapping relationships:

[0038] in, P The probability of collapse at the potential hazard point; D As an indicator of area hazard; S The top plate stability eigenvectors obtained from seepage-stress coupling simulation; M For multi-source monitoring data feature vectors; f It is a probability mapping relationship, which can be trained through supervised learning, semi-supervised learning or transfer learning.

[0039] Model training can use historical collapse cases as positive samples and non-collapse periods as negative samples, and enhance the model's expressive power through feature engineering, time window construction, and feature interaction.

[0040] S5. Based on interpretable analysis methods, extract early warning thresholds under multiple indicator combinations and construct a four-level collapse early warning rule: An interpretable analysis method is employed to assess the impact of each input feature on the probability of landslide, constructing a threshold surface under a multi-indicator combination to form a four-level (red, orange, yellow, and blue) landslide early warning rule, thus achieving interpretability and traceability of the early warning thresholds. The four-level landslide early warning rule is based on the probability of landslide. P With threshold function T Based on the relationship, potential hazards are classified into four warning levels: red, orange, yellow, and blue. The warning level can be determined in the following ways: when P ≥ T red The alert level was red. when T orange ≤ P < T red The current alert level is orange. when T yellow ≤ P < T orange The current alert level is yellow. when P < T yellow The alert level is currently blue.

[0041] in, T red , T orange , T yellow For a multi-index combined threshold derived from the threshold surface, specific values ​​can be determined using methods such as experience, equal division, and natural breakpoints.

[0042] Interpretable analysis methods include any one of feature contribution-based, sensitivity analysis, perturbation analysis, or local interpretation methods, used to calculate the influence of input features on the probability of collapse, and obtain a feature contribution vector:

[0043] in, I D The impact of surface hazard indicators on the probability of collapse; I S The influence of characterizing the stability features of the roof; I M Characterize the impact of monitoring data.

[0044] Based on the trend of feature contribution changes, a threshold surface is constructed under a combination of multiple indicators:

[0045] in, This represents variables in the monitoring data, including but not limited to rainfall. Groundwater level changes InSAR vertical displacement Crack width variation Monitoring indicators that can reflect dynamic changes in the surface or underground environment; Variables representing roof stability characteristics include the safety factor obtained from seepage-stress coupling numerical simulation. Plastic zone volume Vp Maximum displacement of the top plate w max Mechanistic characteristics reflecting the stability of underground structures, such as principal stress distribution; Variables representing area hazard indicators include the area karst collapse hazard index. D And its derived regional-scale geological environmental characteristics.

[0046] 3D threshold mapping function This is used to describe the joint collapse-causing relationship among monitoring data, mechanistic stability characteristics, and surface hazard indicators. Its construction methods include, but are not limited to: 1. Threshold fitting method based on feature contribution segmentation: This method utilizes feature contribution vectors obtained through interpretable analysis methods such as SHAP. Segmented fitting is performed on intervals with different feature combinations to form segmented threshold surfaces.

[0047] 2. Statistical fitting method based on probability distribution: By analyzing the probability of collapse... exist The spatial distribution characteristics are used to construct a threshold surface using kernel density estimation, equal probability curves, or quantile methods.

[0048] 3. Function fitting method based on equal-risk curves: By fitting a function that satisfies... The threshold function is obtained from the isorisk curve (or surface) and used to divide the four-level warning intervals: red, orange, yellow, and blue.

[0049] Threshold surface function The specific mathematical form is not limited; it can be a linear function, a piecewise function, an exponential function, a logistic regression function, a tree model decision boundary, or other function forms that can reflect the risk interface of a multi-indicator combination.

[0050] The threshold surface can be constructed by segmenting feature contribution, probability distribution analysis, and fitting equal-risk curves, without limiting the specific method.

[0051] S6. Output the imminent collapse warning level of the potential hazard point according to the warning rules: Based on the relationship between the probability of imminent collapse and the threshold surface, the system outputs the imminent collapse warning level for potential hazard points. It can further generate warning analysis information including key influencing factors, risk change trends, and remedial recommendations. The output specifically includes: imminent collapse probability, warning level, key influencing factors and their contribution, risk change trends, future risk predictions, possible collapse mechanisms, and remedial recommendations. Remedial recommendations may include engineering measures such as increased monitoring, drainage measures, surface load control, and on-site inspections, without limiting the specific content.

[0052] Example 2: Based on Example 1, but with some differences, the intelligent early warning method for karst land collapse based on the fusion mechanism and multi-source data proposed in this invention will be further explained below with reference to relevant figures and specific examples. The specific content is as follows: Step 1: Area Hazard Assessment and Determination of Key Monitoring Areas: Data from multiple sources, including karst geology, hydrology, topography, and human activities, were collected from the study area and refined into 18 karst ground collapse risk assessment factors.

[0053] An XGBoost model was used to construct a surface hazard assessment model. Eighteen feature factors were input, and the output was a hazard score for each 10 m × 10 m grid cell. The value range is 0–1.

[0054] For example:

[0055] Risk rating The area was designated as a key monitoring area.

[0056] Step 2: Simulation of seepage-stress coupling mechanism under multiple working conditions: (I) Model Construction As shown in Figure 2, based on geological data from the key monitoring area, a three-dimensional seepage-stress coupling model was established using FLAC3D software. The model dimensions are 40 m × 40 m × 25 m, and it includes: Topsoil layer (clay, silt) Top slab soil (sandy clay) Karst cavities (span 3–10 m) Groundwater system (initial water level 10 m) (II) Operating Condition Settings Set up 120 sets of working conditions, as shown in the example below:

[0057] (III) Analog Output Output for each operating condition: Maximum vertical displacement of the top plate 1.2–18.5 mm Plastic region volume : 0.01–2.3 m 3 Safety factor : 0.85–2.10 Failure modes: flexural failure / shear failure / global instability For example:

[0058] These features constitute the stability feature vector. S .

[0059] Step 3: Multi-source monitoring data collection: As shown in Figure 2, monitoring data were collected from typical potential hazard points: Rainfall R ( t ): 0–85 mm / h groundwater level W ( t ): 8–14 m air pressure P ( t ): 980–1020 hPa InSAR vertical displacement D v ( t ): -3.5–+2.1 mm / cycle Surface crack width C ( t ): 0–12 mm Surface load changes L ( t 0–80 kPa For example, 30-day monitoring data for a potential hazard site:

[0060] Construct a monitoring feature vector using the above data. .

[0061] Step 4: Construction of the Collapse Probability Prediction Model: A collapse probability prediction model is constructed using the XGBoost classification model. Input: Surface hazard index D Stability characteristics S (like F s , w max , V p ) Monitoring characteristics M (like R ( t ), W ( t ), D v ( t )) Model outputs probability of collapse P .

[0062] The training set includes: 18 collapse incidents in the past 5 years (positive sample) 1200 monitoring periods without collapse (negative samples) Model performance: AUC = 0.91 Accuracy = 0.87 Recall rate = 0.82 Example of model output for a potential hazard point:

[0063] Step 5: Interpretable Analysis and Threshold Surface Construction: As shown in Figure 3, the XGBoost model is analyzed using the SHAP method to obtain the feature contribution: For example, a potential hazard point:

[0064] Constructing a three-dimensional threshold surface based on SHAP values:

[0065] For example, the fitting yielded: when mm time: high-risk area when mm time: medium risk area Threshold surfaces are used to classify warning levels.

[0066] Step Six: Level Four Collapse Warning Output: As shown in Figure 4, based on the probability of collapse P With threshold surface T Based on the relationship, a level four warning is issued:

[0067] For example, a potential hazard point: Inputs: Rainfall 45 mm, water level +1.6 m, InSAR displacement -2.3 mm. F s =0.92 Model output:

[0068] Threshold surface judgment: falling into the high-risk zone Final Warning: Red Alert And output: Key influencing factors: InSAR displacement (31%), water level (22%) F s (18%) Risk trend: continued to rise over the past week Recommended measures: Conduct on-site inspections immediately, limit surface loads, and implement drainage if necessary.

[0069] The above description is only intended to help understand the method and core essence of the present invention, but the scope of protection of the present invention is not limited thereto. For those skilled in the art, any equivalent substitutions or modifications made within the technical scope disclosed in this invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for intelligent early warning of karst landslides that integrates mechanisms and multi-source data, characterized in that, include: S1. Determine key monitoring areas based on the risk assessment results of karst surface collapse in the area; S2. Construct seepage-stress coupling numerical simulation models under different geometric structures, physical and mechanical parameters and hydrological conditions to obtain the stability characteristics of the roof under multiple working conditions. S3. Collect multi-source monitoring data of potential collapse sites; S4. The area hazard index, roof stability characteristics and multi-source monitoring data are integrated and modeled to obtain the probability of collapse of the hidden danger point; S5. Based on interpretable analysis methods, extract early warning thresholds under multiple index combinations to form a four-level collapse early warning rule; S6. Output the collapse warning level of the potential hazard point according to the warning rules.

2. The method according to claim 1, characterized in that, The seepage-stress coupling numerical simulation described in S2 includes the joint solution of the pore water pressure field and stress field, and outputs at least one of the following: top plate displacement, stress distribution, plastic zone range, and stability index.

3. The method according to claim 1, characterized in that, The seepage-stress coupling numerical simulation model described in S2 generates a roof stability database by changing the roof thickness, cavity span, soil and rock parameters, groundwater level, and rainfall infiltration conditions.

4. The method according to claim 1, characterized in that, The multi-source monitoring data mentioned in S3 includes at least one of the following: rainfall, groundwater level, air pressure, vertical displacement of the ground surface, and other monitoring data that can reflect changes in the surface or underground environment.

5. The method according to claim 1, characterized in that, The fusion modeling described in S4 establishes a collapse probability prediction model using machine learning methods: in, The probability of collapse at the potential hazard point. As an indicator of area hazard, Characteristics of roof stability This is multi-source monitoring data; This is a probability mapping function used to map the area hazard index, the roof stability characteristics obtained from seepage-stress coupling numerical simulation, and multi-source monitoring data into the probability of imminent collapse.

6. The method according to claim 5, characterized in that, The fusion modeling further includes using historical collapse cases to map probability functions. To conduct training or calibration.

7. The method according to claim 1, characterized in that, The interpretable analysis method described in S5 includes any one of the following: feature contribution degree, sensitivity analysis, disturbance analysis, or local interpretation method. It is used to calculate the influence of each input feature on the probability of collapse, obtain a feature contribution degree vector, and determine the warning threshold under a multi-indicator combination based on the feature contribution degree vector. The formula for calculating the feature contribution degree vector is as follows: in, This indicates the impact of the area hazard index on the probability of collapse; This indicates the influence of the roof stability characteristics; This indicates the impact of the monitoring data.

8. The method according to claim 7, characterized in that, Based on the changing trend of the feature contribution vector, a threshold surface under the combination of multiple indicators is constructed: in, These are any three types of variables from monitoring data, roof stability characteristics, and area hazard indicators; It is a three-dimensional threshold mapping function used to describe the joint collapse-causing relationship among monitoring data, mechanism stability characteristics, and surface hazard indicators.

9. The method according to claim 1, characterized in that, The four-level collapse early warning rule described in S5 is based on the probability of collapse. With threshold function Based on this relationship, potential hazard points are divided into four warning levels: red, orange, yellow, and blue. This is a combined variable of monitoring data and stability characteristics.

10. The method according to claim 1, characterized in that, The outputs described in S6 include the probability of collapse, warning level, key influencing factors and their contribution, risk change trend, and disposal recommendations.

Images