Intelligent prediction system for coal mine goaf collapse based on multi-source data fusion

Abstract

The present application relates to the technical field of intelligent treatment of coal mine goaf, and discloses a coal mine goaf collapse intelligent prediction system based on multi-source data fusion, wherein the system obtains goaf exploration data through direct current method, geological radar and two-dimensional horizontal wave seismic method, adopts W-SSIM constrained multi-physical field joint inversion to construct a three-dimensional geological model, and realizes minute-level local smooth evolution by means of Kalman-finite element incremental updating, automatically generates a curtain hole network based on three-dimensional envelope-diffusion radius-Voronoi-Lloyd algorithm combined with drill rig accessibility verification, couples grouting monitoring data by using Bayesian network-MCMC sampling and establishes a four-dimensional collapse probability body, and outputs a 0.5m precision risk heat map through GPU parallel rendering, thereby realizing 28-day rolling prediction in the future. The present application fills the technical blank of digital closed loop of coal mine goaf treatment through the whole-link integrated closed loop architecture of "exploration-modeling-treatment-monitoring-prediction-display".

Description

Technical Field

[0001] This invention relates to the field of intelligent governance technology for coal mine goaf areas, and in particular to an intelligent prediction system for coal mine goaf area collapse based on multi-source data fusion. Background Technology

[0002] With the large-scale mining of coal resources, ground subsidence, building damage, and secondary disasters caused by mine goaf have become increasingly prominent, becoming a key bottleneck restricting the safety and sustainable development of mining areas. Traditional goaf surveys mostly use a single geophysical exploration method for "point-line" surveys, which have limited data dimensions and accuracy. Furthermore, the lack of synchronization and cross-verification mechanisms between different methods leads to significant uncertainties in the understanding of the boundaries, spatial morphology, and degree of danger of hidden goaf areas, making it difficult to meet the needs of precise governance under deep and complex conditions.

[0003] In the 3D geological modeling stage, existing technologies are generally based on a single, static global inversion strategy. When new measurement data is added, the entire model needs to be recalculated, which is time-consuming and computationally expensive. Moreover, "step-like" abrupt changes often occur at the junction of the old and new models. At the same time, the inversion objective function often takes the minimum data residual as the sole criterion, without considering the structural consistency of multiple physical parameters such as resistivity and wave velocity at the geological boundary. This results in the position and shape error of the roof and floor of the goaf generally exceeding ±0.5 m, which brings systematic deviations to the subsequent grouting design and engineering quantity estimation.

[0004] As the mainstream method for goaf treatment, curtain grouting has long relied on engineers' experience for its hole network layout. It adopts a two-dimensional planar pattern of "equal spacing + uniform densification", which ignores the coupling constraints of three-dimensional envelope shape, grout diffusion radius and drilling rig accessibility. This often results in discontinuous curtains due to sparse hole positions or waste of engineering work due to excessively dense hole positions. The construction site also needs to repeatedly adjust the hole positions and move the machine, which increases the cost per hole and prolongs the construction period. Moreover, it still cannot guarantee the effective closure of the goaf boundary by the curtain.

[0005] In terms of risk prediction, traditional methods mainly rely on static rock movement calculations or empirical statistical formulas, which only provide a binary conclusion of "whether it is stable" and lack quantitative answers to "when, where, and what the probability is". In addition, the monitoring data of the grouting process and the geological model are disconnected from each other, failing to form a real-time updated risk field. Managers cannot intervene in advance based on dynamic information. At the same time, existing 3D geological software relies on localized professional workstations, the model size is large and the loading is slow, and remote consultation and mobile viewing are difficult, which seriously restricts the timeliness and accessibility of disaster early warning. Summary of the Invention

[0006] This invention provides an intelligent prediction system for coal mine goaf collapse based on multi-source data fusion to solve existing technical problems.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides an intelligent prediction system for coal mine goaf subsidence based on multi-source data fusion, comprising:

[0009] The multi-source data acquisition module is used to acquire exploration data from coal mine goaf areas and perform preliminary processing and verification.

[0010] The 3D modeling module is used to construct a 3D geological model of the goaf area based on the exploration data using a fusion inversion algorithm, and generate an editable meshed geological body.

[0011] The grouting boundary planning module is used to generate a three-dimensional envelope of the abnormal response of the goaf boundary based on the three-dimensional dynamic model, and then delineate the curtain hole network and borehole depth.

[0012] The drilling scheme planning module generates construction trajectories and drilling control parameters based on the curtain hole pattern and drilling depth.

[0013] The operation monitoring module is used to collect measured data during the drilling and grouting process and perform quantitative evaluation to obtain quantitative evaluation results;

[0014] The risk assessment module is used to couple the three-dimensional geological model and the quantitative assessment results to establish a four-dimensional collapse probability volume and output a risk heat map.

[0015] The process visualization module is used to visually display the operation process of the above modules.

[0016] The beneficial effects of the technical solution provided by this invention include at least the following:

[0017] This invention proposes a "multi-source data fusion-closed-loop governance" architecture, which integrates DC resistivity, ground-penetrating radar and shear wave 2D seismic data through hardware-level millisecond synchronization and cross-validation. This provides multi-dimensional information on resistivity, wave velocity and wave impedance that are complementary for goaf areas, improving the accuracy of hidden cavity detection and realizing a closed-loop iteration of "exploration-modeling-governance-monitoring". This significantly reduces safety blind spots caused by information lag.

[0018] This invention is based on the "W-SSIM constrained cross-gradient joint inversion + Kalman-finite element incremental update" technology. After the system adds new data, it only needs to recalculate the local mesh, which significantly improves the computational efficiency. At the same time, through multi-physics structural consistency constraints, the elevation error of the top and bottom plates is controlled within ±0.2m, providing a high-confidence three-dimensional geological base map for the automatic design of curtain hole depth, avoiding the resource waste and model mutation problems caused by traditional global recalculation.

[0019] This invention utilizes a combined optimization of "three-dimensional envelope-diffusion radius-Voronoi-Lloyd" to automatically generate a double-layer curtain hole network, ensuring that the hole spacing remains within the economic range. This reduces ineffective drilling while maintaining curtain continuity. Combined with rigid body collision detection of the drilling rig, it outputs reachable hole positions and drilling rig station location diagrams in advance, reducing the number of on-site machine relocations and shortening the average construction time per hole, significantly saving material, manpower, and equipment operation and rental costs.

[0020] This invention utilizes the Bayesian network-MCMC method to construct a four-dimensional collapse probability volume, which can continuously output risk heat maps with an accuracy of 0.5m×0.5m×0.5m for the next 1 to 28 days. This upgrades the traditional "static safety assessment" to "spatiotemporal probability prediction," providing quantifiable decision-making basis for mining areas. Combined with lightweight visualization using WebGL+LOD, managers can navigate the million-level grid model in real time on a browser or mobile device without the need for specialized software, enabling remote consultation and immediate early warning, and comprehensively improving the intelligent and refined prevention and control level of subsidence disasters in mining areas. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a system flowchart provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0024] This embodiment provides an intelligent prediction system for coal mine goaf subsidence based on multi-source data fusion. Please refer to... Figure 1 This is a system flowchart provided in an embodiment of the present invention.

[0025] The intelligent prediction system for coal mine goaf collapse based on multi-source data fusion in this embodiment includes:

[0026] I. Multi-source data acquisition module: used to acquire exploration data from coal mine goaf areas and perform preliminary processing and verification;

[0027] The multi-source data acquisition module includes a data acquisition unit, a data preprocessing unit, a data integrity verification unit, and a data compression and transmission unit;

[0028] The data acquisition unit is used to conduct preliminary exploration of the coal mine goaf using direct current resistivity, ground-penetrating radar and shear wave two-dimensional seismic method, and to obtain the resistivity volume of direct current resistivity, the electromagnetic wave velocity volume of ground-penetrating radar and the impedance volume of shear wave two-dimensional seismic wave.

[0029] The data preprocessing unit is used to synchronize the data acquisition process of DC resistivity method, ground-penetrating radar and shear wave 2D seismic method in milliseconds through timestamp alignment and trigger signal control, and further suppress noise and remove outliers from the acquired exploration data. The exploration data includes DC resistivity volume, ground-penetrating radar electromagnetic wave velocity volume and shear wave 2D seismic wave impedance volume.

[0030] The data integrity verification unit is used to verify the integrity of the collected exploration data through redundant acquisition comparison and cross-validation algorithms, mark exploration areas with missing or conflicting data, and trigger the supplementary acquisition process.

[0031] The data compression and transmission unit is used to compress exploration data in real time using a wavelet transform compression algorithm, and transmits the compressed exploration data to the 3D modeling module via a 5G / fiber hybrid network.

[0032] It should be noted that the direct current method involves passing current through power supply electrodes into the underground area and measuring the potential difference and resistivity difference to reveal the distribution of water filling, fissures and cavities in coal mine goaf areas. It has the advantages of large detection depth, lightweight equipment and strong resistance to electromagnetic interference, and is one of the preferred methods for rapid survey of coal mine goaf areas.

[0033] Ground-penetrating radar (GPR) emits high-frequency electromagnetic waves into the ground and receives reflected signals. It analyzes the roof cracks, cavities, and water-rich zones in the goaf based on the amplitude, travel time, and waveform changes of the reflected signals. It can perform continuous scanning, whether vehicle-mounted or handheld, and features high resolution, real-time imaging, and high positioning accuracy.

[0034] Shear wave 2D seismic method: This method involves artificially generating shear waves (S-waves) and recording their propagation time, velocity, and amplitude changes in the strata. It can then invert the elastic modulus and fracture development of the goaf, effectively identifying cavity boundaries and loose roof zones. It combines the advantages of detection depth and structural resolution.

[0035] Redundant acquisition and comparison: This involves repeatedly deploying multiple sets of sensors on the same survey line or point, independently acquiring data, and then performing differential processing to quantify the levels of random noise, system drift, and environmental interference. This ensures data reliability from the source and is a common method for on-site quality control in environmental detection.

[0036] Cross-validation algorithm: This involves spatially overlaying and statistically testing data obtained from different methods (DC electrical resistivity, ground-penetrating radar, and shear wave 2D seismic method) in the same area, and using information entropy or Bayesian criteria to evaluate data consistency.

[0037] Wavelet transform compression algorithm: It uses wavelet basis functions to decompose exploration signals into multi-scale coefficients, retains the main energy through threshold quantization, and achieves lossless compression of more than 10:1. It has the characteristics of low algorithm complexity and suitability for embedded processing, and can significantly reduce 5G / fiber transmission bandwidth and storage costs.

[0038] 5G / fiber hybrid network: This involves deploying 5G macro and micro base stations and fiber optic ring networks in the mining area for efficient data transmission. 5G provides low-latency wireless access with a latency of 20ms, while fiber optics provides backbone bandwidth of over 10Gbps. The two switch seamlessly to ensure real-time transmission and continuous updates of exploration data, making it particularly suitable for the complex electromagnetic environment of the mining area.

[0039] II. 3D Modeling Module: Used to construct a 3D geological model of the goaf based on exploration data using a fusion inversion algorithm, and generate an editable meshed geological body;

[0040] The 3D modeling module includes a multiphysics joint inversion unit, a dynamic update and incremental fusion unit, a gridded geological body generation unit, a priori geological constraint library unit, and a lightweight output interface unit.

[0041] The multiphysics joint inversion unit is used to spatially register exploration data on the same three-dimensional grid, and to iteratively invert the data through a cross-gradient objective function constrained by W-SSIM to obtain a three-dimensional geological model that includes the extent, depth and roof and floor height of the goaf.

[0042] The dynamic update and incremental fusion unit is used to perform local incremental inversion only on the affected grids in the 3D geological model when the exploration data output by the multi-source data acquisition module is updated, and further to perform spatiotemporal fusion update with the new inversion results and the existing 3D geological model through the Kalman filter-finite element coupling framework, and record the version number and timestamp of each update.

[0043] The gridded geological body generation unit is used to partition the three-dimensional geological model using an octree-tetrahedral hybrid grid, dividing the goaf and surrounding rock of the goaf. The grid size is then adaptively refined based on the distance from the boundary of the goaf to generate a gridded geological body. Each grid cell in the gridded geological body is then assigned an editable attribute field to obtain an editable gridded geological body. The attribute fields include lithology code, resistivity, wave velocity, porosity, and void probability.

[0044] The prior geological constraint library unit is used to store hard constraint data of the goaf. The hard constraint data includes existing roadways, existing borehole columns and distributions in the mining area, as well as geological profiles at various depth gradients. During the inversion process, the hard constraint data is embedded into the objective function through a global penalty term as a constraint condition for the generation of the three-dimensional geological model.

[0045] The lightweight output interface unit is used to simplify the LOD of the 3D geological model and output standard format file packages including GeoTIFF, VTK and IFC, which can be called by the grouting boundary planning module, drilling scheme planning module, operation monitoring module, risk assessment module and process visualization module, and supports providing real-time model slicing services in the form of RESTful API.

[0046] It should be noted that the objective function of the cross gradient constrained by W-SSIM is to introduce weighted structural similarity (W-SSIM) as a regularization term on the basis of cross gradient, so as to make multiple physical parameters jump synchronously at the boundary while maintaining the consistency of image structure.

[0047] Local incremental inversion: When a new survey line or survey point is added, only the local area centered on the new data and with an influence radius of 3 times the grid step size is re-inverted, while the rest of the grid is frozen. Nested grids and pre-conditional conjugate gradients are used to accelerate the process, reducing the computational load of the system and effectively meeting the real-time update requirements.

[0048] Kalman filter-finite element coupled framework: The three-dimensional geological model is discretized into finite element state vectors. The covariance is calculated forward by finite element as the prediction step and the state is corrected by data residuals as the update step. Kalman filter is used to fuse the new inversion results with the old model in the spatiotemporal domain. This method can take into account the physical consistency and statistical optimality of the model and achieve millimeter-level smooth transition of model deformation.

[0049] Octree-Tetrahedral Hybrid Mesh: This method uses octree cubes to quickly partition uniform surrounding rock, while inserting tetrahedral partitions for complex interfaces such as goaf boundaries and roadways to obtain a hybrid mesh. This method can balance storage efficiency and geometric fitting accuracy.

[0050] LOD simplification: This method uses the QEM algorithm to fold the distant mesh step by step to generate a multi-level detail model. At the same time, when the viewing distance is greater than 200m, the number of patches is reduced and the frame rate is increased. This method can retain key geological interface features while reducing the amount of computation.

[0051] GeoTIFF: This method embeds geographic coordinates and projection information into TIFF image files. It supports 16-bit floating-point storage of raster attributes such as resistivity and cavity probability, and can be directly called by GIS software. This method can achieve seamless integration of geophysical exploration results with tunnel maps and geological profile maps.

[0052] VTK (Visualization Toolkit Open Source Format): Provides C++ / Python interfaces, supports the output of unstructured tetrahedrons, octrees and attribute fields, and can directly call ParaView for 3D clipping and isosurface extraction. This format can be used to achieve cross-platform rapid visualization of grouting planning and risk heat map rendering.

[0053] IFC (Industry Foundation Classes BIM International Standard): This format is used to write three-dimensional geological models as solid objects, which can be read by software such as Revit and Bentley, realizing integrated collaborative design of "geology-engineering" and effectively guiding subsequent curtain grouting construction.

[0054] RESTful API: This API returns model slices with specified range, precision, and attributes via HTTP GET / POST in JSON format. It supports timestamp version control, and after the 3D modeling module is updated, the client only pulls the difference mesh without downloading the full data, effectively reducing communication traffic and meeting the needs of on-demand WebGL rendering and remote access from mobile devices.

[0055] 3. Grouting Boundary Planning Module: Used to generate a three-dimensional envelope of the abnormal response of the goaf boundary based on the three-dimensional dynamic model, and then delineate the curtain hole network and borehole depth;

[0056] The grouting boundary planning module includes a boundary envelope extraction unit, a curtain hole network generation unit, a hole depth calculation unit, and a drilling rig accessibility verification unit.

[0057] The boundary envelope extraction unit is used to extract the probability gradient maxima surface of a 3D geological model using a 3D Sobel-Canny hybrid edge detection algorithm, generating an initial envelope, and further employing morphological closing operations and... The algorithm fills holes and removes redundant patches from the initial envelope to generate a three-dimensional envelope of the abnormal response at the boundary of the goaf.

[0058] The curtain perforated mesh generation unit is used to generate a double-layer perforated mesh skeleton line at a depth of 0.3 to 1.0 times the envelope thickness on the surface and inside of the three-dimensional envelope of the abnormal response at the goaf boundary, with the grout diffusion radius as the reference. Using empirical formulas as constraints, the hole arrangement of the skeleton lines in a double-layer perforated mesh is optimized through an improved Voronoi–Lloyd iterative algorithm, allowing for arbitrary hole spacing. Finally, the GIS layer of the curtain perforated mesh is output. The GIS layer of the curtain perforated mesh includes the hole number, hole coordinates, and the designed borehole azimuth and inclination angle.

[0059] The borehole depth calculation unit is used to section the 3D geological model based on the curtain borehole network GIS layer, extract the burial depth curves of the goaf roof and floor, and automatically calculate the borehole depth for each hole location with a hard constraint of "penetrating the goaf floor by 0.5m and being 1.0m away from the goaf roof" and an objective function of "minimizing drilling workload". This is achieved when updates to the 3D geological model cause changes in the elevation difference between the roof and floor. The borehole depth is recalculated at 0.2m.

[0060] The drilling rig accessibility verification unit is used to import the drilling rig's equipment parameter library, combine it with the 3D geological model to perform collision detection and accessibility simulation on the curtain hole network, eliminate holes that cannot be constructed due to spatial interference or equipment limitations, and automatically generate a list of alternative holes and a drilling rig station layout diagram. The equipment parameter library includes the drilling rig's minimum turning radius and maximum pitch angle.

[0061] It should be noted that the 3D Sobel–Canny hybrid edge detection algorithm first calculates the gradient vector of each attribute volume in the 3D geological model using a 3D Sobel convolution kernel, and then performs non-maximum suppression and double threshold tracking along its gradient direction to obtain a closed, single-pixel boundary surface. This method combines the noise suppression of Canny with the directional sensitivity of Sobel, and can accurately capture weak anomaly edges in goaf areas, effectively reducing the false detection rate.

[0062] Morphological closing operation and Algorithm: First, voxel-level closure operations are performed on the initial edge surfaces of the 3D geological model, filling internal holes and smoothing burrs with 5×5×5 structural elements. Then, the discrete point cloud is transformed into... Redundant triangular pieces are automatically eliminated by using the rolling ball radius α, while retaining the topologically correct three-dimensional envelope, so that the subsequent hole mesh layout is not affected by "island" facets.

[0063] The improved Voronoi–Lloyd iterative algorithm generates a Voronoi diagram on the envelope surface, using the grouting diffusion radius R as a constraint. In each iteration, the generator is moved to the centroid of its Voronoi element and a boundary projection is made. At the same time, an annealing coefficient is added to avoid local oscillations. This method can ensure that the distance between any two holes is between 0.8R and 1.2R, taking into account both the continuity and economy of the curtain perforated mesh.

[0064] GIS layer: This refers to the output of the curtain wall perforation mesh in Shapefile format, including its hole number, X / Y / Z coordinates, azimuth, and dip angle, as well as the coordinate system and mine roadway information. Figure 1 As such, geological models and tunnels can be directly overlaid in ArcGIS and QGIS, enabling "one map" review and facilitating simultaneous viewing and dynamic adjustments by the design, construction, and supervision parties.

[0065] Collision detection and accessibility simulation: The drilling rig is simplified into a multi-segment rigid body model, and parameters such as minimum turning radius and maximum pitch angle are imported. In the three-dimensional geological-tunnel space, OBB tree is used to perform fast collision detection. For unreachable boreholes, nearby alternative points are automatically searched and drilling rig station layout diagram is output to avoid secondary on-site relocation, thereby shortening the construction period and reducing equipment operating costs.

[0066] IV. Drilling Scheme Planning Module: Generates construction trajectory and drilling control parameters based on curtain hole mesh and drilling depth;

[0067] The drilling scheme planning module includes a construction path generation unit, a borehole trajectory optimization unit, and a control parameter generation unit;

[0068] The construction path generation unit is used to generate construction trajectories based on the curtain hole network GIS layer, with "shortest total footage" and "minimum number of drilling rig relocations" as dual objective functions, through an improved genetic-ant colony hybrid algorithm. The construction trajectory includes the drilling rig station sequence, single-station hole sequence, and estimated pure drilling time.

[0069] The borehole trajectory generation unit is used to perform A* path search in the three-dimensional geological model based on the curtain hole network GIS layer and the borehole depth, and further generate the borehole trajectory by fitting the minimum bending radius constraint of the drill rod of the drilling rig through a quintic spline curve.

[0070] The control parameter generation unit is used to discretize the construction trajectory and drilling trajectory into a sequence of control nodes using the spline trajectory discretization method, generate operating control parameters, and push the operating control parameters to the drilling rig PLC system in JSON format.

[0071] It should be noted that the improved genetic-ant colony hybrid algorithm first uses a genetic algorithm to globally search and generate an initial hole sequence population. Then, the initial hole sequence population is selected, crossovered, and mutated to obtain a better solution. The better solution is then transformed into the initial pheromone distribution of the ant colony. The positive feedback of the ant colony is used to quickly converge to the bi-objective Pareto front of "shortest total advance + minimum relocation". An adaptive pheromone evaporation factor and an elite retention strategy are introduced for iteration. Compared with the single algorithm, this effectively reduces the cost and avoids premature convergence.

[0072] A* Path Search: In the mesh of the 3D geological model, the bending radius of the drill pipe and the fault zone cost are used as heuristic weights to perform the best priority search from the borehole to the target point. The "remaining distance + maximum allowable curvature" dual constraint is used to ensure that the generated trajectory is both the shortest and drillable, and discrete waypoint data is output for subsequent processes.

[0073] Quintic spline curve: Using discrete waypoints generated by A* as nodes, a quintic parametric spline is constructed. The borehole position, direction and curvature are strictly matched at the start and end waypoints to ensure continuous force on the drill rod. The "smoothest" trajectory curve is achieved by minimizing the curvature integral, which can effectively reduce the friction and torque fluctuation of the drill rod and meet the minimum bending radius limit of the drill rig.

[0074] Spline trajectory discretization method: The five-order spline curve is resampled with equal arc length according to the three principles of "displacement-time-curvature" to generate a control node sequence with a step length of 0.1m. Each node is assigned fields such as XYZ coordinates, tool face angle and inclination rate to form a JSON array. The PLC of the drilling equipment can directly read the running posture according to the node control node sequence number. This method can realize closed-loop servo control and reduce the real-time calculation burden of the host computer.

[0075] Drilling rig PLC system: This is the built-in motion control module of the drilling rig equipment. It receives JSON arrays via Ethernet and parses the node sequence into servo motor speed, feed pressure and tool face angle control commands. It performs closed-loop adjustment at a frequency of 1kHz and supports MWD data feedback.

[0076] V. Operation Monitoring Module: Used to collect measured data during the drilling and grouting process and perform quantitative evaluation to obtain quantitative evaluation results;

[0077] The operation monitoring module includes a measurement while drilling and real-time correction unit, a grouting monitoring unit, and a real-time engineering quantity statistics unit;

[0078] The measurement-while-drilling (MWD) and real-time deviation correction unit is used to acquire real-time MWD data at a frequency of 1Hz during the drilling process via wired drill pipe + MWD. ​​The MWD data includes drill bit position, well inclination angle, and azimuth angle, and is matched with the borehole trajectory control parameters through a sliding window. When the horizontal projection deviation... 0.1m / 100m or well inclination angle deviation At 1°, the PID correction algorithm is triggered to output a real-time correction command to the drilling rig PLC system;

[0079] The grouting monitoring unit is used to acquire grouting measurement data in real time through pressure transmitters, turbine flow meters and ultrasonic densitometers during the grouting process of the grouting machine. The grouting measurement data includes grouting pressure, grouting flow rate and grout density.

[0080] The real-time engineering quantity statistics unit is used to calculate drilled linear meters, cumulative pure drilling time, mud consumption and drill bit wear index based on drilling measurement data, and to calculate the filling stone body and grouting pressure-flow-time curve based on grouting measurement data. The results of the above calculations are packaged as quantitative assessment results and pushed to the risk assessment module.

[0081] It should be noted that wired drill pipe + MWD means that a cable is built into the drill string to enable real-time power supply and two-way communication between the bottom hole sensor and the surface. It is not affected by mud pulse attenuation, the data transmission depth is >2000m and the data delay is <0.5s, providing a high-confidence input for automatic deviation correction.

[0082] PID correction algorithm: The trajectory deviation e(t) and its integral and derivative constitute a three-dimensional PID controller, which outputs drilling pressure, rotation speed and tool face correction. After the parameters are tuned by ZN, the horizontal projection error can be pulled back to within 0.1m / 100m within a 3m well section, avoiding frequent start and stop of the drilling tool and extending the life of the drill bit.

[0083] Pressure transmitter: It adopts a diffused silicon + diaphragm isolation structure, with a range of 0–20MPa and an accuracy of ±0.1%FS. It can provide millisecond-level stable data for pressure-flow-time curves. At the same time, it has built-in temperature compensation and overpressure protection, which can monitor grouting pressure changes in real time to prevent formation fracturing or grouting equipment blockage.

[0084] Turbine flow meter: It adopts a magnetoelectric conversion turbine structure. The slurry drives the turbine to cut the magnetic lines of force to generate pulses. After being digitized by its fieldbus, the instantaneous flow rate and cumulative grouting volume are output. The range is 0.5–150L / min and the accuracy is ±0.5%RS.

[0085] Ultrasonic density meter: It uses the time-of-flight method to measure the sound velocity of the slurry and combines it with temperature compensation to calculate the slurry density. The range is 1.0–2.5 g / cm³ and the accuracy is ±0.5 kg / m³. It is a non-contact measurement and can be directly clamped to the outer wall of the pipe. It can detect changes in the water-cement ratio of the slurry in real time to ensure that the strength of the filling stone body is consistent with the design.

[0086] VI. Risk Assessment Module: Used to couple the three-dimensional geological model and quantitative assessment results, establish a four-dimensional collapse probability volume and output a risk heat map;

[0087] The risk assessment module includes a multi-source data coupling unit, a residual cavity volume calculation unit, a grouting and filling progress calculation unit, a formation stress field calculation unit, a collapse probability calculation unit, and a risk heat map rendering unit.

[0088] The multi-source data coupling unit is used to perform millisecond-level synchronous registration of the three-dimensional geological model and the quantitative evaluation results through the spatiotemporal unified indexing algorithm, and to fill the missing grids in the registration process through the Kriging interpolation algorithm to generate the grouting three-dimensional geological model. The grouting three-dimensional geological model includes the residual cavity, the filling progress body and the grouting pressure field.

[0089] The residual cavity volume calculation unit is used to perform volume integration on the residual cavity body in the three-dimensional geological model after grouting, with a micro-element of 0.1m×0.1m×0.1m, to obtain the residual cavity volume of the partitioned and layered areas. Furthermore, the residual cavity evolution volume is predicted in the next 1 day, 7 days, 14 days and 28 days through the time decay function.

[0090] The grouting and filling progress calculation unit is used to predict the filling rate improvement curves for the ungrouted area in the three-dimensional geological model after grouting, based on the filling progress volume, through a three-dimensional convolutional LSTM network.

[0091] The formation stress field calculation unit is used to call the FLAC3D fast solver to perform elastoplastic numerical simulation of the lithology, porosity, and burial depth parameters of the 3D geological model after grouting, and to obtain the minimum principal stress. The stress distribution body is calculated, and the stress redistribution after grouting-stress coupling is calculated in combination with the grouting pressure field, and the stress concentration factor is output.

[0092] The collapse probability calculation unit is used to construct a Bayesian network with residual cavity volume, filling progress, stress concentration factor and historical collapse cases as prior parameters. The collapse probability P of each grid cell in the three-dimensional geological model after grouting is calculated by the MCMC sampling method in the future 1 day, 7 days, 14 days and 28 days, and a four-dimensional collapse probability volume (x,y,z,t) is generated.

[0093] The risk heatmap rendering unit is used to divide the four-dimensional collapse probability volume into a set of horizontal slices with a spacing of 0.5m and a set of vertical cross-sections with a spacing of 10m. It uses WebGL shaders to achieve parallel GPU rendering, outputs the risk heatmap, and pushes it to the process visualization module.

[0094] It should be noted that the spatiotemporal unified indexing algorithm is used to concatenate timestamp data and 3D mesh data into a 64-bit integer key, establish a hash-balanced tree hybrid index, achieve millisecond-level synchronous registration between 3D geological models and monitoring data, and support the rapid extraction of model snapshots at any time, providing a time-aligned data baseline for 4D probabilistic volume calculation.

[0095] Kriging interpolation algorithm: It uses the variation function to characterize the spatial autocorrelation between grids in the 3D geological model, uses known grid attribute fields to weighted estimate of missing nodes, and interpolates and fills parameters such as grouting pressure and filling rate.

[0096] Time decay function: Based on the double exponential model V(t), it is used to predict the trend of the residual cavity volume slowly decreasing over time under the coupling effect of slurry bleeding shrinkage and rock movement, providing physically reasonable trend extrapolation information for risk rolling early warning within 28 days.

[0097] 3D Convolutional LSTM Network: This network adds 3D convolutional gating to the standard LSTM and extracts spatial neighborhood and temporal features from the 3D geological model. It uses the grouting volume, pressure and porosity sequence as input to predict the filling rate of the ungrouted area.

[0098] FLAC3D Fast Solver: Employing explicit Lagrange difference and hybrid parallel GPU acceleration, it performs elastic-plastic-fluid coupled calculations on a million-element model. While maintaining the rock mass strain softening and Mohr-Coulomb yield criterion, it effectively improves the solution speed and meets the timeliness requirements of on-site rolling risk assessment.

[0099] Minimum principal stress The key mechanical indicators for judging roof stripping, water inrush and collapse in goaf areas can be directly used to calculate the stress concentration factor.

[0100] Stress concentration factor: a dimensionless number, defined as the ratio of the local maximum deviatoric stress to the far-field deviatoric stress. When the factor is >1.5, it indicates that the rock mass has entered plastic damage. Combined with the cavity volume and filling rate, it can significantly increase the conditional probability of collapse nodes in the Bayesian network. It is the core driving parameter of the highlighted area in the risk heat map.

[0101] Bayesian network: Connects nodes of “void volume-filling rate-stress concentration factor-collapse” with a directed acyclic graph. The node conditional probability table is obtained from historical case statistics. It supports forward reasoning to predict the collapse probability and reverse diagnosis of disaster-causing factors, providing quantitative numerical basis for governance measures and can be updated online with new data.

[0102] MCMC sampling method: The Metropolis-Hastings algorithm is used to sample the high-dimensional probability space. The random walk avoids the problem of normalization constants; after convergence, the collapse probability distribution of each grid is obtained, with an error of <1% compared with the analytical solution, and it is easy to parallelize. After GPU acceleration, the calculation of a four-dimensional probability volume with millions of nodes is completed in 4 minutes.

[0103] WebGL shader: It uses fragment shaders to traverse 0.5m horizontal slices of a 3D geological model in parallel, maps the collapse probability to HSL gradient colors, and superimposes multiple layers of transparency through a depth peeling algorithm, enabling real-time rotation and scaling at 30fps on the browser.

[0104] VII. Process Visualization Module: Used to visualize the operation process of the above modules;

[0105] The process visualization module includes a 3D scene rendering unit and an interactive construction drawing export unit;

[0106] The 3D scene rendering unit is based on the WebGL+Three.js engine and is used to load 3D geological models, curtain mesh, construction trajectory, quantitative assessment results and risk heat map in real time in the same coordinate system, and to visualize them using the LOD+view frustum culling strategy.

[0107] The interactive construction drawing export unit is used to generate DWG / DXF format planar construction drawings, supports user-selectable scales, and automatically marks hole numbers, design depths, grouting pressures, and risk color block information on the planar construction drawings according to the "Code for Design of Construction Drawings for Coal Mine Shaft and Tunnel Engineering".

[0108] It should be noted that the WebGL+Three.js engine: WebGL is based on GPU hardware acceleration and can achieve high-performance 3D rendering in the browser without plugins. Three.js encapsulates scenes, cameras, geometry and materials on the basis of WebGL, and provides lighting, shadow and post-processing modules. Developers can use this engine to quickly build cross-platform applications and easily integrate geological models, borehole meshes and heat maps to achieve 30fps real-time interactive roaming, meeting the "open and view" needs of mining production sites.

[0109] LOD+view frustum culling strategy: Dynamically switch multi-level detail (LOD) based on the viewpoint distance in the 3D geological model, reducing distant mesh patches by 80% and restoring high resolution when zooming in. Combined with the view frustum culling method, only objects entering the view are rendered, significantly reducing GPU load and providing a smooth user experience for WebGL geological visualization.

[0110] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0111] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] It should also be noted that, in this document, 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. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0114] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A smart prediction system for coal mine goaf subsidence based on multi-source data fusion, characterized in that, include: The multi-source data acquisition module is used to acquire exploration data from coal mine goaf areas and perform preliminary processing and verification. The 3D modeling module is used to construct a 3D geological model of the goaf area based on the exploration data using a fusion inversion algorithm, and generate an editable meshed geological body. The grouting boundary planning module is used to generate a three-dimensional envelope of the abnormal response of the goaf boundary based on the three-dimensional dynamic model, and then delineate the curtain hole network and borehole depth. The drilling scheme planning module generates construction trajectories and drilling control parameters based on the curtain hole pattern and drilling depth. The operation monitoring module is used to collect measured data during the drilling and grouting process and perform quantitative evaluation to obtain quantitative evaluation results; The risk assessment module is used to couple the three-dimensional geological model and the quantitative assessment results to establish a four-dimensional collapse probability volume and output a risk heat map. The process visualization module is used to visually display the operation process of the above modules; The 3D modeling module includes a multiphysics joint inversion unit, a dynamic update and incremental fusion unit, a gridded geological body generation unit, a priori geological constraint library unit, and a lightweight output interface unit. The multiphysics joint inversion unit is used to spatially register the exploration data on the same three-dimensional grid, and to iteratively invert the data through a cross-gradient objective function constrained by W-SSIM to obtain a three-dimensional geological model containing the range, depth and roof and floor height of the goaf. When the exploration data is updated by the multi-source data acquisition module, the dynamic update and incremental fusion unit only performs local incremental inversion on the affected grid and uses the Kalman filter-finite element coupling framework to perform spatiotemporal fusion update with the existing three-dimensional geological model, while recording the version number and timestamp. The gridded geological body generation unit uses an octree-tetrahedral hybrid grid to partition the three-dimensional geological model, divide the goaf and its surrounding rock, and adaptively densify the grid according to the distance from the goaf boundary to generate gridded geological bodies. Each grid unit is assigned an editable attribute field, which includes lithology code, resistivity, wave velocity, porosity and void probability, resulting in an editable gridded geological body. The prior geological constraint library unit is used to store hard constraint data of the goaf area, including existing roadways, borehole columns and their distribution, and geological profiles at various depth gradients in the mining area. During the inversion process, the objective function is embedded through a global penalty term as a constraint condition for the generation of the three-dimensional geological model. The lightweight output interface unit is used to simplify the three-dimensional geological model to LOD and output a standard format file package for the grouting boundary planning module, drilling scheme planning module, operation monitoring module, risk assessment module and process visualization module to call. It also supports providing real-time model slicing services in the form of RESTful API.

2. The intelligent prediction system for coal mine goaf collapse based on multi-source data fusion according to claim 1, characterized in that: The multi-source data acquisition module includes a data acquisition unit, a data preprocessing unit, a data integrity verification unit, and a data compression and transmission unit; The data acquisition unit is used to conduct preliminary exploration of the coal mine goaf using direct current resistivity, ground-penetrating radar and shear wave two-dimensional seismic method, and to obtain the resistivity volume of direct current resistivity, the electromagnetic wave velocity volume of ground-penetrating radar and the impedance volume of shear wave two-dimensional seismic wave. The data preprocessing unit achieves millisecond-level synchronization of data acquisition from DC resistivity, ground-penetrating radar, and shear wave 2D seismic methods through timestamp alignment and trigger signal control, and performs noise suppression and outlier removal on the acquired resistivity volume, electromagnetic velocity volume, and seismic impedance volume. The data integrity verification unit is used to verify the integrity of the collected exploration data through redundant acquisition comparison and cross-validation algorithms, mark the exploration areas with missing or conflicting data, and trigger the supplementary acquisition process. The data compression and transmission unit is used to compress exploration data in real time using a wavelet transform compression algorithm, and transmit the compressed exploration data to the 3D modeling module via a 5G / fiber hybrid network.

3. The intelligent prediction system for coal mine goaf collapse based on multi-source data fusion according to claim 1, characterized in that: The grouting boundary planning module includes a boundary envelope extraction unit, a curtain hole network generation unit, a hole depth calculation unit, and a drilling rig accessibility verification unit. The boundary envelope extraction unit is based on a three-dimensional geological model. It uses three-dimensional Sobel-Canny hybrid edge detection to extract the probability gradient maximum surface to generate an initial envelope. Then, it fills the holes and removes redundant patches through morphological closing operation and α-shape algorithm to obtain the three-dimensional envelope of the abnormal response of the goaf boundary. The curtain perforated mesh generation unit generates a double-layer perforated mesh skeleton line at a thickness of 0.3 to 1.0 times the envelope surface and inside. With the grouting diffusion radius R as a constraint, the improved Voronoi-Lloyd iteration is used to optimize the hole position arrangement so that the hole spacing d∈[0.8R,1.2R]. The output is a GIS layer containing hole position number, coordinates, and borehole azimuth and inclination angle. The borehole depth calculation unit cuts the three-dimensional geological model based on the borehole network GIS layer, extracts the burial depth curves of the top and bottom plates, and automatically calculates the borehole depth with "penetrating the bottom plate by 0.5m and being 1.0m away from the top plate" as a hard constraint and "minimizing the amount of drilling work" as the objective. When the difference in elevation between the top and bottom plates is ≥0.2m, a recalculation is triggered. The drilling rig accessibility verification unit imports the equipment parameter library, combines the three-dimensional geological model to perform collision detection and accessibility simulation on the borehole network, eliminates unworkable boreholes, and automatically generates a list of alternative boreholes and a drilling rig station layout diagram. The parameter library includes the minimum turning radius and the maximum pitch angle.

4. The intelligent prediction system for coal mine goaf subsidence based on multi-source data fusion according to claim 1, characterized in that: The drilling scheme planning module includes a construction path generation unit, a borehole trajectory optimization unit, and a control parameter generation unit. The construction path generation unit is used to generate construction trajectories based on the curtain hole network GIS layer, with "shortest total drilling footage" and "minimum number of machine relocations" as dual objective functions, and through an improved genetic-ant colony hybrid algorithm. The trajectories include drilling rig station sequence, single-station hole sequence, and estimated pure drilling time. The borehole trajectory generation unit is used to combine the curtain hole network GIS layer and the borehole depth to perform A* path search in the three-dimensional geological model, and generate the borehole trajectory by fitting the minimum bending radius constraint of the drill rod through a quintic spline curve. The control parameter generation unit is used to discretize the construction trajectory and drilling trajectory into a sequence of control nodes using the spline trajectory discretization method, generate operating control parameters, and push them to the drilling rig PLC system in JSON format.

5. The intelligent prediction system for coal mine goaf collapse based on multi-source data fusion according to claim 1, characterized in that: The operation monitoring module includes a drilling measurement and real-time correction unit, a grouting monitoring unit, and a real-time engineering quantity statistics unit. The measurement while drilling and real-time deviation correction unit is used during the drilling process of the drilling rig to obtain the drill bit position, well inclination angle and azimuth angle at a frequency of 1Hz through wired drill rod + MWD, and match them with the trajectory control parameters through a sliding window. When the horizontal projection deviation is 0.1m / 100m or the well inclination angle deviation is 1°, the PID algorithm is triggered to output the deviation correction command to the drilling rig PLC. The grouting monitoring unit is used to acquire grouting pressure, flow rate and grout density in real time during the operation of the grouting machine through a pressure transmitter, a turbine flow meter and an ultrasonic density meter. The real-time engineering quantity statistics unit calculates drilled meters, cumulative pure drilling time, mud consumption and drill bit wear index based on drilling data, calculates the filling stone body and pressure-flow-time curve based on grouting data, and packages the results and pushes them to the risk assessment module.

6. The intelligent prediction system for coal mine goaf collapse based on multi-source data fusion according to claim 1, characterized in that: The risk assessment module includes a multi-source data coupling unit, a residual cavity volume calculation unit, a grouting and filling progress calculation unit, a formation stress field calculation unit, a collapse probability calculation unit, and a risk heat map rendering unit. The multi-source data coupling unit is used to register the three-dimensional geological model with the quantitative evaluation results in milliseconds using a spatiotemporal unified indexing algorithm, and to fill the missing grid using Kriging interpolation to generate a post-grouting three-dimensional model containing residual cavities, filling progress, and grouting pressure field. The residual cavity volume calculation unit is used to calculate the volume of residual cavities in the partitioned and layered sections in the three-dimensional model after grouting using a 0.1m×0.1m×0.1m infinitesimal integral, and to predict the evolution volume using a time decay function. The grouting and filling progress calculation unit is used to predict the filling rate improvement curve by using a three-dimensional convolutional LSTM network to extrapolate the ungrouted area in the three-dimensional model after grouting, based on the filling progress volume. The formation stress field calculation unit is used to call the FLAC3D solver to perform elastoplastic numerical simulation of lithology, porosity and burial depth, obtain the minimum principal stress σ3 distribution body, and calculate stress redistribution and stress concentration factor in combination with grouting pressure field. The collapse probability calculation unit is used to construct a Bayesian network with the residual cavity volume, filling progress, stress concentration factor and historical cases as prior parameters, and calculate the future collapse probability P through MCMC sampling to generate a four-dimensional collapse probability volume (x,y,z,t). The risk heatmap rendering unit is used to divide the four-dimensional collapse probability volume into horizontal slices with a spacing of 0.5m and vertical sections with a spacing of 10m, and uses WebGL shaders for GPU parallel rendering to output the risk heatmap and push it to the process visualization module.

7. The intelligent prediction system for coal mine goaf subsidence based on multi-source data fusion according to claim 1, characterized in that: The process visualization module includes a 3D scene rendering unit and an interactive construction drawing export unit; The three-dimensional scene rendering unit is based on the WebGL+Three.js engine and is used to load three-dimensional geological models, curtain perforated meshes, construction trajectories, quantitative assessment results and risk heat maps in real time under the same coordinate system, and to visualize them using the LOD+view frustum culling strategy. The interactive construction drawing export unit is used to generate planar construction drawings.

Images