Multi-scale risk quantitative assessment and dynamic early warning method for slope rockfall

Abstract

The application discloses a slope rockfall multi-scale risk quantitative evaluation and dynamic early warning method, belongs to the field of slope rockfall protection, and comprises the following steps: S1, synchronous acquisition of displacement, temperature, acoustic emission data of the whole process of rockfall gestation-instability-motion, obtaining a standardized feature set; S2, establishment of a multi-physical field feature-rockfall gestation stage mapping relationship, output of the gestation stage and instability probability of the slope rockfall; S3, taking the instability probability as an initial instability criterion of the discrete element method, simulating the whole process of gestation-instability-motion, and quantifying the harm of rockfall to sensitive targets in the accumulation area; S4, integration of multi-physical field real-time data and the quantitative result of step S3, establishment of a multi-level early warning system, and output of the early warning result. The above-mentioned slope rockfall multi-scale risk quantitative evaluation and dynamic early warning method realizes dynamic quantitative evaluation and accurate early warning of the slope rockfall risk through multi-source monitoring data fusion, multi-scale DEM adaptation and machine learning modeling.

Description

Technical Field

[0001] This invention relates to the field of slope rockfall protection technology, and in particular to a method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall. Background Technology

[0002] In the construction of transportation infrastructure such as highways and railways, due to terrain constraints, many routes inevitably pass near mountain slopes or natural slopes. These slopes are subject to long-term erosion by rainwater (affected by average annual rainfall and maximum daily rainfall) and seismic activity, making them prone to rockfall disasters. Rockfalls not only block transportation routes and threaten driving safety, but may also impact houses at the bottom of the slope, injuring passing vehicles and pedestrians, posing serious safety hazards to densely populated areas and sensitive targets. Therefore, accurate monitoring and effective early warning of slope rockfalls have urgent practical engineering significance.

[0003] While various technical approaches have been developed for monitoring rockfalls on slopes in the engineering field, all of them have significant limitations: 1. Manual inspection technology: It requires a large amount of manpower to conduct periodic inspections of the slope, especially for rockfall source areas with a slope greater than 50°. Due to the steep terrain and poor accessibility, the inspection is not only inefficient and costly, but also difficult to capture subtle signs of rockfall incubation (such as the expansion of microcracks in the rock mass and slight displacement). This results in a high degree of randomness in disaster early warning and a low probability of successful early warning, which cannot meet the requirements of real-time and reliability. 2. LiDAR monitoring technology: It relies on high-precision three-dimensional point cloud data to build slope terrain models. Although it can achieve fine depiction of terrain details, the equipment purchase and maintenance costs are high, the technical operation is complex, and it has poor adaptability to the harsh environment of open slopes (such as rainstorms and dense fog), making it difficult to operate stably for a long time. 3. Traditional visual technology monitoring: Rockfall identification is achieved by capturing images of the slope surface with a camera. However, the image quality is highly dependent on the ambient brightness. In strong light, backlight or rainy weather, problems such as blurred images and failure of feature extraction are likely to occur, which greatly increases the difficulty of the perception task and makes it impossible to output effective monitoring data stably. In summary, existing monitoring technologies have significant shortcomings in terms of cost control, environmental adaptability, and multi-source data fusion. Summary of the Invention

[0004] The purpose of this invention is to provide a method for multi-scale quantitative assessment and dynamic early warning of slope rockfall risk, thereby solving the above-mentioned technical problems.

[0005] To achieve the above objectives, this invention provides a method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall, comprising the following steps: S1. For open slopes, synchronously collect displacement, temperature, and acoustic emission data of the entire process of rockfall incubation, instability, and movement to obtain multi-physics raw datasets and slope zoning maps. Then, preprocess and standardize the multi-physics raw datasets to obtain standardized feature sets. S2. Based on the standardized feature set obtained in step S1, use semi-supervised machine learning to establish a mapping relationship between multi-physics field features and rockfall incubation stage, and output the incubation stage and instability probability of slope rockfall. S3. The instability probability output in step S2 is used as the initial instability criterion of the discrete element method. Combined with slope DEM resolution adaptation and rock mass parameter calibration, the entire process of incubation-instability-movement is simulated to quantify the harm of falling rocks to sensitive targets in the accumulation area. S4. Integrate real-time data from multiple physics fields and the quantization results from step S3 to establish a multi-level early warning system and output early warning results.

[0006] Therefore, the beneficial effects of the above-mentioned multi-scale quantitative risk assessment and dynamic early warning method for slope rockfall in this invention are as follows: 1. Significantly improved assessment accuracy: It integrates multi-physical field characteristics such as displacement, microseismic activity, vegetation, and topography, and quantitatively outputs the gestation stage, instability probability, and rockfall kinetic energy through Siamese network and logistic regression model, avoiding the subjectivity and bias of traditional single-indicator or qualitative assessment. 2. Wide scene adaptability: Multi-resolution DEM (30m / 5m / 0.05m) is adapted to slopes at the regional, site, and point scales respectively, taking into account both macro-risk screening and micro-detailed calculation, and is suitable for different types of high and steep slope scenarios such as highways, mines, and scenic spots; 3. Timely and efficient early warning response: Real-time monitoring of crack propagation rate and instability probability changes, and output of early warning signal within seconds after triggering threshold, solving the pain points of high missed rate (30%-40%) and delayed response (5-10 minutes) of traditional manual inspection; 4. Strong decision support and practicality: It generates three-dimensional data of "stage-probability-kinetic energy" and dynamic risk heat map, which intuitively presents the spatiotemporal distribution of risks and provides clear scientific basis for the optimization of protection projects, adjustment of mining sequence and emergency evacuation; 5. Outstanding safety benefits: By identifying high-risk areas in advance and quantifying the exposure risks to personnel and equipment, the casualties and property losses caused by rockfalls can be significantly reduced, helping disaster prevention and control to transform from "passive response" to "proactive prevention".

[0007] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0008] Figure 1 This is a flowchart of the multi-scale quantitative risk assessment and dynamic early warning method for slope rockfall in this invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0010] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0011] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0012] like Figure 1 As shown, the method for multi-scale quantitative assessment and dynamic early warning of slope rockfall risk includes the following steps: S1. For open slopes, synchronously collect displacement, temperature, and acoustic emission data of the entire process of rockfall incubation, instability, and movement to obtain multi-physics raw datasets and slope zoning maps. Then, preprocess and standardize the multi-physics raw datasets to obtain standardized feature sets. Step S1 specifically includes the following steps: S11. Slope Zoning: Based on the slope gradient, the slope is divided into rockfall source area, movement area, and deposition area, resulting in a slope zoning map. The slope gradient of the rockfall source area is shown in the map. slope of the exercise area Accumulation area ; S12. Zonal data acquisition (deploy corresponding monitoring equipment and simultaneously collect multi-physics field data based on the different rockfall behavior characteristics and monitoring needs of rockfall source areas, movement areas, and accumulation areas). Data Acquisition in the Rockfall Source Area: The rockfall source area is characterized by exposed rock mass / sparse vegetation and a concentrated distribution of potentially unstable rock formations. Three sets of microseismometers, five sets of crack gauges, two sets of thermocouples, and one set of UAV monitoring points were deployed. The microseismometers, using a sampling rate of 250Hz, were arranged with an array aperture of 10m and a spacing of 5m to collect acoustic emission signals generated by rock fractures and record event time, amplitude, and frequency. The crack gauges were arranged along the identified crack direction in the source area, spaced 2-3m apart, to collect crack width data with millimeter-level accuracy. Thermocouples were buried 1-2m deep to collect internal slope temperatures, while surface temperature sensors were simultaneously deployed to collect surface temperature data with an accuracy of ±0.5℃. The UAV used was a DJI Phantom 4 Pro. V2.0 equipment acquires point cloud data at a resolution of 0.05m, collected once per quarter, and encrypted to monthly after a fire. NDVI vegetation data is collected simultaneously for subsequent assessment of the slope stabilization effect of vegetation. The collection cycle for rockfall source areas is: once per day during the incubation stage, once per hour during the instability precursor stage, and once per 10 seconds during the instability stage. Data acquisition in the movement zone: The movement zone is dominated by the bouncing and rolling of the blocks, and two sets of UAV monitoring points and three sets of thermocouples are deployed. The UAV parameters are the same as those of the UAV monitoring points deployed in the rockfall source area, and they cover the movement trajectory of the blocks. Point cloud data is collected to capture the precursor displacement of the blocks. The thermocouples are buried at a depth of 0.5-1m to collect temperature data to analyze the impact of slope thermal stress changes on the movement of the blocks. Data acquisition cycle in the movement zone: once a day during the incubation stage, once an hour during the instability precursor stage, and once every 30 seconds during the instability stage.

[0013] Data Acquisition in the Accumulation Zone: The accumulation zone is the area where the block finally accumulates and may contain sensitive targets such as highways and buildings. One set of InSAR ground control points, one set of crack gauges, and one set of environmental parameter sensors are deployed. The InSAR uses Sentinel-1 satellite data, with a retesting cycle of 15 days, combined with regional-scale displacement data collected from the ground control points. The crack gauges are deployed on the surface of the accumulation body to monitor crack changes caused by deformation. Data acquisition frequency for the accumulation zone: once per day during the incubation phase, once every 2 hours during the pre-instability stage, and once every minute during the instability stage. S13. Multi-physics data synchronization is achieved through UTC timestamps, with a time error of ≤1s. Then, PostgreSQL database is used to classify and store the data. Among them, microseismic data is stored in a structured manner according to event ID-time-amplitude-frequency-location, displacement and temperature data are stored with partition labels-monitoring time-coordinates, and NDVI vegetation data is bound to UAV data for storage, forming a multi-physics raw dataset with partition labels. S14. Preprocessing: Noise removal and feature standardization are performed on the original multiphysics dataset. Specifically, for displacement data: phase unwrapping-Gaussian filtering is used to eliminate atmospheric interference for InSAR data, Cloud Compare software is used to remove outliers caused by wind and rain for UAV point clouds, and 5-minute window moving average filtering is used for crack gauge data. Temperature data is filtered by window arithmetic moving average to eliminate short-term temperature fluctuations, and then the least squares method is used to fit the linear trend term. The final denoised data is obtained by residual calculation, eliminating the sudden temperature changes caused by diurnal temperature differences and rainfall. Microseismic data were filtered using a Butterworth third-order low-pass filter (cutoff frequency 40Hz), and valid events were selected using the following formula. : ; In the formula, Indicates the power of the microseismic signal; This represents the average power of ambient noise during periods without events. S15. Extract displacement change rate, temperature gradient, and microseismic event frequency features from the preprocessed multiphysics raw dataset, and standardize them to obtain a standardized feature set. , This represents the standardized value of the displacement; This represents the standardized rate of displacement change, temperature gradient, and frequency of microseismic events. This represents the NDVI normalized value.

[0014] In step S14, the denoising steps for the InSAR data are as follows: First, the coherence coefficient of the original interferogram is screened. Then, a piecewise linear atmospheric phase model is used to fit the atmospheric interference (based on a 15-day remeasurement period of Sentinel-1 data, assuming that the atmospheric phase has a spatially linear distribution over a short period of time). Finally, the phase is unwrapped using a minimum cost flow algorithm to eliminate phase ambiguity caused by atmospheric phase and obtain the true slope displacement. Finally, a two-dimensional Gaussian filter was used to smooth the unwrapped phase map; the actual slope displacement was then analyzed. The calculation formula is as follows: ; in, ; In the formula, This indicates the C-band wavelength of the Sentinel-1 satellite radar; Indicates the continuous true phase after untangling; This represents the minimum cost flow algorithm; Represents the original interference phase, and , Indicates the noise phase; Indicates atmospheric phase, and , Represents the geographic coordinate projection of the slope. , , All represent the coefficients of the atmospheric phase linear model; Indicates terrain phase; Final denoised temperature data The expression is as follows: ; in, ; ; ; ; In the formula, Represents the 1st moving average. One temperature data point; Indicates the first Linear trend term for each temperature data point; Indicates the size of the movable window; This represents the raw temperature data. This represents the index of the temperature data points after the moving average. This represents the index of temperature data points within a single window; and These represent the slope of the trend term and the intercept of the trend term, respectively. Indicates the first The absolute time corresponding to each moving average data point; This represents the total number of data points after the moving average.

[0015] S2. Based on the standardized feature set obtained in step S1, a mapping relationship between multi-physics field features and rockfall incubation stage is established using semi-supervised machine learning. The incubation stage and instability probability of rockfall on the slope are output, providing an initial instability criterion for DEM dynamic simulation and solving the problems of weak rockfall incubation signal and difficulty in stage division on the slope. Step S2 specifically includes the following steps: S21. Select samples with historical rockfall event labels from the standardized feature set, construct a labeled dataset, and divide it into a training set and a validation set; at the same time, select samples without historical labels from the standardized feature set as samples to be classified in semi-supervised learning. S22. Using labeled datasets as anchors and unlabeled datasets as samples, construct anchor-sample pairs to adapt to the similarity learning logic of the Siamese network. S23. Input the anchor point features and sample features into the Siamese network classification model for training, and output the real-time gestation stage. ; S24. Set the independent variable of the logistic regression instability probability model to the real-time gestation stage. And standardized feature set, with the instability probability as the dependent variable. Establish the following regression formula: ; In the formula, Denotes a constant term, and ; , , , and These represent the stage weight, microseismic weight, displacement weight, temperature weight, and vegetation weight, respectively. ; S25. Calculate the AUC value of the logistic regression instability probability model using the validation set, until it is greater than 0.85. Output the trained logistic regression instability probability model and its instability probability. .

[0016] S3. The instability probability output in step S2 is used as the initial instability criterion of the discrete element method. Combined with the slope DEM resolution adaptation and rock mass parameter calibration, the entire process of incubation-instability-movement is simulated to quantify the damage of falling rocks to sensitive targets in the accumulation area (such as highway guardrails, buildings, etc.). Step S3 specifically includes the following steps: S31, DEM resolution adaptation: not less than 10km 2 The slope is selected at a regional scale for macroscopic trajectory simulation; 1-10km 2 The slope is selected based on the site scale for simulating fracturing in the rockfall source area; less than 1km. 2 The slope is selected using a point scale for the calculation of block kinetic energy; S32. For different DEM resolutions, calibrate the normal restitution coefficient, rolling friction coefficient, and vegetation damping: ; ; ; In the formula, Indicates the normal recovery coefficient after calibration; This represents the reference value for the normal restoration coefficient, and ; This represents the calibrated rolling friction coefficient; This represents a reference value for the rolling friction coefficient, and ; Indicates the currently used DEM pixel size; This indicates the calibrated vegetation damping; S33, DEM simulation of the incubation-instability stage: when When the rock mass enters the incubation stage, a contact bond model based on the discrete element method is used to simulate the propagation of cracks within the rock mass, and the extent of crack propagation is obtained. and crack propagation rate ;exist or When triggered Upgrade until The system is judged to have entered the instability phase. At this point, the instability probability is... When the threshold is reached, it is determined that the bonding constraints between the blocks are completely released, the instantaneous state of the block detaching from the rockfall source area is simulated, and the initial velocity of the block is calculated. , Represents gravitational acceleration. Indicates the block height. Indicates the slope gradient of the rockfall source area; among which, and These represent the set threshold values ​​for crack propagation range and crack propagation rate, respectively. S34. DEM Simulation During Motion Phase: Based on Initial Motion Velocity of the Block The stnParabel model was used to simulate the free fall-bouncing-rolling trajectory of the block; and the kinetic energy of the block was calculated considering the fabrication damping. And maximum kinetic energy: ; ; In the formula, Indicates the volume of the block; Indicates the real-time speed of the block; Indicates the height of the block's bounce; Indicates the distance the block has moved; This represents the total distance the block traveled from the rockfall source area to the point where it came to a stop accumulating. S35, Comparison of maximum kinetic energy The maximum kinetic energy exceeding the tolerance threshold of the sensitive target. The trajectory of the corresponding block is marked as a high-risk area.

[0017] In step S33, the specific steps for simulating the propagation of internal cracks in the rock mass using a contact bond model based on the discrete element method during the incubation stage are as follows: First, based on the adapted slope DEM, a rock mass discrete element model is constructed, and particle and contact properties are defined to match the actual structure of the slope rock mass. Step 1: Particle Generation: Based on the DEM terrain contour, generate spherical particles, and ensure that the density of the spherical particles is... Same density as the bulk; Step 2: Defining Contact Relationships: By default, contact and bonding relationships are established between particles, and the normal contact stiffness is set. and tangential contact stiffness ; Step 3: Boundary Condition Setting: Set the bottom particles of the slope to be fixed (simulating foundation constraints), while the sides and top are open (adapting to open slope environments with no lateral constraints). Step 4, Initial stress loading: Applying self-weight stress Simulate the natural stress state of the slope rock mass: ; In the formula, Represents gravitational acceleration; Indicates the burial depth of the particles; Then, based on the physical properties of the rock mass and the instability probability during the incubation stage, particle and bonding parameters are assigned to achieve dynamic decay of bonding strength with the incubation stage: ; ; In the formula, and These represent the normal bond strength and tangential bond strength after calibration, respectively. and These represent the initial normal bond strength and tangential bond strength, respectively; Indicates the adhesion attenuation coefficient; Then, based on discrete element numerical iteration, the internal cohesion failure process of the rock mass is simulated to realize the dynamic evolution of cracks from initiation to propagation; among which, the normal failure rule is as follows: interparticle normal stress Bond fracture occurs, resulting in normal separation cracks; the tangential failure rule is as follows: interparticle tangential shear stress Bonding breaks, resulting in shear cracks; Finally, key features of crack propagation were extracted from the simulation results, including the crack propagation extent. (By connecting the contact points where bond failure occurs consecutively to form crack segments, and statistically analyzing the effective crack coverage area, the crack propagation range can be obtained.) and crack propagation rate The crack distribution map was obtained: ; In the formula, This indicates the longest effective crack length at the end of the simulation; This represents the longest microcrack length at the initial stage of the simulation; This represents the effective propagation time (the duration from the initiation of the first effective crack to the end of the simulation).

[0018] S4. Integrate real-time data from multiple physics fields and the quantization results from step S3 to establish a multi-level early warning system and output early warning results.

[0019] In step S4, the following multi-level early warning system is established: Level 1 warning trigger conditions: , , ; Level 2 warning trigger conditions: , ,and ; Level 3 warning trigger conditions: , ,and ; Level 4 warning trigger conditions: , ,and .

[0020] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for multi-scale quantitative assessment and dynamic early warning of slope rockfall risk, characterized by: Includes the following steps: S1. For open slopes, synchronously collect displacement, temperature, and acoustic emission data of the entire process of rockfall incubation, instability, and movement to obtain multi-physics raw datasets and slope zoning maps. Then, preprocess and standardize the multi-physics raw datasets to obtain standardized feature sets. S2. Based on the standardized feature set obtained in step S1, use semi-supervised machine learning to establish a mapping relationship between multi-physics field features and rockfall incubation stage, and output the incubation stage and instability probability of slope rockfall. S3. The instability probability output in step S2 is used as the initial instability criterion of the discrete element method. Combined with slope DEM resolution adaptation and rock mass parameter calibration, the entire process of incubation-instability-movement is simulated to quantify the harm of falling rocks to sensitive targets in the accumulation area. S4. Integrate real-time data from multiple physics fields and the quantization results from step S3 to establish a multi-level early warning system and output early warning results.

2. The method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall as described in claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Slope Zoning: Based on the slope gradient, the slope is divided into rockfall source area, movement area, and deposition area, resulting in a slope zoning map. The slope gradient of the rockfall source area is shown in the map. slope of the exercise area Accumulation area ; S12, Partition Data Acquisition; Data acquisition in the rockfall source area: 3 sets of microseismometers, 5 sets of crack gauges, 2 sets of thermocouples, and 1 set of UAV monitoring points were deployed. The microseismometers, using a sampling rate of 250Hz, were arranged with an array aperture of 10m and a spacing of 5m to collect acoustic emission signals generated by rock fractures and record event time, amplitude, and frequency. The crack gauges were arranged along the identified crack direction in the source area, with a spacing of 2-3m, to collect crack width data. Thermocouples were buried at a depth of 1-2m to collect internal slope temperature, and temperature sensors were simultaneously deployed on the surface to collect surface temperature. The UAV used a DJI Phantom 4 Pro V2.0 device to acquire 0.05m resolution point cloud data, collected quarterly, with the frequency increased to monthly after a fire, and NDVI vegetation data was collected simultaneously. The rockfall source area data collection cycle was: once a day during the incubation stage, once an hour during the pre-instability stage, and once every 10 seconds during the instability stage. Data acquisition in the motion zone: Deploy 2 sets of UAV monitoring points and 3 sets of thermocouples; the UAV parameters are the same as those of the UAV monitoring points deployed in the rockfall source area, and cover the movement trajectory of the block, collecting point cloud data to capture the precursor displacement of the block movement; the thermocouples are buried at a depth of 0.5-1m; the acquisition cycle in the motion zone is: once a day during the incubation stage, once an hour during the instability precursor stage, and once every 30 seconds during the instability stage; Data acquisition in the accumulation zone: Deployment of 1 set of InSAR ground control points, 1 set of crack gauges, and 1 set of environmental parameter sensors; Among them, InSAR uses Sentinel-1 satellite data, with a retesting cycle of 15 days, combined with regional-scale displacement data collected from ground control points; Crack gauges are deployed on the surface of the accumulation body to monitor crack changes caused by accumulation deformation; Acquisition cycle in the accumulation zone: once a day during the incubation stage, once every 2 hours during the instability precursor stage, and once every 1 minute during the instability stage; S13. Multi-physics data synchronization is achieved through UTC timestamps; then, data is stored in a PostgreSQL database for classification. Among them, microseismic data is stored in a structured manner according to event ID-time-amplitude-frequency-location; displacement and temperature data are stored with partition labels-monitoring time-coordinates; NDVI vegetation data and UAV data are bound and stored together to form a multi-physics raw dataset with partition labels. S14. Preprocessing: Noise removal and feature standardization are performed on the original multiphysics dataset. Specifically, for displacement data: phase unwrapping-Gaussian filtering is used to eliminate atmospheric interference for InSAR data, Cloud Compare software is used to remove outliers caused by wind and rain for UAV point clouds, and 5-minute window moving average filtering is used for crack gauge data. For temperature data, window arithmetic moving average is used to eliminate short-term temperature fluctuations, and then the least squares method is used to fit the linear trend term. The final denoised data is obtained by calculating the residuals. The microseismic data were filtered using a third-order Butterworth low-pass filter, and valid events were selected using the following formula. : ； In the formula, Indicates the power of the microseismic signal; This represents the average power of ambient noise during periods without events. S15. Extract displacement change rate, temperature gradient, and microseismic event frequency features from the preprocessed multiphysics raw dataset, and standardize them to obtain a standardized feature set. , This represents the standardized value of the displacement; This represents the standardized rate of displacement change, temperature gradient, and frequency of microseismic events. This represents the NDVI normalized value.

3. The method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall as described in claim 2, characterized in that: In step S14, the denoising steps for the InSAR data are as follows: First, the coherence coefficient of the original interferogram is screened; then, a piecewise linear atmospheric phase model is used to fit the atmospheric interference; finally, the phase is unwrapped using a minimum cost flow algorithm to eliminate phase ambiguity caused by atmospheric phase, thus obtaining the true slope displacement. Finally, a two-dimensional Gaussian filter was used to smooth the unwrapped phase map; the actual slope displacement was then analyzed. The calculation formula is as follows: ； in, ； In the formula, This indicates the C-band wavelength of the Sentinel-1 satellite radar; Indicates the continuous true phase after untangling; This represents the minimum cost flow algorithm; Represents the original interference phase, and , Indicates the noise phase; Indicates atmospheric phase, and , Represents the geographic coordinate projection of the slope. , , All represent the coefficients of the atmospheric phase linear model; Indicates terrain phase; Final denoised temperature data The expression is as follows: ； in, ； ； ； ； In the formula, Represents the 1st moving average. One temperature data point; Indicates the first Linear trend term for each temperature data point; Indicates the size of the movable window; This represents the raw temperature data. This represents the index of the temperature data points after the moving average. This represents the index of temperature data points within a single window; and These represent the slope of the trend term and the intercept of the trend term, respectively. Indicates the first The absolute time corresponding to each moving average data point; This represents the total number of data points after the moving average.

4. The method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall as described in claim 3, characterized in that: Step S2 specifically includes the following steps: S21. Select samples with historical rockfall event labels from the standardized feature set, construct a labeled dataset, and divide it into a training set and a validation set; at the same time, select samples without historical labels from the standardized feature set as samples to be classified in semi-supervised learning. S22. Using labeled datasets as anchors and unlabeled datasets as samples, construct anchor-sample pairs to adapt to the similarity learning logic of the Siamese network. S23. Input the anchor point features and sample features into the Siamese network classification model for training, and output the real-time gestation stage. ; S24. Set the independent variable of the logistic regression instability probability model to the real-time gestation stage. And standardized feature set, with the instability probability as the dependent variable. Establish the following regression formula: ； In the formula, Denotes a constant term, and ; , , , and These represent the stage weight, microseismic weight, displacement weight, temperature weight, and vegetation weight, respectively. ; S25. Calculate the AUC value of the logistic regression instability probability model using the validation set, until it is greater than 0.

85. Output the trained logistic regression instability probability model and its instability probability. .

5. The method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall as described in claim 4, characterized in that: Step S3 specifically includes the following steps: S31, DEM resolution adaptation: not less than 10km 2 The slope is selected at a regional scale for macroscopic trajectory simulation; 1-10km 2 The slope is selected based on the site scale for simulating fracturing in the rockfall source area; less than 1km. 2 The slope is selected using a point scale for the calculation of block kinetic energy; S32. For different DEM resolutions, calibrate the normal restitution coefficient, rolling friction coefficient, and vegetation damping: ； ； ； In the formula, Indicates the normal recovery coefficient after calibration; This represents the reference value for the normal restoration coefficient, and ; This represents the calibrated rolling friction coefficient; This represents a reference value for the rolling friction coefficient, and ; Indicates the currently used DEM pixel size; This indicates the calibrated vegetation damping; S33, DEM simulation of the incubation-instability stage: when When the rock mass enters the incubation stage, a contact bond model based on the discrete element method is used to simulate the propagation of cracks within the rock mass, and the extent of crack propagation is obtained. and crack propagation rate ;exist or When triggered Upgrade until The system is judged to have entered the instability phase. At this point, the instability probability is... When the threshold is reached, it is determined that the bonding constraints between the blocks are completely released, the instantaneous state of the block detaching from the rockfall source area is simulated, and the initial velocity of the block is calculated. , Represents gravitational acceleration. Indicates the block height. Indicates the slope gradient of the rockfall source area; among which, and These represent the set threshold values ​​for crack propagation range and crack propagation rate, respectively. S34. DEM Simulation During Motion Phase: Based on Initial Motion Velocity of the Block The stnParabel model was used to simulate the free fall-bouncing-rolling trajectory of the block; and the kinetic energy of the block was calculated considering the fabrication damping. And maximum kinetic energy: ； ； In the formula, Indicates the volume of the block; Indicates the real-time speed of the block; Indicates the height of the block's bounce; Indicates the distance the block has moved; This represents the total distance the block traveled from the rockfall source area to the point where it came to a stop accumulating. S35, Comparison of maximum kinetic energy The maximum kinetic energy exceeding the tolerance threshold of the sensitive target. The trajectory of the corresponding block is marked as a high-risk area.

6. The method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall as described in claim 5, characterized in that: In step S33, the specific steps for simulating the propagation of internal cracks in the rock mass using a contact bond model based on the discrete element method during the incubation stage are as follows: First, based on the adapted slope DEM, a rock mass discrete element model is constructed, and particle and contact properties are defined to match the actual structure of the slope rock mass. Step 1: Particle Generation: Based on the DEM terrain contour, generate spherical particles, and ensure that the density of the spherical particles is... Same density as the bulk; Step 2: Defining Contact Relationships: By default, contact and bonding relationships are established between particles, and the normal contact stiffness is set. and tangential contact stiffness ; Step 3: Boundary Condition Setting: Set the bottom particles of the slope to be fixed, while the sides and top are open; Step 4, Initial stress loading: Applying self-weight stress Simulate the natural stress state of the slope rock mass: ； In the formula, Represents gravitational acceleration; Indicates the burial depth of the particles; Then, based on the physical properties of the rock mass and the instability probability during the incubation stage, particle and bonding parameters are assigned to achieve dynamic decay of bonding strength with the incubation stage: ； ； In the formula, and These represent the normal bond strength and tangential bond strength after calibration, respectively. and These represent the initial normal bond strength and tangential bond strength, respectively; Indicates the adhesion attenuation coefficient; Then, based on discrete element numerical iteration, the internal cohesion failure process of the rock mass is simulated to realize the dynamic evolution of cracks from initiation to propagation; among which, the normal failure rule is as follows: interparticle normal stress Bond fracture occurs, resulting in normal separation cracks; the tangential failure rule is as follows: interparticle tangential shear stress The bond breaks, resulting in shear cracks; Finally, key features of crack propagation were extracted from the simulation results, including the crack propagation extent. and crack propagation rate The crack distribution map was obtained: ； In the formula, This indicates the longest effective crack length at the end of the simulation; This represents the longest microcrack length at the initial stage of the simulation; Indicates the effective extension time.

7. The method for multi-scale quantitative risk assessment and dynamic early warning of slope rockfall as described in claim 6, characterized in that: In step S4, the following multi-level early warning system is established: Level 1 warning trigger conditions: , , ; Level 2 warning trigger conditions: , ,and ; Level 3 warning trigger conditions: , ,and ; Level 4 warning trigger conditions: , ,and .

Images