A geological mineral three-dimensional modeling-based precise positioning method and system

Abstract

The application relates to the technical field of geological information, and discloses a method and system for accurate positioning based on three-dimensional modeling of geology and mineral resources. The method comprises the following steps: deploying a laser ranging, microseismic and fiber grating sensing array, and synchronously collecting multi-source deformation and stress data; after time-space alignment and filtering, a millimeter-level precision three-dimensional point cloud model is reconstructed by using an improved ICP algorithm; a safety factor cloud map is calculated by combining a geomechanics constitutive equation, a risk area is automatically circled according to three-level thresholds; and closed-loop control instructions such as voice alarm, heat map push, grouting coordinates and avoidance path are generated. The system is composed of sensing deployment, data preprocessing, point cloud reconstruction, mechanical analysis, risk delineation, instruction distribution and execution units. The application realizes millimeter-level positioning of goaf deformation, sub-hour-level early warning and full-automatic intervention, the early warning accuracy is higher than 95%, the emergency response is shortened to the minute level, and the mining safety and mining efficiency are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of geological information, and particularly relates to a three-dimensional modeling-based precise positioning method and system for geological and mineral resources. BACKGROUND

[0002] With the continuous expansion of mining scale and the continuous promotion of deep resource development, the stability monitoring of goaf has become the core link to ensure the safety production of underground mines. In traditional mining processes such as room-and-pillar method and shrinkage method, the goaf as a key area of stress redistribution of surrounding rock, its morphological evolution and roof subsidence are directly related to the probability of major disasters such as roof fall and rib spalling.

[0003] The current mainstream monitoring means still relies on manual inspection and point sensor arrangement. The former is limited by the danger of the operating environment and the discreteness of data collection, and the latter is difficult to build a complete three-dimensional space situation awareness system due to the lack of spatial coverage density and response lag. Especially in complex geological structure areas, the rock mass structure surface is developed, and the mechanical parameters are significantly heterogeneous. The existing technology cannot realize the spatio-temporal coupling analysis of the morphological dynamic change and micro-fracture precursor signal of the goaf, resulting in the lack of foresight and spatial positioning accuracy in risk warning.

[0004] Among them, the fusion perception technology based on three-dimensional laser scanning and microseismic monitoring is gradually becoming an important development direction of goaf safety evaluation. This technology aims to restore the real-time deformation process and energy release trajectory of the surrounding rock of the goaf through high-precision spatial modeling and continuous microseismic event capture. Its basic principle is to align the spatial geometric data and rock mass fracture signals in a unified coordinate system, so as to identify the stress concentration evolution path and the spatial distribution of potential unstable blocks. However, the existing systems mostly stay at the data collection and visualization level, lack deep coupling mechanism with geomechanical model, and cannot correlate and infer the spatial clustering characteristics, energy release mode of microseismic events, and rock mass structure surface occurrence, strength parameters, resulting in that the warning model stays at the experience threshold judgment stage, and it is difficult to support dynamic risk evolution deduction and active prevention and control decision.

[0005] The prior art has three major defects in constructing the goaf stability evaluation system, including fragmented perception of three-dimensional morphology, isolated interpretation of microseismic signals, and static geological model. The scanning data update cycle is long and cannot be associated with microseismic events synchronously, resulting in disconnection between morphology change and rupture process; the microseismic positioning results are not combined with the mechanical interpretation of rock mass structure characteristics, resulting in high false alarm rate and inability to identify key dangerous blocks; the geological model is mostly static profile or simplified block, without embedding time dimension and real-time monitoring feedback, resulting in lack of evolution deduction ability in stability prediction. Under complex working conditions such as deep high geostress and strong tectonic disturbance, the above defects directly lead to the system unable to form an effective early warning window before the roof fall disaster occurs, and also unable to provide spatially accurate decision basis for support scheme optimization and personnel evacuation path planning, which constitutes a technical bottleneck that needs to be broken through in mine intelligent safety management and control. SUMMARY

[0006] The application provides a geological and mineral three-dimensional modeling precise positioning method and system, which deploys a distributed multi-modal sensor array in the goaf, combines a high-precision time sequence point cloud reconstruction algorithm with a dynamic coupling model of geomechanical parameters, constructs a goaf three-dimensional deformation evolution digital twin in the whole life cycle, realizes millimeter-level spatial positioning and sub-hour-level time warning of the roof subsidence rate, surrounding rock stress redistribution path and local collapse precursor characteristics, and thus establishes an active perception and precise intervention ability of goaf stability risk in the whole three-dimensional space.

[0007] The application provides a geological and mineral three-dimensional modeling precise positioning method, which comprises:

[0008] Laser ranging sensor arrays, microseismic sensor arrays and fiber bragg strain sensor arrays are arranged on the surface of the roof, side slope and key pillar in the goaf according to a preset spatial density, the laser ranging sensor arrays are used to collect normal displacement data of the surface points of the roof and side slope, the microseismic sensor arrays are used to collect spatial coordinate and energy level data of rock mass rupture events, and the fiber bragg strain sensor arrays are used to collect axial and hoop strain distribution data inside the pillar;

[0009] The normal displacement data, spatial coordinate and energy level data, and axial and hoop strain distribution data are synchronously transmitted to a central processing unit through an industrial ring network, the central processing unit performs timestamp alignment, spatial coordinate unified conversion, noise filtering and outlier rejection operations on the original sensor data, and forms a multi-source heterogeneous monitoring data stream with a unified space-time reference;

[0010] Based on the multi-source heterogeneous monitoring data stream, a point cloud dynamic reconstruction module is called to periodically reconstruct the internal space form of the goaf at a preset time interval, the point cloud dynamic reconstruction module adopts an improved iterative closest point algorithm, introduces a geological constraint term and a deformation rate penalty term, so that the registration error of adjacent period point clouds is controlled within 0.5 mm, and a high-fidelity three-dimensional form model containing the roof curvature change gradient and the side displacement vector field is output;

[0011] The high-fidelity three-dimensional form model is input into a geomechanics coupling analysis module, the geomechanics coupling analysis module is embedded with a rock mass mechanics constitutive equation and a damage evolution criterion, combines the ore body occurrence conditions, the surrounding rock grading parameters and the historical mining disturbance records, calculates the current goaf three-dimensional stress field distribution, the plastic zone expansion boundary and the safety factor space cloud picture, the safety factor space cloud picture takes any space point in the goaf as a calculation unit, and outputs the stability margin value of the point under the current load condition;

[0012] Based on the stability margin value, a risk area automatic delineation program is started, the risk area automatic delineation program sets three-level early warning thresholds, marks a yellow early warning area when the stability margin value of a space point is lower than a first threshold, marks an orange early warning area when the stability margin value is lower than a second threshold, and marks a red early warning area when the stability margin value is lower than a third threshold, and the three-level early warning thresholds are obtained by field measurement data calibration according to the ore rock type, the buried depth and the service life;

[0013] The space coordinate ranges of the yellow early warning area, the orange early warning area and the red early warning area and the deformation rate data are input into a warning instruction generation module, the warning instruction generation module automatically generates corresponding control instruction sets according to the warning levels, the control instruction sets include sending a voice alarm instruction to a field broadcast system, pushing a three-dimensional risk heat map instruction to a dispatch center, issuing a local grouting reinforcement coordinate instruction to a support operation unit, and sending an avoidance path re-planning instruction to mining equipment;

[0014] The control instruction sets are distributed to corresponding execution terminals through a wireless communication link, and the closed-loop response of the goaf risk is realized.

[0015] The application provides a geological mineral three-dimensional modeling accurate positioning system, which comprises:

[0016] A distributed multi-modal sensing array deployment unit is used for arranging laser ranging sensor arrays, microseismic sensor arrays and fiber bragg strain sensor arrays on the roof, sides and key pillar surfaces of the goaf at a preset space density, the laser ranging sensor arrays are used for collecting normal displacement data of roof and side surface points, the microseismic sensor arrays are used for collecting spatial coordinates and energy level data of rock mass rupture events, and the fiber bragg strain sensor arrays are used for collecting axial and hoop strain distribution data in the pillar.

[0017] A multi-source data synchronous transmission and preprocessing unit is configured to synchronously transmit the normal displacement data, spatial coordinate and energy level data, and axial and ring strain distribution data to a central processing unit through an industrial ring network, and perform timestamp alignment, spatial coordinate unified conversion, noise filtering and outlier rejection operations on the original sensing data to form a multi-source heterogeneous monitoring data stream with a unified space-time reference;

[0018] A point cloud dynamic reconstruction unit is configured to reconstruct a three-dimensional point cloud of the spatial form inside the goaf based on the multi-source heterogeneous monitoring data stream at a preset time interval, and the point cloud dynamic reconstruction unit uses an improved iterative closest point algorithm, introduces a geological constraint term and a deformation rate penalty term, so that the registration error of adjacent period point clouds is controlled within 0.5 mm, and a high-fidelity three-dimensional form model containing the curvature change gradient of the roof and the displacement vector field of the side slope is outputted;

[0019] A geomechanics coupling analysis unit is configured to input the high-fidelity three-dimensional form model into a calculation module embedded with rock mass mechanical constitutive equation and damage evolution criterion, combine the occurrence conditions of the ore body, the grading parameters of the surrounding rock, and the historical mining disturbance records, calculate the three-dimensional stress field distribution, the plastic zone expansion boundary and the safety factor spatial cloud of the current goaf, and output the stability margin value of the point position under the current load condition with any spatial point of the goaf as a calculation unit;

[0020] A risk area automatic delineation unit is configured to set three-level warning thresholds based on the stability margin value, mark a yellow warning area when the stability margin value of a spatial point is lower than a first threshold, mark an orange warning area when the stability margin value is lower than a second threshold, and mark a red warning area when the stability margin value is lower than a third threshold, and the three-level warning thresholds are obtained by field measurement data calibration according to the type of ore and rock, buried depth, and service life;

[0021] A warning instruction generation and distribution unit is configured to input the spatial coordinate range and deformation rate data of the yellow warning area, orange warning area and red warning area into an instruction generation module, automatically generate a corresponding control instruction set according to the warning level, and the control instruction set includes sending a voice alarm instruction to a field broadcast system, pushing a three-dimensional risk heat map instruction to a dispatch center, issuing a local grouting reinforcement coordinate instruction to a support operation unit, and sending an avoidance path re-planning instruction to mining equipment, and distributing the control instruction set to the corresponding execution terminal through a wireless communication link;

[0022] A closed-loop response execution unit is configured to receive and execute the control instruction set to realize physical intervention and operation process adjustment of the goaf risk.

[0023] Further, the single sensor in the laser ranging sensor array has a measurement range of 0.5 m to 50 m, a repeated measurement accuracy of ±0.1 mm, a sampling frequency of 10 times per second, an installation inclination error of less than 0.5°, a sensor shell protection level higher than IP67, and a working temperature range of -20℃ to +70℃.

[0024] Further, the microseismic sensor array adopts a three-component velocity type sensing element, has a frequency response range of 2 Hz to 1000 Hz, a dynamic range greater than 120 dB, an event positioning accuracy better than 1 m in three-dimensional space, and an energy resolution capable of distinguishing micro-fracture events of 0.1 J to 10 J.

[0025] Further, the fiber grating strain sensor array has a grating length of 10 mm, a strain measurement range of -5000 με to +5000 με, a resolution of 0.1 με, a temperature compensation accuracy of ±0.5℃, and more than 64 sensing points connected in series on a single optical fiber through wavelength division multiplexing technology.

[0026] Further, the improved iterative closest point algorithm in the point cloud dynamic reconstruction module has a target function containing three terms: the first term is the traditional point-to-point distance square sum, the second term is the negative logarithm of the cosine value of the normal angle between adjacent point clouds, and the third term is the square of the displacement vector module of the corresponding points. The weighting coefficients of the three terms are dynamically adjusted according to the current goaf average subsidence rate, the weight distribution strategy is stored in the coefficient adjustment table, and the coefficient adjustment table is obtained offline by training based on historical subsidence data.

[0027] Further, the rock mass mechanical constitutive equation in the geomechanical coupling analysis module adopts a modified Hoek-Brown criterion, and its expression is:

[0028] ;

[0029] wherein is the maximum principal stress, is the minimum principal stress, For the uniaxial compressive strength of rock mass, m is the integrity coefficient of rock mass, reflecting the influence of the structural integrity of rock mass on the strength, the m value of intact rock mass is close to the test value of rock mass (such as granite m≈25), the m value of joint developed rock mass is significantly reduced (such as broken rock mass m≈5), and the specific value is obtained by checking the table through the field rock mass quality classification (such as RMR, Q system) or indoor rock mass test correction; s is the disturbance coefficient of rock mass, representing the weakening degree of external factors such as mining disturbance and groundwater soaking on the strength of rock mass, the s of undisturbed fresh rock mass is 1.0, the s of strongly disturbed or saturated rock mass is less than or equal to 0.1, and the value needs to be calibrated in combination with the mining method (such as blasting, mechanical excavation) and hydrogeological conditions in the field. The damage evolution criterion adopts a cumulative damage model based on strain energy density, and the damage variable D is defined as the ratio of the current cumulative plastic strain energy to the strain energy corresponding to the peak strength of the rock mass.

[0030] Further, the first threshold of the three-level early warning threshold in the risk area automatic delineation program is set to 1.5, the second threshold is set to 1.2, and the third threshold is set to 0.9, which is suitable for medium-hard rock mass, buried depth less than 500m, and service life less than 3 years of goaf, and for soft rock mass or deep mining conditions, the threshold is calibrated by combining field fracturing test and numerical inversion.

[0031] Further, the local grouting reinforcement coordinate instruction in the early warning instruction generation module has a spatial positioning accuracy requirement that the horizontal direction error is less than 0.3m, the vertical direction error is less than 0.2m, the grouting pressure is set to be in the range of 0.5MPa to 3MPa, the initial setting time of the grouting material is controlled to be within 30 minutes, and the final strength is higher than 20MPa.

[0032] Further, the mining equipment avoidance path re-planning instruction in the closed-loop response execution unit adopts an improved A-star algorithm, and the cost function includes three items: the first item is the path length, the second item is the integral risk value of passing through the risk area, and the third item is the total amount of path curvature change, and the weight coefficients of the three items are manually set by the dispatcher according to the current production emergency degree.

[0033] Further, the system further comprises a historical data backtracking analysis unit for storing three-dimensional shape models, stress field distribution maps, early warning records and intervention measure execution logs of all cycles, supporting time axis sliding comparison, deformation trend extrapolation prediction and intervention measure effectiveness evaluation, the data storage period of the historical data backtracking analysis unit is more than 10 years, the data compression ratio is higher than 5:1, and the retrieval response time is less than 3s.

[0034] Compared with the prior art, the advantages and positive effects of the present application are that:

[0035] The application realizes the millimeter-level spatial resolution and minute-level time resolution monitoring of the three-dimensional morphological change of the goaf for the first time by constructing a collaborative architecture of a distributed multi-modal sensor array and a high-precision point cloud dynamic reconstruction algorithm, and solves the fundamental defects that the traditional manual inspection and single-point sensor monitoring cannot form a global three-dimensional situation awareness.

[0036] By introducing a geomechanics coupling analysis module, the apparent deformation data is physically associated with the stress state inside the rock mass, the risk early warning is upgraded from an empirical threshold judgment to a quantitative prediction based on the mechanical mechanism, and the early warning accuracy is improved to more than 95%.

[0037] Through the linkage design of three-level early warning thresholds and closed-loop response execution units, an automatic process from risk identification to physical intervention is realized, the emergency response time is shortened from hours to minutes, and the occurrence of roof falling and rib spalling accidents is effectively avoided.

[0038] After the system is deployed, in the application of a typical metal mine goaf, the roof subsidence monitoring error is less than 0.8mm, the risk area positioning deviation is less than 0.4m, the early warning time is averagely 4 hours, the support material consumption is reduced by 30%, the mining operation interruption time is reduced by 70%, and the mine safety production level and resource mining efficiency are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is the overall technical scheme architecture schematic diagram of the geology and mineral resources three-dimensional modeling accurate positioning method and system proposed by the application;

[0040] Figure 2 is the core principle framework schematic diagram of the collaborative architecture of the distributed multi-modal sensor array and the high-precision time sequence point cloud reconstruction algorithm in the application;

[0041] Figure 3 is the multi-source heterogeneous monitoring data stream preprocessing and space-time reference unification process framework diagram in the application;

[0042] Figure 4 is the stress-deformation joint evolution analysis framework schematic diagram driven by the dynamic coupling model of geomechanical parameters in the application;

[0043] Figure 5 is the risk area automatic delineation and closed-loop response control logic framework diagram driven by the three-level early warning threshold in the application;

[0044] Figure 6 is the multi-level interaction relationship and data stream schematic diagram between the goaf digital twin and the execution terminal in the application; DETAILED DESCRIPTION

[0045] Please refer to Figures 1-6At the surface of the goaf roof, side slope and key pillar, laser ranging sensor arrays, microseismic sensor arrays and fiber bragg strain sensor arrays are arranged according to the preset spatial density. The laser ranging sensor array is used to collect the normal displacement data of the roof and side slope surface points. The microseismic sensor array is used to collect the spatial coordinate and energy level data of the rock mass fracture event. The fiber bragg strain sensor array is used to collect the axial and hoop strain distribution data inside the pillar. The single sensor in the laser ranging sensor array has a measurement range of 0.5m to 50m, a repeat measurement accuracy of ±0.1mm, a sampling frequency of 10 times per second, an installation angle error of less than 0.5°, a sensor shell protection level higher than IP67, and a working temperature range of -20℃ to +70℃.

[0046] The microseismic sensor array adopts a three-component velocity type sensing element, with a frequency response range of 2Hz to 1000Hz, a dynamic range greater than 120dB, an event positioning accuracy better than 1m in three-dimensional space, and an energy resolution that can distinguish micro-fracture events of 0.1J to 10J. The fiber bragg strain sensor array has a grating length of 10mm, a strain measurement range of -5000με to +5000με, a resolution of 0.1με, and a temperature compensation accuracy of ±0.5℃. More than 64 sensing points are connected in series on a single optical fiber through wavelength division multiplexing technology. During the layout process, the laser ranging sensors are arranged in a grid shape along the roof strike with a 3m interval according to the geometric shape of the goaf and the historical collapse record, and the side slope area is arranged in a quincunx shape with a 5m interval.

[0047] The microseismic sensor is arranged radially within a radius of 20m from the pillar center, ensuring that any fracture event is captured by at least 3 sensors simultaneously. The fiber bragg strain sensor is buried every 2m along the axial direction of the pillar, with each group containing 4 evenly distributed sensing points for capturing cross-sectional stress distribution. All sensors are calibrated for zero point and temperature drift before installation, and the spatial coordinates are calibrated after installation to ensure that the measurement origin is strictly aligned with the mine global coordinate system.

[0048] The normal displacement data, spatial coordinate and energy level data, and axial and hoop strain distribution data are transmitted synchronously to the central processing unit through the industrial ring network. The central processing unit performs timestamp alignment, spatial coordinate unified conversion, noise filtering and outlier rejection operations on the original sensor data to form a multi-source heterogeneous monitoring data stream with a unified space-time reference. The timestamp alignment uses the network time protocol synchronization mechanism, with a timing error of less than 1ms. The spatial coordinate unified conversion is achieved through the pre-stored sensor installation coordinate table and the mine geodetic coordinate system conversion matrix, and the 7-parameter Bursa model is introduced in the conversion process to eliminate installation bias and coordinate system rotation error.

[0049] The noise filtering is performed in two stages: the first stage is a hardware level filtering, a combination of moving average and median filtering is used at the sensor end with a window length of 10 sampling points; the second stage is a software level filtering, a wavelet threshold denoising is used at the central processing unit, a db4 wavelet basis is selected, the number of decomposition layers is 5, and a soft threshold function is used. The abnormal value elimination adopts a double check of 3 sigma criterion and isolated forest algorithm, the former eliminates outliers deviating from the mean by three times the standard deviation, and the latter identifies sparse abnormal clusters in high-dimensional space by constructing a random tree forest. The preprocessed data stream is divided by time slice, each time slice has a length of 1 minute and contains all the effective measurement values of the sensors in the period, the data packet header is attached with a timestamp, a sensor number and a data type identifier, and the data body is aligned by fixed bytes for subsequent batch processing.

[0050] Based on the multi-source heterogeneous monitoring data stream, a point cloud dynamic reconstruction module is called to reconstruct the spatial form of the goaf in three dimensions at a preset time interval. The point cloud dynamic reconstruction module uses an improved iterative closest point algorithm, introduces a geological constraint term and a deformation rate penalty term, so that the registration error of adjacent period point clouds is controlled within 0.5 mm, and a high-fidelity three-dimensional form model containing the roof curvature change gradient and the side displacement vector field is output. The objective function of the improved iterative closest point algorithm includes three terms: the first term is the square sum of the traditional point-to-point distance, the second term is the negative logarithm of the cosine value of the normal angle between the corresponding points of adjacent point clouds, and the third term is the square of the displacement vector module length of the corresponding points. The weighting coefficients of the three terms are dynamically adjusted according to the average subsidence rate of the current goaf, and the weight distribution strategy is stored in the coefficient adjustment table, which is obtained by offline training based on historical subsidence data.

[0051] The algorithm execution process is as follows: first, the surface point cloud collected by the laser ranging sensor is extracted from the current time slice data stream, each point contains three-dimensional coordinates and a normal vector; second, the transformation matrix is initialized as an identity matrix, the maximum number of iterations is set to 50, and the convergence threshold is set to 0.01 mm; third, point pair matching is performed, k-d tree is used to accelerate the nearest neighbor search, and the matching point pair needs to satisfy that the normal angle is less than 30°; then, the objective function value is calculated, the transformation matrix is updated, and the transformation matrix update uses the Levenberg-Marquardt optimization algorithm; finally, it is judged whether it converges or not, if it does not converge, the matching and updating steps are repeated, if it converges, the final transformation matrix and the registered point cloud are output.

[0052] The geological constraint term is realized by introducing the roof dip angle prior knowledge, and a penalty term of the angle between the normal vector of the roof point and the preset dip direction is added in the objective function to prevent the distortion deformation that does not conform to the geological law in the registration process. The deformation rate penalty term is realized by limiting the displacement vector module length of the same point between adjacent periods, and the penalty coefficient is inversely proportional to the historical maximum settlement rate, so as to avoid false large deformation caused by instantaneous noise. The output three-dimensional shape model is stored in the form of vertex list and face index, and the vertex attributes include coordinates, normal vector, curvature and displacement vector. The curvature is calculated by local quadratic surface fitting, and the displacement vector is obtained by the coordinate difference between the current point cloud and the corresponding point cloud of the last period.

[0053] The high-fidelity three-dimensional shape model is input into the geomechanical coupling analysis module. The geomechanical coupling analysis module is embedded with rock mass mechanical constitutive equation and damage evolution criterion, combined with ore body occurrence conditions, surrounding rock classification parameters and historical mining disturbance records, to calculate the three-dimensional stress field distribution, plastic zone expansion boundary and safety factor spatial cloud chart of the current goaf. The safety factor spatial cloud chart takes any spatial point in the goaf as a calculation unit, and outputs the stability margin value of the point under the current load condition. The rock mass mechanical constitutive equation adopts the modified Hoek-Brown criterion, and its expression is

[0054] ;

[0055] wherein is the maximum principal stress, is the minimum principal stress, is the uniaxial compressive strength of rock mass, m is the rock mass integrity coefficient, reflecting the influence of rock mass structure integrity on strength, and the value of m of complete rock mass is close to the test value of rock mass (such as granite m≈25), and the value of m of joint developed rock mass is significantly reduced (such as broken rock mass m≈5), and the specific value is obtained by looking up the table of field rock mass quality classification (such as RMR, Q system) or indoor rock mass test correction; s is the rock mass disturbance coefficient, representing the weakening degree of rock mass strength caused by external factors such as mining disturbance and groundwater soaking, and the value of s of undisturbed fresh rock mass is 1.0, and the value of s of strongly disturbed or water-soaked rock mass is less than or equal to 0.1, and the value needs to be calibrated in combination with the mining method (such as blasting, mechanical excavation) and hydrogeological conditions of the mine. The damage evolution criterion adopts the cumulative damage model based on strain energy density, and the damage variable D is defined as the ratio of the current cumulative plastic strain energy to the strain energy corresponding to the peak strength of rock mass. The calculation process adopts the finite element discretization method to divide the goaf space into tetrahedral grids, and the grid size is adaptively adjusted according to the curvature of the shape model, and the grid is encrypted to 0.2m in the large curvature area, and the grid is relaxed to 1m in the flat area.

[0056] The boundary conditions are set according to the actual occurrence state of the mine: the roof applies the overburden rock self-weight stress, the side wall applies the horizontal tectonic stress, and the floor is fixedly constrained. The material parameters are obtained by looking up the table according to the surrounding rock classification results, and the classification is based on the comprehensive evaluation of the uniaxial compressive strength of the rock, the joint development degree and the underground water condition. The implicit incremental iteration method is used for the solver, and the time step is 10 minutes. The convergence standard of each iteration step is that the force residual is less than 1% of the initial residual. The safety factor spatial cloud map output is sampled at a resolution of meters, and the safety factor of each sampling point is obtained by interpolating the calculation results of the four surrounding grid nodes. The plastic zone expansion boundary is determined by tracking the continuous area where the damage variable D is greater than 0.8, and the boundary point coordinates are output in the form of a polyline for subsequent risk area delineation.

[0057] Based on the stability margin value, the risk area automatic delineation program is started. The risk area automatic delineation program sets three warning threshold values. When the stability margin value of a certain spatial point is lower than the first threshold value, it is marked as a yellow warning area; when it is lower than the second threshold value, it is marked as an orange warning area; and when it is lower than the third threshold value, it is marked as a red warning area. The three warning threshold values are obtained by field measurement data according to the type of rock and soil, buried depth and service life. The first threshold value is set to 1.5, the second threshold value is set to 1.2, and the third threshold value is set to 0.9. This threshold value is suitable for medium-hard rock mass, buried depth less than 500m and service life less than 3 years of goaf. For soft rock mass or deep mining conditions, the threshold value is calibrated by field fracturing test and numerical inversion.

[0058] The delineation program execution steps are as follows: first, traverse all the sampling points of the safety factor spatial cloud map, compare the stability margin value with the three threshold values, and assign the corresponding color label; second, spatial clustering is performed on the points with the same color label, and the density-based clustering algorithm is used, with a neighborhood radius of 1m and a minimum point number of 10, to ensure the spatial continuity of the delineated area;

[0059] Thirdly, the geometric center, bounding box, maximum deformation rate and average safety factor of each clustering area are calculated as the area feature vector; finally, the feature vector and the area boundary coordinates are packaged to generate the warning area data package. The yellow warning area indicates that there is potential risk and the monitoring frequency needs to be increased; the orange warning area indicates that the risk is increasing and support plans need to be prepared; the red warning area indicates that it will soon lose stability and personnel need to be evacuated immediately and reinforcement needs to be started. The area data package is pushed to the warning instruction generation module through the message queue and stored in the historical database for subsequent analysis.

[0060] The spatial coordinate ranges of the yellow pre-warning area, the orange pre-warning area, and the red pre-warning area and the deformation rate data are input into the pre-warning instruction generation module. The pre-warning instruction generation module automatically generates corresponding control instruction sets according to the pre-warning levels. The control instruction sets include sending a voice alarm instruction to a field broadcast system, pushing a three-dimensional risk heat map instruction to a dispatch center, issuing a local grouting reinforcement coordinate instruction to a support operation unit, and sending an avoidance path re-planning instruction to mining equipment.

[0061] The content of the voice alarm instruction is customized according to the pre-warning level: the yellow pre-warning plays “local monitoring anomaly in the goaf, please strengthen the patrol”; the orange pre-warning plays “risk in the goaf is rising, the support team is on standby”; and the red pre-warning plays “the goaf is about to lose stability, immediately evacuate the working face”. The three-dimensional risk heat map instruction includes the pre-warning area boundary coordinates, color coding, and deformation rate values, is packaged in a WebGL format, and supports three-dimensional visualization on a large screen in the dispatch center. The local grouting reinforcement coordinate instruction includes the three-dimensional coordinates of the reinforcement points, the grouting pressure, the grouting amount, and the material ratio, and the spatial positioning accuracy requirements are that the horizontal direction error is less than 0.3 m, the vertical direction error is less than 0.2 m, the grouting pressure is set in the range of 0.5 MPa to 3 MPa, the initial setting time of the grouting material is controlled within 30 minutes, and the final strength is higher than 20 MPa.

[0062] The avoidance path re-planning instruction includes the start point, the end point, the intermediate avoidance point sequence, and the path cost evaluation value of the new path, the path generation algorithm uses an improved A-star algorithm, and the cost function includes three items: the first item is the path length, the second item is the integral risk value of passing through the risk area, and the third item is the total amount of path curvature change, and the weight coefficients of the three items are manually set by the dispatcher according to the current production emergency level. All instructions are attached with a time stamp, an instruction number, and an execution priority, and are distributed to corresponding execution terminals through a message bus.

[0063] The control instruction sets are distributed to corresponding execution terminals through a wireless communication link to realize the closed-loop response of the goaf risk. The wireless communication link uses an industrial-grade wireless Mesh network, the node spacing is less than 100 m, the transmission rate is higher than 1 MB / s, and the packet loss rate is less than 1 ‰. After the field broadcast system receives the voice alarm instruction, it immediately plays the pre-recorded voice through the specified area loudspeaker, and the playing volume is automatically adjusted according to the environmental noise to ensure that it is clear and audible within a range of 100 m.

[0064] After receiving the three-dimensional risk heat map instructions, the dispatch center highlights the warning area on the three-dimensional visualization platform, superimposes the deformation rate arrow and the safety factor contour line, and supports mouse hovering to view detailed data. After receiving the local grouting reinforcement coordinate instructions, the support operation unit automatically drives the grouting robot to move to the specified coordinates, adjusts the drilling angle and depth, and injects the quick-setting slurry at the set pressure and flow rate. The grouting process monitors the pressure and flow curve in real time, and automatically stops and alarms in case of abnormality. After receiving the re-planning instruction of the avoidance path, the vehicle-mounted controller immediately interrupts the current path tracking, loads new path data, re-plans the steering and speed curve, and ensures smooth transition to the safe area.

[0065] After completing the instructions, all execution terminals return execution confirmation messages to the central processing unit, including execution time, execution result, and exception code. The central processing unit aggregates all confirmation messages, updates the risk area state, and downgrades the red warning area to yellow if the safety factor returns to 1.5 or above after reinforcement. If the orange warning area continues to deteriorate, it is upgraded to a red warning and triggers a secondary reinforcement instruction. The entire closed-loop response process is recorded to form an intervention log for post-tracing and optimization.

[0066] The historical data backtracking analysis unit stores all three-dimensional shape models, stress field distribution maps, warning records, and intervention measure execution logs for all periods. It supports time axis sliding comparison, deformation trend extrapolation prediction, and intervention measure effectiveness evaluation. The data storage period is more than 10 years, the data compression ratio is higher than 5:1, and the retrieval response time is less than 3s. The storage architecture uses a distributed time series database, and the data is partitioned by time, with each partition containing data for one natural month.

[0067] The compression algorithm uses a combination of difference encoding and dictionary encoding. For coordinate data, the difference between adjacent points is stored, and for safety factor data, hierarchical quantization is used. The retrieval interface supports multiple condition combination queries by time range, spatial range, and warning level, and the query results are presented in the form of three-dimensional animations or statistical charts. The deformation trend extrapolation prediction uses a long short-term memory neural network model, which inputs the deformation rate sequence of the past 30 days and outputs the predicted value for the next seven days. The prediction error is evaluated by the root mean square error, and the model weights are updated regularly with new data.

[0068] The effectiveness of the intervention measures is evaluated by comparing the safety factor change rate before and after the intervention. An effective intervention is defined as a safety factor improvement rate greater than 10%, and an ineffective intervention is defined as an improvement rate less than 5%. The evaluation results are used to optimize grouting parameters and path planning strategies. The backtracking analysis unit also provides data export functions, supporting export to general three-dimensional model formats or table formats for further analysis by third-party software.

[0069] The distributed multi-modal sensor array deployment unit ensures that the sensors work stably for a long time in harsh mine environments. The sensor housing is made of stainless steel, with a corrosion-resistant coating on the surface and filled with fire-retardant silica gel inside, with an anti-shock rating of IK08. The power supply is designed in a intrinsically safe manner, with a working voltage of 24V DC and a power consumption of less than 5W, supporting backup battery power supply and can continue to work for 4 hours after power failure. The communication interface uses industrial Ethernet and RS485 dual redundancy design, automatically switches to the standby link when the main link fails. The installation bracket is designed with adjustable angle, supporting on-site fine tuning of sensor pointing to ensure that the measurement beam is perpendicular to the measured surface. The regular maintenance plan includes zero point calibration once a month, protection level detection once a quarter, and overall performance calibration once a year. The calibration process uses standard gauge blocks and vibration tables to ensure that the measurement accuracy does not drift.

[0070] The multi-source data synchronous transmission and preprocessing unit ensures the integrity and timeliness of the data stream. The industrial ring network uses a ring topology structure, supporting bidirectional data transmission, and single-point failure does not affect the overall communication. The data packet uses checksum and sequence number double protection, and the receiving end requests retransmission when the checksum fails, and triggers the data reissue mechanism when the sequence number is not continuous. The preprocessing module runs on a multi-core server, using a pipeline parallel architecture, with independent threads running in data reception, time alignment, coordinate conversion, filtering, and outlier removal stages, exchanging data through shared memory, with a processing delay of less than 50ms.

[0071] The outlier removal module has a built-in self-learning mechanism that dynamically adjusts the three-sigma threshold and isolation forest parameters based on historical data to adapt to changes in data distribution under different working conditions. The preprocessed data stream is published to the message middleware by topic, for downstream modules such as point cloud reconstruction, mechanical analysis, and early warning generation to subscribe and consume, achieving decoupling and elastic expansion between modules.

[0072] The point cloud dynamic reconstruction unit realizes high-precision three-dimensional morphological evolution tracking. The algorithm runs in a GPU-accelerated environment, using CUDA parallel computing framework to parallelize time-consuming operations such as point cloud matching, objective function calculation, and matrix update, with a single cycle reconstruction time of less than 30s. The roof dip angle prior knowledge in the geological constraint term is obtained through mine geological exploration reports and input in the form of a digital elevation model. The dip angle calculation uses local plane fitting with a fitting window of 5m x 5m. The weight coefficient adjustment table of the deformation rate penalty term contains 10 preset parameters corresponding to 10 intervals of average subsidence rates from 0.1mm / day to 10mm / day. In actual application, the corresponding weight is obtained by looking up the table according to the average subsidence rate calculated in the current period. The output three-dimensional morphological model contains not only geometric information but also timestamps, reconstruction errors, and data source identifiers, facilitating subsequent analysis and tracing. The model is updated every 10 minutes, ensuring that rapid deformation processes can be captured.

[0073] The geomechanics coupling analysis unit provides a physical mechanism driven risk assessment. The finite element mesh generation adopts a front propagation algorithm, which adaptively refines the mesh according to the surface curvature and stress gradient of the morphological model, ensuring the calculation accuracy in critical areas. The material parameter library contains 20 common rock types, each corresponding to a set of m, s, elastic modulus, and Poisson's ratio parameters. The parameter values are determined through a combination of laboratory tests and field inversions.

[0074] The solver supports parallel computing, dividing the computational domain into multiple subdomains, each assigned to an independent computing node. The nodes exchange boundary data through a message passing interface, accelerating the solution of large-scale problems. The safety factor calculation uses the strength reduction method, gradually reducing the rock mass strength parameters until the model loses stability. The instability criterion is a sudden increase in displacement or the penetration of the plastic zone, and the reduction factor is the safety factor. The calculation results are post-processed to generate spatial cloud maps, with a resolution consistent with the morphological model, supporting arbitrary profile cutting and isosurface extraction.

[0075] The risk area automatic delineation unit enables intelligent risk identification. The neighborhood radius and minimum point number in the clustering algorithm are dynamically adjusted according to the size of the goaf. For large goafs, the neighborhood radius is set to 2m and the minimum point number is set to 20. For small goafs, the neighborhood radius is set to 0.5m and the minimum point number is set to 5. The regional feature vector includes not only geometric and mechanical properties but also historical evolution trends. The delineation results are stored optimally through spatial indexing, supporting fast spatial queries. The warning threshold calibration process uses the Bayesian optimization algorithm, using historical accident data as the training set to maximize the warning accuracy and minimize the false alarm rate. The optimization results are stored as a threshold configuration file, supporting on-demand loading for different mines.

[0076] The warning instruction generation and distribution unit ensures the timeliness and accuracy of risk response. The instruction generation uses a template engine to fill in a pre-set template based on different warning levels and regional features, generating structured instruction text. The voice alarm instruction supports multi-language switching, automatically selecting the playback language based on the nationality of the workers. The three-dimensional risk heat map uses progressive loading technology, first transmitting a low-resolution overview, and then loading high-resolution details on demand, reducing network bandwidth usage. The local grouting reinforcement coordinate instruction contains a redundancy check code, which automatically requests retransmission if the decoding fails at the execution terminal. The cost function weight coefficients in the re-planning instruction of the avoidance path are set by the dispatcher through a graphical interface slider, with the interface displaying a preview of the path under different weights in real time to assist decision-making. The instruction distribution uses a publish-subscribe mode, with each execution terminal subscribing to the instruction types it is interested in, ensuring accurate delivery of instructions.

[0077] The closed-loop response execution unit completes the physical world intervention. The grouting robot is equipped with a six-degree-of-freedom manipulator and a high-precision displacement sensor, with a positioning error of less than 0.1 m, and a grouting pressure closed-loop control that adjusts the pump speed in real time to maintain the set pressure. The mining equipment avoidance path tracking uses a model predictive control algorithm that predicts the vehicle state for the next five seconds, optimizes the steering angle and throttle opening, and ensures that the path tracking error is less than 0.5 m. The execution confirmation message contains a snapshot of the device state, such as the grouting pressure curve and vehicle position trajectory, for the central processing unit to verify the effectiveness of the instruction execution. The exception handling mechanism includes instruction timeout retransmission, device fault switching to standby devices, and manual takeover forced mode, ensuring that basic safety functions can still be maintained in extreme cases.

[0078] The entire system works cooperatively through the above-mentioned units to build a three-dimensional deformation evolution digital twin of the goaf throughout its life cycle, achieving millimeter-level spatial positioning and sub-hour-level time warning of the roof subsidence rate, surrounding rock stress redistribution path, and local collapse precursor characteristics, and establishing an active perception and precise intervention capability of goaf stability risk in the entire three-dimensional space.

[0079] It should be noted that, in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0080] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for precise positioning based on three-dimensional modeling of geological minerals, characterized in that, Comprise: Laser ranging sensor array, microseismic sensor array and fiber bragg grating strain sensor array are arranged on the surface of the goaf roof, side slope and key pillar according to the preset spatial density, and multi-source heterogeneous monitoring data stream is collected; the laser ranging sensor array is used for collecting normal displacement data of the surface points of the roof and side slope; Based on the multi-source heterogeneous monitoring data stream, a point cloud dynamic reconstruction module is called to reconstruct the internal space form of the goaf as a three-dimensional point cloud with a preset time interval as a period, an improved iterative closest point algorithm is adopted, and a high-fidelity three-dimensional form model containing roof curvature change gradient and side slope displacement vector field is output; The high-fidelity three-dimensional form model is input into the geomechanics coupling analysis module, and the stability margin value of any spatial point in the goaf under the current load condition is output; Based on the stability margin value, a risk area automatic delineation program is started, the risk area automatic delineation program sets three early warning threshold values, and the early warning area is divided based on the three early warning threshold values; The three early warning threshold values are obtained by field measurement data calibration according to the type of ore and rock, buried depth and service life; The spatial coordinate range and deformation rate data of the early warning area are input into the early warning instruction generation module, and the early warning instruction generation module automatically generates corresponding control instruction set according to the early warning level; The control instruction set is distributed to the corresponding execution terminal through the wireless communication link to realize the closed loop response of the goaf risk.

2. The method according to claim 1, characterized in that, Also include: The normal displacement data, spatial coordinates and energy level data, axial and ring strain distribution data are synchronously transmitted to the central processing unit through the industrial ring network, the central processing unit performs timestamp alignment, spatial coordinate unified conversion, noise filtering and outlier rejection operation on the original sensing data, and forms a multi-source heterogeneous monitoring data stream with unified space-time reference; The timestamp alignment, spatial coordinate unified conversion, noise filtering and outlier rejection operation on the original sensing data, comprising: Synchronize all sensors built-in high-stability crystal oscillator by network time protocol synchronization mechanism, the timing error is less than 1ms; Through the pre-stored sensor installation coordinate table and the conversion matrix of mine geodetic coordinate system, the seven-parameter Bursa model is introduced to perform spatial coordinate unified conversion; Noise filtering is executed in two stages, the first stage uses sliding average and median filtering combination at the sensor end, the window length is 10 sampling points, the second stage uses wavelet threshold denoising at the central processing unit, selects db4 wavelet basis, the decomposition layer number is five, and the threshold function is soft threshold; The outlier rejection adopts three sigma criterion and isolated forest algorithm double check, the former rejects outliers deviating from the mean value by three times of standard deviation, and the latter identifies sparse abnormal clusters in high-dimensional space by constructing random tree forest.

3. The method according to claim 2, wherein, The point cloud dynamic reconstruction module is called to reconstruct the internal space form of the goaf as a three-dimensional point cloud, comprising: Extract the surface point cloud collected by the laser ranging sensor from the current time slice data stream, each point contains three-dimensional coordinates and normal vector; Initialize the transformation matrix to the unit matrix, set the maximum iteration number to fifty, and the convergence threshold to 0.01mm; Perform point pair matching, use k-d tree to accelerate nearest neighbor search, and match point pairs that meet the condition that the normal angle is less than 30°; The target function value is calculated, and the transformation matrix is updated, and the transformation matrix update adopts the Levenberg-Marquardt optimization algorithm; It is judged whether convergence is achieved, if not, the matching and updating steps are repeated, if convergence is achieved, the final transformation matrix and the registered point cloud are outputted; The target function includes three terms: the first term is the sum of squares of point-to-point distances, the second term is the negative logarithm of the cosine value of the normal angle between adjacent point clouds, and the third term is the square of the displacement vector modulus of the corresponding points, and the weighted coefficients of the three terms are dynamically adjusted according to the current goaf average subsidence rate.

4. The method according to claim 3, wherein, The high-fidelity three-dimensional morphological model is input into a geomechanical coupling analysis module, which includes: The goaf space is discretized into tetrahedral grids by using a finite element method, and the grid size is adaptively adjusted according to the curvature of the morphological model; The boundary conditions are set according to the actual occurrence state of the mine, the roof is subjected to the self-weight stress of the overburden rock, the side slope is subjected to the horizontal tectonic stress, and the floor is fixedly constrained; The material parameters are obtained by looking up a table according to the surrounding rock classification results, and the classification is based on the comprehensive evaluation of the uniaxial compressive strength of the rock, the joint development degree, and the underground water condition; The solver adopts an implicit incremental iteration method, and the time step is 10 minutes, and the convergence standard of each iteration step is that the force residual is less than 1% of the initial residual; The output safety factor spatial cloud map is sampled at a meter-level resolution, and the safety factor of each sampling point is obtained by interpolating the calculation results of the four grid nodes around it.

5. The method according to claim 4, wherein, An automatic risk area delineation program is started, which includes: All sampling points of the safety factor spatial cloud map are traversed, and corresponding color labels are assigned according to the comparison between the stability margin values and the three-level threshold values; The points with the same color label are spatially clustered, a density-based clustering algorithm is used, the neighborhood radius is set to 1m, and the minimum number of points is set to 10; The geometric center, bounding box, maximum deformation rate, and average safety factor of each cluster area are calculated as the area feature vector; The feature vector and the area boundary coordinates are packaged to generate a warning area data package.

6. The method according to claim 5, wherein, The warning instruction generation module automatically generates corresponding control instruction sets according to the warning levels, which include: The warning area is divided based on the three-level warning threshold values, including: when the stability margin value of a spatial point is lower than the first threshold value, it is marked as a yellow warning area, when it is lower than the second threshold value, it is marked as an orange warning area, and when it is lower than the third threshold value, it is marked as a red warning area; The content of the voice alarm instruction is customized according to the warning level, the yellow warning plays "local monitoring anomaly in the goaf, please strengthen the patrol", the orange warning plays "risk in the goaf is rising, support teams are on standby", and the red warning plays "the goaf is about to lose stability, immediately evacuate the working face"; The three-dimensional risk heat map instruction includes the warning area boundary coordinates, color coding, and deformation rate value, and is packaged in WebGL format; The local grouting reinforcement coordinate instruction includes the three-dimensional coordinates of the reinforcement points, the grouting pressure, the grouting amount, and the material ratio, and the spatial positioning accuracy requirement is that the horizontal error is less than 0.3m and the vertical error is less than 0.2m; The avoidance path re-planning instruction includes the start point, end point, and intermediate avoidance point sequence of the new path, and the path cost evaluation value, and the path generation algorithm uses an improved A-star algorithm.

7. The method according to claim 6, characterized in that, The control instruction sets are distributed to the corresponding execution terminals through a wireless communication link, which includes: The field broadcast system receives the voice alarm instruction, and plays a pre-recorded voice through a speaker in a specified area. The volume of the played voice is automatically adjusted according to the environmental noise. After receiving the three-dimensional risk heat map instruction, the dispatching center highlights the warning area on the three-dimensional visualization platform and superimposes the deformation rate arrow and the safety factor contour line. After receiving the local grouting reinforcement coordinate instruction, the support operation unit automatically drives the grouting robot to move to the specified coordinate, adjusts the drilling angle and depth, and injects the quick-setting slurry at the set pressure and flow rate. After receiving the re-planning instruction of the avoidance path, the mining equipment immediately interrupts the current path tracking, loads the new path data, and re-plans the steering and speed curve.

8. The method according to claim 7, characterized in that, The three weighting coefficient dynamic adjustment strategies in the improved iterative closest point algorithm are obtained offline by training based on historical settlement data and are stored in a coefficient adjustment table. The corresponding weight is obtained by looking up the table according to the average settlement rate calculated in the current period.

9. A geological mineral three-dimensional modeling precision positioning system based on, characterized in that, The improved iterative closest point algorithm includes: A distributed multi-modal sensor array deployment unit is used to deploy laser ranging sensor arrays, microseismic sensor arrays, and fiber bragg grating strain sensor arrays on the surface of the goaf roof, side slope, and key pillar at a predetermined spatial density. The laser ranging sensor arrays are used to collect normal displacement data of the roof and side slope surface points. The microseismic sensor arrays are used to collect spatial coordinates and energy level data of rock mass fracture events. The fiber bragg grating strain sensor arrays are used to collect axial and hoop strain distribution data inside the pillar. A multi-source data synchronous transmission and preprocessing unit is used to transmit the normal displacement data, spatial coordinates and energy level data, axial and hoop strain distribution data to the central processing unit through an industrial ring network, and perform timestamp alignment, spatial coordinate unified conversion, noise filtering and outlier rejection operations on the original sensor data to form a multi-source heterogeneous monitoring data stream with a unified space-time reference. A point cloud dynamic reconstruction unit is used to reconstruct the internal space form of the goaf based on the multi-source heterogeneous monitoring data stream with a preset time interval as the period. The point cloud dynamic reconstruction unit uses an improved iterative closest point algorithm, introduces a geological constraint term and a deformation rate penalty term, and controls the registration error of adjacent period point clouds within 0.5 mm. A high-fidelity three-dimensional form model containing the roof curvature change gradient and the side slope displacement vector field is output. A geomechanics coupling analysis unit is used to input the high-fidelity three-dimensional form model into the calculation module of the embedded rock mass mechanics constitutive equation and damage evolution criterion, combine the ore body occurrence conditions, surrounding rock classification parameters, and historical mining disturbance records, calculate the current goaf three-dimensional stress field distribution, plastic zone expansion boundary, and safety factor spatial cloud map. The safety factor spatial cloud map takes any spatial point in the goaf as a calculation unit and outputs the stability margin value of the point under the current load condition. The risk area automatic delineation unit is configured to set three-level early warning thresholds based on the stability margin value, mark a yellow early warning area when a certain space point stability margin value is lower than a first threshold, mark an orange early warning area when the stability margin value is lower than a second threshold, and mark a red early warning area when the stability margin value is lower than a third threshold, wherein the three-level early warning thresholds are obtained by field measurement data calibration according to the ore rock type, buried depth and service life; The early warning instruction generation and distribution unit is configured to input the spatial coordinate range of the yellow early warning area, the orange early warning area and the red early warning area and the deformation rate data into an instruction generation module, automatically generate a corresponding control instruction set according to the early warning level, and distribute the control instruction set to a corresponding execution terminal through a wireless communication link, wherein the control instruction set includes sending a voice alarm instruction to a field broadcast system, pushing a three-dimensional risk heat map instruction to a dispatch center, issuing a local grouting reinforcement coordinate instruction to a support operation unit, and sending an avoidance path re-planning instruction to a mining equipment. The closed-loop response execution unit is configured to receive and execute the control instruction set to realize physical intervention and operation process adjustment of the goaf risk.

10. The geological mineral resources three-dimensional modeling precision positioning system based on claim 9, characterized in that, The point cloud dynamic reconstruction unit is configured to: extract surface point clouds collected by a laser ranging sensor from a current time slice data stream, each point containing three-dimensional coordinates and a normal vector; initialize a transformation matrix as an identity matrix, set a maximum iteration number as 50, and set a convergence threshold as 0.01 mm; perform point pair matching, use a k-d tree to accelerate nearest neighbor search, and match point pairs that meet a normal angle less than 30°; calculate a target function value, update the transformation matrix, and use a Levenberg-Marquardt optimization algorithm to update the transformation matrix; determine whether convergence is achieved, repeat the matching and updating steps if convergence is not achieved, and output a final transformation matrix and a registered point cloud if convergence is achieved. The target function includes three terms: a first term is a point pair distance square sum, a second term is a negative logarithm of a corresponding point normal angle cosine value between adjacent point clouds, and a third term is a square of a corresponding point displacement vector module length, and the three weighting coefficients are dynamically adjusted according to the current goaf average subsidence rate.

Images