A gold mine target area optimization method based on spatial constraint modified embedded geochemical anomaly and a system and a memory thereof

Abstract

The application discloses a gold mine target area optimization method based on space constraint modified embedded geochemical anomaly and a system and a memory thereof. The method comprises the following steps: collecting geochemical data of a target area, obtaining a space-sequenced standardized data matrix and an adjacency matrix after preprocessing; adopting global Transformer coding modeling to obtain a hidden state sequence of global element interaction characteristics of each sampling point in the target area, then sequentially constructing a unary potential and a binary potential, combining CRF to establish a geochemical anomaly area identification model of each sampling point in the target area, and then performing anomaly scoring and labeling on the geochemical anomaly area labeled by the model, and finally completing gold mine target area optimization through comprehensive evaluation. The system provided based on the above method contains a computer program memory, and can effectively solve the problems of large amount of calculation and low accuracy in the prior art. Through testing, the area under the ROC curve of the gold mine target area optimization process using the system can reach 0.92, which shows that the system has significant advantages.

Description

Technical Field

[0001] The present invention relates to a gold mine target area optimization method, in particular to a gold mine target area optimization method based on spatially constrained modified embedded geochemical anomaly and a system and memory thereof. Background Art

[0002] In gold exploration, identifying geochemical anomalies is a key step in determining potential gold target areas. While geochemical anomaly identification technology has been continuously developing, traditional methods still have significant drawbacks.

[0003] In the early days, the mean-standard deviation method was a commonly used method for identifying geochemical anomalies. This method is simple to operate and only calculates the mean and standard deviation of geochemical indicators, using a certain multiple of the standard deviation as a threshold to identify anomalies. However, this method is too simple and crude. In actual gold mining exploration scenarios, geochemical data is complex and multidimensional, usually containing stream sediment data of multiple dimensions such as the content of multiple elements such as gold, silver, copper, lead, zinc, and other related geological parameters. The mean-standard deviation method can only analyze a single or a few geochemical indicators and cannot fully and comprehensively utilize multi-dimensional data information. For those geochemical anomaly patterns formed by complex combinations of multiple elements, this method is often difficult to detect, resulting in a large number of potential gold mining targets being missed.

[0004] With the advancement of machine learning technology, algorithms such as support vector machines and decision trees have also been applied to geochemical anomaly identification. These algorithms perform well when processing simple data, but the curse of dimensionality becomes prominent when faced with high-dimensional geochemical data. High-dimensional data means more features, which not only increases computational complexity but also easily leads to model overfitting, resulting in poor generalization of the model to new data. Furthermore, geochemical data exhibits temporal and spatial sequence characteristics, such as changes in geochemical element content over time in the same region and the distribution of geochemical elements at different spatial locations. Traditional machine learning algorithms such as support vector machines and decision trees struggle to effectively mine and utilize this long-term dependency and sequence information, making it difficult to accurately identify geochemical anomalies associated with gold deposits.

[0005] Furthermore, the spatial distribution of geochemical data is not completely random, but rather exhibits a degree of spatial autocorrelation. Existing methods often fail to fully account for this spatial autocorrelation, resulting in an inaccurate reflection of the actual geological conditions when identifying geochemical anomalies and optimizing gold targets. This results in poor accuracy in the final gold targets, which in turn impacts the efficiency and cost of subsequent exploration.

[0006] In summary, when processing high-dimensional geochemical data, existing technologies have serious deficiencies in fully mining data features, utilizing sequence information, and accurately identifying geochemical anomalies. A new method is urgently needed to improve the accuracy and efficiency of gold target area selection. Summary of the Invention

[0007] To address the challenges of existing technologies, the first objective of this invention is to provide a method for optimizing gold target areas based on spatially constrained, modified, and embedded geochemical anomalies. This method comprehensively and deeply explores the features and relationships within geochemical data, aiming to address various challenges inherent in existing technologies when processing high-dimensional geochemical data. This method significantly improves the accuracy and efficiency of gold target area optimization, providing more scientific and reliable technical support for gold exploration.

[0008] A second object of the present invention is to provide a computer-readable memory, which contains a computer program for implementing the above-mentioned uniformity control method and can be read and executed.

[0009] A third objective of the present invention is to provide a system for optimizing gold target areas based on spatially constrained, modified, and embedded geochemical anomalies. This system, leveraging the superior accuracy and efficiency of the aforementioned method, enables precise identification and optimization of gold target areas. This process requires minimal computing power, requiring only a conventional computer, significantly reducing equipment costs. Testing has shown that the area under the receiver operating characteristic (ROC) curve for gold target area optimization using this system can reach 0.92, demonstrating significant advantages.

[0010] In order to achieve the above technical objectives, the present invention provides a method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies, comprising:

[0011] Step S1: Collect geochemical data of the target area and obtain a spatially serialized standardized data matrix and adjacency matrix after preprocessing;

[0012] Step S2: Use the preprocessed data to perform global Transformer encoding modeling to obtain a hidden state sequence containing the global element interaction features of each sampling point in the target area;

[0013] Step S3: Based on the hidden state sequence obtained in step S2, the hidden state is mapped and the geographical distance and element similarity are integrated through the fully connected layer to construct a one-dimensional potential energy and a two-dimensional potential energy, and then the CRF is used to establish a geochemical anomaly area recognition model for each sampling point in the target area;

[0014] Step S4: The geochemical anomaly areas marked by the model obtained in step S3 are scored using the autoregressive characteristics of the Transformer model, and geochemical anomaly areas above the scoring threshold are marked as anomaly areas. A comprehensive evaluation is then performed to complete the optimization of gold target areas.

[0015] As a preferred solution, the geochemical data includes the spatial coordinates of each sampling point, the element type of each sampling point, and the concentration value of the element type.

[0016] As a preferred solution, the preprocessing process is a multi-stage preprocessing process, including: cleaning process, standardization process and adjacency matrix construction process, and the specific process is:

[0017] Step S1-1, cleaning processing: After cleaning the data, the estimated value of the missing elements is filled using the K-nearest neighbor (KNN) interpolation method combined with spatial position. The estimated value of the i-th missing element is:

[0018] Formula 1: ;

[0019] Formula 2: ;

[0020] Step S1-2, standardization: the feature matrix Perform normalization or standardization to eliminate dimensional differences and obtain a standardized feature matrix ;

[0021] Step S1-3, adjacency matrix construction process: construct Voronoi adjacency matrix based on Delaunay triangulation , if and only if two points are Voronoi adjacent, ,provides the topological structure for the subsequent CRF spatial constraints;

[0022] In formulas 1 and 2, for The K spatial nearest neighbors of a point; is the concentration value of the element type; is the weight; for The longitude of the point's spatial location; for The dimension of the point's spatial position; is the bandwidth parameter, which controls the influence of spatial distance on weight.

[0023] As a preferred solution, the global Transformer encoding modeling process is: inputting the preprocessed and serialized standardized data matrix into a globally perceptual Transformer encoder constructed by multiple layers of improved Transformer Blocks to obtain a hidden state sequence containing the global element interaction features of each sampling point in the target area.

[0024] The encoder mechanism allows each position to interact with all positions in the sequence, breaking the limitations of traditional autoregressive models and fully capturing the complex relationships between elements in the data. In this way, it can mine hidden nonlinear relationships in earth data and provide rich and valuable feature information for subsequent analysis.

[0025] As a preferred solution, the first layer of the multi-layer improved Transformer Block The calculation process of the layer is as follows:

[0026] The full-domain attention mechanism is adopted to remove the restriction of the mask matrix M, allowing each position to interact freely with all positions in the sequence. The calculation formula is:

[0027] Formula 3: ;

[0028] Formula 4: ;

[0029] Formula 5: ;

[0030] Formula 6: ;

[0031] In formulas 3 to 6, is the standardized data matrix; is the query matrix; is the bond matrix; is the value matrix; are learnable parameters.

[0032] The attention mechanism employed in this paper enables the model to comprehensively capture the complex relationships between elements in geochemical data, breaking through the limitations of traditional autoregressive models, which can only detect local interactions, and excavating hidden nonlinear relationships between elements. For example, in complex geochemical environments, the relationship between certain trace elements and gold deposit formation may be overlooked by traditional models. However, this global interactive attention mechanism can uncover potential connections between them, providing richer characteristic information for the subsequent accurate identification of geochemical anomalies.

[0033] As a preferred solution, the process of constructing the geochemical anomaly region identification model is as follows:

[0034] Step S3-1: Use the Transformer encoder to output the hidden state sequence , the hidden state is mapped to the initial abnormal probability, i.e., unary potential energy, through the fully connected layer;

[0035] Step S3-2, integrating geographical distance and element similarity to construct a dynamic transfer matrix, i.e., binary potential energy;

[0036] Step S3-3: Optimize the unary potential energy and the binary potential energy through the global constraint of CRF to solve the maximum posterior probability, and obtain .

[0037] As a preferred solution, the calculation process of the unary potential energy is:

[0038] Formula 7: ;

[0039] The calculation process of the binary potential energy is:

[0040] Formula 8: ;

[0041] The formula for solving the maximum posterior probability is:

[0042] Formula 9: ;

[0043] In formulas 8 and 9, , which contains the global element interaction features of the 𝑖th sampling point; is a learnable parameter, corresponding to the label ; is the bias vector, ; is the Voronoi adjacency matrix; is the geographical distance; is the element similarity attenuation coefficient; is the set of all adjacent edges, is the normalization factor.

[0044] The binary potential calculation process of the present invention fully considers the spatial characteristics of geochemical data, leveraging geographic distance and elemental similarity to measure the probability of state transitions between adjacent sampling points. For example, spatially adjacent sampling points with similar elemental concentrations are more likely to share the same state (normal or abnormal). This approach effectively leverages the spatial correlation of data for anomaly identification. Furthermore, in the process of solving the maximum a posteriori probability, the CRF global optimization method determines the most likely state for each sampling point, i.e., whether it is a geochemically anomalous region, by considering both the state probability of each sampling point (unary potential) and the state transition probabilities between adjacent sampling points (binary potential), thereby accurately identifying geochemical anomalies.

[0045] As a preferred solution, the process of scoring the geochemical anomaly area is as follows: using the autoregressive characteristics of the Transformer model, the input high-dimensional geochemical data is processed in steps S2 and S3 and then reconstructed, and the difference between the original input data and the reconstructed data is calculated as the anomaly score. The calculation process is as follows:

[0046] Formula 10: ;

[0047] In formula 10, Indicates the Sampling point The original input value of elements; Indicates the Sampling point The predicted value of an element.

[0048] As a preferred solution, the scoring threshold is determined according to the maximum Youden index. When the abnormality score is higher than the threshold, it is marked as an abnormal area.

[0049] As a preferred solution, the key factors of the comprehensive assessment include the geological structural stability of the target area and the spatial relationship of known gold mines.

[0050] The present invention also provides a storable calculator, including a computer program, which can implement any of the above-mentioned methods for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies.

[0051] The present invention also provides a gold mine target area optimization system based on spatially constrained modified embedded geochemical anomalies, comprising: a programmable logic controller, a central processing unit and the above-mentioned storable calculator; the programmable logic controller collects production process data, inputs it into the central processing unit and executes the computer program on the storable calculator.

[0052] Compared with the prior art, the beneficial technical effects of the technical solution of the present invention are:

[0053] 1) The method provided by the present invention introduces the Transformer model, fully leveraging its self-attention mechanism's ability to capture long-range dependencies in long-sequence data, mining the nonlinear relationships between elements in high-dimensional geochemical data. Combined with the conditional random field (CRF), it utilizes its ability to model data contextual relationships, especially considering the spatial characteristics of geochemical data, to more accurately identify geochemical anomalies. Then, through reconstruction error anomaly scoring and comprehensive evaluation, the optimization of gold mine targets is achieved, significantly improving the accuracy and efficiency of the optimization results.

[0054] 2) In the technical solution provided by the present invention, based on the excellent accuracy and efficiency of the above-mentioned method, accurate identification and optimization of gold mine target areas can be achieved. This process requires extremely little computing power and can be implemented only with a conventional computer, greatly reducing equipment costs. After testing, the area under the ROC curve of the gold mine target area optimization process using this system can reach 0.92, showing its significant advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 1 is a ROC curve diagram of the method provided by Example 1 of the present invention and Comparative Example 1;

[0056] Figure 2 The ROC curve diagram of the method provided by Example 1 of the present invention and Comparative Examples 2 and 3 is shown. DETAILED DESCRIPTION

[0057] To facilitate understanding of the present invention, the present invention will be described in more detail below with reference to the accompanying drawings. It should be noted that the embodiments described are only some embodiments of the present invention and not all embodiments. All other embodiments derived by those skilled in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the invention.

[0058] Example 1

[0059] This example provides a method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies. The Jiaoxibei gold mining area in Shandong Province was selected as the research area, and geochemical measurements of 39 elements in stream sediments were collected from this area as raw data. The known deposit validation set includes 27 large and medium-sized deposits. During data processing and model execution, the powerful computing power of the Nvidia RTX 3090Ti 128GB graphics card was leveraged to ensure efficient and accurate calculations. The specific process is as follows:

[0060] Step S1: Collect geochemical data of the target area and obtain a spatially serialized standardized data matrix and adjacency matrix after preprocessing;

[0061] Based on regional geochemical exploration water system sediment data, the input data includes the geographical coordinates of the sampling points (longitude ,latitude , i=1,...,N) and concentration values ​​of 39 elements (Au, As, Cu, etc.) (d=1,...,39), the above data are preprocessed at multiple levels, including cleaning, standardization and adjacency matrix construction. The specific process is as follows:

[0062] Step S1-1, cleaning processing: After cleaning the data, the estimated value of the missing elements is filled using the K-nearest neighbor (KNN) interpolation method combined with spatial position. The estimated value of the i-th missing element is:

[0063] Formula 1: ;

[0064] Formula 2: ;

[0065] Step S1-2, standardization: the feature matrix Perform normalization or standardization to eliminate dimensional differences and obtain a standardized feature matrix ;

[0066] Step S1-3, adjacency matrix construction process: construct Voronoi adjacency matrix based on Delaunay triangulation , if and only if two points are Voronoi adjacent, ,provides the topological structure for the subsequent CRF spatial constraints;

[0067] In formulas 1 and 2, for The K spatial nearest neighbors of a point; is the concentration value of the element type; is the weight; for The longitude of the point's spatial location; for The dimension of the point's spatial position; is the bandwidth parameter, which controls the influence of spatial distance on weight.

[0068] Step S2: Use the preprocessed data to perform global Transformer encoding modeling to obtain a hidden state sequence containing the global element interaction features of each sampling point in the target area;

[0069] The global Transformer encoding modeling process is as follows: the preprocessed and serialized standardized data matrix is ​​input into the global perception Transformer encoder constructed by the multi-layer improved Transformer Block to obtain the hidden state sequence containing the global element interaction features of each sampling point in the target area; the first layer of the multi-layer improved Transformer Block The calculation process of the layer is as follows:

[0070] The full-domain attention mechanism is adopted to remove the restriction of the mask matrix M, allowing each position to interact freely with all positions in the sequence. The calculation formula is:

[0071] Formula 3: ;

[0072] Formula 4: ;

[0073] Formula 5: ;

[0074] Formula 6: ;

[0075] In formulas 3 to 6, is the standardized data matrix; is the query matrix; is the bond matrix; is the value matrix; are learnable parameters.

[0076] Step S3: Based on the hidden state sequence obtained in step S2, the hidden state is mapped and the geographical distance and element similarity are integrated through the fully connected layer to construct a one-dimensional potential energy and a two-dimensional potential energy, and then the CRF is used to establish a geochemical anomaly area recognition model for each sampling point in the target area;

[0077] The construction process of the geochemical anomaly area identification model is as follows:

[0078] Step S3-1: Use the Transformer encoder to output the hidden state sequence , the hidden state is mapped to the initial abnormal probability, i.e., unary potential energy, through the fully connected layer;

[0079] Step S3-2, integrating geographical distance and element similarity to construct a dynamic transfer matrix, i.e., binary potential energy;

[0080] Step S3-3: Optimize the unary potential energy and the binary potential energy through the global constraint of CRF to solve the maximum posterior probability, and obtain .

[0081] As a preferred solution, the calculation process of the unary potential energy is:

[0082] Formula 7: ;

[0083] The calculation process of the binary potential energy is:

[0084] Formula 8: ;

[0085] The formula for solving the maximum posterior probability is:

[0086] Formula 9: ;

[0087] In formulas 8 and 9, , which contains the global element interaction features of the 𝑖th sampling point; is a learnable parameter, corresponding to the label ; is the bias vector, ; is the Voronoi adjacency matrix; is the geographical distance; is the element similarity attenuation coefficient; is the set of all adjacent edges, is the normalization factor.

[0088] Step S4: Using the autoregressive characteristics of the Transformer model, the geochemical anomaly areas marked by the model obtained in step S3 are scored, and the geochemical anomaly areas above the scoring threshold are marked as anomaly areas. A comprehensive evaluation is then performed to complete the optimization of gold target areas.

[0089] The process of scoring anomaly in geochemical anomaly areas is as follows: using the autoregressive characteristics of the Transformer model, the input high-dimensional geochemical data is processed in steps S2 and S3 and then reconstructed, and the difference between the original input data and the reconstructed data is calculated as the anomaly score. The calculation process is as follows:

[0090] Formula 10: ;

[0091] In formula 10, Indicates the Sampling point The original input value of elements; Indicates the Sampling point The predicted value of an element.

[0092] As a preferred solution, the scoring threshold is determined according to the maximum Youden index. When the abnormality score is higher than the threshold, it is marked as an abnormal area.

[0093] As a preferred solution, the key factors of the comprehensive assessment include the geological structural stability of the target area and the spatial relationship of known gold mines.

[0094] The larger the reconstruction error, the greater the deviation of the geochemical data pattern from the normal pattern in that area, indicating a higher degree of geochemical anomaly. Based on the reconstruction error, an anomaly score is further calculated, and areas with scores above a certain threshold are identified as geochemically anomaly areas. This method of calculating anomaly scores based on reconstruction error provides a more objective and accurate assessment of the degree of anomaly in each area, avoiding the subjectivity and limitations of manually set thresholds.

[0095] Based on the trained CRF model, an anomaly score is calculated. A threshold for the anomaly score is set based on the maximum Youden index. Areas corresponding to data points with an anomaly score greater than the threshold and identified as anomaly by the CRF are then marked as potential gold targets. These potential targets are further comprehensively evaluated, taking into account factors such as their geographic location and surrounding geological structures, to ultimately determine priority gold targets for exploration.

[0096] The Transformer encoder used in this paper has 8 layers, 2 attention heads, 512 hidden layer dimensions, 0.1 dropout ratio, AdamW optimizer, 64 batch size, and 100 training rounds. The CRF regularization coefficient is 10e -4 , to prevent the transfer matrix from overfitting.

[0097] This embodiment also provides a storable calculator, including a computer program, which can implement any of the above-mentioned methods for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies.

[0098] This embodiment also provides a gold mine target area optimization system based on spatially constrained modified embedded geochemical anomalies, including: a programmable logic controller, a central processing unit and the above-mentioned storable calculator; the programmable logic controller collects production process data and inputs it into the central processing unit and executes the computer program on the storable calculator.

[0099] In order to better illustrate the excellent effect of the technical solution provided by the present invention, the present invention introduces the AUC and Morna's I evaluation indicators to perform gold mine target area optimization scoring on Example 1, wherein the Moran's I index is a spatial indicator used to evaluate the spatial aggregation of predicted anomalies. The Moran's I index value range is between -1 and 1. A value greater than 0 indicates a positive correlation, i.e., there is a spatial aggregation phenomenon; a value equal to 0 indicates a random distribution; and a value less than 0 indicates a negative correlation, i.e., a discrete distribution in space. By calculating the Moran's I index, we further understand the spatial distribution characteristics of geochemical anomalies and verify the ability of the present invention method to capture the spatial pattern of geochemical anomalies. The above-mentioned gold mine target area optimization score results are shown in Table 1.

[0100]

[0101] Comparative Example 1

[0102] This comparative example is exactly the same as Example 1, except that principal component analysis (PCA) + local singularity index (Local Singularity Index) are selected for geochemical anomaly identification.

[0103] Comparative Example 2

[0104] This comparative example is exactly the same as Example 1, except that an ablation experiment is performed and only the Transformer model is retained.

[0105] Comparative Example 3

[0106] This comparative example is exactly the same as Example 1, except that an ablation experiment is performed and only the CRF model is retained.

[0107] The present invention compared the AUC values ​​and Morna's I values ​​of Example 1 and Comparative Examples 1 to 3, and the results are shown in Table 2.

[0108]

[0109] The comparative experiments and results presented above clearly demonstrate the excellent AUC performance of the method provided by this invention, demonstrating its significant advantages in identifying correlations between geochemical anomalies and mineral deposits. Furthermore, the Moran's I index demonstrates the excellent performance of this method in capturing the spatial clustering of geochemical anomalies and optimizing gold target areas. Combined with these indicators and comparative experimental results, the feasibility and effectiveness of this method in optimizing gold target areas in-situ are fully demonstrated, as well as its superiority over traditional methods and single models.

Claims

1. A method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies, characterized in that: include: Step S1: Collect geochemical data of the target area and obtain a spatially serialized standardized data matrix and adjacency matrix after preprocessing; Step S2: Use the preprocessed data to perform global Transformer encoding modeling to obtain a hidden state sequence containing the global element interaction features of each sampling point in the target area; Step S3: Based on the hidden state sequence obtained in step S2, the hidden state is mapped and the geographical distance and element similarity are integrated through the fully connected layer to construct a unary potential energy and a binary potential energy, and then a geochemical anomaly area recognition model for each sampling point in the target area is established through the conditional random field (CRF); Step S4: The geochemical anomaly areas marked by the model obtained in step S3 are scored using the autoregressive characteristics of the Transformer model, and geochemical anomaly areas above the scoring threshold are marked as anomaly areas. A comprehensive evaluation is then performed to complete the optimization of gold target areas.

2. The method for selecting a gold target area based on spatially constrained modified embedded geochemical anomalies according to claim 1, characterized in that: The geochemical data includes the spatial coordinates of each sampling point, the element type of each sampling point, and the concentration value of the element type.

3. The method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies according to claim 2, characterized in that: The preprocessing process is a multi-stage preprocessing process, including: cleaning process, standardization process and adjacency matrix construction process, and its specific process is as follows: Step S1-1, cleaning processing: After cleaning the data, the estimated value of the missing elements is filled using the K-nearest neighbor interpolation method combined with spatial position, where the estimated value of the i-th missing element is: Formula 1: ; Formula 2: ; Step S1-2, standardization: the feature matrix Perform normalization or standardization to eliminate dimensional differences and obtain a standardized feature matrix ; Step S1-3, adjacency matrix construction process: construct Voronoi adjacency matrix based on Delaunay triangulation , if and only if two points are Voronoi adjacent, ,provides the topological structure for the subsequent CRF spatial constraints; In formulas 1 and 2, for The K spatial nearest neighbors of a point; is the concentration value of the element type; is the weight; for The longitude of the point's spatial location; for The dimension of the point's spatial position; is the bandwidth parameter, which controls the influence of spatial distance on weight.

4. The method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies according to claim 1, characterized in that: The global Transformer encoding modeling process is as follows: inputting the preprocessed and serialized standardized data matrix into a globally perceptual Transformer encoder constructed by multiple layers of improved Transformer Blocks to obtain a hidden state sequence containing the global element interaction features of each sampling point in the target area; The multi-layer improved Transformer Block The calculation process of the layer is as follows: The full-domain attention mechanism is adopted to remove the restriction of the mask matrix M, allowing each position to interact freely with all positions in the sequence. The calculation formula is: Formula 3: ; Formula 4: ; Formula 5: ; Formula 6: ; In formulas 3 to 6, is the standardized data matrix; is the query matrix; is the bond matrix; is the value matrix; are learnable parameters.

5. The method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies according to claim 1, characterized in that: The construction process of the geochemical anomaly area identification model is as follows: Step S3-1: Use the Transformer encoder to output the hidden state sequence , the hidden state is mapped to the initial abnormal probability, i.e., unary potential energy, through the fully connected layer; Step S3-2, integrating geographical distance and element similarity to construct a dynamic transfer matrix, i.e., binary potential energy; Step S3-3: Optimize the unary potential energy and the binary potential energy through the global constraint of CRF to solve the maximum posterior probability.

6. The method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies according to claim 5, characterized in that: The calculation process of the unary potential energy is: Formula 7: ; The calculation process of the binary potential energy is: Formula 8: ; The formula for solving the maximum posterior probability is: Formula 9: ; In formulas 8 and 9, , which contains the global element interaction features of the 𝑖th sampling point; is a learnable parameter, corresponding to the label ; is the bias vector ; is the Voronoi adjacency matrix; is the geographical distance; is the element similarity attenuation coefficient; is the set of all adjacent edges, is the normalization factor.

7. The method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies according to claim 1, characterized in that: The process of scoring anomaly in geochemical anomaly areas is as follows: using the autoregressive characteristics of the Transformer model, the input high-dimensional geochemical data is processed in steps S2 and S3 and then reconstructed, and the difference between the original input data and the reconstructed data is calculated as the anomaly score. The calculation process is as follows: Formula 10: ; In formula 10, Indicates the Sampling point The original input value of elements; Indicates the Sampling point The predicted value of an element.

8. The method for selecting gold target areas based on spatially constrained modified embedded geochemical anomalies according to claim 7, characterized in that: The scoring threshold is determined based on the maximum Youden index. When the anomaly score is higher than the threshold, it is marked as an abnormal area. The key factors of the comprehensive evaluation include the geological structural stability of the target area and the spatial relationship of known gold mines.

9. A storable calculator comprising a computer program, characterized in that: The computer program can implement the gold target area optimization method based on spatially constrained modified embedded geochemical anomalies as described in any one of claims 1 to 8.

10. A gold target area optimization system based on spatially constrained modified embedded geochemical anomalies, characterized in that: include: A programmable logic controller, a central processing unit and a storable calculator as claimed in claim 9; The programmable logic controller collects production process data and inputs it into the central processing unit and executes the computer program on the storable computer.

Images