A landslide disaster site safety risk assessment decision method and system

Abstract

This invention discloses a method and system for on-site safety risk assessment and decision-making in landslide disasters, belonging to the field of geological disaster risk assessment and emergency rescue technology. This invention identifies potential rockfall risk release points at landslide sites and quantifies their attributes and release probabilities through multi-source monitoring data fusion processing; it constructs a data-driven intelligent optimization system for rockfall simulation model parameters, enabling adaptive adjustment of simulation parameters; and it proposes a scenario-driven five-factor risk assessment system, comprehensively considering disaster hazard and the sensitivity of the affected body to complete the risk assessment of the work area, ultimately generating a visualized risk distribution map and enabling dynamic updates. This invention overcomes the shortcomings of traditional assessment methods, such as staticity, fixed parameters, and single indicators, achieving dynamic and high-precision assessment of landslide disaster risks, significantly improving the accuracy and real-time performance of risk assessment, and providing scientific and reliable decision support for landslide emergency rescue operations.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster analysis technology, specifically relating to a method for on-site safety risk assessment and decision-making in landslide disasters. Background Technology

[0002] Landslides, rockfalls, and other slope geological disasters occur frequently in mountainous areas, transportation hubs, water conservancy projects, and urban construction sites. They are characterized by their suddenness, long secondary disaster chains, and high destructive power, easily causing significant casualties and infrastructure damage. During emergency cleanup and rescue operations following landslides, the slope rock mass is typically in a metastable state. Influenced by external factors such as rainfall infiltration, seismic disturbances, and vibrations from machinery operations, it is highly susceptible to secondary collapses, rockfalls, and instability of accumulated material, directly threatening the lives and property of on-site rescue personnel and equipment. Currently, risk assessment and rescue decision-making technologies for landslide disaster sites are mainly divided into two categories: one is numerical simulation methods based on particle dynamics, with representative tools including RAMMS::Rockfall and RocFall, which calculate the motion path and impact characteristics of falling rocks by constructing rigid body dynamic equations; the other is risk zoning methods based on GIS platforms, which macroscopically classify the risk level of the work area through distance thresholds and empirical statistical models.

[0003] The aforementioned existing technologies have significant technical shortcomings in practical emergency rescue applications: First, the assessment methods are inherently static, unable to dynamically update risk prediction results based on real-time on-site monitoring data, and struggle to respond to dynamic changes in slope stability and environmental conditions, resulting in delayed risk warnings and insufficient timeliness of assessment results. Second, the numerical simulation parameters are fixed, with core calculation parameters often using industry-standard empirical values, failing to adapt to on-site geological and surface conditions, leading to insufficient simulation accuracy of rockfall trajectories and impact kinetic energy, and significant deviations between risk assessment results and actual on-site conditions. Third, the assessment indicators are singular, focusing only on the probability and impact characteristics of the disaster itself, ignoring the risk differences under different triggering scenarios and the differentiated attributes of the disaster-bearing objects, and failing to comprehensively consider the impact of personnel's evacuation capabilities and equipment's impact resistance on actual risks, thus failing to provide targeted and implementable safety decision support for emergency rescue operations. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention constructs a fully intelligent risk assessment and decision-making method that integrates multi-source data fusion, large-model-driven parameter optimization, multi-factor coupled scenario-based evaluation, and dynamic visualization output.

[0005] The technical solution adopted in this invention is as follows:

[0006] This invention provides a method for on-site safety risk assessment and decision-making in landslide disasters, comprising the following steps:

[0007] Step 100: First, acquire multi-source monitoring data and historical disaster data of the landslide disaster site. Then, use a large model to complete the fusion processing of the multi-source data, identify potential rockfall risk release points at the landslide site, quantify the geometric and physical properties of the risk release points and the probability of rockfall release, and generate a set of risk release points.

[0008] Step 200: Then, construct the parameter optimization system of the rockfall simulation model, perform sensitivity analysis on the key parameters of the rockfall simulation model based on the field measured data, screen the core optimization parameters, perform multi-objective optimization solution on the core optimization parameters, establish the mapping relationship between environmental features and preferred simulation parameters through the large model, and obtain the preferred simulation parameter set adapted to the field.

[0009] Step 300: Based on the set of risk release points and the preferred set of simulation parameters, perform rockfall dynamics simulation calculations to obtain the rockfall trajectory, motion parameters and spatial energy distribution data corresponding to the risk release points;

[0010] Step 400: Construct a scenario-driven five-factor risk assessment system. Based on the rockfall trajectory, motion parameters and spatial energy distribution data, combined with the attribute data of the disaster-bearing objects on site and the triggering scenario types such as rainfall and earthquakes, calculate the spatial risk distribution data of the work area, and use a large model to dynamically adjust the weights of the assessment factors according to the triggering scenario type.

[0011] Step 500: Based on the spatial risk distribution data, generate a visualized risk distribution map of the landslide disaster site, and dynamically update the risk distribution by combining real-time monitoring data.

[0012] Preferably, the specific steps of step 100 are as follows:

[0013] Step 101: Obtain multi-source high-resolution data of the landslide site through UAV photogrammetry, InSAR displacement monitoring, surface crack detection, and rainfall and groundwater monitoring equipment. The multi-source high-resolution data includes digital elevation model (DEM), geological structure data, environmental triggering factor data, and historical rockfall records.

[0014] Step 102: Preprocess and extract features from the multi-source high-resolution data using a geographic information system to identify potentially unstable rock or soil units and quantify the geometric and physical properties of each unit, including its three-dimensional location coordinates, volume, shape, and mass.

[0015] Step 103: Based on geological conditions, environmental driving factors and prior information on historical disasters, construct a rockfall release probability function, calculate the rockfall release probability of each potential instability unit, and output a risk release point set that includes geometric and physical properties and rockfall release probability.

[0016] Preferably, in step 200, the Sobol global sensitivity analysis method is used to quantify the sensitivity of the candidate parameters of the rockfall simulation model. The candidate parameters include the sliding friction coefficient, normal recovery coefficient, tangential recovery coefficient, surface roughness, and rolling resistance coefficient. Through first-order sensitivity calculation, the three parameters that contribute the most to the variation of the model output are selected as the core optimization parameters, and the remaining candidate parameters are fixed as empirical values.

[0017] Preferably, in step 200, a multi-objective optimization function integrating trajectory accuracy, velocity accuracy, and endpoint accuracy is constructed, and the NSGA-III algorithm is used to solve the core optimization parameters for multi-objective optimization. Among them, the trajectory error is calculated by using the Dynamic Time Warping (DTW) algorithm to calculate the matching error between the simulated trajectory and the measured trajectory, the velocity error is calculated by weighting the root mean square error after time series alignment with the peak velocity difference, and the endpoint error is obtained by normalizing the three-dimensional spatial distance between the simulated endpoint and the measured endpoint.

[0018] Preferably, in step 200, an environmental feature vector containing slope, roughness, humidity, vegetation coverage, lithology, and meteorological conditions is constructed, and an ensemble learning model is used to establish a mapping relationship between the environmental feature vector and the optimal parameters for rockfall simulation; through a weighted fusion algorithm of new and old parameters, the simulation parameters are adjusted in real time to adapt to changes in the environment.

[0019] Preferably, in step 400, the rockfall hazard is calculated based on the rockfall trajectory, motion parameters and spatial energy distribution data. The rockfall hazard is calculated by coupling three factors: rockfall probability, rockfall coverage and rockfall destructive force.

[0020] Among them, the rockfall coverage rate is calculated by probabilistic diffusion of the interpolated trajectory points using an anisotropic Gaussian kernel, and the rockfall destructive force is calculated by combining the translational kinetic energy and rotational kinetic energy of the rockfall, and by combining the adaptive energy kernel function of the surface conditions to calculate the spatial energy distribution.

[0021] Preferably, in step 400, calculating the risk avoidance capability factor for personnel within the work area specifically involves:

[0022] The effective evacuation speed of personnel is calculated by combining the on-site terrain and surface factors. The time required for personnel evacuation is calculated based on the evacuation path length and the effective evacuation speed. The probability of successful evacuation is calculated by combining the arrival time of falling rocks. Finally, the personnel risk index is obtained based on the evacuation failure rate and the personnel vulnerability coefficient.

[0023] Preferably, in step 400, calculating the resistance factor of the equipment within the work area specifically involves:

[0024] Based on the ratio of the destructive force of falling rocks to the equipment's impact resistance threshold energy, the probability of equipment damage is calculated using an exponential probability function; among which, the equipment's impact resistance threshold energy, damage rate parameter, and shape parameter can be set individually for different types of equipment.

[0025] Preferably, in step 500, based on spatial risk distribution data, a personnel risk assessment matrix and an equipment risk assessment matrix are constructed respectively, dividing the work area into low-risk, medium-risk, and high-risk zones, and generating corresponding safety measures recommendations for personnel operations and equipment use for different risk level zones; at the same time, based on real-time updated multi-source monitoring data, the risk release point set, simulation parameters, risk distribution data, and visualized risk distribution map are automatically refreshed, and the assessment weights are dynamically adjusted according to the trigger scenario type identified by real-time monitoring data, realizing dynamic updates of the entire risk assessment process.

[0026] The present invention also provides a landslide disaster site safety risk assessment and decision-making system, characterized in that it includes a processor and a memory, wherein the memory stores a computer-executable program, and the computer-executable program, when run by the processor, performs the steps of the landslide disaster site safety risk assessment and decision-making method described above.

[0027] The beneficial effects of this invention are as follows:

[0028] (1) This invention addresses the core defects of traditional landslide disaster on-site risk assessment technology, namely static nature, fixed parameters, and single nature. It constructs a full-process intelligent risk assessment decision-making method that integrates multi-source data fusion, large model-driven parameter optimization, multi-factor coupled scenario-based assessment, and dynamic visualization output. This achieves three core technology upgrades for landslide disaster risk assessment: from static assessment to dynamic assessment, from experience parameters to intelligent optimization, and from single indicators to multi-factor coupling. This significantly improves the accuracy, real-time performance, and on-site applicability of landslide on-site risk assessment.

[0029] (2) This invention realizes the fusion processing of multi-source heterogeneous monitoring data through a large model, and constructs an automated risk release point identification system. It can accurately identify potential rockfall source areas and quantify their geometric and physical properties and release probability, providing high-precision initial input for subsequent simulation calculation and risk assessment, and solving the problems of low efficiency and insufficient accuracy in identifying potential risk sources in traditional methods.

[0030] (3) This invention constructs a data-driven intelligent optimization system for rockfall simulation model parameters. It filters core optimization parameters through global sensitivity analysis, solves the optimal parameter set for on-site adaptation by combining multi-objective optimization algorithm, and establishes the mapping relationship between environmental features and optimal simulation parameters through large model and ensemble learning. This realizes the real-time adaptive adjustment of simulation parameters as the environment changes, solves the core problems of fixed parameters, poor on-site adaptability and insufficient simulation accuracy in traditional simulation methods, and provides a reliable numerical simulation basis for risk assessment.

[0031] (4) This invention constructs a three-dimensional visualization and full-process dynamic update mechanism for risk distribution, which can generate a high-resolution spatial risk distribution map and shorten the warning time to the minute level. At the same time, it can automatically refresh the full-process risk assessment results in combination with real-time monitoring data, effectively reducing the misjudgment and underreporting rate of risk assessment, and can provide accurate spatial risk guidance and scientific intelligent decision support for landslide clearing and emergency rescue operations.

[0032] (5) This invention realizes the dynamic coupling calculation of the arrival time of falling rocks and the evacuation time of personnel. Through the refined modeling of the effective evacuation speed, evacuation path and response time of personnel, it accurately quantifies the success rate of personnel avoidance at different spatial locations. Unlike the traditional risk assessment mode that only considers spatial coverage, it realizes the full-chain risk quantification of the spatiotemporal characteristics of disasters and the response capability of disaster-bearing bodies, which significantly improves the adaptability of risk assessment results to on-site emergency operations.

[0033] (6) This invention constructs a two-dimensional risk assessment matrix that differentiates personnel and equipment, establishes risk quantification models and grading judgment standards for the disaster-bearing characteristics of personnel and equipment respectively, and can directly output grading control measures that are adapted to different work objects, solving the problem of disconnect between traditional technical risk zoning and on-site operation control, and has strong engineering practicality. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the steps of the method in an embodiment of the present invention;

[0035] Figure 2 This is a structural diagram of the intelligent optimization method for Rockfall simulation model parameters in this embodiment of the invention;

[0036] Figure 3 This is a diagram showing the equipment risk assessment matrix composed of risk probability and equipment damage rate in the method of this embodiment of the invention;

[0037] Figure 4 This is a first schematic diagram of the trajectory and kinetic energy distribution of falling rocks on a riverbank in an embodiment of the present invention;

[0038] Figure 5 This is a second schematic diagram of the trajectory and kinetic energy distribution of falling rocks on a riverbank in an embodiment of the present invention;

[0039] Figure 6 This is a first schematic diagram of a rockfall risk zoning map for a certain riverbank in an embodiment of the present invention;

[0040] Figure 7 This is a second schematic diagram of a rockfall risk zoning map for a certain river beach in an embodiment of the present invention. Detailed Implementation

[0041] The present invention will be further explained below with reference to the figures and specific embodiments.

[0042] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below in conjunction with the diagrams (such as flowcharts, data flow diagrams, system architecture diagrams, etc.) in the embodiments of this application. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0043] Therefore, the following detailed description of the embodiments of this application provided in the figures is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0044] In the description of the embodiments of this application, the terms "step 100," "step 200," etc., are used only for the convenience of describing the phased characteristics and logical hierarchy of the technical solution, and do not indicate or imply that these steps must be executed strictly in the order shown. Unless it is explicitly stated that there are data dependencies, input-output relationships, or logical sequence conditions between the steps, some steps may be executed in parallel, in a different order, iteratively, or in a combined manner, and these variations in execution methods are all within the protection scope of this application.

[0045] Furthermore, the terms "first," "second," etc., are used only to distinguish different data sources, parameter sets, model instances, or processing stages, and should not be construed as indicating or implying relative importance. The terms "model," "module," and "unit" represent a logical division of functions, and can be software algorithm components, data processing sub-processes, or functional hardware units. Their implementation methods (such as local computing or cloud computing) should not constitute a limitation on this application.

[0046] In the description of this application, the terms "acquisition," "processing," "computation," and "prediction," etc., should be interpreted broadly and can refer to various data operations, algorithmic reasoning, or model calculations in a computer system. The terms "input," "output," and "feedback" indicate the flow of data between different stages of a method or between different components of a system; this can be real-time streaming or batch interaction via storage media. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0047] Example 1:

[0048] This embodiment discloses a safety risk assessment and decision-making method for landslide disaster sites. Addressing the real-time assessment and safety decision-making needs of secondary collapses and rockfall risks when slopes are in a metastable state during emergency landslide cleanup and rescue operations, this method overcomes the core technical shortcomings of existing risk assessment methods, such as static nature, fixed parameters, and single assessment indicators. It achieves dynamic, high-precision, and scenario-based full-process assessment of rockfall risks at landslide disaster sites, providing scientific and practical intelligent decision support for the safety of on-site rescue personnel and equipment deployment.

[0049] Reference Figure 1 The landslide disaster site safety risk assessment and decision-making method described in this embodiment specifically includes the following steps:

[0050] Step 100: First, acquire multi-source monitoring data and historical disaster data of the landslide disaster site. Then, use a large model to complete the fusion processing of the multi-source data, identify potential rockfall risk release points at the landslide site, quantify the geometric and physical properties of the risk release points and the probability of rockfall release, and generate a set of risk release points.

[0051] Specifically, in this step, real-time monitoring data and historical disaster data of the landslide site are obtained simultaneously through the deployment of multiple types of monitoring equipment on site and the retrieval of historical archives. The multi-source monitoring data includes slope topography data, displacement and deformation data, geological structure data, and environmental meteorological data. The historical disaster data includes records of historical rockfall events and disaster cause data in the landslide area and areas with similar geological conditions.

[0052] A multi-source data fusion model is used to complete the spatiotemporal registration, feature fusion, and noise filtering of heterogeneous multi-source data, solving the problems of dimensional differences and spatiotemporal scale mismatch between different data sources. Based on the fused multi-source data, rock blocks and soil units with instability risk within the slope area are identified as potential rockfall risk release points. The geometric and physical attributes such as the three-dimensional spatial location, volume, mass, and shape of each risk release point are quantified one by one. At the same time, combined with slope deformation characteristics, environmental driving factors, and historical disaster prior information, the rockfall release probability of each risk release point is calculated. Finally, the attribute data and release probabilities of all risk release points are integrated to output a standardized risk release point set, providing initial input for subsequent simulation calculations.

[0053] Step 200: Then, construct the parameter optimization system of the rockfall simulation model, perform sensitivity analysis on the key parameters of the rockfall simulation model based on the field measured data, screen the core optimization parameters, perform multi-objective optimization on the core optimization parameters, and learn the nonlinear mapping relationship between the field environment characteristics and the optimal simulation parameters through an integrated learning model to obtain the preferred simulation parameter set adapted to the field.

[0054] In this step, a full-process intelligent parameter optimization system is constructed for the key input parameters of the rockfall dynamics simulation model. First, based on the field-measured data such as the rockfall trajectory, endpoint position, and velocity, a global sensitivity analysis is performed on the candidate parameters affecting the simulation output results. The contribution of each candidate parameter to the variation of the simulation results is quantified, and the parameters with the most significant impact on simulation accuracy are selected as core optimization parameters. The remaining parameters are fixed as empirical values ​​adapted to the field geological conditions. Based on the selected core optimization parameters, a multi-objective optimization function is constructed with the goal of minimizing the error between the simulation results and the field-measured data. The optimal parameter solution adapted to the current field conditions is obtained through a multi-objective optimization algorithm. At the same time, the nonlinear mapping relationship between the field environmental characteristics and the optimal simulation parameters is learned through a large model, and an intelligent mapping model from environmental characteristics to the optimized simulation parameters is established to realize the adaptive adjustment of simulation parameters as the environment changes. Finally, the optimized simulation parameter set adapted to the field geology and surface environment is output.

[0055] Step 300: Based on the set of risk release points and the optimized set of simulation parameters, perform rockfall dynamics simulation calculations to obtain the rockfall trajectory, motion parameters and spatial energy distribution data corresponding to the risk release points.

[0056] In this step, the risk release point set generated in step 100 is used as the initial conditions for simulation, and the optimized simulation parameter set obtained in step 200 is used as the model input parameters. The rigid body dynamics rockfall simulation model is used to perform Monte Carlo stochastic simulation calculations. Multiple independent simulations are performed for each risk release point, and the three-dimensional motion trajectory, instantaneous velocity, acceleration, rotational angular velocity and other motion parameters of the rockfall corresponding to each risk release point are output, as well as the spatial energy distribution data such as kinetic energy distribution and impact energy during the rockfall process, providing a quantitative dynamic calculation basis for subsequent risk assessment.

[0057] Step 400: Construct a scenario-driven five-factor risk assessment system. Based on the rockfall trajectory, motion parameters, and spatial energy distribution data, combined with the attribute data of the disaster-bearing objects on site and the triggering scenario types such as rainfall and earthquakes, calculate the spatial risk distribution data of the work area, and dynamically adjust the weights of the assessment factors according to the triggering scenario type through a large model.

[0058] In this step, a scenario-driven risk assessment system is constructed, comprising five core factors: rockfall probability, rockfall coverage, rockfall destructive force, personnel evacuation capability, and equipment resistance capability. Specifically, based on rockfall release probability, trajectory coverage, and spatial impact energy, three disaster-end factors—rockfall probability, rockfall coverage, and rockfall destructive force—are calculated to obtain a comprehensive rockfall hazard index. Based on the evacuation conditions and response capabilities of on-site personnel, as well as the impact resistance performance of the operating equipment, two disaster-bearing end factors—personnel evacuation capability and equipment resistance capability—are calculated to obtain a comprehensive disaster-bearing body risk sensitivity index. Simultaneously, the triggering scenario types at the current landslide site are acquired, including rainfall scenarios, earthquake disturbance scenarios, and mechanical operation disturbance scenarios. Through a scenario-based risk reasoning engine, the weights of the five assessment factors are dynamically adjusted according to the disaster occurrence mechanism and risk characteristics of different triggering scenarios. Finally, combining the rockfall hazard and the disaster-bearing body risk sensitivity, the risk value of each spatial point within the operating area is calculated, generating spatial risk distribution data for the entire operating area.

[0059] Step 500: Based on the spatial risk distribution data, generate a visualized risk distribution map of the landslide disaster site, and dynamically update the risk distribution by combining real-time monitoring data.

[0060] In this step, based on the spatial risk distribution data obtained in step 400, and combined with the digital elevation model of the landslide site, a high-resolution three-dimensional visualized risk distribution map is generated. The risk level of different spatial locations within the work area is intuitively displayed through color gradients. At the same time, key information such as the location of risk release points, rockfall trajectories, and energy distribution fields are overlaid. A dynamic update mechanism is established to acquire real-time monitoring data at fixed time intervals and automatically repeat the calculation process from step 100 to step 400 to refresh the risk release point set, simulation parameters, spatial risk distribution data, and visualized risk distribution map. This enables real-time dynamic updates of the risk assessment results and ensures the timeliness of risk warnings during emergency operations.

[0061] Example 2:

[0062] This embodiment is a preferred implementation of the technical solution described in Embodiment 1, specifically targeting an emergency rescue operation scenario for landslide disasters along mountain trunk highways. In this scenario, the slope is a layered rock slope, and during the landslide clearing operation, the slope is in a metastable state, making it susceptible to secondary rockfalls due to factors such as rainfall and vibrations from operating machinery, directly threatening the safety of highway rescue personnel, equipment, and temporary traffic vehicles. This embodiment further improves the accuracy of risk assessment and its on-site applicability by optimizing and refining the technical details of each step. The specific steps are as follows:

[0063] Step 100: Acquire multi-source monitoring data and historical disaster data from the landslide disaster site. Use a large model to fuse the multi-source data, identify potential rockfall risk release points at the landslide site, quantify the geometric and physical properties of these risk release points and their rockfall release probability, and generate a risk release point set. This includes the following sub-steps:

[0064] Step 101 involves acquiring multi-source, high-resolution data of the landslide site using UAV photogrammetry, InSAR displacement monitoring, surface crack detection, and rainfall and groundwater monitoring equipment. Specifically, UAVs equipped with oblique photography cameras acquire slope images, generating a Digital Elevation Model (DEM) and a Digital Surface Model (DSM) with a spatial resolution better than 5 cm. Ground-based InSAR monitoring equipment acquires millimeter-level displacement data and displacement acceleration anomaly data across the entire slope area, with a monitoring frequency of once every 5 minutes. Before monitoring, terrain leveling and atmospheric phase correction are performed. During rainfall, a dual-channel differential algorithm is used to suppress atmospheric phase noise, ensuring monitoring accuracy better than ±1 mm, meeting the deformation identification requirements for the metastable state of the slope. Surface crack detection is performed using 3D laser scanning and ground-penetrating radar, acquiring data on crack distribution, width, and depth. Real-time rainfall intensity and groundwater infiltration depth data are acquired using rain gauges and pore water pressure sensors. Simultaneously, geological survey reports for the area are retrieved to obtain geological structure data such as joints, faults, and lithology, as well as historical rockfall records for the past 20 years for this road section.

[0065] Step 102: The multi-source high-resolution data is preprocessed and features are extracted using Geographic Information System (GIS) software. Preprocessing includes coordinate system one, spatiotemporal registration, noise filtering, and outlier removal. Based on the preprocessed data, the slope is divided into several independent rock / soil units using a slope unit discretization algorithm. Potentially unstable units with free faces and joint cuts are identified using a geometric feature extraction algorithm. The volume of each unit is estimated using a stereochemical method, with the volume range controlled between 0.01 and 10 m³. The shape of the unit is determined to be spherical, columnar, or sheet-like using a geometric fitting algorithm. The unit mass is calculated by combining lithological density parameters. Finally, the three-dimensional position coordinates (x, y, z), volume V, shape, and mass m of each potential unstable unit are quantified.

[0066] Step 103: Based on geological conditions, environmental driving factors, and prior information on historical disasters, construct the rockfall release probability function, as follows:

[0067]

[0068] Where Δu(x,y,z) represents the local displacement acceleration anomaly value obtained from InSAR monitoring, R(x,y,z) represents the intensity of rainfall and groundwater infiltration, G(x,y,z) is the geological structure discontinuity index, quantified from joint density and fault distribution data, and H represents historical disaster prior information; the rockfall release probability of each potential instability unit is calculated through this release probability function, and the final output risk release point set is as follows:

[0069] Let i = 1, 2, ..., N, where N is the total number of potential risk release points, and C i For the unit shape parameters.

[0070] Step 200: Construct a parameter optimization system for the rockfall simulation model. Based on field measurement data, perform sensitivity analysis on the key parameters of the rockfall simulation model, screen core optimization parameters, and perform multi-objective optimization on the core optimization parameters. Establish a mapping relationship between environmental features and preferred simulation parameters through a large model to obtain a set of preferred simulation parameters adapted to the field.

[0071] Reference Figure 2 In this step, the Sobol global sensitivity analysis method is used to quantify the sensitivity of candidate parameters of the RAMMS::Rockfall rockfall simulation model. These candidate parameters include the sliding friction coefficient μ and the normal restitution coefficient e. n Tangential coefficient of restitution e t Surface roughness ξ, rolling resistance coefficient μ rA low-discrepancy sampling matrix was generated using Sobol sequences, with the number of samples N=1024. Basic matrices A and B, as well as a parameter substitution matrix, were constructed. By using first-order sensitivity calculations, the contribution of individual parameters to the model output variation was quantified, and the three parameters with the highest contribution to the model output variation were finally selected: the sliding friction coefficient μ, the normal restitution coefficient e, and the sliding friction coefficient μ. n Tangential coefficient of restitution e t Surface roughness ξ is used as the core optimization parameter, and the remaining tangential restitution coefficients e t Rolling resistance coefficient μ r Fixed to the empirical value corresponding to the lithology.

[0072] Furthermore, a multi-objective optimization function integrating trajectory accuracy, velocity accuracy, and endpoint accuracy is constructed, as follows:

[0073]

[0074] Where w1, w2, and w3 are weighting coefficients, and w1 + w2 + w3 = 1; trajectory error E trajectory The Dynamic Time Warping (DTW) algorithm is used to calculate the matching error between the simulated trajectory and the measured trajectory. A distance matrix is ​​constructed to quantify the spatial distance between points on the simulated and measured trajectories. The overall matching error is obtained by solving for the optimal time alignment path. The velocity error E... velocity The root mean square error (RMS) after time series alignment is used for weighted calculation with the peak velocity difference. First, linear interpolation is performed to align the time nodes of the simulated and measured velocity time series. The RMS error of the time series is then calculated, and the relative error of the peak velocity is matched. The final velocity error is obtained by weighted averaging. The endpoint error E... endpoint The three-dimensional spatial distance between the simulated endpoint and the measured endpoint was used for calculation, and normalization was performed using 5% of the total slope length.

[0075] The NSGA-III algorithm combined with a Kriging surrogate model is used to solve the multi-objective optimization problem for the core optimization parameters. First, a Kriging surrogate model is constructed based on the Sobol sampling results to establish a surrogate mapping relationship between the core optimization parameters and the multi-objective optimization function, replacing the time-consuming rockfall simulation model for iterative optimization. An initial population is randomly generated within the physical constraints of each parameter, with a population size of 100. The objective function value is quickly calculated for the parameter set corresponding to each individual using the surrogate model. The goodness-of-fit R-value of the surrogate model is measured. 2 With a value of ≥0.95, the time required for a single optimization iteration is reduced by more than 85% compared to direct simulation, ensuring the timeliness requirement of minute-level updates.

[0076] Individuals are hierarchically sorted based on Pareto dominance. New individuals are generated through tournament selection, simulated binary crossover, and polynomial mutation operations. The maximum number of iterations is set to 200 generations. Optimization is terminated when the Pareto front remains stable for 30 consecutive generations. The compromise solution in the Pareto optimal solution set is output as the optimal value of the core optimization parameters.

[0077] Furthermore, an environmental feature vector is constructed, including slope, roughness, humidity, vegetation cover, lithology, and meteorological conditions, as follows:

[0078]

[0079] The mapping relationship between feature vectors and optimal parameters for rockfall simulation is established. The ensemble learning model uses random forest, support vector regression (SVR), gradient boosting tree (XGBoost), and backpropagation neural network as base learners, and a linear regression model as a meta-learner. The meta-learner fuses the outputs of each base learner to obtain the predicted optimal simulation parameters. Simultaneously, a weighted fusion algorithm for new and old parameters is used, as follows:

[0080]

[0081] The repetition factor, ranging from 0.2 to 0.5, Param old Param is the simulation parameter from the previous cycle. pred The parameter values ​​predicted by the ensemble learning model are adapted to dynamic changes in environmental conditions while maintaining parameter stability.

[0082] Step 300: Based on the set of risk release points and the optimized set of simulation parameters, perform rockfall dynamics simulation calculations to obtain the rockfall trajectory, motion parameters and spatial energy distribution data corresponding to the risk release points.

[0083] In this step, the three-dimensional coordinates, mass, and shape of each release point within the risk release point set are used as initial conditions, along with the optimized sliding friction coefficient μ and normal recovery coefficient e. n Surface roughness ξ is the core simulation parameter. Monte Carlo simulation is performed using the RAMMS::Rockfall model, and the number of simulations N is set for each risk release point. sim =1000 simulations, outputting the three-dimensional motion trajectory of the falling rock, instantaneous velocity v, acceleration a, rotational angular velocity ω, and translational kinetic energy, rotational kinetic energy and total kinetic energy of each point during the motion of the falling rock. Integrate all simulation results to obtain the set of falling rock motion trajectories, motion parameter time series and spatial energy distribution data of the entire operation area corresponding to each risk release point.

[0084] Step 400: Construct a scenario-driven five-factor risk assessment system. Based on the rockfall trajectory, motion parameters, and spatial energy distribution data, combined with the attribute data of the disaster-bearing objects on site and the triggering scenario types such as rainfall and earthquakes, calculate the spatial risk distribution data of the work area, and dynamically adjust the weights of the assessment factors according to the triggering scenario type through a large model.

[0085] First, based on the trajectory, motion parameters, and spatial energy distribution data of falling rocks, the hazard of falling rocks is calculated. The hazard is obtained by coupling three factors: falling rock probability, falling rock coverage, and falling rock destructive force. Specifically:

[0086]

[0087] Where D(⋅) is the consequence severity function, and M is the total number of risk release points. Let be the probability of a rockfall at the i-th release point. Let i be the rockfall coverage rate corresponding to the i-th release point. Let be the destructive force of the falling rocks corresponding to the i-th release point.

[0088] Among them, rockfall coverage rate The calculation process is as follows: First, linear interpolation is performed on the trajectory points of each simulated rockfall at fixed intervals to ensure uniform distribution of trajectory points and avoid underestimation of coverage due to sparse trajectory points in the high-speed movement segment. Then, an anisotropic Gaussian kernel function is used to perform probability diffusion on the interpolated trajectory points. The bandwidth of the kernel function is set to three times the bandwidth along the rockfall direction and perpendicular to the direction of movement, allowing for greater probability diffusion along the direction of movement and narrower diffusion perpendicular to the direction of movement. Finally, the probability is calculated using the formula:

[0089]

[0090] The rockfall coverage rate was calculated, where Let be the number of trajectory points in the k-th simulation. Let J be the coordinates of the j-th trajectory point in the k-th simulation. It is an anisotropic Gaussian kernel.

[0091] Destructive power of falling rocks Combining the translational and rotational kinetic energy of the falling rock, using the formula:

[0092]

[0093] Calculate the total kinetic energy of the falling rocks at each point, and then use the formula:

[0094]

[0095] Calculate the spatial energy distribution, where Let be the total kinetic energy at point j in the k-th simulation. The scene kinetic energy enhancement coefficient, It is an energy kernel function that adapts to surface conditions, and its bandwidth is positively correlated with surface roughness and vegetation cover.

[0096] Secondly, the risk avoidance capability factor of personnel within the work area is calculated, specifically: the effective evacuation speed of personnel is calculated by combining the site topography and surface factors as follows:

[0097]

[0098] Where v0 is the baseline walking speed of personnel under ideal flat conditions, with a value of 1.2 m / s, s is the local slope, V is the quantified value of surface roughness (0-1), and a and b are empirical constants with values ​​of 0.05 and 0.8, respectively; the time required for personnel evacuation is calculated based on the evacuation path length and effective evacuation speed as follows:

[0099]

[0100] Where L(x,y) is the shortest evacuation path length from spatial point (x,y) to the nearest safe zone, and G cong The evacuation efficiency factor, influenced by day / night cycle, weather, and population density, ranges from 1.0 to 2.0. (T) response The time for personnel to respond to the warning is 3-10 seconds; this is combined with the time T for the falling rock to reach the location. impact The probability of successful evacuation of personnel (x, y) is calculated using the sigmoid function as follows:

[0101]

[0102] Where τ is the time margin coefficient, with a value of 2s; finally, based on the evacuation failure rate and the personnel vulnerability coefficient, the personnel risk index is obtained as follows:

[0103]

[0104] in The vulnerability coefficient for personnel ranges from 0 to 1 and is determined by considering the age structure, health status, and operational experience of the on-site rescue personnel.

[0105] Simultaneously, the resistance factor of equipment within the work area is calculated, specifically: based on the ratio of the destructive force of falling rocks to the equipment's impact resistance threshold energy, the probability of equipment damage is calculated using an exponential probability function as follows:

[0106]

[0107] in λ is the impact threshold energy of the equipment, which is the critical energy at which the equipment begins to show significant damage. It is set separately for different types of equipment such as excavators, loaders, and dump trucks; λ is the damage rate parameter, which controls the probability of damage at the threshold energy, and takes a value of 1.0; β is the shape parameter, which describes the steepness of the change in the probability of damage with energy, and takes a value of 3.0.

[0108] Finally, the triggering scenario type at the current site is obtained, including normal operation scenario, rainfall scenario, earthquake aftershock scenario, and strong mechanical disturbance scenario. Based on the risk characteristics of different scenarios, the weights of five factors, namely rockfall probability, rockfall coverage, rockfall destructive force, personnel avoidance ability, and equipment resistance capability, are dynamically adjusted through a large model. For example, the weights of rockfall probability and rockfall destructive force are increased in the rainfall scenario, and the weights of rockfall coverage are increased in the earthquake scenario. Finally, the spatial risk distribution data of each grid cell in the operation area is calculated by combining the rockfall hazard, personnel risk index, and equipment damage probability. The grid cell size is set to 0.5m×0.5m.

[0109] Step 500: Based on the spatial risk distribution data, generate a visualized risk distribution map of the landslide disaster site, and dynamically update the risk distribution by combining real-time monitoring data.

[0110] In this step, refer to Figure 3 Based on spatial risk distribution data, personnel risk assessment matrices and equipment risk assessment matrices are constructed separately. Personnel risk index and rockfall hazard are divided into low, medium, and high intervals, forming a 3×3 personnel risk assessment matrix, which divides the work area into low-risk, medium-risk, and high-risk zones for personnel. Similarly, risk probability (rockfall probability × rockfall coverage) and equipment damage probability are divided into low, medium, and high intervals, forming a 3×3 equipment risk assessment matrix, which divides the work area into low-risk, medium-risk, and high-risk zones for equipment. Corresponding safety measures are generated for different risk levels: personnel are prohibited from entering high-risk personnel zones and remote operation is mandatory; work time is limited in medium-risk personnel zones, with enhanced real-time monitoring and personal protective equipment; normal work is permitted in low-risk personnel zones, with protective equipment worn according to regulations; heavy-duty protective equipment or unmanned operation equipment is used in high-risk equipment zones; standard equipment protection is implemented in medium-risk equipment zones, with real-time monitoring of equipment status; and standard operating procedures are followed in low-risk equipment zones.

[0111] Meanwhile, a minute-level dynamic update mechanism is established, acquiring real-time on-site monitoring data every 5 minutes, including InSAR displacement data, rainfall data, and groundwater data. The risk release point set, simulation parameters, spatial risk distribution data, and visualized risk distribution map are automatically refreshed. The assessment weights are dynamically adjusted based on the trigger scenario type identified by the real-time monitoring data, realizing dynamic updates throughout the entire risk assessment process. When a high-risk area is identified to expand or the risk level is upgraded, an early warning is automatically triggered.

[0112] Example 3:

[0113] This embodiment is a case study of the application of the method described in this invention in a specific engineering scenario, specifically an emergency cleanup operation for a landslide on the left bank of a river in a mountainous area of ​​Southwest China. The method corresponds to the present invention. Figures 4 to 7 The results show the visualization of rockfall trajectory and kinetic energy distribution, as well as risk zoning. This area is a mid-mountain canyon landform, with a steeply sloping layered limestone slope on the left bank of the river. The slope has a vertical height of 85m and a slope angle of 55°-70°. The toe of the slope is a 120m wide and 80m long river terrace, which is the main work area for clearing the landslide debris. Heavy rainfall during the 2024 flood season triggered a landslide on this slope, accumulating approximately 12,000 m³ of landslide debris in the river terrace area, blocking one-third of the river's flood discharge section, necessitating urgent clearing operations. During the clearing operation, the remaining unstable rock mass at the rear edge of the slope is in a metastable state, highly susceptible to secondary rockfalls due to rainfall and vibrations from machinery operations, directly threatening the safety of personnel and equipment in the river terrace work area. This embodiment uses the landslide disaster on-site safety risk assessment and decision-making method described in this invention to conduct on-site risk management. The specific implementation process and application effects are as follows:

[0114] 1. Obtain multi-source monitoring data and historical disaster data of the landslide disaster site, complete the fusion processing of multi-source data through a large model, identify potential rockfall risk release points on the slope, quantify the geometric and physical properties and rockfall release probability of each risk release point, and generate a risk release point set.

[0115] The specific implementation process is as follows:

[0116] (1) Using a drone equipped with a five-lens oblique photography camera, aerial surveys were conducted on the slope and river beach work area. The aerial ground resolution was 3cm, generating a digital elevation model (DEM) and a digital orthophoto (DOM) covering an area of ​​1.2km².

[0117] Ground-based InSAR equipment was used to monitor the slope displacement over the entire area, with a monitoring frequency of once every 2 minutes and a monitoring accuracy of ±0.5 mm, to obtain local displacement acceleration anomaly data of the slope; ground-penetrating radar and vibrating wire crack gauge were used to detect surface cracks and obtain data on crack distribution, development depth and propagation rate.

[0118] Real-time data collection of on-site rainfall intensity and groundwater depth was achieved using tipping bucket rain gauges and pore water pressure sensors. At the same time, 1:50,000 geological survey reports and historical rockfall disaster records for the area over the past 15 years were retrieved to complete multi-dimensional, multi-source data collection.

[0119] (2) The multi-source high-resolution data were preprocessed by ArcGIS geographic information system, including coordinate system I, spatiotemporal registration, noise filtering and outlier removal; the slope was divided into 25cm×25cm grid units by the slope unit discretization algorithm, and 47 potentially unstable rock block units with free face and two sets of joints were identified by the geometric feature extraction algorithm.

[0120] The volume of each unit was quantified using stereoscopic methods, ranging from 0.08 m³ to 7.2 m³, with 12 unstable rock masses having a volume greater than 1 m³. The shape of the rock blocks was determined to be mainly spherical and columnar using a geometric fitting algorithm. The mass of each rock block was calculated based on the natural density of limestone (2.7 g / cm³), ranging from 216 kg to 19440 kg. Finally, the geometric and physical properties of the three-dimensional position coordinates, volume, shape, and mass of each potential unstable unit were quantified.

[0121] Step 103: Based on the slope limestone geological conditions, rainfall infiltration driving factors, and prior information on historical disasters, construct the rockfall release probability function as shown in Example 2, where Δu(x,y,z) is the local displacement acceleration anomaly value obtained by InSAR monitoring, R(x,y,z) is the rainfall and groundwater infiltration intensity, G(x,y,z) is the geological structure discontinuity index quantified by joint density and fault distribution, and H is prior information on historical disasters;

[0122] This function calculates the rockfall release probability of 47 potentially unstable units, ranging from 0.03 to 0.82. Among them, there are 11 high-risk release points with a release probability greater than 0.5. The final output is a standardized set of risk release points. This provides high-precision initial input for subsequent simulation calculations.

[0123] 2. Construct a parameter optimization system for the rockfall simulation model. Based on the historical rockfall measurement data from the field, conduct sensitivity analysis on the key parameters of the RAMMS::Rockfall rockfall simulation model, screen the core optimization parameters, perform multi-objective optimization on the core optimization parameters, establish the mapping relationship between environmental features and the optimal simulation parameters through the large model, and obtain the optimal simulation parameter set adapted to the field.

[0124] The specific implementation process is as follows:

[0125] Twelve historical rockfall events with complete imagery and trajectory records were selected from the slope. Three-dimensional rockfall trajectories, endpoint locations, velocity, and slope environment data were collected as the measured dataset. The sliding friction coefficient μ and normal coefficient of restitution e were also selected. n Tangential coefficient of restitution e t Surface roughness ξ, rolling resistance coefficient μ r As candidate parameters, the Sobol global sensitivity analysis method was used to quantify sensitivity. A low-difference sampling matrix was generated using Sobol sequences, and the number of samples was set to N=1024. The first-order sensitivity index of each parameter was calculated.

[0126] The results show that the sliding friction coefficient μ has a first-order sensitivity index of 0.42, and the normal restitution coefficient e n The first-order sensitivity index is 0.28, and the first-order sensitivity index of surface roughness ξ is 0.17. The combined contribution of these three parameters reaches 87%, therefore they are selected as the core optimization parameters. The remaining tangential recovery coefficient e... t Rolling resistance coefficient μ r The value is fixed as an empirical value for limestone slopes, i.e., e. t =0.85, μ r =0.03.

[0127] Construct a multi-objective optimization function Fobj=0.4⋅E that integrates trajectory accuracy, velocity accuracy, and endpoint accuracy. trajectory +0.3⋅E velocity +0.3⋅E endpoint The trajectory error E trajectory The Dynamic Time Warping (DTW) algorithm is used to calculate the matching error between the simulated trajectory and the measured trajectory, specifically the velocity error E. velocity The endpoint error E is calculated by weighting the root mean square error after time series alignment with the peak velocity difference. endpoint The three-dimensional spatial distance between the simulated endpoint and the measured endpoint, normalized by 5% to the total slope length, was used to obtain the optimal solution. The NSGA-III algorithm was employed for multi-objective optimization of the core parameters. An initial population size of 100 and a maximum of 200 iterations were set. Optimization terminated when the Pareto front remained stable for 30 consecutive generations. The final optimal solution for the core parameters was obtained: sliding friction coefficient μ = 0.38, normal restitution coefficient e... n =0.42, surface roughness ξ=0.15.

[0128] Construct an environmental feature vector F that includes slope, roughness, soil moisture, vegetation cover, limestone lithology, and rainfall intensity. vector=[slope,roughness,moisture,vegetation,geology,weather], adopts an ensemble learning model with random forest, support vector regression, XGBoost, and BP neural network as base learners and linear regression as meta learner to establish the mapping relationship between environmental feature vectors and optimal simulation parameters. The model fit R2=0.94.

[0129] Param, a weighted fusion algorithm combining old and new parameters new (t)=(1−α)⋅Param old +α⋅Param pred The update weight α=0.3 is set to enable real-time adaptive adjustment of simulation parameters as rainfall and humidity change, and finally output the optimal set of simulation parameters that are suitable for the river beach-slope scenario.

[0130] 3. Based on the set of risk release points and the optimized simulation parameter set, perform rockfall dynamics simulation calculations to obtain the rockfall trajectory, motion parameters, and spatial energy distribution data corresponding to each risk release point, as per this embodiment. Figure 4 , Figure 5 The results of the rockfall trajectory and kinetic energy distribution.

[0131] The specific implementation process is as follows:

[0132] Using the three-dimensional coordinates, mass, and shape of 47 risk release points as initial simulation conditions, and the optimized set of preferred simulation parameters as model input parameters, a Monte Carlo stochastic simulation was performed using the RAMMS::Rockfall model. The simulation was performed N times for each risk release point. sim =1000 times, a total of 47000 simulation calculations have been completed.

[0133] Simulation results show that the maximum horizontal movement distance of the falling rocks is 118m, which can completely cover the river beach operation area; the maximum instantaneous velocity of the falling rocks on the slope is 32.6m / s, and the average velocity when they reach the river beach operation area is 18.2m / s; the total kinetic energy of the falling rocks reaching the river beach area ranges from 0.8kJ to 1280kJ, of which high-energy falling rocks with a value greater than 500kJ account for 12.3%, mainly concentrated in the area in the middle of the river beach directly opposite the slope.

[0134] Based on the simulation results, the system outputs the three-dimensional motion trajectory, instantaneous velocity, acceleration, rotational angular velocity, and other motion parameters of all falling rocks, as well as the spatial kinetic energy distribution data of the entire working area, generating data such as... Figure 4 , Figure 5 The diagram showing the trajectory and kinetic energy distribution of the falling rocks is shown. Figure 4 This is a planar distribution map of the rockfall trajectory. Figure 5 The resulting thermal distribution map of the kinetic energy in the rockfall space provides a quantitative kinetic basis for subsequent risk assessment.

[0135] 4. Construct a scenario-driven five-factor risk assessment system. Based on the rockfall trajectory, motion parameters and spatial energy distribution data, combined with the attribute data of the disaster-bearing objects in the river beach operation area and the rainfall trigger scenario type, calculate the spatial risk distribution data of the operation area, and use a large model to dynamically adjust the weight of the assessment factors according to the trigger scenario type.

[0136] The specific implementation process is as follows:

[0137] This implementation targets scenarios triggered by rainfall during the flood season. A risk assessment system was constructed, incorporating five core factors: rockfall probability, rockfall coverage, rockfall destructive force, personnel evacuation capability, and equipment resistance capability. The riverbank operation area was divided into 0.5m × 0.5m grid units for point-by-point calculation. First, the rockfall hazard was calculated based on the coupled calculation of three factors: rockfall probability, rockfall coverage, and rockfall destructive force. The formula is as follows:

[0138]

[0139] Before calculating the rockfall coverage, linear interpolation was performed on the simulated trajectory points at a fixed interval of 0.5m to avoid underestimating the coverage due to the sparseness of trajectory points in the high-speed movement segment. An anisotropic Gaussian kernel function was used to perform probability diffusion on the interpolated trajectory points, and the kernel function bandwidth was set to 3 times the vertical movement direction along the rockfall movement direction. Finally, the coverage was calculated using the formula in Example 2. The probability of falling rocks; the destructive force of falling rocks combines the translational and rotational kinetic energy of the falling rocks, and the total kinetic energy of the falling rocks is calculated using the formula in Example 2, combined with the kinetic energy enhancement coefficient M for the rainfall scenario. s =1.2, the surface condition adaptive energy kernel function, obtained through the formula in Example 2. The spatial energy distribution is calculated, where the bandwidth of the energy kernel function is positively correlated with surface roughness and vegetation cover.

[0140] Secondly, the risk sensitivity of the disaster-bearing body was calculated, and quantitative calculations were carried out on the personnel's evacuation capability and the equipment's resistance capability. Regarding personnel evacuation capability, a baseline walking speed of v0 = 1.2 m / s was set under ideal flat conditions, with empirical constants a = 0.05 and b = 0.8. Combined with the terrain and surface roughness of the riverbank work area, the effective evacuation speed range for personnel in the work area was calculated to be 0.6-1.1 m / s. The right bank embankment of the river was designated as the safe zone, and the shortest evacuation path length from each grid point to the safe zone was calculated using the Dijkstra algorithm, combined with the daytime operation evacuation efficiency factor G. cong =1.2, Personnel response time T response =5s, the evacuation time required is calculated using the formula in Example 2, ranging from 4.2s to 28.6s; combined with the time T for the falling rocks to reach each grid point. impactGiven (x, y), with a time margin τ = 2s, the probability of successful personnel evacuation is calculated using the sigmoid function, and finally combined with the vulnerability coefficient VUL of the personnel. pop =0.9, resulting in a personnel risk index range of 0.02-0.97 for each grid point.

[0141] Regarding equipment resistance, the core equipment for this operation consisted of excavators, loaders, and dump trucks, and impact resistance threshold energy was set for each: Excavator E thresh =800kJ, Loader E thresh =500kJ, dump truck E thresh =300kJ, with a uniform damage rate parameter λ=1.0 and shape parameter β=3.0, the probability of equipment damage at each grid point was calculated using the formula in Example 2, ranging from 0.01 to 0.99.

[0142] Finally, the weights of the assessment factors are dynamically adjusted using a large model for flood season rainfall scenarios, increasing the weight of rockfall probability and rockfall destructive force. Ultimately, rockfall hazard, personnel risk index, and equipment damage probability are coupled to generate spatial risk distribution data for the entire riverbank operation area.

[0143] 5. Based on the spatial risk distribution data, a visualized risk distribution map of the landslide disaster site is generated, and the risk distribution is dynamically updated by combining real-time monitoring data, corresponding to this embodiment. Figure 6 , Figure 7 The results of rockfall risk zoning.

[0144] The specific implementation process is as follows: Based on spatial risk distribution data, personnel risk assessment matrices and equipment risk assessment matrices are constructed separately. The risk of rockfall, personnel risk index, risk probability (rockfall probability × rockfall coverage), and equipment damage probability are all divided into three intervals: low (0-0.3), medium (0.3-0.7), and high (0.7-1.0). A 3×3 risk assessment matrix is ​​constructed, dividing the 120m×80m riverbank operation area into high, medium, and low risk level zones, corresponding to... Figure 6 , Figure 7 Risk partitioning visualization results:

[0145] High-risk area: Located in the core area directly opposite the riverbank slope, covering an area of ​​approximately 1860 square meters, accounting for 19.4% of the total work area. Figure 6 , Figure 7 The area marked in red indicates that the rockfall coverage is greater than 60%, the average kinetic energy of the falling rocks is greater than 400 kJ, the probability of successful personnel evacuation is less than 30%, and the probability of damage to dump trucks is greater than 85%.

[0146] Medium-risk area: Located in the transition zone between the high-risk area and the middle of the riverbank, covering an area of ​​approximately 3720 square meters, accounting for 38.7% of the total work area. Figure 6 , Figure 7 The area marked in yellow indicates that the rockfall coverage rate in this area is 20%-60%, the average kinetic energy of the falling rocks is 100-400kJ, the probability of successful personnel evacuation is 30%-70%, and the probability of equipment damage is 20%-85%.

[0147] Low-risk area: Located on the riverbank side away from the slope, covering an area of ​​approximately 4020 square meters, accounting for 41.9% of the total work area, and is an attached area. Figure 6 Appendix Figure 7 The area marked in green indicates that the rockfall coverage rate in this area is less than 20%, the average kinetic energy of the falling rocks is less than 100kJ, the probability of successful personnel evacuation is greater than 70%, and the probability of equipment damage is less than 20%.

[0148] Recommended safety measures for different risk levels: High-risk areas are off-limits to personnel, and only remotely operated excavators are used for clearing debris; medium-risk areas limit the number of workers and the duration of continuous operation, with manual slope inspections conducted every 30 minutes, and workers wearing personal protective equipment such as safety helmets and anti-fall vests throughout the process; low-risk areas are designated as areas for equipment parking, personnel rest, and material storage, and are operated according to standard safety operating procedures.

[0149] Simultaneously, a 5-minute dynamic update mechanism is established to access ground-based InSAR displacement data and rain gauge monitoring data in real time, automatically refreshing the risk release point set, simulation parameters, spatial risk distribution data, and visualized risk distribution map; when the hourly rainfall exceeds 10mm, the scene weight correction is automatically triggered, the risk zoning results are updated, and early warning information is pushed to the on-site command center.

[0150] The application results of this embodiment show that: using the method described in this invention, several high-risk rockfall release points on the slope were accurately identified, the simulated rockfall trajectory matched the actual rockfall endpoint on site with a high degree of accuracy, and the accuracy was significantly improved compared with the traditional empirical parameter simulation method; the risk zoning results were highly consistent with the actual rockfall impact range on site, and through graded risk management, no rockfall injuries or equipment damage occurred during the 22-day emergency cleanup operation, and the river clearing task was completed, verifying the engineering applicability and practical application effectiveness of the method of this invention in landslide scenarios.

[0151] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A landslide disaster site safety risk assessment decision method, characterized in that, Includes the following steps: Step 100: First, acquire multi-source monitoring data and historical disaster data of the landslide disaster site. Then, use a large model to complete the fusion processing of the multi-source data, identify potential rockfall risk release points at the landslide site, quantify the geometric and physical properties of the risk release points and the probability of rockfall release, and generate a set of risk release points. Step 200: Then, construct the parameter optimization system of the rockfall simulation model, perform sensitivity analysis on the key parameters of the rockfall simulation model based on the field measured data, screen the core optimization parameters, perform multi-objective optimization solution on the core optimization parameters, establish the mapping relationship between environmental features and simulation parameters through the large model, and obtain a set of simulation parameters adapted to the field. Step 300: Based on the risk release point set and simulation parameter set, perform rockfall dynamics simulation calculation to obtain the rockfall trajectory, motion parameters and spatial energy distribution data corresponding to the risk release point; Step 400: Construct a scenario-driven five-factor risk assessment system. The five-factor risk assessment system includes a scenario-driven risk assessment system with five core factors: rockfall probability, rockfall coverage, rockfall destructive force, personnel evacuation capability, and equipment resistance capability. Based on the rockfall trajectory, motion parameters, and spatial energy distribution data, combined with the attribute data of the disaster-bearing objects on site and the types of triggering scenarios such as rainfall and earthquakes, the spatial risk distribution data of the work area is calculated, and the weight of the evaluation factors is dynamically adjusted according to the triggering scenario type through a large model. Step 500: Based on the spatial risk distribution data, generate a visualized risk distribution map of the landslide disaster site, and combine it with real-time monitoring data to complete the dynamic update of the risk distribution; In step 400, the rockfall hazard is calculated based on the rockfall trajectory, motion parameters and spatial energy distribution data. The rockfall hazard is calculated by coupling three factors: rockfall probability, rockfall coverage and rockfall destructive force. Among them, the rockfall coverage rate is calculated by probabilistic diffusion of the interpolated trajectory points using an anisotropic Gaussian kernel, and the rockfall destructive force is calculated by combining the translational kinetic energy and rotational kinetic energy of the rockfall, and by combining the adaptive energy kernel function of the surface conditions to calculate the spatial energy distribution. In step 400, the risk avoidance capability factor of personnel within the work area is calculated, specifically as follows: The effective evacuation speed of personnel is calculated by combining the on-site terrain and surface factors. The time required for personnel evacuation is calculated based on the evacuation path length and the effective evacuation speed. The probability of successful evacuation is calculated by combining the arrival time of falling rocks. Finally, the personnel risk index is obtained based on the evacuation failure rate and the personnel vulnerability coefficient. In step 400, the resistance factor of the equipment within the work area is calculated, specifically as follows: Based on the ratio of the destructive force of falling rocks to the equipment's impact resistance threshold energy, the probability of equipment damage is calculated using an exponential probability function; among which, the equipment's impact resistance threshold energy, damage rate parameter, and shape parameter can be set individually for different types of equipment.

2. The landslide disaster site safety risk assessment decision method according to claim 1, characterized in that, The specific steps of step 100 are as follows: Step 101: Obtain multi-source high-resolution data of the landslide site through UAV photogrammetry, InSAR displacement monitoring, surface crack detection, and rainfall and groundwater monitoring equipment. The multi-source high-resolution data includes digital elevation model (DEM), geological structure data, environmental triggering factor data, and historical rockfall records. Step 102: Preprocess and extract features from the multi-source high-resolution data using a geographic information system to identify potentially unstable rock or soil units and quantify the geometric and physical properties of each unit, including its three-dimensional location coordinates, volume, shape, and mass. Step 103: Based on geological conditions, environmental driving factors and prior information on historical disasters, construct a rockfall release probability function, calculate the rockfall release probability of each potential instability unit, and output a risk release point set that includes geometric and physical properties and rockfall release probability.

3. The landslide disaster site safety risk assessment and decision-making method according to claim 1, characterized in that, In step 200, the Sobol global sensitivity analysis method is used to quantify the sensitivity of the candidate parameters of the rockfall simulation model. The candidate parameters include sliding friction coefficient, normal recovery coefficient, tangential recovery coefficient, surface roughness, and rolling resistance coefficient. Through first-order sensitivity calculation, the three parameters that contribute the most to the variation of the model output are selected as core optimization parameters, and the remaining candidate parameters are fixed as empirical values.

4. The landslide disaster site safety risk assessment and decision-making method according to claim 3, characterized in that, In step 200, a multi-objective optimization function is constructed that integrates trajectory error, velocity error, and endpoint error. The NSGA-III algorithm is used to solve the core optimization parameters for multi-objective optimization. Specifically, the trajectory error is calculated by using the Dynamic Time Warping (DTW) algorithm to calculate the matching error between the simulated trajectory and the measured trajectory. The velocity error is calculated by weighting the root mean square error after time series alignment with the peak velocity difference. The endpoint error is obtained by normalizing the three-dimensional spatial distance between the simulated endpoint and the measured endpoint.

5. The landslide disaster site safety risk assessment and decision-making method according to claim 1, characterized in that, In step 200, an environmental feature vector containing slope, roughness, humidity, vegetation coverage, lithology, and meteorological conditions is constructed. An ensemble learning model is used to establish a mapping relationship between the environmental feature vector and the optimal parameters for rockfall simulation. Through a weighted fusion algorithm of new and old parameters, the simulation parameters are adjusted in real time to adapt to changes in the environment.

6. The landslide disaster site safety risk assessment and decision-making method according to claim 1, characterized in that, In step 500, based on spatial risk distribution data, personnel risk assessment matrices and equipment risk assessment matrices are constructed respectively, dividing the work area into low-risk, medium-risk, and high-risk zones, and generating corresponding safety measures recommendations for personnel operations and equipment use for different risk level zones; at the same time, based on real-time updated multi-source monitoring data, the risk release point set, simulation parameters, risk distribution data, and visualized risk distribution map are automatically refreshed, and the assessment weights are dynamically adjusted according to the trigger scenario types identified by real-time monitoring data, realizing dynamic updates of the entire risk assessment process.

7. A landslide disaster on-site safety risk assessment and decision-making system based on a large model, characterized in that, It includes a processor and a memory, the memory storing a computer-executable program, which, when run by the processor, performs the steps of the landslide disaster site safety risk assessment and decision-making method as described in any one of claims 1 to 6.

Images