Multi-source data fusion prediction method and system for concealed ore body

Abstract

The invention discloses a hidden ore body multi-source data fusion prediction method and system, and belongs to the technical field of geophysical exploration data processing, and the method comprises the steps: carrying out the standardization of obtained multi-source exploration data, and generating standardized data; quantifying errors of the standardized data in spatial positions and attribute values to generate a multi-source data set; constraining the multi-source data set according to a physical rule to obtain physical constraint embedded data; performing multi-scale feature extraction on the physical constraint embedded data to generate a multi-source feature map; performing cross-source and cross-scale association on the multi-source feature map to generate a fusion feature vector; and based on the fusion feature vector, predicting the mineralization probability, the physical property parameters and the uncertainty distribution, and generating a prediction result map. According to the method, multi-source data standardization, physical law constraint embedding, multi-scale feature fusion and uncertainty quantification technologies are adopted, and a three-dimensional mineralization prediction result with reasonable geology and controllable risk can be generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geophysical exploration data processing, and in particular to a concealed ore body multi-source data fusion prediction method and system. BACKGROUND

[0002] In the field of geophysical exploration, gravity, magnetic, electrical and geochemical methods are often used to detect underground concealed ore bodies, and geological interpretation data provides structural and lithological background information. The existing technology realizes ore body positioning prediction through data conversion and simple superposition analysis.

[0003] Conventional methods usually perform independent processing on multi-source data, and only fuse through spatial superposition or weighted average, without fully considering the physical correlation between data acquisition errors and physical parameters. Feature extraction is usually limited to a single scale, making it difficult to cooperatively identify local anomalies and regional background.

[0004] The defects of the existing technology include fusion deviation caused by projection system and dimensional differences of multi-source data, lack of quantification of spatial position errors and attribute value fluctuation ranges in the data acquisition process, lack of physical law constraint mechanism leading to geological irrationality of the prediction results, and ignoring the influence of multi-scale correlation and uncertainty propagation on the reliability of the prediction. SUMMARY

[0005] To solve the above problems, the present application provides a concealed ore body multi-source data fusion prediction method and system, which adopts multi-source data standardization, physical law constraint embedding, multi-scale feature fusion and uncertainty quantification technology, and can generate three-dimensional mineralization prediction results that are geologically reasonable and risk controllable.

[0006] The above object can be achieved by the following scheme:

[0007] A concealed ore body multi-source data fusion prediction method, comprising: obtaining gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data to obtain multi-source exploration data; standardizing the multi-source exploration data to generate standardized data of a unified dimension; quantifying the errors of the standardized data in spatial position and attribute value to generate a multi-source data set containing uncertainty features representing data quality; constraining the multi-source data set according to physical rules to obtain physically constrained embedded data satisfying geological rules; performing multi-scale feature extraction on the physically constrained embedded data to generate multi-source feature maps containing different spatial range features; performing cross-source and cross-scale correlation on the multi-source feature maps to generate a fusion feature vector integrating information of each data source; and predicting mineralization probability, physical parameters and uncertainty distribution based on the fusion feature vector to generate a prediction result map.

[0008] Optionally, the generating the standardized data of uniform dimension includes: converting the multi-source exploration data to a same spatial grid coordinate system, eliminating projection differences, and generating spatially aligned data; performing normalization processing on attribute values in the spatially aligned data, eliminating dimension differences, and generating normalized data; determining importance weight coefficients of each data source according to a geological background of the exploration area, and performing a weighted fusion operation on the normalized data to generate the standardized data of uniform dimension.

[0009] Optionally, the generating the multi-source data set with uncertainty features representing data quality includes: calculating coordinate offsets of each spatial position point in the standardized data according to positioning accuracy of a data acquisition device; calculating attribute value fluctuation ranges of each spatial position point in the standardized data according to data calibration records and repeated measurement results; and attaching the coordinate offsets and the attribute value fluctuation ranges to corresponding spatial position points to generate the multi-source data set with uncertainty features representing data quality.

[0010] Optionally, the obtaining the physically constrained embedded data satisfying geological rules includes: embedding numerical constraint conditions of the physical rules into the multi-source data set with uncertainty features to generate a constrained data set; and correcting data in the constrained data set that exceeds the numerical constraint conditions to obtain the physically constrained embedded data satisfying geological rules.

[0011] Optionally, the generating the multi-source feature map containing different spatial range features includes: using a feature extraction method with a variable analysis range to process the physically constrained embedded data to obtain feature information corresponding to geological structures and anomalies; and using a spatial weighting mechanism to fuse the feature information to generate the multi-source feature map containing different spatial range features.

[0012] Optionally, the generating the fusion feature vector integrating information of each data source includes: evaluating mutual relationships between different location features and different scale features in the multi-source feature map to obtain cross-source and cross-scale correlation information; using error information representing data quality in the multi-source data set with uncertainty features, and combining the cross-source and cross-scale correlation information, fusing cross-source features in conflict to generate the fusion feature vector integrating information of each data source.

[0013] Optionally, the generating the prediction result map includes: performing mineralization probability regression calculation on the fusion feature vector to generate three-dimensional mineralization probability distribution data; performing physical property parameter inversion calculation on the fusion feature vector to generate three-dimensional physical property parameter estimation data; performing error propagation calculation on the fusion feature vector to generate three-dimensional uncertainty distribution data; and superimposing the three-dimensional mineralization probability distribution data, the three-dimensional physical property parameter estimation data, and the three-dimensional uncertainty distribution data to a same spatial coordinate system to generate the prediction result map.

[0014] Optionally, the method further comprises: performing a physical rule compliance verification operation on the prediction result map to generate an abnormal area identification map; performing a spatial clustering analysis operation on the prediction result map to generate a preset potential target area distribution map; and determining a priority order of subsequent verification work according to an overlapping area of the abnormal area identification map and the preset potential target area distribution map.

[0015] Optionally, the performing of the physical rule compliance verification operation comprises: extracting a property correlation rule based on the physical rule, calculating a preset range of a property parameter, comparing the property parameter estimated value in the prediction result map with the preset range, identifying an out-of-range position that exceeds the preset range, lowering a mineralization probability confidence of the out-of-range position to generate a confidence adjustment result, and extracting all spatial positions with the lowered mineralization probability confidence based on the confidence adjustment result to generate the abnormal area identification map.

[0016] Based on the same inventive concept, the present application also provides a concealed ore body multi-source data fusion prediction system, which comprises: a multi-source data acquisition module for acquiring gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data to obtain multi-source exploration data; a standardization processing module for standardizing the multi-source exploration data to generate standardized data of a unified dimension; an uncertainty quantification module for quantifying errors of the standardized data in spatial position and attribute value to generate a multi-source data set with uncertainty characteristics representing data quality; a physical constraint embedding module for embedding the multi-source data set according to physical rules to obtain physical constraint embedded data satisfying geological rules; a multi-scale feature extraction module for extracting multi-scale features from the physical constraint embedded data to generate multi-source feature maps containing different spatial range features; a cross-source feature fusion module for correlating the multi-source feature maps across sources and scales to generate a fusion feature vector integrating information of each data source; and a three-dimensional mineralization prediction module for predicting mineralization probability, property parameters and uncertainty distribution based on the fusion feature vector to generate a prediction result map.

[0017] Compared with the prior art, the present application has the following advantages:

[0018] Through spatial alignment and attribute normalization processing of multi-source exploration data, the projection difference and dimension heterogeneity of original data are eliminated, a standardized data basis of a unified dimension is established, and the compatibility and calculation consistency of multi-source data fusion are significantly improved.

[0019] The introduction of physical rules for boundary correction of data forces the property parameters to comply with geological rules, effectively suppresses abnormal value interference that violates the common sense of geophysics, and enhances the geological rationality and interpretability of the prediction result.

[0020] Adopt multi-scale feature extraction and cross-source and cross-scale dynamic weighted fusion mechanism, cooperatively capture local mineralization anomaly and regional tectonic background characteristics, and improve the recognition ability of weak signal and complex mineralization mode.

[0021] Synchronously output three-dimensional prediction results of mineralization probability, physical parameters and uncertainty distribution, quantify the prediction confidence range, provide risk controllable decision basis for drilling target area optimization, and optimize exploration engineering deployment efficiency.

[0022] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application can be realized and obtained by the structure indicated in the specification, claims and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0024] Figure 1 is a flowchart of a multi-source data fusion prediction method for concealed ore body of an embodiment of the present application.

[0025] Figure 2 is a multi-source exploration data space alignment and normalization schematic diagram of an embodiment of the present application.

[0026] Figure 3 is a comparison diagram of attribute values before and after physical rule constraint correction of an embodiment of the present application.

[0027] Figure 4 is a three-dimensional space mineralization probability distribution diagram of an embodiment of the present application.

[0028] Figure 5 is a structure schematic diagram of a multi-source data fusion prediction system for concealed ore body of an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical scheme in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0030] Referring to Figure 1 One embodiment of the present application proposes a concealed ore body multi-source data fusion prediction method, which adopts multi-source data standardization, physical law constraint embedding, multi-scale feature fusion and uncertainty quantification technology, and can generate three-dimensional mineralization prediction results that are geologically reasonable and risk controllable.

[0031] The method of the embodiment specifically comprises:

[0032] Obtaining multi-source exploration data by acquiring gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data;

[0033] Standardizing the multi-source exploration data to generate standardized data of unified dimension;

[0034] Quantifying the errors of the standardized data in spatial position and attribute value to generate a multi-source data set containing uncertainty features representing data quality;

[0035] Constraining the multi-source data set according to physical rules to obtain physically constrained embedded data satisfying geological rules;

[0036] Extracting multi-scale features from the physically constrained embedded data to generate multi-source feature maps containing different spatial range features;

[0037] Correlating the multi-source feature maps across sources and scales to generate a fusion feature vector integrating information of each data source;

[0038] Based on the fusion feature vector, predicting mineralization probability, physical parameters and uncertainty distribution to generate a prediction result map.

[0039] Specifically, by integrating gravity, magnetic, electrical, geochemical and geological interpretation multi-source exploration data, first, the standardized data is generated by eliminating the data space projection and dimensional difference; then the spatial position offset and attribute value fluctuation range in the data collection process are quantified to form a data set containing uncertainty characteristics; the data is corrected to ensure that the physical parameters meet the geological rules; multi-scale feature extraction technology is used to capture local anomalies and regional structural features, and the multi-scale response is fused through a spatial weighting mechanism; based on the mutual correlation analysis between features and dynamic weighting of data quality indicators, cross-source and cross-scale feature fusion is realized; finally, the integrated feature vector is used to simultaneously predict the three-dimensional mineralization probability, physical parameters and their uncertainty distribution, and the comprehensive prediction result map is generated. The present application significantly improves the geological rationality and reliability of concealed ore body prediction, effectively overcomes the fusion deviation caused by the heterogeneity of multi-source data; the physical law embedding mechanism suppresses abnormal values that violate geological common sense, enhancing the interpretability of the prediction results; the multi-scale feature fusion technology cooperatively identifies local mineralization anomalies and regional metallogenic background, improving the weak signal detection capability; the uncertainty quantification and propagation analysis clearly define the confidence range of the prediction results, providing a risk-controllable target area optimization basis for drilling verification, and optimizing the exploration decision-making efficiency as a whole.

[0040] Optionally, generating standardized data of uniform dimension includes:

[0041] Converting the multi-source exploration data to the same spatial grid coordinate system, eliminating the projection difference, and generating spatially aligned data;

[0042] Performing normalization processing on the attribute values in the spatially aligned data to eliminate dimensional differences and generate normalized data;

[0043] Determining the importance weight coefficient of each data source according to the geological background of the exploration area, and performing a weighted fusion operation on the normalized data to generate standardized data of uniform dimension.

[0044] Specifically, first, gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data are obtained as multi-source exploration data, which come from different exploration equipment or geological surveys and may have different spatial reference systems and attribute units. To generate standardized data of uniform dimension, the first step is to convert the multi-source exploration data to the same spatial grid coordinate system, which is a pre-set regular grid system, such as the Universal Transverse Mercator projection coordinate system. Projection conversion is performed using geographic information system tools to eliminate the projection difference between different data sources, ensuring that all data points are located on the same spatial position grid, thereby generating spatially aligned data. The second step is to perform normalization processing on the attribute values in the spatially aligned data. Normalization refers to scaling different dimensional attribute values to a unified dimensionless range, for example, using the min-max normalization method, the formula is:

[0045] ,

[0046] in, These are the normalized attribute values. Represents the original attribute value. and These represent the minimum and maximum attribute values ​​of the data source at all spatial locations within the exploration area, respectively. These values ​​are calculated by traversing spatially aligned data. This operation eliminates dimensional differences between gravity, magnetic, electrical, and geochemical data, for example, converting milligal, nanot, and ohmmeter units to values ​​between 0 and 1 to generate normalized data. Figure 2 As shown, all multi-source exploration data exhibit a normalized distribution within a unified, regular spatial grid. The third step is to determine the importance weight coefficients of each data source based on the geological background of the exploration area. The geological background includes regional rock types, structural features, and known mineralization information, provided by a geological expert knowledge base or historical exploration reports. The weight coefficients of each data source are determined using expert scoring, the Analytic Hierarchy Process (AHP), or statistical analysis. The AHP method includes: constructing a hierarchical structure of data source importance; generating a criterion layer comparison matrix based on expert judgment; calculating weights using the eigenvector method and verifying consistency. The final output is a set of coefficients whose weights sum to 1. For example, magnetic data is assigned a higher weight in igneous rock regions and geochemical data is assigned a higher weight in sedimentary rock regions, ensuring that the sum of the weight coefficients is 1. Then, a weighted fusion operation is performed on the normalized data, using the following formula:

[0047] ,

[0048] in, Represents standardized data. Indicates the first Weight coefficients of each data source, Indicates the first Normalized data from multiple data sources Given the total number of data sources, this operation sums the weighted data at points with the same spatial location. Since the normalized data is dimensionless and the weighting coefficients are pure numerical values, the fusion process is dimensionless, generating standardized data with unified dimensions. Through the above operations, the overall method achieves spatial and attribute consistency of multi-source data, eliminates the heterogeneity of the original data, lays the foundation for subsequent uncertainty quantification and feature extraction, improves the reliability and accuracy of concealed ore body prediction, reduces error propagation caused by data mismatch, and enhances the geological rationality of the prediction results.

[0049] Optionally, the multi-source dataset containing uncertain features that characterize data quality includes:

[0050] According to the positioning accuracy of the data acquisition equipment, the coordinate offset of each spatial position point in the standardized data is calculated;

[0051] According to the data calibration record and the repeated measurement result, the attribute value fluctuation range of each spatial position point in the standardized data is calculated;

[0052] The coordinate offset and the attribute value fluctuation range are attached to the corresponding spatial position point to generate a multi-source data set with uncertainty characteristics representing data quality.

[0053] Specifically, first, starting from the generated standardized data of uniform dimension, the standardized data has been processed by spatial alignment and attribute normalization to ensure that all data points are located in the same spatial grid coordinate system and the attribute values are dimensionless values. To generate a multi-source data set with uncertainty characteristics representing data quality, first, according to the positioning accuracy of the data acquisition equipment, the coordinate offset of each spatial position point in the standardized data is calculated, and the positioning accuracy refers to the inherent position error range of the equipment during data acquisition, for example, the horizontal accuracy of a GPS receiver is ±0.5 meters or the angle measurement accuracy of a total station is ±0.1 degrees. These accuracy values are directly extracted from the technical specification book or calibration certificate provided by the equipment manufacturer; for each spatial position point, the coordinate offset directly uses the positioning accuracy value of the equipment, for example, if a point is collected using GPS, its coordinate offset is set to 0.5 meters, indicating that the true position of the point may fluctuate within this offset range. Then, according to the data calibration record and the repeated measurement result, the attribute value fluctuation range of each spatial position point in the standardized data is calculated, the data calibration record refers to the error data recorded during the periodic verification of the equipment, for example, the calibration report of an electrical method instrument shows that the measurement error is ±2%, and the repeated measurement result refers to the attribute value data set collected at the same position multiple times, and the standard deviation is calculated by statistical analysis of these values to quantify the fluctuation; for each spatial position point, the attribute value fluctuation range is calculated using the standard deviation formula, the formula is:

[0054]

[0055] ​​​​​​​​​​​and attribute value fluctuation range Additional to the corresponding spatial position points in the standardized data, the specific operation is to add two new attribute fields for each point when storing data, respectively storing and value, thereby forming a multi-source data set containing uncertainty characteristics, which not only retains the original standardized data, but also contains quantified spatial position and attribute value error information. This method effectively quantifies the inherent uncertainty of multi-source exploration data, enhances the reliability of subsequent feature extraction and fusion process, avoids prediction bias caused by data quality fluctuations, improves the accuracy of concealed ore body positioning and the reliability of geological interpretation, and provides basic support for error propagation analysis.

[0056] Optionally, the physical constraint embedded data satisfying the geological rule comprises:

[0057] Embedding the numerical constraint condition set by the physical rule into the multi-source data set containing uncertainty characteristics to generate a constrained data set;

[0058] Correcting the data in the constrained data set that exceeds the numerical constraint condition to obtain the physical constraint embedded data satisfying the geological rule.

[0059] Specifically, starting from the multi-source data set containing uncertainty characteristics representing data quality, which contains standardized attribute values and additional spatial position coordinate offsets and attribute value fluctuation ranges, first, a preset physical rule is obtained, which is based on the numerical constraint condition set by the geological physical rule, for example, the density parameter corresponding to the gravity data, the magnetic susceptibility parameter corresponding to the magnetic method data and other physical parameters, define the reasonable value range of these parameters, the minimum value and the maximum value are obtained through geological expert knowledge base or historical exploration data statistics, to ensure compliance with the actual rock physical law; then embed the numerical constraint condition into the multi-source data set containing uncertainty characteristics, and the specific operation is to add a constraint range field for each spatial position point of the physical parameter, store the corresponding and , to generate a constrained data set, which not only retains the original attribute value, coordinate offset and attribute value fluctuation range, but also contains constraint boundary information; then correct the data in the constrained data set that exceeds the numerical constraint condition, and the correction method adopts a boundary adjustment strategy, the formula is:

[0060] ,

[0061] wherein, represents the original attribute value in the constrained data set, which is directly read from the data set,

[0062] and represent the minimum and maximum values of the numerical constraint obtained from the embedded constraint range field, the correction operation ensures that all attribute values are limited in the closed interval of to , since the attribute values have been normalized to dimensionless values, and are also dimensionless, the correction process is dimensionless conflict; the corrected data that satisfy the physical constraint embedded data of geological rules, the data inherits the uncertainty characteristics of the original data set but the attribute values meet the physical law constraints. As shown in Figure 3 , the attribute values falling outside the constraint interval are corrected to the interval boundary, which highlights the technical core of the boundary adjustment strategy. This method effectively eliminates the abnormal values in the data that violate the geological rules, enhances the geological consistency of multi-source data, provides more reliable input for subsequent feature extraction, improves the stability and geological interpretation rationality of the concealed ore body prediction model, and reduces the prediction error caused by physical contradictions.

[0063] Optionally, the generating of the multi-source feature map containing different spatial range features comprises:

[0064] using a feature extraction method with variable analysis range to process the physical constraint embedded data to obtain feature information corresponding to geological structures and anomalies;

[0065] using a spatial weighting mechanism to fuse the feature information to generate a multi-source feature map containing different spatial range features.

[0066] Specifically, starting from the generated physical constraint embedded data that satisfies the geological rules, the data has been corrected for abnormal values by physical rules to ensure that all attribute values meet the geological physical range, while preserving the spatial position coordinate offset and attribute value fluctuation range information. To generate a multi-source feature map containing different spatial range features, a variety of feature extraction methods with variable analysis range can be used to achieve, for example, a multi-scale analysis method based on wavelet transform or a convolutional neural network (CNN) using a multi-scale convolution kernel to process the physical constraint embedded data. The method uses a multi-scale sliding window technology, and the window radius r is preset to small scale , medium scale and large scale three levels, for example meters, meters, meters, for each spatial position point , the statistical characteristics of the attribute values in the window with different radii centered on are calculated respectively, including the arithmetic mean , , and standard deviation , , The formula is:

[0067] ,

[0068] ,

[0069] in It is the arithmetic mean of the attribute values ​​within the window. Standard deviation Indicates the number of spatial points within the window. Indicates the first The attribute values ​​of each point are directly read from the physically constrained embedded data. This operation extracts feature information corresponding to local geological structures and regional anomalies. Small-scale windows capture the details of mineralized bodies, while large-scale windows reflect tectonic background features. Next, a spatial weighting mechanism is used to fuse the feature information. First, the fusion weights of features at each scale are calculated. , , It is inversely proportional to the average value of the attribute value fluctuation range within the corresponding window, as shown in the formula:

[0070] ,

[0071] ,

[0072] in, Indicates the first in the window The range of attribute value fluctuations for each point is obtained from the dataset. Indicates the number of spatial points within the window. This represents the average value of the attribute's fluctuation range. For scale The fusion weight, To prevent division by zero, a small constant is set to 1e-5. Then, the weights are normalized so that the sum is 1. Finally, multi-scale features are weighted and fused to generate location points. The final feature vector This vector contains feature responses from different spatial ranges, forming a multi-source feature map covering the entire exploration area. This method effectively integrates multi-scale geological information, adaptively highlights the contribution of reliable data, enhances the ability to identify the spatial distribution patterns of concealed ore bodies, improves the accuracy of the prediction model in coordinating the perception of local anomalies and regional background, and provides highly discriminative input features for cross-source feature fusion.

[0073] Optionally, generating a fused feature vector that integrates information from various data sources includes:

[0074] The relationships between features at different locations and at different scales in the multi-source feature map are evaluated to obtain cross-source and cross-scale association information.

[0075] The error information representing data quality in the multi-source data set with uncertainty features is used, and the cross-source and cross-scale correlation information is combined to fuse the conflicting cross-source features, to generate a fusion feature vector integrating information of each data source.

[0076] Specifically, starting from the generated multi-source feature map, the feature map contains feature responses of each spatial position point at different spatial scales, and attribute value fluctuation range information of the original data points is obtained from the generated multi-source data set with uncertainty features. To generate a fusion feature vector integrating information of each data source, first, the mutual relationship between different position features and different scale features in the multi-source feature map is evaluated, specifically, for any two feature vectors in the feature map and , the global cross-correlation value is calculated, and the formula is:

[0077] ,

[0078] wherein represents the total number of spatial position points in the exploration area, represents the feature at position ,

[0079] represents the feature at position , which is directly read from the multi-source feature map, and represent the average values of feature and feature at all position points, which are obtained by calculating the arithmetic mean values of feature and feature at all position points, and the operation generates a symmetric cross-correlation matrix , and the element of the symmetric cross-correlation matrix takes a value between -1 and 1, as the cross-source and cross-scale correlation information. Then, the error information representing data quality in the multi-source data set with uncertainty features is used, specifically, the attribute value fluctuation range corresponding to each position point is directly obtained from the data set, and the cross-correlation matrix is combined to fuse the conflicting cross-source features. For the feature vector of position point , the conflict adjustment weight is calculated, and the formula is:

[0080] ,

[0081] in Represents the total number of features. Indicates position The range of attribute value fluctuations, Representation of features With features The cross-correlation value from the matrix The weighting formula, obtained from [source], ensures high data quality. Features that are small and highly correlated with other features receive greater weight; finally, the position... All features are weighted and fused to generate a comprehensive feature value. The combined feature values ​​of all location points constitute the fused feature vector. This method effectively utilizes data quality indicators and feature correlations to dynamically adjust the fusion weights, significantly reducing interference caused by multi-source data conflicts, enhancing the geological consistency and reliability of the feature vector, and providing high-confidence input features for the three-dimensional prediction of concealed ore bodies.

[0082] Optionally, the generated prediction result map includes:

[0083] Mineralization probability regression calculation is performed on the fused feature vector to generate three-dimensional mineralization probability distribution data;

[0084] Perform physical property parameter inversion calculations on the fused feature vectors to generate three-dimensional physical property parameter estimation data;

[0085] Error propagation calculations are performed on the fused feature vectors to generate three-dimensional uncertainty distribution data;

[0086] The three-dimensional mineralization probability distribution data, the three-dimensional physical property parameter estimation data, and the three-dimensional uncertainty distribution data are superimposed on the same spatial coordinate system to generate the prediction result map.

[0087] Specifically, starting with the generated fused feature vector, which integrates information from various data sources and includes feature values ​​of spatial location points, mineralization probability regression is first performed on the fused feature vector. A pre-trained probabilistic regression model is used, which can be Logistic Regression, Support Vector Machine (SVM), or a complex Deep Neural Network (DNN). Its core function is to map the fused feature vector to mineralization probability values. Before deployment, this model has learned mineralization patterns from labeled samples through supervised learning. Its technical essence is to fit the complex relationship between multi-source features and mineralization state through a nonlinear function. The formula is:

[0088] ,

[0089] in Indicates the probability of mineralization. denotes activation function, denotes the dimensional feature value in the fusion feature vector, and is a model parameter, which is trained by known mineralization point and non-mineralization point samples, as shown in the formula: Figure 4 The calculation generates a mineralization probability value for each location point in the three-dimensional space grid, forming a three-dimensional mineralization probability distribution data. Secondly, the fusion feature vector is executed to perform physical parameter inversion calculation, and the physical parameters include density, magnetic susceptibility, and resistivity. A pre-trained multivariate linear inversion model is used to establish the model by supervised training of the drilling core physical property measurement data covering the exploration area and the feature vector corresponding to the spatial position. The mathematical essence is to construct a multivariate linear relationship between the fusion features and the target physical parameters. Specifically, the model takes the fusion feature vector generated by cross-source and cross-scale correlation as input, uses the coefficient matrix obtained by pre-training to linearly weight and combine the features, and simultaneously outputs the estimated value of multiple physical parameters at each grid location in the three-dimensional space. The formula is:

[0090] ,

[0091] wherein denotes the estimated value of the th physical parameter, denotes the dimensional feature value in the fusion feature vector, and is an inversion model parameter, which is trained by drilling core physical property measurement data and corresponding position feature vectors. Then, error propagation calculation is performed. First, the attribute value fluctuation range of the location point is extracted from the generated feature data set containing uncertainty, and the variance of the mineralization probability estimate is calculated. The formula is:

[0092] ,

[0093] wherein denotes the mineralization probability, is the probability regression model coefficient, is the attribute value fluctuation range, and the variance of the physical parameter estimate is:

[0094] ,

[0095] wherein To invert the model parameters, the calculation generates the mineralization probability variance and physical property parameter variance of each position point, forming a three-dimensional uncertainty distribution data. Finally, the three-dimensional mineralization probability distribution data, three-dimensional physical property parameter estimation data and three-dimensional uncertainty distribution data are superimposed according to the same spatial grid coordinates, and the prediction result graph is generated through the three-dimensional visualization system, wherein each voxel in the graph contains the mineralization probability value, the physical property parameter value and the corresponding uncertainty value. The method realizes the collaborative prediction of mineralization probability, physical property parameter and uncertainty, and the generated result graph can intuitively display the spatial distribution of the prospecting target area, while quantifying the reliability of the prediction result, providing a reliable basis for drilling verification, and significantly improving the exploration efficiency of concealed ore bodies.

[0096] Optionally, the method further comprises:

[0097] performing a physical law compliance verification operation on the prediction result graph to generate an abnormal area identification graph;

[0098] performing a spatial clustering analysis operation on the prediction result graph to generate a target area distribution graph of preset potential;

[0099] determining the priority order of subsequent verification work according to the overlapping area of the abnormal area identification graph and the target area distribution graph of preset potential.

[0100] Specifically, starting from the generated prediction result graph, the graph contains three-dimensional mineralization probability distribution data, three-dimensional physical property parameter estimation data and three-dimensional uncertainty distribution data. First, a physical law compliance verification operation is performed to extract the physical property correlation rule, which stores the reasonable range boundary of the geological physical property parameter, for example, the density range 2.5-3.0 g / cm³ or the magnetic susceptibility range SI, these boundary values and are derived from regional geological research reports or rock physics databases, based on statistical analysis of the physical property test results of hundreds of core samples of specific lithology (such as granite, diorite) in the region, taking the 95% confidence interval as the reasonable range; the physical property parameter estimation value of each spatial position point in the prediction result graph is compared with the and of the corresponding parameter, when or , the position is marked as out-of-limit; the confidence degree of the mineralization probability of the out-of-limit position is lowered, and a linear penalty function is used to adjust the mineralization probability value , wherein , takes the closer one of or , , the operation makes the degree of over-limit greater, the probability value is reduced more; all the adjusted position points are extracted to generate an abnormal area identification map, which is a binary raster data, the over-limit position is marked as 1, and the rest is 0. Then, a spatial aggregation analysis operation is performed, the three-dimensional morphological closing operation is used to process the mineralization probability distribution data, the structure element is a sphere with a radius of 200 meters, and adjacent high-probability areas are connected; the volume-weighted average probability value of each connected region is calculated , the formula is:

[0101] ,

[0102] wherein, is the volume of the voxel in the connected region, is the voxel mineralization probability value; the regions with and cubic meters are screened, and a target area distribution map of preset potential is generated. Finally, according to the spatial overlap relationship between the abnormal area identification map and the target area distribution map of the preset potential, the target area is divided into two categories: the target area without overlap with the abnormal area is marked as the first priority verification area, and the target area overlapping with the abnormal area is marked as the second priority verification area, and a target area spatial distribution list with a priority label is output. This method can realize the geological rationality verification and target area optimization grading of the prediction result, effectively identify and inhibit the false mineralization signal of the contradictory region, highlight the spatially continuous high-confidence target area, provide a scientific basis for drilling engineering deployment, and significantly reduce the exploration decision risk.

[0103] Optionally, the performing a physical law compliance verification operation comprises:

[0104] extracting a physical property correlation rule based on the physical rule, and calculating a preset range of the physical property parameter;

[0105] comparing the physical property parameter estimated value in the prediction result map with the preset range, and identifying an over-limit position that exceeds the preset range;

[0106] lowering the mineralization probability confidence of the over-limit position, and generating a confidence adjustment result;

[0107] based on the confidence adjustment result, extracting all spatial positions with lowered mineralization probability confidence, and generating an abnormal area identification map.

[0108] Specifically, starting from the generated prediction result map, the map contains three-dimensional mineralization probability distribution data and three-dimensional physical property parameter estimation data. First, the physical property correlation rule is extracted based on the physical rule, the rule stores the physical property parameter range boundary value provided by the regional geological research report as the preset range, including the minimum density , the maximum density , the minimum magnetic susceptibility , the maximum value of magnetic susceptibility , etc., the range of each parameter is obtained by statistically measuring the properties of the known geological body. Then the estimated value of the property parameter of each spatial position point in the prediction result map is compared with the preset range of the corresponding parameter, and when or , the position is determined to be out of limit. For each out-of-limit position, the out-of-limit degree of the position is calculated, and the formula is:

[0109] ,

[0110] wherein represents the estimated value of the property parameter of the position, which is directly read from the prediction result map, takes the boundary value of or , which is closer, represents the length of the reasonable range interval, that is . Then the confidence degree of the mineralization probability is lowered, and a linear penalty function is used to adjust the mineralization probability value , wherein represents the original mineralization probability value, which is obtained from the prediction result map. The operation ensures that the greater the out-of-limit degree, the greater the reduction in the probability value. After generating the confidence degree adjustment result, all spatial position points of are extracted to form a binary raster data, and the adjusted positions are marked as 1 and the unadjusted positions are marked as 0. Finally, an anomaly area identification map is output. The method can realize dynamic checking of the prediction result and the geophysical law, automatically identify and suppress unreliable areas of the property parameter anomaly, significantly improve the geological reasonableness of the mineralization probability distribution, provide more reliable basis for target area optimization, and reduce the risk of exploration engineering deployment.

[0111] Based on the same inventive concept, as shown in Figure 5 , the present application also provides a multi-source data fusion prediction system for concealed ore bodies, which comprises:

[0112] a multi-source data acquisition module for acquiring multi-source exploration data obtained by gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data;

[0113] a standardization processing module for standardizing the multi-source exploration data to generate standardized data of uniform dimension;

[0114] an uncertainty quantification module for quantifying the error of the standardized data in spatial position and attribute value to generate a multi-source data set with uncertainty characteristics representing data quality;

[0115] ​a physical property constraint embedding module, configured to constrain the multi-source data set according to physical rules to obtain physical constraint embedding data meeting geological rules;

[0116] a multi-scale feature extraction module, configured to perform multi-scale feature extraction on the physical constraint embedding data to generate multi-source feature maps containing features of different spatial ranges;

[0117] a cross-source feature fusion module, configured to perform cross-source and cross-scale correlation on the multi-source feature maps to generate a fusion feature vector integrating information of all data sources;

[0118] a three-dimensional mineralization prediction module, configured to predict mineralization probability, physical property parameters and uncertainty distribution based on the fusion feature vector to generate a prediction result map.

[0119] To verify the feasibility of the present application in implementation, the present application is applied to a concealed ore body exploration project in a copper-gold polymetallic ore concentration area in eastern China. The surface overburden in the area is relatively thick, and the known mineralization clues are limited, so it is difficult to accurately locate the deep concealed ore body by using traditional exploration methods. The project aims to use the method and system of the present application to fuse multi-source geophysical and geochemical data in the area to delineate drilling verification target areas.

[0120] In the present embodiment, the exploration project team deploys the "concealed ore body multi-source data fusion prediction system" of the present application. First, the multi-source exploration data covering about 50 square kilometers of the entire exploration area are obtained by using the multi-source data acquisition module, including: 1:50000 scale gravity exploration data (unit: milligal), airborne magnetic exploration data (unit: nT), magnetotelluric sounding (MT) data (unit: ohm meter), high-density soil geochemical survey data covering the entire area (copper, gold and other element contents, unit: ppm), and existing regional geological interpretation maps.

[0121] The system first processes the above multi-source exploration data by using the standardization processing module. All the data are converted to the unified CGCS2000 spatial grid coordinate system to generate spatially aligned data. Subsequently, the attribute values in the spatially aligned data, such as gravity values, magnetic anomaly values, resistivity values and element contents, are subjected to minimum-maximum normalization processing to scale them to the dimensionless range of 0 to 1. According to the geological background of the area being a porphyry-skarn type copper-gold mine, the importance weight coefficients of each data source are determined by geologists, for example, the magnetic data and the electrical data closely related to mineralization are given higher weights (0.4 and 0.3, respectively) , ), the gravity data and the geochemical data are given lower weights (0.2 and 0.1, respectively) , ), and finally the standardized data of unified dimensions are generated by weighted fusion.

[0122] Next, the uncertainty quantification module processes the normalized data. According to the records of the data acquisition equipment, the GPS positioning accuracy of the ground gravity measurement points is ±1.2 meters, which is used as the coordinate offset of the spatial position. Through statistical analysis of the repeated measurement data of the exploration base points, the fluctuation range of the attribute values of each data source is calculated, for example, the standard deviation of the gravity measurement value of a certain area is 0.08 milligal, which is normalized as the fluctuation range of the attribute value of the area. These uncertainty information is attached to the normalized data set.

[0123] The physical law constraint embedding module corrects the data based on physical rules. The rules define the reasonable range of rock physical parameters related to mineralization in this area, for example, the density range of the rock mass related to copper mineralization should be 2.75-3.40 g / cm³. The system performs boundary correction on all points in the data set whose corresponding density values exceed this range, ensuring that the data conforms to the geological rules.

[0124] Subsequently, the multi-scale feature extraction module uses a three-level sliding window (small-scale radius 100 meters, medium-scale radius 400 meters, large-scale radius 1200 meters) to process the physically constrained embedded data, extracting feature information of different spatial ranges, such as local high-density anomalies, medium-scale structure-controlled low-resistivity anomaly zones, and large-scale rock mass boundaries, and performing spatial weighting according to the uncertainty of the data in each window to generate multi-source feature maps.

[0125] The cross-source feature fusion module evaluates the mutual relationship of the multi-source feature maps. Analysis shows that there is a strong spatial positive correlation (cross-correlation value ) between the medium-scale magnetic anomaly high-value area and the low resistivity area, indicating that this feature combination may indicate sulfide enrichment. The system uses this correlation information and the uncertainty of the data to weight and fuse the features, generating a fusion feature vector that integrates information from all data sources.

[0126] Finally, the three-dimensional mineralization prediction module generates mineralization probability, physical parameter (density, magnetic susceptibility, resistivity) and uncertainty distribution maps of the underground three-dimensional space of the exploration area based on the fusion feature vector through a pre-trained regression and inversion model. The prediction result map clearly delineates three high mineralization probability (>0.7) areas, marked as target area , target area and target area .

[0127] To further verify and optimize the target areas, the system performs physical law compliance verification and spatial aggregation analysis on the prediction result map. In the verification, it is found that target area The predicted density value locally reaches 3.8 g / cm3, which exceeds the upper limit of the known rock density in this area (3.4 g / cm3), and the system determines that there is a physical contradiction in this area, generates an abnormal area identification map, and accordingly reduces the target area The mineralization probability confidence of the target area and the target area are determined as the level priority verification area, and the target area is downgraded to the level priority verification area.

[0128] Table 1: Example of multi-source data standardization and uncertainty quantification table

[0129] Data point ID Original data type Original value Normalized value Weight coefficient Coordinate offset (m) Attribute value fluctuation range (after normalization) P001 Gravity 15.2 mGal 0.65 0.20 ±1.2 0.04 P001 Magnetic method 350 nT 0.82 0.35 ±2.5 0.02 P001 Electric method 80 Ω·m 0.21 0.35 ±5.0 0.05 P002 Gravity 12.8 mGal 0.43 0.20 ±1.2 0.05 P002 Magnetic method 120 nT 0.35 0.35 ±2.5 0.03 P002 Electric method 250 Ω·m 0.75 0.35 ±5.0 0.06

[0130] Table 2: Predicted target area characteristics and preferred verification results table

[0131] Target area number Predicted central depth (m) Average mineralization probability (original) Average predicted density (g / cm³) Physical law compliance Final priority Target area A 650 0.82 3.15 Compliant Class A Target area B 800 0.78 3.65 Non-compliant (over limit) Class B Target area C 550 0.75 3.20 Compliant Class A Comparative method target area 780 0.79 - Not verified High

[0132] Table 3: Comparison of target area verification effects table

[0133] Evaluation index The method of the present application Traditional single anomaly stacking method Effectiveness improvement Number of Class A target areas delineated 2 3 - False target area identification rate 100% (target area B identified) 0% 100% improvement Agreement degree of predicted physical properties and drilling core 92% About 70% 22% improvement Reliability of drilling verification recommendations High Medium Significant improvement

[0134] The above Tables 1 to 3 record the actual application data of the present application in the hidden ore exploration project, and details the performance of the system in data processing, feature fusion, three-dimensional prediction and target area optimization, etc. As can be seen from Table 1, the system effectively unifies the exploration data of different sources and different dimensions into a standardized framework, and quantifies the spatial and attribute uncertainty of each data point, laying a foundation for subsequent reliable fusion. Table 2 clearly shows the target area optimization process of the present application. The system not only predicts the high mineralization probability area, but also successfully identifies the physical contradiction in the target area by physical law compliance verification. If the traditional method is used, only the high probability anomaly superposition, the target area will be considered as a preset potential target, and the present application reduces it through intelligent verification, effectively avoiding the investment of expensive drilling cost in unreasonable target area. The comparison data in Table 3 further confirms the superiority of the present application. Through physical law constraint and uncertainty quantification, the present application improves the identification rate of false target area to 100%, and the prediction physical property is highly consistent with the core measured data of subsequent drilling verification, reaching 92%, which is significantly better than the traditional method. This shows that the present application can generate more reliable and lower risk exploration target area with geological significance, greatly improving the success rate and efficiency of hidden ore body exploration.

[0135] It should be noted that the electrical connection between the various units described above does not necessarily indicate a direct connection, and the indirect connection mode can also be applied to the embodiments of the present application as long as the purpose of the present application is achieved. The above is only an exemplary embodiment of the present application, and cannot limit the scope of the present application.

[0136] That is, any equivalent changes and modifications made in accordance with the teachings of the present application are still within the scope of the present application. Other embodiments of the present application will be readily apparent to those skilled in the art upon considering the description and practice of the true principles of the disclosure. The present application is intended to cover any variations, uses, or adaptive changes of the present application that follow the general principles of the present application and include common knowledge or conventional techniques in the art that are not described in the present application.

Claims

1. A multi-source data fusion prediction method for concealed ore bodies, characterized in that, The method comprises: obtaining multi-source exploration data by acquiring gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data; standardizing the multi-source exploration data to generate standardized data of uniform dimensions; quantifying errors of the standardized data in spatial position and attribute value to generate a multi-source data set with uncertainty features representing data quality; constraining the multi-source data set according to physical rules to obtain physically constrained embedded data satisfying geological rules; extracting multi-scale features from the physically constrained embedded data to generate multi-source feature maps containing different spatial range features; cross-source and cross-scale correlation of the multi-source feature maps to generate a fusion feature vector integrating information from each data source; based on the fusion feature vector, predicting mineralization probability, physical parameters and uncertainty distribution to generate a prediction result map.

2. The method according to claim 1, characterized in that, The generation of standardized data of uniform dimensions comprises: converting the multi-source exploration data to the same spatial grid coordinate system to eliminate projection differences and generate spatially aligned data; performing normalization processing on attribute values in the spatially aligned data to eliminate dimensional differences and generate normalized data; determining the importance weight coefficients of each data source according to the geological background of the exploration area, and performing weighted fusion operation on the normalized data to generate standardized data of uniform dimensions.

3. The method according to claim 2, characterized in that, The generation of a multi-source data set with uncertainty features representing data quality comprises: calculating the coordinate offset of each spatial position point in the standardized data according to the positioning accuracy of the data acquisition equipment; calculating the attribute value fluctuation range of each spatial position point in the standardized data according to the data calibration records and repeated measurement results; attaching the coordinate offset and the attribute value fluctuation range to the corresponding spatial position point to generate a multi-source data set with uncertainty features representing data quality.

4. The method according to claim 3, characterized in that, The generation of physically constrained embedded data satisfying geological rules comprises: embedding the numerical constraint conditions set by the physical rules into the multi-source data set with uncertainty features to generate a constrained data set; correcting the data in the constrained data set that exceeds the numerical constraint conditions to obtain physically constrained embedded data satisfying geological rules.

5. The method according to claim 4, characterized in that, The generation of multi-source feature maps containing different spatial range features comprises: using a feature extraction method with variable analysis range to process the physically constrained embedded data to obtain feature information corresponding to geological structures and anomalies; using a spatial weighting mechanism to fuse the feature information to generate multi-source feature maps containing different spatial range features.

6. The method according to claim 5, characterized in that, The generation of a fusion feature vector integrating information from each data source comprises: evaluating the mutual relationship between different position features and different scale features in the multi-source feature maps to obtain cross-source and cross-scale correlation information; using error information representing data quality in the multi-source data set with uncertainty features, and combining the cross-source and cross-scale correlation information, to fuse cross-source features with conflicts to generate a fusion feature vector integrating information from each data source.

7. The method according to claim 6, characterized in that, The generation of a prediction result map comprises: performing mineralization probability regression calculation on the fusion feature vector to generate three-dimensional mineralization probability distribution data; Performing physical property parameter inversion calculation on the fusion feature vector to generate three-dimensional physical property parameter estimation data; Performing error propagation calculation on the fusion feature vector to generate three-dimensional uncertainty distribution data; Superimposing the three-dimensional mineralization probability distribution data, the three-dimensional physical property parameter estimation data and the three-dimensional uncertainty distribution data to the same spatial coordinate system to generate the prediction result map.

8. The method according to claim 7, characterized in that, Also includes: Performing physical rule compliance verification operation on the prediction result map to generate anomaly area identification map; Performing spatial aggregation analysis operation on the prediction result map to generate target area distribution map of preset potential; According to the overlapping area of the anomaly area identification map and the target area distribution map of preset potential, determine the priority order of subsequent verification work.

9. The method according to claim 8, characterized in that, The execution of the physical rule compliance verification operation includes: Based on the physical rule, extract the physical property association rule, calculate the preset range of the physical property parameter; Compare the physical property parameter estimation value in the prediction result map with the preset range, identify the out-of-limit position that exceeds the preset range; Lower the mineralization probability confidence of the out-of-limit position, generate confidence adjustment result; Based on the confidence adjustment result, extract all spatial positions with lowered mineralization probability confidence to generate anomaly area identification map.

10. A system for predicting a concealed ore body by multi-source data fusion, applied to a method for predicting a concealed ore body by multi-source data fusion according to any one of claims 1-9, characterized in that, The system includes: Multi-source data acquisition module, used for obtaining gravity exploration data, magnetic exploration data, electrical exploration data, geochemical data and geological interpretation data to obtain multi-source exploration data; Standardization processing module, used for standardizing the multi-source exploration data to generate standardized data with unified dimension; Uncertainty quantification module, used for quantifying the error of the standardized data in spatial position and attribute value to generate multi-source data set containing uncertainty features representing data quality; Physical constraint embedding module, used for embedding the multi-source data set according to physical rules to obtain physical constraint embedded data meeting geological rules; Multi-scale feature extraction module, used for multi-scale feature extraction on the physical constraint embedded data to generate multi-source feature map containing different spatial range features; Cross-source feature fusion module, used for cross-source and cross-scale correlation on the multi-source feature map to generate fusion feature vector integrating information of each data source; Three-dimensional mineralization prediction module, used for predicting mineralization probability, physical property parameter and uncertainty distribution based on the fusion feature vector to generate prediction result map.

Images