A mine slope stability dynamic evaluation method based on unmanned aerial vehicle laser radar

Abstract

This application relates to a dynamic assessment method for mine slope stability based on UAV lidar. The method includes: establishing a rock mechanics numerical model of the mine slope based on the slope's geometric and geological information at the initial monitoring time from a high-precision time-series point cloud sequence; establishing a spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope; based on a data assimilation algorithm, using the spatial coordinate mapping relationship, fusing the complete slope displacement observation field at the current monitoring time with the predicted state of the rock mechanics numerical model, and inverting and updating the initial mechanical parameter field to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model. This drives the rock mechanics numerical model to predict the stability of the slope over a preset future period, obtaining the future deformation field and stability indicators of the slope. This method can improve the dynamic risk assessment and early warning capabilities of slopes.

Description

Technical Field

[0001] This invention belongs to the field of slope stability assessment technology, and in particular relates to a dynamic assessment method for mine slope stability based on UAV lidar. Background Technology

[0002] With the development of technologies in the field of mine engineering safety monitoring and numerical simulation, UAV lidar three-dimensional monitoring technology and rock mechanics numerical simulation technology based on finite element method / discrete element method have emerged. UAV lidar technology collects high-precision three-dimensional point cloud data of slope surface to reveal the geometric changes of slope. Combined with numerical simulation technology, mechanical models are constructed to explore the instability mechanism of slope and predict future stability, providing theoretical support for slope safety assessment.

[0003] In traditional mine slope stability assessment, during the monitoring phase, the three-dimensional point cloud data acquired by UAV lidar is mostly used for post-event description of slope deformation and anomaly alarms, which can only achieve a "phenomenal" presentation of the current state of the slope and remain at the level of "knowing what is happening". In the simulation phase, technicians obtain rock mechanics parameters through limited field or indoor tests, use them as initial inputs to construct numerical models, and then deduce the slope instability law and stability trend.

[0004] However, the monitoring data mentioned above is not effectively used for model optimization and can only play a post-event feedback role, failing to provide dynamic support for prediction. The rock mechanics parameters on which the numerical model relies are difficult to represent the spatiotemporal heterogeneity and anisotropy of complex rock masses, and even more so, they cannot reflect their time-varying characteristics in dynamic processes such as weathering, unloading, and excavation. This results in a large deviation between the model prediction results and the actual situation, making it difficult to meet the actual needs of slope dynamic risk assessment and early warning. Summary of the Invention

[0005] Therefore, it is necessary to provide a dynamic assessment method for mine slope stability based on UAV lidar that can improve the dynamic risk assessment and early warning capabilities of slopes, addressing the aforementioned technical issues.

[0006] Firstly, this application provides a dynamic assessment method for mine slope stability based on UAV lidar, including: The raw scanning data of the UAV lidar of the target slope under multiple time-series monitoring periods were acquired, and the raw scanning data of the UAV lidar under each time-series monitoring period were solved and registered to obtain a high-precision time-series point cloud sequence under the same coordinate system. A spatiotemporally continuous four-dimensional geometric evolution model of a slope is constructed based on a high-precision time-series point cloud sequence, and the complete slope displacement observation field at the current monitoring time is extracted based on the four-dimensional geometric evolution model of the slope. A rock mechanics numerical model of a mine slope is established based on the slope geometry and geological information at the initial monitoring time in a high-precision time-series point cloud sequence. The spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope is also established. The rock mechanics numerical model includes the initial mechanical parameter field. Based on the data assimilation algorithm, the complete slope displacement observation field at the current monitoring time is fused with the predicted state of the rock mechanics numerical model by utilizing the spatial coordinate mapping relationship, and the initial mechanical parameter field is inverted and updated to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model. Using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, the rock mechanics numerical model is driven to perform stability extrapolation and prediction for a preset future time period, thereby obtaining the future deformation field and stability index of the slope.

[0007] In one embodiment, a spatiotemporally continuous four-dimensional geometric evolution model of the slope is constructed based on a high-precision time-series point cloud sequence, and the complete slope displacement observation field at the current monitoring time is extracted based on the four-dimensional geometric evolution model of the slope, including: Calculate the three-dimensional coordinate difference between each point in the high-precision time-series point cloud sequence and the corresponding point at the initial monitoring time to obtain the actual displacement vector of each point at the corresponding monitoring time. Use the spatial coordinates of each point and the monitoring time as sample inputs and the actual displacement vector as sample outputs to construct a training sample set. The pre-defined multi-layer feedforward neural network model is trained based on the training sample set, and the network parameters are optimized by minimizing the pre-defined loss function through the backpropagation algorithm to obtain the four-dimensional geometric evolution model of the slope. The four-dimensional geometric evolution model of the slope takes three-dimensional spatial coordinates and normalized time as inputs and outputs a three-dimensional displacement vector. The loss function is obtained using the following formula: in, Spatial coordinates; For monitoring time; It is a multi-layer feedforward neural network model; These are model parameters; This is the actual displacement vector; The square of the L2 norm; The regularization coefficient is used. To apply to model parameters Preset constraints; By querying the current monitoring time based on the four-dimensional geometric evolution model of the slope, a complete slope displacement observation field covering the entire slope space can be obtained.

[0008] In one embodiment, a rock mechanics numerical model of the mine slope is established based on the slope geometry and geological information at the initial monitoring time in a high-precision time-series point cloud sequence. A spatial coordinate mapping relationship is established between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope, including: Based on the slope point cloud data at the initial monitoring time and the reconstructed three-dimensional model of the slope surface, and combined with borehole-profile geological data, a three-dimensional geological entity model is constructed; the three-dimensional geological entity model includes the main strata, lithological boundaries and dominant structural planes; The three-dimensional geological entity model is meshed, and constitutive models of elastic-plastic and joint materials are assigned to different lithological regions and structural planes according to lithological boundaries and dominant structural planes, so as to obtain a numerical model of rock mechanics. Based on regional geological data, in-situ test data, and indoor rock sample test results, an initial value is set for the empirical parameters of different material regions in the rock mechanics numerical model, resulting in an initial mechanical parameter field. The empirical parameters include cohesion, internal friction angle, elastic modulus, and Poisson's ratio. Based on the index relationship between the spatial coordinates of all calculation nodes in the rock mechanics numerical model and the spatial domain defined by the four-dimensional geometric evolution model of the slope, the spatial coordinate mapping relationship can be obtained by using coordinate interpolation to obtain the displacement observation value from the four-dimensional geometric evolution model of the slope for any calculation node.

[0009] In one embodiment, based on a data assimilation algorithm, the complete slope displacement observation field at the current monitoring time is fused with the predicted state of the rock mechanics numerical model using spatial coordinate mapping relationships. The initial mechanical parameter field is then inverted and updated to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model, including: The current model state variables and the current rock mechanics parameter field of the rock mechanics numerical model are combined to obtain the amplified state vector. In the first assimilation cycle, the current model state variables and the current rock mechanics parameter field are the initial model state and the initial mechanical parameter field, respectively. In subsequent assimilation cycles, the current model state variables and the current rock mechanics parameter field are the rock mechanics numerical model state and the updated rock mechanics parameter field after the previous assimilation cycle. Using the amplified state vector as the initial condition, a mechanical calculation for a monitoring period is performed using a rock mechanics numerical model to obtain the state variables of the prediction model. The state variables of the prediction model are then combined with the unchanged parameter field to obtain the predicted amplified state vector. By using spatial coordinate mapping relationships, the corresponding positions of each calculation node of the rock mechanics numerical model in the four-dimensional geometric evolution model of the slope are determined, and the displacement observation values ​​of each calculation node at the current monitoring time are queried to obtain the observed displacement vector. An ensemble Kalman filter algorithm is used to fuse the predicted amplified state vector with the observed displacement vector to obtain a state vector set, and the updated rock mechanics parameter field and rock mechanics numerical model state are obtained by updating the state vector set. The corresponding steps for updating are as follows: The amplified state vector can be obtained using the following formula: in, Let i be the predicted augmented state vector of the i-th state vector. To observe the displacement vector, For the random perturbation applied to the observed displacement vector, To extract the observation operators corresponding to the computational node displacements from the state vector, The Kalman gain matrix is ​​calculated based on the covariance of the forecast set and the covariance of the observation error. The analysis expands the state vector to represent the updated i-th state vector; Calculate the mean of all analyzed amplified state vector sets, and use the distribution of the parameter field part and the model state variable part as the updated rock mechanics parameter field and rock mechanics numerical model state.

[0010] In one embodiment, using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, the rock mechanics numerical model is driven to perform stability prediction for a preset future time period, obtaining the future deformation field and stability indicators of the slope, including: The updated rock mechanics numerical model is obtained by replacing the values ​​of the corresponding material regions or structural surfaces in the rock mechanics parameter field with the values ​​of each parameter. The state of the rock mechanics numerical model is set as the initial condition for the prediction calculation of the rock mechanics numerical model; the state of the rock mechanics numerical model includes displacement field, stress field and plastic state variables; On the rock mechanics numerical model with updated parameters, load changes and / or boundary condition changes for a future preset time period are set, and the rock mechanics numerical model is driven to perform numerical simulation calculations based on the initial conditions to obtain numerical simulation calculation results. The slope displacement field, maximum shear stress distribution area, and plastic strain cloud map for a future preset time period are extracted from the numerical simulation results to obtain the future deformation field; Based on the stress state and updated rock mechanics parameter field obtained from numerical simulation, the overall safety factor of the slope under the current and future conditions is calculated using the strength reduction method, thus obtaining the stability index.

[0011] In one embodiment, the raw scanning data of the UAV lidar under each time-series monitoring period are processed and registered to obtain a high-precision time-series point cloud sequence in the same coordinate system, including: The raw scanning data of UAV lidar under each time series monitoring period are processed to obtain the initial three-dimensional point cloud under each time series monitoring period. Outlier removal and statistical filtering are performed on the initial three-dimensional point cloud under each time series monitoring period to obtain the standard three-dimensional point cloud under each time series monitoring period. Using the standard 3D point cloud of the first time-series monitoring cycle as a fixed reference benchmark, and using artificially placed targets or natural feature points that can be stably identified in each time-series monitoring cycle as control points, the standard 3D point cloud of each time-series monitoring cycle is initially registered to the coordinate system of the fixed reference benchmark through rigid body transformation, thus obtaining the preliminary spatiotemporal registration 3D point cloud of each time-series monitoring cycle. Based on the preliminary spatiotemporal registration 3D point clouds under each time series monitoring period, the iterative nearest point algorithm is used to perform fine registration on the preliminary spatiotemporal registration 3D point clouds of two adjacent time series monitoring periods to obtain a high-precision time series point cloud sequence. Fine registration corresponds to optimizing the registration transformation parameters by minimizing the distance error between corresponding points in the preliminary spatiotemporal registration 3D point clouds, and then performing registration according to the optimized registration transformation parameters.

[0012] In one embodiment, the method further includes: In response to receiving the new batch of raw UAV lidar scanning data collected in the next time-series monitoring cycle, the raw UAV lidar scanning data is calculated and registered to obtain an updated high-precision time-series point cloud sequence. The updated high-precision time-series point cloud sequence was used to incrementally train the four-dimensional geometric evolution model of the slope, and the complete slope displacement observation field at the current monitoring time was extracted. The updated rock mechanics parameter field and rock mechanics numerical model state obtained from the previous time series monitoring period are used as new initial conditions for inversion update and stability prediction to obtain new future deformation field and stability index.

[0013] Secondly, this application also provides a dynamic assessment device for mine slope stability based on UAV lidar, comprising: The point cloud data module is used to acquire the raw scanning data of UAV lidar under multiple time-series monitoring periods of the target slope, and to perform calculation and registration processing on the raw scanning data of UAV lidar under each time-series monitoring period to obtain a high-precision time-series point cloud sequence under the same coordinate system. The slope four-dimensional continuous deformation field modeling module is used to construct a spatiotemporally continuous slope four-dimensional geometric evolution model based on a high-precision time-series point cloud sequence, and extract the complete slope displacement observation field at the current monitoring time based on the slope four-dimensional geometric evolution model. The rock mechanics module is used to establish a rock mechanics numerical model of a mine slope based on the slope geometry and geological information at the initial monitoring time in a high-precision time-series point cloud sequence, and to establish the spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope; the rock mechanics numerical model includes the initial mechanical parameter field; The inversion update module is used to fuse the complete slope displacement observation field at the current monitoring time with the predicted state of the rock mechanics numerical model based on the data assimilation algorithm and the spatial coordinate mapping relationship, and to invert and update the initial mechanical parameter field to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model. The fusion prediction module is used to drive the rock mechanics numerical model to perform stability extrapolation and prediction for a preset future period, using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, so as to obtain the future deformation field and stability index of the slope.

[0014] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above-mentioned dynamic assessment methods for mine slope stability based on UAV lidar.

[0015] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods for dynamic assessment of mine slope stability based on UAV lidar.

[0016] The aforementioned dynamic assessment method for mine slope stability based on UAV lidar utilizes parameter fields and model states to drive numerical models for future stability prediction. By constructing a closed-loop system with bidirectional coupling of "observation-simulation" and dynamic parameter optimization, it achieves continuous learning and approximation of the mechanical properties of mine slopes, significantly improving the accuracy and foresight of stability state assessment and instability prediction. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a flowchart illustrating the dynamic assessment method for mine slope stability based on UAV lidar of the present invention. Figure 2 This is a flowchart illustrating the steps of step S102. Figure 3 It is a schematic diagram of the sub - steps of step S103; Figure 4 It is a structural composition diagram of the device for dynamically evaluating the stability of mine slopes based on UAV lidar of the present invention. Specific embodiments

[0019] In order to make the purpose, technical solutions and advantages of this application clearer, the following further elaborates on this application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0020] In one embodiment, as Figure 1 shown, a method for dynamically evaluating the stability of mine slopes based on UAV lidar is provided. In this embodiment, this method is illustrated by taking its application to a terminal as an example. It can be understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is realized through the interaction between the terminal and the server. In this embodiment, the method includes the following steps: S101. Obtain the original UAV lidar scan data of the target slope in multiple time - series monitoring periods, and perform calculation and registration processing on the original UAV lidar scan data in each time - series monitoring period to obtain a high - precision time - series point cloud sequence in the same coordinate system.

[0021] Schematically, set the initial monitoring time and subsequent time - series monitoring nodes . The time interval between adjacent monitoring periods can be dynamically adjusted according to the slope stability state. During the active deformation period, it can be shortened to 1 - 3 days, and during the stable period, it can be extended to 1 - 2 weeks. The flight route adopts a strip - overlapping flight mode, the overlap rate between adjacent flight lines is not less than 30%, and the heading overlap rate is not less than 80% to ensure that there is no data blind area in the entire slope area; the laser point cloud density of the lidar meets not less than 50 points per square meter to ensure the ability to capture minute deformations. The lidar device carried by the UAV integrates a high - precision GNSS module and an IMU inertial measurement unit. The GNSS positioning accuracy is better than ±1 cm, and the angular velocity measurement accuracy of the IMU is better than ±0.1° / h.

[0022] Furthermore, use the calculation software supporting the lidar device for calculation. During the calculation process, error correction terms such as earth curvature correction and atmospheric refraction correction need to be introduced to eliminate the influence of systematic errors on the coordinate accuracy. After the calculation is completed, multi - stage denoising and filtering processing are performed on the original point cloud data. Furthermore, point cloud registration processing is performed through coarse registration and fine registration to obtain a high - precision time - series point cloud sequence in the same coordinate system .

[0023] S102. Construct a spatiotemporally continuous four-dimensional geometric evolution model of the slope based on a high-precision time-series point cloud sequence, and extract the complete slope displacement observation field at the current monitoring time based on the four-dimensional geometric evolution model of the slope.

[0024] To illustrate, the time dimension t is normalized, mapping the initial monitoring time t0 to t=0, and the current latest monitoring time t k Mapped to t=1, any intermediate monitoring time t i The normalized time is calculated as t i =(t i -t0) / (t k -t0). Optionally, the four-dimensional geometric evolution model employs a neural network architecture based on Neural Radiation Field (NeRF) extensions. This system can learn continuous spatiotemporal mapping relationships from a small amount of discrete observation data, supporting displacement queries at any spatiotemporal point. The neural network's input layer contains four neurons, corresponding to spatial coordinates x, y, and z, and normalized time t, respectively. The hidden layers consist of 6-8 layers, each with 128-256 neurons, and use the ReLU activation function. To enhance the network's ability to fit nonlinear deformation relationships, the output layer contains four neurons. The first three neurons output the displacement vector of the spatiotemporal point relative to the initial time t0. The fourth neuron outputs a weight value w, which ranges from [0,1]. This weight value represents whether the point exists at the query time. If the point is missing in part of the temporal point cloud due to slope peeling or occlusion, the weight value is close to 0; otherwise, it is close to 1. After training, the neural network... It has the ability to continuously query in space and time. For any given spatial point (x, y, z) and any time t, the displacement vector of that point can be obtained by inputting the network. .

[0025] Furthermore, based on the trained four-dimensional geometric evolution model, the current monitoring time t is extracted. k Complete slope displacement observation field During the extraction process, a uniformly distributed three-dimensional grid of points is generated based on the three-dimensional geometry of the slope, covering the entire spatial area of ​​the slope. The spatial coordinates (x, y, z) of each grid point and the normalized time t=1 at the current moment are input into the model to obtain the displacement vector of each grid point. Simultaneously, based on the spatial gradient analysis of the displacement field, the location information of the potential rupture surface is inferred, i.e., the gradient tensor of the displacement field in three-dimensional space is calculated. This tensor contains the partial derivatives of the displacement field along the x, y, and z coordinate axes, fully characterizing the spatial rate of change of the displacement field. ,in, These are the complete slope displacement observation fields. The gradient tensor for the displacement components along the three coordinate axes is calculated using the finite difference method. Based on the constructed 3D grid points, displacement data of six neighboring grid points are selected around each grid point. The partial derivatives are solved by the ratio of the displacement difference between neighboring points to the grid step size, i.e., the grid resolution. For example, for a grid point... , for ,in This represents the grid step size.

[0026] By performing eigenvalue decomposition on the gradient tensor, the principal strain and shear strain components of the strain tensor are extracted. The formation and development of a potential fracture surface are inevitably accompanied by local shear strain concentration or abrupt changes in displacement gradient. That is, due to the significant difference in displacement between the rock masses on both sides of the fracture surface, the continuity of the displacement field in this region is interrupted, manifested as abnormal peak values ​​in the eigenvalues ​​of the gradient tensor. A shear strain threshold is set. and displacement gradient abrupt change threshold The shear strain threshold is determined based on the rock mass type and mechanical properties, while the displacement gradient abrupt change threshold is set based on the point cloud observation accuracy and model fitting error. The entire three-dimensional grid of the slope displacement observation field is traversed, and points with shear strain greater than [a certain value] are selected. Or the magnitude of the displacement gradient is greater than Anomalies are identified in the grid, and these anomalies represent the contour points of potential fracture surfaces. Spatial clustering analysis is performed on these anomalies using the density-based clustering algorithm (DBSCAN). Anomalies with a spatial distance less than twice the grid step size are grouped into the same cluster, with each cluster corresponding to a potential local fracture region. For each cluster, principal component analysis (PCA) is used to extract the dominant direction of the region, determining the orientation and dip of the fracture surface. Then, moving least squares is used to fit a surface to the contour points within the clusters, constructing a three-dimensional geometric model of the potential fracture surface. During the fitting process, the goal is to minimize the sum of squared distances from the contour points to the fitted surface, ensuring that the rupture surface model can accurately reflect the spatial morphology of the discontinuous displacement region.

[0027] S103. Based on the slope geometry and geological information at the initial monitoring time in the high-precision time-series point cloud sequence, establish a rock mechanics numerical model of the mine slope, and establish the spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope; the rock mechanics numerical model includes the initial mechanical parameter field.

[0028] High-precision point cloud at the initial monitoring time Based on this, a three-dimensional solid geometric model of the slope is constructed using detailed geological exploration data, which must include borehole columnar sections, geological profiles, joint and fracture survey data, and other information. Specifically, point cloud processing software is used to... Surface reconstruction is performed to generate a triangular mesh surface model. The surface smoothness is optimized using the Poisson reconstruction algorithm to ensure that the model can accurately reflect the original topographic features of the slope. Then, it is imported into 3D geological modeling software. The thickness, burial depth and spatial distribution range of each rock layer are determined based on borehole data. The geometric morphology of the stratigraphic interface is supplemented based on the geological profile map. At the same time, the main structural surfaces, such as the attitude, spacing and extension length of joints and faults, are accurately depicted in the model with reference to the joint and fracture survey results.

[0029] The initial mechanical parameter field setting needs to reflect the spatial heterogeneity of the rock mass. That is, based on the geological survey report, the rock mass types of each rock layer and structural plane of the slope should be identified. For different rock mass types, indoor mechanical tests should be carried out to obtain basic mechanical parameters such as uniaxial compressive strength, elastic modulus, Poisson's ratio, cohesion, and friction angle. For structural planes, their shear strength parameter, i.e., cohesion, should be determined through direct shear tests. and friction angle Furthermore, a spatial interpolation algorithm is used to extend the discrete parameter values ​​obtained from the experiment into a continuous spatial parameter field. ,in Let cohesion be the spatial distribution function. Let be the spatial distribution function of the friction angle. Let be the spatial distribution function of the elastic modulus. Let be the spatial distribution function of Poisson's ratio.

[0030] Spatial coordinate mapping is achieved by constructing a bidirectional mapping function G to establish the spatial association between the two coordinates. This is applicable to any node or element integration point in the numerical model's computational mesh. G, through a spatial coordinate transformation algorithm, maps the observed displacement data to the unified coordinate system of the four-dimensional geometric evolution model of the slope, obtaining the corresponding spatial coordinates (x, y, z). This allows the extraction of observed displacement data from the four-dimensional geometric evolution model. Conversely, for any spatial point (x, y, z) in the four-dimensional geometric evolution model, the inverse mapping of G reveals its corresponding position in the numerical model grid, thus mapping observed displacement data to the numerical model grid. The mapping function must be constructed based on a unified geodetic coordinate system. Coordinate system differences are eliminated through coordinate transformation parameters, which are calibrated using several common control points to ensure a mapping error of less than 0.01m.

[0031] S104. Based on the data assimilation algorithm, the complete slope displacement observation field at the current monitoring time is fused with the predicted state of the rock mechanics numerical model by utilizing the spatial coordinate mapping relationship, and the initial mechanical parameter field is inverted and updated to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model.

[0032] To illustrate, Ensemble Kalman Filtering (EnKF) is used as the data assimilation algorithm to achieve bidirectional coupling between observational data and the numerical model. Specifically, it involves using the model state updated from the previous monitoring period. and parameter field As initial conditions, boundary conditions such as self-weight stress boundary and constraint boundary, as well as load conditions such as excavation load and rainfall infiltration load, are set for the numerical model. The rock mechanics numerical simulation software is then used for forward simulation until the current monitoring time t. k Receive forecast status This state reflects the current slope condition predicted by the model based on historical information when no new observation data is introduced.

[0033] For example, to achieve synchronous updates of state and parameters, an expanded state-parameter joint vector is constructed. ,in For the model state vector, This is the parameter vector. The joint vector for the forecast phase. In this approach, the state vector represents the current forecast result, while the parameter vector uses the optimal values ​​from the previous period's update, ensuring the continuity of parameter updates. To characterize model uncertainty, a forecast set containing N members is generated. The ensemble generation employs a parameter perturbation method, focusing on the forecast joint vector. For the model parameter vector By applying random perturbations that conform to a normal distribution, and by making small perturbations to boundary conditions and initial stress fields, we can ensure that each set member represents the possible true state and parameter combination of the model, and the overall distribution of the set can reflect the uncertainty range of the model.

[0034] An observation operator H is defined to adapt the model state to the observation data. The observation operator extracts the physical quantity, i.e., the displacement field, from the model state vector, which is consistent with the data type of the observation data. For the i-th member in the forecast set... Its state vector The numerical model contains displacement data for each node. The node displacements are extracted using the observation operator H to obtain the predicted displacement field for that member. Simultaneously, by utilizing spatial coordinate mapping relationships, the complete slope displacement observation field at the current monitoring moment is... Interpolation is mapped onto the nodes of the numerical model to form observation vectors. Based on the point cloud observation accuracy and the four-dimensional model fitting error, a statistical method is used to estimate the observation error covariance matrix. The diagonal elements of the matrix represent the variance of the observed displacements at each node, while the off-diagonal elements represent the covariance of the observed errors at different nodes.

[0035] The analysis step is the core of data assimilation. It updates the forecast ensemble by fusing observational information. Specifically, it calculates the mean of the forecast ensemble. Covariance Matrix The covariance matrix characterizes the degree of dispersion among members of the forecast ensemble, reflecting the uncertainty of the model state and parameters. The Kalman gain matrix is ​​then calculated. H T The Kalman gain matrix is ​​the transpose of the observation operator H. It balances the weights of prediction uncertainty and observation uncertainty. When the observation accuracy is high, i.e., R... k When the value is small, the Kalman gain is large, and the contribution of observed data to the update results is stronger; when the model prediction accuracy is high, that is... When the value is small, the Kalman gain is small, and the weight of the model prediction results is larger. Update each set member. ,in To obtain from the observation error distribution The perturbation vector, randomly selected from the set, is used to maintain the dispersion characteristics of the updated set and avoid excessive concentration of set members, which could lead to the invalidation of the uncertain representation. After the update is completed, the updated set is calculated. mean ,in The updated model state. This represents the optimal rock mass mechanical parameter field obtained through inversion.

[0036] S105. Using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, drive the rock mechanics numerical model to perform stability extrapolation and prediction for a preset future period, and obtain the future deformation field and stability index of the slope.

[0037] The optimal rock mass mechanical parameter field obtained by data assimilation and model state The rock mechanics numerical model is injected, and the digital twin kernel is updated to achieve high-fidelity simulation capabilities. Before initiating stability prediction, the boundary conditions and external load settings of the model are refined. If engineering activities such as excavation and support are involved during the prediction period, the spatiotemporal range and load characteristics of these activities must be accurately simulated in the model. If environmental loads such as rainfall and earthquakes are involved, meteorological forecast data and seismic activity background must be combined to convert them into load parameters that the model can recognize. These parameters are then applied to the model through load time history curves to ensure that the prediction process can accurately reflect the external forces acting on the slope in the future.

[0038] Furthermore, the numerical model is driven to perform forward extrapolation, calculating the mechanical response of the slope in future time periods to obtain the future deformation field. , T represents the preset future time period length and stress field distribution. The future deformation field can visually display the displacement evolution trend of the slope within the predicted time period, including the displacement magnitude, direction, and deformation concentration areas; the stress field distribution can be used to analyze the stress concentration inside the slope and identify potential failure initiation areas.

[0039] The above-mentioned dynamic assessment method for mine slope stability based on UAV lidar realizes the dynamic coupling of monitoring data and numerical model, effectively overcoming the problems of low utilization rate of monitoring data and difficulty in reflecting spatiotemporal heterogeneity and time-varying characteristics of rock mechanics parameters in traditional methods. It improves the dynamism, accuracy and reliability of slope stability assessment and prediction results, and meets the actual needs of dynamic risk assessment and early warning of slopes.

[0040] In one embodiment, such as Figure 2 As shown, a spatiotemporally continuous four-dimensional geometric evolution model of a slope is constructed based on a high-precision time-series point cloud sequence. Based on this model, the complete slope displacement observation field at the current monitoring moment is extracted, including: S201. Calculate the three-dimensional coordinate difference between each point in the high-precision time-series point cloud sequence and the corresponding point at the initial monitoring time to obtain the actual displacement vector of each point at the corresponding monitoring time. Use the spatial coordinates of each point and the monitoring time as sample inputs and the actual displacement vector as sample outputs to construct a training sample set.

[0041] Indicatively, based on the already registered high-precision time-series point cloud sequences, the mapping relationship between all observation points in each time-series point cloud and the corresponding points in the initial monitoring time point cloud is clarified. This mapping relationship is determined based on the unified coordinate system established in the previous fine registration through a spatial nearest neighbor search algorithm. That is, for any time-series point cloud P i Any observation point P in ij In the initial point cloud P0, find the nearest reference point P that satisfies the registration consistency requirement. 0j To ensure that the two points are observation points at the same physical location at different times, calculate the actual displacement vector of the observation point at the corresponding monitoring time. The calculation logic is to extract the observation points P respectively. ij Three-dimensional spatial coordinates With reference point P 0j Three-dimensional spatial coordinates Calculate the difference along each of the three-dimensional coordinate axes, i.e. This vector fully represents the observation point from the initial time to t. i The spatial location changes over time. When constructing the sample set, the spatial coordinates of each observation point are... With the corresponding monitoring time t i Combined as sample input vectors; the calculated actual displacement vectors As a sample output vector.

[0042] S202. Train the preset multilayer feedforward neural network model according to the training sample set, and optimize the network parameters by minimizing the preset loss function through the backpropagation algorithm to obtain the four-dimensional geometric evolution model of the slope; the four-dimensional geometric evolution model of the slope inputs three-dimensional spatial coordinates and normalized time and outputs a three-dimensional displacement vector.

[0043] The loss function is obtained using the following formula: in, Spatial coordinates; For monitoring time; It is a multi-layer feedforward neural network model; These are model parameters; This is the actual displacement vector; The square of the L2 norm; The regularization coefficient is used. To apply to model parameters The preset constraints.

[0044] Illustrative loss function for Where i represents the index of the time-series point cloud, and j represents the point cloud P in the i-th period. i Index of observation points; For the point cloud P in period i i The spatial coordinates of the j-th observation point; t i Let be the normalized time corresponding to the point cloud in period i; For the point cloud P in period i i The three-dimensional coordinate vector of the j-th observation point; For the initial point cloud P0 and The corresponding reference point coordinate vector, which is determined by the registered point cloud mapping relationship; Represents the square of the Euclidean distance; This is the regularization coefficient, used to balance the weights of the displacement error term and the regularization term; For neural network weight parameters The regularization term is L2 regularization, and the expression is: Its function is to limit the range of values ​​of the weight parameters, avoid the model from overfitting the training data, and ensure the smoothness of the deformation field in time and space.

[0045] During training, backpropagation is used to optimize parameters. This involves batch inputting input vectors from the training sample set into the model and calculating the predicted displacement vectors through forward propagation. The error between the predicted and actual displacement vectors is calculated based on the loss function. Using the chain rule, the error is backpropagated from the output layer to the input layer, and the gradients of each layer's parameters with respect to the loss function are calculated sequentially. A stochastic gradient descent (SGD) optimizer is used to update the model parameters. The learning rate is initially set to 0.001 and dynamically adjusted using a learning rate decay strategy to ensure parameter convergence and stability in the later stages of training. The training iteration termination condition is set when the loss function value on the validation set does not decrease for 100 consecutive iterations, or when the total number of iterations reaches 3000-5000. When the training meets the termination condition, the current optimal model parameters are saved. This yields a four-dimensional geometric evolution model of the slope with spatiotemporal mapping capabilities.

[0046] S203. Based on the four-dimensional geometric evolution model of the slope, query the current monitoring time to obtain a complete slope displacement observation field covering the entire slope space.

[0047] As an illustration, the time information of the current monitoring moment is converted into normalized time parameters in the interval [0,1] according to the normalization rules used during model training. To achieve displacement field coverage of the entire slope space, a three-dimensional computational grid was constructed using a structured grid generation method based on the three-dimensional spatial range of the slope determined by the initial point cloud. The grid points needed to completely cover the main area of ​​the slope and a buffer zone within a 10-meter radius to ensure no blind spots in displacement observation. Furthermore, the three-dimensional spatial coordinates (x, y, z) of each generated grid point were compared with the normalized current monitoring time. The vectors are combined to form the model query input vectors, which are then batch-input into the trained four-dimensional geometric evolution model of the slope. The model calculates through forward propagation, outputting a three-dimensional displacement vector corresponding to each grid point. This vector directly represents the displacement state of that spatial point relative to the initial time at the current monitoring moment. The displacement vectors of all grid points are integrated to form a complete slope displacement observation field based on a three-dimensional grid.

[0048] In one embodiment, such as Figure 3 As shown, a rock mechanics numerical model of a mine slope is established based on the slope geometry and geological information at the initial monitoring time in a high-precision time-series point cloud sequence. Furthermore, a spatial coordinate mapping relationship is established between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope, including: S301. Based on the slope point cloud data at the initial monitoring time and the reconstructed three-dimensional model of the slope surface, and combined with borehole-profile geological data, a three-dimensional geological entity model is constructed; the three-dimensional geological entity model includes the main strata, lithological boundaries and dominant structural planes.

[0049] Based on high-precision slope point cloud data at the initial monitoring time, a three-dimensional surface reconstruction algorithm is used to construct a three-dimensional model of the slope surface. Subsequently, borehole-profile geological data are integrated. The borehole data must include the borehole location coordinates, borehole depth, stratigraphic thickness, lithological name, and lithological description for each borehole. By accurately mapping the borehole location coordinates to the corresponding positions on the surface three-dimensional model, and using the borehole depth data as a vertical constraint, the burial depth and vertical distribution pattern of each stratum are determined. The profile geological data includes geological profile maps of different orientations of the slope. By spatially aligning the topographic lines and stratigraphic boundaries of the profile maps with the surface three-dimensional model and borehole data, the lateral extension range of the strata and the spatial morphology of lithological boundaries are supplemented, thus achieving three-dimensional fusion of borehole data and profile data. The identification and characterization of dominant structural surfaces can be based on field joint and fracture survey data and structural surface descriptions in geological exploration reports. Structural surfaces that control slope stability, i.e., dominant structural surfaces, are selected. Screening indicators include the orientation stability, extension length, spacing, and angle relationship between the structural surfaces and the slope surface. Through spatial geometric modeling, parameters such as the direction, dip, and dip angle of the dominant structural surfaces are accurately integrated into the three-dimensional geological entity model to ensure that the spatial location of the structural surfaces is consistent with the actual geological conditions. Finally, a complete three-dimensional geological entity model containing the main stratigraphic distribution, clear lithological boundaries, and dominant structural surfaces is formed.

[0050] S302. Mesh the three-dimensional geological entity model, and assign corresponding constitutive models of elastic-plastic and joint materials to different lithological regions and structural planes according to lithological boundaries and dominant structural planes, to obtain the rock mass mechanics numerical model.

[0051] Optionally, the mesh size can be differentiated for different regions. For example, denser meshes with a mesh size of 1-3m are used in areas with drastic lithological changes, near dominant structural planes, and areas with concentrated potential slope deformation to accurately capture local mechanical responses. In stable areas with homogeneous lithology and far from structural planes, the mesh size can be appropriately increased to 3-5m to improve computational efficiency while ensuring accuracy. The mesh type should be either tetrahedral unstructured mesh or hexahedral structured mesh. Hexahedral meshes are preferred for large, homogeneous lithological areas, while tetrahedral meshes are used for areas with complex topography or dense structural planes. After mesh generation, it must be ensured that the mesh completely matches the lithological boundaries and dominant structural planes, with no cross-boundary meshes.

[0052] Furthermore, for various rock mass regions, corresponding elastoplastic constitutive models are assigned according to lithological types, such as granite, sandstone, and shale, to reflect the elastic deformation stage and mechanical response after plastic yielding of the rock mass. For dominant structural planes, constitutive models of joint materials are assigned, specifically considering the normal compression and shear slip characteristics of the structural planes. By setting the normal stiffness, shear stiffness, and shear strength parameters of the structural planes, the influence of the structural planes on the overall mechanical behavior of the slope is simulated. Finally, through the combination of mesh and constitutive models, a rock mechanics numerical model that can truly reflect the geomechanical characteristics of the slope is formed.

[0053] S303. Based on regional geological data, in-situ test data, and indoor rock sample test results, an initial value is set for the empirical parameters of different material regions in the rock mechanics numerical model to obtain the initial mechanical parameter field; the empirical parameters include cohesion, internal friction angle, elastic modulus, and Poisson's ratio.

[0054] Indicatively, the initial values ​​of empirical parameters should primarily be set using in-situ test data and laboratory rock sample test results, with regional geological data serving as supplementary verification. In-situ tests should be conducted for different lithological regions, including plate load tests to obtain the elastic modulus and in-situ direct shear tests to obtain cohesion and internal friction angle. Laboratory rock sample tests should select representative rock samples from each lithological region and perform uniaxial compressive strength tests, triaxial shear tests, and elastic wave tests to obtain parameters such as elastic modulus, Poisson's ratio, cohesion, and internal friction angle. Regional geological data is mainly used for comparative analysis of parameters of similar lithologies. If in-situ test data is insufficient, recommended parameter ranges from geological reports of similar mines in the surrounding area or relevant industry standards can be referenced and corrected based on the weathering degree and integrity coefficient of the slope rock mass.

[0055] Furthermore, during the parameter assignment process, different material regions are assigned values ​​in zones based on lithological boundaries and structural plane distribution. Hard lithological regions are assigned higher elastic modulus and strength parameters, while soft lithological regions are assigned lower elastic modulus and strength parameters. Poisson's ratio is set to 0.2-0.35 according to the density of the lithology. The parameters of the dominant structural planes are set independently, with their strength parameters being significantly lower than those of the surrounding rock mass, and their elastic modulus being 1 / 5-1 / 10 of that of the surrounding rock mass. This zoned assignment forms an initial mechanical parameter field with a spatially uneven distribution.

[0056] S304. Based on the index relationship between the spatial coordinates of all calculation nodes in the rock mechanics numerical model and the spatial domain defined by the four-dimensional geometric evolution model of the slope, the spatial coordinate mapping relationship can be obtained by using coordinate interpolation to obtain the displacement observation value from the four-dimensional geometric evolution model of the slope for any calculation node.

[0057] First, the spatial coordinates of the computational nodes in the rock mechanics numerical model and the spatial domain of the slope four-dimensional geometric evolution model are unified to the same geodetic coordinate system. Specifically, a spatial index relationship between the computational nodes and the spatial domain of the four-dimensional geometric evolution model is constructed. The KD-tree indexing algorithm is used to divide the spatial domain of the four-dimensional geometric evolution model, establishing an efficient spatial retrieval structure. This allows for the rapid location of any computational node in the rock mechanics numerical model within the spatial domain of the four-dimensional geometric evolution model. Linear interpolation or cubic spline interpolation is used for coordinate interpolation. When a computational node is located exactly on a grid node of the four-dimensional geometric evolution model, its displacement observation value is directly extracted. When a computational node is located inside a grid cell, its displacement observation value is calculated through interpolation based on the displacement observation values ​​of the four or eight vertices of that grid cell. Through this process, a precise spatial association is achieved between the computational nodes of the rock mechanics numerical model and the slope four-dimensional geometric evolution model. Any computational node can quickly obtain its corresponding displacement observation value through this mapping relationship.

[0058] In one embodiment, based on a data assimilation algorithm, the complete slope displacement observation field at the current monitoring time is fused with the predicted state of the rock mechanics numerical model using spatial coordinate mapping relationships. The initial mechanical parameter field is then inverted and updated to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model, including: S11. Combine the current model state variables and the current rock mechanics parameter field of the rock mechanics numerical model to obtain the amplified state vector. In the first assimilation cycle, the current model state variables and the current rock mechanics parameter field are the initial model state and the initial mechanical parameter field, respectively. In subsequent assimilation cycles, the current model state variables and the current rock mechanics parameter field are the rock mechanics numerical model state and the updated rock mechanics parameter field after the previous assimilation cycle.

[0059] Indicatively, the current model state variables include all dynamic response parameters of the rock mechanics numerical model at the current moment, specifically covering physical quantities such as displacement, velocity, stress, and plastic strain of each calculation node. These state variables are vectorized according to the order of the calculation nodes to form a one-dimensional state vector X. The current rock mechanics parameter field includes spatially distributed parameters such as cohesion, internal friction angle, elastic modulus, and Poisson's ratio. These are also vectorized according to the order of model elements or calculation nodes to form a one-dimensional parameter vector. The expanded state vector Z is constructed by vector concatenation, that is... Here, T represents the transpose of the vector, ensuring that the state variables and parameter fields form a continuous numerical sequence in the amplified vector. In the first assimilation cycle, since no updates or iterations have been performed, the current model state variables adopt the initial model state of the rock mechanics numerical model, obtained from model initialization calculations, including initial stress fields, initial displacements being zero, and other initial conditions. The current rock mechanics parameter field directly uses the initial mechanical parameter field established in the previous cycle. In subsequent assimilation cycles, both the current model state variables and the current rock mechanics parameter field adopt the updated results output from the previous assimilation cycle, ensuring the temporal continuity of the assimilation process and the cumulative effect of parameter iterations, so that the amplified state vector can truly reflect the current cognitive state of the model.

[0060] S12. Using the amplified state vector as the initial condition, perform mechanical calculations for a monitoring period using the rock mechanics numerical model to obtain the state variables of the prediction model. Combine the state variables of the prediction model with the unchanged parameter field to obtain the predicted amplified state vector.

[0061] Using the constructed amplified state vector as input, the state vector portion is taken as the initial state, and the parameter vector portion is taken as the parameter basis for mechanical calculation. These are substituted into the rock mechanics numerical model to perform forward mechanical calculations. During the calculation process, the stability of boundary conditions and external loads must be maintained, and the model evolution is driven only by the initial state and parameter field. The core of the mechanical calculation is based on the rock mass constitutive model and the mechanical properties of structural surfaces. The equilibrium equations of the rock mass are solved numerically and iteratively to obtain the dynamic response of the model after one monitoring cycle, i.e., the predicted model state variables. This variable contains updated state information such as displacement, stress, and velocity at each calculation node at the predicted time. Since the physical changes in the parameter field are relatively slow within a monitoring period, and its updates are achieved through subsequent assimilation processes, the predicted amplified state vector... The construction adopts a combination of the predicted state and the original parameter field, that is... ,in To expand the unchanged rock mechanics parameter field vector in the state vector, and to ensure the stability of the parameter field during the prediction stage.

[0062] S13. By using spatial coordinate mapping relationships, determine the corresponding positions of each calculation node of the rock mechanics numerical model in the four-dimensional geometric evolution model of the slope, and query the displacement observation values ​​of each calculation node at the current monitoring time to obtain the observed displacement vector.

[0063] For each computation node in the numerical model, its spatial coordinates in the model coordinate system are converted into coordinates in the unified coordinate system of the four-dimensional geometric evolution model of the slope through the forward operation of the mapping function, so as to accurately locate the corresponding position of the computation node in the four-dimensional model space domain.

[0064] S14. Using the ensemble Kalman filter algorithm, the predicted amplified state vector and the observed displacement vector are fused to obtain a state vector set, and the updated rock mechanics parameter field and rock mechanics numerical model state are obtained by updating the state vector set.

[0065] The corresponding steps for updating are as follows: The amplified state vector can be obtained using the following formula: in, Let i be the predicted augmented state vector of the i-th state vector. To observe the displacement vector, For the random perturbation applied to the observed displacement vector, To extract the observation operators corresponding to the computational node displacements from the state vector, The Kalman gain matrix is ​​calculated based on the covariance of the forecast set and the covariance of the observation error. The analysis expands the state vector to represent the updated i-th state vector; Calculate the mean of all analyzed amplified state vector sets, and use the distribution of the parameter field part and the model state variable part as the updated rock mechanics parameter field and rock mechanics numerical model state.

[0066] First, a set of predicted amplification state vectors is generated, and then... By adjusting the parameter field vector Apply a normally distributed random perturbation to generate a forecast set. According to the observation operator For the predicted state vector in the set members By filtering and extracting, the displacement prediction vector corresponding to each member is obtained. The above update formula is applied to each member of the forecast set to obtain N updated analytical augmentation state vector members, which constitute the analytical set. Then, the mean of the analysis set is calculated, that is, the arithmetic mean of the corresponding elements of each member is calculated, to obtain the mean of the analysis amplified state vector. The mean vector is split according to the previously set splicing rules, and the part corresponding to the original parameter field vector is the updated rock mechanics parameter field. The part corresponding to the original state vector is the updated state of the rock mechanics numerical model. .

[0067] In one embodiment, using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, the rock mechanics numerical model is driven to perform stability prediction for a preset future time period, obtaining the future deformation field and stability indicators of the slope, including: S21. Replace the values ​​of the corresponding material regions or structural surfaces in the rock mechanics numerical model with the values ​​of each parameter in the rock mechanics parameter field to obtain the rock mechanics numerical model with updated parameters.

[0068] The parameter replacement process adopts the principle of precise matching by region, which clarifies the unique identifier of each material region and dominant structural surface in the rock mechanics numerical model, and replaces the original parameter settings in the model one by one with the parameter values ​​of the corresponding identified regions in the updated rock mechanics parameter field.

[0069] S22. Set the state of the rock mechanics numerical model as the initial condition for prediction calculation of the rock mechanics numerical model; the state of the rock mechanics numerical model includes displacement field, stress field and plastic state variables.

[0070] Specifically, displacement field data is directly applied as the initial displacement to the corresponding calculation node, replacing the default zero initial displacement in the prediction calculation; stress field data is loaded into the model as the initial stress field to ensure that the stress state at the start of the prediction calculation is consistent with the actual stress level of the current slope, avoiding distortion of the prediction results due to initial stress deviation; plastic state variables are assigned values ​​one by one according to the model unit, so that the model can inherit the cumulative effect of plastic damage in the early stage and accurately reflect the mechanical response law of the rock mass in the subsequent stress process.

[0071] S23. Set the load changes and / or boundary condition changes for a future preset time period on the rock mechanics numerical model after parameter update, and drive the rock mechanics numerical model to perform numerical simulation calculations based on the initial conditions to obtain the numerical simulation calculation results.

[0072] Based on the updated parameters and initial conditions, the rock mechanics numerical model is driven to perform forward numerical simulation calculations. The calculation process adopts a numerical iterative algorithm that matches the model constitutive structure. The time step is set according to the degree of load change and the model stability requirements to ensure that the calculation converges and the accuracy meets the requirements. Finally, the numerical simulation calculation results cover the model state at each time point within the prediction period.

[0073] S24. Extract the slope displacement field, maximum shear stress distribution area, and plastic strain cloud map for a future preset time period from the numerical simulation calculation results to obtain the future deformation field.

[0074] In a schematic manner, slope displacement field data at preset time intervals are extracted from the calculation results for each moment, including the spatial distribution and magnitude of the three-dimensional displacement components, forming a time-series displacement field set. Simultaneously, the maximum shear stress distribution area data for each moment is extracted to locate the spatial position, range, and peak value of shear stress concentration, clarifying the potential stress failure initiation area within the slope. Furthermore, plastic strain cloud map data is extracted to obtain the spatial distribution, accumulation degree, and development trend of plastic strain, intuitively reflecting the damage evolution state of the rock mass. The extracted displacement field, maximum shear stress distribution area, and plastic strain cloud map are integrated to construct a complete future deformation field. The displacement field reflects the overall deformation trend, the maximum shear stress distribution area reveals the mechanical driving factors of deformation, and the plastic strain cloud map reflects the rock mass damage caused by deformation. These three elements corroborate each other, comprehensively characterizing the deformation evolution law of the slope in the future period.

[0075] S25. Based on the stress state and updated rock mechanics parameter field obtained from numerical simulation, the overall safety factor of the slope under the current and future conditions is calculated using the strength reduction method, and the stability index is obtained.

[0076] Simultaneously, the stability indices of the slope are calculated, with core indices including the safety factor (FoS) and the development of plastic strain. The safety factor is calculated using the strength reduction method, by gradually reducing the shear strength parameters of the rock mass and structural surfaces until the model reaches a limit equilibrium state. The strength reduction factor at this point is the safety factor. ,in This represents the actual shear strength of the rock mass. The shear strength required to maintain slope stability is determined by a safety factor of less than 1.2, indicating that the slope is in an unstable state, and less than 1.0, indicating that the slope has failed. The development of plastic strain is determined by analyzing the distribution range, thickness, and penetration of the plastic zone in the model.

[0077] In one embodiment, the raw scanning data of the UAV lidar under each time-series monitoring period are processed and registered to obtain a high-precision time-series point cloud sequence in the same coordinate system, including: S31. Solve the raw scanning data of UAV lidar under each time series monitoring period to obtain the initial three-dimensional point cloud under each time series monitoring period, and perform outlier removal and statistical filtering on the initial three-dimensional point cloud under each time series monitoring period to obtain the standard three-dimensional point cloud under each time series monitoring period.

[0078] The raw scanning data from UAV lidar includes laser echo signals, GNSS positioning data, and IMU inertial measurement data. The calculation process requires a multi-source data fusion algorithm to accurately determine the three-dimensional coordinates of the laser footprints. Specifically, GNSS positioning data, IMU attitude data, and laser echo data are aligned by timestamp to eliminate spatiotemporal asynchrony errors during data acquisition. Based on the lidar ranging principle, and combining the real-time attitude parameters provided by the IMU and the position parameters provided by the GNSS, the laser footprints are transformed from the local radar coordinate system to the geodetic coordinate system using a spatial coordinate transformation formula, resulting in the initial three-dimensional point cloud for each time-series monitoring period.

[0079] Furthermore, outlier removal employs a radius filtering algorithm. A reasonable radius threshold and minimum number of neighboring points are set. Neighborhood analysis is performed on each point in the initial 3D point cloud. If a point has fewer neighboring points within the set radius than the minimum number of neighboring points, it is identified as an outlier and removed to eliminate isolated noise points generated by dynamic interference such as birds and dust during flight. Statistical filtering, based on the spatial distribution characteristics of the point cloud, calculates the average distance between each point and its K nearest neighbors, thus obtaining the mean and standard deviation of the average distances of all points. A filtering threshold is set to the mean plus three times the standard deviation. Points with a distance greater than this threshold are identified as noise points and removed, further purifying the point cloud data.

[0080] S32. Using the standard three-dimensional point cloud of the first time-series monitoring cycle as a fixed reference benchmark, and using artificially placed targets or natural feature points that can be stably identified in each time-series monitoring cycle as control points, the standard three-dimensional point cloud of each time-series monitoring cycle is initially registered to the coordinate system of the fixed reference benchmark through rigid body transformation, so as to obtain the preliminary spatiotemporal registration three-dimensional point cloud of each time-series monitoring cycle.

[0081] In a schematic manner, a standard 3D point cloud from the first time-series monitoring period is selected as a fixed reference benchmark, with its coordinate system adopting the local independent geodetic coordinate system. The selection of control points must meet the requirements of stability and recognizability. Among them, artificially deployed targets are high-reflectivity spherical targets, which are evenly distributed throughout the entire slope area, with no less than 3 targets within each 1000-square-meter area. The target positions must avoid obstructed areas and easily deformable areas to ensure that they can be stably identified by the lidar in all time-series monitoring periods. Natural feature points are automatically extracted from the point cloud using a corner detection algorithm, selecting points with stable spatial morphology and high grayscale contrast, such as rock edges and corners of fixed structures. After extraction, they must be verified through multiple time series to ensure that their recognition success rate in the point cloud of each period is ≥95%.

[0082] Furthermore, corresponding control points are identified in the standard 3D point cloud to be registered and the reference datum point cloud, respectively, and the correspondence between control points is established. An error equation is constructed based on the least squares method. With the goal of minimizing the coordinate deviation of the control points in the two coordinate systems, the rotation matrix R and translation vector T of the rigid body transformation are solved. The coordinates of all points in the standard 3D point cloud to be registered are then substituted into the transformation formula. ,in The original coordinates of the points to be registered. To transform the coordinates, preliminary registration is completed. After registration, the registration error of the control points needs to be calculated to ensure that the average error is less than 0.1m and the maximum error is less than 0.3m, meeting the accuracy requirements of preliminary spatiotemporal registration, and obtaining the preliminary spatiotemporal registration 3D point cloud for each time series monitoring period.

[0083] S33. Based on the preliminary spatiotemporal registration three-dimensional point clouds under each time series monitoring period, the iterative nearest point algorithm is used to perform fine registration on the preliminary spatiotemporal registration three-dimensional point clouds of two adjacent time series monitoring periods to obtain a high-precision time series point cloud sequence. Fine registration corresponds to optimizing the registration transformation parameters by minimizing the distance error between corresponding points in the preliminary spatiotemporal registration three-dimensional point clouds, and performing registration according to the optimized registration transformation parameters.

[0084] Fine registration takes the preliminary spatiotemporal registration of 3D point clouds from two adjacent time-series monitoring periods as the processing object, and assumes the previous time-series point cloud as the source point cloud P. s The subsequent temporal point cloud is the target point cloud P. t The core of the Iterative Closest Point Algorithm is to continuously refine the registration transformation parameters through iterative optimization, thereby reducing the distance error between two point clouds. Specifically, from the source point cloud P... s A certain number of sampling points are randomly selected from the target point cloud P, and the KD-tree nearest neighbor search algorithm is used to search for the target point cloud P. t Find the nearest point for each sampling point and establish an initial corresponding point pair. Based on the initial corresponding point pairs, use the Singular Value Decomposition (SVD) method to solve for the optimal rotation matrix. Translation vector To minimize the sum of squared Euclidean distances between corresponding pairs of points, i.e., to minimize the objective function. , where P s，i For the source point cloud sampling points, P t，i Let n be the nearest point to the target point cloud, and n be the number of sampling points.

[0085] After each iteration, the average distance error of all corresponding point pairs is calculated. If the error exceeds a set threshold or the number of iterations has not reached the maximum, sampling points are reselected and the above process is repeated until the error meets the threshold requirement or the maximum number of iterations is reached. At this point, the optimized registration transformation parameters are obtained. The optimized transformation parameters are then applied to the entire source point cloud P. s This completes the precise registration of two adjacent temporal point clouds.

[0086] In one embodiment, the method further includes: S41. In response to receiving the new batch of raw scanning data from the UAV lidar collected in the next time-series monitoring cycle, the raw scanning data from the UAV lidar is processed for settlement and registration to obtain an updated high-precision time-series point cloud sequence.

[0087] The data processing follows the same multi-source data fusion strategy as previous periods to obtain the initial 3D point cloud for the new period, followed by standardized filtering and registration processing.

[0088] S42. Using the updated high-precision time-series point cloud sequence, the four-dimensional geometric evolution model of the slope is incrementally trained, and the complete slope displacement observation field at the current monitoring time is extracted.

[0089] Based on the updated high-precision time-series point cloud sequence, the actual displacement vector of each observation point in the new point cloud relative to the corresponding point at the initial time is calculated to form a new training sample set. The incremental training of the slope four-dimensional geometric evolution model adopts the original parameter fine-tuning mode, loading the model parameters Φ obtained from the previous training as the initial weights, and only combining the new training samples with some original representative samples to form an incremental training set, so as to reduce the amount of computation and retain the deformation rules already learned by the model.

[0090] S43. Using the updated rock mechanics parameter field and rock mechanics numerical model state obtained from the previous time series monitoring period as new initial conditions, perform inversion updates and stability predictions to obtain new future deformation fields and stability indices.

[0091] The updated rock mechanics parameter field and rock mechanics numerical model state, obtained by assimilating data from the previous time-series monitoring period, are directly used as the new initial conditions for this inversion update. This eliminates the need to reconstruct the initial mechanical parameter field and initial model state, ensuring the continuity of model evolution. The new displacement observation field extracted from the updated high-precision time-series point cloud sequence serves as the observation data. Based on the new updated parameter field and model state, load changes and boundary conditions are adjusted according to the latest engineering planning and environmental prediction data to drive the rock mechanics numerical model for forward simulation calculations. New future deformation fields are extracted from the calculation results, and a new overall safety factor is calculated using the strength reduction method, forming a new stability index.

[0092] 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.

[0093] Based on the same inventive concept, this application also provides a device for dynamically assessing the stability of mine slopes based on UAV lidar, used to implement the aforementioned method for dynamically assessing the stability of mine slopes based on UAV lidar. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for dynamically assessing the stability of mine slopes based on UAV lidar provided below can be found in the limitations of the method for dynamically assessing the stability of mine slopes based on UAV lidar described above, and will not be repeated here.

[0094] In one exemplary embodiment, such as Figure 4 As shown, a dynamic assessment device for mine slope stability based on UAV lidar is provided, comprising: The point cloud data module 401 is used to acquire the raw scanning data of UAV lidar under multiple time-series monitoring periods of the target slope, and to perform calculation and registration processing on the raw scanning data of UAV lidar under each time-series monitoring period to obtain a high-precision time-series point cloud sequence under the same coordinate system. The slope four-dimensional continuous deformation field modeling module 402 is used to construct a spatiotemporally continuous slope four-dimensional geometric evolution model based on a high-precision time-series point cloud sequence, and to extract the complete slope displacement observation field at the current monitoring time based on the slope four-dimensional geometric evolution model. The rock mechanics module 403 is used to establish a rock mechanics numerical model of a mine slope based on the slope geometry and geological information at the initial monitoring time in a high-precision time-series point cloud sequence, and to establish the spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope; the rock mechanics numerical model includes the initial mechanical parameter field; The inversion update module 404 is used to integrate the complete slope displacement observation field at the current monitoring time with the predicted state of the rock mechanics numerical model based on the data assimilation algorithm and the spatial coordinate mapping relationship, and to invert and update the initial mechanical parameter field to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model. The fusion prediction module 405 is used to drive the rock mechanics numerical model to perform stability extrapolation and prediction for a preset future period, using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, so as to obtain the future deformation field and stability index of the slope.

[0095] In one embodiment, the slope four-dimensional continuous deformation field modeling module 402 is also used for: Calculate the three-dimensional coordinate difference between each point in the high-precision time-series point cloud sequence and the corresponding point at the initial monitoring time to obtain the actual displacement vector of each point at the corresponding monitoring time. Use the spatial coordinates of each point and the monitoring time as sample inputs and the actual displacement vector as sample outputs to construct a training sample set. The pre-set multi-layer feedforward neural network model is trained based on the training sample set, and the network parameters are optimized by minimizing the pre-set loss function through the backpropagation algorithm to obtain the four-dimensional geometric evolution model of the slope. By querying the current monitoring time based on the four-dimensional geometric evolution model of the slope, a complete slope displacement observation field covering the entire slope space can be obtained.

[0096] In one embodiment, the rock mechanics module 403 is further configured to: Based on the slope point cloud data at the initial monitoring time and the reconstructed three-dimensional model of the slope surface, combined with borehole-profile geological data, a three-dimensional geological entity model was constructed. The three-dimensional geological entity model is meshed, and constitutive models of elastic-plastic and joint materials are assigned to different lithological regions and structural planes according to lithological boundaries and dominant structural planes, so as to obtain a numerical model of rock mechanics. Based on regional geological data, in-situ test data, and indoor rock sample test results, an initial value is set for the empirical parameters of different material regions in the rock mechanics numerical model to obtain the initial mechanical parameter field. Based on the index relationship between the spatial coordinates of all calculation nodes in the rock mechanics numerical model and the spatial domain defined by the four-dimensional geometric evolution model of the slope, the spatial coordinate mapping relationship can be obtained by using coordinate interpolation to obtain the displacement observation value from the four-dimensional geometric evolution model of the slope for any calculation node.

[0097] In one embodiment, the inversion update module 404 is further configured to: The current model state variables of the rock mechanics numerical model are combined with the current rock mechanics parameter field to obtain the amplified state vector; Using the amplified state vector as the initial condition, a mechanical calculation for a monitoring period is performed using a rock mechanics numerical model to obtain the state variables of the prediction model. The state variables of the prediction model are then combined with the unchanged parameter field to obtain the predicted amplified state vector. By using spatial coordinate mapping relationships, the corresponding positions of each calculation node of the rock mechanics numerical model in the four-dimensional geometric evolution model of the slope are determined, and the displacement observation values ​​of each calculation node at the current monitoring time are queried to obtain the observed displacement vector. An ensemble Kalman filter algorithm is used to fuse the predicted amplified state vector with the observed displacement vector to obtain a state vector set. The updated rock mechanics parameter field and the state of the rock mechanics numerical model are then obtained based on the state vector set.

[0098] In one embodiment, the fusion prediction module 405 is further configured to: The updated rock mechanics numerical model is obtained by replacing the values ​​of the corresponding material regions or structural surfaces in the rock mechanics parameter field with the values ​​of each parameter. The state of the rock mechanics numerical model is set as the initial condition for the prediction calculation of the rock mechanics numerical model; the state of the rock mechanics numerical model includes displacement field, stress field and plastic state variables; On the rock mechanics numerical model with updated parameters, load changes and / or boundary condition changes for a future preset time period are set, and the rock mechanics numerical model is driven to perform numerical simulation calculations based on the initial conditions to obtain numerical simulation calculation results. The slope displacement field, maximum shear stress distribution area, and plastic strain cloud map for a future preset time period are extracted from the numerical simulation results to obtain the future deformation field; Based on the stress state and updated rock mechanics parameter field obtained from numerical simulation, the overall safety factor of the slope under the current and future conditions is calculated using the strength reduction method, thus obtaining the stability index.

[0099] In one embodiment, the point cloud data module 401 is further configured to: The raw scanning data of UAV lidar under each time series monitoring period are processed to obtain the initial three-dimensional point cloud under each time series monitoring period. Outlier removal and statistical filtering are performed on the initial three-dimensional point cloud under each time series monitoring period to obtain the standard three-dimensional point cloud under each time series monitoring period. Using the standard 3D point cloud of the first time-series monitoring cycle as a fixed reference benchmark, and using artificially placed targets or natural feature points that can be stably identified in each time-series monitoring cycle as control points, the standard 3D point cloud of each time-series monitoring cycle is initially registered to the coordinate system of the fixed reference benchmark through rigid body transformation, thus obtaining the preliminary spatiotemporal registration 3D point cloud of each time-series monitoring cycle. Based on the preliminary spatiotemporal registration of 3D point clouds under each time series monitoring period, the iterative nearest point algorithm is used to perform fine registration of the preliminary spatiotemporal registration of 3D point clouds between two adjacent time series monitoring periods, resulting in a high-precision time series point cloud sequence.

[0100] In one embodiment, an iterative update module is also included, for: In response to receiving the new batch of raw UAV lidar scanning data collected in the next time-series monitoring cycle, the raw UAV lidar scanning data is calculated and registered to obtain an updated high-precision time-series point cloud sequence. The updated high-precision time-series point cloud sequence was used to incrementally train the four-dimensional geometric evolution model of the slope, and the complete slope displacement observation field at the current monitoring time was extracted. The updated rock mechanics parameter field and rock mechanics numerical model state obtained from the previous time series monitoring period are used as new initial conditions for inversion update and stability prediction to obtain new future deformation field and stability index.

[0101] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps in the above method embodiments.

[0102] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0103] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0104] 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 dynamic assessment method for mine slope stability based on UAV lidar, characterized in that, The method includes: The raw scanning data of the UAV lidar of the target slope under multiple time-series monitoring periods were acquired, and the raw scanning data of the UAV lidar under each time-series monitoring period were solved and registered to obtain a high-precision time-series point cloud sequence under the same coordinate system. A spatiotemporally continuous four-dimensional geometric evolution model of the slope is constructed based on the high-precision time-series point cloud sequence, and the complete slope displacement observation field at the current monitoring time is extracted based on the four-dimensional geometric evolution model of the slope. Based on the slope geometry and geological information at the initial monitoring time in the high-precision time-series point cloud sequence, a rock mechanics numerical model of the mine slope is established, and the spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope is established; the rock mechanics numerical model includes an initial mechanical parameter field; Based on the data assimilation algorithm, the complete slope displacement observation field at the current monitoring time is fused with the predicted state of the rock mechanics numerical model using the spatial coordinate mapping relationship, and the initial mechanical parameter field is inverted and updated to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model. Using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, the rock mechanics numerical model is driven to perform stability extrapolation and prediction for a preset future time period, thereby obtaining the future deformation field and stability index of the slope.

2. The method according to claim 1, characterized in that, The process involves constructing a spatiotemporally continuous four-dimensional geometric evolution model of the slope based on the high-precision time-series point cloud sequence, and extracting the complete slope displacement observation field at the current monitoring time based on the four-dimensional geometric evolution model of the slope, including: Calculate the three-dimensional coordinate difference between each point in the high-precision time-series point cloud sequence and the corresponding point at the initial monitoring time to obtain the actual displacement vector of each point at the corresponding monitoring time. Use the spatial coordinates of each point and the monitoring time as sample inputs and the actual displacement vector as sample outputs to construct a training sample set. The preset multilayer feedforward neural network model is trained based on the training sample set, and the preset loss function is minimized through the backpropagation algorithm to optimize the network parameters, thereby obtaining the four-dimensional geometric evolution model of the slope; the four-dimensional geometric evolution model of the slope is input with three-dimensional spatial coordinates and normalized time, and outputs a three-dimensional displacement vector. The loss function is obtained using the following formula: in, Spatial coordinates; For monitoring time; It is a multi-layer feedforward neural network model; These are model parameters; This is the actual displacement vector; The square of the L2 norm; The regularization coefficient is used. To apply to model parameters Preset constraints; Based on the slope four-dimensional geometric evolution model, the current monitoring time is queried to obtain the complete slope displacement observation field covering the entire slope space.

3. The method according to claim 2, characterized in that, The process of establishing a rock mechanics numerical model of the mine slope based on the slope geometry and geological information at the initial monitoring time in the high-precision time-series point cloud sequence, and establishing the spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope, includes: Based on the slope point cloud data at the initial monitoring time and the reconstructed three-dimensional model of the slope surface, and combined with borehole-profile geological data, a three-dimensional geological entity model is constructed; the three-dimensional geological entity model includes the main strata, lithological boundaries and dominant structural planes; The three-dimensional geological entity model is meshed, and constitutive models of elastic-plastic and joint materials are assigned to different lithological regions and structural planes according to the lithological boundaries and dominant structural planes, so as to obtain the rock mass mechanics numerical model. Based on regional geological data, in-situ test data, and indoor rock sample test results, an initial value is set for the empirical parameters of different material regions in the rock mechanics numerical model, thus obtaining the initial mechanical parameter field; the empirical parameters include cohesion, internal friction angle, elastic modulus, and Poisson's ratio; Based on the index relationship between the spatial coordinates of all calculation nodes in the rock mechanics numerical model and the spatial domain defined by the slope four-dimensional geometric evolution model, the spatial coordinate mapping relationship can be obtained when the displacement observation value can be obtained from the slope four-dimensional geometric evolution model through coordinate interpolation for any calculation node.

4. The method according to claim 3, characterized in that, The data assimilation algorithm, utilizing the spatial coordinate mapping relationship, fuses the complete slope displacement observation field at the current monitoring time with the predicted state of the rock mechanics numerical model, and inverts and updates the initial mechanical parameter field to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model, including: The current model state variables and the current rock mechanics parameter field of the rock mechanics numerical model are combined to obtain the amplified state vector; wherein, in the first assimilation cycle, the current model state variables and the current rock mechanics parameter field are the initial model state and the initial mechanical parameter field, respectively; in subsequent assimilation cycles, the current model state variables and the current rock mechanics parameter field are the rock mechanics numerical model state and the updated rock mechanics parameter field after the previous assimilation cycle. Using the amplified state vector as the initial condition, the rock mechanics numerical model is used to perform mechanical calculations for a monitoring period to obtain the prediction model state variables. The prediction model state variables are then combined with the unchanged parameter field to obtain the prediction amplified state vector. By using the spatial coordinate mapping relationship, the corresponding positions of each calculation node of the rock mechanics numerical model in the four-dimensional geometric evolution model of the slope are determined, and the displacement observation values ​​of each calculation node at the current monitoring time are queried to obtain the observed displacement vector. An ensemble Kalman filter algorithm is used to fuse the predicted amplified state vector with the observed displacement vector to obtain a state vector set, and the updated rock mass mechanical parameter field and the state of the rock mass mechanical numerical model are obtained by updating the state vector set. The update corresponds to the following steps: The amplified state vector can be obtained using the following formula: in, Let i be the predicted augmented state vector of the i-th state vector. To observe the displacement vector, For the random perturbation applied to the observed displacement vector, To extract the observation operators corresponding to the computational node displacements from the state vector, The Kalman gain matrix is ​​calculated based on the covariance of the forecast set and the covariance of the observation error. The analysis expands the state vector to represent the updated i-th state vector; Calculate the mean of all the analyzed amplified state vector sets, and use the distribution of the parameter field part and the model state variable part as the updated rock mechanics parameter field and the state of the rock mechanics numerical model.

5. The method according to claim 3, characterized in that, The process of using the rock mass mechanics parameter field and the state of the rock mass mechanics numerical model as initial conditions to drive the rock mass mechanics numerical model to perform stability extrapolation and prediction for a preset future time period, thereby obtaining the future deformation field and stability indices of the slope, includes: The values ​​of the corresponding material regions or structural surfaces in the rock mechanics numerical model are replaced by the values ​​of each parameter in the rock mechanics parameter field to obtain the rock mechanics numerical model with updated parameters. The state of the rock mechanics numerical model is set as the initial condition for the prediction calculation of the rock mechanics numerical model; the state of the rock mechanics numerical model includes displacement field, stress field and plastic state variables; On the rock mechanics numerical model after parameter update, load changes and / or boundary condition changes for a future preset time period are set, and the rock mechanics numerical model is driven to perform numerical simulation calculations based on the initial conditions to obtain numerical simulation calculation results. The slope displacement field, maximum shear stress distribution area, and plastic strain cloud map for a future preset time period are extracted from the numerical simulation calculation results to obtain the future deformation field; Based on the stress state and the updated rock mass mechanical parameter field obtained from the numerical simulation, the overall safety factor of the slope under the current and future conditions is calculated using the strength reduction method, thus obtaining the stability index.

6. The method according to claim 1, characterized in that, The process of solving and registering the raw scanning data of the UAV lidar under each time-series monitoring period yields a high-precision time-series point cloud sequence in the same coordinate system, including: The raw scanning data of UAV lidar under each time series monitoring period are processed to obtain the initial three-dimensional point cloud under each time series monitoring period. Outlier removal and statistical filtering are performed on the initial three-dimensional point cloud under each time series monitoring period to obtain the standard three-dimensional point cloud under each time series monitoring period. Using the standard three-dimensional point cloud of the first time-series monitoring cycle as a fixed reference benchmark, and using artificially placed targets or natural feature points that can be stably identified in each time-series monitoring cycle as control points, the standard three-dimensional point cloud of each time-series monitoring cycle is initially registered to the coordinate system of the fixed reference benchmark through rigid body transformation, thus obtaining the preliminary spatiotemporal registration three-dimensional point cloud of each time-series monitoring cycle. Based on the preliminary spatiotemporal registration three-dimensional point clouds under each time series monitoring period, the iterative nearest point algorithm is used to perform fine registration on the preliminary spatiotemporal registration three-dimensional point clouds of two adjacent time series monitoring periods to obtain the high-precision time series point cloud sequence; the fine registration corresponds to optimizing the registration transformation parameters by minimizing the distance error between corresponding points in the preliminary spatiotemporal registration three-dimensional point clouds, and performing registration according to the optimized registration transformation parameters.

7. The method according to claim 1, characterized in that, The method further includes: In response to receiving the new batch of raw UAV lidar scanning data collected in the next time-series monitoring cycle, the raw UAV lidar scanning data is processed for settlement and registration to obtain the updated high-precision time-series point cloud sequence. Using the updated high-precision time-series point cloud sequence, the slope four-dimensional geometric evolution model is incrementally trained, and the new complete slope displacement observation field at the current monitoring time is extracted. The updated rock mechanics parameter field and rock mechanics numerical model state obtained from the previous time series monitoring period are used as new initial conditions for inversion update and stability prediction to obtain the new future deformation field and stability index.

8. A dynamic assessment device for mine slope stability based on UAV lidar, characterized in that, The device includes: The point cloud data module is used to acquire the raw scanning data of UAV lidar under multiple time-series monitoring periods of the target slope, and to perform calculation and registration processing on the raw scanning data of UAV lidar under each time-series monitoring period to obtain a high-precision time-series point cloud sequence under the same coordinate system. The slope four-dimensional continuous deformation field modeling module is used to construct a spatiotemporally continuous slope four-dimensional geometric evolution model based on the high-precision time-series point cloud sequence, and to extract the complete slope displacement observation field at the current monitoring time based on the slope four-dimensional geometric evolution model. The rock mechanics module is used to establish a rock mechanics numerical model of the mine slope based on the slope geometry and geological information at the initial monitoring time in the high-precision time-series point cloud sequence, and to establish the spatial coordinate mapping relationship between the computational grid of the rock mechanics numerical model and the four-dimensional geometric evolution model of the slope; the rock mechanics numerical model includes an initial mechanical parameter field; The inversion update module is used to fuse the complete slope displacement observation field at the current monitoring time with the predicted state of the rock mechanics numerical model based on the data assimilation algorithm and the spatial coordinate mapping relationship, and to invert and update the initial mechanical parameter field to obtain the rock mechanics parameter field and the state of the rock mechanics numerical model. The fusion prediction module is used to drive the rock mechanics numerical model to perform stability extrapolation and prediction for a preset future time period, using the rock mechanics parameter field and the state of the rock mechanics numerical model as initial conditions, so as to obtain the future deformation field and stability index of the slope.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images