Three-dimensional metallogenic prediction method, device, equipment and medium
By combining the SE-DNN model and the 3D ellipsoidal kriging model with the COPF framework, multi-dimensional weights and correlation potential functions were constructed, which solved the problems of spatial correlation and nonlinear fitting in three-dimensional mineralization prediction and achieved high-precision exploration of deep mineral resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-04-24
- Publication Date
- 2026-07-14
AI Technical Summary
Existing three-dimensional mineralization prediction methods struggle to balance spatial correlation with nonlinear fitting and integrate local features with global patterns when faced with complex deep geological environments. This results in predictions that deviate from the actual geological background and rely on prior geological assumptions, leading to insufficient prediction accuracy.
The SE-DNN model is used to process the mineralization control factor matrix. Combined with the 3D ellipsoidal kriging model and the COPF framework, multi-dimensional weights are constructed through correlation potential and interaction potential functions. The mean field approximation is used for posterior distribution inference, and the mean of mineralization variables is iteratively optimized through a recurrent convolution module to achieve end-to-end training.
It improves the accuracy and geological rationality of three-dimensional mineralization prediction, reduces reliance on prior geological assumptions, adapts to the geological conditions of different types of deposits, and provides accurate prediction of continuous mineralization variables.
Smart Images
Figure CN122089512B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mineral resource exploration technology, and in particular to a three-dimensional mineralization prediction method, apparatus, equipment and medium. Background Technology
[0002] As mineral resource exploration extends to deeper areas, three-dimensional metallogenic prediction has become a key technical means to solve the problem of resource shortage. Although traditional geostatistical methods and machine learning models have been widely used in the field of metallogenic prediction, they still reveal many insurmountable shortcomings when faced with complex deep geological environments. First, while traditional geostatistical methods can consider spatial correlations, their ability to fit nonlinear metallogenic laws is limited, and they rely too heavily on prior geological assumptions, making it difficult to adapt to the changing deep mineralization conditions. On the other hand, mainstream machine learning models mostly use the independent and identically distributed assumption, ignoring the inherent spatial dependence between mineralized volumes formed by geological structures, ore-forming fluid migration, etc., resulting in prediction results that deviate from the actual geological background. Second, existing three-dimensional metallogenic prediction models generally suffer from localized modeling limitations, either focusing only on the feature representation of a single volume or only considering the simple correlation of neighboring volumes, making it difficult to capture the overall pattern of the three-dimensional distribution of deep deposits.
[0003] In summary, current three-dimensional mineralization prediction methods have significant shortcomings in balancing spatial correlation and nonlinear fitting, integrating local features and global patterns, and improving the accuracy of deep prediction. Summary of the Invention
[0004] In order to at least solve one of the technical problems existing in the prior art, the present invention provides a three-dimensional mineralization prediction method, apparatus, equipment and medium.
[0005] One aspect of the present invention provides a three-dimensional mineralization prediction method, comprising:
[0006] Acquire multi-source mineral control information, fuse and standardize the multi-source mineral control information to obtain key mineral control factors, and then construct a mineral control factor matrix;
[0007] The mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix using the SE-DNN model, and the correlation potential function is constructed based on the mineralization reference value.
[0008] The spatial weights, feature similarity weights, and kriging coupling coefficients of voxels are calculated using a 3D ellipsoidal kriging model, and then an interaction potential function that integrates multi-dimensional weights is constructed.
[0009] The COPF global energy function is constructed based on the correlation potential function and the interaction potential function. The mean field approximation is used to infer the posterior distribution and obtain the initial mean and variance of the volume element mineralization variable.
[0010] The volumetric mineralization variables are trained end-to-end through recurrent convolution to optimize the learnable parameters, and the mean of the volumetric mineralization variables is iteratively updated until the convergence condition is met, thus obtaining the three-dimensional mineralization regression prediction results.
[0011] According to the aforementioned three-dimensional mineralization prediction method, multi-source mineralization control information is acquired, and this information is then fused and standardized to obtain key mineralization control factors. A mineralization control factor matrix is then constructed, including:
[0012] Acquire multi-source mineral control information and extract the 3D coordinates of the target mining area;
[0013] Data cleaning and Z-score normalization were performed on multi-source mineral control information, and then feature importance was used for evaluation and screening to obtain key mineral control factors.
[0014] Using the 3D coordinates of the target mining area as the basic spatial feature, the key ore-controlling factors are arranged according to the volume element dimension to obtain the ore-controlling factor matrix.
[0015] According to the aforementioned three-dimensional mineralization prediction method, the mineralization control factor matrix is processed using the SE-DNN model to obtain mineralization reference values for volume elements. Based on these mineralization reference values, a correlation potential function is constructed, including:
[0016] A SE-DNN model is constructed, and a mean squared error loss function is built using the mineralization control factor matrix and real mineralization variables as labels. The Adam optimizer is used for training until the loss converges. The mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix through the trained SE-DNN model.
[0017] Based on the consistency between the predicted and reference values of mineralization, a quadratic function is used to construct the correlation potential function. for:
[0018] ;
[0019] in, For learnable correlation potential weights, For the first Mineralization prediction values of individual elements, Indicates the first Reference value for mineralization of individual elements. Indicates the first Individual element.
[0020] According to the aforementioned three-dimensional mineralization prediction method, the SE-DNN model includes a feature extraction layer and a squeeze-excited attention module. The feature extraction layer consists of multiple fully connected layers, with ReLU as the activation function and the input dimension matching the number of columns in the mineralization control factor matrix. The squeeze-excited attention module aggregates channel information through global average pooling, models inter-channel dependencies through a lightweight gating mechanism, and recalibrates features through output channel weights.
[0021] According to the aforementioned three-dimensional mineralization prediction method, a 3D ellipsoidal kriging model is used to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of volume elements, thereby constructing an interaction potential function that integrates multi-dimensional weights, including:
[0022] The anisotropic spatial kernel is determined by the 3D ellipsoid variogram, and the spatial similarity between adjacent voxels is obtained:
[0023] ;
[0024] in, and They are respectively body elements , 3D coordinates For space kernel bandwidth parameters, Representing adjacent volume elements and The heterogeneous space core between them, in which the volume element As the target element, element As adjacent volumes of the target volume;
[0025] The Euclidean distance is calculated based on the high-dimensional ore-controlling feature vector of the volume element, and the feature kernel is obtained through Gaussian kernel function transformation. Among them, the characteristic kernel Used to quantify feature domain similarity:
[0026] ;
[0027] in , For body element , High-dimensional ore-controlling feature vectors, Indicates Euclidean distance. The feature kernel bandwidth parameter;
[0028] Construct the covariance matrix based on known mineralized samples. Calculate the covariance vector between the target volume element and the known mineralized sample. The Kriging coupling coefficient was obtained by solving the 3D ellipsoidal Kriging model. for:
[0029] ;
[0030] in, Covariance matrix The inverse matrix, Kriging coupling coefficient Used to quantify the spatial influence weights of adjacent voxels on the target voxel;
[0031] By fusing the heterogeneous space kernel, the characteristic kernel, and the kriging coupling coefficients, the interaction potential function is obtained. for:
[0032] ;
[0033] in, For learnable interaction potential weights, and For balance coefficient, For the first Mineralization prediction values for individual elements.
[0034] According to the aforementioned three-dimensional mineralization prediction method, a COPF global energy function is constructed based on the correlation potential function and the interaction potential function. A mean-field approximation is used for posterior distribution inference to obtain the initial mean and variance of the volumetric mineralization variables, including:
[0035] A COPF global energy function is constructed based on the correlation potential function and the interaction potential function to determine the prediction accuracy and spatial continuity between volumes. for:
[0036] ;
[0037] in, For the first The correlation potential function of individual elements, Adjacent volume elements and The interaction potential function, Given the known ore-controlling characteristics, Unknown mineralization state The total number of volume elements;
[0038] The joint posterior distribution of mineralization variables is defined based on the Gibbs distribution. for:
[0039] ;
[0040] in, This is the partition function, used to ensure the normalization of the posterior distribution;
[0041] Using the mean-field approximation, the joint posterior distribution is decomposed into a product of independent Gaussian marginal distributions:
[0042] ;
[0043] Each factor in the mean-field approximation satisfies By minimizing the KL divergence Iterative optimization of the mean With variance ;
[0044] Based on the quadratic property of the potential function, the logarithmic probability of the approximated mean field is derived, thus obtaining the mean. Iterative update formula for:
[0045] ;
[0046] The initial mean was determined using the mineralization reference value output by the SE-DNN model. Represents the correlation potential weight. For the related potential, The total number of interaction potential functions, Index for the interaction potential function, For the first Interaction potential weights of each interaction potential function.
[0047] According to the aforementioned three-dimensional mineralization prediction method, the volumetric mineralization variables are trained end-to-end through recurrent convolution to optimize the learnable parameters, iteratively updating the mean of the volumetric mineralization variables until the convergence condition is met, thus obtaining the three-dimensional mineralization regression prediction results, including:
[0048] According to the mean With variance Construct a recurrent convolution module, and update the mean by performing neighborhood information transfer, compatibility transformation and fusion update during iteration according to a preset number of iterations;
[0049] Using the global energy function as the optimization objective, the gradient descent algorithm is employed for backpropagation to simultaneously optimize the parameters and associated potential weights of the SE-DNN model. Interaction potential weights and balance coefficient , And L2 regularization is applied to the parameters of the SE-DNN model, as well as the correlation potential weights. and interaction potential weights Add non-negativity constraints;
[0050] Calculate the mean of all voxels in two adjacent iterations. The maximum absolute change is determined. If the change is less than a preset threshold or reaches a preset number of iterations, the iteration is terminated, and the ore-forming regression prediction result is obtained.
[0051] Another aspect of the present invention provides a three-dimensional mineralization prediction device, comprising:
[0052] The first module is used to acquire multi-source mineral control information, fuse and standardize the multi-source mineral control information to obtain key mineral control factors, and then construct a mineral control factor matrix.
[0053] The second module is used to process the mineralization control factor matrix using the SE-DNN model to obtain the mineralization reference value of the volume element, and to construct the correlation potential function based on the mineralization reference value;
[0054] The third module is used to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of voxels using a 3D ellipsoidal kriging model, and then construct an interaction potential function that integrates multi-dimensional weights.
[0055] The fourth module is used to construct the COPF global energy function based on the correlation potential function and the interaction potential function, and to use the mean field approximation to infer the posterior distribution, thereby obtaining the initial mean and variance of the volume element mineralization variable.
[0056] The fifth module is used to optimize the learnable parameters by performing recurrent convolution and end-to-end training on the volumetric mineralization variables, and to iteratively update the mean of the volumetric mineralization variables until the convergence condition is met, so as to obtain the three-dimensional mineralization regression prediction results.
[0057] Another aspect of the present invention provides an electronic device, including a processor and a memory;
[0058] The memory is used to store programs;
[0059] The processor executes the program to implement the method as described above.
[0060] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the methods described above.
[0061] The beneficial effects of this invention are as follows: This invention, through a kriging-enhanced COPF framework, adapts to the regression prediction needs of continuous mineralization variables, breaking through the application limitations of traditional classification-based CRFs. By synergistic constraints of correlation potential and interaction potential functions, it balances the prediction accuracy of individual volumes with the spatial continuity between volumes, improving the geological rationality of the prediction results. It integrates SE-DNN and a 3D ellipsoidal kriging model. SE-DNN effectively extracts high-dimensional mineralization-controlling features through an attention mechanism, generating accurate mineralization reference values. The 3D ellipsoidal kriging model achieves anisotropic spatial correlation modeling, combining feature kernels and kriging coupling coefficients to construct multi-dimensional weights, strengthening spatial continuity constraints, and solving the problem of insufficient spatial correlation modeling in traditional methods. The problem involves using a mean field approximation to infer posterior distributions, decomposing complex joint distributions into independent Gaussian marginal distributions to reduce computational complexity; iteratively optimizing the mean through a recurrent convolution module to balance local geological heterogeneity and global spatial continuity, thereby improving inference efficiency and accuracy; employing an end-to-end training model to simultaneously optimize SE-DNN parameters, potential function weights, and balance coefficients, achieving an organic integration of geostatistical priors and data-driven learning, reducing the loss of high-value mineralization information; and adapting ore-controlling factors and model parameters to different types of deposits (metallic and non-metallic) and geological conditions of exploration areas, providing accurate and continuous mineralization variable prediction results for deep mineral resource exploration without relying on complex prior geological assumptions. Attached Figure Description
[0062] Figure 1 This is a schematic diagram of the three-dimensional mineralization prediction process according to an embodiment of the present invention;
[0063] Figure 2 This is a schematic diagram of the mineral control factor matrix construction process according to an embodiment of the present invention;
[0064] Figure 3 This is a schematic diagram of the end-to-end training process according to an embodiment of the present invention;
[0065] Figure 4 This is a scatter plot comparing the performance of the model in this embodiment of the invention with mainstream machine learning models, where (a) is... (a) is a scatter plot comparing the performance over time; (b) is a scatter plot comparing the performance over time. Scatter plot comparing performance over time; (c) is The performance comparison scatter plot at different times; (d) is... Scatter plot comparing performance over time;
[0066] Figure 5 This is a schematic diagram of a three-dimensional mineralization prediction device according to an embodiment of the present invention. Detailed Implementation
[0067] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings. Throughout the description, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions. In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no specific meaning in themselves. Therefore, "module," "part," or "unit" can be used interchangeably. Terms such as "first," "second," etc., are used only to distinguish technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the sequential relationship of the indicated technical features. In the following description, the consecutive reference numerals for method steps are for ease of review and understanding. Adjusting the implementation order of steps, in conjunction with the overall technical solution of the present invention and the logical relationship between the various steps, will not affect the technical effect achieved by the technical solution of the present invention. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0068] refer to Figure 1 A schematic diagram of a three-dimensional mineralization prediction process, including but not limited to steps S100~S500:
[0069] S100: Obtain multi-source mineral control information, fuse and standardize the multi-source mineral control information to obtain key mineral control factors, and then construct a mineral control factor matrix.
[0070] In some embodiments, reference Figure 2 The schematic diagram shown illustrates the process of constructing the mineral control factor matrix, which includes, but is not limited to, steps S110 to S130:
[0071] S110, acquire multi-source mineral control information, and extract the 3D coordinates of the target mining area;
[0072] S120 performs data cleaning and Z-score normalization on multi-source mineral control information, and then uses feature importance to evaluate and screen key mineral control factors.
[0073] S130 uses the 3D coordinates of the target mining area as the spatial basis feature, and arranges the key ore-controlling factors according to the volume element dimension to obtain the ore-controlling factor matrix.
[0074] In some embodiments, multi-source mineralization control information is fused and standardized to construct a mineralization control factor matrix, while the 3D coordinate data of all volume elements are extracted as spatial basic features. All extracted mineralization-related information retains its continuous characteristics to avoid information loss due to discretization. Standardization processing involves preprocessing the extracted multi-source mineralization control information, including data cleaning (using box plots to remove outliers and linear interpolation to supplement missing values) and normalization (using the Z-score method to transform data of different dimensions to a scale with a mean of 0 and a variance of 1), ensuring data consistency and usability, and laying the foundation for subsequent model training. Key factor screening and matrix construction: Key mineralization control factors with significant influence on continuous mineralization variables (grade) are screened using feature engineering methods such as random forest feature importance assessment and Pearson correlation analysis, and redundant features are removed. The quantified key mineralization control factors are arranged according to the volume element dimension to form a mineralization control factor matrix. ,in The total number of discrete elements in the study area. This represents the number of key ore-controlling factors.
[0075] For example, in the practice of three-dimensional mineral prospect modeling of a gold ore belt in a certain mining area, it is necessary to take the ore-controlling laws as the core, systematically construct a predictive variable system, and comprehensively extract multi-dimensional mineralization information. Specifically, the predictive variables cover four key ore-controlling dimensions: First, the characteristics of tectonic expansion space, including dip-direction expansion space (exFdip), dip-direction mineralization precipitation favorability (preFdip), strike-direction expansion space (exFstr), and strike-direction mineralization precipitation favorability (preFstr), accurately depicting the mineralization insitu spatial conditions generated by tectonic deformation; Second, the characteristics of alteration zone boundaries, quantifying the distance between the ore body and the strongly developed alteration zone through the lower boundary distance (thinAlt_down) and upper boundary distance (thinAlt_up). The study examines four key aspects of mineralization: spatial correlation, fluid migration and enrichment characteristics (including fluid migration distance (disFluid), shear-direction fluid migration distance (disFluid_JJ), fluid flux intensity (fluxFluid), and shear-direction fluid flux intensity (fluxFluid_JJ), quantifying the control of fluid activity on mineralization using Darcy's fluid dynamics principles; and fault-controlled mineralization characteristics (including the distance of the main fault (Dist) and the fault azimuth (dvF, the horizontal angle between the fault plane and true north, reflecting the distribution direction of the fault system and the background of fluid migration). By systematically analyzing and optimizing these predictive variables, and considering the specific geological background of the deposit, accurate extraction and quantitative characterization of mineralization-controlling information can be achieved, providing solid data support for three-dimensional mineral prospect modeling and effectively revealing the control mechanism of mineralization-controlling factors on the spatial distribution of ore bodies.
[0076] S200, the mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix using the SE-DNN model, and the correlation potential function is constructed based on the mineralization reference value.
[0077] In some embodiments, an SE-DNN model is constructed, and a mean squared error loss function is built using the mineralization control factor matrix and real mineralization variables as labels. The model is trained using an Adam optimizer until the loss converges. The mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix through the trained SE-DNN model. Based on the consistency between the predicted mineralization value and the mineralization reference value, a quadratic function is used to construct a correlation potential function. for:
[0078] ;
[0079] in, For learnable correlation potential weights, For the first Mineralization prediction values of individual elements, Indicates the first Reference value for mineralization of individual elements. Indicates the first Individual element.
[0080] The SE-DNN model includes a feature extraction layer and a squeeze-excitation attention module. The feature extraction layer consists of multiple fully connected layers, with ReLU as the activation function and the input dimension matching the number of columns in the mining factor matrix. The squeeze-excitation attention module aggregates channel information through global average pooling, models inter-channel dependencies through a lightweight gating mechanism, and recalibrates features through output channel weights.
[0081] In some embodiments, an SE-DNN model is constructed, wherein the model consists of a feature extraction layer and a squeeze-excitation (SE) attention module. The feature extraction layer contains three fully connected layers with 256, 128, and 64 neurons respectively, all using ReLU as the activation function. The input layer dimension matches the number of columns (m) of the ore-controlling factor matrix, and is used to extract high-dimensional nonlinear ore-controlling features. The SE attention module is used to adaptively calibrate the channel feature weights, enhance the expressive power of useful features, and suppress redundant information.
[0082] SE module implementation:
[0083] ① Squeezing operation: Squeezing the feature map output from the last layer of the feature extraction layer. Global average pooling is performed to aggregate the features of each channel into a scalar value, thus obtaining global channel information. This enables the global aggregation of channel information.
[0084] ② Activation Operation: Construct a lightweight gating mechanism consisting of two fully connected layers. The first fully connected layer compresses the channel dimension to 1 / 16 of the original dimension, using ReLU activation function. The second fully connected layer restores the dimension to its original size, using Sigmoid activation function, and outputs the channel weights. ,in , Let be a learnable parameter matrix, where This is the learnable parameter matrix (dimensionality reduction) for the first fully connected layer. This is the learnable parameter matrix (increased dimensionality) for the second fully connected layer. Use the Sigmoid activation function;
[0085] ③ Recalibration operation: Adjust channel weights With original features Element-wise multiplication is performed to obtain the recalibrated features. This strengthens the characteristic channels that are effective for mineralization prediction.
[0086] Model training and reference value generation: The ore-controlling factor matrix X is divided into a training set (70%) and a validation set (30%). The training set is input into the SE-DNN model, and a mean squared error (MSE) loss function is constructed using the actual ore grade as the label. The Adam optimizer (learning rate set to 1e-3) is used to train the model, and the network parameters are iteratively updated until the validation set loss converges. The trained SE-DNN model is then used to process the ore-controlling factors of all volumes, and the mineralization reference values of each volume are output. .
[0087] Correlation potential function construction: To constrain the consistency between the predicted values of individual volumes and the reference values of SE-DNN, a quadratic function form is used to construct the correlation potential function, expressed as follows:
[0088] ;
[0089] in These are learnable weight parameters used to adjust the contribution strength of the correlation potential. For the first The mineralization prediction value of an individual element. This quadratic form ensures that when the prediction value... Close to reference value At this point, the correlation potential function reaches its minimum value, driving the regression process to converge toward a geologically reasonable direction.
[0090] S300 uses a 3D ellipsoidal kriging model to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of voxels, and then constructs an interaction potential function that integrates multi-dimensional weights.
[0091] In some embodiments, the anisotropic spatial kernel is determined by the 3D ellipsoid variogram function to obtain the spatial similarity between adjacent voxels:
[0092] ;
[0093] in, and They are respectively body elements , 3D coordinates For space kernel bandwidth parameters, Representing adjacent volume elements and The heterogeneous space core between them, in which the volume element As the target element, element As adjacent volumes of the target volume; express and Anisotropic space kernel, wherein the anisotropic space kernel is constructed based on the variogram function of a 3D ellipsoid. Used to characterize volume elements and The spatial similarity between them includes rotating the 3D spatial coordinates using the Euler angle rotation matrix to adjust the azimuth, dip, and deflection angles to match the anisotropic distribution of the mineralization; then the Euclidean distance between the transformed voxels is calculated and substituted into the Gaussian kernel function to obtain the spatial kernel.
[0094] Extracting recalibrated features from the SE-DNN model As a high-dimensional ore-controlling feature vector of the volume element, calculate the volume element. and Euclidean distance of eigenvectors The feature kernel is obtained through Gaussian kernel function transformation:
[0095] ;
[0096] in , For body element , High-dimensional ore-controlling feature vectors, Indicates Euclidean distance. The feature kernel bandwidth parameter;
[0097] Construct a covariance matrix based on known mineralized samples (including true grade labels). Calculate the covariance vector between the target volume element and the known mineralized sample. The Kriging coupling coefficient was obtained by solving the 3D ellipsoidal Kriging model. for:
[0098] ;
[0099] in, Covariance matrix The inverse matrix, Kriging coupling coefficient Used to quantify the spatial influence weights of adjacent voxels on the target voxel, ensuring the geological rationality of spatial correlation modeling;
[0100] By fusing the heterogeneous space kernel, the characteristic kernel, and the kriging coupling coefficients, the interaction potential function is obtained. for:
[0101] ;
[0102] in, For learnable interaction potential weights, and For balance coefficient, For the first The mineralization prediction value of each individual element is used to adjust the contribution ratio of the three weighting terms. This function ensures the spatial continuity of the prediction results by penalizing the difference in prediction values between adjacent elements, while taking into account spatial similarity, feature similarity, and geostatistical prior.
[0103] For example, the feature kernel bandwidth parameter For example, calculating the feature kernel Construct the covariance matrix based on known Au grade samples. The Kriging coupling coefficients are obtained by solving. Set the balance coefficient , Construct the interaction potential function, where The initial value is set to 0.5.
[0104] S400, based on the correlation potential function and the interaction potential function, constructs the COPF global energy function, and uses the mean field approximation to infer the posterior distribution, thus obtaining the initial mean and variance of the volume element mineralization variable.
[0105] In some embodiments, COPF (Cooperative Potential Field) is a cooperative potential field model. A global energy function for COPF, which constructs the volumetric prediction accuracy and spatial continuity between volumetric elements, is built based on the correlation potential function and the interaction potential function. for:
[0106] ;
[0107] in, Given the known ore-controlling characteristics, Unknown mineralization state The total number of volume elements, For the first The correlation potential function of individual elements, Adjacent volume elements and The interaction potential function;
[0108] The joint posterior distribution of mineralization variables is defined based on the Gibbs distribution. for:
[0109] ;
[0110] in, The partition function is used to ensure the normalization of the posterior distribution. The initial value was set to 0.6, and the mean field was used for approximate inference. The mean field was iteratively constructed into a recurrent convolution module, with T=10 iterations. The global energy function was used as the optimization objective, and the Adam optimizer (learning rate 5e-4) was used to optimize the parameters end-to-end. After 200 iterations, the global energy converged. During the iteration process, the maximum absolute change of the mean between two adjacent rounds gradually decreased from the initial 0.12 to 0.008 (less than the convergence threshold 1e-3), and the iteration was terminated.
[0111] Using the mean-field approximation, the joint posterior distribution is decomposed into a product of independent Gaussian marginal distributions:
[0112] ;
[0113] Each factor in the mean-field approximation satisfies By minimizing the KL divergence Iterative optimization of the mean With variance ;
[0114] Based on the quadratic property of the potential function, the logarithmic probability of the approximated mean field is derived, thus obtaining the mean. Iterative update formula for:
[0115] ;
[0116] in, Represents the correlation potential weight. For the related potential, The total number of interaction potential functions, Index for the interaction potential function, For the first The interaction potential weights of the interaction potential functions are set during the initial iteration. That is, the reference value output by SE-DNN.
[0117] S500 optimizes the learnable parameters by performing recurrent convolution and end-to-end training on the volumetric mineralization variables, iteratively updating the mean of the volumetric mineralization variables until the convergence condition is met, thus obtaining the three-dimensional mineralization regression prediction results.
[0118] In some embodiments, referring to the end-to-end training process diagram shown in Figure 3, the process includes, but is not limited to, steps S510-S530:
[0119] S510, based on the mean With variance Construct a recurrent convolution module, and update the mean by performing neighborhood information transfer, compatibility transformation and fusion update during iteration according to a preset number of iterations;
[0120] S520 uses the global energy function as the optimization objective and employs the gradient descent algorithm for backpropagation to simultaneously optimize the parameters and associated potential weights of the SE-DNN model. Interaction potential weights and balance coefficient , And L2 regularization is applied to the parameters of the SE-DNN model, as well as the correlation potential weights. and interaction potential weights Add non-negativity constraints;
[0121] S530, calculate the mean of all volume elements in two adjacent iterations. The maximum absolute change is determined. If the change is less than a preset threshold or reaches a preset number of iterations, the iteration is terminated, and the ore-forming regression prediction result is obtained.
[0122] For example, the recurrent convolution module is constructed as follows: the mean field iteration process is built into a recurrent convolution module, with the number of iterations T set to 8-10. Each iteration includes three core steps: ① Neighborhood information transfer: calculating the spatial kernel, feature kernel, and kriging coupling coefficient between the target voxel and all neighboring voxels; ② Compatibility transformation: performing a weighted transformation on the mean of the neighboring voxels through an interaction potential function; ③ Fusion update: updating the mean of the target voxel based on the constraints of the correlation potential function. The mean output of the previous iteration serves as the input for the next iteration, forming a recurrent transfer mechanism.
[0123] End-to-end parameter optimization is performed using the global energy function as the optimization objective. The Adam optimizer is used for end-to-end training, simultaneously optimizing all learnable parameters, including the SE-DNN network parameters and correlation potential weights. Interaction potential weights and balance coefficient , To avoid overfitting, L2 regularization is applied to the SE-DNN network parameters; to ensure the physical meaning of the parameters, L2 regularization is applied to... Apply nonnegativity constraints; calculate the gradients of each parameter through gradient backpropagation, and iteratively update the parameters until the global energy function converges.
[0124] Convergence judgment and result output: Calculate the maximum absolute change in the mean μ of all volume elements in two adjacent iterations. .like If the convergence threshold is less than the preset convergence threshold or the number of iterations reaches the preset maximum value T, the iteration terminates; otherwise, it returns to continue iterating. The final output is the mean value of the mineralization variable for all voxels. This refers to the three-dimensional mineralization regression prediction result, which is a continuous value and can directly reflect the three-dimensional spatial distribution law of mineralization variables such as ore grade.
[0125] Validation reference for prediction results of embodiments of the present invention Figure 4 The model performance comparison chart shows a scatter plot comparing the performance of the model of this invention with mainstream machine learning models (the horizontal axis represents the actual mineralization value, and the vertical axis represents the model's predicted value). The dashed line y=x represents the ideal fitting state, where (a) is... (a) is a scatter plot comparing the performance over time; (b) is a scatter plot comparing the performance over time. Scatter plot comparing performance over time; (c) is The performance comparison scatter plot at different times; (d) is... A scatter plot comparing the performance of the method in this embodiment is presented. The prediction results are compared with traditional multilayer perceptron, random forest, and support vector machine models, using the coefficient of determination (R²), mean squared error (MSE), and root mean square error (RMSE) as evaluation metrics. The results show that the method of this invention has an R² of 0.926, an MSE of 0.12, and an RMSE of 0.346; the best performing traditional machine learning model is the multilayer perceptron, with an R² of 0.849, an MSE of 0.244, and an RMSE of 0.494. Therefore, the performance of the model in this invention is superior. The highest value (0.926) and the smallest error in each category indicate that the prediction performance is significantly better than other models, verifying its accuracy advantage in mineralization prediction. Therefore, the prediction accuracy of the method in this invention is significantly better than that of traditional methods, indicating that the spatial continuity of the prediction results is more in line with the geological mineralization law.
[0126] Figure 5 This is a schematic diagram of a three-dimensional mineralization prediction device according to an embodiment of the present invention. The device includes a first module 510, a second module 520, a third module 530, a fourth module 540, and a fifth module 550.
[0127] The system comprises five modules: The first module acquires multi-source mineralization control information, fuses and standardizes this information to obtain key mineralization control factors, and then constructs a mineralization control factor matrix. The second module processes the mineralization control factor matrix using the SE-DNN model to obtain mineralization reference values for voxels, and constructs a correlation potential function based on these reference values. The third module uses a 3D ellipsoidal kriging model to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of voxels, and then constructs an interaction potential function that integrates multi-dimensional weights. The fourth module constructs a COPF global energy function based on the correlation and interaction potential functions, uses a mean field approximation for posterior distribution inference, and obtains the initial mean and variance of the voxel mineralization variables. The fifth module optimizes the learnable parameters by performing recurrent convolution and end-to-end training on the voxel mineralization variables, iteratively updating the mean of the voxel mineralization variables until convergence conditions are met, thus obtaining the three-dimensional mineralization regression prediction results.
[0128] Exemplarily, with the cooperation of the first, second, third, fourth, and fifth modules in the device, the embodiment device can implement any of the aforementioned three-dimensional mineralization prediction methods, namely, acquiring multi-source mineralization control information, fusing and standardizing the multi-source mineralization control information to obtain key mineralization control factors, and then constructing a mineralization control factor matrix; processing the mineralization control factor matrix using the SE-DNN model to obtain the mineralization reference values of the voxels, and constructing a correlation potential function based on the mineralization reference values; calculating the spatial weights, feature similarity weights, and kriging coupling coefficients of the voxels using a 3D ellipsoidal kriging model, and then constructing an interaction potential function that fuses multi-dimensional weights; constructing a COPF global energy function based on the correlation potential function and the interaction potential function, and using the mean field approximation for posterior distribution inference to obtain the initial mean and variance of the voxel mineralization variables; optimizing the learnable parameters by performing recurrent convolution and end-to-end training on the voxel mineralization variables to iteratively update the mean of the voxel mineralization variables until the convergence condition is met, and obtaining the three-dimensional mineralization regression prediction result. The beneficial effects of this invention are as follows: The Kriging-enhanced COPF framework adapts to the regression prediction needs of continuous mineralization variables, breaks through the application limitations of traditional classification CRF, and improves the geological rationality of the prediction results by synergistic constraints of correlation potential and interaction potential functions, taking into account both the prediction accuracy of a single volume element and the spatial continuity between volumes; it integrates SE-DNN and the 3D ellipsoidal Kriging model. SE-DNN effectively extracts high-dimensional mineralization-controlling features through an attention mechanism to generate accurate mineralization reference values; the 3D ellipsoidal Kriging model realizes anisotropic spatial correlation modeling, and constructs multi-dimensional weights by combining feature kernels and Kriging coupling coefficients, strengthening spatial continuity constraints and solving the problem of insufficient spatial correlation modeling in traditional methods; The posterior distribution inference is approximated using a mean field, decomposing the complex joint distribution into independent Gaussian marginal distributions to reduce computational complexity. The mean is iteratively optimized through a recurrent convolution module to balance local geological heterogeneity and global spatial continuity, improving inference efficiency and accuracy. An end-to-end training mode is adopted to simultaneously optimize SE-DNN parameters, potential function weights, and balance coefficients, achieving an organic integration of geostatistical priors and data-driven learning, reducing the loss of high-value mineralization information. By adjusting ore-controlling factors and model parameters, the model can be adapted to different types of deposits (metallic and non-metallic) and geological conditions of exploration areas, providing accurate prediction results of continuous mineralization variables for deep mineral resource exploration without relying on complex prior geological assumptions.
[0129] This invention also provides an electronic device, which includes a processor and a memory;
[0130] The memory stores the program;
[0131] The processor executes a program to perform the aforementioned three-dimensional mineralization prediction method; the electronic device has the function of carrying and running the three-dimensional mineralization prediction software system provided in the embodiments of the present invention, such as a personal computer, minicomputer, mainframe, workstation, network or distributed computing environment, standalone or integrated computer platform, or communicating with charged particle tools or other imaging devices, etc.
[0132] This invention also provides a computer-readable storage medium storing a program that is executed by a processor to implement the three-dimensional mineralization prediction method described above.
[0133] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented in the embodiments of this invention. Alternative embodiments are contemplated, in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.
[0134] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned three-dimensional mineralization prediction method.
[0135] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, considering the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed in the embodiments of the invention, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.
[0136] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0137] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can include, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0138] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0139] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0140] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0141] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
[0142] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.
Claims
1. A three-dimensional mineralization prediction method, characterized in that, include: Acquire multi-source mineral control information, fuse and standardize the multi-source mineral control information to obtain key mineral control factors, and then construct a mineral control factor matrix; The mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix using the SE-DNN model, and the correlation potential function is constructed based on the mineralization reference value. The spatial weights, feature similarity weights, and kriging coupling coefficients of voxels are calculated using a 3D ellipsoidal kriging model, and then an interaction potential function that integrates multi-dimensional weights is constructed. The COPF global energy function is constructed based on the correlation potential function and the interaction potential function. The mean field approximation is used to infer the posterior distribution and obtain the initial mean and variance of the volume element mineralization variable. The volumetric mineralization variables are trained end-to-end through recurrent convolution, and the learnable parameters are optimized to iteratively update the mean of the volumetric mineralization variables until the convergence condition is met, thus obtaining the three-dimensional mineralization regression prediction results. The process of acquiring multi-source mineral control information, fusing and standardizing the multi-source mineral control information to obtain key mineral control factors, and then constructing a mineral control factor matrix includes: Acquire multi-source mineral control information and extract the 3D coordinates of the target mining area; Data cleaning and Z-score normalization were performed on multi-source mineral control information, and then feature importance was used for evaluation and screening to obtain key mineral control factors. Using the 3D coordinates of the target mining area as the basic spatial feature, the key ore-controlling factors are arranged according to the volume element dimension to obtain the ore-controlling factor matrix; The mineralization control factor matrix is processed using the SE-DNN model to obtain mineralization reference values for volume elements. Based on these reference values, a correlation potential function is constructed, including: A SE-DNN model is constructed, and a mean squared error loss function is built using the mineralization control factor matrix and real mineralization variables as labels. The Adam optimizer is used for training until the loss converges. The mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix through the trained SE-DNN model. Based on the consistency between the predicted and reference values of mineralization, a quadratic function is used to construct the correlation potential function. for: in, For learnable correlation potential weights, For the first Mineralization prediction values of individual elements, Indicates the first Reference value for mineralization of individual elements. Indicates the first Individual element; A 3D ellipsoidal kriging model is used to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of voxels, and then an interaction potential function integrating multi-dimensional weights is constructed, including: The anisotropic spatial kernel is determined by the 3D ellipsoid variogram, and the spatial similarity between adjacent voxels is obtained: in, and They are respectively body elements , 3D coordinates For space kernel bandwidth parameters, Representing adjacent volume elements and The heterogeneous space core between them, in which the volume element As the target element, element As adjacent volumes of the target volume; The Euclidean distance is calculated based on the high-dimensional ore-controlling feature vector of the volume element, and the feature kernel is obtained through Gaussian kernel function transformation. Among them, the characteristic kernel Used to quantify feature domain similarity: in , For body element , High-dimensional ore-controlling feature vectors, Represents Euclidean distance. The feature kernel bandwidth parameter; Construct the covariance matrix based on known mineralized samples. Calculate the covariance vector between the target volume element and the known mineralized sample. The Kriging coupling coefficient was obtained by solving the 3D ellipsoidal Kriging model. for: in, Covariance matrix The inverse matrix, Kriging coupling coefficient Used to quantify the spatial influence weights of adjacent voxels on the target voxel; By fusing the heterogeneous space kernel, the characteristic kernel, and the kriging coupling coefficients, the interaction potential function is obtained. for: in, For learnable interaction potential weights, and For balance coefficient, For the first Mineralization prediction values for individual elements.
2. The three-dimensional mineralization prediction method according to claim 1, characterized in that, The SE-DNN model includes a feature extraction layer and a squeeze-excited attention module. The feature extraction layer consists of multiple fully connected layers, with ReLU as the activation function and the input dimension matching the number of columns in the mining factor matrix. The squeeze-excited attention module aggregates channel information through global average pooling, models inter-channel dependencies through a lightweight gating mechanism, and recalibrates features through output channel weights.
3. The three-dimensional mineralization prediction method according to claim 1, characterized in that, The process of constructing the COPF global energy function based on the correlation potential function and the interaction potential function, and using the mean field approximation for posterior distribution inference to obtain the initial mean and variance of the volumetric mineralization variables includes: A COPF global energy function is constructed based on the correlation potential function and the interaction potential function to determine the prediction accuracy and spatial continuity between volumes. for: in, For the first The correlation potential function of individual elements, Adjacent volume elements and The interaction potential function, Given the known ore-controlling characteristics, Unknown mineralization state The total number of volume elements; The joint posterior distribution of mineralization variables is defined based on the Gibbs distribution. for: in, This is the partition function, used to ensure the normalization of the posterior distribution; Using the mean-field approximation, the joint posterior distribution is decomposed into a product of independent Gaussian marginal distributions: Each factor in the mean-field approximation satisfies By minimizing the KL divergence Iterative optimization of the mean With variance ; Based on the quadratic property of the potential function, the logarithmic probability of the approximated mean field is derived, thus obtaining the mean. Iterative update formula for: The initial mean was determined using the mineralization reference value output by the SE-DNN model. Represents the correlation potential weight. For the related potential, The total number of interaction potential functions, Index for the interaction potential function, For the first Interaction potential weights of each interaction potential function.
4. The three-dimensional mineralization prediction method according to claim 3, characterized in that, The volumetric mineralization variables are optimized through recurrent convolution and end-to-end training to iteratively update the mean of the volumetric mineralization variables until convergence conditions are met, resulting in a three-dimensional mineralization regression prediction result, including: According to the mean With variance Construct a recurrent convolution module, and update the mean by performing neighborhood information transfer, compatibility transformation and fusion update during iteration according to a preset number of iterations; Using the global energy function as the optimization objective, the gradient descent algorithm is employed for backpropagation to simultaneously optimize the parameters and associated potential weights of the SE-DNN model. Interaction potential weights and balance coefficient , And L2 regularization is applied to the parameters of the SE-DNN model, as well as the correlation potential weights. and interaction potential weights Add non-negativity constraints; Calculate the mean of all voxels in two adjacent iterations. The maximum absolute change is determined. If the change is less than a preset threshold or reaches a preset number of iterations, the iteration is terminated, and the ore-forming regression prediction result is obtained.
5. A three-dimensional mineralization prediction device, characterized in that, include: The first module is used to acquire multi-source mineral control information, fuse and standardize the multi-source mineral control information to obtain key mineral control factors, and then construct a mineral control factor matrix. The second module is used to process the mineralization control factor matrix using the SE-DNN model to obtain the mineralization reference value of the volume element, and to construct the correlation potential function based on the mineralization reference value; The third module is used to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of voxels using a 3D ellipsoidal kriging model, and then construct an interaction potential function that integrates multi-dimensional weights. The fourth module is used to construct the COPF global energy function based on the correlation potential function and the interaction potential function, and to use the mean field approximation to infer the posterior distribution, thereby obtaining the initial mean and variance of the volume element mineralization variable. The fifth module is used to optimize the learnable parameters by performing recurrent convolution and end-to-end training on the volume mineralization variables, and to iteratively update the mean of the volume mineralization variables until the convergence condition is met, so as to obtain the three-dimensional mineralization regression prediction results. The process of acquiring multi-source mineral control information, fusing and standardizing the multi-source mineral control information to obtain key mineral control factors, and then constructing a mineral control factor matrix includes: Acquire multi-source mineral control information and extract the 3D coordinates of the target mining area; Data cleaning and Z-score normalization were performed on multi-source mineral control information, and then feature importance was used for evaluation and screening to obtain key mineral control factors. Using the 3D coordinates of the target mining area as the basic spatial feature, the key ore-controlling factors are arranged according to the volume element dimension to obtain the ore-controlling factor matrix; The mineralization control factor matrix is processed using the SE-DNN model to obtain mineralization reference values for volume elements. Based on these reference values, a correlation potential function is constructed, including: A SE-DNN model is constructed, and a mean squared error loss function is built using the mineralization control factor matrix and real mineralization variables as labels. The Adam optimizer is used for training until the loss converges. The mineralization reference value of the volume element is obtained by processing the mineralization control factor matrix through the trained SE-DNN model. Based on the consistency between the predicted and reference values of mineralization, a quadratic function is used to construct the correlation potential function. for: in, For learnable correlation potential weights, For the first Mineralization prediction values of individual elements, Indicates the first Reference value for mineralization of individual elements. Indicates the first Individual element; A 3D ellipsoidal kriging model is used to calculate the spatial weights, feature similarity weights, and kriging coupling coefficients of voxels, and then an interaction potential function integrating multi-dimensional weights is constructed, including: The anisotropic spatial kernel is determined by the 3D ellipsoid variogram, and the spatial similarity between adjacent voxels is obtained: in, and They are respectively body elements , 3D coordinates For space kernel bandwidth parameters, Representing adjacent volume elements and The heterogeneous space core between them, in which the volume element As the target element, element As adjacent volumes of the target volume; The Euclidean distance is calculated based on the high-dimensional ore-controlling feature vector of the volume element, and the feature kernel is obtained through Gaussian kernel function transformation. Among them, the characteristic kernel Used to quantify feature domain similarity: in , For body element , High-dimensional ore-controlling feature vectors, Indicates Euclidean distance. The feature kernel bandwidth parameter; Construct the covariance matrix based on known mineralized samples. Calculate the covariance vector between the target volume element and the known mineralized sample. The Kriging coupling coefficient was obtained by solving the 3D ellipsoidal Kriging model. for: in, Covariance matrix The inverse matrix, Kriging coupling coefficient Used to quantify the spatial influence weights of adjacent voxels on the target voxel; By fusing the heterogeneous space kernel, the characteristic kernel, and the kriging coupling coefficients, the interaction potential function is obtained. for: in, For learnable interaction potential weights, and For balance coefficient, For the first Mineralization prediction values for individual elements.
6. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the three-dimensional mineralization prediction method as described in any one of claims 1-4.
7. A computer-readable storage medium, characterized in that, The storage medium stores a program, which is executed by a processor to implement the three-dimensional mineralization prediction method as described in any one of claims 1-4.