An intelligent identification and visual analysis method for ore-forming mechanism

Abstract

The application provides a metallogenic mechanism intelligent identification and visual analysis method, relates to the field of geological resource exploration, and comprises the following steps: constructing a three-dimensional metallogenic probability body model based on multivariate geological data; delimiting a high-probability metallogenic target area based on the three-dimensional metallogenic probability body model, and extracting present mineral assemblage characteristics and whole-rock geochemical component data corresponding to the target area from a high-dimensional feature tensor; performing metallogenic environment thermodynamic inversion based on the principle of Gibbs free energy minimization, in combination with the present mineral assemblage characteristics and the whole-rock geochemical component data, to obtain mineral precipitation evolution sequence data; and performing four-dimensional space-chemical collaborative visualization and interactive decision of the metallogenic mechanism through the three-dimensional metallogenic probability body model and the mineral precipitation evolution sequence data to generate a drilling decision scheme. Through the deep integration of "data driving" and "mechanism verification", the application realizes four-dimensional dynamic reconstruction of the metallogenic mechanism, and significantly improves the accuracy of deep prospecting.

Description

Technical Field

[0001] This application relates to the field of geological resource exploration, and particularly to an intelligent recognition and visualization analysis method for ore-forming mechanisms. Background Art

[0002] With the increasing depletion of shallow mineral resources, prospecting work is accelerating the expansion towards deep buried deposits. Facing a large amount of heterogeneous and multi-source geological data, although artificial intelligence technology has been widely used in ore-forming prediction, there are still significant limitations in practical applications of the existing technologies. First of all, most of the existing AI prospecting methods are based on "data correlation" for statistical prediction, lacking effective constraints on physical and chemical ore-forming mechanisms. Due to the widespread phenomenon of "different substances with the same spectrum" in geological bodies, pure data-driven models are extremely likely to misjudge non-ore-induced anomalies as ore-induced anomalies, resulting in a high "false positive" rate of prediction results, being difficult to explain geological causality, and unable to meet the requirements of precise drilling in the deep part. Secondly, traditional ore-forming mechanism research (such as thermodynamic simulation) is mostly limited to academic theoretical explanations, and usually adopts "forward simulation" to restore the process. The existing technologies lack a method that can use the three-dimensional geological mineral characteristics observed today as constraints to "inverse back-invert" the paleotemperature, pressure and fluid composition during the ore-forming period. This leads to the difficulty of deeply integrating mature theoretical models with actual exploration data and being unable to directly feed back to prospecting practice. Finally, traditional geological visualization systems mainly focus on static three-dimensional geometric modeling, lacking the ability to display the dynamic evolution process and thermodynamic properties of ore formation. In particular, the existing systems cannot achieve real-time linkage analysis between the "three-dimensional geological space view" and the "two-dimensional thermodynamic phase diagram", resulting in the difficulty for geological personnel to visually evaluate the thermodynamic ore-forming favorability of the target area in three-dimensional space, and the data analysis and mechanism research being in a separated state. To sum up, there is an urgent need to develop an intelligent system that can deeply integrate data-driven deep learning prediction with mechanism-driven thermodynamic inversion and can intuitively display ore-forming mechanisms and processes through visualization means. Summary of the Invention

[0003] The purpose of the present invention is to: in order to overcome the deficiencies in the existing technology that the ore-forming prediction method relying solely on data driving lacks mechanism constraints, resulting in a high "false positive" rate, traditional thermodynamic simulations are mostly forward deductions and it is difficult to inverse back-invert paleoenvironmental parameters, and the existing geological visualization systems lack the ability of spatial-chemical linkage analysis, to provide an intelligent recognition and visualization analysis method for ore-forming mechanisms.

[0004] The above object of this application is achieved through the following technical solutions: S1: Construct a three-dimensional ore-forming probability volume model based on multi-source geological data, specifically including: Construct a high-dimensional feature tensor according to the multi-source geological data obtained for the target exploration area; A three-dimensional convolutional neural network with an embedded spatial attention mechanism is used to extract and classify features from high-dimensional feature tensors, and output a three-dimensional mineralization probability model of the target region. S2: Based on the three-dimensional mineralization probability volume model, high-probability mineralization target areas are delineated, and the current mineral assemblage characteristics and whole-rock geochemical composition data corresponding to the target areas are extracted from the high-dimensional feature tensor. S3: Based on the principle of minimizing Gibbs free energy, combined with current mineral assemblage characteristics and whole-rock geochemical composition data, thermodynamic inversion of the mineralization environment is performed to obtain mineral precipitation evolution sequence data; S4: Through the three-dimensional mineralization probability model and mineral precipitation evolution sequence data, four-dimensional spatial-chemical collaborative visualization and interactive decision-making of mineralization mechanism are carried out to generate drilling decision schemes.

[0005] An electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to perform a method for intelligent identification and visualization analysis of mineralization mechanisms.

[0006] A computer-readable storage medium storing instructions that, when executed, perform a method for intelligent identification and visualization analysis of mineralization mechanisms.

[0007] The beneficial effects of the technical solution provided in this application are: 1) Significantly reduced the false positive rate of mineralization prediction and improved the reliability of predictions. Existing technologies mostly rely solely on AI to mine statistical correlations in data, making them susceptible to interference from "heterogeneous objects with similar spectra." This invention creatively introduces "thermodynamic inversion" as a verification step. That is, it requires not only that the AI ​​"sees" the mineral (geometric shape similar to data characteristics), but also that thermodynamics "calculates" the mineral (physicochemical mechanism is reasonable). Only when the AI ​​prediction result matches the mineralization environment calculated by minimizing Gibbs free energy is it considered a valid target area. This dual constraint mechanism of "data + mechanism" effectively eliminates false anomalies that do not conform to the mineralization law.

[0008] 2) This invention achieves a reverse inversion from "current observations" to "paleoenvironmental parameters," filling a gap in the application of thermodynamic techniques in resource exploration. Traditional geochemical simulations are mostly forward theoretical deductions, which are difficult to directly serve mineral exploration. This invention proposes an automatic inversion algorithm based on the minimization of Gibbs free energy, which can use readily available rock and mineral data to reverse-engineer paleotemperature, paleopressure, and fluid properties at the time of mineralization. This transforms thermodynamic research from a purely theoretical laboratory setting into a practical tool for quantitatively assessing the deep extension potential of mineral deposits.

[0009] 3) It provides a visually interactive means for spatial-chemical collaboration to assist scientific decision-making. Addressing the limitation of traditional geological software in intuitively displaying chemical processes, this invention constructs a linked interface between three-dimensional geological space and two-dimensional thermodynamic phase diagrams. Geologists can directly view the thermodynamic stability of ore bodies in the three-dimensional model, or delineate favorable mineralization areas in the phase diagram to determine their spatial location. This intuitive interactive method significantly lowers the barrier to understanding complex mineralization mechanisms, providing a more scientific and intuitive basis for decision-making regarding deep borehole deployment.

[0010] 4) This invention enables the reconstruction and four-dimensional visualization of the dynamic mineralization process. It not only inverts static environmental parameters but also reconstructs the dynamic evolution sequence of mineral precipitation through reaction path simulation. By superimposing the temporal evolution process onto a three-dimensional spatial model, users can gain insight into the evolutionary trends of ore-forming fluids and the mineralization zoning patterns, overcoming the limitations of traditional methods that can only display the static final state.

[0011] 5) Effectively solves the problem of deep fusion of diverse and heterogeneous geological data. This invention employs three-dimensional voxelization and tensor construction techniques to uniformly map discrete borehole data, continuous geophysical field data, and sparse geochemical data into the same mathematical space. Furthermore, through a 3D-CNN network with an embedded attention mechanism, it effectively extracts weak nonlinear mineral exploration information from deep regions, thus solving the problem that traditional linear statistical methods struggle to handle complex geological features. Attached Figure Description

[0012] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a step diagram of an embodiment of this application; Figure 2 This is a schematic diagram of the layout of the first software interaction interface in the embodiments of this application; Figure 3 This is a schematic diagram of the layout of the second software interaction interface in an embodiment of this application; Figure 4 This is a schematic diagram of the layout of the third software interaction interface in the embodiments of this application; Figure 5 This is a schematic diagram of the electronic device structure in the embodiments of this application. Detailed Implementation

[0013] To provide a clearer understanding of the technical features, objectives, and effects of this application, the specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0014] The embodiments of this application provide a method for intelligent identification and visualization analysis of mineralization mechanisms.

[0015] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of an intelligent identification and visualization analysis method for mineralization mechanisms in an embodiment of this application, including: S1: Construct a three-dimensional metallogenic probability model based on multivariate geological data, specifically including: A high-dimensional feature tensor is constructed based on the multi-dimensional geological data of the target exploration area. A three-dimensional convolutional neural network with an embedded spatial attention mechanism is used to extract and classify features from high-dimensional feature tensors, and output a three-dimensional mineralization probability model of the target region. S2: Based on the three-dimensional mineralization probability volume model, high-probability mineralization target areas are delineated, and the current mineral assemblage characteristics and whole-rock geochemical composition data corresponding to the target areas are extracted from the high-dimensional feature tensor. S3: Based on the principle of minimizing Gibbs free energy, combined with current mineral assemblage characteristics and whole-rock geochemical composition data, thermodynamic inversion of the mineralization environment is performed to obtain mineral precipitation evolution sequence data; S4: Through the three-dimensional mineralization probability model and mineral precipitation evolution sequence data, four-dimensional spatial-chemical collaborative visualization and interactive decision-making of mineralization mechanism are carried out to generate drilling decision schemes.

[0016] This application, by adopting the above-mentioned technical solution, aims to overcome the shortcomings of existing technologies, such as the lack of mechanistic constraints in data-driven mineralization prediction methods leading to high false positive rates, the difficulty of reverse inversion of paleoenvironmental parameters in traditional thermodynamic simulations, and the lack of spatial-chemical linkage analysis capabilities in geological visualization systems. This invention provides an intelligent identification and visualization analysis method for mineralization mechanisms based on multivariate data fusion and thermodynamic inversion. This method uses a deep learning model to mine nonlinear mineralization patterns in multivariate geological data, predicts mineralization probabilities, and delineates high-potential target areas. Subsequently, based on the delineated target area locations, it extracts current mineral assemblage characteristics from observational data as constraints, uses the Gibbs free energy minimization principle to reverse invert the thermodynamic environment of the mineralization period, and constructs a visualized linkage decision-making mechanism, thereby improving the accuracy and scientific rigor of deep mineral exploration.

[0017] This application, employing the aforementioned technical solution, utilizes three-dimensional tensor fusion of multivariate data and AI to initially identify mineralization anomalies. It then innovatively introduces thermodynamic inversion for mechanism verification. Based on the principle of Gibbs free energy minimization, this technology reverse-engineers the paleotemperature, pressure, and fluid properties during mineralization, forming a dual criterion of "data-driven + mechanism-constrained" analysis, significantly reducing false positives. Simultaneously, it constructs a linked visualization interface between a spatial geological model and a thermodynamic phase diagram, intuitively presenting complex chemical processes and providing quantitative and reliable decision-making basis for deep mineral exploration.

[0018] Step S1 includes: The multivariate geological data are uniformly mapped to a voxel grid in a three-dimensional Cartesian coordinate system to construct a high-dimensional feature tensor, specifically including: Multi-source geological data includes: geological borehole data, geophysical inversion data, and geochemical sampling data; Establish a unified three-dimensional voxel mesh; Based on data density, interpolation strategies are selected for discrete geological borehole data and geophysical inversion data: When the data density is greater than the first preset threshold, a geological body is constructed and voxelized using an explicit modeling method based on geological rules. When the data density is less than the second preset threshold, a geostatistical interpolation algorithm is used to convert it into a voxel lithology coding channel; For continuous geophysical inversion data, a three-dimensional interpolation algorithm is used to map the geophysical inversion data from its original grid and sample it onto the three-dimensional voxel grid to form a physical property channel; Continuous geophysical inversion data includes gravity, magnetic, and electromagnetic data; For sparse geochemical sampling data, an interpolation algorithm that can characterize the spatial variability of element concentration is used to extend it into a three-dimensional element concentration channel. The lithology coding channel, physical property channel, and elemental concentration channel were normalized or standardized respectively. The processed channel data are stacked along the channel dimension to form a shape. The multi-channel high-dimensional feature tensor, where C is the total number of channels, and X, Y, and Z are the dimensions of the three-dimensional voxel grid.

[0019] In one specific embodiment of this application, step S1 aims to solve the spatial fusion problem of diverse heterogeneous data and utilize deep learning to mine deep nonlinear mineralization characteristics, providing accurate spatial constraints for subsequent thermodynamic inversion, such as... Figure 2 As shown.

[0020] S101: Acquisition, Cleaning, and Standardization Preprocessing of Multivariate Geological Data The system collects multi-source, multi-scale basic geological data of the target exploration area and performs digitization and cleaning.

[0021] Basic geological data includes surface geological maps at different scales (vector format, containing stratigraphic boundaries, fault structures, and igneous rock distribution) and borehole logging data (including the three-dimensional coordinates of the borehole opening, inclinometer data, and micromineralogical and structural features of layered lithology). Paper maps must be vectorized; borehole data must be filtered to remove outliers caused by data entry errors. Geophysical inversion data: Collect regional gravity anomaly and aeromagnetic anomaly data, as well as geophysical inversion profile data that can reflect deep structures, such as resistivity and polarizability models of deep high-power induced polarization (IP) and inversion profile data of controlled source audio-frequency magnetotelluric sounding (CSAMT). Geochemical data: The system collects geochemical sampling data of rocks of different lithologies, measurement data of stream sediments, and mineral content and elemental analysis data of borehole core samples (elemental analysis mainly focuses on mineralization indicator elements and alteration indicator elements). Data standardization: Establish a unified three-dimensional Cartesian coordinate system (e.g., using the CGCS2000 coordinate system and Gauss-Kruger projection). Utilize coordinate transformation algorithms to project all two-dimensional planar data (basic geological and geophysical exploration maps) and one-dimensional linear data (boreholes) into this three-dimensional spatial coordinate system, ensuring consistent spatial reference.

[0022] S102: Construction of high-dimensional feature tensors based on three-dimensional voxel technology: To address the differences in spatial resolution and dimensionality between different source data, this embodiment employs three-dimensional voxelization technology to construct a unified data cube.

[0023] Mesh generation: The target subsurface space is divided into a regular three-dimensional voxel mesh. Considering the accuracy requirements of deep mineral exploration and the balance of computational resources, a voxel resolution of L is preferred. W H (e.g., 20m) 20m 20m).

[0024] Multi-channel attribute mapping (ChannelMapping): Channel A (Geological Structure Channel): Using explicit modeling techniques or indicator kriging, geological boundaries and borehole lithology are extrapolated to each voxel, assigning it a discrete lithology code value (e.g., 1 represents granodiorite, 2 represents marble, and 3 represents plagioclase amphibolite).

[0025] Channel B (Physical Field Channel): For continuous geophysical inversion data, three-dimensional linear interpolation or spline interpolation algorithms are used to resample the data from the original inversion grid onto a unified voxel grid, forming channels for physical properties such as density, magnetic susceptibility, and resistivity.

[0026] Channel C (Geochemical Channel): For sparsely distributed geochemical sampling data, in order to preserve local anomaly characteristics, fractal interpolation or inverse distance weighting (IDW) is preferred to extend it to three-dimensional space to form concentration field channels for each element.

[0027] High-dimensional tensor construction: Z-score normalization or min-max normalization is performed on each continuous channel data to eliminate dimensional differences. Finally, all channels are stacked in the depth direction to construct a high-dimensional feature tensor of size (C,D,H,W), where C is the total number of feature channels, and D,H,W are the depth, height, and width in three-dimensional space.

[0028] S103: Constructing a 3DU-Net recognition model with an embedded attention mechanism: This embodiment designs an improved three-dimensional convolutional neural network (3D-CNN) for mineralization prediction, aiming to automatically extract nonlinear mineralization features from high-dimensional tensors.

[0029] Network Architecture: The 3DU-Net architecture is adopted as the backbone network, which consists of an encoder and a decoder. The encoder extracts multi-scale deep geological features (background field) through continuous convolution and downsampling operations. The decoder restores spatial resolution and locates mineralization anomalies through upsampling and feature concatenation operations.

[0030] Spatial Attention Module: A spatial attention module is embedded in the skip connection layer between the encoder and decoder. This module adaptively weights the feature maps by calculating the spatial weight matrix of the feature maps, automatically suppressing the feature responses of the background surrounding rocks (non-mineralized areas), and significantly enhancing the feature weights of key ore-controlling areas such as fault intersections, rock mass contact zones, and geophysical and geochemical anomalies, thereby improving the model's ability to capture weak information at depth.

[0031] Loss function optimization: To address the severe sample imbalance problem in geological prospecting, characterized by "very few ore body voxels and very large surrounding rock voxels," the model training does not use the conventional cross-entropy loss function but instead employs the FocalLoss loss function. This function reduces the weight of easily classified samples (a large number of surrounding rocks), forcing the model to focus on the scarce ore body samples that are difficult to classify, thereby improving the prediction recall.

[0032] S104: Model Training, Mineralization Probability Prediction, and Constraint Extraction Data augmentation and training: The voxel model of the known ore deposit is randomly rotated, mirrored, and Gaussian noise is added to expand the training dataset, and the above 3DU-Net model is trained until convergence.

[0033] Mineralization probability prediction: Input the feature tensor of the region to be predicted into the trained model, and output a mineralization probability volume with the same size as the input space. The voxel value is between [0,1], and the closer the value is to 1, the greater the probability of mineralization.

[0034] Key feature extraction (aimed at providing observational constraints for the thermodynamic inversion in step S2): Target delineation: Set a probability threshold (e.g., P>0.75), binarize the mineralization probability volume, and generate a spatial mask for target areas with high mineralization potential.

[0035] Feature backtracking and extraction: Using the spatial mask of the high mineralization potential target area, backtracking to the original high-dimensional feature tensor constructed in step S102, and extracting the corresponding geological and geochemical attributes within the mask range.

[0036] Data aggregation: Statistically analyze the attribute values ​​of each voxel within the masked area, and extract the "current mineral assemblage characteristics" (based on lithological statistics of geological channels to determine the occurrence state of the main host rocks and altered minerals) and "whole-rock geochemical average composition" (based on the average elemental concentration of geochemical channels) of the target area.

[0037] Output: These true observations extracted from the raw data will be passed to step S2 as “Ground Truth” to constrain the thermodynamic inversion process.

[0038] Step S1 also includes: The three-dimensional convolutional neural network adopts a 3D U-Net architecture, and a spatial attention module is embedded in the skip connections of the corresponding layers of the encoder and decoder of the three-dimensional convolutional neural network. The training of the three-dimensional convolutional neural network uses the Focal Loss loss function.

[0039] As one embodiment, the spatial attention module is used to calculate the spatial weight matrix of the feature map and adaptively weight the multi-scale feature map transmitted by the encoder in terms of spatial dimension to enhance the feature response of key ore-controlling parts such as the intersection of fault structures and geophysical and geochemical anomaly areas; the training of the three-dimensional convolutional neural network adopts the Focal Loss loss function to alleviate the training difficulties caused by the severe imbalance in the number of voxels of target bodies (such as ore bodies) and non-target bodies (such as surrounding rocks) in three-dimensional geological modeling.

[0040] Step S3 includes: S31: Using current mineral assemblage characteristics and whole-rock geochemical composition data as observational constraints, a multi-component, multi-phase thermodynamic system is constructed through thermodynamic modeling units; S32: Based on the principle of minimizing Gibbs free energy, a forward model is constructed, and combined with a multi-component multiphase thermodynamic system, the theoretical equilibrium mineral assemblage is simulated under different temperature, pressure and fluid composition conditions; S33: Construct the objective function and calculate the difference between the theoretical equilibrium mineral assemblage and the observational constraints; S34: Iteratively adjust the paleotemperature, paleopressure and initial fluid composition parameters using an optimization algorithm until the objective function converges to the minimum value, and invert the paleometallogenic environment parameters at the time of ore deposit formation; S35: Reaction path simulation based on ancient mineralization environment parameters to reconstruct mineral precipitation evolution sequence data.

[0041] In one specific embodiment of this application, step S3 aims to utilize physicochemical mechanisms to verify the mechanism of the target area delineated in step S1, and to reverse-engineer key environmental parameters and evolution processes during the formation of the ore deposit, such as... Figure 3 As shown. The specific implementation process includes the following sub-steps: S301: Construction and Mapping of Multicomponent Thermodynamic Systems First, the extracted geological information is converted into a chemical model required for thermodynamic calculations.

[0042] Input data mapping: From the delineated high-potential target area, select the most representative set of voxels (or the target area center voxel). Extract the current mineral assemblage observations at that location from the original high-dimensional feature tensor. ) and the abundance of major and trace elements in the whole rock.

[0043] Example data (selected from a large rare metal deposit in Northwest China): Rock mineral types and proportions:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] Whole-rock geochemical composition of major and trace elements:

[0056]

[0057] ;

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068] System Definition: Independent components are defined based on the whole-rock major and trace element geochemical composition of the rock. They define the minimum set of independent chemical components required to describe the geological body, for example:

[0069] in Represents electrons and is used to handle redox reactions.

[0070] Species sets are defined based on rock and mineral types and proportions, including pure mineral phases (such as quartz and pyrite), solid solution phases (such as feldspar and chlorite solid solutions), and aqueous solution species (such as...). , ) and gas phase (such as , ).

[0071] Database access: Load standard thermodynamic databases (such as SUPCRT92 or LLNL databases) to obtain standard Gibbs free energy data Δ for each species over a wide temperature and pressure range (e.g., 0-600℃, 1-5000 bar). and equilibrium constant.

[0072] S302: Constructing a forward simulation engine for minimizing Gibbs free energy (GEM) A forward model is established as the core computational unit for inversion iteration. The engine's function is to perform calculations under a given set of assumed ore-forming environment parameters (temperature T, pressure P, initial fluid composition). ), calculate the theoretical mineral assemblage when the system reaches thermodynamic equilibrium.

[0073] Mathematical model (minimization of Gibbs free energy in multi-component, multi-phase, and multi-system conditions): Under isothermal and isobaric conditions (T, P), closed or open geochemical systems tend to have a total Gibbs free energy. minimize.

[0074] The optimization problem is described as follows: For a closed geochemical system, the objective function is:

[0075] For the development of aqueous solution-mineral systems:

[0076] Parameter details: For the first The amount of substance (mol) of each species (ion, complex, mineral, gas). This is the decision variable to be solved.

[0077] It is the standard chemical potential of a substance.

[0078] R is the ideal gas constant. .

[0079] T is the absolute temperature of the system. ).

[0080] For the first The activity of a species is used to correct for biases in non-ideal systems.

[0081] For aqueous solutions ,in The activity coefficient is usually calculated using the Debye-Hückel or Pitzer equations and is a function of the ionic strength I. It is molar mass concentration ( ). It is the standard concentration state ( This is used to ensure that the logarithmic terms are dimensionless internally. For pure minerals / solids... That is, the chemical potential of a solid is usually determined by the standard state (except for solid solutions). For pure water... .

[0082] Constraints: 1) Conservation of mass:

[0083] in For stoichiometric matrix elements, Let be the total number of moles of the j-th element.

[0084] 2) Charge conservation: The total charge of the system is zero (using the principle of electroneutrality to constrain ion concentration).

[0085] 3) Nonnegativity constraint: ≥0 (the amount of substance cannot be negative).

[0086] In an open environment, inert components maintain mass conservation, and their total amount depends on the initial chemical composition. In contrast, active components are controlled by an external fluid reservoir; their quantity within the system is not conserved, but rather they freely enter and exit based on chemical potential equilibrium.

[0087] Open geochemical systems, objective function:

[0088] Parameter details: This is the total Gibbs free energy within the system (calculated as a closed system).

[0089] It is a set of active components.

[0090] For the first in the external fluid pool The chemical potential of a certain active component.

[0091] Constraints: In open systems, the constraints of charge conservation and nonnegativity still need to be observed, but the constraint of mass conservation only applies to inert components.

[0092] Solution Algorithm: The interior point method or sequential quadratic programming (SQP) is used to solve the above nonlinear constrained optimization problem, and the theoretical equilibrium mineral assemblage under the assumed environmental conditions is output. ).

[0093] S303: Constructing the objective function for the “simulation-observation” difference To perform the inversion, an evaluation function must be defined to measure the reasonableness of the "hypothetical paleoenvironment." The criterion for reasonableness should be whether the theoretical mineral assemblages generated by this environment are consistent with the extracted actual mineral assemblages.

[0094] Objective function definition: J(T, P, ) =

[0095] in: and These represent the mole fraction or percentage of mineral content for the k-th mineral in the thermodynamic simulation results and actual observations (from step S1), respectively.

[0096] The weighting coefficients are used. This embodiment specifically employs a non-uniform weighting strategy, assigning high weights to indicator minerals. >1.0 (Determined based on the main prospecting targets of different deposits. For example, for gold deposits, the main targets are gold, chalcopyrite, galena, and pyrite), while common gangue minerals (such as quartz, feldspar, or calcite) are assigned low weights. <0.5), this setting ensures that the inversion results preferentially match the mineralization characteristics, rather than just fitting the characteristics of the surrounding rocks.

[0097] λ This is a regularization term used to prevent overfitting of parameters.

[0098] S304: Iterative Inversion of Metallogenic Paleoenvironment Parameters The purpose of this step is to automatically find the objective function using optimization algorithms. The minimized environmental parameters, namely the "most likely mineralized paleoenvironment".

[0099] Inversion variable (independent variable): The parameter to be determined is the paleotemperature of the mineralization period ( ), ancient pressure ( pH and oxygen fugacity (log) of the initial ore-forming fluid f O2 ) and the concentration of ligands related to mineralization.

[0100] Iterative process: Initialization: within a reasonable geological range (e.g.) Randomly generate the initial population or initial point.

[0101] Iterative optimization: A hybrid genetic algorithm is employed. First, a global search is performed using a genetic algorithm to avoid getting trapped in local optima; once the region nearing the optimal solution is reached, gradient descent is used for a finer search. Each iteration calls the GEM engine from step S302 for computation. And use S303 to calculate the error. .

[0102] Convergence criterion: When the objective function value J is less than a preset threshold Or stop when the maximum number of iterations is reached.

[0103] Output: Output the optimal solution. This represents the most likely paleoenvironmental parameters for mineralization in the target area.

[0104] S305: Simulation of Dynamic Reaction Path in Ore Formation After obtaining the optimal static initial conditions, the dynamic process of mineralization is further simulated to support the four-dimensional visualization in step S4.

[0105] Simulation mechanism: Based on the initial fluid obtained through inversion, the simulation simulates its process of rising along the fracture, boiling and cooling, or interacting with the surrounding rock. Reaction progress variable ξ and temperature gradient variable are set. or pressure gradient variable As a driving force.

[0106] Evolutionary calculation: Gradually change ξ, T, or P (e.g., simulate from...) Cool to The equilibrium state of the system at each tiny step size is continuously calculated using the GEM principle.

[0107] Results Generation: A time-series mineral precipitation map is generated, recording which minerals began to precipitate, reached their peak, and ceased precipitation as the evolution process progressed. This sequence data serves as direct evidence of the "mineralization process" and is passed to step S4 for constructing a virtual borehole evolution chart.

[0108] The forward model constructed based on the principle of minimizing Gibbs free energy specifically includes: Under constant temperature and pressure conditions, with the goal of minimizing the total Gibbs free energy of the system, an optimization model is constructed to solve for the equilibrium molar amounts of each species in a multi-component multiphase thermodynamic system. The constraints of the optimization model include: For a closed system environment, the mass conservation constraints, charge conservation and nonnegativity constraints of all components are satisfied; For open system environments, the following constraints must be met: mass conservation of inert components, chemical potential or activity limitation of active components, and charge conservation and nonnegativity. The active components include: , and ; The total Gibbs free energy of the system Represented as:

[0109] in, This represents the total number of species within the system. For the first The number of moles of each species Standard chemical potential, The gas constant is... Thermodynamic temperature For activity; The objective function is constructed as follows: The weighted sum of squares of deviations between the mole fraction of each mineral phase in the theoretical equilibrium mineral assemblage and the mole fraction of the actual mineral phase is calculated; wherein, the weight coefficient assigned to the mineralization indicator metallic minerals is higher than the weight coefficient of gangue minerals, and the objective function also includes a regularization term for constraining the range or smoothness of the inversion parameters. The reaction path simulation specifically includes: using the optimal paleomagnetic environment parameters obtained by inversion as the initial state of the system, and introducing reaction progress variables or temperature and pressure gradient variables; By using the principle of minimizing Gibbs free energy, the equilibrium state of the calculation system at different evolution stages is serialized, generating mineral precipitation evolution sequence data that changes with the evolution process.

[0110] Step S4 includes: Spatial registration and fusion of the three-dimensional mineralization probability model and mineral sedimentation evolution sequence data; By using visualization units and combining spatially registered and fused data, a linked interactive interface is constructed that includes a three-dimensional geological spatial view and a two-dimensional thermodynamic phase diagram. A drilling decision-making scheme is generated by using a linked interactive interface and normalized comprehensive evaluation indicators. The interactive interface has the following characteristics: Establish a two-way mapping relationship between the three-dimensional voxel index and the two-dimensional thermodynamic state space; When a user selects a specific spatial region in the 3D geological spatial view, the 2D thermodynamic phase diagram view projects and displays the set of thermodynamic state parameter points corresponding to that region in real time. When a user defines a specific mineral stability region or parameter range in a two-dimensional thermodynamic phase diagram using interactive components, the three-dimensional geological space view filters and highlights the space voxels that meet the thermodynamic conditions in real time. The comprehensive evaluation index is a weighted sum of the mineralization probability and the normalized thermodynamic mineralization favorability; The thermodynamic mineralization favorability is used to characterize the physicochemical trend of mineral precipitation and is obtained by calculating the reaction affinity or Gibbs free energy change of the target mineralization reaction under the current thermodynamic conditions.

[0111] As one embodiment, the system normalizes the thermodynamic mineralization favorability and maps it to a dimensionless interval with the same mineralization probability. The system weights and fuses the mineralization probability with the normalized thermodynamic mineralization favorability to generate a comprehensive evaluation index and maps it to a three-dimensional decision cloud map. Areas that simultaneously meet the requirements of high mineralization probability and high thermodynamic mineralization favorability are marked as recommended drilling target areas.

[0112] This application provides an intelligent identification and dynamic inversion system for mineralization mechanisms, comprising: an intelligent identification module for voxelizing and tensor construction of multivariate geological data, predicting three-dimensional mineralization probabilities using a deep learning model, and backtracking to extract the current mineral assemblage features and whole-rock geochemical composition corresponding to high-probability target areas from the high-dimensional feature tensor; a thermodynamic inversion module for constructing a forward model based on the principle of minimizing Gibbs free energy, using the current mineral assemblage features as constraints, and iteratively solving parameters such as paleotemperature, paleopressure, and fluid composition at the time of ore deposit formation through an optimization algorithm, and performing reaction path simulation to reconstruct the mineralization evolution sequence; and a visualization and interaction module for constructing a two-way linkage analysis interface between the three-dimensional geological model and the thermodynamic phase diagram, and generating a visualization of the mineralization evolution process and a drilling decision scheme based on normalized comprehensive evaluation indicators.

[0113] As one embodiment, step S3 aims to construct an integrated visualization analysis terminal. By establishing a two-way mapping mechanism between three-dimensional geological space (geometric field) and two-dimensional thermodynamic and chemical space (mechanical field), it assists technicians in intuitively identifying true and false geological anomalies and making exploration decisions. Figure 4 As shown. The specific implementation process includes the following sub-steps: S401: Spatial Registration, Fusion and Normalization of Multidimensional Data The static geometric data generated in step S1 and the dynamic physicochemical data obtained by inversion in step S2 are subjected to unified spatial mapping and data cleaning.

[0114] Data loading and structure definition: The system receives the ore-forming probability model output in step S1. ).

[0115] The system receives the paleothermodynamic field data output in step S3, including the scalar field (paleotemperature). Ancient pressure , (value) and dynamic evolution sequence data (mineral precipitation amount as a function of reaction progress) generated in S305. (The change curve).

[0116] Attribute fusion (constructing a supervoxel): Each spatial voxel is assigned a multidimensional attribute vector. To support four-dimensional analysis, the vector contains static and dynamic components:

[0117] in For mineralization reaction affinity, This is the time series dataset of mineral precipitation corresponding to this location.

[0118] Data normalization processing (core logical correction): targeting mineralization probability With thermodynamic parameters (such as affinity) To address the issue of dimensional inconsistency, the Min-Max normalization method is used to map the thermodynamic parameters to... Interval:

[0119] like If the value is normalized, it is set to 0 or a negative value to distinguish the mineralization favorability.

[0120] Dynamic Transfer Function Settings: The system interface provides a "Properties-Visual" mapping editor. Users can set joint filtering rules. Rule Example 1: Only when (AI predicts high mineralization probability) and When the thermodynamics is extremely favorable, the opacity of the voxel is set to 1.0 and the color is red (preferably for ore-forming target areas).

[0121] Rule Example 2: When (AI predicts high probability of mineralization) but When the thermodynamics is unfavorable, the opacity of the voxel is set to 0.5 and the color is gray (false anomaly area).

[0122] Geological effects: Through GPU-accelerated volume rendering, the system interface automatically filters out false areas that "look like minerals (AI misjudgment), but the chemical conditions cannot support minerals (mechanistic denial)".

[0123] S402: Constructing an interactive interface linking three-dimensional geological space and two-dimensional thermodynamic phase diagrams. A linkage interaction mode based on "bi-directional indexing" technology was designed. The system maintains a mapping table in memory, which maps the IDs of three-dimensional voxels to the coordinates of state points in the two-dimensional phase diagram.

[0124] 1) Positive Linkage: From Space to Mechanism Operation: In the system's 3D view, users can use the mouse to perform ray casting picking, click on an anomaly with a high probability of mineralization, or select a region of interest along a fault zone.

[0125] Response: When the user selects a region, the two-dimensional thermodynamic phase diagram interface (e.g.) Image or The phase diagram immediately highlights the projection of the state point corresponding to the selected voxel.

[0126] Decision Analysis: Geologists observe the distribution of these points on the facies diagram. If the state points are densely concentrated within the "metallogenic stable region," it indicates that the AI ​​not only predicts mineralization in the area, but the paleoenvironment is indeed conducive to mineral precipitation and deposit formation, with extremely high reliability. If the state points fall within the "metallogenic unstable" region, it indicates that only alteration or weak mineralization may have occurred, but the physicochemical conditions for mineralization have not been met, and the area is judged as weakly mineralized or non-mineralized.

[0127] 2) Reverse linkage: From Mechanism to Space Operation: In the system's thermodynamic phase diagram interface, the user uses the selection tool to check the "optimal ore-forming temperature and pressure range" (e.g., , Alternatively, click on "Stability Zone for Specific Minerals (e.g., Niobite)".

[0128] Response: In the system's 3D view, the system automatically searches for all voxels and immediately highlights or displays voxels that meet the thermodynamic screening criteria, while the remaining voxels that do not meet the criteria are hidden or made transparent.

[0129] Decision analysis: This operation can visually demonstrate how "invisible" chemical fields are distributed in "visible" geometric space, helping geologists track the migration fronts and sedimentation centers of ore-forming fluids. S403: Virtual Drilling Analysis of Ore-forming Process Evolution To transform abstract inversion data into engineering language, the system provides virtual drilling functionality, supporting “slice-style” four-dimensional analysis of deep data.

[0130] Operation: Users can interactively draw a virtual drilling trajectory (design hole) in a 3D scene, and set the drilling depth and orientation angle.

[0131] Response: The system performs spatial interpolation along the trajectory, extracts the attribute data of each sampling point on the path and the associated S305 evolution sequence, and pops up the "Metallogenic Evolution Profile".

[0132] Chart content: The board contains three parallel tracks.

[0133] 1) Track 1 (lithology prediction): shows the distribution of lithology and structure along the borehole identified by S1.

[0134] 2) Orbit 2 (Paleoenvironment Inversion): Displays the inverted paleotemperature curve ( ), paleopressure curves ( ) and fluid Characteristic values ​​such as value change curves.

[0135] 3) Orbit 3 (Dynamic Mineral Sedimentation Sequence): One of the core features of this invention. For key mineralized sections in the borehole, the "time-sedimentation" relationship is displayed in the form of a heatmap or packing map, that is, how the mineral assemblage at this depth changes as the reaction progresses (evolution time) (e.g., early potassium feldspar, sodium feldspar alteration). Mid-term chloritization and muscovification accompanied by niobium and rare earth mineral precipitation Late quartz-calcite vein stage accompanied by minor niobium and rare earth mineralization.

[0136] Geological significance: Geologists can use this system to determine whether there are "multi-stage superposition" or "mineralization zoning" phenomena at depth, thereby optimizing the borehole depth design (for example, although only quartz veins and a small amount of niobium and rare earth mineralization are seen at the current depth, the evolution sequence shows that the deep fluid has a tendency to extend to the niobium and rare earth mineralization at depth. The system can suggest deepening the borehole to assist geologists in making decisions and optimizing the design of the borehole project).

[0137] S404: Generate an engineering drilling decision map based on dual constraints of "AI + thermodynamics". This is the final output of the system, providing suggestions for engineering construction.

[0138] Algorithm logic: The system background runs a comprehensive evaluation model to calculate the mineralization confidence score for each voxel. ):

[0139] in N represents the weighting coefficient. It is the normalized thermodynamic mineralization favorability.

[0140] Visualized output (decision-making system): Based on the score and individual indicator thresholds, the system generates a three-color decision cloud map (Traffic Light System). 1) Red area (preferred target area): This area represents a high probability of AI prediction and thermodynamic conditions that fully support the precipitation of ore-forming materials. The system automatically extracts the coordinates of this area as the preferred drilling verification point, calculates and outputs the suggested preferred drilling verification hole location (including three-dimensional coordinates, azimuth, and dip).

[0141] 2) Yellow area (secondary target area): This indicates a geological anomaly, but thermodynamic inversion shows the environment is in a critical state. The system will mark this area and suggest actions such as low-cost surface geochemical exploration, electrical verification, or sampling verification.

[0142] 3) Gray area (exclusion zone): This indicates that thermodynamic inversion shows that the area is an unfavorable environment for mineralization, and it is determined to be a non-mineralized anomaly. It is recommended to avoid it.

[0143] This application also discloses an electronic device. (See reference...) Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0144] The communication bus 502 is used to enable communication between these components.

[0145] The user interface 503 may include a display screen, and optionally, the user interface 503 may also include a standard wired interface or a wireless interface.

[0146] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0147] This application also discloses a computer-readable storage medium storing multiple instructions adapted for loading by a processor to execute the aforementioned intelligent identification and visualization analysis method for mineralization mechanisms.

[0148] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure.

[0149] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for intelligent identification and visualization analysis of mineralization mechanisms, characterized in that, The method includes the following steps: S1: Construct a three-dimensional metallogenic probability model based on multivariate geological data, specifically including: A high-dimensional feature tensor is constructed based on the multi-dimensional geological data of the target exploration area. A three-dimensional convolutional neural network with an embedded spatial attention mechanism is used to extract and classify features from high-dimensional feature tensors, and output a three-dimensional mineralization probability model of the target region. S2: Based on the three-dimensional mineralization probability volume model, high-probability mineralization target areas are delineated, and the current mineral assemblage characteristics and whole-rock geochemical composition data corresponding to the target areas are extracted from the high-dimensional feature tensor. S3: Based on the principle of minimizing Gibbs free energy, combined with current mineral assemblage characteristics and whole-rock geochemical composition data, thermodynamic inversion of the mineralization environment is performed to obtain mineral precipitation evolution sequence data; S4: Through the three-dimensional mineralization probability model and mineral precipitation evolution sequence data, four-dimensional spatial-chemical collaborative visualization and interactive decision-making of mineralization mechanism are carried out to generate drilling decision schemes.

2. The intelligent identification and visualization analysis method for mineralization mechanisms as described in claim 1, characterized in that, Step S1 includes: The multivariate geological data are uniformly mapped to a voxel grid in a three-dimensional Cartesian coordinate system to construct a high-dimensional feature tensor, specifically including: Multi-source geological data includes: geological borehole data, geophysical inversion data, and geochemical sampling data; Establish a unified three-dimensional voxel mesh; Based on data density, interpolation strategies are selected for discrete geological borehole data and geophysical inversion data: When the data density is greater than the first preset threshold, a geological body is constructed and voxelized using an explicit modeling method based on geological rules. When the data density is less than the second preset threshold, a geostatistical interpolation algorithm is used to convert it into a voxel lithology coding channel; For continuous geophysical inversion data, a three-dimensional interpolation algorithm is used to map the geophysical inversion data from its original grid and sample it onto the three-dimensional voxel grid to form a physical property channel; Continuous geophysical inversion data includes gravity, magnetic, and electromagnetic data; For sparse geochemical sampling data, an interpolation algorithm that can characterize the spatial variability of element concentration is used to extend it into a three-dimensional element concentration channel. The lithology coding channel, physical property channel, and elemental concentration channel were normalized or standardized respectively. The processed channel data are stacked along the channel dimension to form a shape. The multi-channel high-dimensional feature tensor, where C is the total number of channels, and X, Y, and Z are the dimensions of the three-dimensional voxel grid.

3. The intelligent identification and visualization analysis method for mineralization mechanisms as described in claim 1, characterized in that, Step S1 also includes: The three-dimensional convolutional neural network adopts a 3D U-Net architecture, and a spatial attention module is embedded in the skip connections of the corresponding layers of the encoder and decoder of the three-dimensional convolutional neural network. The training of the three-dimensional convolutional neural network uses the Focal Loss loss function.

4. The intelligent identification and visualization analysis method for mineralization mechanisms as described in claim 1, characterized in that, Step S3 includes: S31: Using current mineral assemblage characteristics and whole-rock geochemical composition data as observational constraints, a multi-component, multi-phase thermodynamic system is constructed through thermodynamic modeling units; S32: Based on the principle of minimizing Gibbs free energy, a forward model is constructed, and combined with a multi-component multiphase thermodynamic system, the theoretical equilibrium mineral assemblage is simulated under different temperature, pressure and fluid composition conditions; S33: Construct the objective function and calculate the difference between the theoretical equilibrium mineral assemblage and the observational constraints; S34: Iteratively adjust the paleotemperature, paleopressure and initial fluid composition parameters using an optimization algorithm until the objective function converges to the minimum value, and invert the paleometallogenic environment parameters at the time of ore deposit formation; S35: Reaction path simulation based on ancient mineralization environment parameters to reconstruct mineral precipitation evolution sequence data.

5. The intelligent identification and visualization analysis method for mineralization mechanisms as described in claim 4, characterized in that, The forward model constructed based on the principle of minimizing Gibbs free energy specifically includes: Under constant temperature and pressure conditions, with the goal of minimizing the total Gibbs free energy of the system, an optimization model is constructed to solve for the equilibrium molar amounts of each species in a multi-component multiphase thermodynamic system. The constraints of the optimization model include: For a closed system environment, the mass conservation constraints, charge conservation and nonnegativity constraints of all components are satisfied; For open system environments, the following constraints must be met: mass conservation of inert components, chemical potential or activity limitation of active components, and charge conservation and nonnegativity. The active components include: , and ; The total Gibbs free energy of the system Represented as: in, This represents the total number of species within the system. For the first The number of moles of each species Standard chemical potential, The gas constant is Thermodynamic temperature For activity; The objective function is constructed as follows: The weighted sum of squares of deviations between the mole fraction of each mineral phase in the theoretical equilibrium mineral assemblage and the mole fraction of the actual mineral phase is calculated; wherein, the weight coefficient assigned to the mineralization indicator metallic minerals is higher than the weight coefficient of gangue minerals, and the objective function also includes a regularization term for constraining the range or smoothness of the inversion parameters. The reaction path simulation specifically includes: using the optimal paleomagnetic environment parameters obtained by inversion as the initial state of the system, and introducing reaction progress variables or temperature and pressure gradient variables; By using the principle of minimizing Gibbs free energy, the equilibrium state of the calculation system at different evolution stages is serialized, generating mineral precipitation evolution sequence data that changes with the evolution process.

6. The intelligent identification and visualization analysis method for mineralization mechanisms as described in claim 1, characterized in that, Step S4 includes: Spatial registration and fusion of the three-dimensional mineralization probability model and mineral sedimentation evolution sequence data; By using visualization units and combining spatially registered and fused data, a linked interactive interface is constructed that includes a three-dimensional geological spatial view and a two-dimensional thermodynamic phase diagram. A drilling decision-making scheme is generated by using a linked interactive interface and normalized comprehensive evaluation indicators. The interactive interface has the following characteristics: Establish a two-way mapping relationship between the three-dimensional voxel index and the two-dimensional thermodynamic state space; When a user selects a specific spatial region in the 3D geological spatial view, the 2D thermodynamic phase diagram view projects and displays the set of thermodynamic state parameter points corresponding to that region in real time. When a user defines a specific mineral stability region or parameter range in a two-dimensional thermodynamic phase diagram using interactive components, the three-dimensional geological space view filters and highlights the space voxels that meet the thermodynamic conditions in real time. The comprehensive evaluation index is a weighted sum of the mineralization probability and the normalized thermodynamic mineralization favorability; The thermodynamic mineralization favorability is used to characterize the physicochemical trend of mineral precipitation and is obtained by calculating the reaction affinity or Gibbs free energy change of the target mineralization reaction under the current thermodynamic conditions.

7. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to perform the intelligent identification and visualization analysis method for mineralization mechanisms as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a computer, perform the intelligent identification and visualization analysis method for mineralization mechanisms as described in any one of claims 1-6.

Images