Deep mineral resource high-precision intelligent detection and evaluation system and method

Abstract

The application discloses a kind of deep mineral resources high-precision intelligent detection evaluation system and method.The method includes: obtaining prior geological model and multi-source original detection data, regulation and control acquisition process, evaluate data quality and mark data conflict, generate multi-dimensional standardized dataset;Construct shallow constraint model, filter shallow interference, limit model modification range, generate three-dimensional physical property model set;Establish the spatial correlation domain of surface and underground, carry out multi-stage physical tracing diagnosis, mark physical property mutation interface, form multi-source anomaly fusion result set;Evaluate the resolution confidence of each detection depth, carry out boundary sharpening on the marked physical property mutation interface region, generate three-dimensional mineralization probability cloud model;Infer geological uncertainty distribution, generate candidate virtual drill hole, carry out comprehensive optimization, generate drill trajectory control instruction, according to drilling verification result local update, generate termination or steering instruction, form closed-loop iteration link.

Description

Technical Field

[0001] This invention relates to the field of intelligent exploration technology for deep mineral resources, and more specifically, to a high-precision intelligent detection and evaluation system and method for deep mineral resources. Background Technology

[0002] As deep mineral resource exploration develops towards multi-source three-dimensional exploration involving "space-ground-well", the fusion of massive heterogeneous data and the high cost of deep drilling verification are often disconnected, resulting in low positioning accuracy of deep ore bodies and difficulty in controlling exploration costs.

[0003] Existing 3D modeling and drilling planning mostly adopt static strategies, which have many shortcomings. For example, shallow interference during modeling can easily contaminate deep results, and conflicts between multiple data sources rely solely on mathematical smoothing without physical traceability, leading to distortion of deep models and blurred ore body boundaries. Drilling layout is highly dependent on human experience, resulting in low exploration efficiency and a high risk of wasting funds on ineffective drilling. Summary of the Invention

[0004] The main objective of this invention is to provide a high-precision intelligent detection and assessment system and method for deep mineral resources, so as to overcome the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution.

[0006] According to one aspect of the present invention, a high-precision intelligent detection and assessment system for deep mineral resources is provided, comprising: Multidimensional data acquisition and standardization module: acquires prior geological models and collects multi-source raw exploration data, reverse-derives acquisition control commands to regulate energy during the acquisition process and adds depth stratification markers; after acquisition, evaluates data quality and marks data conflicts, and generates a multidimensional standardized dataset; Shallow and deep constraint and deep inversion module: Based on a multidimensional standardized dataset, a shallow constraint model is constructed, shallow interference is filtered out through depth layer labeling, and inversion permission commands are issued to limit the scope of model modification, generating a set of three-dimensional physical property models; Multi-source spatial correlation and source tracing diagnosis module: Based on a three-dimensional physical property model set, a spatial correlation domain between the surface and underground is established; then, data conflict identifiers are read to perform multi-stage physical source tracing diagnosis, and physical property abrupt change interfaces are confirmed and marked to form a multi-source anomaly fusion result set; Boundary sharpening module: Based on the multi-source anomaly fusion result set, it evaluates the resolution confidence of each detection depth, and then combines the confidence with the source tracing diagnosis results to perform multi-source data fusion, and sharpens the boundaries of the marked physical property change interface areas to generate a three-dimensional mineralization probability cloud model. Uncertainty extrapolation and drilling closed-loop iteration module: Based on the three-dimensional mineralization probability cloud model, the geological uncertainty distribution is extrapolated to generate candidate virtual boreholes; comprehensive optimization is carried out by combining surface construction feasibility and engineering cost to generate borehole trajectory control instructions; the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated according to the drilling verification results, and exploration termination or reversal instructions are generated to form a closed-loop iteration link.

[0007] Furthermore, the methods for generating the multidimensional standardized dataset include: A priori geological model is acquired and multi-source raw detection data are collected simultaneously to construct a five-dimensional unified data container with grid nodes. During the acquisition process, the transmission current and reception time of the physical acquisition equipment are deduced in reverse based on the physical penetration rule to generate acquisition control commands and adaptively regulate the acquisition energy. After the data acquisition is completed, the actual penetration depth is deduced by reverse calculation based on the actual transmission current and reception time, and corresponding depth layer markers are added to each detection data. The quality of each detection data is evaluated by a multi-factor composite rule of physical quality, and data conflict detection and labeling are performed based on the comparison results of the difference between multi-source detection data in the same spatial grid and the instrument nominal error, so as to generate a multi-dimensional standardized dataset.

[0008] Furthermore, the method for constructing the shallow constraint model includes: Based on a multidimensional standardized dataset, physical properties with shallow depth classification labels are extracted, spatial variation feature analysis is performed, and then the range is extracted based on the spatial variation features. Spatial interpolation is performed on each grid node to construct a shallow three-dimensional physical hard shell. The data coverage of grid nodes is evaluated based on the distribution of probe data within each spatial grid. A three-level adaptive degradation process is then performed to match the corresponding constraint type and allowable adjustment range for each grid node, generating a shallow constraint model with control labels.

[0009] Furthermore, the method for generating the three-dimensional property model set includes: The physical properties of deep grid nodes in the multidimensional standardized dataset are extracted to obtain the comprehensive physical property gradient modulus. Then, the physical property abrupt change interface is identified by the boundary judgment rule, and a suspected physical property abrupt change interface marker is generated for the grid nodes on the interface. Read the depth layering labels of all grid nodes in the five-dimensional unified data container, assign mask values ​​to different layer regions, construct a three-dimensional mask matrix, and filter all grid nodes marked as shallow regions; The multidimensional standardized dataset is divided into a surface exploration dataset and a downhole exploration dataset. Then, an inversion permission control command is issued to couple the shallow constraint model with the 3D mask matrix, lock the shallow grid nodes and open the modification permissions of the middle and deep grid nodes, and drive the inversion engine to perform independent inversion of the surface and downhole data respectively, generating a multi-source deep independent 3D physical property model set containing the surface 3D physical property model and the downhole 3D physical property model.

[0010] Furthermore, the method for forming the multi-source anomaly fusion result set includes: Based on the three-dimensional physical property model set, the grid nodes with overlapping spatial coordinates in the three-dimensional physical property models of the surface and underground are extracted to construct a spatial association domain, and the spatial consistency index is obtained to identify low consistency areas. For regions with low consistency, data conflict markers are read, triggering a three-stage progressive physical source diagnosis: the first stage investigates static effects to eliminate topographic distortion, the second stage investigates volumetric effects to generate differentiated fusion weight instructions, and the third stage investigates geological phase changes to generate physical property mutation interface confirmation markers, forming a multi-source anomaly fusion result set.

[0011] Furthermore, the methods for performing multi-source data fusion include: Based on the multi-source anomaly fusion result set, three-dimensional depth slices are divided, the built-in physical resolution attenuation rule library is queried, the resolution confidence factor within each detection depth range is quantified, and the physical attributes of each network node are nonlinearly mapped to the basic mineralization probability. Then, the resolution confidence factor and the basic mineralization probability are coupled together to obtain the initial mineralization probability, which is assigned to the corresponding grid node and then integrated into the initial three-dimensional mineralization probability cloud model.

[0012] Furthermore, the method for generating the three-dimensional mineralization probability cloud model includes: Based on the multi-source anomaly fusion result set, the grid nodes carrying the physical property abrupt change interface confirmation mark in the initial three-dimensional mineralization probability cloud model are located and denoted as phase transition nodes. The mineralization probability is enhanced and phase transition boundary probability enhancement labels are added to them. Furthermore, for phase transition nodes and their neighboring grid nodes, the spatial angle between the probability gradient direction and the physical property gradient direction is extracted, and the consistency preservation logic or forced projection sharpening logic is triggered according to the size of the spatial angle. If the forced projection sharpening logic is triggered, the probability gradient direction will be forcibly projected to the physical property gradient direction to redistribute the mineralization probability value. Based on the final mineralization probability after probability enhancement and gradient sharpening, a three-dimensional mineralization probability cloud model is updated and formed.

[0013] Furthermore, the method for generating the drilling trajectory control command includes: Based on the three-dimensional mineralization probability cloud model, the information entropy of each grid node is extracted to infer the distribution of geological uncertainty field, and candidate virtual boreholes are generated in the area where the information entropy exceeds the limit. Obtain surface construction condition data, conduct surface construction feasibility assessment on candidate virtual boreholes, and screen out candidate boreholes that meet the actual construction conditions; Furthermore, for candidate boreholes, a two-factor game mechanism of information benefits and engineering costs is used to dynamically and comprehensively optimize based on the current exploration stage, and generate borehole trajectory control instructions.

[0014] Furthermore, the method of locally updating the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model based on drilling verification results, and generating exploration termination or reversal commands, includes: The real physical property data of the rock core extracted by the drilling rig executing the drilling trajectory control command are obtained as the drilling verification result, and the physical property values ​​of the corresponding grid nodes in the multi-source anomaly fusion result set are extracted as the model prediction values; then the relative prediction deviation rate between the real physical property data and the model prediction values ​​is obtained. When the relative prediction deviation rate does not exceed the limit, the normal mode is triggered to perform probability assimilation on the local neighborhood grid of the borehole; when the relative prediction deviation rate exceeds the limit, the emergency mode is triggered to expand the grid reconstruction range and generate exploration turning instructions; after executing the normal mode or the emergency mode, the corresponding regions of the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated. After each update, an iterative termination determination mechanism is used to determine the termination based on the global grid average information entropy and cumulative engineering cost. If the determination conditions are met, an exploration termination command is generated and executed.

[0015] According to another aspect of the present invention, a high-precision intelligent detection and assessment method for deep mineral resources is provided, comprising: S1: Obtain the prior geological model and collect multi-source raw exploration data; reverse-engineer the acquisition control command to regulate the energy of the acquisition process and add depth layer markers; after the acquisition is completed, evaluate the data quality and mark data conflicts, and generate a multidimensional standardized dataset. S2: Based on a multidimensional standardized dataset, a shallow constraint model is constructed. Shallow interference is filtered out by depth layer labeling, and an inversion permission command is issued to limit the scope of model modification, generating a three-dimensional physical property model set. S3: Based on the three-dimensional physical property model set, establish the spatial correlation domain between the surface and the well; then read the data conflict identifier to perform multi-stage physical source tracing diagnosis, and confirm the physical property change interface to form a multi-source anomaly fusion result set; S4: Based on the multi-source anomaly fusion result set, evaluate the resolution confidence of each detection depth, and then combine the confidence with the source tracing diagnosis results to perform multi-source data fusion, and sharpen the boundaries of the marked physical property change interface areas to generate a three-dimensional mineralization probability cloud model with enhanced labels. S5: Based on the three-dimensional mineralization probability cloud model, the distribution of geological uncertainties is deduced, and candidate virtual boreholes are generated; combined with the feasibility of surface construction and engineering costs, comprehensive optimization is performed to generate borehole trajectory control instructions; based on the drilling verification results, the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated, and exploration termination or reversal instructions are generated to form a closed-loop iterative link.

[0016] Compared with existing technologies, this invention focuses on the scenario of deep mineral exploration and realizes an automated closed loop of "three-dimensional probabilistic modeling - intelligent drilling optimization - local model correction", which significantly improves the accuracy of deep mineral exploration and the efficiency of capital utilization.

[0017] Specifically, firstly, this invention constructs a shallow three-dimensional physical hard shell and, in conjunction with a three-dimensional mask matrix, issues inversion permissions to forcibly lock the shallow mesh, thereby physically cutting off the propagation path of shallow interference to the deep part and ensuring the high fidelity of the deep physical property model.

[0018] Secondly, this invention addresses the three-stage progressive physical source tracing diagnosis triggered by multi-source data conflicts, and introduces resolution confidence coupling and gradient forced projection sharpening logic to overcome the loss of physical meaning caused by mathematical smoothing and eliminate ore body boundary smoothing artifacts.

[0019] Furthermore, this invention uses information entropy to deduce the uncertainty field, combines the feasibility of surface construction with engineering costs to perform a two-factor game optimization, and performs spatial exclusion verification to ensure that the borehole trajectory is both the optimal solution for information gain and the physically feasible optimal solution for engineering.

[0020] Finally, based on the deviation rate between the actual drilled physical properties and the predicted values, this invention adaptively triggers conventional probability assimilation or emergency grid reconstruction modes, and sets an iteration termination threshold based on exploration specifications, thereby achieving efficient scheduling of computing resources and intelligent exploration closed loop.

[0021] This invention achieves high-precision reconstruction of deep models and dynamic optimal control of drilling costs across the entire data fusion and drilling verification chain. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a high-precision intelligent detection and assessment system for deep mineral resources according to one embodiment of the present invention; Figure 2 This is a schematic diagram of a high-precision intelligent detection and evaluation method for deep mineral resources in one embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 Please see Figure 1 As shown, this embodiment provides a high-precision intelligent detection and assessment system for deep mineral resources, including: Multidimensional data acquisition and standardization module: acquires prior geological models and collects multi-source raw exploration data, reverse-derives acquisition control commands to regulate energy during the acquisition process and adds depth stratification markers; after acquisition is completed, evaluates data quality and marks data conflicts, and generates a multidimensional standardized dataset.

[0025] Shallow and deep constraint and deep inversion module: Based on a multidimensional standardized dataset, a shallow constraint model is constructed, shallow interference is filtered through depth layer labeling, and inversion permission commands are issued to limit the scope of model modification, generating a set of three-dimensional physical property models.

[0026] Multi-source spatial correlation and source tracing diagnosis module: Based on the three-dimensional physical property model set, a spatial correlation domain between the surface and underground is established; then, data conflict identifiers are read to perform multi-stage physical source tracing diagnosis, and physical property mutation interfaces are confirmed and marked to form a multi-source anomaly fusion result set.

[0027] Boundary sharpening module: Based on the multi-source anomaly fusion result set, it evaluates the resolution confidence of each detection depth, and then combines the confidence with the source tracing diagnosis results to perform multi-source data fusion, and sharpens the boundaries of the marked physical property change interface regions to generate a three-dimensional mineralization probability cloud model with enhanced labels.

[0028] Uncertainty extrapolation and drilling closed-loop iteration module: Based on the three-dimensional mineralization probability cloud model, the geological uncertainty distribution is extrapolated to generate candidate virtual boreholes; comprehensive optimization is carried out by combining surface construction feasibility and engineering cost to generate borehole trajectory control instructions; the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated according to the drilling verification results, and exploration termination or reversal instructions are generated to form a closed-loop iteration link.

[0029] Methods for generating multidimensional normalized datasets include: A three-dimensional geological grid model, constructed from regional geological maps, historical boreholes, and geological profiles, is acquired and designated as the prior geological model. Simultaneously, multi-source raw exploration data from satellite remote sensing, airborne geophysical exploration, ground geophysical exploration, and downhole exploration are collected.

[0030] To prevent storage redundancy caused by massive, multi-source, heterogeneous data, a five-dimensional unified data container is constructed based on an octree dynamic sparse grid. The five dimensions are defined as: three-dimensional spatial coordinates (X, Y, Z, all in meters), time dimension (in seconds), and physical attribute dimension (the dimension depends on the detection type, such as resistivity in ohm-meters and density in grams per cubic centimeter).

[0031] During the grid generation process, only spatial regions with measured data or significant changes in prior geological properties are subdivided using octrees, while blank areas without data are kept in a coarse grid state, thereby reducing memory usage while ensuring spatial resolution.

[0032] The collected multi-source raw detection data is standardized and mapped: through unified coordinate system transformation and spatial interpolation, it is projected and distributed to the discrete grid nodes (i.e. leaf grid nodes of the octree) of the five-dimensional unified data container, thereby realizing the alignment and structured storage of multi-source data under a unified spatial, temporal and physical attribute benchmark.

[0033] Meanwhile, during the data acquisition process, acquisition control commands are generated by reverse derivation based on physical penetration rules to drive the physical acquisition equipment to perform energy adaptive regulation.

[0034] The specific method for reverse derivation based on the physical penetration rule is as follows: Taking the transient electromagnetic method (TEM) as an example, its physical penetration rule is expressed as: the detection depth is equal to the calibration coefficient multiplied by the fifth root of the product of the transmission current and the reception time (where the calibration coefficient is calibrated by actual measurement through a standard borehole of known depth); based on this rule, reverse derivation is performed: after setting the target detection depth, the required transmission current is equal to the fifth power of the quotient of the target depth divided by the calibration coefficient, and then divided by the current reception time.

[0035] The required transmission current is monitored and calculated in real time. If the current value is less than or equal to the rated maximum current of the equipment, the current value is directly issued as a control command.

[0036] If the current value exceeds the rated maximum current, the current limiting fallback strategy is triggered: the transmit current is forcibly locked to the rated maximum current, and the receive time is extended in reverse to compensate for energy attenuation.

[0037] The extended reception time is equal to the fifth power of the quotient of the target depth divided by the calibration coefficient, and then divided by the rated maximum current.

[0038] After the data acquisition is completed, the actual penetration depth is calculated by reverse calculation based on the actual transmission current and reception time. Based on this, a depth layer marker is added to each detection data (for example, the current industry generally divides 0 to 500 meters into shallow depth, 500 to 1000 meters into middle depth, and greater than 1000 meters into deep depth).

[0039] After data collection, each piece of detection data is evaluated for quality using a multi-factor composite rule for physical quality. The specific evaluation logic is as follows: the measured signal-to-noise ratio is divided by the theoretical optimal signal-to-noise ratio of the corresponding detection data to obtain the normalized signal-to-noise ratio value; the measured number of coverages is divided by the reference number of coverages (for example, 4 times) to obtain the normalized number of coverages value; finally, the weighted sum of the normalized signal-to-noise ratio value and the normalized number of coverages value is taken as the physical quality composite identifier.

[0040] The weighting of the weighted sum is dynamically mapped to the detection depth. In shallow detection, due to severe electromagnetic interference, the signal-to-noise ratio weight is set to 0.7 and the coverage count weight is 0.3. In deep detection, due to severe signal attenuation, increasing the coverage count is more critical to improving the signal superposition effect. Therefore, the weighting is dynamically adjusted to a signal-to-noise ratio weight of 0.4 and a coverage count weight of 0.6.

[0041] Detect conflicts in physical attribute data that have overlapping spatial coordinates (i.e., fall into the same octagonal leaf network) but originate from different detection methods or different survey lines.

[0042] Calculate the absolute difference of physical attribute data from multiple sources within the same grid. If the absolute difference is greater than the instrument’s nominal measurement error, it is determined to be a data conflict. Generate a data conflict marker that represents the original appearance of the conflict and completely preserve the data sources, acquisition time, and original values ​​of both conflicting parties in the data container for subsequent physical source tracing and diagnosis.

[0043] If the absolute difference is less than or equal to the nominal measurement error, it is determined to be a normal fluctuation, and its arithmetic mean is taken as the final physical property value of the grid.

[0044] Finally, the five-dimensional standardized probe data, which included physical quality verification markers, depth layer markers, and data conflict markers, were integrated to generate a multi-dimensional standardized dataset.

[0045] Methods for constructing shallow constraint models include: Based on a multidimensional normalized dataset, for regions labeled as shallow in depth classification, spatial variation features of physical properties are extracted and analyzed to construct a shallow three-dimensional physical hard shell. Specifically: First, a spatial variogram is constructed to quantify the spatial autocorrelation of physical properties. The quantification logic is as follows: for any given spatial lag distance, the corresponding variogram value is equal to the sum of squares of the physical property interpolations of all sample point pairs ÷ (number of sample point pairs × 2).

[0046] It should be noted that each grid node is a sample point, and each sample point carries its five-dimensional standardized probe data. Spatial lag distance refers to the spatial distance between two sample points. The calculation logic of the variogram is: find all pairs of points with the same distance or within a narrow band, and calculate the variogram value using their attribute differences.

[0047] As the spatial lag distance gradually increases, the variogram value will tend to a stable base value (i.e., a stable value of the global variance of physical properties); when the variogram value first reaches 95% of the base value, the corresponding spatial lag distance is extracted as the "range"; the range represents the maximum effective radius for maintaining the spatial correlation of shallow strata physical properties.

[0048] Using this range as the spatial smoothing radius, kriging spatial interpolation is performed on the physical properties of the shallow grid nodes to generate a spatially continuously variable shallow strata layering interface, thereby constructing an initial shallow three-dimensional physical hard shell model.

[0049] To prevent insufficient shallow constraints from causing the subsequent deep inversion engine to diverge, it is necessary to evaluate the data coverage of each grid node in the shallow physical shell based on the distribution of the probe data in each spatial grid, and to perform a three-level adaptive degradation process.

[0050] The data coverage rate of a grid node is equal to the number of valid probe data points actually acquired within the spatial grid where the grid node is located, divided by the theoretically expected number of full-coverage data points for that grid.

[0051] The three-level adaptive degradation process includes a first-level hard constraint, a second-level soft constraint, and a third-level elimination and filling. The specific processing logic is as follows: Level 1 hard constraint: When the data coverage of a grid node is greater than or equal to 80%, the data of that grid node is considered sufficient. In subsequent inversion iterations, the physical attribute value of that grid node is forcibly locked to the measured interpolation result and cannot be modified, serving as the absolute benchmark for deep inversion.

[0052] Secondary soft constraint: When the data coverage of a grid node is greater than or equal to 20% and less than 80%, it is determined that the data of that node is locally missing. The physical properties of the grid node are allowed to be adjusted within the constraint range of 85% to 115% of the background physical property value (i.e., the background value is allowed to fluctuate by 15%). This design can both retain the macro-constraint of the shallow model and give it the freedom to correct local errors in the inversion.

[0053] Level 3 Removal and Filling: When the data coverage of a grid node is less than 20%, the grid node is determined to be an invalid data blind spot, triggering the missing data removal mechanism: the original outlier value of the grid node is forcibly discarded and filled using the median of the global physical attribute. At the same time, the constraint range of the grid node is widened to ±50% of the background value to eliminate the numerical singularity interference caused by the extremely low coverage.

[0054] After three levels of degradation processing, control labels containing constraint type, allowable adjustment range and initial physical property values ​​are generated for each mesh node in the shallow 3D physical hard shell, generating a shallow constraint model with control labels.

[0055] Methods for generating 3D property model sets include: Extract the physical properties of deep mesh nodes from the multidimensional standardized dataset and obtain the comprehensive physical property gradient modulus. Specifically: for any target mesh node in the depth region, extract the physical property values ​​of its adjacent mesh nodes in the three orthogonal directions (X, Y, Z) in three-dimensional space; for any direction, the rate of change of physical property in that direction = (physical property of the target node - physical property of adjacent nodes) ÷ spatial mesh step size in that direction; the comprehensive physical property gradient modulus is equal to the square root of the sum of the squares of the rates of change of physical property in each direction.

[0056] Then, the boundary determination rules are used to identify the physical property abrupt change interface (referring to the interface of macroscopic geological structure or lithological abrupt change surface, etc., where the physical properties of underground rocks change drastically): the comprehensive physical property gradient modulus is compared with the preset gradient threshold (based on the physical property difference of typical lithological interfaces in regional historical geological exploration data, such as 10Ω·m / m). When the comprehensive physical property gradient modulus is greater than the gradient threshold, the grid node is identified as being located on the physical property abrupt change interface, a suspected physical property abrupt change interface marker is generated for it, and the three-dimensional gradient direction vector of the grid node is recorded simultaneously.

[0057] To prevent shallow, complex near-surface interference signals from contaminating the deep inversion results, the depth layering labels of all grid nodes in the five-dimensional unified data container are read. For grid nodes marked as shallow regions, the mask value is set to 0 (representing shielding); for grid nodes marked as middle and deep regions, the mask value is set to 1 (representing openness). This generates a three-dimensional mask matrix that is completely consistent with the spatial topology of the five-dimensional unified data container.

[0058] The multidimensional standardized dataset is divided into surface exploration dataset and downhole exploration dataset according to the acquisition source. Inversion permission control instructions are issued to the external inversion engine. The shallow constraint model is coupled with the three-dimensional mask matrix. During the iteration process of the inversion engine, the physical attribute values ​​of the shallow grid nodes are forcibly locked to the initial values ​​or hard constraint values ​​in the shallow constraint model, and modification is prohibited. Modification permissions are only granted to the middle and deep grid nodes.

[0059] The optimization objective of the inversion engine is reconstructed as follows: minimizing the sum of squared residuals between the actual observed data and the forward simulation data, while adding a model roughness regularization term, and all update operations are subject to the permission control of the three-dimensional mask matrix.

[0060] The inversion engine receives surface exploration datasets and downhole exploration datasets respectively, and independently performs forward simulation and optimization iteration under the same access control and mask constraints to generate surface 3D physical property models and downhole 3D physical property models respectively.

[0061] The iteration status of the inversion engine is monitored in real time. When the root mean square error of the data is less than 5% and the rate of change of the objective function is less than 1% for three consecutive iterations, the inversion is considered to have converged.

[0062] If the convergence condition is not met even after the number of iterations reaches the preset maximum threshold (e.g., 100 times), an abnormal fallback mechanism is triggered: the iteration is forcibly terminated and the optimal solution of the current iteration step is output; finally, the surface and downhole models obtained by different detection methods (e.g., transient electromagnetic method, gravity exploration, etc.) are summarized to obtain a set of independent three-dimensional physical property models of multi-source deep parts.

[0063] It should be noted that the inversion engine is a commonly used built-in solver in this field specifically designed to execute geophysical inversion algorithms. Its working principle is as follows: it receives actual exploration data (such as electromagnetic field attenuation curves and gravity anomalies) and an initial geological model, performs forward modeling based on the current 3D model to simulate a set of "theoretical exploration data"; then it compares the differences (residuals) between the "theoretical data" and the "actual exploration data"; using optimization algorithms (such as the Gauss-Newton method, conjugate gradient method, etc.), it adjusts the physical property values ​​of each grid node in the model to continuously reduce the residuals. When the residuals are less than a set threshold (i.e., convergence), the iteration stops, and the final subsurface model is output.

[0064] Physical properties refer to the physical attributes of rocks or ores (i.e., physical properties such as resistivity and density). The purpose of a three-dimensional physical model is to discretize these physical attributes in three-dimensional space, dividing them into countless three-dimensional grids (voxels), and assigning a specific physical property value to each grid node. The model set is because this embodiment uses multi-source detection data from space, ground, and wells. Different detection methods invert different models of physical attributes, so the final result is a set.

[0065] A typical set of three-dimensional property models usually includes: Three-dimensional resistivity grid model (obtained by transient electromagnetic method or magnetotelluric method, which is extremely sensitive to aquifers and metal sulfide ore bodies).

[0066] Three-dimensional density grid model (obtained from gravity exploration inversion, used to identify rock mass density differences and fault structures).

[0067] Three-dimensional magnetic susceptibility grid model (obtained from magnetic exploration inversion, used to delineate magnetic ore bodies such as magnetite and basic-ultrabasic rock bodies).

[0068] Three-dimensional wave velocity grid model (obtained from seismic exploration inversion, used to characterize stratigraphic lithological interfaces and structural fracture zones).

[0069] The methods for forming a multi-source anomaly fusion result set include: Based on the three-dimensional physical property model set, grid nodes that completely overlap in spatial coordinates between the surface three-dimensional model and the downhole three-dimensional physical property model are extracted in three-dimensional space to construct a spatial association domain.

[0070] For any overlapping grid node in the spatial association domain, extract its surface physical attribute value and downhole physical attribute value; Spatial consistency index = (2 × surface physical attribute value × downhole physical attribute value) ÷ (square of surface physical attribute value + square of downhole physical attribute value).

[0071] It should be noted that the formula for calculating the spatial consistency index is a general metric derived from the direct application of the Dice coefficient in the continuous positive number space. In other words, it is a variant of the cosine similarity in the positive number space. Its purpose is to quantify the relative consistency between two positive numbers by using the ratio of the harmonic mean to the arithmetic mean.

[0072] When the spatial consistency index is less than the consistency threshold (generally set between 0.7 and 0.9; if the instrument accuracy is high or the geological conditions are simple, the threshold is set high (e.g., 0.85); if the geological conditions are complex or the electromagnetic interference is large, the threshold is set low (e.g., 0.75)), the node is judged to have multi-source data anomalies and is marked as a low consistency area.

[0073] Read the data conflict markers in the low consistency region, and trigger a three-stage progressive physical tracing diagnosis for abnormal nodes with the markers.

[0074] First-order diagnosis is used to investigate static effects (terrain distortion): extract the surface elevation data of the area where the abnormal grid node is located, and obtain the terrain relief (equal to the maximum surface elevation value minus the minimum surface elevation value within the local window where the node is located); if the terrain relief is greater than the relief threshold (this threshold can be set by forward modeling, constructing a pure terrain model containing different terrain reliefs for forward modeling, and when the simulated outlier value exceeds the instrument noise level, the corresponding terrain elevation difference is the threshold), then it is determined that the multi-source data anomaly is caused by the static effect of severe near-surface terrain undulation; call the built-in terrain correction algorithm commonly used in this field to eliminate the distortion of the physical attribute value of the grid node, and terminate the subsequent diagnosis stage.

[0075] Second-order diagnosis is used to investigate volume effect (detection range difference): If first-order diagnosis is not triggered, the ratio of the surface detection volume and the downhole detection volume corresponding to the grid node is calculated to obtain the detection volume ratio; if the detection volume ratio > the volume ratio threshold (which can be obtained by using the sensitivity matrix analysis method, by extracting the sensitivity matrix (JacobianMatrix) inside the inversion engine, spatially integrating the sensitivity of the surface and downhole devices on the target grid, obtaining their respective equivalent detection volumes, and recording their detection volume ratio as the volume ratio threshold), then the anomaly is determined to be caused by the physical response volume asymmetry (volume effect) of the two detection methods, and a differentiated fusion weight instruction is generated accordingly: downhole fusion weight = detection volume ratio ÷ (1 + detection volume ratio), surface fusion weight = 1 ÷ (1 + detection volume ratio), and the subsequent diagnosis stage is terminated.

[0076] The third-order diagnosis is used to investigate geological phase transitions (real abrupt changes in physical properties): If the first two-order diagnoses are not triggered, the anomaly is determined to originate from a real underground geological phase transition; the local physical property gradient modulus is obtained (the calculation method is the same as described above, equal to the square root of the sum of the squares of the spatial variation rates of physical properties in each orthogonal direction within the local area); if the local physical property gradient modulus is greater than the gradient threshold (this threshold is equal to the difference in physical properties on both sides of a typical lithological interface ÷ the three-dimensional mesh compensation, for example, if the resistivity difference on both sides of a fault is 500 Ω·m and the mesh step size is 50m, then the gradient threshold is set to 10 Ω·m / m), then it is confirmed as a real geological phase transition, and a physical property abrupt change interface confirmation mark is generated for the mesh node.

[0077] After completing the diagnosis and undergoing physical arbitration, a multi-source anomaly fusion result set is finally formed, which includes: a seamless fusion three-dimensional physical property model set after physical arbitration, a physical property mutation interface confirmation mark set, a differential fusion weight matrix, a diagnosis traceability log, and a quality control report.

[0078] The fusion 3D physical property model set includes a terrain correction model (first-order diagnostic trigger area, physical property model after eliminating terrain static effect distortion), a weighted fusion model (second-order diagnostic trigger area, physical property model recalculated according to differentiated fusion weight instructions (high downhole weight, low surface weight), effectively overcoming the volume average effect), and a high-fidelity preservation model (third-order diagnostic trigger area (real geological phase transition), which preserves the high-resolution physical property characteristics of downhole exploration and is not smoothed by low-resolution surface data).

[0079] The physical property mutation interface confirmation mark set is a direct product of the third-order diagnosis. It exists in the form of three-dimensional spatial coordinates and attribute labels, including spatial coordinates (recording the three-dimensional coordinates of all grid nodes confirmed as real geological phase transitions), interface normal vector (recording the three-dimensional gradient direction vector of the mutation interface), and phase transition type label (labels automatically classified according to the physical property change characteristics, such as "high resistivity mutation - suspected water-bearing fault" and "low resistivity mutation - suspected metallic ore body boundary")).

[0080] The differentiated fusion weight matrix is ​​a product of second-order diagnosis and is a weight allocation matrix that is completely consistent with the three-dimensional grid topology. It records the specific weight ratio of surface data and downhole data for each conflicting grid node in the spatial association domain during fusion (e.g., node A: surface weight 0.1, downhole weight 0.9). This matrix is ​​used to provide transparent parameter basis for possible subsequent secondary fusion or artificial intervention.

[0081] The diagnostic traceability log and quality control report include: diagnostic path record (records the specific diagnostic path of each abnormal node (e.g., node 1024 → consistency index 0.65 → first-order not triggered → second-order triggered (volume ratio = 15) → termination of diagnosis)), abnormal fallback record (records the nodes that trigger the abnormal fallback mechanism during the inversion or fusion process and their processing status), and overall quality assessment (including macro-quality control indicators such as the overall root mean square error (RMSR) of the fused model and the spatial consistency improvement ratio).

[0082] Methods for multi-source data fusion include: Based on the fused 3D physical property model set in the multi-source anomaly fusion result set, 3D depth slices are divided along the Z-axis (depth direction) with a fixed step size. For the dominant geophysical exploration method on which the fusion model is based, the built-in physical resolution attenuation rule library (which stores the physical resolution attenuation calculation model of each geophysical exploration method and the corresponding attenuation-related parameters) is queried to calculate its theoretical spatial resolution at the current depth.

[0083] Because the attenuation laws of different physical fields are fundamentally different, the calculation methods for the theoretical spatial resolution of various geophysical exploration methods are also different.

[0084] For example, taking the gravity and magnetic method, its resolution decreases exponentially with depth. The commonly used calculation model is: Theoretical spatial resolution = R0 × e -z÷D Where R0 is the initial resolution of the land surface, e is the natural constant, z is the current depth, and D is the stored corresponding attenuation depth constant.

[0085] The theoretical spatial resolution is normalized and compared with the maximum resolution of this detection method across the entire detection depth range to generate a dimensionless resolution confidence factor (with a value ranging from 0 to 1).

[0086] Since the integrated three-dimensional physical property model contains physical dimensions such as resistivity and density, it needs to be uniformly mapped to dimensionless mineralization probability.

[0087] The typical physical property response range of the target mineral is obtained by calling the regional rock physical statistical database, and then nonlinear mapping is performed through the Sigmoid membership function.

[0088] Basic mineralization probability = 1 ÷ (1 + exp(-k × (the physical attribute value of the current grid node - the median of the interval))); Where k is a shape parameter that controls the steepness of the mapping (calibrated by the historical borehole ore-bearing rate), and exp is a natural exponential function.

[0089] For the grid nodes within the fusion model, confidence coupling correction is performed: the product of the resolution confidence factor and the basic mineralization probability is calculated as the initial mineralization probability.

[0090] By traversing all grid nodes in the three-dimensional space, the initial mineralization probability is assigned to the corresponding grid node, and finally the initial three-dimensional mineralization probability cloud model is generated.

[0091] If a grid node has an empty basic mineralization probability due to data missing data marked in the diagnostic traceability log, an anomaly fallback strategy is triggered: read the background physical attribute value in the shallow 3D physical hard shell constraint model, and use the background value to obtain the basic mineralization probability for filling.

[0092] Methods for generating three-dimensional mineralization probability cloud models include: Based on the multi-source anomaly fusion result set, all grid nodes carrying property change interface confirmation markers in the initial three-dimensional mineralization probability cloud model are located and denoted as phase transition nodes, and phase transition boundary probability enhancement labels are attached to them.

[0093] For nodes with enhanced phase transition boundary probability labels, perform a calculation to increase the probability of mineralization: enhanced mineralization probability = min(1, initial mineralization probability + enhancement step size).

[0094] It should be noted that the enhancement step size is set based on regional historical borehole data and demonstration physical statistics. The mean difference between the ore body phase transition boundary and the background surrounding rock after the mineralization probability mapping is calculated. Then, the enhancement step size is set to 0.5 to 1.5 times the mean difference of this probability to ensure that the probability value of the phase transition node can significantly jump to the high probability range.

[0095] By designing a cutoff function of min(1.0,…), we can ensure that the enhanced mineralization probability value is strictly limited to the physical range of 0 to 1, preventing the probability value from overflowing, thereby directly highlighting the mineralization potential of the real geological phase transition area at the numerical level.

[0096] To address the issue of blurred ore body boundaries (i.e., smoothing artifacts) caused by traditional interpolation fusion, it is necessary to verify whether the direction of change of the probability field is consistent with the direction of change of the real physical field.

[0097] Therefore, for phase transition nodes and their first-order neighboring grid nodes, probability gradient direction vectors (obtained based on probability cloud models, representing the trend of probability change) and property gradient direction vectors (obtained based on a fused three-dimensional property model set, or the interface normal vectors in the property abrupt change interface confirmation mark set can be directly used, representing the trend of material property change) are extracted respectively.

[0098] Then, the spatial angle between the two vectors is calculated using the cosine formula of the spatial angle. This spatial angle quantifies the degree of normal deviation between the mathematical probability contour line and the actual physical property interface.

[0099] Set a direction consistency threshold. If the spatial angle is less than or equal to the direction consistency threshold, it is determined that the probability gradient and the physical property gradient are basically consistent. This triggers the consistency retention logic, which retains the current probability value without modification.

[0100] If the spatial angle is greater than the direction consistency threshold, a boundary smoothing artifact is determined to exist, triggering the forced projection sharpening logic. The probability gradient direction is forcibly projected onto the property gradient direction, and a corrected probability gradient vector is calculated and generated. The specific calculation logic is as follows: extract the magnitude of the original probability gradient vector of the current phase transition node (i.e., the intensity of the probability space rate of change) and the unit direction vector of the property gradient vector (i.e., the normal of the real property interface), perform scalar multiplication on the two, and generate the corrected probability gradient vector.

[0101] Subsequently, the mineralization probability values ​​of the phase transition node and its neighborhood are redistributed based on the corrected probability gradient vector, so that the normal of the redistributed probability contour lines remains parallel to the normal of the real physical property abrupt interface, thereby eliminating boundary smoothing artifacts.

[0102] If the original probability gradient vector magnitude is found to be 0 during the calculation, an anomaly fallback strategy is triggered, skipping the sharpening correction and directly retaining the original initial mineralization probability value.

[0103] It should be noted that the set directional consistency threshold must be greater than the inherent numerical truncation deviation caused by mesh discretization (usually between 15° and 25°) in order to identify boundary smoothing artifacts generated by the interpolation algorithm. It is usually set to 30° or 45°.

[0104] By traversing all target nodes, the final mineralization probability value after probability enhancement and gradient sharpening is updated to the 3D mesh, ultimately forming a 3D mineralization probability cloud model with clear boundaries and phase transition boundary probability enhancement labels.

[0105] The methods for generating borehole trajectory control commands include: Based on the three-dimensional mineralization probability cloud model, the information entropy of each grid node is extracted (the probability value of the grid node being in the mineralization and non-mineralization states is multiplied by its base-2 logarithm, summed, and the negative value is taken to obtain the information entropy) to characterize the overall geological uncertainty field distribution.

[0106] An uncertainty threshold is set based on the statistical quantile of the prior geological model (e.g., extracting the top 20% of the global information entropy). When the information entropy of a certain grid node is greater than the uncertainty threshold, it is marked as an area with excessive information entropy, and a candidate virtual borehole is generated at the location of that grid node.

[0107] Obtain surface construction condition data and conduct a physical surface construction feasibility assessment of candidate virtual boreholes: The assessment is composed of a linear superposition of a road accessibility factor (proportional to the inverse of the distance to the nearest road) and a water source security factor (proportional to the inverse of the distance to the nearest water source), and the weight coefficients of each factor are pre-calibrated using the analytic hierarchy process.

[0108] The physical limit parameters of the drilling rig and industry drilling safety construction specifications are used as the basis for setting the safety construction threshold to evaluate candidate virtual boreholes.

[0109] If the calculated feasibility score is lower than the set safe construction threshold, the location is determined not to meet the conditions for drilling rig entry, and the candidate hole is directly eliminated; the remaining holes are the candidate holes that meet the actual construction conditions.

[0110] For the selected candidate boreholes, a two-factor game mechanism considering information benefits and engineering costs is used to dynamically optimize the process based on the current exploration stage, generating borehole trajectory control instructions. Specifically, the optimization logic of the two-factor game mechanism is as follows: The spatial covariance radius (i.e., the three-dimensional spatial range within which the borehole data can effectively radiate and constrain) is calculated based on the spatial variogram. Assuming that real data is obtained from the borehole, the probability distribution of each grid node after updating within the spatial covariance radius is mathematically derived using spatial interpolation algorithms (such as variance reduction in Kriging). The information entropy of the grid within the influence range before drilling (prior) and after simulated drilling (posterior) is calculated respectively. The sum of the prior information entropy is subtracted from the sum of the posterior information entropy to obtain the expected reduction in information entropy of the candidate borehole.

[0111] The optimal value of the candidate borehole is obtained by multiplying the expected reduction in information entropy and the drilling cost by their respective dynamic weighting coefficients and then subtracting them.

[0112] The weighting coefficients are adaptively adjusted according to the current exploration stage (high information weight in the early general survey stage and high cost weight in the later detailed survey stage), and a marginal benefit feedback mechanism is introduced. When the actual benefits of historical boreholes are consistently lower than expected, the information weight of that area is reduced by a set step size.

[0113] Based on the optimal values ​​of all candidate holes, the top K candidate holes with the highest optimal values ​​are selected as the initial optimal batch; where K refers to the quantity.

[0114] To avoid overlapping borehole information, a spatial exclusion check is performed: if the spatial distance between any two boreholes in a batch is less than the spatial covariance radius, the low-scoring borehole is discarded and replaced by a subsequent high-scoring borehole.

[0115] The finalized selection of multiple preferred boreholes, along with their corresponding 3D terrain and borehole inclination and azimuth constraints, are packaged into a borehole trajectory control command and sent to the physical drilling rig.

[0116] The methods for locally updating the 3D physical property model set and the 3D mineralization probability cloud model based on drilling verification results, and generating exploration termination or reversal commands, include: After the physical drilling rig executes the issued drilling trajectory control command, it obtains the real physical property data of the drill core (referring to the real laboratory physical property measurement value, such as the real resistivity). Then, it extracts the region to which the drill trajectory passes through the grid node in the fused three-dimensional physical property model set in the multi-source anomaly fusion result set belongs, matches and extracts the physical property value of the node in the corresponding sub-model (terrain correction model, weighted fusion model or high-fidelity preservation model) as the model prediction value.

[0117] Calculate the absolute value of the difference between the actual physical property value and the model predicted physical property value, and then divide it by the actual physical property value to obtain the relative prediction deviation rate; simultaneously read the phase change type label that the borehole trajectory passes through in the physical property change interface confirmation mark set; compare the actual lithology revealed by drilling with the label to generate the phase change label hit rate.

[0118] Based on the error tolerance limits specified in the mineral exploration specifications, a deviation tolerance threshold is set, and different closed-loop update modes are triggered according to the relative prediction deviation rate and the phase change tag hit rate.

[0119] When the relative prediction deviation rate is less than or equal to the deviation tolerance threshold and the phase change label is hit, the model is determined to be reliable overall, triggering the normal mode, which only performs probability assimilation updates on the grid nodes in the local neighborhood of the borehole.

[0120] When the relative prediction deviation rate exceeds the deviation tolerance threshold, or when the phase change label is misjudged (e.g., predicted as a metallic ore body but actually a water-bearing fault), it is determined that there is a systematic deviation in the prior geological model, triggering the emergency mode. The grid reconstruction radius is doubled, the differential fusion weight matrix of the area is reset, and an exploration redirection command is generated. At the same time, a prompt is sent to the operator to re-execute the prior geological model and locally encrypted acquisition of multi-source original exploration data.

[0121] After executing the normal mode or emergency mode, the corresponding regions of the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated, forming a closed-loop iterative link.

[0122] After each partial update, the macro-level quality control indicators in the diagnostic traceability log and quality control report are updated. An iterative termination mechanism is used to determine the closed-loop termination, considering two conditions: First, the average information entropy of the global three-dimensional grid is reduced to below the accuracy threshold (this threshold is calculated in reverse based on the confidence level of the reserve level required by the exploration specifications of the target mineral type, or by taking the statistical benchmark of the historical final state model of similar mineral deposits).

[0123] Second, the root mean square error of the overall data in the diagnostic traceability log and quality control report converges to the allowable range (this range is based on the ore body delineation error tolerance specified in the exploration specifications, and is set by superimposing the natural physical variation coefficient of the rock to accommodate geological noise).

[0124] When both conditions are met simultaneously, or when the cumulative project cost reaches the budget limit, an exploration termination command is generated, ending the closed-loop iteration link.

[0125] Example 2 Please see Figure 2 As shown, for parts not described in detail in this embodiment, please refer to the description in Embodiment 1. A high-precision intelligent detection and assessment method for deep mineral resources is provided, including: S1: Obtain the prior geological model and collect multi-source raw exploration data, reverse derive the acquisition control command to regulate the energy of the acquisition process and add depth layer markers; after the acquisition is completed, evaluate the data quality and mark data conflicts, and generate a multidimensional standardized dataset.

[0126] S2: Based on a multidimensional standardized dataset, a shallow constraint model is constructed. Shallow interference is filtered out by deep layer labeling, and inversion permission commands are issued to limit the scope of model modification, generating a set of three-dimensional physical property models.

[0127] S3: Based on the three-dimensional physical property model set, establish the spatial correlation domain between the surface and the well; then read the data conflict identifier to perform multi-stage physical source tracing diagnosis, and confirm the physical property mutation interface to form a multi-source anomaly fusion result set.

[0128] S4: Based on the multi-source anomaly fusion result set, evaluate the resolution confidence of each detection depth, and then combine the confidence with the source tracing diagnosis results to perform multi-source data fusion, and sharpen the boundaries of the marked physical property change interface regions to generate a three-dimensional mineralization probability cloud model with enhanced labels.

[0129] S5: Based on the three-dimensional mineralization probability cloud model, the distribution of geological uncertainties is deduced, and candidate virtual boreholes are generated; combined with the feasibility of surface construction and engineering costs, comprehensive optimization is performed to generate borehole trajectory control instructions; based on the drilling verification results, the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated, and exploration termination or reversal instructions are generated to form a closed-loop iterative link.

[0130] Example 3 This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the high-precision intelligent detection and evaluation system for deep mineral resources described above.

[0131] Since the electronic device described in this embodiment is the one used to implement the high-precision intelligent detection and assessment method for deep mineral resources in this application embodiment, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the high-precision intelligent detection and assessment method for deep mineral resources described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any electronic device used by those skilled in the art to implement the high-precision intelligent detection and assessment method for deep mineral resources in this application embodiment falls within the scope of protection of this application.

[0132] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0133] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A high-precision intelligent detection and assessment system for deep mineral resources, characterized in that, include: Multidimensional data acquisition and standardization module: acquires prior geological models and collects multi-source raw exploration data, reverse-derives acquisition control commands to regulate energy during the acquisition process and add depth stratification markers; After data collection is completed, data quality is evaluated and data conflicts are marked to generate a multidimensional standardized dataset. Shallow and deep constraint and deep inversion module: Based on a multidimensional standardized dataset, a shallow constraint model is constructed, shallow interference is filtered out through depth layer labeling, and inversion permission commands are issued to limit the scope of model modification, generating a set of three-dimensional physical property models; Multi-source spatial correlation and source tracing diagnosis module: Based on a three-dimensional physical property model set, a spatial correlation domain between the surface and underground is established; then, data conflict identifiers are read to perform multi-stage physical source tracing diagnosis, and physical property abrupt change interfaces are confirmed and marked to form a multi-source anomaly fusion result set; Boundary sharpening module: Based on the multi-source anomaly fusion result set, it evaluates the resolution confidence of each detection depth, and then combines the confidence with the source tracing diagnosis results to perform multi-source data fusion, and sharpens the boundaries of the marked physical property change interface areas to generate a three-dimensional mineralization probability cloud model. Uncertainty extrapolation and drilling closed-loop iteration module: Based on the three-dimensional mineralization probability cloud model, the geological uncertainty distribution is extrapolated to generate candidate virtual boreholes; comprehensive optimization is carried out by combining surface construction feasibility and engineering cost to generate borehole trajectory control instructions; the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated according to the drilling verification results, and exploration termination or reversal instructions are generated to form a closed-loop iteration link.

2. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 1, characterized in that, The methods for generating multidimensional standardized datasets include: A priori geological model is acquired and multi-source raw detection data are collected simultaneously to construct a five-dimensional unified data container with grid nodes. During the acquisition process, the transmission current and reception time of the physical acquisition equipment are deduced in reverse based on the physical penetration rule to generate acquisition control commands and adaptively regulate the acquisition energy. After the data acquisition is completed, the actual penetration depth is deduced by reverse calculation based on the actual transmission current and reception time, and corresponding depth layer markers are added to each detection data. The quality of each detection data is evaluated by a multi-factor composite rule of physical quality, and data conflict detection and labeling are performed based on the comparison results of the difference between multi-source detection data in the same spatial grid and the instrument nominal error, so as to generate a multi-dimensional standardized dataset.

3. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 2, characterized in that, The methods for constructing shallow constraint models include: Based on a multidimensional standardized dataset, physical properties with shallow depth classification labels are extracted, spatial variation feature analysis is performed, and then the range is extracted based on the spatial variation features. Spatial interpolation is performed on each grid node to construct a shallow three-dimensional physical hard shell. The data coverage of grid nodes is evaluated based on the distribution of probe data within each spatial grid. A three-level adaptive degradation process is then performed to match the corresponding constraint type and allowable adjustment range for each grid node, generating a shallow constraint model with control labels.

4. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 3, characterized in that, The methods for generating the three-dimensional property model set include: The physical properties of deep grid nodes in the multidimensional standardized dataset are extracted to obtain the comprehensive physical property gradient modulus. Then, the physical property abrupt change interface is identified by the boundary judgment rule, and a suspected physical property abrupt change interface marker is generated for the grid nodes on the interface. Read the depth layering labels of all grid nodes in the five-dimensional unified data container, assign mask values ​​to different layer regions, construct a three-dimensional mask matrix, and filter all grid nodes marked as shallow regions; The multidimensional standardized dataset is divided into a surface exploration dataset and a downhole exploration dataset. Then, an inversion permission control command is issued to couple the shallow constraint model with the 3D mask matrix, lock the shallow grid nodes and open the modification permissions of the middle and deep grid nodes, and drive the inversion engine to perform independent inversion of the surface and downhole data respectively, generating a multi-source deep independent 3D physical property model set containing the surface 3D physical property model and the downhole 3D physical property model.

5. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 4, characterized in that, The methods for forming the multi-source anomaly fusion result set include: Based on the three-dimensional physical property model set, the grid nodes with overlapping spatial coordinates in the three-dimensional physical property models of the surface and underground are extracted to construct a spatial association domain, and the spatial consistency index is obtained to identify low consistency areas. For regions with low consistency, read data conflict markers to trigger a three-stage progressive physical source diagnosis: the first stage investigates static effects to eliminate topographic distortion, the second stage investigates volumetric effects to generate differentiated fusion weight instructions, and the third stage investigates geological phase changes to generate physical property mutation interface confirmation markers, forming a multi-source anomaly fusion result set.

6. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 5, characterized in that, The methods for performing multi-source data fusion include: Based on the multi-source anomaly fusion result set, three-dimensional depth slices are divided, the built-in physical resolution attenuation rule library is queried, the resolution confidence factor within each detection depth range is quantified, and the physical attributes of each network node are nonlinearly mapped to the basic mineralization probability. Then, the resolution confidence factor and the basic mineralization probability are coupled together to obtain the initial mineralization probability, which is assigned to the corresponding grid node and then integrated into the initial three-dimensional mineralization probability cloud model.

7. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 6, characterized in that, The methods for generating the three-dimensional mineralization probability cloud model include: Based on the multi-source anomaly fusion result set, the grid nodes carrying the physical property abrupt change interface confirmation mark in the initial three-dimensional mineralization probability cloud model are located and denoted as phase transition nodes. The mineralization probability is enhanced and phase transition boundary probability enhancement labels are added to them. Furthermore, for phase transition nodes and their neighboring grid nodes, the spatial angle between the probability gradient direction and the physical property gradient direction is extracted, and the consistency preservation logic or forced projection sharpening logic is triggered according to the size of the spatial angle. If the forced projection sharpening logic is triggered, the probability gradient direction will be forcibly projected to the physical property gradient direction to redistribute the mineralization probability value. Based on the final mineralization probability after probability enhancement and gradient sharpening, a three-dimensional mineralization probability cloud model is updated and formed.

8. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 7, characterized in that, The methods for generating borehole trajectory control commands include: Based on the three-dimensional mineralization probability cloud model, the information entropy of each grid node is extracted to infer the distribution of geological uncertainty field, and candidate virtual boreholes are generated in the area where the information entropy exceeds the limit. Obtain surface construction condition data, conduct surface construction feasibility assessment on candidate virtual boreholes, and screen out candidate boreholes that meet the actual construction conditions; Furthermore, for candidate boreholes, a two-factor game mechanism of information benefits and engineering costs is used to dynamically and comprehensively optimize based on the current exploration stage, and generate borehole trajectory control instructions.

9. The high-precision intelligent detection and assessment system for deep mineral resources according to claim 8, characterized in that, The method of locally updating the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model based on drilling verification results, and generating exploration termination or reversal instructions, includes: The real physical property data of the rock core extracted by the drilling rig executing the drilling trajectory control command are obtained as the drilling verification result, and the physical property values ​​of the corresponding grid nodes in the multi-source anomaly fusion result set are extracted as the model prediction values; then the relative prediction deviation rate between the real physical property data and the model prediction values ​​is obtained. When the relative prediction deviation rate does not exceed the limit, the normal mode is triggered to perform probability assimilation on the local neighborhood grid of the borehole; when the relative prediction deviation rate exceeds the limit, the emergency mode is triggered to expand the grid reconstruction range and generate exploration turning instructions; after executing the normal mode or the emergency mode, the corresponding regions of the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated. After each update, an iterative termination determination mechanism is used to determine the termination based on the global grid average information entropy and cumulative engineering cost. If the determination conditions are met, an exploration termination command is generated and executed.

10. A high-precision intelligent detection and assessment method for deep mineral resources, implemented based on the high-precision intelligent detection and assessment system for deep mineral resources as described in any one of claims 1 to 9, characterized in that, include: S1: Obtain the prior geological model and collect multi-source raw exploration data, reverse-derive the acquisition control command to regulate the energy of the acquisition process and add depth layer markers; After data collection is completed, data quality is evaluated and data conflicts are marked to generate a multidimensional standardized dataset. S2: Based on a multidimensional standardized dataset, a shallow constraint model is constructed. Shallow interference is filtered out by depth layer labeling, and an inversion permission command is issued to limit the scope of model modification, generating a three-dimensional physical property model set. S3: Based on the three-dimensional physical property model set, establish the spatial correlation domain between the surface and the well; then read the data conflict identifier to perform multi-stage physical source tracing diagnosis, and confirm the physical property change interface to form a multi-source anomaly fusion result set; S4: Based on the multi-source anomaly fusion result set, evaluate the resolution confidence of each detection depth, and then combine the confidence with the source tracing diagnosis results to perform multi-source data fusion, and sharpen the boundaries of the marked physical property change interface areas to generate a three-dimensional mineralization probability cloud model with enhanced labels. S5: Based on the three-dimensional mineralization probability cloud model, the distribution of geological uncertainties is deduced, and candidate virtual boreholes are generated; combined with the feasibility of surface construction and engineering costs, comprehensive optimization is performed to generate borehole trajectory control instructions; based on the drilling verification results, the three-dimensional physical property model set and the three-dimensional mineralization probability cloud model are locally updated, and exploration termination or reversal instructions are generated to form a closed-loop iterative link.

Images