Method and system for predicting spatial distribution of mineralization domain based on geological prior knowledge

Abstract

The application provides a mineralization domain spatial distribution prediction method and system based on geological prior knowledge, and relates to the technical field of mineral exploration, wherein the method comprises the following steps: obtaining drilling core assay analysis data and magnetic survey data of a target mining area; then processing geochemical data to obtain an element concentration data set, and performing a potential field conversion on the magnetic survey data to generate a magnetization direction data set; then using a fractal algorithm to identify geochemical anomaly information from the concentration data set and determine an anomaly range, fusing the anomaly range and the magnetization direction data set, and then performing three-dimensional visualization modeling to construct a three-dimensional geological model; finally, extracting feature data in the three-dimensional geological model and inputting the feature data into a machine learning prediction model for processing to obtain a distribution map representing the existence probability of the mineralization domain in the space of the mining area. The application improves the accuracy of predicting the spatial position of a concealed mineralization domain.

Description

Technical Field

[0001] This application relates to the field of mineral exploration technology, and in particular to a method and system for predicting the spatial distribution of mineralized domains based on prior geological knowledge. Background Technology

[0002] In the field of mineral exploration, accurately predicting the spatial distribution of concealed mineralization zones is of direct significance for improving mineral exploration efficiency and reducing exploration risks, and is one of the key technologies for locating deep resources.

[0003] In existing technologies, a common method is to analyze the geochemical data of borehole samples and delineate anomalous areas based on empirical thresholds to infer the extent of mineralization. This method is effective for shallow or simple mineralized bodies. Another method combines geophysical exploration techniques, such as processing magnetic survey data to indirectly reflect the morphology of underground rock masses and assist in identifying favorable mineralization areas, thus enhancing the diversity of prediction data.

[0004] However, the existing methods mentioned above face certain limitations in the prediction process: on the one hand, geochemical data processing usually relies on empirical thresholds, which can easily lead to misjudgments or omissions in deep environments with complex background values ​​or weak anomalies; on the other hand, the lithological information obtained from geophysical inversion and geochemical anomalies are mostly qualitative or loosely correlated, and the two have not formed a quantitative fusion in three-dimensional space, resulting in uncertainty in the spatial location of mineralization centers. Therefore, existing technologies have limited accuracy in predicting deep concealed mineralization domains. Summary of the Invention

[0005] This application provides a method and system for predicting the spatial distribution of mineralized domains based on prior geological knowledge, in order to solve the problem of low accuracy in predicting the spatial location of concealed mineralized domains in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for predicting the spatial distribution of mineralization domains based on prior geological knowledge, comprising:

[0007] Based on prior geological knowledge, core analysis data and magnetic measurement data from boreholes in the target mining area are obtained;

[0008] The core analysis data were analyzed to obtain elemental concentration data, and the elemental concentration data were preprocessed to obtain the first dataset.

[0009] The magnetic measurement data is subjected to potential field transformation to obtain transformation parameters, and the magnetization direction is calculated based on the transformation parameters. A second dataset is then generated based on the magnetization direction.

[0010] Based on the first dataset, a fractal algorithm is used to identify geochemical anomaly information from the first dataset, and the anomaly range is determined based on the geochemical anomaly information;

[0011] Based on the anomaly range and the second dataset, fused data is generated, and three-dimensional visualization modeling is performed on the fused data to construct a three-dimensional geological model;

[0012] The three-dimensional feature data in the three-dimensional geological model is input into the machine learning prediction model. The machine learning prediction model processes the three-dimensional feature data to obtain a probability distribution map characterizing whether there is a mineralized domain in the target mining area.

[0013] Optionally, the step of inputting the three-dimensional feature data from the three-dimensional geological model into a machine learning prediction model, and processing the three-dimensional feature data through the machine learning prediction model to obtain a probability distribution map characterizing the existence of mineralized domains within the target mining area, includes:

[0014] Extract three-dimensional feature data from the three-dimensional geological model;

[0015] The three-dimensional feature data is input into the feature encoding module of the machine learning prediction model. Through the multi-layer convolution operation of the feature encoding module, spatial feature extraction and dimensionality reduction are performed on the three-dimensional feature data to generate a geological feature map. The geological feature map is used to reflect the local structural features of geochemical and geophysical data in three-dimensional space.

[0016] The geological feature map is input into the context modeling module of the machine learning prediction model. The correlation weights between features at different spatial locations in the geological feature map are calculated through the three-dimensional attention mechanism of the context modeling module.

[0017] The context modeling module uses the correlation weights to perform weighted fusion of the geological feature map to generate a fused feature map, which is used to reflect the cooperative distribution pattern between geological anomalies and magnetic anomalies in three-dimensional space.

[0018] The fused feature map is input into the task interpretation module of the machine learning prediction model. The fused feature map is processed through the fully connected layer of the task interpretation module to generate a probability distribution map.

[0019] Optionally, the step of processing the fused feature map through the fully connected layer of the task interpretation module to generate a probability distribution map includes:

[0020] The fused feature map is transformed into a one-dimensional feature vector;

[0021] The one-dimensional feature vector is input into the first fully connected layer, and the one-dimensional feature vector is mapped to a high-dimensional space through the first fully connected layer to obtain high-dimensional features. The high-dimensional features are then nonlinearly activated to generate activated geological features. The high-dimensional features are used to reflect the multi-scale geological correlation of the fused feature map.

[0022] The activated geological features are input into the second fully connected layer, and the activated geological features are dimensionality-reduced and mapped through the second fully connected layer to obtain compressed features. The compressed features are used to reflect the synergistic distribution characteristics of geochemical anomalies and magnetic anomalies.

[0023] The compressed features are input into the third fully connected layer, and the compressed features are mapped to the target dimension through the third fully connected layer to generate preliminary prediction results;

[0024] The preliminary prediction results are normalized to generate a probability distribution map. Each value in the probability distribution map corresponds to a location in three-dimensional space and is used to characterize the probability that a mineralization domain exists at the corresponding location.

[0025] Optionally, the step of identifying geochemical anomaly information from the first dataset using a fractal algorithm, and determining the anomaly range based on the geochemical anomaly information, includes:

[0026] From the first dataset, identify regions where the element concentration value is greater than a preset concentration threshold, and calculate the geometric features of the regions.

[0027] Fractal algorithms are used to analyze the correlation between the geometric features and spatial scale, and to determine the fractal relationship;

[0028] Based on the fractal relationship, a target concentration threshold is selected from the concentration threshold set as an anomaly detection threshold;

[0029] The first dataset is processed based on the anomaly discrimination threshold to obtain geochemical anomaly information;

[0030] Three-dimensional connectivity analysis is performed on the geochemical anomaly information to identify and label the independent spatial connected components of the geochemical anomaly information in three-dimensional space;

[0031] The range of anomalies is determined based on the spatial distribution pattern of the independent spatial interconnections.

[0032] Secondly, this application provides a spatial distribution prediction system for mineralization domains based on prior geological knowledge, including:

[0033] The acquisition module is used to acquire core analysis data and magnetic measurement data from boreholes in the target mining area based on prior geological knowledge.

[0034] The analysis module is used to analyze the core analysis data, obtain elemental concentration data, and preprocess the elemental concentration data to obtain a first dataset.

[0035] The conversion module is used to perform potential field conversion on the magnetic measurement data to obtain conversion parameters, calculate the magnetization direction based on the conversion parameters, and generate a second dataset according to the magnetization direction;

[0036] The identification module is used to identify geochemical anomaly information from the first dataset using a fractal algorithm, and to determine the anomaly range based on the geochemical anomaly information.

[0037] The generation module is used to generate fused data based on the anomaly range and the second dataset, and to perform three-dimensional visualization modeling on the fused data to construct a three-dimensional geological model;

[0038] The input module is used to input the three-dimensional feature data in the three-dimensional geological model into the machine learning prediction model, and to process the three-dimensional feature data through the machine learning prediction model to obtain a probability distribution map characterizing whether there is a mineralized domain in the target mining area.

[0039] Thirdly, this application provides an electronic device, comprising:

[0040] Memory, used to store computer programs;

[0041] A processor, configured to execute the computer program to implement the steps of the method for predicting the spatial distribution of mineralization domains based on prior geological knowledge as described in the first aspect above.

[0042] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the mineralization domain spatial distribution prediction method based on geological prior knowledge as described in the first aspect above.

[0043] The technical solution provided in this application has the following beneficial effects:

[0044] This application acquires and analyzes geochemical and magnetic data to generate quantitative datasets reflecting elemental distribution and rock magnetization characteristics, respectively. Secondly, by applying fractal algorithms to process the geochemical data, it can more accurately identify geochemical anomalies related to mineralization and effectively distinguish background interference. Then, it integrates the anomaly range and magnetization direction data and performs three-dimensional visualization modeling to construct a three-dimensional geological model that simultaneously reflects geochemical and geophysical characteristics. Finally, the three-dimensional feature data from this model is input into a machine learning prediction model for processing, thereby directly generating a prediction map characterizing the spatial distribution probability of mineralization domains.

[0045] Furthermore, this application extracts feature data from the three-dimensional geological model, uses a feature encoding module to perform spatial feature extraction and dimensionality reduction, and generates a geological feature map that reflects the local structural features of the data. Then, the geological feature map is input into the context modeling module, and the correlation weights between features at different spatial locations are calculated through a three-dimensional attention mechanism. The weights are then used for weighted fusion to generate a fused feature map that reflects the synergistic distribution law of geological and magnetic anomalies. Finally, the fused feature map is processed by the fully connected layer of the task interpretation module to generate a probability distribution map.

[0046] Furthermore, by extracting local features in stages, modeling spatial context associations, and performing final interpretation, it is possible to fully explore and integrate deep correlation information in multi-source 3D geological data, thereby improving the accuracy and reliability of identifying spatial distribution patterns of mineralization domains.

[0047] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 A flowchart illustrating a method for predicting the spatial distribution of mineralization domains based on prior geological knowledge, provided for embodiments of this application;

[0050] Figure 2 A schematic diagram illustrating a specific implementation of a method for predicting the spatial distribution of mineralization domains based on prior geological knowledge, provided in this application embodiment;

[0051] Figure 3 This is a schematic diagram of the structure of a mineralization domain spatial distribution prediction system based on geological prior knowledge, provided in an embodiment of this application. Detailed Implementation

[0052] To address the problems of existing technologies, this application proposes a method for predicting the spatial distribution of mineralization domains based on prior geological knowledge. The core of this method lies in the phased quantitative analysis and deep synergistic fusion of geochemical and geophysical data. First, a fractal algorithm is applied to geochemical data for anomaly identification to distinguish mineralization-related anomalies from background field interference. Then, the identified geochemical anomaly range is fused with magnetization direction data obtained from magnetic survey data inversion in three-dimensional space to construct a unified geological model. Finally, a machine learning model is used to process the three-dimensional feature data in this model to generate a probability map representing the spatial distribution of mineralization domains. This method abandons the traditional prediction model that relies on empirical thresholds and qualitative correlations. It enhances the objectivity of anomaly identification through fractal algorithms, achieves spatial integration of multi-source data through three-dimensional modeling, and uses machine learning models to uncover deep distribution patterns. This achieves quantitative fusion and synergistic interpretation of geochemical and geophysical evidence at the data level, effectively overcoming the limited accuracy of predictions for deep, concealed mineralization domains in existing technologies and improving the reliability of prediction results.

[0053] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] The core of this application is to provide a method for predicting the spatial distribution of mineralization domains based on prior geological knowledge. A flowchart of one specific implementation method is shown below. Figure 1 As shown, the method includes:

[0055] Step 101: Based on prior geological knowledge, obtain core analysis data and magnetic measurement data from boreholes in the target mining area.

[0056] In step 101, geological prior knowledge refers to the understanding that mineralized elements obtained in a known mining area exhibit vertical zonation, which guides the targeted collection of subsequent data; core analysis data refers to the data obtained by laboratory chemical analysis of core samples obtained from boreholes, which includes information on the content of various ore-forming elements and associated elements; magnetic data is physical field data reflecting the magnetic characteristics of underground rocks obtained by deploying measuring equipment in the borehole.

[0057] In this embodiment of the application, based on the prior geological knowledge of the known vertical zonation of mineralized elements, it is determined that two types of data need to be collected in specific boreholes in the target mining area. One type is rock samples used to analyze elemental content, and core analysis data is obtained through laboratory analysis of these samples. The other type is magnetic data directly collected by arranging measuring devices in the same borehole.

[0058] Step 102: Analyze the core analysis data to obtain elemental concentration data, and preprocess the elemental concentration data to obtain the first dataset.

[0059] Among them, the element concentration data are specific values ​​obtained by analyzing core test data, representing the content of specific ore-forming elements and associated elements; the first dataset is a set of data generated after standardization and three-dimensional spatial interpolation, used to characterize the continuous distribution of multiple ore-forming elements and associated elements in three-dimensional space within the target mining area. Each data point in this dataset corresponds to a three-dimensional spatial location and contains the standardized concentration values ​​of one or more elements.

[0060] The process of analyzing the core analysis data to obtain element concentration data is as follows: First, core analysis data containing information on the content of multiple elements is received from the laboratory; then, the content values ​​of each ore-forming element and associated element are extracted from the core analysis data, and these values ​​are organized into a list indexed by element type and sample location. This list is the element concentration data.

[0061] In this embodiment, step 102 includes the following process:

[0062] Step 1021: Standardize the element concentration data to obtain standardized concentration data.

[0063] In step 1021, the standardized concentration data is the elemental concentration data obtained after standardization. This data is obtained by subtracting the average value from the original concentration value of each element and then dividing it by the standard deviation, thereby eliminating the influence caused by the difference in dimensions and absolute values ​​between different elements, so that the concentration values ​​of different elements can be compared and analyzed on the same scale.

[0064] In this embodiment of the application, the element concentration data is standardized by performing standardization processing, that is, calculating the mean and standard deviation of the concentration of each element, and then subtracting the mean of the corresponding element from each original concentration value and dividing by the standard deviation of the element, thereby obtaining standardized concentration data, which makes the concentration values ​​of different elements comparable.

[0065] Step 1022: Place the standardized concentration data in a three-dimensional spatial grid, and for locations in the three-dimensional spatial grid without data, interpolate based on the neighboring standardized concentration data to generate three-dimensional concentration field data, and use the three-dimensional concentration field data as the first dataset.

[0066] In step 1022, the three-dimensional spatial grid is a grid system composed of regular units divided in the three-dimensional space of the target mining area, and each grid unit is called a volume element.

[0067] In this embodiment, a three-dimensional spatial grid covering the target mining area is first established. Then, the existing standardized concentration data is assigned to the corresponding voxel positions in the three-dimensional spatial grid according to their corresponding actual spatial coordinates. Next, for voxel positions in the three-dimensional spatial grid that do not have directly measured data, a spatial interpolation algorithm, such as Kriging interpolation, is used. This algorithm estimates the concentration values ​​of these blank voxels based on the standardized concentration data at nearby known voxel positions, taking into account spatial distance and correlation. After estimating and filling all blank voxels, a complete and continuous three-dimensional concentration field data is generated, which serves as the first dataset for subsequent steps.

[0068] Through the above steps, this application transforms discrete point element concentration data into unified and comparable three-dimensional continuous field data, providing a standardized and spatially continuous data foundation for subsequent anomaly identification and analysis.

[0069] Step 103: Perform potential field transformation on the magnetic measurement data to obtain transformation parameters, calculate the magnetization direction based on the transformation parameters, and generate a second dataset according to the magnetization direction.

[0070] Among them, the transformation parameter is a mathematical factor that characterizes the spatial positional relationship between the observation point and the underground magnetic source in the frequency domain, the magnetization direction refers to the spatial orientation of the total magnetization intensity of the underground rock unit, and the second dataset is a set of data representing the spatial distribution of magnetization direction in each local area in three-dimensional space.

[0071] In this embodiment, step 103 includes the following process:

[0072] Step 1031: Perform frequency domain conversion on the magnetic measurement data, and calculate the conversion parameters based on the converted magnetic measurement data and the positional relationship between the observation location and the magnetic source.

[0073] In step 1031, the positional relationship between the observation location and the magnetic source refers to the relative geometric orientation and distance relationship between the spatial coordinates of each measurement point in the borehole where magnetic data is collected and the spatial range of the possible underground magnetic geological body. This relationship is used to establish a physical and mathematical model between the observation data and the underground source body in the frequency domain potential field conversion.

[0074] In this embodiment, the magnetic measurement data is first subjected to a Fourier transform to convert it from the spatial domain to the frequency domain, resulting in frequency domain magnetic measurement data. Then, in the frequency domain, the conversion parameter is derived based on the mathematical relationship in the potential field theory, according to the spatial relationship between the coordinates of the observation point and the assumed magnetic source body. This conversion parameter is essentially a filter factor related to the spatial frequency.

[0075] Step 1032: Calculate the converted magnetic measurement data using the conversion parameters, and inversely estimate the magnetization direction of multiple local regions in three-dimensional space.

[0076] In step 1032, a local region refers to a volume unit formed by dividing the three-dimensional space into regular or irregular grids, or a spatial block artificially set during the inversion calculation process to represent the magnetization state of a specific underground area based on the geological background or data processing requirements. The division can be based on the volume element boundary of the three-dimensional spatial grid or on the spatial resolution set when discretizing the continuous space.

[0077] In this embodiment of the application, the conversion parameters calculated in step 1031 are used to perform filtering calculations on the frequency domain magnetic measurement data. This process is completed in the frequency domain. Then, an inverse Fourier transform is performed on the filtered result to convert it from the frequency domain back to the spatial domain, thereby obtaining the magnetization direction estimates of a series of discrete local regions in three-dimensional space.

[0078] Step 1033: Place the magnetization direction of each local region in a three-dimensional spatial grid to generate the second dataset.

[0079] In this embodiment of the application, the magnetization direction of each local region estimated by inversion in step 1032 is assigned to the corresponding voxel position in the three-dimensional spatial grid according to its corresponding actual spatial coordinates; for voxels in the grid that do not have direct inversion results, a spatial interpolation method is used to fill in the magnetization direction based on the existing magnetization direction of the neighboring voxels, and finally a complete second dataset is generated in which each voxel contains magnetization direction information.

[0080] Through the above steps, this application transforms the original magnetic measurement data into three-dimensional directional field data that can quantitatively and continuously reflect the spatial distribution characteristics of magnetic properties in deep underground rock masses, providing a key physical field basis for subsequent multi-source information fusion.

[0081] Step 104: Based on the first dataset, use a fractal algorithm to identify geochemical anomaly information from the first dataset, and determine the anomaly range based on the geochemical anomaly information.

[0082] Geochemical anomaly information refers to regional information where elemental concentrations are higher than background values ​​in space, potentially indicating mineralization. The anomaly range is a spatial region delineated in three-dimensional space based on geochemical anomaly information, which is considered to be related to mineralization.

[0083] In this embodiment, step 104 includes the following process:

[0084] Step 1041: Determine the regions in the first dataset where the element concentration value is greater than the preset concentration threshold, and calculate the geometric features of the regions.

[0085] In step 1041, the preset concentration threshold set is a set of multiple candidate concentration thresholds used to initially screen possible abnormal regions; geometric features refer to the morphological and structural attributes of these initially screened regions in three-dimensional space, such as the volume or surface area of ​​the region at a specific concentration threshold.

[0086] In this embodiment, a preset concentration threshold set is first set, which includes multiple candidate concentration thresholds. For example, the set includes concentration values ​​of 1.5, 2.0, and 2.5. Then, for each concentration threshold in the preset concentration threshold set, all voxels with concentration values ​​greater than the threshold are found in the first dataset. The set of these voxels constitutes a candidate abnormal region. Next, the geometric features of each candidate abnormal region are calculated. For example, the total number of voxels contained in the region is calculated as its volume feature, or the outer surface area of ​​the region is calculated as its morphological feature.

[0087] In practical applications, for target mining area A, the first dataset is three-dimensional concentration field data, with a preset concentration threshold set of [1.2, 1.5, 1.8, 2.0]. First, select a concentration threshold of 1.2, find all voxels in the first dataset with a concentration value greater than 1.2, assuming there are 8500 voxels in total, calculate the volume feature of this area as 8500 voxels, and estimate its surface area feature by calculating the number of voxel faces contained on the outer surface of this area. Similarly, select concentration thresholds of 1.5, 1.8, and 2.0 respectively, and repeat the above process to obtain the corresponding area and its geometric feature data under different thresholds.

[0088] Step 1042: Use a fractal algorithm to analyze the relationship between the geometric features and the spatial scale, and determine the fractal relationship.

[0089] In step 1042, fractal relationship refers to the power-law relationship of geometric features as the concentration threshold changes, which is usually expressed as a linear relationship in logarithmic coordinates.

[0090] In this embodiment of the application, the result obtained in step 1041 is processed using a fractal algorithm. The specific process is as follows: the regional geometric features, such as volume, calculated for each concentration threshold in step 1041 are associated with the corresponding concentration threshold; then, in a double logarithmic coordinate system, data points are plotted with the logarithm of the concentration threshold as the abscissa and the logarithm of the regional geometric features as the ordinate; next, linear fitting is performed on these data points, and the slope of the fitted line is the fractal dimension. This fractal dimension and the fitted line together characterize the fractal relationship of the element concentration distribution.

[0091] In practical applications, continuing the previous example, for target mining area A, four concentration thresholds [1.2, 1.5, 1.8, 2.0] and their corresponding regional volumes [8500, 5200, 3000, 1500] individual elements have been obtained. In a double logarithmic coordinate system, the horizontal axis is the natural logarithm of the concentration thresholds, and the vertical axis is the natural logarithm of the volume, resulting in four data points [ln(1.2), ln(8500)], [ln(1.5), ln( [5200)], [ln(1.8), ln(3000)], [ln(2.0), ln(1500)]; Perform least squares linear fitting on these data points to obtain a fitted line with a slope of -2.1. This slope is an estimate of the fractal dimension D. The equation of the fitted line is ln(V) = -2.1 × ln(C) + b, where V is the volume, C is the concentration threshold, and b is the intercept, which is the determined fractal relationship.

[0092] Step 1043: Based on the fractal relationship, select the target concentration threshold from the concentration threshold set as the anomaly discrimination threshold.

[0093] In this embodiment of the application, based on the fractal relationship determined in step 1042, the law of change of geometric features with concentration threshold is analyzed. Generally, on the fractal relationship curve, when the concentration threshold reaches a certain critical value, the trend of change of geometric features will deviate significantly. This critical concentration threshold is selected as the target concentration threshold, which is used as the final anomaly discrimination threshold to distinguish between geochemical background and anomaly.

[0094] In practical applications, continuing with the previous example, observing the fitted line and data points, it is found that when the concentration threshold increases from 1.8 to 2.0, the actual volume data points [ln(2.0), ln(1500)] deviate significantly from the fitted line determined by the first three points [ln(1.2), ln(8500)], [ln(1.5), ln(5200)], [ln(1.8), ln(3000)]. This indicates that the concentration threshold of 1.8 is a turning point in the trend. Therefore, the concentration threshold of 1.8 is selected as the target concentration threshold, i.e., the anomaly detection threshold.

[0095] Step 1044: Process the first dataset based on the anomaly discrimination threshold to obtain geochemical anomaly information.

[0096] In this embodiment of the application, the anomaly discrimination threshold selected in step 1043 is used to judge each voxel in the first dataset; all voxels in the first dataset whose element concentration values ​​are greater than the anomaly discrimination threshold are marked, and the set of these marked voxels constitutes geochemical anomaly information.

[0097] In practical applications, for target mining area A, the anomaly detection threshold has been determined to be 1.8. Traverse each voxel in the first dataset. If the element concentration value of a voxel is greater than 1.8, then mark the voxel as 1, which indicates an anomaly. Otherwise, mark it as 0, which indicates background. The spatial distribution data composed of all voxels marked as 1 is the geochemical anomaly information.

[0098] Step 1045: Perform three-dimensional connected component analysis on the geochemical anomaly information to identify and label the independent spatial connected components of the geochemical anomaly information in three-dimensional space.

[0099] In step 1045, an independent spatial connected entity refers to a set of volumes that are connected to each other in three-dimensional space and have consistent properties, as identified through three-dimensional connected domain analysis.

[0100] In this embodiment of the application, a three-dimensional connected component analysis is performed on the geochemical anomaly information obtained in step 1044. Specifically, starting from the voxel marked as an anomaly, it is checked whether the voxels adjacent to it in the six directions of front-back, left-right, up-down in three-dimensional space are also anomaly voxels. If so, they are classified into the same connected component. By traversing all the aomaly voxels, all spatially connected aomaly voxels are aggregated, and each aggregate is assigned a unique label number, thereby identifying and marking multiple independent spatial connected components.

[0101] In practical applications, following the previous example, we perform three-dimensional connectivity analysis on the marked geochemical anomaly information. Assume that three independent spatial connectivity components are identified, numbered 1, 2, and 3, where connectivity component 1 contains 1200 anomaly elements, connectivity component 2 contains 800 anomaly elements, and connectivity component 3 contains 500 anomaly elements.

[0102] Step 1046: Determine the anomaly range based on the spatial distribution pattern of the independent spatial interconnected bodies.

[0103] In this embodiment of the application, based on the actual position and geometric shape of each independent spatial connected body identified in step 1045 in three-dimensional space, such as the spatial coordinate range, volume and three-dimensional shape occupied by each connected body, the anomaly range used to characterize the possible mineralization area is finally delineated.

[0104] In practical applications, continuing the previous example, based on the volume element spatial coordinates of independent spatial connected bodies 1, 2, and 3, the minimum outer envelope cube or convex hull range of each connected body in three-dimensional space is determined. These ranges together constitute the final anomaly range. For example, the range of connected body 1 is approximately between X coordinate [100, 150], Y coordinate [50, 90], and Z coordinate [-200, -180].

[0105] This application, through the above steps, uses a fractal algorithm to adaptively determine the anomaly discrimination threshold, effectively overcoming the limitations of traditional empirical thresholds. Furthermore, by performing three-dimensional connectivity analysis of anomaly information, it accurately delineates the spatial morphology of the anomaly, thereby achieving a more objective and accurate identification and spatial delineation of geochemical anomalies.

[0106] Step 105: Based on the anomaly range and the second dataset, generate fused data, and perform three-dimensional visualization modeling on the fused data to construct a three-dimensional geological model.

[0107] Among them, the fused data is a unified format three-dimensional dataset that simultaneously contains spatial location information of geochemical anomalies and magnetization direction information; the three-dimensional geological model is a digital entity constructed based on multi-source geological data through computer three-dimensional modeling technology, which can quantitatively characterize the geometric shape, attribute distribution and interrelationship of geological elements such as underground rock masses, structures and mineralization bodies in three-dimensional space. The model integrates geochemical anomaly information, geophysical characteristics and other data into a unified three-dimensional spatial framework, and intuitively presents the spatial structure of geological bodies in the form of volume elements, meshes or surface meshes.

[0108] In this embodiment, step 105 includes the following process:

[0109] Step 1051: Obtain the spatial location information corresponding to the abnormal range.

[0110] In step 1051, spatial location information refers to the set of three-dimensional spatial coordinates occupied by the anomaly range, specifically manifested as the position index or specific spatial coordinates of each volume element constituting the anomaly range in the three-dimensional spatial grid.

[0111] In this embodiment of the application, information is extracted from the abnormal range determined in step 1046. Specifically, the coordinate indices of all the volume elements contained in each independent spatial connected body that is identified as abnormal are obtained in the three-dimensional spatial grid. The set of these coordinate indices is the spatial location information corresponding to the abnormal range.

[0112] In practical applications, following the example of step 104, the anomaly range of the target mining area A includes three independent spatial connected bodies: No. 1, No. 2, and No. 3. The volume element coordinate index range of connected body No. 1 is 100 to 150 in the X direction, 50 to 90 in the Y direction, and -200 to -180 in the Z direction. The coordinate indices of all volume elements of these three connected bodies are obtained to form a spatial location information set S.

[0113] Step 1052: Under a unified three-dimensional spatial coordinate system, spatially associate the spatial location information with the second dataset, and merge the spatial association results to generate fused data.

[0114] In this embodiment, firstly, it is ensured that the spatial location information and the second dataset use the same three-dimensional spatial coordinate system and the same grid division; then, each coordinate position in the spatial location information set S is paired with the magnetization direction data stored at the same coordinate position in the second dataset; finally, the paired information corresponding to each coordinate position, that is, the position belongs to the anomalous range and has a specific magnetization direction, is merged as a data unit to generate a new dataset, which is the fused data.

[0115] In practical applications, continuing the previous example, the second dataset is three-dimensional magnetization direction field data, and its grid coordinate system is consistent with the coordinate system of the spatial location information S. For each coordinate index in S, such as coordinate [120, 70, -190], the magnetization direction data at that coordinate in the second dataset is found, assuming it is the direction vector [0.34, 0.85, 0.40]. The coordinate [120, 70, -190] and its corresponding direction vector [0.34, 0.85, 0.40] are merged into a data record. This operation is performed on all coordinates in S, and all the generated data records are collected together to form the fused data F.

[0116] Step 1053: Input the fused data into the feature enhancement network, which generates enhanced fused data by strengthening the regional features in the fused data that are spatially consistent with the second dataset.

[0117] In step 1053, the feature enhancement network is designed as an encoder-decoder architecture, and a three-dimensional spatial attention module is introduced at the end of the encoder. Specifically, the encoder consists of four three-dimensional convolutional layers, each with a kernel size of 3x3x3 and a stride of 2, used to progressively extract and compress the spatial features of the fused data. The three-dimensional spatial attention module receives the feature map output by the encoder, first obtains channel attention weights through a global average pooling layer, and then obtains spatial attention weights through a three-dimensional convolutional layer. The two weights are multiplied together and then element-wise weighted with the original feature map to generate the attention-enhanced feature map. The decoder consists of four 3D transposed convolutional layers, used to upsample the feature maps to restore them to the original input size, and finally output enhanced fused data. The training process of this network is as follows: a training sample set is constructed using a large amount of historical exploration data from known mining areas, where each sample contains the original fused data and corresponding key collaborative region labels annotated by experts. During training, mean squared error is used as the loss function, and the backpropagation algorithm and Adam optimizer are used to iteratively adjust the network parameters so that the feature values ​​of key collaborative regions in the enhanced fused data output by the network are as close as possible to the labels annotated by experts, thereby learning the mapping relationship that strengthens spatial consistency.

[0118] In this embodiment, the feature enhancement network is a convolutional neural network with a three-dimensional attention module. First, the fused data F is converted into a three-dimensional tensor format suitable for network input, where each voxel location contains anomaly markers and magnetization direction vector information. Then, the network's three-dimensional attention module analyzes the spatial consistency between the anomaly markers at each voxel location and the magnetization direction features of surrounding voxels, and calculates a higher attention weight for regions with high consistency. Next, the network uses these attention weights to reweight and aggregate the features in the original fused data, thereby highlighting the regional features where geochemical anomalies and magnetic anomalies coexist spatially. After network processing, enhanced fused data is output, which retains the original information but enhances key co-existing features.

[0119] In practical applications, the fused data F is input into a feature enhancement network. This network processes F, assuming that in the coordinate region [125, 75, -185], the anomaly is marked as 1 and the magnetization direction of the surrounding region all points to the center of the region. The network's 3D attention module calculates the attention weight for this region as 0.95. The network uses this weight to enhance the features of this region. For another region at coordinate [140, 60, -195], the anomaly is marked as 1 but the surrounding magnetization direction is chaotic. The network calculates the attention weight as 0.30 and weakens its features. Finally, the network outputs enhanced fused data, in which the feature values ​​of key collaborative regions are improved.

[0120] Step 1054: Perform three-dimensional graphics rendering processing on the enhanced fusion data, and generate a three-dimensional geological model based on the rendering results using a voxel reconstruction model. The three-dimensional geological model is used to present the spatial correspondence between the anomaly range and the second dataset.

[0121] In step 1054, the structure design of the voxel reconstruction model is based on a conditional generative adversarial network framework, where the generator adopts a 3D U-Net structure and the discriminator adopts a 3D convolutional network. Specifically, the generator's encoding path contains four downsampling blocks, each consisting of a 3D convolutional layer, a batch normalization layer, and an activation function, used to extract multi-scale features from the enhanced fused data. The decoding path contains four upsampling blocks, each consisting of a 3D transposed convolutional layer, a batch normalization layer, and an activation function, and features corresponding to the scale of the encoding path are fused through skip connections, ultimately outputting a 3D probability field representing the likelihood that each voxel belongs to the surface of the geological model. The discriminator consists of five 3D convolutional layers, used to determine whether the input 3D data comes from the surface probability field generated by the generator or not. The model outputs a true / false value based on real geological model surface data. The training process involves using augmented fusion data paired with the feature enhancement network training set, along with corresponding real geological model surface data generated by high-precision 3D laser scanning or geological modeling software, as training samples. During training, the generator and discriminator undergo adversarial training. The generator aims to minimize the difference between its generated probability field and the real surface data while simultaneously deceiving the discriminator, which aims to accurately distinguish between the generated and real data. By alternately optimizing the generator and discriminator using a hybrid loss function that includes adversarial and reconstruction losses, and employing the Adam optimizer for parameter updates, the generator is ultimately able to reconstruct a high-quality 3D geological model surface based on the input augmented fusion data.

[0122] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the corresponding structural design adopted for the internal structure of the feature enhancement network and the voxel reconstruction model. The corresponding settings can be made according to the actual situation.

[0123] In this embodiment, the enhanced fusion data is first rendered in three dimensions. Specifically, a specific color and transparency are assigned to each location in the data according to its attributes, and lighting effects are added to generate a preliminary three-dimensional visualization image. Then, the rendering result is input into a voxel reconstruction model. This model uses surface rendering or volume rendering techniques to convert the rendered voxel data into a three-dimensional solid model with a defined geometric surface, i.e., a three-dimensional geological model. This model can intuitively display the superposition and correspondence between the geochemical anomaly range and magnetization direction data in three-dimensional space.

[0124] In practical applications, the enhanced fusion data is rendered in 3D. The anomalous area is rendered as a semi-transparent red block, and the magnetization direction is represented by a blue arrow. The direction and length of the arrow correspond to the direction vector. The rendering generates a 3D scene image containing the red anomalous block and the blue directional arrow. Then, the rendering result is input into the Marching Cubes voxel reconstruction model. The model extracts the surface mesh of the red anomalous block and integrates it with the blue arrow direction data to construct a 3D geological entity model M containing the surface and internal direction field of the anomalous body.

[0125] Through the steps described above, this application not only achieves spatial fusion of geochemical and geophysical data, but also highlights the intrinsic spatial correlation between the two through feature enhancement networks, and ultimately constructs an intuitive and comprehensive three-dimensional geological model, providing a highly integrated spatial data foundation for geological interpretation and subsequent prediction.

[0126] Step 106: Input the three-dimensional feature data in the three-dimensional geological model into the machine learning prediction model, and process the three-dimensional feature data through the machine learning prediction model to obtain a probability distribution map characterizing whether there is a mineralized domain in the target mining area.

[0127] The machine learning prediction model is structured as a deep neural network consisting of a feature encoding module, a context modeling module, and a task interpretation module. Specific examples of each layer are as follows: The feature encoding module consists of three 3D convolutional layers. The first layer uses 32 3×3×3 convolutional kernels with a stride of 1 and padding of 1, resulting in an output feature map with the same size but 32 channels. This is followed by a ReLU activation function and a max-pooling layer with 2×2×2 pooling kernels. The second layer uses 64 3×3×3 convolutional kernels with a stride of 1 and padding of 1, resulting in 64 output channels, followed by ReLU and max-pooling. The third layer uses 128 3×3×3 convolutional kernels with a stride of 1 and padding of 1, resulting in 128 output channels, followed by ReLU and max-pooling, ultimately outputting the geological feature map. The context modeling module... The module employs a 3D multi-head self-attention mechanism. First, the input feature map is projected into query, key, and value vectors through three different linear layers. Eight attention heads are set, each with an attention dimension of 16. Attention weights are obtained by calculating the dot product of the query and key, scaling, and normalization exponential functions. These weights are then weighted and summed with the value vectors before being fused through a linear layer to produce the fused feature map. The task interpretation module consists of three fully connected layers. The first fully connected layer takes the flattened length of the fused feature map as input (e.g., 92160) and outputs a dimension of 2048, using ReLU activation. The second fully connected layer takes a 2048-dimensional input and outputs a 512-dimensional output, also using ReLU activation. The third fully connected layer takes a 512-dimensional input and outputs the same dimension as the total number of predicted target voxels, with no activation function, outputting the preliminary prediction result.

[0128] The training process of this model is as follows: A large number of 3D geological model samples from known mining areas and their corresponding real mineralization distribution labels are collected, and the samples are divided into training and validation sets. During training, a binary cross-entropy loss function is used as the loss function, the Adam optimizer is used for parameter updates, the initial learning rate is set to 0.001, and batch training is employed. In each training cycle, the training set samples are input into the model to obtain a predicted probability distribution map, the loss value between the predicted map and the real labels is calculated, the gradient is calculated using the backpropagation algorithm, and the model parameters are updated. Simultaneously, the model performance is monitored on the validation set to prevent overfitting. After multiple iterations of training until the loss converges, a well-trained machine learning prediction model is finally obtained.

[0129] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the structural design of modules and other components used in the internal structure of the machine learning prediction model. These can be set according to the actual situation.

[0130] A probability distribution map is a three-dimensional grid data map in which each grid cell contains a value between 0 and 1, which represents the probability of a mineralized domain existing at the spatial location represented by that grid cell.

[0131] In this embodiment, step 106 includes the following process, such as... Figure 2 As shown:

[0132] Step 1061: Extract three-dimensional feature data from the three-dimensional geological model.

[0133] In step 1061, three-dimensional feature data refers to a set of data extracted from a three-dimensional geological model that can characterize geochemical and magnetic anomalies and their spatial relationships, such as the elemental concentration value of each volume element, the magnetization direction vector, and geometric morphological indices calculated from the surface or interior of the model.

[0134] In this embodiment of the application, the element concentration anomaly marker value and magnetization direction vector value contained in each volume element in the three-dimensional geological model are extracted; at the same time, the model is spatially analyzed to extract the distance from each volume element to the surface of the nearest anomaly body, the curvature of the local area, and other derived features; all these feature values ​​are organized into a multi-dimensional feature tensor according to the volume element position, and this tensor is the three-dimensional feature data.

[0135] In practical applications, it is assumed that the three-dimensional geological model M has 100,000 voxels. Two basic features of each voxel are extracted: anomaly marker and magnetization direction vector. At the same time, the nearest distance from each voxel to the surface of the red anomaly block in model M is calculated as a derived feature. Finally, each voxel corresponds to a 5-dimensional feature vector, namely the anomaly marker, the three components of the magnetization direction vector, and the distance to the anomaly surface. The set of feature vectors of all voxels constitutes the three-dimensional feature data Df, which has a shape of 100,000 by 5.

[0136] Step 1062: Input the three-dimensional feature data into the feature encoding module of the machine learning prediction model. Through the multi-layer convolution operation of the feature encoding module, perform spatial feature extraction and dimensionality reduction on the three-dimensional feature data to generate a geological feature map. The geological feature map is used to reflect the local structural features of geochemical and geophysical data in three-dimensional space.

[0137] In step 1062, the geological feature map is a three-dimensional data tensor with lower spatial resolution and higher feature dimension obtained after processing by the encoding module.

[0138] In this embodiment, the feature encoding module includes three three-dimensional convolutional layers. First, the three-dimensional feature data Df is input into the first three-dimensional convolutional layer, which uses 32 convolutional kernels of size 3x3x3 for convolution and is processed by an activation function to obtain the first feature map. Then, the first feature map is input into the second three-dimensional convolutional layer, which uses 64 convolutional kernels of size 3x3x3 for convolution and is downsampled to obtain the second feature map. Finally, the second feature map is input into the third three-dimensional convolutional layer, which uses 128 convolutional kernels of size 3x3x3 for convolution and is downsampled again to output the final geological feature map. The spatial size of the geological feature map is smaller than that of the input data Df, but the number of feature channels at each location is increased to 128, thus comprehensively encoding the geochemical and geophysical structural information of the local area.

[0139] In practical applications, the shape of the three-dimensional feature data Df is 100,000 by 5. It is reshaped into a three-dimensional grid, for example, the grid size is 50 by 50 by 40, and each voxel has 5 feature channels. After being input into the feature encoding module, it goes through three layers of convolution and downsampling. The size of the output geological feature map becomes 12 by 12 by 10, and the number of feature channels becomes 128. This geological feature map is the encoded local structural feature representation.

[0140] Step 1063: Input the geological feature map into the context modeling module of the machine learning prediction model, and calculate the correlation weights between features at different spatial locations in the geological feature map through the three-dimensional attention mechanism of the context modeling module.

[0141] In this embodiment, the context modeling module first linearly projects the input geological feature map into a query vector, a key vector, and a value vector; then, it calculates the dot product similarity between the query vector and all key vectors, divides the dot product result by a scaling factor, and applies a normalization function to obtain the relevance weight between each query location and all other locations.

[0142] In practical applications, the quality feature map has a size of 12 x 12 x 10 and 128 feature channels. After projection, it is assumed that the query vector, key vector, and value vector all have a dimension of 64. For each voxel position i in the feature map, the dot product of its query vector and the key vectors of all voxel positions j (including itself) is calculated, and the dot product formula is the similarity. ,in Indicates the position of the volume element and similarity, It is the query vector at position i. It is the key vector at position j. This represents the dot product operation; then for each i, all Scaling / Scaling ,in The scaled similarity is then represented by a normalization function, such as a normalized exponential function, to convert the similarity into weight values. , of which all For any i, all weights The sum is 1; finally, the set of correlation weights between each voxel position i and all other positions j is obtained.

[0143] It should be noted that this embodiment does not specifically limit the specific expression used for the normalization exponent function; it can be set according to the actual situation.

[0144] Step 1064: The context modeling module uses the correlation weight to perform weighted fusion of the geological feature map to generate a fused feature map, which is used to reflect the cooperative distribution law between geological anomalies and magnetic anomalies in three-dimensional space.

[0145] In this embodiment, the context modeling module utilizes the relevance weights calculated in step 1063. The value vector at each location j in the geological feature map is weighted and summed to generate a new feature representation for each location i. This process incorporates global spatial context information; specifically, the new feature... ,in Location indicated in the geological feature map The new feature representation vector, It is the position in the quality feature map. The value vector, This indicates summing over all positions j; after performing this operation over all positions i, the output new feature map is the fused feature map. This fused feature map enhances spatially correlated features and suppresses uncorrelated features, thus more clearly reflecting the spatial synergy between geochemical anomalies and magnetic anomalies.

[0146] In practical applications, the calculated weights are used. Sum value vector According to the formula Calculate the new feature for each position i; assuming position i = 100, then calculate... 1440 represents the total number of voxels in the geological feature map, which is 12×12×10. After performing the above calculations on all 1440 voxel locations, a fused feature map with the same size of 12×12×10 but updated features is generated. This map emphasizes geochemical anomaly regions that are spatially strongly correlated with magnetic anomaly features.

[0147] Step 1065: Input the fused feature map into the task interpretation module of the machine learning prediction model, and process the fused feature map through the fully connected layer of the task interpretation module to generate a probability distribution map.

[0148] Among them, each fully connected layer, such as the first, second, and third fully connected layers, is a specific, layered network structure component within the task interpretation module that can perform linear transformations and dimensional mappings.

[0149] Step 1065 may specifically include the following steps:

[0150] A1: Convert the fused feature map into a one-dimensional feature vector.

[0151] In step A1, the one-dimensional feature vector is a long vector obtained by flattening the three-dimensional fused feature map in spatial order.

[0152] In this embodiment of the application, all voxels of the fused feature map are arranged in a predetermined order, and the 64-dimensional features of each voxel are joined end to end to obtain a one-dimensional feature vector with a length of 12×12×10×64 equal to 92160.

[0153] In practical applications, the fused feature map is 12×12×10 in size, with a feature dimension of 64; flattening it yields a one-dimensional feature vector with a length of 92160. .

[0154] A2: The one-dimensional feature vector is input into the first fully connected layer, and the one-dimensional feature vector is mapped to a high-dimensional space through the first fully connected layer to obtain high-dimensional features. The high-dimensional features are then nonlinearly activated to generate activated geological features. The high-dimensional features are used to reflect the multi-scale geological correlation of the fused feature map.

[0155] In this embodiment, the first fully connected layer is a linear transformation layer with a weight matrix of size 92160 x 2048; the one-dimensional feature vector... Input to this layer, through matrix multiplication operations ,in It is a weight matrix. It is a bias vector, which yields a high-dimensional feature of 2048 dimensions. Then on By applying nonlinear activation functions, such as the ReLU activation function, activated geological features can be generated.

[0156] In practical applications, The length is 92160; the weight matrix W1 of the first fully connected layer has a size of 2048×92160, and the bias vector b1 has a length of 2048; calculate the high-dimensional features. This results in a 2048-dimensional vector; then... Each element is activated using the ReLU activation function, which sets the element value to 0 if it is less than 0, and otherwise leaves it unchanged, thus generating activated geological features.

[0157] A3: The activated geological features are input into the second fully connected layer. The activated geological features are then subjected to dimensionality reduction mapping through the second fully connected layer to obtain compressed features. The compressed features are used to reflect the synergistic distribution characteristics of geochemical anomalies and magnetic anomalies.

[0158] In this embodiment, the second fully connected layer is a dimension-reduced linear transformation layer with a weight matrix of size 2048 x 512. The activated geological features are input into this layer and processed through matrix multiplication. ,in It is a weight matrix. This represents the activated geological feature vector. The bias vector yields a 512-dimensional compressed feature. This feature is trained to learn a low-dimensional representation that best represents the cooperative distribution pattern.

[0159] A4: Input the compressed features into the third fully connected layer, and map the compressed features to the target dimension through the third fully connected layer to generate preliminary prediction results.

[0160] In step A4, the preliminary prediction result is the raw numerical vector output by the last fully connected layer in the machine learning prediction model. This vector has not yet undergone probability normalization. Each numerical element corresponds to the initial prediction score of a voxel position in the target 3D space. The magnitude of this score is positively correlated with the model's judgment that a mineralized domain exists at that position, but it has not yet been constrained to the standard probability interval.

[0161] In this embodiment, the target dimension is consistent with the total number of voxels in the final output probability distribution map; the weight matrix of the third fully connected layer is 512 x N, where N is the total number of output voxels; the 512-dimensional compressed feature... Input to this layer, through matrix multiplication operations ,in This represents the weight matrix of the third fully connected layer. Let represent the bias vector of the third fully connected layer, resulting in an N-dimensional preliminary prediction vector. Each element corresponds to an initial predicted value for a volume element.

[0162] A5: Normalize the preliminary prediction results to generate a probability distribution map. Each value in the probability distribution map corresponds to a location in three-dimensional space and is used to characterize the probability that a mineralization domain exists at the corresponding location.

[0163] In this embodiment of the application, the preliminary prediction results are normalized to generate a probability distribution map. The specific process is as follows: First, the exponent value of each numerical element in the preliminary prediction result vector is calculated; then, the sum of the exponent values ​​of all numerical elements is calculated; next, for each numerical element, its corresponding exponent value is divided by the sum of all previously calculated exponent values, thereby converting the value of each element to between 0 and 1, and the sum of the converted values ​​of all elements is 1; finally, these converted values ​​are rearranged into a three-dimensional grid data according to their corresponding original three-dimensional voxel space order, and this three-dimensional grid data is the final probability distribution map.

[0164] Through the above steps, this application utilizes a deep learning model to extract and fuse local and global features of three-dimensional geological data layer by layer, and finally outputs a quantitative spatial probability distribution map of mineralization domains covering the entire target mining area, thereby realizing intelligent and quantitative prediction of the spatial location of concealed mineralization domains.

[0165] Figure 3 A schematic diagram of the structure of a mineralization domain spatial distribution prediction system based on prior geological knowledge provided in this application embodiment is shown below. Figure 3 As shown, the system includes:

[0166] The acquisition module 31 is used to acquire core analysis data and magnetic measurement data from boreholes in the target mining area based on prior geological knowledge.

[0167] Analysis module 32 is used to analyze the core analysis data, obtain elemental concentration data, and preprocess the elemental concentration data to obtain a first dataset.

[0168] The conversion module 33 is used to perform potential field conversion on the magnetic measurement data to obtain conversion parameters, calculate the magnetization direction based on the conversion parameters, and generate a second dataset according to the magnetization direction.

[0169] The identification module 34 is used to identify geochemical anomaly information from the first dataset using a fractal algorithm, and to determine the anomaly range based on the geochemical anomaly information.

[0170] The generation module 35 is used to generate fused data based on the anomaly range and the second dataset, and to perform three-dimensional visualization modeling on the fused data to construct a three-dimensional geological model.

[0171] Input module 36 is used to input the three-dimensional feature data in the three-dimensional geological model into the machine learning prediction model, and process the three-dimensional feature data through the machine learning prediction model to obtain a probability distribution map characterizing whether there is a mineralized domain in the target mining area.

[0172] The mineralization domain spatial distribution prediction system based on geological prior knowledge in this application is used to implement the aforementioned mineralization domain spatial distribution prediction method based on geological prior knowledge. Therefore, the specific implementation of the mineralization domain spatial distribution prediction system based on geological prior knowledge can be found in the embodiment section of the mineralization domain spatial distribution prediction method based on geological prior knowledge above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0173] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described methods for predicting the spatial distribution of mineralization domains based on prior geological knowledge.

[0174] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for predicting the spatial distribution of mineralization domains based on prior geological knowledge.

[0175] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0176] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the mineralization domain spatial distribution prediction method based on prior geological knowledge.

[0177] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0178] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0179] The foregoing has provided a detailed description of a method and system for predicting the spatial distribution of mineralization domains based on prior geological knowledge, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for predicting the spatial distribution of mineralization domains based on prior geological knowledge, characterized in that, include: Based on prior geological knowledge, core analysis data and magnetic measurement data from boreholes in the target mining area are obtained; The core analysis data were analyzed to obtain elemental concentration data, and the elemental concentration data were preprocessed to obtain the first dataset. The magnetic measurement data is subjected to potential field transformation to obtain transformation parameters, and the magnetization direction is calculated based on the transformation parameters. A second dataset is then generated based on the magnetization direction. Based on the first dataset, a fractal algorithm is used to identify geochemical anomaly information from the first dataset, and the anomaly range is determined based on the geochemical anomaly information; Based on the anomaly range and the second dataset, fused data is generated, and three-dimensional visualization modeling is performed on the fused data to construct a three-dimensional geological model; The three-dimensional feature data in the three-dimensional geological model is input into the machine learning prediction model. The machine learning prediction model processes the three-dimensional feature data to obtain a probability distribution map characterizing whether there is a mineralized domain in the target mining area.

2. The method according to claim 1, characterized in that, The step of inputting the three-dimensional feature data from the three-dimensional geological model into a machine learning prediction model, and processing the three-dimensional feature data through the machine learning prediction model to obtain a probability distribution map characterizing the existence of mineralized domains within the target mining area includes: Extract three-dimensional feature data from the three-dimensional geological model; The three-dimensional feature data is input into the feature encoding module of the machine learning prediction model. Through the multi-layer convolution operation of the feature encoding module, spatial feature extraction and dimensionality reduction are performed on the three-dimensional feature data to generate a geological feature map. The geological feature map is used to reflect the local structural features of geochemical and geophysical data in three-dimensional space. The geological feature map is input into the context modeling module of the machine learning prediction model. The correlation weights between features at different spatial locations in the geological feature map are calculated through the three-dimensional attention mechanism of the context modeling module. The context modeling module uses the correlation weights to perform weighted fusion of the geological feature map to generate a fused feature map, which is used to reflect the cooperative distribution pattern between geological anomalies and magnetic anomalies in three-dimensional space. The fused feature map is input into the task interpretation module of the machine learning prediction model. The fused feature map is processed through the fully connected layer of the task interpretation module to generate a probability distribution map.

3. The method according to claim 2, characterized in that, The step of processing the fused feature map through the fully connected layer of the task interpretation module to generate a probability distribution map includes: The fused feature map is transformed into a one-dimensional feature vector; The one-dimensional feature vector is input into the first fully connected layer, and the one-dimensional feature vector is mapped to a high-dimensional space through the first fully connected layer to obtain high-dimensional features. The high-dimensional features are then nonlinearly activated to generate activated geological features. The high-dimensional features are used to reflect the multi-scale geological correlation of the fused feature map. The activated geological features are input into the second fully connected layer, and the activated geological features are dimensionality-reduced and mapped through the second fully connected layer to obtain compressed features. The compressed features are used to reflect the synergistic distribution characteristics of geochemical anomalies and magnetic anomalies. The compressed features are input into the third fully connected layer, and the compressed features are mapped to the target dimension through the third fully connected layer to generate preliminary prediction results; The preliminary prediction results are normalized to generate a probability distribution map. Each value in the probability distribution map corresponds to a location in three-dimensional space and is used to characterize the probability that a mineralization domain exists at the corresponding location.

4. The method according to claim 1, characterized in that, The step of identifying geochemical anomaly information from the first dataset using a fractal algorithm and determining the anomaly range based on the geochemical anomaly information includes: From the first dataset, identify regions where the element concentration value is greater than a preset concentration threshold, and calculate the geometric features of the regions. Fractal algorithms are used to analyze the correlation between the geometric features and spatial scale, and to determine the fractal relationship; Based on the fractal relationship, a target concentration threshold is selected from the concentration threshold set as an anomaly detection threshold; The first dataset is processed based on the anomaly discrimination threshold to obtain geochemical anomaly information; Three-dimensional connectivity analysis is performed on the geochemical anomaly information to identify and label the independent spatial connected components of the geochemical anomaly information in three-dimensional space; The range of anomalies is determined based on the spatial distribution pattern of the independent spatial interconnections.

5. The method according to claim 1, characterized in that, The process of generating fused data based on the anomaly range and the second dataset, and performing 3D visualization modeling on the fused data to construct a 3D geological model, includes: Obtain the spatial location information corresponding to the anomaly range; Under a unified three-dimensional spatial coordinate system, the spatial location information is spatially correlated with the second dataset, and the spatial correlation results are merged to generate fused data; The fused data is input into a feature enhancement network, which generates enhanced fused data by strengthening the regional features in the fused data that are spatially consistent with the second dataset. The enhanced fused data is subjected to three-dimensional graphics rendering processing, and based on the rendering results, a three-dimensional geological model is generated using a voxel reconstruction model. The three-dimensional geological model is used to present the spatial correspondence between the anomaly range and the second dataset.

6. The method according to claim 1, characterized in that, The preprocessing of the element concentration data to obtain the first dataset includes: The element concentration data is standardized to obtain standardized concentration data; The standardized concentration data is placed in a three-dimensional spatial grid. For locations in the three-dimensional spatial grid where there is no data, interpolation is performed based on the neighboring standardized concentration data to generate three-dimensional concentration field data. The three-dimensional concentration field data is then used as the first dataset.

7. The method according to claim 1, characterized in that, The process of performing a potential field transformation on the magnetic measurement data to obtain transformation parameters, calculating the magnetization direction based on the transformation parameters, and generating a second dataset according to the magnetization direction includes: The magnetic measurement data is frequency domain converted, and the conversion parameters are calculated based on the converted magnetic measurement data and the positional relationship between the observation location and the magnetic source. The magnetization directions of multiple local regions in three-dimensional space are estimated by using the conversion parameters to calculate the converted magnetic measurement data. The magnetization direction of each local region is placed in a three-dimensional spatial grid to generate a second dataset.

8. A spatial distribution prediction system for mineralization domains based on prior geological knowledge, characterized in that, include: The acquisition module is used to acquire core analysis data and magnetic measurement data from boreholes in the target mining area based on prior geological knowledge. The analysis module is used to analyze the core analysis data, obtain elemental concentration data, and preprocess the elemental concentration data to obtain a first dataset. The conversion module is used to perform potential field conversion on the magnetic measurement data to obtain conversion parameters, calculate the magnetization direction based on the conversion parameters, and generate a second dataset according to the magnetization direction; The identification module is used to identify geochemical anomaly information from the first dataset using a fractal algorithm, and to determine the anomaly range based on the geochemical anomaly information. The generation module is used to generate fused data based on the anomaly range and the second dataset, and to perform three-dimensional visualization modeling on the fused data to construct a three-dimensional geological model; The input module is used to input the three-dimensional feature data in the three-dimensional geological model into the machine learning prediction model, and to process the three-dimensional feature data through the machine learning prediction model to obtain a probability distribution map characterizing whether there is a mineralized domain in the target mining area.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for predicting the spatial distribution of mineralization domains based on geological prior knowledge as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the method for predicting the spatial distribution of mineralization domains based on prior geological knowledge as described in any one of claims 1 to 7.

Images