Method and system for predicting the scope of complex rock mass collapse disaster

Abstract

The application relates to a complex rock mass collapse disaster range prediction method and system. The method comprises the following steps: obtaining multi-dimensional monitoring data of a target rock mass to extract rock mass attribute features; performing geological modeling based on the rock mass attribute features to obtain a rock mass geological model; determining a collapse starting condition according to the rock mass geological model by using a machine learning model; combining the rock mass geological model and the collapse starting condition, simulating a rock mass collapse motion trajectory by using a kinematics-dynamics joint simulation method; and evaluating a collapse influence range based on the motion trajectory and generating a disaster range evaluation report. By using the method, the accuracy and reliability of complex rock mass collapse disaster range prediction can be significantly improved through multi-dimensional data fusion, machine learning accurate identification of starting conditions and dynamic simulation simulation.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster monitoring and early warning, and in particular relates to a method and system for predicting the range of complex rock mass collapse disasters. Background Technology

[0002] With the continuous development of geological disaster monitoring and early warning technologies, traditional rock mass stability assessment methods based on geological exploration and sensor monitoring have emerged. These technologies, through the deployment of displacement gauges, stress sensors, and other equipment, can initially identify surface deformation of the rock mass. However, their characteristic lies in their reliance on empirical models and local observations, making it difficult to comprehensively reflect the internal dynamics under complex geological structures (such as steep slopes with well-developed joints). In traditional techniques, when dealing with the risk of rock mass collapse, geological surveys combined with simplified mechanical analysis (such as the limit equilibrium method or numerical simulation) are typically used to predict the probability of collapse and its impact range. However, current or traditional methods have the following problems: they cannot accurately capture the characteristics of accumulated damage and sudden instability within the rock mass, resulting in insufficient real-time identification of critical state transitions and delayed early warnings; the prediction of the disaster range is affected by the complexity of the rock mass structure, the uncertainty of the collapse movement path, and the randomness of energy transfer, leading to large prediction errors, low accuracy of risk zoning, and difficulty in quantifying the movement trajectory and energy diffusion boundaries, thus affecting the accuracy of disaster prevention work. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for predicting the range of complex rock mass collapse disasters that can solve the above problems.

[0004] Firstly, this application provides a method for predicting the extent of complex rock mass collapse hazards, including:

[0005] Acquire multidimensional monitoring data of the target rock mass, and extract rock mass property characteristics based on the multidimensional monitoring data;

[0006] Based on the characteristics of rock mass properties, geological modeling is performed to obtain a geological model of the rock mass;

[0007] Based on the rock mass geological model, a machine learning model is used to determine the collapse initiation conditions of the target rock mass;

[0008] Based on the rock mass geological model and collapse initiation conditions, a kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement and obtain the rock mass collapse movement trajectory.

[0009] Based on the rock mass collapse trajectory, the impact range of the collapse is assessed, resulting in a rock mass collapse hazard range assessment report.

[0010] In one embodiment, the rock mass property characteristics include rock mass structural characteristics, rock mass mechanical characteristics, and rock mass geometric morphology characteristics;

[0011] Based on the characteristics of the rock mass properties, geological modeling is performed to obtain the rock mass geological model, including:

[0012] Based on the geometric morphological characteristics of the rock mass, a three-dimensional model was created to obtain the first model;

[0013] Based on the characteristics of the rock mass structure, a topological network of the rock mass structure is constructed in the first model to obtain the second model;

[0014] Based on the second model, the third model is obtained by clustering blocks according to the rock mass structure topology network;

[0015] Based on the rock mass mechanics characteristics, the mechanical parameters of each block in the third model are calibrated to obtain the rock mass geological model.

[0016] In one embodiment, the machine learning model includes a feature mapping unit, an attention interaction unit, and a decision output unit. The feature mapping unit includes a convolutional neural network with 3×3 convolutional kernels. The attention interaction unit adopts a channel attention mechanism based on the SE structure and includes a global average pooling layer, two fully connected layers with ReLU activation function, and an output layer with Sigmoid activation function. The decision output unit includes a fully connected layer that parameterizes the probability distribution for classifying the state of the block, where the block state categories include stable, critical, and unstable.

[0017] Based on the rock mass geological model, a machine learning model is used to determine the collapse initiation conditions of the target rock mass, including:

[0018] Based on the rock mass geological model, structural topological features, mechanical features and geometric morphological features associated with each block are extracted through feature mapping units to form a block association feature set;

[0019] Based on the block-related feature set, a global average pooling layer of the attention interaction unit is used for pooling to obtain the feature mean vector.

[0020] Based on the feature mean vector, a fully connected layer with attention interaction units is used to perform feature transformation to obtain the transformed feature vector;

[0021] Based on the transformed feature vector, channel weights are assigned to the output layer of the attention interaction unit to obtain a high-dimensional feature vector;

[0022] The high-dimensional feature vector is input into the decision output unit to determine the block state category;

[0023] For each block, the block state category is associated with the corresponding block association feature to generate a block state-feature correspondence;

[0024] Based on the correspondence between block state and feature, the triggering parameter combination conditions for rock mass collapse are determined by analyzing the synergistic relationship of the associated features between blocks, thus obtaining the collapse initiation conditions.

[0025] In one embodiment, based on a rock mass geological model and collapse initiation conditions, a kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement, obtaining the rock mass collapse trajectory, including:

[0026] Based on the structural topological, mechanical, and geometric characteristics of each block in the rock mass geological model, a set of kinematic-dynamic coupled equations is established. The set of equations includes translational equations, rotational equations, collision response equations, and energy dissipation equations, which are used to characterize the block's sliding / rolling mode switching, collision transient response, separation condition criterion, and energy dissipation process related to the friction-recovery coefficient.

[0027] Based on the collapse initiation conditions, determine the initial motion parameters of each block in the rock mass geological model;

[0028] Based on the initial motion parameters, under the constraints of the coupled equation system, an explicit time integration algorithm is adopted. The block satisfies the preset stability condition or exceeds the simulation boundary as the stopping condition. Iterative calculation is performed to obtain the three-dimensional coordinate sequence to generate the motion path of the block.

[0029] By summarizing the movement paths of each block, the trajectory of the rock mass collapse can be obtained.

[0030] In one embodiment, the initial motion parameters of each block in the rock mass geological model are determined based on the collapse initiation conditions, including:

[0031] Based on structural topological and geometric features, the initial spatial coordinate parameters of each block are determined;

[0032] Based on the initial spatial coordinate parameters and collapse initiation conditions, the initial velocity components of each block are calculated according to the translation equation.

[0033] Based on the structural topological characteristics, the initial angular velocity parameters of each block are calculated using rotational equations;

[0034] Based on the collision response equation, the initial contact state parameters between each block are generated;

[0035] The initial spatial coordinate parameters, initial velocity components, initial angular velocity parameters, and contact state parameters are vector-superimposed to construct the first block motion parameter matrix;

[0036] The dynamic consistency of the first block motion parameter matrix is ​​verified, and the second block motion parameter matrix is ​​obtained by eliminating abnormal parameter groups that do not conform to the law of conservation of energy.

[0037] Based on the preset confidence interval, the Monte Carlo method is used to fit the probability distribution of the motion parameter matrix of the second block to generate the initial motion parameters.

[0038] In one embodiment, the energy dissipation equation is expressed as follows:

[0039]

[0040] in, The total amount of energy dissipated. At the initial moment, The termination time, Based on the motion velocity of the i-th block geometric parameters of the j-th contact surface The dynamic friction coefficient, Let g be the mass of the i-th block, and g be the acceleration due to gravity. Let be the tangential velocity component of block i along the contact surface. Based on collision energy The collision restitution coefficient function, For contact stiffness parameters, Let be the change in the angular velocity of the block caused by the j-th contact surface.

[0041] In one embodiment, based on the rock mass collapse trajectory, an assessment of the collapse impact range is performed to obtain a rock mass collapse hazard range assessment report, including:

[0042] Based on the rock mass collapse trajectory, block trajectory range parameters are extracted. The block trajectory range includes the block movement path boundary, the maximum block diffusion distance, and the block collapse accumulation range.

[0043] The range weights of the block trajectory range parameters are determined according to the preset weighting rules.

[0044] Based on the range weight, the influence radius of the block trajectory range parameter is determined, and the range of rock mass collapse disaster is obtained;

[0045] Based on the preset report template, the block trajectory range parameters, range weights, and rock collapse hazard range are integrated to obtain a rock collapse hazard range assessment report.

[0046] Secondly, this application also provides a system for predicting the extent of complex rock mass collapse hazards, including:

[0047] The rock mass feature extraction module is used to acquire multidimensional monitoring data of the target rock mass and extract rock mass attribute features based on the multidimensional monitoring data.

[0048] The attribute feature modeling module is used to perform geological modeling based on the rock mass attribute features to obtain a rock mass geological model;

[0049] The collapse condition determination module is used to determine the collapse initiation conditions of the target rock mass based on the rock mass geological model and using a machine learning model.

[0050] The collapse motion simulation module is used to simulate rock mass collapse motion based on rock mass geological model and collapse initiation conditions, using kinematic-dynamic joint simulation method to obtain the rock mass collapse motion trajectory;

[0051] The impact range assessment module is used to assess the impact range of rock collapse based on the rock collapse movement trajectory, and to obtain a rock collapse hazard range assessment report.

[0052] The aforementioned method and system for predicting the extent of complex rock mass collapse disasters acquires multi-dimensional monitoring data of the target rock mass and extracts its property characteristics, enabling the capture of internal damage accumulation and sudden instability features, overcoming the shortcomings of traditional monitoring methods that rely on surface observation. Based on the rock mass property characteristics, geological modeling is performed to obtain a rock mass geological model, enhancing the characterization of complex geological structures (such as slopes with well-developed joints) and reducing errors caused by structural uncertainties. A machine learning model is used to determine the collapse initiation conditions, enabling real-time identification of the critical transition from steady state to instability, enhancing the timeliness and accuracy of early warnings. Based on the rock mass geological model and collapse initiation conditions, a kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement, obtaining the movement trajectory, quantifying the collapse path and energy diffusion boundary, and reducing prediction randomness. Based on the movement trajectory, the collapse impact range is assessed and a report is generated, delineating the disaster-affected area and improving the reliability of risk zoning. This method, through hierarchical design, integrates previously isolated monitoring, modeling, and simulation technologies into an organic whole, solving the problem of the fragmentation of "data-model-simulation-evaluation" in traditional methods. It provides a quantifiable and iterative integrated framework for predicting complex rock mass hazards. This composite strategy improves the timeliness and reliability of early warnings, meeting the needs of the geological hazard monitoring field for precision and systematization. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart of the method for predicting the range of complex rock mass collapse disasters according to the present invention;

[0055] Figure 2 This is a structural diagram of the complex rock mass collapse disaster range prediction system of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] In one embodiment, such as Figure 1 As shown, a method for predicting the extent of complex rock mass collapse hazards is provided. This embodiment illustrates the application of this method to a hazard extent prediction terminal. It is understood that this method can also be applied to a hazard extent prediction server, and further to a system including both a hazard extent prediction terminal and a hazard extent prediction server, and is implemented through the interaction between the two. In the implementation environment, the hardware architecture of this application mainly includes terminal equipment (such as a slope monitoring station or mobile data acquisition terminal), a server, and a network communication module (such as a 5G or IoT device). The hardware can be connected via wired or wireless networks.

[0058] The application scenarios of this application include: when there is a need to predict the disaster range of complex rock mass collapse hazard areas (such as mountainous scenic areas, coastal rock slopes, mountain transportation routes, mine rock slopes, and water conservancy project reservoir rock slopes), the terminal device acquires multi-dimensional monitoring data in real time and transmits it to the server via the network; the server performs geological modeling, machine learning model analysis, and kinematic-dynamic joint simulation based on the received data to calculate the collapse initiation conditions and movement trajectory; the server feeds back the generated disaster range assessment report to the terminal device, realizing data interaction and collaborative decision-making between the terminal device and the server, meeting the real-time and accuracy requirements of geological disaster early warning, and improving the efficiency of disaster prevention and mitigation.

[0059] In this embodiment, the method includes the following steps:

[0060] S01: Obtain multidimensional monitoring data of the target rock mass, and extract rock mass property characteristics based on the multidimensional monitoring data.

[0061] Optionally, the multidimensional monitoring data (spatiotemporal and / or physical dimensions) refers to multi-source information reflecting rock mass dynamics acquired by the disaster extent prediction terminal through integrated monitoring methods (such as ground sensor networks, remote sensing imaging, borehole exploration, or IoT devices). This includes displacement (using GNSS displacement monitoring stations with sampling frequencies of 1Hz to 100Hz), stress (using fiber optic strain sensors with a range of 0-50 MPa), and acoustic emission (frequency range 1-100 Hz). Data from ground-penetrating radar or laser scanning (point density ≥ 100 points / square meter) and UAV aerial surveys (ground resolution ≤ 5 cm) are used to capture the characteristics of rock mass surface deformation, internal stress evolution, fracture propagation, and damage accumulation; exploration and test data are obtained through on-site core drilling, indoor rock mechanics tests, and on-site in-situ rock mass tests; rock mass property characteristics are quantitative parameters extracted from the raw data by the disaster range prediction terminal after processing, which may include rock mass structural characteristics (such as joint density, fracture topology, etc., which can be obtained by using a region growth algorithm based on 3D scan point cloud data for fracture identification), rock mass mechanical characteristics (such as compressive strength, friction coefficient, etc., rock mass mechanical characteristics calibrated through laboratory tests), and rock mass geometric morphology characteristics (such as slope curvature, slope height, slope angle, block spatial dimensions, block centroid coordinates, free face morphology, etc., which can be calculated by establishing a rock mass digital elevation model based on 3D scan point cloud data). The disaster range prediction terminal can perform data preprocessing on multidimensional monitoring data (such as filtering, denoising, and coordinate normalization), and use feature extraction algorithms (such as principal component analysis, wavelet transform, or convolutional neural networks) to reduce the dimensionality of the preprocessed data and mine key information, converting multidimensional monitoring data signals into rock mass attribute features that can be used for geological modeling.

[0062] S02. Based on the rock mass property characteristics, geological modeling is performed to obtain the rock mass geological model.

[0063] Optionally, geological modeling is the process by which the disaster range prediction terminal transforms the rock mass property characteristics into a high-precision digital model of the rock mass using computer-aided technology. This may include modeling steps such as three-dimensional geometric reconstruction, structural network construction, block division, and mechanical parameter assignment. The rock mass geological model is a digital entity that integrates the rock mass's geometric morphology, structural topology, and mechanical properties, and can be used for subsequent collapse dynamics analysis. In practice, the disaster range prediction terminal can complete geological modeling in various ways, such as generating an initial geometric model based on the geometric features of the rock mass property characteristics, constructing a rock mass joint network and dividing it into discrete blocks according to structural features, and calibrating the material parameters (such as elastic modulus and density) of each block according to mechanical features, forming a geological model that can simulate the dynamic response of the rock mass, providing a reliable basis for determining the collapse initiation conditions.

[0064] S03, based on the rock mass geological model, uses a machine learning model to determine the collapse initiation conditions of the target rock mass.

[0065] Optionally, the machine learning model is a computational algorithm used by the disaster range prediction terminal to learn collapse patterns from rock mass characteristics through a data-driven approach. This can be a neural network, decision tree, or ensemble learning model. The machine learning model is used to adaptively extract the nonlinear mapping between features and states. The collapse initiation condition is a combination of triggering parameters that cause the rock mass to transition from a steady state to instability. These parameters may include critical displacement (trained based on data from historical collapse cases), stress threshold (calibrated through laboratory experiments based on mechanical characteristics in the rock mass geological model and data from historical collapse cases), or block-to-block coordinated motion parameters (analyzing the coordinated relationships of associated features among blocks (correlation of parameters such as displacement and stress in time and space) to determine the parameter combination that triggers collapse. For example, when multiple blocks are simultaneously in a "critical" state and their movement directions or stress distributions show high coordination, the coordinated parameters are deemed effective). These parameters are used to quantify the instability risk. In practice, the disaster range prediction terminal can extract block-level or multi-scale feature sets based on the rock mass geological model as model input; the disaster range prediction terminal can perform feature analysis and pattern recognition through machine learning models, output the block state probability (such as stable, critical or unstable) or directly predict the triggering parameters, and determine the collapse triggering conditions by analyzing the correlation between the model output and the features.

[0066] S04. Based on the rock mass geological model and collapse initiation conditions, the kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement and obtain the rock mass collapse movement trajectory.

[0067] Optionally, the kinematic-dynamic co-simulation method is a numerical simulation technique that combines kinematic equations (such as translational and rotational equations) and dynamic equations (such as collision response and energy dissipation equations) to quantify the switching of sliding / rolling modes, transient response to collisions, and energy diffusion processes in block motion. The disaster range prediction terminal can establish a set of coupled kinematic-dynamic equations based on the structural topological, mechanical, and geometric characteristics of each block in the rock mass geological model. The terminal can determine the initial motion parameters (including spatial coordinates, velocity components, angular velocity, and contact state) of each block according to the collapse initiation conditions, and iteratively calculate using explicit time integration algorithms (such as the central difference method or the Verlet algorithm), stopping when the block reaches a stable condition or exceeds the boundary, generating a three-dimensional coordinate sequence for each block to form a motion path. The terminal can summarize all block paths to obtain the overall rock mass collapse trajectory, which can be used to characterize the spatiotemporal evolution of the collapse path.

[0068] S05. Based on the rock mass collapse trajectory, an assessment of the collapse impact range is conducted to obtain a rock mass collapse hazard range assessment report.

[0069] Optionally, the landslide impact range assessment is a process of quantifying the disaster diffusion boundary based on motion trajectory parameters. The aim is to comprehensively consider multi-dimensional parameters such as the block movement path boundary, maximum diffusion distance, and accumulation range, and delineate the disaster area through weight allocation and impact radius calculation. The rock mass landslide disaster range assessment report is an output document obtained by the disaster range prediction terminal by integrating trajectory parameters, weight coefficients, and disaster range parameters, used for risk zoning and disaster prevention decision-making. The disaster range prediction terminal can extract block-level parameters (such as path envelope and termination point distribution) from the motion trajectory, calculate the impact weight of each parameter using weighting rules (such as weighting methods based on block movement energy or block volume), and determine the impact radius of the disaster range through geometric interpolation or probability density estimation based on the impact weights. An assessment report is generated based on a preset template (such as a standardized format including parameter tables, spatial distribution maps, and risk levels).

[0070] In one embodiment, the rock mass property characteristics include rock mass structural characteristics, rock mass mechanical characteristics, and rock mass geometric morphology characteristics;

[0071] Based on the characteristics of the rock mass properties, geological modeling is performed to obtain the rock mass geological model, including:

[0072] S11, based on the geometric morphology characteristics of the rock mass, a three-dimensional model was created to obtain the first model.

[0073] Optionally, when performing geological modeling, the disaster range prediction terminal can utilize the geometric features of the rock mass and perform three-dimensional reconstruction using modeling software such as Geomagic Studio or AutoCAD to restore the true geometric shape and free face features of the rock mass, thus obtaining a first model that only contains the geometric shape (such as a rock mass digital elevation model constructed using the Delaunay triangulation algorithm).

[0074] S12. Based on the characteristics of the rock mass structure, a topological network of the rock mass structure is constructed in the first model to obtain the second model.

[0075] Optionally, the disaster extent prediction terminal, based on data from rock mass structural characteristics such as joint attitude, joint density, fracture connectivity, and microfracture development, employs a Discrete Fracture Network (DFN) model. In the first model, macroscopic joint surfaces and a random microfracture network are simultaneously constructed to form a second model that integrates geometric morphology and structural information. (Where, macroscopic joint surfaces are defined based on measured joint traces, including attitude, trace length, and connectivity; joint connectivity is determined by the following criteria: joint surface spacing ≤ 0.05m, dip angle difference ≤ 5°, and strike difference ≤ 1m.) 0°, ensuring consistency between the joint network and the field survey results; random microfracture network: a Monte Carlo random simulation method can be used to generate a microfracture network that conforms to the statistical laws of the field based on the measured microfracture development density and orientation distribution, thus restoring the heterogeneous and discontinuous characteristics of the rock mass; based on the constructed macro-micro fracture network, a rock mass structure topology network is constructed, with blocks as nodes and joint surfaces as edges, defining the adjacency, contact, and connectivity relationships between nodes, forming a second model that integrates the geometric morphology and complete structural characteristics of the rock mass.

[0076] S13, based on the second model, performs block clustering based on the rock mass structure topology network to obtain the third model.

[0077] Optionally, the disaster range prediction terminal can use a graph clustering algorithm to discretize the rock mass into independent block units using joint surfaces and fracture surfaces as cutting boundaries. At the same time, it can perform class aggregation and merging based on the topological adjacency relationship, spatial location, and size characteristics between blocks to avoid over-segmentation. After the block division is completed, a unique identifier, spatial adjacency relationship, and contact type (surface contact / line contact / point contact) for each block are defined to obtain the third model of discrete blockization.

[0078] S14. Based on the rock mass mechanical characteristics, the mechanical parameters of each block in the third model are calibrated to obtain the rock mass geological model.

[0079] Optionally, the disaster range prediction terminal can combine laboratory test data related to rock mass mechanical characteristics to calibrate mechanical parameters for each block unit in the third model (including block unit mechanical parameters: such as calibrating the rock mass elastic modulus, Poisson's ratio, density, uniaxial compressive strength, tensile strength, and fracture toughness after Hoek-Brown criterion scale effect reduction; joint surface mechanical parameters: such as calibrating the joint surface normal stiffness, tangential stiffness, internal friction angle, cohesion, and residual strength parameters, with parameter values ​​based on the results of field joint direct shear tests; at the same time, it can also set deterioration coefficients for mechanical parameters: such as taking a deterioration coefficient of 0.6~0.8 for the cohesion of the rock mass under saturated state and a deterioration coefficient of 0.8~0.9), to achieve coupling and matching between parameters and the model, forming a rock mass geological model integrating geometric morphology, structural topology, and mechanical properties. After the rock mass geological model is constructed, the disaster range prediction terminal can perform the following three levels of verification to ensure the reliability of the model: geometric verification, the RMSE between the model and the measured terrain is ≤5%, and the block size and free face morphology are consistent with the site; structural verification, the relative error between the joint density, attitude distribution, connectivity, etc. of the joint network and the field survey results is ≤10%; mechanical verification, through random uniaxial compression numerical simulation of rock mass, the relative error between the numerical simulation results and the field rock mass strength test results is ≤15%, verifying the rationality of the mechanical parameter calibration.

[0080] In one embodiment, the machine learning model includes a feature mapping unit, an attention interaction unit, and a decision output unit. The feature mapping unit includes a convolutional neural network with 3×3 convolutional kernels. The attention interaction unit adopts a channel attention mechanism based on the SE structure and includes a global average pooling layer, two fully connected layers with ReLU activation function, and an output layer with Sigmoid activation function. The decision output unit includes a fully connected layer that parameterizes the probability distribution for classifying the state of the block. The block state categories include stable, critical, and unstable.

[0081] Optionally, the machine learning model can be a physically constrained graph neural network model, including a feature mapping unit, an attention interaction unit, a physical constraint unit, and a decision output unit. The feature mapping unit uses a 3×3 convolutional neural network with convolutional kernels for local feature perception, effectively capturing the structural topological, mechanical, and geometric features of the blocks. The convolutional stride and padding parameters of the kernels can be adaptively adjusted according to the input data dimension to ensure the completeness of feature extraction. The attention interaction unit can employ a channel attention mechanism based on an SE (Squeeze-and-Excitation) structure. The second fully connected layer (also using the ReLU activation function) further refines feature correlations and outputs a transformed feature vector. The ReLU function ensures the sparsity and efficiency of model training, adaptively strengthening the contribution of key feature channels to calibrate feature importance and improve the machine learning model's sensitivity to critical states. The decision output unit can map features to a state probability space through a fully connected layer. Its probability distribution parameterization can use the Softmax function or a similar method, outputting the probability value of each block belonging to one of four states (e.g., stable, damage evolution, critical instability, and instability). The classification threshold is set based on rock mass grading standards and / or historical data. When a certain state P has the highest probability and is greater than the corresponding classification threshold, it can be determined as the corresponding state.

[0082] The training dataset for the machine learning model can consist of three types of data, with a total sample size of ≥2000 sets and a training set / validation set / test set ratio of 7:2:1. These include: a historical collapse case dataset: ≥300 complex rock mass collapse cases from around the world with complete monitoring data, geological survey data, and records of the entire collapse process, which are then standardized to form samples; a physical model test dataset: ≥700 sets of data on the entire process of rock mass instability under different lithologies, joint conditions, and triggering conditions obtained through indoor rock mass collapse similar model tests; and a numerical simulation dataset: ≥1000 sets of rock mass instability evolution samples under different geological conditions and triggering conditions generated through discrete element numerical simulation, covering various complex scenarios such as hard rock, soft rock, jointed rocks, and high ground stress. Model training can use the Adam optimizer with an initial learning rate of 0.001 and a learning rate decay coefficient of 0.95, decaying once every 10 rounds, and early stopping (patience value = 15 rounds) to prevent overfitting; the loss function consists of two parts: classification loss and physical constraint loss. The physical constraint loss is based on the residual calculation of the rock mechanics constitutive equation, so that the model output conforms to the laws of mechanics and physics, rather than pure data correlation fitting.

[0083] For unique rock mass parameters (rare or anomalous geological conditions, such as unusual joint patterns or extreme mechanical properties), this machine learning model incorporates a coping mechanism through its design and training methods: the attention interaction unit adopts an SE structure, which can adaptively allocate channel weights, strengthen important features, and suppress noise. For rare parameters, this mechanism refines feature correlations through global average pooling and fully connected layers, reducing the interference of rare parameters on model decisions and improving fault tolerance to anomalous conditions. Optionally, in practice, the dataset can be expanded through synthetic data or transfer learning, or multidimensional data fusion can be used to integrate rare parameters into the overall feature set, reducing their isolated impact. For rare parameters, the probability value can reflect uncertainty, and users can combine it with confidence intervals for risk assessment. For example, if rare parameters cause ambiguous state probabilities, they can be marked as requiring manual verification. Training employs early stopping and possible built-in regularization (such as weight decay) to prevent the model from overfitting to common patterns, thereby maintaining its generalization ability to rare parameters. This machine learning model achieves a certain degree of reliability under specific conditions through validation set optimization.

[0084] Based on the rock mass geological model, a machine learning model is used to determine the collapse initiation conditions of the target rock mass, including:

[0085] S21, based on the rock mass geological model, extracts the structural topological features, mechanical features and geometric morphological features of each block through feature mapping units to form a block association feature set.

[0086] Optionally, the disaster range prediction terminal can be based on a rock mass geological model and use a feature mapping unit composed of a convolutional neural network with 3×3 convolutional kernels to extract structural topological features, mechanical features, and geometric morphological features. Based on the structural topological features, mechanical features, and geometric morphological features, it can construct rock mass topology map data with blocks as nodes and contact relationships between blocks as edges, forming a block association feature set.

[0087] S22, based on the block association feature set, uses the global average pooling layer of the attention interaction unit for pooling processing to obtain the feature mean vector.

[0088] Optionally, the disaster extent prediction terminal can input the block association feature set into the attention interaction unit, and perform dimensionality reduction and aggregation of the features through a global average pooling layer to obtain a feature mean vector that can represent the overall feature level.

[0089] S23, based on the feature mean vector, uses a fully connected layer with attention interaction units to perform feature transformation, and obtains the transformed feature vector.

[0090] Optionally, the disaster extent prediction terminal can input the feature mean vector into two fully connected layers with ReLU activation function for feature transformation, extracting deep correlation information to generate transformed feature vectors.

[0091] S24, based on the transformed feature vector, uses the output layer of the attention interaction unit to assign channel weights to obtain a high-dimensional feature vector.

[0092] Optionally, the disaster range prediction terminal can use an activation function to assign weights to the feature channels of each transformed feature vector in the output layer of the Sigmoid, thereby strengthening the contribution of key features, suppressing noise features, and forming a high-dimensional transformed feature vector.

[0093] S25, input the high-dimensional feature vector into the decision output unit to determine the block state category.

[0094] Optionally, the disaster range prediction terminal can input high-dimensional feature vectors into the fully connected layer with probability distribution parameterization in the decision output unit, and output the probability value of each block belonging to one of the four states: stable, damage evolution, critical instability, and instability.

[0095] S26. For each block, associate the block state category with the corresponding block association feature to generate the block state-feature correspondence.

[0096] Optionally, for each block, the disaster range prediction terminal can bind its state category with the corresponding associated features to generate a block state-feature correspondence.

[0097] S27. Based on the block state-feature correspondence, by analyzing the synergistic relationship of the associated features between blocks, the trigger parameter combination conditions for rock mass collapse initiation are determined, and the collapse initiation conditions are obtained.

[0098] Optionally, the disaster range prediction terminal can analyze the synergistic effect of the correlation characteristics between different blocks (which can be determined by Pearson correlation coefficient). For example, based on the Pearson correlation coefficient between the mechanical parameter threshold of the critical state block and the structural topological connectivity, the mechanical parameters exceeding the preset coefficient threshold are determined as the triggering parameters for rock mass collapse, and the collapse initiation conditions are obtained by combining them.

[0099] In one embodiment, based on a rock mass geological model and collapse initiation conditions, a kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement, obtaining the rock mass collapse trajectory, including:

[0100] S31. Based on the structural topological, mechanical, and geometric characteristics of each block in the rock mass geological model, a set of kinematic-dynamic coupled equations is established. The set of equations includes translational equations, rotational equations, collision response equations, and energy dissipation equations, which are used to characterize the block's sliding / rolling mode switching, collision transient response, separation condition criteria, and energy dissipation process related to the friction-recovery coefficient.

[0101] Optionally, the disaster range prediction terminal can establish a set of coupled equations (including translational equations, rotational equations, collision response equations, fragmentation evolution equations, and energy dissipation equations) based on the structural topological, mechanical, and geometric characteristics of each block in the rock mass geological model. These equations characterize the block's sliding / rolling mode switching, collision transient response, fragmentation process, separation condition criteria, and the entire energy dissipation process. The translational equations can be established based on Newton's second law to characterize the translational motion of the block's center of mass; the rotational equations can be established based on the Newton-Euler equations to characterize the block's rotation; the collision response equations can be established based on the law of conservation of momentum and contact mechanics theory to characterize the normal / tangential contact force, contact deformation, and springback characteristics of collisions between blocks; and the fragmentation evolution equations can be established based on rock fracture mechanics and the bonded particle model (BPM) to characterize crack propagation and fragmentation processes during block collisions. Together, these equations cover core processes such as sliding / rolling switching and collision response, enhancing the model's robustness.

[0102] S32, based on the collapse initiation conditions, determine the initial motion parameters of each block in the rock mass geological model.

[0103] Optionally, the disaster range prediction terminal can determine the initial motion parameters of each block at the time of collapse initiation, such as initial spatial coordinates, velocity components, angular velocity, and contact state, based on the critical parameters (such as critical displacement and stress threshold) in the collapse initiation conditions through dynamic analysis.

[0104] S33, based on the initial motion parameters, under the constraints of the coupled equation system, adopts an explicit time integration algorithm, and uses the block satisfying the preset stability condition or exceeding the simulation boundary as the stopping condition to perform iterative calculations to obtain a three-dimensional coordinate sequence to generate the motion path of the block.

[0105] Optionally, the disaster range prediction terminal can, based on initial motion parameters and under the constraints of coupled equations, employ explicit time integration algorithms such as the central difference method. Its maximum time step cannot exceed 1 / 20 of the elastic wave propagation time corresponding to the smallest block size. The block velocity must be ≤0.001 m / s (this velocity can be considered a stability condition; the velocity value can be derived from the block kinetic energy dissipation theory, and stability is considered when the block's kinetic energy is below 1% of its potential energy) or exceed a preset simulation boundary (the simulation space range defined in the kinematic-dynamic co-simulation, used to determine the simulation termination point). The boundary can be based on the target rock mass topography in the rock mass geological model or set according to actual needs. For example, in the horizontal direction, it can start from the top line of the rock mass and extend downwards along the slope for a distance not less than 5 times the slope height, and extend laterally for a distance not less than 3 times the width of the collapse source area; in the vertical direction, it can extend to the bedrock layer 10m below the surface. This boundary is used to cover most potential movement paths while avoiding unbounded calculations. Using as the stopping criterion, iterative calculations are performed, and each iteration outputs the three-dimensional coordinates of the block, forming a continuous three-dimensional coordinate sequence and generating the complete movement path of a single block.

[0106] S34, summarize the movement paths of each block to obtain the rock mass collapse trajectory.

[0107] Optionally, the disaster range prediction terminal can aggregate the movement paths of all blocks and integrate the data to obtain the rock mass collapse movement trajectory (such as the movement path, velocity time history, acceleration time history, kinetic energy time history, collision and fragmentation process, and deposition process data of the blocks) that can reflect the spatiotemporal evolution of the rock mass collapse.

[0108] In one embodiment, the initial motion parameters of each block in the rock mass geological model are determined based on the collapse initiation conditions, including:

[0109] S41. Based on structural topological features and geometric morphological features, determine the initial spatial coordinate parameters of each block.

[0110] Optionally, the disaster range prediction terminal can use information such as the block adjacency relationship in the structural topology features and the block size and spatial location in the geometric features to perform coordinate normalization processing through the geodetic coordinate system, and calibrate the spatial coordinate parameters, centroid position, moment of inertia and other parameters of each block to obtain the initial spatial coordinate parameters.

[0111] S42, based on the initial spatial coordinate parameters and collapse initiation conditions, calculates the initial velocity components of each block according to the translation equation.

[0112] Optionally, the disaster range prediction terminal can combine the initial spatial coordinate parameters and the critical instability state corresponding to the collapse initiation conditions, calculate the initial kinetic energy of the block when it becomes unstable based on the rock mass instability energy release rate and Griffith strength criterion, and obtain the initial velocity components of each block along the three spatial axes by combining the translation equation decomposition.

[0113] S43, based on the structural topological characteristics, uses rotation equations to calculate the initial angular velocity parameters of each block.

[0114] Optionally, the structural topology features reflect the spatial adjacency relationship between blocks, the joint surface cutting method, and the contact interlocking characteristics (such as surface contact, point contact, and no direct contact separated by joints and fissures). The disaster range prediction terminal can use the rotation equation, combined with the parameters in the structural topology features that characterize the rotational constraints of the blocks (such as the moment of inertia of the blocks about the contact point, the rotational resistance torque provided by the joint surface, the lever arm data of the contact point between blocks, etc.), to calculate the initial angular velocity parameters of each block about the three axes of space through torque balance analysis.

[0115] S44, generate the initial contact state parameters between the blocks according to the collision response equation.

[0116] Optionally, the disaster range prediction terminal can determine the contact type (such as point contact, surface contact, line contact, etc.) by judging the spatial relative position and contact fit degree between blocks based on the collision response equation and the geometric morphology and structural topology characteristics of each block in the rock geological model. It can then solve for parameters such as normal contact force, tangential contact force and contact stiffness of the corresponding contact type and integrate them to generate the initial contact state parameters between each block.

[0117] S45, the initial spatial coordinate parameters, initial velocity components, initial angular velocity parameters and contact state parameters are vector-superimposed to construct the first block motion parameter matrix.

[0118] Optionally, the disaster range prediction terminal can number the blocks and match the initial spatial coordinate parameters, initial velocity components, initial angular velocity parameters, and contact state parameters based on the block numbers. After normalizing the dimensions of various parameters for the same block number, vector superposition is performed. For example, a first block motion parameter matrix is ​​constructed with the block number as the row dimension and various motion parameters as the column dimension, so as to realize the systematic integration and quantitative representation of the initial motion parameters of each block.

[0119] S46, Perform dynamic consistency verification on the motion parameter matrix of the first block, and obtain the motion parameter matrix of the second block by eliminating abnormal parameter groups that do not conform to the law of conservation of energy.

[0120] Optionally, the disaster range prediction terminal can perform dynamic consistency verification on the first block motion parameter matrix, such as verifying whether each block parameter group satisfies the law of conservation of energy, determining whether the sum of the block's kinetic energy, potential energy, and contact energy is within a preset reasonable physical range, and eliminating abnormal parameter groups that do not meet the law requirements or exceed physical limits to correct the first block motion parameter matrix and obtain the second block motion parameter matrix.

[0121] S47. Based on the preset confidence interval, the Monte Carlo method is used to fit the probability distribution of the motion parameter matrix of the second block to generate the initial motion parameters.

[0122] Optionally, the disaster range prediction terminal can use the Monte Carlo method to fit the parameters in the second matrix to a normal distribution; the preset confidence interval is the statistical confidence level used in the probability distribution fitting, preferably 95%. This interval is used to quantify the uncertainty of the initial motion parameters and ensure the reliability of the parameter estimation. Optionally, in the Monte Carlo method, a normal distribution can be fitted using a 95% confidence interval, with a sampling number of no less than 1000, and outlier parameter groups with energy deviations greater than 5% are removed. The 95% confidence interval is based on the statistical significance standard, and the normality of the parameter distribution is verified by combining the KS test (p-value > 0.05), generating the initial motion parameter interval and optimal initial motion parameters for each block, which can improve the dynamic consistency.

[0123] In one embodiment, S51, the expression for the energy dissipation equation is:

[0124]

[0125] in, The total amount of energy dissipated. At the initial moment, The termination time, Based on the motion velocity of the i-th block geometric parameters of the j-th contact surface The dynamic friction coefficient, Let g be the mass of the i-th block, and g be the acceleration due to gravity. Let be the tangential velocity component of block i along the contact surface. Based on collision energy The collision restitution coefficient function, For contact stiffness parameters, Let be the change in the angular velocity of the block caused by the j-th contact surface.

[0126] Optionally, this energy dissipation equation is used to quantify the total energy loss generated by block sliding friction and contact collisions during rock mass collapse. It is a core component of the kinematic-dynamic coupled equation set, providing crucial energy constraints for calculating the evolution of block motion states. The equation is in integral form (the integration interval is the collapse initiation time). Until the end time Integrating two types of core dissipative energy, among which The term characterizes the frictional dissipation of the block sliding along the contact surface, and the dynamic friction coefficient. With the speed of block movement and contact surface geometry parameters Dynamically adjust the block mass mi, gravitational acceleration g, and tangential velocity components. Together they determine the intensity of frictional dissipation; The term characterizes the energy dissipation generated by block collisions, and the collision restitution coefficient function. Collision energy Quantitative mapping, contact stiffness parameters (Dimensions are N / m, values ​​can be calibrated through rock indentation tests, range is...) The change in rotational angular velocity caused by the contact surface The collision directly affects the degree of energy dissipation; the total energy dissipation is obtained by integrating the two factors. This equation is closely related to the mechanical parameters of the rock mass geological model. Structural topological features The geometric morphology-derived parameters, together with the translational equation, rotational equation and collision response equation, form a complete coupled constraint system. By calculating energy loss, it ensures that the evolution of the block's motion state (such as velocity and angular velocity) conforms to physical laws, providing reliable energy evolution data for kinematic-dynamic joint simulation and supporting the subsequent collapse trajectory of the rock-formed body.

[0127] In one embodiment, based on the rock mass collapse trajectory, an assessment of the collapse impact range is performed to obtain a rock mass collapse hazard range assessment report, including:

[0128] S61. Based on the rock mass collapse trajectory, extract the block trajectory range parameters. The block trajectory range includes the block movement path boundary, the maximum block diffusion distance, and the block collapse accumulation range.

[0129] Optionally, the disaster range prediction terminal can extract block trajectory range parameters from the motion trajectory using data processing tools (such as the Python trajectory analysis library): including block motion path boundaries, maximum block diffusion distance, block collapse and accumulation range, maximum block motion speed, and block impact energy; wherein, the block motion path boundaries can be obtained by fitting the spatial outer envelope of all block motion trajectories, and the coordinates of the outermost boundary of the trajectory are calibrated using a polygon fitting algorithm; the maximum block diffusion distance can be calculated by taking the starting position of each block as the origin, calculating the straight-line distance from its motion termination point to the origin, and taking the maximum value as the parameter; the block collapse and accumulation range is determined by statistically analyzing the coordinate clustering results of the final static positions of all blocks to determine the boundary and area of ​​the accumulation area.

[0130] S62, determine the range weight of the block trajectory range parameter according to the preset weight rules.

[0131] Optionally, the disaster range prediction terminal can assign range weights according to preset weighting rules. These weighting rules can be set based on the contribution of parameters to the disaster impact (a combination of Analytic Hierarchy Process (AHP) and entropy weighting can be used to avoid the subjectivity of fixed weights; or the subjective weights of each parameter can be determined by referring to national standard assessment criteria and using AHP, and at least 10 experts with over 10 years of experience in geological disaster assessment can be invited to compare and score the contribution of each parameter to the disaster impact, construct a judgment matrix, and obtain the subjective weights after passing a consistency test (CR < 0.1); or the weights can be based on Monte Carlo simulations). Using 1000 sample data points, the objective weights of each parameter are calculated using the entropy weighting method. The degree of variation of each parameter is quantified based on its information entropy; the smaller the information entropy, the greater the weight. Subjective and objective weights can be coupled based on the principle of minimum discriminative information to obtain combined weights, the sum of which is 1. In the preferred embodiment, after combined weighting, the weights for the maximum diffusion distance of the block are 0.35~0.45, the impact energy of the block is 0.25~0.35, the boundary of the block's movement path is 0.1~0.2, and the collapse and accumulation range of the block is 0.1~0.2. These weights can be dynamically adjusted according to the actual scenario of the target slope.

[0132] S63, based on the range weight, determines the influence radius of the block trajectory range parameter, and obtains the range of rock mass collapse disaster.

[0133] Optionally, the disaster range prediction terminal can calculate the comprehensive impact radius based on the range weight by weighted summation, and combine it with GIS spatial analysis tools to delineate a polygonal area based on the impact radius, thus obtaining the impact radius R of the rock mass collapse disaster range (calculated using the formula...). ,in The range weights for the block trajectory range parameters. The maximum diffusion distance for each parameter, The topographic impedance coefficient is determined based on slope vegetation cover, building obstruction, and topographic relief, with a value ranging from 0.5 to 1.0, and 1.0 when there is no obstruction. After the comprehensive impact radius calculation is completed, GIS spatial analysis tools can be used to generate a buffer zone downwards along the slope, starting from the collapse source area. Simultaneously, the boundary of the block movement path and the accumulation range can be combined to delineate the rock mass collapse hazard area. For example, the buffer zone width can be set to R±10%, and the hazard area can be divided into high-risk, medium-risk, and low-risk zones based on the block impact energy. Based on Monte Carlo simulation sample data, different exceedance probabilities (5%, 10%, 50%) of hazard areas can be generated to achieve probabilistic risk assessment rather than a single deterministic result.

[0134] S64. Based on the preset report template, integrate the block trajectory range parameters, range weights, and rock collapse hazard range to obtain a rock collapse hazard range assessment report.

[0135] Optionally, the disaster range prediction terminal can call a preset report template and integrate the aforementioned block trajectory range parameters, range weights, and disaster range information to generate a rock mass collapse disaster range assessment report. The report content may include: project overview, target rock mass geological conditions, monitoring data description, rock mass geological model construction status, collapse initiation condition identification results, motion simulation results, disaster range and hazard zoning, disaster prevention and mitigation recommendations, and also includes a parameter detail table, a disaster range spatial distribution map, and a risk level map. The report can be directly used for engineering review and disaster prevention decision-making.

[0136] In a preferred embodiment, the method further includes a dynamic closed-loop early warning step: based on real-time updated multi-dimensional monitoring data, the rock mass property characteristics and rock mass geological model are updated every hour to re-identify the collapse initiation conditions. When it is determined that the rock mass has entered a critical instability state, motion simulation and range assessment are automatically triggered to generate a dynamically updated disaster range assessment report and early warning information, which are pushed to on-site management personnel and emergency management departments to achieve dynamic closed-loop early warning of collapse disasters.

[0137] In one specific embodiment of this application, a prediction of the extent of complex rock collapse disasters was carried out on a rock slope in a mountainous scenic area. The specific steps are as follows:

[0138] By integrating sensor networks to collect multidimensional monitoring data of the target rock mass, including displacement data, stress data, and laser point cloud data, and based on the multidimensional monitoring data, the rock mass structural features and geometric features are extracted, and the rock mass mechanical features are calibrated according to experimental data.

[0139] The first model was generated by Delaunay triangulation based on the geometric morphology of the rock mass. The second model was formed by constructing a joint topology network based on the structural characteristics of the rock mass (the connectivity criteria for the joint topology network are: joint spacing < 0.05 meters and dip angle difference < 5°). The second model was divided into blocks using a region growing clustering algorithm (curvature change rate < 0.01 rad / m) to obtain the third model. The parameters of each block were calibrated based on mechanical characteristics and the normality was verified by KS test (p value > 0.05), thus constructing the geological model of the rock mass.

[0140] A pre-trained machine learning model is used to determine the collapse initiation conditions: a block-related feature set is extracted using a feature mapping unit with a 3×3 convolutional kernel; based on the block-related feature set, channel weights are assigned through an attention interaction unit to generate a high-dimensional feature vector; based on the high-dimensional feature vector, the probability of the block state is output through a decision output unit. For each block, a state-feature correspondence is generated, and the collapse initiation conditions are determined by analyzing collaborative parameters (e.g., displacement change rate > 0.1 mm / s and stress concentration factor > 1.5 are considered as collapse initiation conditions).

[0141] Based on the rock mass geological model and collapse initiation conditions, a kinematic-dynamic co-simulation method is used to simulate the rock mass collapse motion. A coupled equation set is established (including translational equations, rotational equations, collision response equations, and energy dissipation equations), where the integration interval of the energy dissipation equation is from collapse initiation to termination, and the contact stiffness parameter... The range is 1×10 6 –1×10 9 N / m. Based on the initiation conditions, initial motion parameters were generated using the Monte Carlo method (sampling 1000 times, 95% confidence interval), and an explicit time integration algorithm (time step of 100 μs) was used to iteratively calculate the motion paths of each block with a block velocity ≤0.001 m / s as the stopping condition, and the rock mass collapse motion trajectory was obtained by summarizing the results.

[0142] Assessment of landslide impact range based on movement trajectory: Extracting block trajectory range parameters (such as movement path boundary, maximum diffusion distance, and accumulation range), and assigning range weights (movement path boundary is...). =0.4, maximum diffusion distance is =0.3, stacking range is =0.3), calculate the radius of influence. The data is then integrated into a preset report template to generate a disaster extent assessment report. This embodiment achieves a disaster extent prediction error of less than 5% through multi-dimensional data fusion and dynamic simulation, thus improving the reliability of early warnings.

[0143] The aforementioned method for predicting the extent of complex rock mass collapse disasters overcomes the problems of traditional methods that rely on local observations and ignore multidimensional feature correlations by acquiring multidimensional monitoring data of the target rock mass and extracting its structural, mechanical, and geometric features. It accurately captures the characteristics of internal damage accumulation and sudden instability within the rock mass. Based on these attribute features, a geological model of the rock mass is formed through 3D modeling, structural topology network construction, block clustering, and mechanical parameter calibration, improving the representation ability of complex geological structures and reducing errors caused by structural uncertainties. A machine learning model containing feature mapping, attention interaction, and decision output units is employed to determine collapse initiation conditions through feature extraction, weight allocation, and state classification, achieving real-time identification of critical state transitions and solving the problems of empirical models. The previous method suffered from delayed early warning and inability to capture sudden instability. Based on the rock mass geological model and collapse initiation conditions, this method establishes a set of kinematic-dynamic coupled equations, determines initial motion parameters, and uses an explicit time integration algorithm for joint simulation. This quantifies the movement path and energy diffusion boundary of the collapse body, reduces prediction randomness, and overcomes the problem of traditional simplified mechanical models ignoring energy dissipation and collision response. By extracting block trajectory range parameters and combining them with weight rules to assess the impact range, an assessment report is generated, improving the accuracy of risk zoning. This effectively solves the problems of large prediction errors, difficulty in quantifying trajectory and energy boundaries, and scattered prediction results in traditional methods, which are difficult to directly guide disaster prevention work. This improves the accuracy and reliability of predicting the range of complex rock mass collapse disasters.

[0144] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0145] Based on the same inventive concept, this application also provides a complex rock mass collapse disaster range prediction system for implementing the aforementioned method for predicting the range of complex rock mass collapse disasters. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the complex rock mass collapse disaster range prediction system provided below can be found in the limitations of the complex rock mass collapse disaster range prediction method described above, and will not be repeated here.

[0146] In one exemplary embodiment, such as Figure 2As shown, a system for predicting the extent of complex rock mass collapse hazards is provided, including:

[0147] The rock mass feature extraction module 101 is used to acquire multidimensional monitoring data of the target rock mass and extract rock mass attribute features based on the multidimensional monitoring data.

[0148] The attribute feature modeling module 102 is used to perform geological modeling based on the rock mass attribute features to obtain a rock mass geological model;

[0149] The collapse condition determination module 103 is used to determine the collapse initiation conditions of the target rock mass based on the rock mass geological model and using a machine learning model.

[0150] The collapse motion simulation module 104 is used to simulate rock mass collapse motion based on the rock mass geological model and collapse initiation conditions, using a kinematic-dynamic joint simulation method to obtain the rock mass collapse motion trajectory.

[0151] The impact range assessment module 105 is used to assess the impact range of rock collapse based on the rock collapse movement trajectory and obtain a rock collapse disaster range assessment report.

[0152] In one embodiment, the rock mass attribute features in the attribute feature modeling module 102 include rock mass structural features, rock mass mechanical features, and rock mass geometric features;

[0153] The attribute feature modeling module 102 is also used for:

[0154] Based on the geometric morphological characteristics of the rock mass, a three-dimensional model was created to obtain the first model;

[0155] Based on the characteristics of the rock mass structure, a topological network of the rock mass structure is constructed in the first model to obtain the second model;

[0156] Based on the second model, the third model is obtained by clustering blocks according to the rock mass structure topology network;

[0157] Based on the rock mass mechanics characteristics, the mechanical parameters of each block in the third model are calibrated to obtain the rock mass geological model.

[0158] In one embodiment, the machine learning model in the collapse condition determination module 103 includes a feature mapping unit, an attention interaction unit, and a decision output unit. The feature mapping unit includes a convolutional neural network with 3×3 convolutional kernels. The attention interaction unit adopts a channel attention mechanism based on the SE structure and includes a global average pooling layer, two fully connected layers with ReLU activation function, and an output layer with Sigmoid activation function. The decision output unit includes a fully connected layer that parameterizes the probability distribution for classifying the state of the block. The block state categories include stable, critical, and unstable.

[0159] The collapse condition determination module 103 is also used for:

[0160] Based on the rock mass geological model, structural topological features, mechanical features and geometric morphological features associated with each block are extracted through feature mapping units to form a block association feature set;

[0161] Based on the block-related feature set, a global average pooling layer of the attention interaction unit is used for pooling to obtain the feature mean vector.

[0162] Based on the feature mean vector, a fully connected layer with attention interaction units is used to perform feature transformation to obtain the transformed feature vector;

[0163] Based on the transformed feature vector, channel weights are assigned to the output layer of the attention interaction unit to obtain a high-dimensional feature vector;

[0164] The high-dimensional feature vector is input into the decision output unit to determine the block state category;

[0165] For each block, the block state category is associated with the corresponding block association feature to generate a block state-feature correspondence;

[0166] Based on the correspondence between block state and feature, the triggering parameter combination conditions for rock mass collapse are determined by analyzing the synergistic relationship of the associated features between blocks, thus obtaining the collapse initiation conditions.

[0167] In one embodiment, the collapse motion simulation module 104 is further configured to:

[0168] Based on the structural topological, mechanical, and geometric characteristics of each block in the rock mass geological model, a set of kinematic-dynamic coupled equations is established. The set of equations includes translational equations, rotational equations, collision response equations, and energy dissipation equations, which are used to characterize the block's sliding / rolling mode switching, collision transient response, separation condition criterion, and energy dissipation process related to the friction-recovery coefficient.

[0169] Based on the collapse initiation conditions, determine the initial motion parameters of each block in the rock mass geological model;

[0170] Based on the initial motion parameters, under the constraints of the coupled equation system, an explicit time integration algorithm is adopted. The block satisfies the preset stability condition or exceeds the simulation boundary as the stopping condition. Iterative calculation is performed to obtain the three-dimensional coordinate sequence to generate the motion path of the block.

[0171] By summarizing the movement paths of each block, the trajectory of the rock mass collapse can be obtained.

[0172] In one embodiment, the collapse motion simulation module 104 is further configured to:

[0173] Based on the collapse initiation conditions, the initial motion parameters of each block in the rock mass geological model are determined, including:

[0174] Based on structural topological and geometric features, the initial spatial coordinate parameters of each block are determined;

[0175] Based on the initial spatial coordinate parameters and collapse initiation conditions, the initial velocity components of each block are calculated according to the translation equation.

[0176] Based on the structural topological characteristics, the initial angular velocity parameters of each block are calculated using rotational equations;

[0177] Based on the collision response equation, the initial contact state parameters between each block are generated;

[0178] The initial spatial coordinate parameters, initial velocity components, initial angular velocity parameters, and contact state parameters are vector-superimposed to construct the first block motion parameter matrix;

[0179] The dynamic consistency of the first block motion parameter matrix is ​​verified, and the second block motion parameter matrix is ​​obtained by eliminating abnormal parameter groups that do not conform to the law of conservation of energy.

[0180] Based on the preset confidence interval, the Monte Carlo method is used to fit the probability distribution of the motion parameter matrix of the second block, generating the initial motion parameters.

[0181] In one embodiment, the expression for the energy dissipation equation in the collapse motion simulation module 104 is:

[0182]

[0183] in, The total amount of energy dissipated. At the initial moment, The termination time, Based on the motion velocity of the i-th block geometric parameters of the j-th contact surface The dynamic friction coefficient, Let g be the mass of the i-th block, and g be the acceleration due to gravity. Let be the tangential velocity component of block i along the contact surface. Based on collision energy The collision restitution coefficient function, For contact stiffness parameters, Let be the change in the angular velocity of the block caused by the j-th contact surface.

[0184] In one embodiment, the influence range assessment module 105 is further configured to:

[0185] Based on the rock mass collapse trajectory, block trajectory range parameters are extracted. The block trajectory range includes the block movement path boundary, the maximum block diffusion distance, and the block collapse accumulation range.

[0186] The range weights of the block trajectory range parameters are determined according to the preset weighting rules.

[0187] Based on the range weight, the influence radius of the block trajectory range parameter is determined, and the range of rock mass collapse disaster is obtained;

[0188] Based on the preset report template, the block trajectory range parameters, range weights, and rock collapse hazard range are integrated to obtain a rock collapse hazard range assessment report.

[0189] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for predicting the extent of complex rock mass collapse hazards, characterized in that, The method includes: Acquire multidimensional monitoring data of the target rock mass, and extract rock mass property characteristics based on the multidimensional monitoring data; Based on the aforementioned rock mass properties, geological modeling is performed to obtain a rock mass geological model; Based on the geological model of the rock mass, a machine learning model is used to determine the collapse initiation conditions of the target rock mass; Based on the geological model of the rock mass and the collapse initiation conditions, the kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement and obtain the rock mass collapse movement trajectory. Based on the rock mass collapse trajectory, the impact range of the collapse is assessed, resulting in a rock mass collapse hazard range assessment report.

2. The method according to claim 1, characterized in that, The rock mass properties include rock mass structural features, rock mass mechanical features, and rock mass geometric features; The geological modeling based on the rock mass property characteristics, resulting in a rock mass geological model, includes: Based on the geometric morphological characteristics of the rock mass, a three-dimensional model was created to obtain the first model; Based on the rock mass structural characteristics, a rock mass structural topology network is constructed in the first model to obtain the second model; Based on the second model, and according to the rock mass structure topology network, block clustering is performed to obtain the third model; Based on the rock mass mechanical characteristics, the mechanical parameters of each block in the third model are calibrated to obtain the rock mass geological model.

3. The method according to claim 2, characterized in that, The machine learning model includes a feature mapping unit, an attention interaction unit, and a decision output unit. The feature mapping unit comprises a convolutional neural network with 3×3 convolutional kernels. The attention interaction unit adopts a channel attention mechanism based on the SE structure and includes a global average pooling layer, two fully connected layers with ReLU activation function, and an output layer with Sigmoid activation function. The decision output unit includes a fully connected layer with probability distribution parameterization for classifying the state of blocks, where the block state categories include stable, critical, and unstable. The process of determining the collapse initiation conditions of the target rock mass using a machine learning model based on the rock mass geological model includes: Based on the rock mass geological model, the structural topological features, mechanical features and geometric morphological features associated with each block are extracted through the feature mapping unit to form a block-related feature set; Based on the block association feature set, the global average pooling layer of the attention interaction unit is used for pooling to obtain the feature mean vector. Based on the mean feature vector, the fully connected layer of the attention interaction unit is used to perform feature transformation to obtain the transformed feature vector; Based on the transformed feature vector, channel weights are assigned using the output layer of the attention interaction unit to obtain a high-dimensional feature vector; The high-dimensional feature vector is input into the decision output unit to determine the block state category; For each block, the block state category is associated with the corresponding block association feature to generate a block state-feature correspondence; Based on the block state-feature correspondence, by analyzing the synergistic relationship of the associated features between each block, the trigger parameter combination conditions for rock mass collapse initiation are determined, and the collapse initiation conditions are obtained.

4. The method according to claim 3, characterized in that, Based on the geological model of the rock mass and the collapse initiation conditions, a kinematic-dynamic joint simulation method is used to simulate the rock mass collapse movement, obtaining the rock mass collapse movement trajectory, including: Based on the structural topological, mechanical, and geometric features of each block in the geological model of the rock mass, a set of kinematic-dynamic coupled equations is established. The set of equations includes translational equations, rotational equations, collision response equations, and energy dissipation equations, which are used to characterize the block's sliding / rolling mode switching, collision transient response, separation condition criteria, and energy dissipation process related to the friction-recovery coefficient. Based on the collapse initiation conditions, determine the initial motion parameters of each block in the rock mass geological model; Based on the initial motion parameters, under the constraints of the coupled equations, an explicit time integration algorithm is used to perform iterative calculations with the block satisfying a preset stability condition or exceeding the simulation boundary as the stopping condition, to obtain a three-dimensional coordinate sequence to generate the motion path of the block. By summing the movement paths of each block, the trajectory of the rock mass collapse is obtained.

5. The method according to claim 4, characterized in that, The determination of the initial motion parameters of each block in the rock mass geological model based on the collapse initiation conditions includes: Based on the structural topological features and the geometric morphological features, the initial spatial coordinate parameters of each block are determined; Based on the initial spatial coordinate parameters and the collapse initiation conditions, the initial velocity components of each block are calculated according to the translation equation. Based on the aforementioned structural topological features, the initial angular velocity parameters of each block are calculated using the aforementioned rotation equation; Based on the collision response equation, the initial contact state parameters between the blocks are generated. The initial spatial coordinate parameters, the initial velocity components, the initial angular velocity parameters, and the contact state parameters are vector-superimposed to construct the first block motion parameter matrix; The first block motion parameter matrix is ​​subjected to dynamic consistency verification. By eliminating abnormal parameter groups that do not conform to the law of conservation of energy, the second block motion parameter matrix is ​​obtained. Based on the preset confidence interval, the Monte Carlo method is used to fit the probability distribution of the second block motion parameter matrix to generate the initial motion parameters.

6. The method according to claim 5, characterized in that, The expression for the energy dissipation equation is: in, The total amount of energy dissipated. At the initial moment, The termination time, Based on the motion velocity of the i-th block geometric parameters of the j-th contact surface The dynamic friction coefficient, Let g be the mass of the i-th block, and g be the acceleration due to gravity. Let be the tangential velocity component of block i along the contact surface. Based on collision energy The collision restitution coefficient function, For contact stiffness parameters, Let be the change in the angular velocity of the block caused by the j-th contact surface.

7. The method according to claim 4, characterized in that, The assessment of the impact range of the rock mass collapse, based on the rock mass collapse trajectory, yields a rock mass collapse hazard range assessment report, including: Based on the rock mass collapse trajectory, block trajectory range parameters are extracted, including the block movement path boundary, the maximum block diffusion distance, and the block collapse accumulation range. The range weights of the block trajectory range parameters are determined according to preset weighting rules. Based on the range weight, the influence radius of the block trajectory range parameter is determined to obtain the range of rock mass collapse disaster; Based on the preset report template, the block trajectory range parameters, the range weights, and the rock collapse disaster range are integrated to obtain the rock collapse disaster range assessment report.

8. A system for predicting the extent of complex rock mass collapse hazards, characterized in that, The system includes: The rock mass feature extraction module is used to acquire multidimensional monitoring data of the target rock mass and extract rock mass attribute features based on the multidimensional monitoring data. The attribute feature modeling module is used to perform geological modeling based on the rock mass attribute features to obtain a rock mass geological model; The collapse condition determination module is used to determine the collapse initiation conditions of the target rock mass based on the rock mass geological model and using a machine learning model. The collapse motion simulation module is used to simulate rock mass collapse motion based on the rock mass geological model and the collapse initiation conditions, using a kinematic-dynamic joint simulation method to obtain the rock mass collapse motion trajectory; The impact range assessment module is used to assess the impact range of the rock mass collapse based on the rock mass collapse trajectory, and to obtain a rock mass collapse disaster range assessment report.

Images