A method for constructing a mineral resource prediction model based on a complex geological background

Abstract

The application discloses a kind of based on complex geological background's mineral resources prediction model construction method, it is related to mineral resources prediction technical field, based on the geological background of unknown mining area Construction of non-mineralization interference three-dimensional ideal geological background field, three-dimensional ideal geological background field is used to show the mine field of non-mineralization area;Based on exploration data to construct three-dimensional real geological field, compare three-dimensional real geological field and three-dimensional ideal geological background field to obtain first difference distribution data;Extract the geological features of known mineralization area, generate interference factor based on geological features;Interference factor is introduced into ideal geological background field to generate interference coupling background field, extract the difference between interference coupling background field and ideal geological background field as second difference distribution data;Based on first difference distribution data and second difference distribution data to construct mineral resources prediction model, the present application can adapt to the prediction demand of different geological background, significantly improve the intelligent level and application flexibility of mineral resources prediction system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mineral resource prediction, and particularly relates to a mineral resource prediction model construction method based on a complex geological background. BACKGROUND

[0002] The mainstream technical route in the current mineral resource prediction field mainly includes a data-driven method based on statistics and an expert experience model based on geological concepts. The data-driven method such as the evidence weight method, the random forest, the support vector machine and other machine learning algorithms realizes spatial distribution prediction of mineral resources through statistical analysis and feature mining of a large amount of exploration data and known deposit data.

[0003] The existing technology performs outstandingly in a region with sufficient data and clear mine area types, but generally relies on historical data and statistical correlation, and is difficult to fully express the dynamic evolution of complex geological processes, and has poor adaptability to unknown or rare ore-forming types, and limited interpretability. Meanwhile, the expert knowledge model based on geological concepts relies on the deep understanding of experts on deposit types and geological ore-controlling factors, and uses genetic models and deposit models to make ore prediction. This kind of method is widely used in areas with mature ore-forming geological understanding, but has strong subjectivity, and the modeling process relies on expert experience and qualitative analysis, and is difficult to systematically integrate multi-source complex data, and the model has insufficient generalization and generalization ability, and is difficult to meet the fine prediction needs of complex geological backgrounds such as deep and concealed deposits.

[0004] Therefore, it is of great significance to develop a mineral resource prediction model construction method based on a complex geological background. SUMMARY

[0005] The present application aims to provide a mineral resource prediction model construction method based on a complex geological background to solve the problems in the background art.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a mineral resource prediction model construction method based on a complex geological background, comprising:

[0007] Constructing a three-dimensional ideal geological background field without mineralization interference based on the geological background of unknown mine areas, and the three-dimensional ideal geological background field is used to show the mine field of non-mineralization areas;

[0008] Constructing a three-dimensional real geological field based on exploration data, and obtaining first difference distribution data by comparing the three-dimensional real geological field and the three-dimensional ideal geological background field;

[0009] Extracting geological features of known mineralization areas, and generating interference factors based on the geological features;

[0010] The interference factor is introduced into the ideal geological background field to generate an interference coupling background field, and a difference between the interference coupling background field and the ideal geological background field is extracted as second difference distribution data;

[0011] A mineral resource prediction model is constructed based on the first difference distribution data and the second difference distribution data.

[0012] In a preferred embodiment, the step of constructing the three-dimensional ideal geological background field of non-mineralization interference is:

[0013] Geological background data of an unknown mining area are collected, and a three-dimensional ideal geological background field is constructed based on an ideal background field generation model;

[0014] The ideal background field generation model includes a lithology distribution simulator, a tectonic framework generator, a physical property field calculator, and a voxel segmenter;

[0015] The lithology distribution simulator uses geological background data to constrain lithology transition probability based on a Markov random field algorithm;

[0016] The tectonic framework generator generates a fracture network that conforms to mechanical laws based on B-spline surface fitting of stress field simulation to generate a tectonic generation environment;

[0017] The physical property field calculator calculates the density, magnetic susceptibility, and resistivity fields of the unknown mining area based on rock physics empirical formula and joint field forward modeling;

[0018] The voxel segmenter segments the three-dimensional ideal geological background field into multiple ideal voxels.

[0019] In a preferred embodiment, the step of constructing the three-dimensional real geological field is:

[0020] Exploration data of an unknown mining area are obtained, and a three-dimensional real geological field is constructed based on a real background field generation model using the exploration data;

[0021] The real background field generation model includes a data converter, an anomaly converter, and a voxel segmenter;

[0022] The data converter converts discrete exploration data into three-dimensional raster data with the same resolution as the three-dimensional ideal geological background field based on a Kriging interpolation algorithm;

[0023] The anomaly converter extracts effective anomaly bodies that conform to geological laws based on statistical threshold analysis and morphological processing;

[0024] The voxel segmenter segments the three-dimensional real geological field into multiple real voxels.

[0025] In a preferred embodiment, the step of obtaining the first difference distribution data by comparing the real geological background field with the three-dimensional ideal geological background field is:

[0026] Voxel registration is performed between the real geological background field and the three-dimensional ideal geological background field;

[0027] Voxel differences are calculated between the corresponding ideal voxels and the actual voxels. Voxel differences include differences in lithological classification, relative deviations in geochemistry, and absolute differences in geophysics.

[0028] Based on morphological opening operations and connected component analysis, real-world voxels with differences are connected to form effective outlier regions as the first differential distribution data.

[0029] In a preferred embodiment, the step of extracting geological features of a known mineralized area and generating interference factors based on those features is as follows:

[0030] A geological knowledge graph of a mining area is constructed based on the known geological data and background data of the existing mining area. The geological knowledge graph of a mining area includes element nodes and element association edges.

[0031] Geological features are extracted from geological data based on feature extraction algorithms;

[0032] Vectorize the element nodes and geological features and project them onto the matching space;

[0033] In the matching space, the matching element node with the highest matching degree is obtained based on the dual-tower attention matching model;

[0034] Multiple interference factors are constructed based on the matching feature nodes. The interference factors include the voxel range constraint, the attribute assigner, and the attribute modifier.

[0035] In a preferred embodiment, the step of introducing the interference factor into the ideal geological background field to generate the interference coupled background field is as follows:

[0036] The action voxel range constraint converts the action range defined by multiple interference factors into voxel index coordinates of a three-dimensional ideal geological background field to obtain the action target voxel;

[0037] The attribute allocator generates target interference factors by analyzing the lithological code, tectonic location, and physical property threshold of the target voxel and injecting corresponding attributes into the corresponding interference factors.

[0038] The target interference factor is deployed into the target voxel, and the corresponding attribute stored in the target interference factor is released based on the attribute modifier.

[0039] The updated voxel is obtained by modifying the attributes of the target voxel based on the corresponding attributes;

[0040] Examine the attribute deviation between the updated voxel and its neighboring voxels, and define two voxels whose attribute deviation exceeds a preset safety threshold as conflicting voxels;

[0041] Based on the conflict resolution mechanism, the properties of the conflict voxels are modified until all target voxels are updated, generating an interference coupling background field.

[0042] In a preferred embodiment, the step of extracting the difference between the interference coupling background field and the ideal geological background field as the second difference distribution data is as follows:

[0043] The interference coupling background field is segmented into multiple interference coupling voxels;

[0044] Voxel differences are calculated for the corresponding ideal voxels and interference coupled voxels. Voxel differences include lithological classification differences, geochemical relative deviations, and geophysical absolute differences.

[0045] Based on morphological opening operations and connected component analysis, the interfering coupled voxels with differences are connected to form effective outlier regions as the second differential distribution data.

[0046] In a preferred embodiment, the step of constructing a mineral resource prediction model based on the first differential distribution data and the second differential distribution data is as follows:

[0047] Voxel-level spatial coupling analysis was performed on the first and second differential distribution data to calculate the overlap rate and difference intensity of the overlapping anomaly region.

[0048] Based on the overlap rate and the intensity of the difference, the overlapping abnormal areas are divided into three levels, A, B, and C, according to a preset threshold.

[0049] A three-dimensional mineralization probability field is generated by weighted fusion of the overlapping anomaly regions of the first and second differential distribution data;

[0050] The three-dimensional mineralization probability field is converted into voxel index coordinates of a three-dimensional ideal geological background field, and a mineral resource prediction model is generated based on the voxel index coordinates mapped to the three-dimensional ideal geological background field.

[0051] Mineral resource prediction models include mineral distribution, mineral type identification, mineral probability values, and confidence levels.

[0052] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0053] 1. This invention achieves cross-validation of mineral prediction under complex geological backgrounds by constructing a dual difference comparison between an ideal geological background field and a real geological background field, combined with the theoretical response mode of the interference coupling field. The first difference analysis comprehensively captures potential mineralization signals by comparing the spatial anomaly distribution of the real field and the ideal field; the second difference analysis verifies the geological rationality of the anomalies based on the theoretical interference model. This dual validation mechanism can effectively distinguish mineralization-related anomalies from non-mineralization geological noise. It adopts a voxel-level spatial coupling algorithm, which is suitable for the identification and evaluation of weak anomalies under complex geological backgrounds. It can effectively overcome the multi-solution problem common in traditional single anomaly analysis methods and significantly improve the reliability of prediction results.

[0054] 2. This invention, through the construction of a knowledge graph-driven interference factor system, realizes a closed loop of knowledge transformation from geological features to prediction models. It automatically identifies key ore-controlling elements using a multi-scale feature extraction algorithm and achieves intelligent matching with unknown mining areas through a matching algorithm. Based on clearly defined voxel ranges, spatial constraints, propagation mechanisms, and parameter modification rules, the interference factor can perform differentiated simulations for different geological processes and mineralization processes. Through the interference factor mechanism, data obtained from different exploration methods and expert geological knowledge can be deeply integrated to form a parameterized and spatialized interference model. This knowledge-driven method breaks through the limitations of traditional empirical models, can adapt to the prediction needs of different geological backgrounds, and significantly improves the intelligence level and application flexibility of mineral resource prediction systems. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0056] Figure 1 This is a flowchart of the method of the present invention.

[0057] Figure 2 This is a logic block diagram of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0059] Example 1, please refer toFigure 1 and Figure 2 As shown in this embodiment, a method for constructing a mineral resource prediction model based on a complex geological background includes:

[0060] S1. Construct a three-dimensional ideal geological background field without mineralization interference based on the geological background of the unknown mining area. The three-dimensional ideal geological background field is used to display the mining area without mineralization.

[0061] S2. Construct a three-dimensional real geological field based on exploration data, and obtain the first difference distribution data by comparing the three-dimensional real geological field with the three-dimensional ideal geological background field.

[0062] S3. Extract the geological features of known mineralized areas and generate interference factors based on the geological features;

[0063] S4. Introduce the interference factor into the ideal geological background field to generate the interference coupled background field, and extract the difference between the interference coupled background field and the ideal geological background field as the second difference distribution data.

[0064] S5. Construct a mineral resource prediction model based on the first differential distribution data and the second differential distribution data;

[0065] As described in steps S1-S5 above, the mainstream technical routes in the field of mineral resource prediction mainly include data-driven methods based on statistics and expert experience models based on geological concepts. Data-driven methods, such as the evidence weight method, random forest, support vector machine and other machine learning algorithms, realize the spatial distribution prediction of mineral resources through statistical analysis and feature mining of a large amount of exploration data and known mineral deposit data.

[0066] Existing technologies perform well in areas with abundant data and clearly defined mining types, but they generally rely on historical data and statistical correlations, making it difficult to fully express the dynamic evolution of complex geological processes. They also have poor adaptability to unknown or rare mineralization types and limited interpretability. Meanwhile, expert knowledge models based on geological concepts rely on experts' deep understanding of deposit types and geological ore-controlling factors, and use genetic models, deposit models, and other methods for mineral exploration prediction. These methods are widely used in areas with mature metallogenic geology, but they are highly subjective. The modeling process depends on expert experience and qualitative analysis, making it difficult to systematically integrate complex multi-source data. Furthermore, the models lack generalization and extensibility, making it difficult to meet the refined prediction needs of complex geological backgrounds such as deep and concealed deposits.

[0067] This invention achieves cross-validation of mineral prediction under complex geological backgrounds by constructing a dual difference comparison between an ideal geological background field and a real geological background field, combined with the theoretical response mode of the interference coupling field. The first difference analysis comprehensively captures potential mineralization signals by comparing the spatial anomaly distribution of the real field and the ideal field; the second difference analysis verifies the geological rationality of the anomalies based on the theoretical interference model. This dual validation mechanism can effectively distinguish mineralization-related anomalies from non-mineralization geological noise. It adopts a voxel-level spatial coupling algorithm, which is suitable for the identification and evaluation of weak anomalies under complex geological backgrounds. It can effectively overcome the multi-solution problem common in traditional single anomaly analysis methods and significantly improve the reliability of prediction results.

[0068] By constructing a knowledge graph-driven interference factor system, a closed loop of knowledge transformation from geological features to prediction models is realized. A multi-scale feature extraction algorithm automatically identifies key ore-controlling elements, and a matching algorithm enables intelligent matching with unknown mining areas. Based on clearly defined voxel ranges, spatial constraints, propagation mechanisms, and parameter modification rules, the interference factor can perform differentiated simulations for different geological processes and mineralization processes. Through the interference factor mechanism, data obtained from different exploration methods and expert geological knowledge can be deeply integrated to form a parameterized and spatialized interference model. This knowledge-driven approach breaks through the limitations of traditional empirical models, adapts to the prediction needs of different geological backgrounds, and significantly improves the intelligence level and application flexibility of the mineral resource prediction system.

[0069] In one embodiment, step S1 of constructing a three-dimensional ideal geological background field free from mineralization interference includes:

[0070] S11. Collect geological background data of unknown mining areas and construct a three-dimensional ideal geological background field based on the ideal background field generation model;

[0071] S12. The ideal background field generation model includes a lithology distribution simulator, a structural framework generator, a physical property field calculator, and a voxel segmenter.

[0072] S13, The lithology distribution simulator is based on the Markov random field algorithm and uses geological background data to constrain the lithology transfer probability;

[0073] S14. The lattice generator is based on B-spline surface fitting stress field simulation to generate a fracture network that conforms to mechanical laws.

[0074] S15, the physical property field calculator calculates the density, magnetic susceptibility and resistivity fields of unknown mining areas based on empirical rock physics formulas and potential field forward modeling.

[0075] S16. The voxel segmenter divides the three-dimensional ideal geological background field into multiple ideal voxels;

[0076] As described in steps S11-S16 above, basic geological data of the unknown mining area are collected, including regional stratigraphic columnar section, standard profile lithological sequence, regional tectonic outline map and physical property parameter library. Based on this, an ideal background field generation model is invoked. This model uses geostatistics as a framework, defines the three-dimensional spatial range through boundary conditions (such as basin outlines and unconformities), and initializes the grid resolution. The ideal background field generation model includes a lithology distribution simulator, a structural lattice generator, a physical property field calculator, and a voxel segmenter. The lithology distribution simulator uses the Markov random field algorithm, constructing a conditional probability field based on borehole core data and surface outcrop lithology maps as prior knowledge. This algorithm simulates the spatial correlation of lithological units. For example, when sandstone exists around a certain area, the probability that it is shale itself is significantly increased. Through iterative calculation, a three-dimensional lithological skeleton model that conforms to geological laws (such as sedimentary facies transitions and intrusive contact zones) is generated. The structural lattice generator is used for stress field reconstruction and fracture network generation. Stress field reconstruction generates a smooth stress distribution field by inputting the principal stress direction and intensity parameters of the region and using B-spline surface technology to simulate the stress state of the crust. Fracture network generation automatically generates fracture zones in stress concentration areas according to the Coulomb fracture criterion (rock fractures under specific compressive and shear stresses). The algorithm ensures that the fracture morphology conforms to geomechanical principles (such as the dip angle of normal faults) and adds secondary fractures through fractal algorithms to enhance realism. The physical property calculator includes a rock physics conversion module and a geophysical forward modeling module. The rock physics conversion module matches empirical formulas based on lithology type (e.g., using seismic wave velocity to calculate density and porosity to estimate resistivity). The geophysical forward modeling module generates a density field and uses a gravity forward modeling algorithm to infer the distribution of gravity anomalies. Magnetic anomalies are simulated based on a magnetic susceptibility field. Multi-field data are cross-checked to ensure consistency of physical property parameters. The voxel segmenter discretizes the continuous three-dimensional geological field into a regular cubic mesh. Each voxel stores attribute labels including but not limited to lithology code, density value, magnetic susceptibility value, and resistivity. Rate values ​​and structural markers (such as whether it is located in a fault zone) are used. Further, adaptive mesh technology can be used to optimize storage efficiency. The four core components work together to generate a three-dimensional ideal geological background field. The lithology simulator generates the original lithology framework, the structural framework generator adds a fault / fold system (cuts and modifies lithological units), the physical property field calculator allocates physical property parameters according to lithology-structure association rules (e.g., rock density decreases in fault zones), and the voxel segmenter outputs a structured mesh model. All mineralization and alteration-related parameter inputs (such as mineralization element content and alteration mineral assemblage) are disabled throughout the process. Only regional background geological features are retained. The final output three-dimensional model is an ideal geological background field without mineralization interference, providing a pure background for subsequent mineralization anomaly identification.

[0077] In one embodiment, step S2 of constructing a three-dimensional realistic geological field includes:

[0078] S21. Obtain exploration data of unknown mining areas and use the exploration data to construct a three-dimensional realistic geological field based on the real background field generation model.

[0079] S22. The real-world background field generation model includes a data converter, an anomaly converter, and a voxel segmenter.

[0080] S23. The data converter is based on the Kriging interpolation algorithm to convert discrete exploration data into three-dimensional raster data with the same resolution as the three-dimensional ideal geological background field.

[0081] S24. The anomaly converter extracts valid anomalies that conform to geological laws through statistical threshold analysis and morphological processing.

[0082] S25, The voxel segmenter divides the three-dimensional real geological field into multiple real voxels;

[0083] As described in steps S21-S25 above, multi-source exploration data of the unknown mining area is obtained, including but not limited to borehole cores, geochemical sampling points, geophysical survey lines, and remote sensing interpretation results. This data is then input into a customized real-world background field generation model. This model employs a modular pipeline design to ensure the entire process of transforming raw data into a three-dimensional realistic geological field. The real-world background field generation model includes three core processors: a data converter to solve the problem of spatial continuity of discrete data; an anomaly converter to extract geologically significant anomalous signals; and a voxel segmenter to construct a computable three-dimensional mesh voxel. The data converter is implemented using a Kriging algorithm framework, automatically optimizing interpolation parameters based on the spatial distribution characteristics of sampling points. By analyzing the spatial autocorrelation of exploration points, it dynamically adjusts the influence weights in different directions and introduces geological boundary constraints (such as fault lines and lithological contact zones) to prevent... To prevent erroneous interpolation across geological units, the system outputs a 3D raster data volume perfectly aligned with the ideal geological background field. Resolution is forcibly matched through voxel size parameters. The anomaly converter processing logic includes a multi-level anomaly filtering mechanism. Statistical threshold stratification is used to independently calculate the background distribution for each geophysical / chemical parameter. The cumulative frequency curve inflection point method is used to determine the lower limit of anomalies, distinguishing between the background field and the anomaly field. Morphological optimization is performed by first performing erosion to eliminate isolated noise points, and then performing dilation to connect spatially adjacent potential anomaly areas. Geological regularity verification is performed by fusing tectonic mineralization control models (such as anomaly zones extending along fault strikes) and cross-validating the spatial coupling of multi-parameter anomalies (such as overlapping areas of high resistivity + Cu element anomalies). The voxel segmenter performs attribute inheritance segmentation, cutting the 3D raster data according to a preset voxel size (such as 50m×50m×20m). Each real voxel inherits the multi-dimensional attribute labels of the original raster.

[0084] In one embodiment, step S2, which involves comparing the real geological background field with the three-dimensional ideal geological background field to obtain the first difference distribution data, includes:

[0085] S26. Perform voxel registration between the real geological background field and the three-dimensional ideal geological background field;

[0086] S27. Calculate the voxel difference between the corresponding ideal voxels and the actual voxels. The voxel difference includes lithological classification differences, geochemical relative deviations and geophysical absolute differences.

[0087] S28. Based on morphological opening operation and connected component analysis, real voxels with differences are connected into effective outlier regions as the first difference distribution data;

[0088] As described in steps S26-S28 above, ensuring strict alignment of voxel spatial positions between the real geological background field and the three-dimensional ideal geological background field requires the mandatory use of the mining area geodetic coordinate system and elevation datum, and assigning globally unique three-dimensional index coordinates to all voxels. Furthermore, during the registration process, tectonic displacement compensation and key stratigraphic calibration can be introduced. Tectonic displacement compensation identifies fault displacement caused by tectonic movement in the real field, establishes local displacement field models on both sides of the fault, and dynamically adjusts voxel positions to achieve tectonic alignment. Key stratigraphic calibration involves selecting landmark strata (such as those containing ore). Using the rock layer and volcanic ash layer as calibration layers, a thin-plate spline interpolation algorithm is employed to spatially overlap the actual geological background field with the real geological background field. Differential processing is used to calculate difference types, including lithological classification differences, geochemical relative deviations, and geophysical absolute differences. Lithological classification differences can be compared using semantic encoding: a value of 1 indicates inconsistency in the main lithological class, 0.5 indicates inconsistency in the subclass, and 0 indicates complete consistency, used to identify lithological variations caused by alteration and mineralization. Geochemical relative deviations are calculated element-wise, determining the difference between element values ​​in the actual geological background field and the three-dimensional ideal geological background field. Dividing by the elemental standard deviation of the three-dimensional ideal geological background field is used to capture the enrichment intensity of ore-forming elements. The geophysical absolute difference is processed by field type, and the absolute difference is obtained by magnetic method, while the sign is preserved in gravity field, which is used to locate mineralization-related physical property anomalies. Furthermore, independent calculation threads are started for cross-fault voxels to avoid the propagation of structural boundary errors. After identifying real voxels with differences, morphological opening operation is used to optimize the difference voxel space formed by the real voxels with differences. The difference voxel space is scanned using spherical structure elements, and all isolated anomalies not completely covered by the structure elements (such as <8 points) are deleted. The analysis uses a three-dimensional region growing algorithm to optimize the connectivity of each differential voxel space, making its boundaries smoother. Finally, through geological rule constraints (including but not limited to deleting areas that overlap with known non-mineral interference sources by more than 70%, shielding geochemical false anomalies distributed along modern rivers, and retaining only anomaly areas within 2km of ore-controlling structures), an effective anomaly area is generated as the first differential distribution data.

[0089] In one embodiment, step S3, which involves extracting geological features of a known mineralized area and generating interference factors based on those features, includes:

[0090] S31. Construct a geological knowledge graph of the mining area based on the known geological data and background data of the existing mining area. The geological knowledge graph of the mining area includes element nodes and element association edges.

[0091] S32. Extracting geological features from geological data based on feature extraction algorithms;

[0092] S33. Vectorize the element nodes and geological features and project them onto the matching space;

[0093] S34. In the matching space, the matching element node with the highest matching degree is obtained based on the dual-tower attention matching model.

[0094] S35. Construct multiple interference factors based on matching feature nodes. The interference factors include a voxel range constraint, an attribute assigner, and an attribute modifier.

[0095] As described in steps S31-S35 above, a geological knowledge graph of the mining area is constructed based on existing geological and background data. This graph includes element nodes representing geological elements such as ore bodies, alteration zones, and ore-controlling structures, as well as element association edges describing their genesis, spatial relationships, and attribute connections. Multi-scale geological features are extracted from the exploration data using feature extraction algorithms. These algorithms include processing geophysical raster data using convolutional neural networks and resolving topological relationships using graph neural networks. The semantics of element nodes are vectorized using word embedding technology, and element nodes and element association edges are transformed into relation vectors using graph embedding technology. Simultaneously, an autoencoder is used to compress geological features into feature vectors, which are then projected onto a unified matching space. A dual-tower attention matching model is deployed in this space. —The left tower inputs the feature node vector, and the right tower inputs the geological feature vector. The interaction weight between features and nodes is calculated through a multi-layer attention mechanism, and the feature node with the highest matching degree is finally output (such as the skarn-type copper mineralization system node). Based on the matching nodes, interference factors are constructed: the action voxel range constraint generates an ellipsoidal action domain according to the spatial pattern of node association (such as mineralization zoning radius), the attribute allocator parses the node attribute modification rules (such as reducing the resistivity of marble contact zone voxels to 50-100Ω·m) to generate an operation instruction sequence carrying parameters, and the attribute modifier encapsulates the atomic operation functions of the execution instructions (such as gradual interpolation of physical property values ​​and replacement of lithological codes). The three work together to form a structured interference factor that can be implanted into the geological background field.

[0096] In one embodiment, step S4, which introduces the interference factor into the ideal geological background field to generate the interference coupled background field, includes:

[0097] S41. The action voxel range constraint converts the action range defined by multiple interference factors into voxel index coordinates of a three-dimensional ideal geological background field to obtain the action target voxel.

[0098] S42. The attribute allocator generates target interference factors by analyzing the lithological coding, structural location, and physical property thresholds of the target voxels and injecting corresponding attributes into the corresponding interference factors.

[0099] S43. Deploy the target interference factor into the target voxel, and release the corresponding attribute stored in the target interference factor based on the attribute modifier;

[0100] S44. Modify the attributes of the target voxel based on the corresponding attributes to obtain the updated voxel;

[0101] S45. Examine the attribute deviation between the updated voxel and its neighboring voxels, and define two voxels whose attribute deviation exceeds a preset safety threshold as conflicting voxels.

[0102] S46. Based on the conflict resolution mechanism, modify the properties of the conflict voxels until all target voxels have been updated to generate the interference coupling background field.

[0103] As described in steps S41-S46 above, the fusion of interference factors and the ideal geological background field is achieved through a five-order coupling process. First, the voxel range constraint converts the three-dimensional domain of the interference factors (such as fault buffer zones and mineralization halos) into voxel index coordinates. Further, if multiple interference factors have overlapping domains, a distance-weighted algorithm can be used to generate a composite domain. The distance from the center voxel of the domain to the overlapping area is calculated, and the modification weight of the center voxel on the composite domain is calculated based on the distance. Then, the attribute allocator assigns interference factors based on the lithological coding, structural location, and physical property thresholds of the target voxel. The corresponding attribute's steps are as follows: The attribute allocator sets up an interference factor attribute library. For the address data of the target voxel, the attribute allocator dynamically extracts matching alteration rules, element enrichment equations, and property adjustment amounts from the interference factor attribute library to generate a target interference factor carrying geological process parameters. Subsequently, the attribute modifier injects the target interference factor into the corresponding voxel: performing lithological alteration transformation (e.g., preserving the original rock structure and superimposing 15% pyrite), reconstructing the geochemical field according to the equation, and gradually adjusting the property parameters. After the target interference factor modifies the target voxel, a 26-neighborhood voxel verification unit is established to detect and update the voxel and its surroundings. The properties of voxels are abruptly altered, and the dimensions for testing include, but are not limited to, lithological compatibility, physical property gradient, and elemental zoning. Testing indicators include, but are not limited to, coding difference, density change rate, and elemental ratio gradient. If a testing indicator exceeds a preset threshold, the voxel is identified as a conflicting voxel. The properties of the conflicting voxels are then revised through a conflict resolution mechanism. This mechanism includes: for lithological abrupt changes, inserting transitional lithological voxels into the contact zone (e.g., generating a chlorite zone between andesite and sericite); for physical property jumps, initiating a thermal diffusion model to smooth the transition, triggering smoothing filtering to recalculate the physical property distribution; and for elemental inversion zoning, iteratively correcting the migration equation and adjusting the activity. The method generates a geologically reasonable interference coupling background field by resolving all conflicts and using parameters. Furthermore, Gaussian smoothing can be applied to the boundaries of updated voxels in the interference coupling background field. Based on the above analysis, this method achieves precise spatial-attribute-genetic triple coupling: spatial constraint transformation of the acting voxels ensures positioning accuracy; dynamic attribute injection mechanism based on lithology-structure-physical property linkage achieves geological process adaptation; and a multi-parameter neighborhood verification and genetically driven conflict resolution mechanism ensures the three-dimensional geological rationality of the updated voxels and the background field. Finally, a quantifiable and verifiable interference coupling background field is generated, providing a high-fidelity simulation basis for mineral resource prediction.

[0104] In one embodiment, step S4, which extracts the difference between the interference-coupled background field and the ideal geological background field as the second difference distribution data, includes:

[0105] S47. Divide the interference coupling background field into multiple interference coupling voxels;

[0106] S48. Perform voxel difference calculations on the corresponding ideal voxels and interference coupled voxels. Voxel differences include lithological classification differences, geochemical relative deviations, and geophysical absolute differences.

[0107] S49. Based on morphological opening operation and connected component analysis, interfering coupled voxels with differences are connected into effective anomaly regions as the second differential distribution data.

[0108] As described in steps S47-S49 above, the interfering coupled background field is divided into interfering coupled voxels that are perfectly aligned with the spatial index of the three-dimensional ideal geological background field using a voxel segmenter; difference analysis is performed on the registered voxels using the same parallel architecture as the first difference calculation: lithological classification differences are based on the comparison between the modified lithological code and the ideal code; geochemical relative deviation is calculated using the (coupled field value - ideal value) / ideal background standard deviation model; geophysical absolute differences are processed separately for magnetic / gravity / resistivity fields (e.g., absolute difference is taken for magnetic methods, and positive difference in gravity fields indicates high-density anomalies); subsequently... The process of anomaly optimization guided by geological laws is initiated. First, a spherical structural element is set with the minimum volume of the mineralized domain as the diameter to perform an opening operation (after erosion to eliminate isolated noise points, directional expansion is used to repair the anomaly boundary). Then, intelligent aggregation of anomaly regions is performed based on three-dimensional connected domain analysis. A region growth algorithm constrained by the stratigraphic dip is used to force the merging of neighboring anomalies with a spacing of less than twice the voxel size. At the same time, a special filter for simulated anomalies is activated to remove anomaly clumps with a spatial matching degree of less than 60% with known mineralization patterns. The identification of alteration zoning ring structures is strengthened. Finally, the effective anomaly area is output as the second differential distribution data.

[0109] In one embodiment, step S5, which involves constructing a mineral resource prediction model based on the first differential distribution data and the second differential distribution data, includes:

[0110] S51. Perform voxel-level spatial coupling analysis on the first and second differential distribution data to calculate the overlap rate and difference intensity of the overlapping anomaly region.

[0111] S52. Based on the overlap rate and the intensity of the difference, the overlapping abnormal area is divided into three levels, A, B and C, according to a preset threshold.

[0112] S53. A three-dimensional mineralization probability field is generated by weighted fusion of the overlapping anomaly areas of the first and second differential distribution data;

[0113] S54. Convert the three-dimensional mineralization probability field into the voxel index coordinates of the three-dimensional ideal geological background field, and generate a mineral resource prediction model based on the voxel index coordinates mapped to the three-dimensional ideal geological background field.

[0114] S55. The mineral resource prediction model includes mineral distribution, mineral type identification, mineral probability value, and confidence level.

[0115] As described in steps S51-S55 above, voxel-level spatial coupling analysis is performed on the first and second difference distribution data. Overlapping anomaly areas are identified through three-dimensional spatial Boolean operations, and their overlap rate is calculated as: number of overlapping voxels / total number of anomaly voxels × 100%. The difference intensity is calculated by taking the geometric mean of lithological, geophysical, and geochemical differences. Based on a preset threshold, the overlapping areas are divided into three levels: Level A (e.g., overlap rate ≥ 70% and difference intensity > 85 percentile), Level B (e.g., overlap rate 40%-70% and intensity > 60 percentile), and Level C (e.g., overlap rate < 40% or insufficient intensity). A three-dimensional mineralization probability field is generated through dynamic weighted fusion: Level A areas are assigned a 70% weight to the first difference data (strengthening the contribution of real anomalies), while Level B / C areas are assigned a 5% weight. The model employs a 0% / 30% weighting (emphasizing simulation of anomalies) and applies probability decay to non-overlapping areas (the probability value decreases by 20% for every 100m increase in distance from the boundary). It rigorously matches the probability field voxel index coordinates with the ideal 3D geological background field, using bilinear interpolation to map them to the background field grid nodes. The final output mineral resource prediction model includes: 1. Mineral distribution triangular network entities (voxels with probability > 0.6 are aggregated into closed surfaces), 2. Mineral type identifiers (automatically labeled based on the knowledge graph nodes of the matching interference factors), 3. Probability distribution (0-1 continuous distribution), and 4. Confidence level (confidence level of Grade A area = 0.9 × data coverage coefficient, Grade C area = 0.5 × simulation matching degree). Furthermore, the mineral resource prediction model supports hierarchical visualization based on probability thresholds and interface calls for reserve estimation.

[0116] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing a mineral resource prediction model based on complex geological backgrounds, characterized in that, Based on the geological background of the unknown mining area, a three-dimensional ideal geological background field without mineralization interference is constructed. The three-dimensional ideal geological background field is used to display the mining area in the non-mineralized area. A three-dimensional realistic geological field is constructed based on exploration data, and the first difference distribution data is obtained by comparing the three-dimensional realistic geological field with the three-dimensional ideal geological background field. Voxel registration is performed between the real geological background field and the three-dimensional ideal geological background field; Voxel differences are calculated between the corresponding ideal voxels and the actual voxels. Voxel differences include differences in lithological classification, relative deviations in geochemistry, and absolute differences in geophysics. Based on morphological opening operations and connected component analysis, real voxels with differences are connected to form effective outlier regions as the first differential distribution data; Extract geological features from known mineralized areas and generate interference factors based on these features; The interference factor is introduced into the ideal geological background field to generate the interference coupled background field, and the difference between the interference coupled background field and the ideal geological background field is extracted as the second difference distribution data. The interference coupling background field is segmented into multiple interference coupling voxels; Voxel differences are calculated for the corresponding ideal voxels and interference coupled voxels. Voxel differences include lithological classification differences, geochemical relative deviations, and geophysical absolute differences. Based on morphological opening operations and connected component analysis, interfering coupled voxels with differences are connected to form effective outlier regions as the second differential distribution data; A mineral resource prediction model is constructed based on the first and second differential distribution data. Voxel-level spatial coupling analysis was performed on the first and second differential distribution data to calculate the overlap rate and difference intensity of the overlapping anomaly region. Based on the overlap rate and the intensity of the difference, the overlapping abnormal areas are divided into three levels, A, B, and C, according to a preset threshold. A three-dimensional mineralization probability field is generated by weighted fusion of the overlapping anomaly regions of the first and second differential distribution data; The three-dimensional mineralization probability field is converted into voxel index coordinates of a three-dimensional ideal geological background field, and a mineral resource prediction model is generated based on the voxel index coordinates mapped to the three-dimensional ideal geological background field. Mineral resource prediction models include mineral distribution, mineral type identification, mineral probability values, and confidence levels.

2. The method for constructing a mineral resource prediction model based on a complex geological background according to claim 1, characterized in that, The steps for constructing a three-dimensional ideal geological background field free from mineralization interference are as follows: Collect geological background data of unknown mining areas and construct a three-dimensional ideal geological background field based on the ideal background field generation model; The ideal background field generation model includes a lithology distribution simulator, a structural framework generator, a physical property field calculator, and a voxel segmenter; The lithology distribution simulator is based on the Markov random field algorithm and uses geological background data to constrain the lithology transfer probability. The lattice generator is based on B-spline surface fitting stress field simulation to generate a fracture network that conforms to mechanical laws. The physical property field calculator calculates the density, magnetic susceptibility, and resistivity fields of unknown mineral areas based on empirical rock physics formulas and potential field forward modeling. The voxel segmenter divides the three-dimensional ideal geological background field into multiple ideal voxels.

3. The method for constructing a mineral resource prediction model based on a complex geological background according to claim 1, characterized in that, The steps for constructing a three-dimensional realistic geological field based on exploration data are as follows: Acquire exploration data from unknown mining areas, and use the exploration data to generate a three-dimensional realistic geological field based on a real background field model; The real-world context generation model includes a data converter, an anomaly converter, and a voxel segmenter. The data converter, based on the Kriging interpolation algorithm, transforms discrete exploration data into three-dimensional raster data with the same resolution as the ideal three-dimensional geological background field. The anomaly converter is based on the extraction of valid anomalies that conform to geological patterns through statistical threshold analysis and morphological processing. The voxel segmenter divides the 3D real geological field into multiple real voxels.

4. The method for constructing a mineral resource prediction model based on a complex geological background according to claim 1, characterized in that, The step of extracting geological features of known mineralized areas and generating interference factors based on these geological features is as follows: A geological knowledge graph of a mining area is constructed based on the known geological data and background data of the existing mining area. The geological knowledge graph of a mining area includes element nodes and element association edges. Geological features are extracted from geological data based on feature extraction algorithms; Vectorize the element nodes and geological features and project them onto the matching space; In the matching space, the matching element node with the highest matching degree is obtained based on the dual-tower attention matching model; Multiple interference factors are constructed based on the matching feature nodes. The interference factors include the voxel range constraint, the attribute assigner, and the attribute modifier.

5. The method for constructing a mineral resource prediction model based on a complex geological background according to claim 1, characterized in that, The step of introducing the interference factor into the ideal geological background field to generate the interference coupled background field is as follows: The action voxel range constraint converts the action range defined by multiple interference factors into voxel index coordinates of a three-dimensional ideal geological background field to obtain the action target voxel; The attribute allocator generates target interference factors by analyzing the lithological code, tectonic location, and physical property threshold of the target voxel and injecting corresponding attributes into the corresponding interference factors. The target interference factor is deployed into the target voxel, and the corresponding attribute stored in the target interference factor is released based on the attribute modifier. The updated voxel is obtained by modifying the attributes of the target voxel based on the corresponding attributes; Examine the attribute deviation between the updated voxel and its neighboring voxels, and define two voxels whose attribute deviation exceeds a preset safety threshold as conflicting voxels; The properties of conflict voxels are modified based on the conflict resolution mechanism until all target voxels have been updated, generating an interference coupling background field.

Images