Full-life-cycle underground engineering geological safety risk collaborative management and control method and system

Abstract

The invention discloses a full-life-cycle underground engineering geological safety risk collaborative management and control method and system. The method comprises the following steps: performing regional geological structure identification and geological risk identification on a target construction region, and constructing a regional geological digital model; regional construction monitoring information is obtained, and construction risk analysis and cavity prediction are performed in combination with a regional geological digital model, so that regional construction risk analysis information is obtained; constructing a carrier risk assessment model, and analyzing whether the current construction risk affects a building object on the ground or not through the regional construction risk analysis information and the regional construction monitoring information to obtain carrier risk assessment information; and based on the regional construction risk analysis information and the carrier risk assessment information, judging whether construction risk management and control need to be carried out, and if so, generating a risk management and control scheme to carry out regional construction safety risk collaborative management and control. And timeliness, comprehensiveness and decision reliability requirements of underground engineering construction management and control in a complex geological environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of underground engineering risk control, and particularly to a collaborative control method and system for geological safety risks of underground engineering in the whole life cycle. Background Art

[0002] With the acceleration of the urbanization process, the scale of underground space development continues to expand, and the geological safety risks faced by projects such as tunnels and underground utility tunnels are becoming increasingly prominent. Traditional underground engineering risk control mainly relies on empirical judgment and static geological reports. During the construction stage, the deformation of surrounding rocks is recorded through manual inspections, and the safety of the support structure is estimated using a simplified limit equilibrium theory. Currently, the geological risk control of underground engineering faces three major technical bottlenecks: the fragmentation of data in the whole life cycle, the lag in dynamic response, and the difficulty in quantifying the transmission of multi-source risks. Traditional methods rely on static geological exploration data to construct two-dimensional profile models, which are difficult to dynamically correct by integrating real-time monitoring information during the construction period, resulting in deficiencies in construction risk identification and response. At the same time, most traditional underground engineering risk management focuses on geological changes, thus only considering the control of underground engineering and failing to consider the disaster damage of buildings above or near the construction area.

[0003] In view of the requirements for the timeliness, comprehensiveness, and decision-making reliability of risk control in the whole life cycle under complex geological environments, it is urgent to construct an intelligent control framework for multi-system collaboration. Therefore, a collaborative control method and system for geological safety risks of underground engineering in the whole life cycle are proposed to solve the above problems. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and provides a collaborative control method and system for geological safety risks of underground engineering in the whole life cycle, with the important purpose of meeting the requirements for the timeliness, comprehensiveness, and decision-making reliability of underground engineering construction control under complex geological environments.

[0005] To achieve the above object, the first aspect of the present invention provides a collaborative control method for geological safety risks of underground engineering in the whole life cycle, including: Conduct geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, identify regional geological structures and mark geological risks, and construct a regional geological digital model; Real-time monitor the construction area through a set heterogeneous sensor network to obtain regional construction monitoring information, and combine it with the regional geological digital model to conduct construction risk analysis and cavity prediction to obtain regional construction risk analysis information; Construct a carrier risk assessment model, and analyze whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information to obtain carrier risk assessment information; Based on the regional construction risk analysis information and the carrier risk assessment information, it is determined whether construction risk control is required. If so, a risk control plan is generated for collaborative control of regional construction safety risks.

[0006] In this solution, the geological monitoring and surface terrain scanning of the target area are carried out to obtain regional geological monitoring information and surface terrain point cloud data, and regional geological structure identification and geological risk identification are carried out to construct a regional geological digital model, which specifically includes: The surface terrain point cloud data is obtained by using a surface scanning device to scan the target area from multiple angles according to a preset scanning plan. The iterative closest point algorithm is introduced to register the regional point cloud scanning data of the target area to a unified engineering coordinate system and eliminate interference points, and an initial digital surface model is constructed; The geological monitoring of the target area is carried out to obtain regional geological monitoring information, and the obtained regional geological monitoring information is preprocessed to eliminate abnormal data and supplement missing values; Feature extraction is carried out on the preprocessed regional geological monitoring information, the regional geological features of the target area are extracted and converted into a vector point set associated with geographical coordinates, and the Kriging interpolation algorithm is used for spatial distribution calculation; Taking the sensor coordinates as control points, based on the semi-variogram analysis of the spatial autocorrelation of lithology parameters, a physical property distribution matrix of each cubic meter of rock mass is established to obtain a rock mass property matrix. The rock mass property matrix and the initial digital surface model are linearly weighted and fused in the depth direction to generate a three-dimensional geological grid model.

[0007] In this solution, the geological monitoring and surface terrain scanning of the target area are carried out to obtain regional geological monitoring information and surface terrain point cloud data, and regional geological structure identification and geological risk identification are carried out to construct a regional geological digital model, which also includes: The three-dimensional geological grid model is obtained, and the three-dimensional geological grid model is input into a pre-trained three-dimensional convolutional neural network for structure identification. The first convolutional layer traverses the entire three-dimensional geological grid model based on a preset sliding window, and performs feature scanning on local rock mass voxel units to obtain local rock mass feature vectors; The local rock mass feature vectors are input into the second layer and compressed through a maximum pooling operation, and the maximum eigenvalue is taken for output. Finally, the fully connected layer is used to map the high-dimensional features into a lithology boundary probability map; The wave velocity difference and RQD value of each voxel unit are obtained through the lithology boundary probability map. When the wave velocity difference or RQD value of adjacent voxel units is greater than a preset threshold, the adjacent voxel units are marked as candidate points of geological discontinuity surfaces; For each candidate point of the geological discontinuity, extract the wave velocity and rock mass index values within the preset voxel range around the point, calculate the parameter change rates in the three dimensions of X, Y, and Z in the spatial rectangular coordinate system respectively, and generate a parameter gradient vector representing the mutation direction of the geological parameters; Introduce the marching cubes algorithm to connect each candidate point of the geological discontinuity to generate a fault surface, combine the parameter gradient vector to generate a spatial triangular grid, calculate the engineering fault distance value of each section through the spatial triangular grid, and eliminate the sections with the engineering fault distance value less than the preset threshold to construct a fault grid model; Based on the fault grid model, use the end points where the fault lines intersect as the seed source points, perform three-dimensional spatial extension calculation along the dominant direction of the rock mass joint surface, and generate the risk score of each voxel unit by traversing each voxel to extract the risk evaluation parameters for weighted calculation, where the risk evaluation parameters include wave velocity anomaly, RQD attenuation rate, fracture density, and seepage convergence intensity; Perform geological risk level identification according to the calculated risk scores, and output the final regional geological digital model after completing the geological risk identification.

[0008] In this solution, the regional construction monitoring information is obtained by real-time monitoring of the construction area through the set heterogeneous sensor network, and the construction risk analysis and cavity prediction are carried out in combination with the regional geological digital model to obtain the regional construction risk analysis information, which specifically includes: Construct the target area according to the preset construction plan and obtain the regional construction monitoring information by real-time monitoring the construction data of the construction area through the set heterogeneous sensor network, and perform data preprocessing on the regional construction monitoring information; Obtain the regional geological digital model, input the preprocessed regional construction monitoring information into the regional geological digital model, use the spatial hashing algorithm to match the three-dimensional coordinates of the sensors with each risk unit in the regional geological model, and generate a construction disturbance vector sequence with geological attribute tags; Conduct a geological response coupling simulation analysis according to the construction disturbance vector sequence, construct an energy transfer model based on the discrete element-finite element hybrid calculation framework and input the construction disturbance vector sequence, and the energy transfer model includes a discrete element module and a finite element module; In the discrete element module, use the rock mass risk unit as the basic particle swarm, and simulate the propagation and attenuation process of the drilling influence according to the Hertz contact theory, where the energy attenuation rate is dynamically controlled by the rock mass damping coefficient; In the finite element module, discretize the interface between the support structure and the surrounding rock into shell-body coupling units, dynamically solve the redistribution of the contact stress under the action of the disturbance load based on the Lagrangian algorithm, and calibrate the corresponding interface as an abnormal interface when the peak value of the shear stress at the coupling interface exceeds the shear strength of the rock mass stored in the geological model; Obtain the coupled simulation results of geological responses, input the regional construction monitoring information into the gradient boosting decision tree model for void prediction, analyze the probability of voids appearing in the current construction area, and output the regional construction void prediction information; Generate a stress redistribution contour map based on the coupled simulation results of geological responses, calibrate the area exceeding the preset disturbance stress increase amplitude as a risk area, and calibrate the risk level according to the void probability based on the regional construction void prediction information to obtain the regional construction risk analysis information.

[0009] In this solution, the carrier risk assessment model is constructed. By analyzing whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and the regional construction monitoring information, the carrier risk assessment information is obtained, which specifically includes: Introduce a data network, obtain historical construction risk instances through the big data network, extract features from the historical construction risk instances, divide the extracted historical construction risk instance features according to the construction risk categories, generate several clusters and construct a historical construction risk feature set; Build a carrier risk assessment model based on the Bayesian inference network. During the model training process, introduce a node conditional probability table to optimize the carrier risk assessment model, construct a training data set using the historical construction risk feature set and adopt the variational Bayesian inference algorithm to train the node conditional probability table; Establish a validation data set according to the historical construction risk feature set to validate the carrier risk assessment model. When the model output result meets the expectation, retain the model parameters and output the carrier risk assessment model after training; Obtain the regional construction risk analysis information and the regional construction monitoring information, and input them into the trained carrier risk assessment model to analyze the risk probability of the regional ground carrier when a construction risk occurs; Generate a construction risk feature map at the current moment using the regional construction risk analysis information and the regional construction monitoring information input into the model, calculate the spatial relationship between the construction risk and the ground carrier, and perform topological association with the construction risk feature map at the current moment; Initialize the risk nodes through the topologically associated construction risk feature map at the current moment and obtain the node transition probability distribution from the node conditional probability table. Use the node transition probability distribution to conduct risk conduction inference to obtain the carrier risk assessment information.

[0010] In this solution, based on the regional construction risk analysis information and the carrier risk assessment information, judge whether construction risk control is required. If so, generate a risk control plan for collaborative control of regional construction safety risks, which specifically includes: Retrieve several historical construction risk control examples through historical data retrieval, extract features from each historical construction risk control example, and obtain the historical regional construction risk types, historical regional geological risk characteristics, historical construction risk control plan characteristics, and historical risk control time limit characteristics of each historical construction risk control example; Generate entity triples based on the historical regional construction risk types, historical regional geological risk characteristics, and historical construction risk control plan characteristics, and construct a risk control knowledge graph according to the entity triples; Use MetaPath random walk to perform representation learning on the risk control knowledge graph and obtain entity nodes corresponding to the historical regional geological risk characteristics, generate meta-paths based on the connections between entity nodes, and use the historical risk control time limit characteristics as the attached features of the meta-paths to form a heterogeneous information network; Perform deep learning and training on the heterogeneous information network through a graph neural network, obtain the risk assessment information of the carrier of the regional construction risk analysis information, and input it into the trained graph neural network to generate target nodes; In the graph neural network, perform third-order neighborhood sampling on the target nodes to obtain neighbor nodes and secondary neighbor nodes, obtain the attention coefficients between the neighbor nodes and the secondary neighbor nodes through the attention mechanism, and perform weighted update on the feature vectors corresponding to the neighbor nodes to obtain the updated neighbor nodes; Concatenate the updated neighbor nodes and the target nodes through the shared attention parameter mechanism, update the feature vector of the target using a preset activation function, and output the candidate risk control plan in the current construction scenario through the updated target node feature vector; Extract the risk control time limit characteristics corresponding to each candidate risk control plan, judge them against the expected time limit, screen the candidate risk control plans that meet the expected time limit for sorting, and perform construction risk control assistance according to the sorting results.

[0011] The second aspect of the present invention provides a full-life-cycle underground engineering geological safety risk collaborative control system, which includes: a memory and a processor. The memory contains a program for the full-life-cycle underground engineering geological safety risk collaborative control method. When the program for the full-life-cycle underground engineering geological safety risk collaborative control method is executed by the processor, the following steps are implemented: Conduct geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, conduct regional geological structure identification and geological risk marking, and construct a regional geological digital model; Real-time monitor the construction area through a set heterogeneous sensor network to obtain regional construction monitoring information, and combine it with the regional geological digital model to conduct construction risk analysis and cavity prediction to obtain regional construction risk analysis information; Construct a carrier risk assessment model, and analyze whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information, so as to obtain the carrier risk assessment information; Based on the regional construction risk analysis information and the carrier risk assessment information, judge whether construction risk control is required. If so, generate a risk control plan for collaborative control of the regional construction safety risk.

[0012] The present invention discloses a method and system for collaborative control of geological safety risks in the whole life cycle of underground engineering, including: identifying the regional geological structure and marking geological risks in the target construction area, and constructing a regional geological digital model; obtaining regional construction monitoring information, combining the regional geological digital model to conduct construction risk analysis and cavity prediction, so as to obtain regional construction risk analysis information; constructing a carrier risk assessment model, and analyzing whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information, so as to obtain the carrier risk assessment information; based on the regional construction risk analysis information and the carrier risk assessment information, judge whether construction risk control is required. If so, generate a risk control plan for collaborative control of the regional construction safety risk. It meets the requirements of timeliness, comprehensiveness and decision-making reliability of underground engineering construction control in complex geological environments. Brief Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions in the embodiments or exemplary examples of the present invention, the following will briefly introduce the drawings required for use in the embodiments or exemplary descriptions. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the drawings shown without creative efforts.

[0014] Figure 1 It is a flowchart of a method for collaborative control of geological safety risks in the whole life cycle of underground engineering provided by an embodiment of the present invention; Figure 2 It is a flowchart of a method for collaborative control of construction safety risks in underground engineering provided by an embodiment of the present invention; Figure 3 It is a block diagram of a system for collaborative control of geological safety risks in the whole life cycle of underground engineering provided by an embodiment of the present invention; The realization, functional characteristics and advantages of the object of the present invention will be further described in conjunction with the embodiments with reference to the drawings. Detailed Embodiments

[0015] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0016] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0017] Figure 1 A flow chart of a collaborative control method for geological safety risks of underground engineering in the whole life cycle provided by an embodiment of the present invention; As Figure 1 shown, the present invention provides a flow chart of a collaborative control method for geological safety risks of underground engineering in the whole life cycle, including: S102, performing geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, performing regional geological structure identification and geological risk marking, and constructing a regional geological digital model; S104, obtaining regional construction monitoring information by real-time monitoring of the construction area through a set heterogeneous sensor network, and combining with the regional geological digital model to perform construction risk analysis and cavity prediction to obtain regional construction risk analysis information; S106, constructing a carrier risk assessment model, and analyzing whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and the regional construction monitoring information to obtain carrier risk assessment information; S108, judging whether construction risk control is required based on the regional construction risk analysis information and the carrier risk assessment information. If so, generating a risk control plan for collaborative control of regional construction safety risks.

[0018] Further, in a preferred embodiment of the present invention, the performing geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, performing regional geological structure identification and geological risk marking, and constructing a regional geological digital model specifically includes: Using a surface scanning device to perform multi-angle scanning on the target area according to a preset scanning plan to obtain surface terrain point cloud data, introducing the iterative closest point algorithm to register the regional point cloud scanning data of the target area to a unified engineering coordinate system and removing interference points, and constructing an initial digital surface model; Performing geological monitoring on the target area to obtain regional geological monitoring information, and performing data preprocessing on the obtained regional geological monitoring information to remove abnormal data and supplement missing values; Extract the features of the preprocessed regional geological monitoring information, extract the regional geological features of the target area and convert them into a vector point set associated with geographical coordinates, and use the Kriging interpolation algorithm for spatial distribution calculation; Taking the sensor coordinates as control points, based on the semi-variogram analysis of the spatial autocorrelation of lithology parameters, establish the physical property distribution matrix of each cubic meter of rock mass to obtain the rock mass property matrix, and linearly weight and fuse the rock mass property matrix and the initial digital surface model in the depth direction to generate a three-dimensional geological grid model.

[0019] It should be noted that based on a preset scanning scheme, a multi-rotor UAV equipped with lidar is used to perform multi-angle surface scanning on the target area to obtain a raw topographic point cloud dataset with centimeter-level accuracy. The scanning process follows the well-shaped flight path design, with the single-point positioning accuracy controlled within ±2 cm and the point cloud density not less than 200 points per square meter. The iterative closest point algorithm (ICP) is used to automatically register the multi-flight path scanning data: taking the coordinates of the engineering control network reference station as the conversion target, and through minimizing the root mean square error (RMSE < 5 cm) of the Euclidean distance between the corresponding point clouds, the point clouds of each flight path are uniformly converted to the engineering coordinate system. The registered point clouds are filtered to remove the interference of non-topographic points such as vegetation cover and temporary facilities through an improved moving surface filtering algorithm, and a high-fidelity digital surface model is generated based on the Poisson surface reconstruction technology to clearly depict the surface undulation characteristics and the contours of artificial structures. Synchronously deploy a multi-type geological monitoring sensor network, including in-hole strain gauges (buried depth 0 - 50 m), cross-hole seismic wave velocity gauge arrays, groundwater pressure sensors, etc., to continuously collect parameters such as rock mass deformation rate, elastic wave velocity, and seepage pressure. The original monitoring data goes through two levels of preprocessing: first, use box plot statistics to identify and remove abnormal jump points caused by temperature drift and electrical interference; then use the spatio-temporal Kriging interpolation algorithm to supplement signal interruptions or missing values to ensure the spatio-temporal continuity of the data. The processed structured data extracts seven-dimensional geological feature parameters such as triaxial strain energy density, wave velocity gradient modulus, and seepage instability coefficient through wavelet packet decomposition, and each feature parameter is bound to its spatial geographical coordinates (longitude, latitude, elevation) to form a vector feature point set with position tags.

[0020] Subsequently, the vector point set is input into the spatial interpolation engine, and Kriging spatial distribution calculation is performed with the three-dimensional coordinates of the monitoring points as control points. The spatial autocorrelation of lithology parameters is quantified through the semi-variogram function: the spatial variation characteristics of wave velocity and strain parameters are analyzed to determine the standard model parameters of nugget effect value of 0.12, range of 15 m, and sill value of 1.8. Interpolation calculation is carried out on the regular grid points of 0.5 m×0.5 m×0.5 m in three-dimensional space according to this model to generate the physical property distribution matrix of each cubic meter of rock mass (including mechanical indexes such as dynamic elastic modulus, Poisson's ratio, permeability coefficient, joint density, etc.), and its spatial resolution and error distribution are strictly matched with the borehole calibration results (vertical error <5%, horizontal error <8%). Finally, three-dimensional integration of surface topography and rock mass properties is achieved through linear weighted fusion in the depth direction: for the shallow area of 0-5 m underground, the weight coefficient of the digital surface model is set to 0.8 (emphasizing terrain constraint), and the weight of the rock mass property matrix is 0.2; for the medium depth of 5-20 m, the balanced weight is 0.5:0.5; for the area deeper than 20 m, the rock mass properties are dominant (weight 0.9). The fusion process performs multi-level LOD processing under the support of the octree spatial index structure to generate a three-dimensional geological grid model. This model is stored in a tensor structure, and each voxel unit contains 12-dimensional attribute fields such as terrain elevation, lithology code, and mechanical parameters, providing a fully parameterized digital base for subsequent intelligent identification of geological risks.

[0021] Further, in a preferred embodiment of the present invention, the geological monitoring of the target area and the surface topography scanning to obtain regional geological monitoring information and surface topography point cloud data, perform regional geological structure identification and geological risk marking and construct a regional geological digital model, further include: Obtain a three-dimensional geological grid model, input the three-dimensional geological grid model into a pre-trained three-dimensional convolutional neural network for structure identification, traverse the entire three-dimensional geological grid model through the first convolutional layer based on a preset sliding window, perform feature scanning on local rock mass voxel units, and obtain local rock mass feature vectors; Input the local rock mass feature vectors into the second layer for compression through max-pooling operation, take the maximum eigenvalue for output, and finally use the fully connected layer to map the high-dimensional features into a lithology boundary probability map; Obtain the wave velocity difference and RQD value of each voxel unit through the lithology boundary probability map. When the wave velocity difference or RQD value of adjacent voxel units is greater than a preset threshold, mark the adjacent voxel units as candidate points for geological discontinuity surfaces; Extract the wave velocity and rock mass quality index values within the preset voxel range around each candidate point for geological discontinuity surface, calculate the parameter change rates in the X, Y, and Z dimensions of the spatial rectangular coordinate system respectively, and generate a parameter gradient vector characterizing the mutation direction of geological parameters; The marching cubes algorithm is introduced to connect the candidate points of each geological discontinuity to generate a fault surface, and a spatial triangular grid is generated by combining the parameter gradient vectors. The engineering throw values of each section are calculated through the spatial triangular grid, and the sections with engineering throw values less than the preset threshold are removed to construct a fault grid model; Based on the fault grid model, taking the end points where the fault lines intersect as the seed source points, three-dimensional space extension calculation is carried out along the dominant direction of the rock mass joint surface. The risk evaluation parameters are extracted by traversing each voxel, and weighted calculation is performed to generate the risk scores of each voxel unit, where the risk evaluation parameters include wave velocity anomaly degree, RQD attenuation rate, fracture density, and seepage convergence intensity; The geological risk level is marked according to the calculated risk scores, and the final regional geological digital model is output after the geological risk marking is completed.

[0022] It should be noted that the three-dimensional geological grid model is used as the input data source of the pre-trained three-dimensional convolutional neural network (3D-CNN). The first convolutional layer uses a 3×3×3-sized cubic convolution kernel and traverses the entire model space in the form of a sliding window. The convolution kernel performs deep feature scanning on local rock mass voxel units during movement, and focuses on extracting 9-dimensional feature parameters such as wave velocity gradient, joint density, and stress concentration degree, and outputs a local rock mass feature vector matrix composed of 128-channel feature maps. The feature matrix is input into the second processing unit, and the data dimension is compressed through a 2×2×2 maximum pooling operation: the most significant geological anomaly signal (retaining the maximum value) is extracted within each cubic region, and the data volume is reduced to 1 / 8 of the original size. Then, the pooled feature map is input into the fully connected layer, and the hyperbolic tangent activation function is used to map the high-dimensional features into a lithology boundary probability map. This probability map quantifies the possibility of lithology mutation at each spatial position (value range 0-1), so as to clearly identify geological interfaces such as sandstone-mudstone boundaries and fault fracture zones.

[0023] On the other hand, discontinuity detection is carried out based on the lithology boundary probability map: calculate the absolute value of the wave velocity difference between adjacent voxel units (measured across the unit boundary) and the jump amplitude of the rock quality designation (RQD). When it is detected that the wave velocity difference > 500 m / s or the attenuation rate of the RQD value > 30% within three consecutive voxels, this group of voxels is marked as candidate points of geological discontinuities, and their three-dimensional coordinates and abnormal parameter values are recorded. Perform parameter gradient field analysis on each candidate point: extract the wave velocity and RQD values within the range of 3×3×3 voxels around it, and calculate the parameter change rate in the XYZ axes of the spatial rectangular coordinate system respectively. The X-direction gradient is calculated by dividing the wave velocity difference between the left and right adjacent voxels by the voxel spacing, the Y-direction calculates the difference between the front and back voxels, and the Z-direction analyzes the vertical change, and finally synthesizes a three-dimensional gradient vector. This vector points to the direction where the geological parameters deteriorate most significantly, and the modulus length characterizes the degree of drastic change. Subsequently, the marching cubes algorithm is introduced to construct the fault space surface: a cubic calculation domain is generated with the discontinuity candidate points as vertices, and the intersection points of the isosurface are located by linear interpolation on the edges of the cube. Connect adjacent intersection points according to the spatial continuity of the gradient vector G, and generate triangular patches according to the right-hand rule. The engineering fault displacement value is automatically calculated for each triangular patch: identify pairs of lithology feature matching points on the fault surface, calculate the vertical fault displacement based on the unit normal vector N, and calculate the horizontal fault displacement in combination with the strike vector T to synthesize the three-dimensional comprehensive fault displacement value. Automatically eliminate the micro-fracture surfaces with fault displacements less than the preset threshold, and retain the main fault zones to form a fault network model.

[0024] Start risk zoning based on the fault network model: use the intersection endpoints of the fault lines as seed sources, and perform a three-dimensional region growing algorithm along the dominant directions of the joint surfaces (within the range of strike ±15° and dip ±10°). During the growth process, traverse and cover the voxel units, and synchronously calculate four key parameters, namely the wave velocity anomaly (the multiple of the standard deviation relative to the regional background value), the RQD attenuation rate (the linear deterioration slope along the depth direction), the fracture density (the number of micro-cracks identified in the borehole CT image per cubic meter), and the seepage convergence intensity (the divergence value of the groundwater flow vector field). The risk assessment calculation formula is: risk score = 0.4×wave velocity anomaly + 0.3×RQD attenuation rate + 0.2×fracture density + 0.1×seepage convergence intensity. Voxels with a risk score > 0.7 are marked red (high-risk area), those in the range of 0.5 - 0.7 are marked yellow (warning area), and the rest of the area remains the reference color. Carry out geological risk level identification based on the calculated risk scores, and output the final regional geological digital model after completing the geological risk identification.

[0025] Furthermore, in a preferred embodiment of the present invention, the regional construction monitoring information is obtained by real-time monitoring of the construction area through a set heterogeneous sensor network, and construction risk analysis and cavity prediction are carried out in combination with the regional geological digital model to obtain regional construction risk analysis information, which specifically includes: Construct the target area according to a preset construction plan and monitor the construction data of the construction area in real time through a set heterogeneous sensor network to obtain regional construction monitoring information, and perform data preprocessing on the regional construction monitoring information; Obtain a regional geological digital model, input the preprocessed regional construction monitoring information into the regional geological digital model, and use the spatial hashing algorithm to match the three-dimensional coordinates of the sensors with each risk unit in the regional geological model to generate a construction disturbance vector sequence with geological attribute tags; Conduct a geological response coupling simulation analysis based on the construction disturbance vector sequence, construct an energy transfer model based on a discrete element-finite element hybrid calculation framework and input the construction disturbance vector sequence, where the energy transfer model includes a discrete element module and a finite element module; In the discrete element module, use the rock mass risk unit as the basic particle swarm, and simulate the propagation attenuation process of the drilling impact according to the Hertz contact theory, where the energy attenuation rate is dynamically controlled by the rock mass damping coefficient; In the finite element module, discretize the interface between the support structure and the surrounding rock into a shell-body coupling unit, and dynamically solve the redistribution of contact stress under the action of disturbance loads based on the Lagrangian algorithm. When the peak value of the shear stress at the coupling interface exceeds the shear strength of the rock mass stored in the geological model, calibrate the corresponding interface as an abnormal interface; Obtain the geological response coupling simulation results, input them into the gradient boosting decision tree model in combination with the regional construction monitoring information for cavity prediction, analyze the probability of cavities appearing in the current construction area, and output regional construction cavity prediction information; Generate a stress redistribution cloud map based on the geological response coupling simulation results, calibrate the area exceeding the preset disturbance stress increase amplitude as a risk area, and calibrate the risk level according to the cavity probability based on the regional construction cavity prediction information to obtain regional construction risk analysis information.

[0026] It should be noted that the construction data of the construction area is monitored in real time through a set heterogeneous sensor network to obtain regional construction monitoring information. The temperature drift compensation algorithm is used to eliminate environmental interference, and then the wavelet threshold denoising technology is used to filter out the vibration noise of construction machinery. Finally, the preprocessed regional construction monitoring information is output. The preprocessed monitoring data is input into the regional geological digital model, and the physical coordinates of the sensors are accurately matched with the risk units in the geological model through the spatial hashing index engine. Three-dimensional hash bucket division (bucket size 0.5m³) is used to retrieve the risk voxels within a radius of 1.2m around the target unit at a speed of 0.1ms level, and a construction disturbance vector sequence V = {timestamp, spatial coordinates, vibration acceleration, support stress, pore water pressure, risk level code} with geological attribute tags is generated.

[0027] Construct an energy transfer model based on the discrete element - finite element hybrid calculation framework and input the sequence of construction disturbance vectors, and use the discrete element and finite element hybrid calculation architecture to conduct energy propagation modeling: in the discrete element calculation domain, regard the risk units divided from the rock mass as a dynamic particle swarm, and simulate the propagation attenuation process based on the contact mechanics principle, where the energy attenuation rate is dynamically controlled by the rock mass damping coefficient; in the finite element calculation domain, discretize the contact surface between the support structure and the surrounding rock into coupling units, and dynamically solve the redistribution of contact stress under the action of disturbance loads based on the Lagrangian algorithm. When the peak value of the interface shear stress exceeds the shear strength threshold of the rock mass stored in the geological model, it is automatically marked as the interface slip risk area and the warning coordinates are output. Furthermore, obtain the coupled simulation results of geological response, and input the regional construction monitoring information into the gradient boosting decision tree model for cavity prediction. Extract six - dimensional characteristic parameters through the gradient boosting decision tree algorithm, including key indicators such as the relative increase in disturbance stress, the cumulative energy of microseismic events, the dynamic attenuation rate of rock mass integrity, and the abnormal deflection of seepage direction, and particularly introduce the fault space constraint factor to strengthen the geological correlation, and output the grid - shaped cavity probability distribution map, clearly showing the future potential cavity development locations and risk levels. Generate a stress redistribution cloud map based on the coupled simulation results of geological response, calibrate the area exceeding the preset disturbance stress increase as the risk area, and calibrate the risk level according to the cavity probability based on the regional construction cavity prediction information to obtain the regional construction risk analysis information.

[0028] The rock mass damping coefficient is calibrated in real - time by inverting the cross - hole seismic wave velocity field, and the calculation formula is: , where, is the rock mass damping coefficient, is the longitudinal wave velocity, is the stress difference between adjacent rock layers.

[0029] Furthermore, in a preferred embodiment of the present invention, to construct the carrier risk assessment model, analyze whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information, and obtain the carrier risk assessment information, which specifically includes: Introduce a data network, obtain historical construction risk instances through the big data network, extract the characteristics of the historical construction risk instances, classify the extracted characteristics of the historical construction risk instances according to the construction risk categories, generate several clusters and construct a historical construction risk feature set; Build a carrier risk assessment model based on the Bayesian inference network. During the model training process, introduce a node conditional probability table to optimize the carrier risk assessment model, construct a training data set using the historical construction risk feature set and use the variational Bayesian inference algorithm to train the node conditional probability table; A validation dataset is established based on the historical construction risk feature set to validate the carrier risk assessment model. When the model output result meets the expectation, the model parameters are retained and the carrier risk assessment model after training is output. Obtain regional construction risk analysis information and regional construction monitoring information, and input them into the trained carrier risk assessment model to analyze the risk probability of the regional ground carrier when construction risks occur. Generate a construction risk feature map at the current moment using the regional construction risk analysis information and regional construction monitoring information input into the model, calculate the spatial relationship between the construction risk and the ground carrier, and perform topological association with the construction risk feature map at the current moment. Initialize risk nodes through the construction risk feature map after topological association and obtain the node transition probability distribution from the node conditional probability table. Use the node transition probability distribution to conduct risk conduction inference to obtain the carrier risk assessment information.

[0030] It should be noted that relying on the engineering big data platform, a global historical construction risk case library is constructed, and a number of underground engineering risk event cases are collected through distributed crawler technology. Multidimensional feature extraction is performed on the original cases, including core elements such as focusing on construction disturbance intensity, geological anomaly types, support structure responses, and disaster evolution paths, to form a structured dataset of multidimensional feature vectors. The clustering algorithm is used to divide the cases into six main clusters according to the risk mechanism (such as fault activation type seepage, large deformation of soft rock, joint surface slip, etc.), and a labeled historical construction risk feature knowledge base is established. Based on the Bayesian theory, a carrier risk assessment network model framework is constructed, and the network nodes cover four-layer topological structures of construction disturbance factors, rock mass deterioration indicators, support state parameters, and ground building vulnerability. A three-layer optimization mechanism is introduced: first, the node association rules are initialized according to the principles of geotechnical mechanics (such as the positive correlation between blasting vibration and fault activation, and the hysteresis effect between support stress and ground settlement), secondly, the prior conditional probabilities set by industry expert experience are embedded, and finally, the variational Bayesian inference algorithm is used for deep learning. Through thousands of iterations of probability inference on the risk evolution path, the parameters of the conditional probability table between nodes are gradually corrected. Model validation is implemented through the pressure test of an independent case set: input 500 groups of risk scenario features that have not participated in training. When the error rate between the risk probability output by the model and the measured disaster result is stable within the range of ±7%, it is determined that the engineering application standard is met, and the trained model is output.

[0031] Subsequently, real-time input of construction risk analysis information and high-precision monitoring data. The model first fuses multi-dimensional information to generate a spatio-temporal linkage feature map, which is based on a geographic information system and maps dynamic parameters such as the current construction disturbance intensity, fault activation state, and density of superficial rock mass fissures. Calculate the three-dimensional spatial relationship between construction risks (such as void areas, etc.) and ground disaster-bearing bodies, and establish a risk conduction topology chain in combination with the joint plane attitude data in the geological model. Initialize the risk node state through the topologized feature map, call the transfer rules from the trained conditional probability table, and implement risk conduction deduction: simulate the disaster chain process of construction vibration → fault dislocation → superficial fissure expansion → building foundation settlement, and output three types of indicators: the probability of ground building inclination, the risk level of road cracking, and the deformation threshold of underground pipelines. The finally generated risk assessment report of the bearing body integrates spatial position labels, catastrophe path maps, and probability prediction values, and intuitively displays the ground sensitive targets affected by construction through a red-orange-yellow three-color risk zoning map.

[0032] Figure 2 It is a block diagram of a full-life-cycle underground engineering geological safety risk collaborative control system provided by an embodiment of the present invention; As Figure 2 shown, the present invention provides a block diagram of a full-life-cycle underground engineering geological safety risk collaborative control system, including: S202, obtain a number of historical construction risk control instances through historical data retrieval, extract features from each historical construction risk control instance, and obtain the historical regional construction risk types, historical regional geological risk characteristics, historical construction risk control plan characteristics, and historical risk control timeliness characteristics of each historical construction risk control instance; S204, generate entity triples based on the historical regional construction risk types, historical regional geological risk characteristics, and historical construction risk control plan characteristics, and construct a risk control knowledge graph according to the entity triples; S206, use MetaPath random walk to perform representation learning on the risk control knowledge graph and obtain entity nodes corresponding to the historical regional geological risk characteristics, generate meta-paths according to the connections between entity nodes, and use the historical risk control timeliness characteristics as the attached features of the meta-paths to form a heterogeneous information network; S208, perform deep learning and training on the heterogeneous information network through a graph neural network, obtain the risk assessment information of the carrier of the regional construction risk analysis information, and input it into the trained graph neural network to generate target nodes; S210, in the graph neural network, perform third-order neighborhood sampling on the target nodes to obtain neighbor nodes and secondary neighbor nodes, obtain the attention coefficients between the neighbor nodes and the secondary neighbor nodes through the attention mechanism, and perform weighted update on the feature vectors corresponding to the neighbor nodes to obtain the updated neighbor nodes; S212, concatenate the updated neighbor nodes with the target node through the shared attention parameter mechanism, update the feature vector of the target using a preset activation function, and output the candidate risk control and management solutions in the current construction scenario through the updated feature vector of the target node; S214, extract the risk control and management time-limit features corresponding to each candidate risk control and management solution, judge them against the expected time limit, screen the candidate risk control and management solutions that meet the expected time limit for ranking, and provide assistance for construction risk control and management based on the ranking results.

[0033] It should be noted that historical risk control and management cases of underground projects are retrieved through the engineering data center for structured feature extraction. Each case is disassembled into a four-dimensional feature vector: historical regional construction risk types (such as fault water inrush, large deformation of soft rock, regional building settlement, etc.), historical regional geological risk characteristics (including quantitative indicators such as fault throw / dip angle / RQD attenuation rate, etc.), historical construction risk control and management solution characteristics, and historical risk control and management time-limit characteristics (response delay hours, disposal duration days, etc.). Using entity relationship modeling technology, the first three types of features are constructed into a risk type - geological feature - control and management solution entity triple, and a risk control and management knowledge graph covering 3 types of entities is constructed. The MetaPath random walk strategy is used for graph representation learning: starting from the geological risk feature node, 1000 rounds of walk sampling are carried out along the geological feature → risk event → control and management solution path. During this process, the historical risk control and management time-limit characteristics are transformed into the attached attributes of the meta-path, and finally a heterogeneous information network with spatio-temporal constraints is formed. The heterogeneous information network is input into the graph neural network for deep learning and training to obtain a graph neural network that meets the expectations.

[0034] Subsequently, obtain the risk assessment information of the carrier of the regional construction risk analysis information, input it into the trained graph neural network to generate a target node, and obtain the neighbor nodes directly associated with the target node (such as cases with the same fault characteristics) and secondary neighbor nodes (cases with similar rock mass structures) through third-order neighborhood sampling. The multi-head attention mechanism is used to calculate the node association strength, that is, the attention coefficient. After updating the neighbor node features through weighted aggregation, they are concatenated with the target node feature vector through the parameter sharing mechanism, and a new feature of the target node integrating global knowledge is generated through the ReLU activation function. This feature vector is mapped through a fully connected layer to output candidate risk control and management solutions. Extract the risk control and management time-limit features corresponding to each candidate risk control and management solution, judge them against the expected time limit, screen the candidate risk control and management solutions that meet the expected time limit for ranking, and provide assistance for construction risk control and management based on the ranking results. Improve the timeliness, comprehensiveness, and decision-making reliability requirements for underground project construction control in complex geological environments.

[0035] Figure 3A collaborative control system 3 for geological safety risks in the whole life cycle of underground engineering provided by an embodiment of the present invention, the system includes: a memory 31 and a processor 32. The memory 31 contains a program for the method of collaborative control of geological safety risks in the whole life cycle of underground engineering. When the program for the method of collaborative control of geological safety risks in the whole life cycle of underground engineering is executed by the processor 32, the following steps are implemented: Conduct geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, conduct regional geological structure identification and geological risk marking, and construct a regional geological digital model; Real-time monitor the construction area through a set of heterogeneous sensor networks to obtain regional construction monitoring information, and combine with the regional geological digital model to conduct construction risk analysis and cavity prediction to obtain regional construction risk analysis information; Construct a carrier risk assessment model, and analyze whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information to obtain carrier risk assessment information; Based on the regional construction risk analysis information and carrier risk assessment information, judge whether construction risk control is required. If so, generate a risk control plan for collaborative control of regional construction safety risks.

[0036] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be electrical, mechanical, or other forms.

[0037] The units described as separate components above may or may not be physically separated. The components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0038] In addition, each functional unit in the embodiments of the present invention can be all integrated in one processing unit, or each unit can be separately used as a unit, or two or more units can be integrated in one unit; the above integrated units can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0039] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including those of the above method embodiments. The foregoing storage medium includes: various media that can store program codes, such as removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0040] Alternatively, if the above integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as removable storage devices, ROM, RAM, magnetic disks, or optical discs.

[0041] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A collaborative control method for geological safety risks in underground engineering throughout the whole life cycle, characterized in that, Including: Conduct geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, identify regional geological structures and mark geological risks, and construct a regional geological digital model; Real-time monitor the construction area through a set heterogeneous sensor network to obtain regional construction monitoring information, combine it with the regional geological digital model to conduct construction risk analysis and cavity prediction, and obtain regional construction risk analysis information; Construct a carrier risk assessment model, and analyze whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information to obtain carrier risk assessment information; Based on the regional construction risk analysis information and carrier risk assessment information, judge whether construction risk control is required. If so, generate a risk control plan to conduct collaborative control of regional construction safety risks.

2. The collaborative control method for geological safety risks of underground engineering throughout the life cycle according to claim 1, characterized in that, The conduct of geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, identify regional geological structures and mark geological risks, and construct a regional geological digital model specifically includes: Use surface scanning equipment to conduct multi-angle scanning on the target area according to a preset scanning plan to obtain surface terrain point cloud data, introduce the iterative closest point algorithm to register the regional point cloud scanning data of the target area to a unified engineering coordinate system and remove interference points, and construct an initial digital surface model; Conduct geological monitoring on the target area to obtain regional geological monitoring information, and perform data preprocessing on the obtained regional geological monitoring information to remove abnormal data and supplement missing values; Extract features from the preprocessed regional geological monitoring information, extract the regional geological features of the target area and convert them into a vector point set associated with geographical coordinates, and use the Kriging interpolation algorithm for spatial distribution calculation; Take the sensor coordinates as control points, analyze the spatial autocorrelation of lithology parameters based on the semi-variogram, establish a physical property distribution matrix of each cubic meter of rock mass to obtain a rock mass property matrix, and linearly weight and fuse the rock mass property matrix and the initial digital surface model in the depth direction to generate a three-dimensional geological grid model.

3. A collaborative control method for geological safety risks in underground engineering throughout the life cycle according to claim 1, characterized in that, The conduct of geological monitoring and surface terrain scanning on the target area to obtain regional geological monitoring information and surface terrain point cloud data, identify regional geological structures and mark geological risks, and construct a regional geological digital model also includes: Obtain a three-dimensional geological grid model, input the three-dimensional geological grid model into a pre-trained three-dimensional convolutional neural network for structure identification, traverse the entire three-dimensional geological grid model based on a preset sliding window through the first convolutional layer, and perform feature scanning on local rock mass voxel units to obtain local rock mass feature vectors; Input the local rock mass feature vectors into the second layer for compression through maximum pooling operation, take the maximum eigenvalue for output, and finally use the fully connected layer to map the high-dimensional features into a lithology boundary probability map; Obtain the wave velocity difference and RQD value of each voxel unit through the lithology boundary probability map. When the wave velocity difference or RQD value of adjacent voxel units is greater than a preset threshold, mark the adjacent voxel units as candidate points for geological discontinuity surfaces; For each candidate point of the geological discontinuity, extract the wave velocity and rock mass index values within the preset voxel range around this point, calculate the parameter change rates in the three dimensions of X, Y, and Z in the space rectangular coordinate system respectively, and generate a parameter gradient vector characterizing the mutation direction of the geological parameters; Introduce the marching cubes algorithm to connect each candidate point of the geological discontinuity to generate a fault surface, combine the parameter gradient vector to generate a spatial triangular grid, calculate the engineering throw values of each section through the spatial triangular grid, and remove the sections with engineering throw values less than the preset threshold to construct a fault grid model; Based on the fault grid model, use the end points where the fault lines intersect as the seed source points, perform three-dimensional space extension calculation along the dominant direction of the rock mass joint surface, and generate the risk scores of each voxel unit by traversing each voxel to extract risk evaluation parameters for weighted calculation, where the risk evaluation parameters include wave velocity anomaly degree, RQD attenuation rate, fracture density, and seepage convergence intensity; Perform geological risk level identification according to the calculated risk scores, and output the final regional geological digital model after completing the geological risk identification.

4. A collaborative control method for geological safety risks in underground engineering throughout the life cycle according to claim 1, characterized in that, The regional construction monitoring information is obtained by real-time monitoring of the construction area through the set heterogeneous sensor network, and construction risk analysis and cavity prediction are carried out in combination with the regional geological digital model to obtain regional construction risk analysis information, specifically including: Carry out construction on the target area based on the preset construction plan and obtain the regional construction monitoring information by real-time monitoring of the construction data in the construction area through the set heterogeneous sensor network, and perform data preprocessing on the regional construction monitoring information; Obtain the regional geological digital model, input the preprocessed regional construction monitoring information into the regional geological digital model, and use the spatial hashing algorithm to match the three-dimensional coordinates of the sensors with each risk unit in the regional geological model to generate a construction disturbance vector sequence with geological attribute tags; Perform geological response coupling simulation analysis according to the construction disturbance vector sequence, construct an energy transfer model based on the discrete element-finite element hybrid calculation framework and input the construction disturbance vector sequence, and the energy transfer model includes a discrete element module and a finite element module; In the discrete element module, take the rock mass risk unit as the basic particle swarm, and simulate the propagation and attenuation process of the drilling influence according to the Hertz contact theory, where the energy attenuation rate is dynamically controlled by the rock mass damping coefficient; In the finite element module, discretize the interface between the support structure and the surrounding rock into a shell-body coupling unit, and dynamically solve the redistribution of the contact stress under the action of the disturbance load based on the Lagrangian algorithm. When the peak value of the shear stress at the coupling interface exceeds the shear strength of the rock mass stored in the geological model, mark the corresponding interface as an abnormal interface; Obtain the geological response coupling simulation result, input it into the gradient boosting decision tree model in combination with the regional construction monitoring information for cavity prediction, analyze the probability of cavities appearing in the current construction area, and output the regional construction cavity prediction information; Generate a stress redistribution cloud map based on the geological response coupling simulation result, mark the area exceeding the preset disturbance stress increase amplitude as a risk area, and perform risk level calibration according to the cavity probability based on the regional construction cavity prediction information to obtain the regional construction risk analysis information.

5. A collaborative control method for geological safety risks of underground engineering throughout the life cycle according to claim 1, characterized in that, The construction of the carrier risk assessment model analyzes whether the current construction risk affects the building objects on the ground through the regional construction risk analysis information and regional construction monitoring information, and obtains the carrier risk assessment information, which specifically includes: Introduce a data network, obtain historical construction risk instances through the big data network, extract the characteristics of the historical construction risk instances, classify the extracted historical construction risk instance characteristics according to the construction risk categories, generate several clusters and construct a historical construction risk feature set; Build a carrier risk assessment model based on the Bayesian inference network. During the model training process, introduce a node conditional probability table to optimize the carrier risk assessment model, construct a training data set using the historical construction risk feature set and use the variational Bayesian inference algorithm to train the node conditional probability table; Establish a validation data set according to the historical construction risk feature set to verify the carrier risk assessment model. When the model output result meets the expectation, retain the model parameters and output the carrier risk assessment model after training; Obtain the regional construction risk analysis information and regional construction monitoring information, and input them into the trained carrier risk assessment model to analyze the risk probability of the regional ground carrier when a construction risk occurs; Generate a construction risk feature map at the current moment using the regional construction risk analysis information and regional construction monitoring information input into the model, calculate the spatial relationship between the construction risk and the ground carrier, and perform topological association with the construction risk feature map at the current moment; Perform risk node initialization through the topologically associated construction risk feature map at the current moment and obtain the node transition probability distribution from the node conditional probability table. Use the node transition probability distribution to perform risk conduction inference to obtain the carrier risk assessment information.

6. A collaborative control method for geological safety risks in underground engineering throughout the life cycle according to claim 1, characterized in that, Judge whether construction risk control is required based on the regional construction risk analysis information and carrier risk assessment information. If so, generate a risk control plan for collaborative control of regional construction safety risks, which specifically includes: Obtain several historical construction risk control instances through historical data retrieval, extract the characteristics of each historical construction risk control instance, and obtain the historical regional construction risk type, historical regional geological risk characteristics, historical construction risk control plan characteristics, and historical risk control time limit characteristics of each historical construction risk control instance; Generate entity triples based on the historical regional construction risk type, historical regional geological risk characteristics, and historical construction risk control plan characteristics, and construct a risk control knowledge graph according to the entity triples; Use MetaPath random walk to perform representation learning on the risk control knowledge graph and obtain the entity nodes corresponding to the historical regional geological risk characteristics. Generate meta-paths based on the connections between entity nodes, and use the historical risk control time limit characteristics as the attached features of the meta-paths to form a heterogeneous information network; Perform deep learning and training on the heterogeneous information network through a graph neural network, obtain the regional construction risk analysis information carrier risk assessment information, and input it into the trained graph neural network to generate target nodes; In a graph neural network, third-order neighborhood sampling is performed on the target node to obtain neighbor nodes and secondary neighbor nodes. The attention coefficients between the neighbor nodes and the secondary neighbor nodes are obtained through an attention mechanism, and the feature vectors corresponding to the neighbor nodes are weighted and updated to obtain the updated neighbor nodes. The updated neighbor nodes and the target node are vector-concatenated through a shared attention parameter mechanism, and the feature vector of the target is updated using a preset activation function. The candidate risk control and management solutions in the current construction scenario are output through the updated feature vector of the target node. The risk control and management time-effect characteristics corresponding to each candidate risk control and management solution are extracted and judged against the expected time-effect. The candidate risk control and management solutions that meet the expected time-effect are screened and sorted, and construction risk control and management assistance is provided according to the sorting results.

7. A collaborative control system for geological safety risks in underground engineering throughout the entire life cycle, characterized in that, The system includes: a memory and a processor. The memory contains a program for the collaborative risk control and management method of the geological safety of underground engineering throughout the life cycle. When the program for the collaborative risk control and management method of the geological safety of underground engineering throughout the life cycle is executed by the processor, the following steps are implemented: Geological monitoring and surface terrain scanning are performed on the target area to obtain regional geological monitoring information and surface terrain point cloud data, and regional geological structure identification and geological risk marking are carried out to construct a regional geological digital model. The construction area is monitored in real time through a set heterogeneous sensor network to obtain regional construction monitoring information. Combining with the regional geological digital model, construction risk analysis and cavity prediction are carried out to obtain regional construction risk analysis information. A carrier risk assessment model is constructed, and whether the current construction risk affects the building objects on the ground is analyzed through the regional construction risk analysis information and the regional construction monitoring information to obtain carrier risk assessment information. Based on the regional construction risk analysis information and the carrier risk assessment information, it is judged whether construction risk control and management are required. If so, a risk control and management solution is generated for collaborative risk control and management of the regional construction safety risk.

Images