Method and device for evaluating ore-bearing nature of geochemical anomaly

Abstract

The application provides a geochemical exploration anomaly ore-bearing property evaluation method and device, obtains geochemical exploration information of a target anomaly region to be evaluated in ore-bearing property, obtains reconstruction concentration information of a chemical element at each sampling point according to the geochemical exploration information and a trained physical information neural network model, wherein the trained physical information neural network model comprises a physical simulation layer embedded with a fractional order reaction-diffusion equation; and then the target anomaly region is evaluated in ore-bearing property according to original concentration information and the reconstruction concentration information of the chemical element at each sampling point. The fractional order reaction-diffusion equation describing the ore-forming geochemical process is learned and simulated through the physical information neural network model, so that the reconstruction concentration information is more in line with the ore-forming regularity, and field verification or drilling is performed when the ore-bearing property evaluation grade of the target anomaly region is high, thereby reducing the exploration cost.

Description

Technical Field

[0001] This application relates to the field of metal mineral exploration technology, and in particular to a method and apparatus for evaluating the mineralization of geochemical anomalies. Background Technology

[0002] "Geochemical anomaly mineralization assessment" is a process that systematically identifies whether an elemental anomaly zone discovered by geochemical exploration (geochemical exploration) is caused by an economically valuable ore body (or mineralization), and preliminarily assesses its mineral exploration potential and economic significance.

[0003] The formation of metallic minerals is essentially the result of the extraordinary enrichment of specific elements during geological processes. Geochemical data directly or indirectly reveals the enrichment and migration patterns of elements in the Earth's crust. Therefore, geochemical analysis can directly reveal the spatial distribution and anomalous concentrations of target elements (such as copper and gold) and their associated elements (such as arsenic and antimony), thereby transforming macroscopic geological mineralization theories into microscopic, quantifiable element enrichment signals. This allows for the efficient and economical delineation of anomalous areas related to mineralization, providing crucial direct evidence for the selection of prospecting targets and the prediction of deep ore bodies.

[0004] However, existing geochemical anomaly identification techniques, such as statistical analysis, filtering / interpolation, machine learning, and empirical thresholding, are typically used to delineate anomalous areas that may contain ore bodies from a large background field. They cannot assess mineralization, requiring systematic engineering verification and sampling analysis. This necessitates engineering verification and sampling analysis of all delineated anomaly areas, resulting in high costs and low efficiency. Furthermore, although existing technologies use traditional reaction-diffusion models to simulate mineralization processes and evaluate mineralization, these models cannot reflect long-range nonlocal interactions, leading to inaccurate mineralization assessments. Summary of the Invention

[0005] This application provides a method and apparatus for evaluating the mineralization of geochemical anomalies, in order to solve the technical problems mentioned in the background art.

[0006] Firstly, this application provides a method for evaluating the mineralization of geochemical anomalies, including: Obtain geochemical information of the target anomaly area to be evaluated for mineralization, wherein the geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area; Based on the geochemical exploration information and the trained physical information neural network model, the reconstructed concentration information of chemical elements at each sampling point is obtained. The trained physical information neural network model includes a physical simulation layer embedded with a fractional reaction-diffusion equation. The physical simulation layer simulates the diffusion term, convection term, and reaction term in the fractional reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension. Based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point, the Lyapunov index and self-organization-other-organization ratio of the target anomaly region are obtained; The mineralization level of the geochemical anomaly in the target anomaly area is obtained based on the Lyapunov index and the self-organization-other-organization ratio.

[0007] Secondly, this application provides a geochemical anomaly mineralization evaluation device, characterized in that it includes: The acquisition module is used to acquire geochemical information of the target anomaly area to be evaluated for mineralization. The geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area. The model processing module is used to obtain the reconstructed concentration information of chemical elements at each sampling point based on the geochemical exploration information and the trained physical information neural network model. The trained physical information neural network model includes a physical simulation layer embedded with a fractional reaction-diffusion equation. The physical simulation layer simulates the diffusion term, convection term, and reaction term in the fractional reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension. The evaluation module is used to obtain the Lyapunov index and self-organization-other-organization ratio of the target anomaly region based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point, and to obtain the geochemical anomaly mineralization level of the target anomaly region based on the Lyapunov index and the self-organization-other-organization ratio.

[0008] Thirdly, this application provides an electronic device, including: a processor and a memory; The memory stores the instructions that the computer executes; The processor executes computer execution instructions stored in memory, causing the processor to perform the method as described in any of the first aspects.

[0009] Fourthly, embodiments of this application provide a readable storage medium including a program or instructions that, when run on a computer, execute the method described in any of the first aspects above.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the first aspects.

[0011] Compared with existing technologies, the geochemical anomaly mineralization evaluation method and apparatus provided in this application have the following advantages: 1. By introducing a fractional-order reaction-diffusion equation into the physical information neural network model, and using the fractional-order exponent α to capture the nonlocality and self-organized reaction characteristics in the diffusion process, the reconstructed concentration information can reflect the geochemical field affected by long-term geological evolution, thereby improving the authenticity of mineralization evaluation and providing support for field verification and drilling.

[0012] 2. For each anomalous region, mineralization is evaluated, and the boundary conditions of the anomalous region are used as Dirichlet boundary conditions to make the boundary corresponding to the reconstructed concentration information more consistent with geological significance.

[0013] 3. The coupling error of multiple chemical elements is reduced by the spectral separability module, thereby improving the robustness of multi-element fusion.

[0014] 4. LE and SvL provide dynamic and fractal basis for prioritizing anomalous areas, facilitating exploration decision-making. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the structure of a physical information neural network model provided in an embodiment of this application; Figure 2 A flowchart of a method for evaluating the mineralization of geochemical anomalies provided in an embodiment of this application; Figure 3 This is a flowchart illustrating the internal operation of a physical information neural network model provided in one embodiment of this application. Figure 4 This is a practical application effect diagram provided for one embodiment of this application; Figure 5A schematic diagram of the structure of a geochemical anomaly mineralization evaluation device provided in an embodiment of this application. Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application are described clearly and completely below. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of this application.

[0018] Current geochemical anomaly identification techniques, such as statistical analysis, filtering / interpolation, machine learning, and empirical thresholding, are used to determine anomaly regions but cannot evaluate mineralization. They require systematic engineering verification and sampling analysis, resulting in high costs and low efficiency. Furthermore, when using traditional reaction-diffusion models to simulate mineralization processes and evaluate mineralization, these models cannot reflect long-range nonlocal interactions, leading to inaccurate mineralization assessments.

[0019] Therefore, to address the technical problems existing in the prior art, this application provides a method and apparatus for evaluating the mineralization of geochemical anomalies. Since the diffusion process described by the traditional reaction-diffusion equation is only valid in a homogeneous, isotropic ideal medium, actual mineralization processes, such as fracture zones, fracture networks, and porous media, are extremely heterogeneous and anisotropic. Therefore, this application introduces a fractional-order exponent, namely the Hausdorff dimension α, into the traditional reaction-diffusion equation to obtain a fractional-order reaction-diffusion equation. This fractional-order equation mathematically describes the path dependence and spatially nonlocal interactions through the nonlocal integral kernel in fractional calculus. Then, the fractional-order reaction-diffusion equation is used as the loss function during the training of the physical information neural network model. By integrating the fractional-order reaction-diffusion equation into the physical information neural network model, the physical information neural network model follows the physical laws of the mineralization process when reconstructing the concentration field of the anomalous region and solving the fractional-order reaction-diffusion equation. This makes the output data interpretable and realizes the goal of judging the self-organization criticality or steady state of geochemical anomaly patches from the dynamic level based on the theory of "chaotic edge" and conducting mineralization evaluation, thereby improving the effectiveness of mineralization evaluation.

[0020] To make the subsequent explanation of the geochemical anomaly mineralization evaluation method of this application clearer, the fractional-order reaction-diffusion equation will be explained first.

[0021] The traditional reaction-diffusion equation is expressed by Equation 1: Formula 1 In Formula 1, the first, second, and third terms on the right side of the equation represent the diffusion, convection, and reaction terms, respectively. C i ( x , y , t ) represents chemical elements i In time t The spatial coordinates corresponding to a given time (in geochemistry, these are often longitude, latitude, or plane coordinates) x , y Concentration at ) D i Represents chemical elements i The effective diffusion coefficient, Represents the Laplace operator. v i ( x , y , t ) represents chemical elements i In time t The spatial coordinates corresponding to a given time (in geochemistry, these are often longitude, latitude, or plane coordinates) x , y The velocity field corresponding to the location (applicable to convection / Darcy flow scenarios).

[0022] The reaction-diffusion equation for the reaction in Formula 1 is a dynamic model. The inputs are the initial and boundary conditions, and the output is the concentration field. C i ( x , y , t Over time t The evolution of the ore-forming process is important, but the current geochemical field is a cumulative result of long-term geodynamic evolution. Therefore, its evolutionary history (i.e., time t) and its initial concentration and boundary conditions cannot be reconstructed using Equation 1. However, on the other hand, since the current geochemical field is a cumulative result of long-term geodynamic evolution, when time t is the current time, the complex dynamics of the ore-forming process can be combined with the currently measured geochemical data. Therefore, in order to make Equation 1 applicable to geochemical data processing, the left side of the equation needs to be 0, and the current state needs to be regarded as the "steady state".

[0023] Furthermore, in order to capture the nonlocality and self-organized reaction characteristics in the diffusion process, based on Equation 1, a fractional-order reaction-diffusion equation is obtained by introducing a fractional-order exponent, namely the Hausdorff dimension α, into the diffusion term and incorporating the differential into the reaction-diffusion equation. The fractional-order reaction-diffusion equation introduces memory (temporal nonlocality) and spatial nonlocality through the Hausdorff dimension α, thereby describing the geological medium and mineralization process.

[0024] In summary, the fractional-order reaction-diffusion equation can be expressed by Equation 2: Formula 2 in, Let represent the Laplace operator, which is solved in Fourier space for ease of integration, where α∈(0,2). Given the complexity and variability of physicochemical reactions and environmental conditions (such as energy, volatile matter, temperature, pressure, and material composition) related to mineralization, the reaction terms are defined at the following levels: Formula 3 in, k i ( x , y ) is a spatial correlation function. The second half of the term represents the precipitation process with self-reinforcing characteristics during mineralization. This process reflects the increase in elemental concentration throughout the region and is the result of the combined effects of crystal nucleation and growth, autocatalytic precipitation, and erosion.

[0025] For the parameters in Formulas 2 and 3, namely a, b, c, v i ( x , y ), k i ( x , y ), n , e i , D i These can be used as parameters for a physical information neural network model, obtained through training. The physical information neural network model solves for a, b, c, and ... v i ( x , y ), k i ( x , y ), n , e i , D i The process is also the process of integrating fractional-order reaction-diffusion equations into a physical information neural network model.

[0026] Based on the fractional-order reaction-diffusion equation, the structure of the physical information neural network model in this application is as follows: Figure 1As shown, it includes: an input layer 110, a physical simulation layer 120 and an output layer 130. The physical simulation layer simulates the diffusion term, convection term and reaction term in the fractional-order reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional-order reaction-diffusion equation.

[0027] based on Figure 1 The training process of the physical information neural network model shown in the diagram is as follows: S101. Input the geochemical exploration information as training data into the initial physical information neural network model to obtain the predicted concentration information of chemical elements at each sampling point.

[0028] In this step, taking the target area as an example, when evaluating the mineralization of the predicted target anomaly area, the geochemical information of the target anomaly area is used to train the initial physical information neural network model. The geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area.

[0029] The target area includes a background area and multiple abnormal areas, which are regions within the target area where metal ore may be present. It should be noted that the target area and abnormal areas mentioned in this embodiment can be patches obtained through superpixel segmentation technology, or patches obtained through other segmentation methods.

[0030] In this embodiment, geochemical information can be geochemical information of an anomalous area or geochemical information of a self-organized geochemical patch (i.e., the target area).

[0031] Among them, the location information of the sampling point is the coordinate point corresponding to the sampling point, which corresponds to ( x , y For any sampling point containing multiple chemical elements, the concentration of each chemical element at each sampling point is collected, corresponding to... C i ( x , y The boundary information of the target anomaly region includes the coordinates of the boundary of the target anomaly region and the concentration of the chemical element corresponding to each coordinate point. That is, the concentration of the chemical element corresponding to each coordinate point is denoted as: C i (x, y) = C i，d The segmentation boundary along the target anomaly region.

[0032] in, C i，d Represents any target anomaly region d Chemical elements at the boundary iThe concentration is determined, and a set of boundary points is generated for each target anomaly region. x , y , C i ( x , y The Dirichlet boundary conditions constitute the anomaly region of the target.

[0033] After the geochemical information is input as training data into the initial physical information neural network model, the initial physical information neural network model processes the geochemical information. Since the physical simulation layer simulates the diffusion term, convection term, and reaction term in the fractional-order reaction-diffusion equation, the initial physical information neural network model follows the fractional-order reaction-diffusion equation when processing the geochemical information, so that the output of the predicted value of the concentration information of chemical elements at each sampling point conforms to the physical laws of the mineralization process.

[0034] S102. Based on the original concentration information, predicted concentration information and loss function of chemical elements at each sampling point, train the initial physical information neural network model to obtain the trained physical information neural network model.

[0035] The loss function includes the physical residual loss function, the boundary condition loss function, and the initial condition loss function. The physical residual loss function is obtained by using the concentration information prediction value and the fractional-order reaction-diffusion equation. The boundary condition loss function is obtained by using the concentration information prediction value and the original concentration information at the boundary of the target anomaly region. The initial condition loss function is obtained by using the average value of the concentration information prediction value and the average value of the original concentration information.

[0036] In this step, such as Figure 1 As shown, the physical information neural network model also includes: a parameter solving layer 140 and a loss function processing layer 150. The parameter solving layer 140 is used to solve for parameters a, b, c, ... in the fractional-order reaction-diffusion equation. v i ( x , y ), k i ( x , y ), n , e i , D i The loss function processing layer 150 is used to process the physical residual loss function, boundary condition loss function and initial condition loss function. When the loss function is less than the preset loss function or the number of iterations of the physical information neural network model reaches the preset number of iterations, the trained physical information neural network model is obtained.

[0037] Specifically, since the predicted concentration information of chemical elements at each sampling point follows the fractional-order reaction-diffusion equation, the predicted concentration information is input into Formula 2, which should be 0. Therefore, the physical residual loss function is obtained through the predicted concentration information and the fractional-order reaction-diffusion equation. The physical residual loss function is expressed by the formula: Formula 4 Where PDE represents the right-hand side of Equation 2, and M and N represent the number of coordinate points on the X-axis and Y-axis, respectively. c i This represents the predicted concentration of chemical elements at each sampling point, with [•] indicating the real part. The closer the LossPDE value is to 0, the smaller the residual, indicating that the physical information neural network model follows the fractional-order reaction-diffusion equation.

[0038] The trained physical information neural network model needs to be constrained to ensure it can output predicted concentrations of chemical elements at each sampling point. Furthermore, to ensure that the predicted concentrations of chemical elements at each sampling point within each target anomaly region closely approximate the original concentrations at the boundary, a boundary condition loss function needs to be considered. The boundary condition loss function is expressed by the following formula: Formula 5 Where MB and NB represent the number of coordinate points on the boundary of the target anomaly region along the X-axis and Y-axis, respectively. c i ( x m , y n ) represents the coordinates of a point located on the boundary of the target anomaly region. x m , y n Concentration information prediction at location ) C i ( x m , y n ) represents the coordinates of a point located on the boundary of the target anomaly region. x m , y n The original concentration information at ().

[0039] Furthermore, for the sampling points within the target anomaly region, the concentration information exhibits spatial continuity and uniformity. Therefore, it is necessary to ensure that the predicted concentration values ​​of chemical elements at each sampling point output by the trained physical information neural network model also possess spatial continuity and uniformity. This prevents significant anomalies in the predicted concentration values ​​of chemical elements at any sampling point within the target anomaly region, which could lead to large discrepancies in the predicted concentration values ​​among the sampling points within the target anomaly region. In this case, an initial conditional loss function can be set to constrain the solution of the physical information neural network model by comparing the average of the predicted concentration values ​​with the average of the original concentration values. The initial conditional loss function is expressed by the formula: Formula 6 Wherein, MI and NI represent the number of coordinate points located in the target anomaly region along the X-axis and Y-axis, respectively. C i ( x 0, y 0) represents the initial conditions of the target anomaly region, generally the average value of the original concentration information, i.e. C i ( x 0, y 0) = mean ( C i，d ).

[0040] Therefore, the loss function is: Loss=λ1*LossPDE+λ2*LossBC+λ3*LossIC Formula 7 Wherein, λ1, λ2, and λ3 represent the weights of LossPDE, LossBC, and LossIC in the loss function Loss, respectively, which can be obtained through model training. For example, λ1, λ2, and λ3 can take values ​​of 1.00, 0.75, and 0.25, respectively.

[0041] It should be noted that the reason why there is a difference between the predicted concentration information of chemical elements at each sampling point and the original concentration information is that although the original concentration information is the true value, it contains noise due to sampling error, analysis error, etc.

[0042] When two target anomaly regions overlap, the predicted concentration values ​​of chemical elements at each sampling point corresponding to the overlapping region are averaged.

[0043] In summary, the well-trained physical information neural network model, through the physical residual loss function, boundary condition loss function, and initial condition loss function, ensures that the reconstructed concentration information of chemical elements at each sampling point follows the fractional-order reaction-diffusion equation. The reconstructed concentration information obtained is both in line with physical laws and based on geological experience.

[0044] Figure 2 This is a flowchart illustrating a method for evaluating the mineralization of geochemical anomalies provided in an embodiment of this application. Figure 2 As shown, the method includes: S201. Obtain geochemical information of the target anomaly area to be evaluated for mineralization.

[0045] The geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area.

[0046] S202. Based on geochemical exploration information and a trained physical information neural network model, obtain the reconstructed concentration information of chemical elements at each sampling point.

[0047] The trained physical information neural network model includes a physical simulation layer embedded with a fractional-order reaction-diffusion equation. The physical simulation layer simulates the diffusion, convection, and reaction terms in the fractional-order reaction-diffusion equation, so that the reconstruction process of the concentration information follows the fractional-order reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension.

[0048] S203. Based on the original and reconstructed concentration information of chemical elements at each sampling point, obtain the Lyapunov index and self-organization-other-organization ratio of the target anomaly region.

[0049] In this step, the reconstructed concentration information of chemical elements at each sampling point is obtained according to S301 and S302. There is a residual between the reconstructed concentration information and the original concentration information. This residual not only indicates that the reconstructed concentration information and the original concentration information have noise and uncertainty, but also reflects the multiple solutions of the fractional reaction-diffusion equation. That is, due to wind and rain, the ore body that was originally formed by self-organization disappears or its scale is significantly reduced. The residual between the reconstructed concentration information and the original concentration information is called the other-organization property.

[0050] Self-organized critical behavior controls the formation mode and location of high-grade ore bodies (large in scale and rich in grade): energy is slowly and continuously input into the mineralization system, while a threshold barrier prevents energy dissipation to the sink area, thus maintaining an enhanced energy gradient. Meanwhile, regarding the phenomenon in this embodiment where the reconstructed concentration information deviates from the fractional-order reaction-diffusion equation, the other-organization represents the uncontrollable function in geochemical behavior, i.e., the portion explained by the fractional-order reaction-diffusion equation. Based on this, the Lyapunov index LE and the self-organization-other-organization ratio SvL are used for mineralization evaluation. The self-organization-other-organization ratio represents the relative proportion or contribution rate of self-organization to observed other-organization.

[0051] One specific implementation of the Lyapunov exponent LE is as follows: S11. Obtain the Jacobian matrix at each sampling point from the initial position to the target position based on the reconstructed concentration information.

[0052] Specifically, the initial position is generally a sampling point on the boundary of the target anomaly region, and the target position is a sampling point located at the center of the target anomaly region. For the sampling points between the initial position and the target position, the Jacobian matrix J of each sampling point is obtained. The method for obtaining the Jacobian matrix J can refer to existing technologies, which will not be elaborated here.

[0053] S12. Based on the Jacobian matrix at each sampling point from the initial position to the target position, the perturbation vector corresponding to the initial position, and the perturbation step size, obtain the perturbation vector corresponding to the target position.

[0054] Specifically, the perturbation vector corresponding to the initial position is set manually. δ r0 And set the perturbation step size, which is generally less than or equal to the distance between two adjacent sampling points. Calculate the perturbation propagation according to the following formula, that is, the perturbation vector corresponding to the position after each perturbation step size: Formula 8 Among them, J( x n , y n )express( x n , y n The Jacobian matrix at () This represents the perturbation vector after n moves. This represents the perturbation vector after the perturbation has been propagated once more.

[0055] The perturbation vector corresponding to the target position is obtained by progressively applying Formula 8. .

[0056] S13. Based on the disturbance vector corresponding to the initial position, the disturbance vector corresponding to the target position, and the total number of disturbance propagation steps, obtain the Lyapunov exponent. The total number of disturbance propagation steps is obtained based on the step distance from the initial position to the target position and the disturbance step length.

[0057] Specifically, the Lyapunov index is obtained according to the formula:

[0058] Where, N ε This represents the total number of perturbation propagation steps, obtained by the step distance from the initial position to the target position and the perturbation step length.

[0059] For the self-organizing-other-organizing ratio (SvL), one specific implementation is as follows: S21. Obtain the concentration residual based on the original concentration information and reconstructed concentration information of chemical elements at each sampling point.

[0060] S22. Obtain the self-organized to other-organized ratio based on the concentration residual.

[0061] For S21 and S22, specifically, the SvL value (self-organization-other-organization ratio, where other-organization represents inert factors in the evolution process) of each target anomaly region is determined, the concentration residuals of the original concentration information and the reconstructed concentration information are obtained, the concentration residuals are corrected, and the self-organization-other-organization ratio is obtained.

[0062] In practice, the concentration residual D is usually used to replace the SvL value. If the box dimension of the residual image of the reconstructed concentration information and the original concentration information falls within the typical statistical range of natural landscapes, that is, the concentration residual D is within 1.2–1.6, it indicates that there are additional self-organizing processes or dynamic driving processes (such as fracture systems) that have affected the formation of the geochemical field. Conversely, if the dimension value is close to an integer, that is, the concentration residual D is close to 1.0 or 2.0, it reflects a strong regularity. Such characteristics are more likely to originate from geological processes without mineralization potential (such as sedimentation or mixing).

[0063] S204. Based on the Lyapunov index and the self-organization-other-organization ratio, the mineralization level of the geochemical anomaly in the target anomaly area is obtained.

[0064] In this step, the following two evaluation schemes are used to evaluate the mineralization: Option 1: Level A (High Priority): LE is infinitely close to 0, and SvL is high (>SvLCut) (corresponding to residual concentration D within 1.2–1.6).

[0065] Category B (Medium Priority): LE greater than 0 and SvL moderate (i.e., residual concentration difference D between 1.1 and 1.2 / 1.6 and 1.7).

[0066] Grade C (low priority): LE is less than 0 and SvL is low (i.e., residual D is close to 1 or 2 (possibly due to deposition or noise)).

[0067] As LE approaches 0, the closer LE is to 0, the more it indicates a self-organized critical state or the edge of chaos, representing the most favorable dynamic environment for mineralization. If LE > 0, the disturbance is exponentially amplified, the system is spatially chaotic, and it may correspond to a region of intense, multi-stage tectonic-hydrothermal superposition and alteration, which is extremely complex and has a high risk of mineral exploration. If LE < 0, the disturbance decays rapidly, the system is spatially stable, and it may correspond to a homogeneous sedimentary background or a region that has undergone intense homogenization, with low mineralization potential.

[0068] Option 2: Tier A (Highest Priority): f *LE+SvL normalization > 0.7 f This is the weighting coefficient; the closer it is to 0, the larger the value.

[0069] Tier B (Medium Priority): f *LE+SvL normalization is between 0.35 and 0.7.

[0070] Tier C (lower priority): f *LE+SvL normalization is less than 0.35.

[0071] To verify the effectiveness of the geochemical anomaly mineralization evaluation method proposed in this embodiment, a prediction effect was obtained using a region in Inner Mongolia as an example. The results are as follows: Figure 4 As shown, the mineralization evaluation method for geochemical anomalies in this embodiment is used to obtain the mineralization evaluation of each anomaly region in the target area.

[0072] In this embodiment, geochemical information of the target anomaly region to be evaluated for mineralization is obtained. Based on the geochemical information and a trained physical information neural network model, the reconstructed concentration information of chemical elements at each sampling point is obtained. The trained physical information neural network model includes a physical simulation layer embedded with a fractional-order reaction-diffusion equation. The physical simulation layer simulates the diffusion, convection, and reaction terms in the fractional-order reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional-order reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension. Then, based on the original and reconstructed concentration information of chemical elements at each sampling point, the Lyapunov index and self-organization-other-organization ratio of the target anomaly region are obtained, thereby obtaining the mineralization level of the geochemical anomaly region. By using a physical information neural network model, fractional-order reaction-diffusion equations describing mineralization geochemical processes are learned and simulated, making the reconstructed concentration information more consistent with mineralization laws. This allows for the acquisition of dynamic characteristics LE and SvL that reveal the mineralization process, enabling direct evaluation of the target anomaly area based on its "mineralization" rather than its "existence." Consequently, when the mineralization evaluation level of the target anomaly area is high, field verification or drilling can be conducted, reducing exploration costs.

[0073] Optionally, the physical simulation layer's corresponding operations include: convolution operations, residual connection operations, and offset operations based on dual deformable convolutional layers. Furthermore, as follows: Figure 3 As shown, one specific implementation of S201 is as follows: S2011. By simulating the physical process corresponding to the diffusion term through convolution operations based on a dual deformable convolutional layer, feature extraction is performed on geochemical information to obtain diffusion features.

[0074] Specifically, a dual deformable convolutional layer (without activation function) is used to simulate the diffusion term. The kernel size, stride, and dilatation rate of the first deformable convolutional layer are 3×3, 1, and 2, respectively, and the kernel size, stride, and dilatation rate of the second deformable convolutional layer are 5×5, 2, and 1, respectively. By setting the deformable convolution, the position of the convolutional kernel can be adjusted according to the boundary shape of the target abnormal region, and the boundary of the target abnormal region can be extracted to obtain diffusion features.

[0075] S2012. Simulate the physical process corresponding to the reaction term through residual connection operation, perform multiple feature extractions on diffusion features to obtain intermediate features, and perform feature extraction again on the intermediate features and diffusion features after feature fusion to obtain reaction features.

[0076] Specifically, the residual connection operation corresponds to the reaction term in the fractional reaction-diffusion equation. It extracts features from the diffusion features. After obtaining the first intermediate feature out1 by out1=ReLU(Conv1(c)), it extracts features from the first intermediate feature again by out2=Conv2(out) to directly obtain the second intermediate feature out2. Then, it extracts features from the feature fusion of the diffusion features, the first intermediate feature, and the second intermediate feature by out3=ReLU(BatchNorm, out2+residual), normalizes them, and obtains the reaction feature out3 by the activation function.

[0077] S2013. Simulate the physical process corresponding to the convection term through offset operation, and obtain the convection characteristics based on the reaction characteristics and the changes in chemical element concentrations in the first and second directions.

[0078] The angle between the first direction and the second direction is 90°.

[0079] Specifically, the first direction is the X-axis, and the second direction is the Y-axis. The convection term is handled using an "offset + learnable bias" approach, as shown below:

[0080] Where z represents the reaction characteristics before the update. for x The velocity component in the axial direction, for y The velocity component in the axial direction, This is the result of the reaction feature z being rolled 1 pixel along the x-axis before the update. This is the result of the reaction feature z being rolled 1 pixel along the y-axis before the update. and These represent the backward differences in the x-axis and y-axis directions, respectively.

[0081] S2014. Based on the convection characteristics, obtain the reconstructed concentration information of chemical elements at each sampling point.

[0082] Specifically, the convection features are processed using Conv3 (3×3 kernel size) + BatchNorm + ReLU activation function + Conv4 (1×1 kernel size) to obtain structural integration features and output reconstructed concentration information.

[0083] In this embodiment, by setting the operation process corresponding to the physical simulation layer to include: convolution operation based on a double deformable convolution layer, residual connection operation, and offset operation, the trained physical information neural network model obtains reconstructed concentration information based on the fractional-order reaction-diffusion equation. Furthermore, through this design, complex nonlinear relationships can be fitted, and gradient flow can be improved. For the reaction term, the initial input usually represents the baseline concentration, while each intermediate feature (i.e., the residual) captures the amount of concentration change, which is more in line with the inherent laws of mineralization geochemistry.

[0084] Optional, such as Figure 1 As shown, the physical information neural network model includes a coding layer 160 with embedded spectral separable modules; therefore, as Figure 3 As shown, prior to S2011, the method also included: S31. Based on the spectral separability module, the chemical elements in the geochemical information are decoupled to obtain the decoupled geochemical characteristics.

[0085] S32. Input the decoupled geochemical features as geochemical information into the physical simulation layer.

[0086] Specifically, for metallic ores, there are many types of chemical elements, usually more than 20 to 30. However, most of these chemical elements are not related to mineralization. Furthermore, there are cases where high concentrations of elements (such as Fe and Ca) mask low concentrations of important mineralizing elements (such as Au and Ag), which can lead to distorted mineralization assessments.

[0087] Therefore, by using the spectrally separable module (SSM), chemical elements in geochemical information are decoupled, features of each chemical element are extracted, important element combinations and spatial patterns among multiple chemical elements are obtained, and the chemical elements in geochemical information are reduced in dimensionality to obtain the features of chemical elements that are truly related to mineralization, i.e., geochemical features. This reduces multi-element coupling errors and improves the robustness of multi-element fusion.

[0088] Figure 5 This is a schematic diagram of the structure of a geochemical anomaly mineralization evaluation device provided in an embodiment of this application, as shown below. Figure 5As shown, the geochemical anomaly mineralization evaluation device includes: an acquisition module 501, a model processing module 502, and an evaluation module 503. Optionally, the geochemical anomaly mineralization evaluation device may also include: a training module 504.

[0089] The acquisition module 501 is used to acquire geochemical information of the target anomaly area to be evaluated for mineralization. The geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area. The model processing module 502 is used to obtain the reconstructed concentration information of chemical elements at each sampling point based on the geochemical exploration information and the trained physical information neural network model. The trained physical information neural network model includes a physical simulation layer embedded with a fractional reaction-diffusion equation. The physical simulation layer simulates the diffusion term, convection term, and reaction term in the fractional reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension. Evaluation module 503 is used to obtain the Lyapunov index and self-organization-other-organization ratio of the target anomaly area based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point, and to obtain the geochemical anomaly mineralization level of the target anomaly area based on the Lyapunov index and the self-organization-other-organization ratio.

[0090] Optionally, the operation process corresponding to the physical simulation layer includes: convolution operation, residual connection operation and offset operation based on the dual deformable convolution layer; The model processing module 502 obtains the reconstructed concentration information of chemical elements at each sampling point based on the geochemical information and the trained physical information neural network model, specifically for: The physical process corresponding to the diffusion term is simulated by the convolution operation based on the dual deformable convolution layer, and the geochemical information is used to extract features to obtain diffusion features. The physical process corresponding to the reaction term is simulated by the residual connection operation. The diffusion feature is extracted multiple times to obtain intermediate features. The intermediate features and the diffusion feature after feature fusion are extracted again to obtain the reaction feature. The physical process corresponding to the convection term is simulated by the offset operation. The convection characteristics are obtained based on the reaction characteristics and the changes in chemical element concentration in the first and second directions, where the angle between the first and second directions is 90°. Based on the convection characteristics, the reconstructed concentration information of chemical elements at each sampling point is obtained.

[0091] Optionally, the trained physical information neural network model includes an encoding layer with embedded spectral separable modules; The model processing module 502 simulates the physical process corresponding to the diffusion term through the convolution operation based on the dual deformable convolution layer, extracts features from the geochemical information, and before obtaining the diffusion features, it is also used for: The chemical elements in the geochemical information are decoupled according to the spectral separability module to obtain the decoupled geochemical characteristics. The decoupled geochemical features are input as geochemical information into the physical simulation layer.

[0092] Optionally, before obtaining the reconstructed concentration information of chemical elements at each sampling point based on the geochemical information and the trained physical information neural network model, the training module 504 is used for: The geochemical exploration information is used as training data and input into the initial physical information neural network model to obtain the predicted concentration information of chemical elements at each sampling point; The initial physical information neural network model is trained based on the original concentration information of chemical elements at each sampling point, the predicted concentration information, and the loss function to obtain the trained physical information neural network model. The loss function includes a physical residual loss function, a boundary condition loss function, and an initial condition loss function. The physical residual loss function is obtained through the predicted concentration information and the fractional-order reaction-diffusion equation. The boundary condition loss function is obtained through the predicted concentration information at the boundary of the target anomaly region and the original concentration information. The initial condition loss function is obtained through the average value of the predicted concentration information and the average value of the original concentration information.

[0093] Optionally, the evaluation module 503 obtains the Lyapunov index of the target anomaly region based on the original concentration information and the reconstructed concentration information of the chemical elements at each sampling point, specifically for: Based on the reconstructed concentration information, obtain the Jacobian matrix at each sampling point from the initial position to the target position; Based on the Jacobian matrix at each sampling point from the initial position to the target position, the perturbation vector corresponding to the initial position, and the perturbation step size, the perturbation vector corresponding to the target position is obtained. The Lyapunov exponent is obtained based on the perturbation vector corresponding to the initial position, the perturbation vector corresponding to the target position, and the total number of perturbation propagation steps. The total number of perturbation propagation steps is obtained based on the step distance from the initial position to the target position and the perturbation step size.

[0094] Optionally, based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point of the evaluation module 503, the self-organization-other-organization ratio of the target anomaly region is obtained, specifically for: Based on the original concentration information and the reconstructed concentration information of the chemical elements at each sampling point, the concentration residual is obtained; The self-organized to other-organized ratio is obtained based on the concentration residual.

[0095] The geochemical anomaly mineralization evaluation device provided in this application embodiment can be referred to the above method embodiment for its specific implementation process. Its implementation principle and technical effect are similar, and will not be repeated here.

[0096] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may be a server, such as... Figure 6 As shown, the electronic device includes a processor 601 and a memory 602.

[0097] The memory 602 stores computer-executed instructions.

[0098] The processor 601 executes the computer execution instructions stored in the memory 602, causing the processor 601 to perform the method described in any of the above embodiments.

[0099] The electronic device provided in this application embodiment can be referred to the above method embodiment for its specific implementation process. The implementation principle and technical effect are similar, and will not be repeated here.

[0100] In the above Figure 6 In the illustrated embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0101] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage.

[0102] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0103] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method shown in the above-described method embodiments.

[0104] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0105] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0106] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for evaluating the mineralization potential of geochemical anomalies, characterized in that, include: Obtain geochemical information of the target anomaly area to be evaluated for mineralization, wherein the geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area; Based on the geochemical exploration information and the trained physical information neural network model, the reconstructed concentration information of chemical elements at each sampling point is obtained. The trained physical information neural network model includes a physical simulation layer embedded with a physical simulation layer corresponding to the fractional reaction-diffusion equation. The physical simulation layer simulates the diffusion term, convection term, and reaction term in the fractional reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension. Based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point, the Lyapunov index and self-organization-other-organization ratio of the target anomaly region are obtained; The mineralization level of the geochemical anomaly in the target anomaly area is obtained based on the Lyapunov index and the self-organization-other-organization ratio.

2. The method according to claim 1, characterized in that, The physical simulation layer includes the following operations: convolution operation, residual connection operation, and offset operation based on a dual deformable convolutional layer; The process of obtaining the reconstructed concentration information of chemical elements at each sampling point based on the geochemical information and the trained physical information neural network model includes: The physical process corresponding to the diffusion term is simulated by the convolution operation based on the dual deformable convolution layer, and the geochemical information is used to extract features to obtain diffusion features. The physical process corresponding to the reaction term is simulated by the residual connection operation. The diffusion feature is extracted multiple times to obtain intermediate features. The intermediate features and the diffusion feature after feature fusion are extracted again to obtain the reaction feature. The physical process corresponding to the convection term is simulated by the offset operation. The convection characteristics are obtained based on the reaction characteristics and the changes in chemical element concentration in the first and second directions, where the angle between the first and second directions is 90°. Based on the convection characteristics, the reconstructed concentration information of chemical elements at each sampling point is obtained.

3. The method according to claim 2, characterized in that, The trained physical information neural network model includes an encoding layer with embedded spectral separable modules; Before simulating the physical process corresponding to the diffusion term through the convolution operation based on the dual deformable convolution layer to extract features from the geochemical information and obtain diffusion features, the method further includes: The chemical elements in the geochemical information are decoupled according to the spectral separability module to obtain the decoupled geochemical characteristics. The decoupled geochemical features are input as geochemical information into the physical simulation layer.

4. The method according to any one of claims 1-3, characterized in that, Before obtaining the reconstructed concentration information of chemical elements at each sampling point based on the geochemical information and the trained physical information neural network model, the method further includes: The geochemical exploration information is used as training data and input into the initial physical information neural network model to obtain the predicted concentration information of chemical elements at each sampling point; The initial physical information neural network model is trained based on the original concentration information of chemical elements at each sampling point, the predicted concentration information, and the loss function to obtain the trained physical information neural network model. The loss function includes a physical residual loss function, a boundary condition loss function, and an initial condition loss function. The physical residual loss function is obtained through the predicted concentration information and the fractional-order reaction-diffusion equation. The boundary condition loss function is obtained through the predicted concentration information at the boundary of the target anomaly region and the original concentration information. The initial condition loss function is obtained through the average value of the predicted concentration information and the average value of the original concentration information.

5. The method according to claim 1, characterized in that, The step of obtaining the Lyapunov index of the target anomaly region based on the original concentration information and the reconstructed concentration information of the chemical elements at each sampling point includes: Based on the reconstructed concentration information, obtain the Jacobian matrix at each sampling point from the initial position to the target position; Based on the Jacobian matrix at each sampling point from the initial position to the target position, the perturbation vector corresponding to the initial position, and the perturbation step size, the perturbation vector corresponding to the target position is obtained. The Lyapunov exponent is obtained based on the perturbation vector corresponding to the initial position, the perturbation vector corresponding to the target position, and the total number of perturbation propagation steps. The total number of perturbation propagation steps is obtained based on the step distance from the initial position to the target position and the perturbation step size.

6. The method according to claim 1, characterized in that, Based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point, the self-organized-other-organized ratio of the target anomaly region is obtained, including: Based on the original concentration information and the reconstructed concentration information of the chemical elements at each sampling point, the concentration residual is obtained; The self-organized to other-organized ratio is obtained based on the concentration residual.

7. A geochemical anomaly mineralization evaluation device, characterized in that, include: The acquisition module is used to acquire geochemical information of the target anomaly area to be evaluated for mineralization. The geochemical information includes the location information of each sampling point in the target anomaly area, the original concentration information of chemical elements at each sampling point, and the boundary information of the target anomaly area. The model processing module is used to obtain the reconstructed concentration information of chemical elements at each sampling point based on the geochemical exploration information and the trained physical information neural network model. The trained physical information neural network model includes a physical simulation layer embedded with a fractional reaction-diffusion equation. The physical simulation layer simulates the diffusion term, convection term, and reaction term in the fractional reaction-diffusion equation, so that the reconstruction process of the reconstructed concentration information follows the fractional reaction-diffusion equation. The diffusion term introduces the Hausdorff dimension. The evaluation module is used to obtain the Lyapunov index and self-organization-other-organization ratio of the target anomaly region based on the original concentration information and the reconstructed concentration information of chemical elements at each sampling point, and to obtain the geochemical anomaly mineralization level of the target anomaly region based on the Lyapunov index and the self-organization-other-organization ratio.

8. An electronic device, characterized in that, include: Processor and memory; The memory stores the instructions that the computer executes; The processor executes computer execution instructions stored in memory, causing the processor to perform the method according to any one of claims 1-6.

9. A readable storage medium, characterized in that, include: A program or instruction that, when run on a computer, executes the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1-6.

Images