Intelligent prospecting method and device based on multi-modal data and control theory model

Abstract

The embodiment of the application provides an intelligent ore-prospecting method and device based on multi-modal data and a control theory model, which comprises the following steps: obtaining a multi-modal data body by fusing and aligning multi-source heterogeneous data of a target mining area; extracting key geological slow variables from the data body as system order parameters based on the Haken slave principle and sparse statistical learning, and classifying the key geological slow variables into resistance and driving force control variables for gridding; then, applying the Thom cusp catastrophe theory to calculate a mutation discriminant value in the whole area; finally, intelligently delineating and optimizing an ore-prospecting target area through anomaly area identification, uncertainty analysis and Bayesian updating driven by real drilling data, and a high-confidence exploration target. The application can improve the accuracy and efficiency of deep ore prospecting.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to an intelligent mineral exploration method and apparatus based on multimodal data and cybernetics models. Background Technology

[0002] Traditional mineral exploration methods rely on limited data sources, facing limitations such as insufficient resolution, high subjectivity, and low efficiency, especially in the exploration of deep, concealed mineral deposits. With technological advancements, multimodal data fusion technology has become a research hotspot. By integrating data from multiple sources, including satellite remote sensing, geophysics, geology, and geochemical exploration, it enhances our understanding of underground geological structures. However, in practical applications, this technology still faces challenges such as data heterogeneity, inconsistent quality, and information loss.

[0003] At the same time, AI models have shown great potential in the field of mineral exploration due to their powerful data processing and pattern recognition capabilities. They can not only automatically extract data features and identify mineralization anomalies, but also build predictive models and optimize mineral exploration schemes. However, traditional AI models lack physical constraints and lack predictive accuracy and physical consistency.

[0004] Therefore, there is an urgent need for an intelligent mineral exploration method based on multimodal data and cybernetics models, which can improve the accuracy and efficiency of deep mineral exploration. Summary of the Invention

[0005] To address the problems in the existing technology, this application provides an intelligent mineral exploration method and apparatus based on multimodal data and cybernetics models, which can improve the accuracy and efficiency of deep mineral exploration.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides an intelligent mineral exploration method based on multimodal data and a cybernetics model, comprising: Acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, and perform georegistration and resolution unification on the standardized multimodal heterogeneous data to determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. A dimensionality reduction model is constructed based on the Haken enslavement principle, and a geological gradient constraint term is introduced into the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model. A sparse learning loss function is constructed based on the physical constraint dimensionality reduction model, and the sparse learning loss function is used to extract slow variables from the multimodal data volume to determine multiple corresponding geological slow variables. These multiple geological slow variables are classified into resistance variables and driving force variables. The weights of each resistance variable and each driving force variable are calculated using the ridge regression algorithm, and the variables after weight calculation are normalized to determine the corresponding resistance control variables and driving force control variables. The target ore-forming region is gridded and assigned values ​​based on the resistance control variables and the driving force control variables to determine the corresponding gridded coefficient field. The value of each grid point in the gridded coefficient field is calculated based on the Tom cusp catastrophe model to determine the cusp catastrophe discriminant value of the target ore-forming region. The final prospecting target is determined based on the cusp catastrophe discriminant value.

[0007] Furthermore, the construction of a dimensionality reduction model based on the Haken enslavement principle, and the introduction of a geological gradient constraint term into the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model, includes: The key intrinsic parameters controlling the evolution of the ore-forming system are set as slow variables, and a large number of original observation parameters in the multimodal data volume are regarded as fast variables. A dimensionality reduction model is constructed, which is the standard Haken enslavement equation describing the evolution rate of fast variables. An explicit geological constraint term is added to the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model. The geological gradient constraint term is a digital gradient field determined based on prior geological knowledge, which characterizes the spatial continuity prior of stratigraphic dip angle, lithological boundary or tectonic boundary, and is used to ensure that key geological constraints are not lost during the dimensionality reduction process.

[0008] Furthermore, the construction of the sparse learning loss function guided by the physical constraint dimensionality reduction model includes: The data fitting term is constructed based on the physical constraint dimensionality reduction model described above, which is used to measure the error of reconstructing the original observation data by slow variables; Based on the physical constraint dimensionality reduction model, sparse constraint terms are constructed to screen slow variables that have a key impact on mineralization and suppress redundant features. Guided by the physical constraint dimensionality reduction model, a prior knowledge constraint term is constructed to introduce the regional mineralization experience represented by the geological gradient constraint term in the form of weighted priors, so as to guide the model learning to converge toward a more geologically reasonable direction. Guided by the physical constraint dimensionality reduction model, a spatial smoothing constraint term is constructed to constrain the smoothness of the changes of the extracted slow variables in adjacent spatial locations, so as to suppress data noise and enhance the regional continuity of the results. A loss function is constructed based on the data fitting term, the sparse constraint term, the prior knowledge constraint term, and the spatial smoothness constraint term. Then, the corresponding weight coefficients are set for each constraint term in the loss function using cross-validation to determine the corresponding sparse learning loss function.

[0009] Furthermore, the step of extracting slow variables from the multimodal data volume using the sparse learning loss function to determine multiple corresponding geological slow variables includes: The multimodal data volume is fused according to a preset data fusion algorithm to construct a surface observable proxy index. The surface observable proxy index is a surface observable representation of deep slow variables in geological data. The data fusion algorithm includes: adjusting the basic weights of the multimodal data to be fused by penalizing or rewarding them based on the cloud cover degree and InSAR data point density of the multimodal data to be fused. Mutual information analysis and Granger causality test are performed on the observable proxy indicators of the land surface to determine the indicators whose mutual information is greater than the first threshold and whose Granger causality test value is less than the second threshold, which are then used as the candidate feature set for slow variable calculation. The candidate feature set is solved by the sparse learning loss function to determine the corresponding multiple slow geological variables.

[0010] Further, the process of classifying the multiple geological slow variables into resistance variables and driving force variables, calculating the weights of each resistance variable and each driving force variable using the ridge regression algorithm, and normalizing the variables after weight calculation to determine the corresponding resistance control variables and driving force control variables includes: Based on the physical role of slow variables in geological processes, the aforementioned geological slow variables are classified into resistance variables and driving force variables, respectively. Based on prior geological knowledge of the target mining area, prior weights are assigned to each of the aforementioned resistance variables and driving force variables. Ridge regression is used to optimize the prior weights with the goal of minimizing prediction error, thereby determining the corresponding weights of each of the aforementioned resistance variables and driving force variables, and constraining the sum of the weight coefficients to be 1. The classified drag and driving force variables are standardized using Z-score to determine the corresponding normalized drag and driving force components. The normalized resistance components are weighted and summed according to the weights of the resistance variables to determine the corresponding resistance control variables. The normalized driving force components are weighted and summed according to the weights of the driving force variables to determine the corresponding driving force control variables.

[0011] Further, determining the final mineral exploration target based on the cusp mutation discriminant value includes: When the cusp mutation discrimination value is greater than 0, the target mineralization area is determined to be in a stable state, indicating that there is no industrial ore body; When the cusp mutation discrimination value is equal to 0, the target mineralization area is determined to be in a critical state, indicating that there is small-scale mineralization. When the cusp mutation discrimination value is less than 0, it is determined that a cusp mutation has occurred in the target mineralization area, indicating the presence of an industrial ore body. When the cusp mutation discrimination value is within a preset critical ambiguity range, a secondary discrimination process is initiated to determine the final mineral exploration target. The secondary discrimination process includes: Initiate the time dimension verification process, use multi-temporal synthetic aperture radar interferometry data to calculate the rate of change of the cusp mutation discrimination value over time. If the absolute value of the rate of change is greater than the preset annual rate of change threshold, the region is determined to be a dynamic evolution zone, which may be a false anomaly. Initiate the spatial correlation test process, calculate the spatial clustering index of the negative anomaly area with the known mineral deposits and the cusp mutation discrimination value. If the clustering index exceeds the preset confidence threshold within the preset spatial distance, the mineral exploration confidence of the anomaly area is improved.

[0012] Furthermore, the step of georegistering and unifying the resolution of the multimodal heterogeneous data to form a three-dimensional aligned data volume includes: The multimodal heterogeneous data is converted to a preset geodetic coordinate system and subjected to sub-pixel level geometric fine correction to determine the corresponding corrected multimodal data so that the planar positioning error of the data layer is less than 1 pixel. A spatial resolution benchmark is set, and the corrected multimodal data with a resolution higher than the spatial resolution benchmark are unified to the benchmark according to the downsampling algorithm that protects high-frequency information. The corrected multimodal data with a resolution lower than the spatial resolution benchmark are unified to the benchmark according to the upinterpolation algorithm based on terrain constraints. The vertical stratigraphic levels of each modal data, after being standardized in resolution, are matched and aligned with standard geological stratigraphic levels to determine the corresponding three-dimensional aligned multimodal data volume.

[0013] Secondly, this application provides an intelligent mineral exploration device based on multimodal data and a cybernetics model, comprising: A multi-source data processing module is used to acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, perform georegistration and resolution unification on the standardized multimodal heterogeneous data, and determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. The slow variable extraction module is used to construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine multiple corresponding geological slow variables. The multiple geological slow variables are classified into resistance variables and driving force variables. The weights of each resistance variable and each driving force variable are calculated according to the ridge regression algorithm, and the variables after weight calculation are normalized to determine the corresponding resistance control variables and driving force control variables. The ore body determination module is used to assign gridded values ​​to the target ore-forming region based on the resistance control variables and the driving force control variables, determine the corresponding gridded coefficient field, calculate the value of each grid point of the gridded coefficient field according to the Tom cusp catastrophe model, determine the cusp catastrophe discrimination value of the corresponding target ore-forming region, and determine the final prospecting target point based on the cusp catastrophe discrimination value.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent mineral exploration method based on multimodal data and a cybernetics model.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the intelligent mineral exploration method based on multimodal data and a cybernetics model.

[0016] Fifthly, this application provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the steps of the intelligent mineral exploration method based on multimodal data and a cybernetics model.

[0017] As can be seen from the above technical solution, this application provides an intelligent mineral exploration method and device based on multimodal data and cybernetics models. By fusing and aligning multi-source heterogeneous data of the target mining area, a multimodal data volume is obtained. Based on the Haken enslavement principle and sparse statistical learning, key geological slow variables are extracted from the data volume as system order parameters and classified as resistance and driving force control variables for gridding. Then, the Tom cusp catastrophe theory is applied to calculate the catastrophe discriminant value in the gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, the mineral exploration target area is intelligently delineated and optimized, and high-confidence exploration targets are obtained, thereby improving the accuracy and efficiency of deep mineral exploration. Attached Figure Description

[0018] 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.

[0019] Figure 1 This is one of the flowcharts illustrating the intelligent mineral exploration method based on multimodal data and cybernetics models in the embodiments of this application; Figure 2 This is a structural diagram of the intelligent mineral exploration device based on multimodal data and cybernetics model in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0020] Figure label: Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0023] Considering the numerous challenges of traditional mineral exploration methods and the lack of real physical constraints in emerging AI-based model data processing approaches, this application provides an intelligent mineral exploration method and apparatus based on multimodal data and cybernetics models. By fusing and aligning multi-source heterogeneous data from a target mining area, a multimodal data volume is obtained. Based on the Haken enslavement principle and sparse statistical learning, key geological slow variables are extracted from the data volume as system order parameters and categorized as resistance and driving force control variables for gridding. Then, Thomson's cusp catastrophe theory is applied to calculate catastrophe discriminant values ​​across the entire gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, the mineral exploration target area is intelligently delineated and optimized, resulting in high-confidence exploration targets. This improves the accuracy and efficiency of deep mineral exploration.

[0024] To improve the accuracy and efficiency of deep mineral exploration, this application provides an embodiment of an intelligent mineral exploration method based on multimodal data and a cybernetics model. See [link to embodiment]. Figure 1 The intelligent mineral exploration method based on multimodal data and cybernetics models specifically includes the following: Step S101: Acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, and perform georegistration and resolution unification on the standardized multimodal heterogeneous data to determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. Optionally, in this embodiment, during the data acquisition and aggregation stage, a complete set of information covering the target mineralized area is systematically collected from multiple sensors and databases. Specifically: Satellite remote sensing data includes multispectral (such as Landsat, Sentinel-2), hyperspectral (such as ASTER), thermal infrared and radar (such as Sentinel-1 InSAR) imagery, as well as high-precision digital elevation models (DEMs).

[0025] Geophysical data encompasses field values ​​or inverted physical property parameters obtained from gravity, magnetic, electromagnetic, and seismic exploration.

[0026] Geochemical data are primarily derived from the determination of elemental content in surface soils, rocks, or river sediments.

[0027] Geological data includes maps of strata, structures, and lithology, as well as point and line observation data such as boreholes and profiles.

[0028] Because these data differ significantly in format, coordinate system, resolution, dimensions, and semantics, they are in a heterogeneous state. Therefore, standardization preprocessing is necessary.

[0029] Specifically, the data format is uniformly converted to a common geographic information system (GIS) or a scientific computing compatible format (such as GeoTIFF, NetCDF); the numerical data of geophysical and geochemical data are denoised and outliers are removed, and normalized according to the background values ​​of the whole region (such as Z-score normalization or range normalization) to eliminate the influence of dimensions and transform them into dimensionless and comparable indicators; geological category data are coded and converted from vector to raster, and the lithology, structure and other attribute information in geological maps are converted into raster layers with classification codes.

[0030] Next, georeferencing is performed to convert all data layers to the same precise geodetic coordinate system and projection system. Then, resolution unification is performed by setting a benchmark spatial resolution suitable for regional mineralization prediction. Data with resolutions higher and lower than this resolution are unified to ensure that all data are expressed and processed on the same spatial scale, avoiding information distortion or analytical bias caused by scale mismatch.

[0031] The data layers processed through the above steps are spatially aligned in three dimensions, forming a three-dimensional aligned multimodal data volume. Each three-dimensional grid cell (volume element) of the data volume contains standardized and registered attribute values ​​from multiple sources, including remote sensing, geophysics, geochemistry, and geology. This provides the input basis for subsequent slow variable extraction and gridding.

[0032] Step S102: Construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine the corresponding multiple geological slow variables. Classify the multiple geological slow variables into resistance variables and driving force variables, calculate the weights of each resistance variable and each driving force variable according to the ridge regression algorithm, and normalize each variable after weight calculation to determine the corresponding resistance control variables and driving force control variables. Optionally, in this embodiment, this step converts the original geological observation data into core control variables for mineralization and ultimately extracts seven key sequence parameters.

[0033] Specifically, this step is guided by the Haken principle of enslavement.

[0034] First, at the underlying logic level, the mineralization / reservoir formation system is essentially a dissipative structure far from equilibrium (Prigogine dissipative model), with continuous input of deep energy, matter and fluid forming nonlinear positive feedback (hypercycle), and its self-organizing criticality produces cusp mutations.

[0035] Based on the above-mentioned mineralization laws, we know that in complex systems, a few "slow variables" (characteristic parameters) determine the evolutionary behavior of a large number of "fast variables." Fast variables are "enslaved" by slow variables. Therefore, Haken's enslavement principle is used to explain how "slow variables" dominate the dynamic behavior of "fast variables" in complex systems, thereby realizing the formation of system self-organization and macroscopic ordered structure.

[0036] This leads to the Haken enslavement dimensionality reduction model: Let the system state variables be: slow variables (order parameters): Fast variables:

[0037] The system can then be written as:

[0038]

[0039] in, It is a fast variable The time derivative of represents the rate of evolution of the fast variable; The decay coefficient represents the decay rate of a fast variable. 1. Ensure that fast variables are enslaved by slow variables. Fast variables represent the original observation parameters (such as remote sensing data points) in a multimodal data volume. Coupled function, representing slow variable The impact on fast variables; Slow variables (order parameters) represent core system parameters, such as geological slow variables; because 1 (Fast variables decay quickly), we can let:

[0040] Thus, the fast variable is "enslaved" by the slow variable, and the system behavior is entirely determined by... control.

[0041] Based on the aforementioned Haken enslavement dimensionality reduction model, in the context of geological prospecting, "fast variables" correspond to massive, high-frequency surface observation data points that may be affected by local interference or noise (such as the reflectance of a single pixel or the geochemical measurement value at a single point), while "slow variables" are the core geophysical parameters that control whether the entire metallogenic system undergoes a qualitative change from "mineralized" to "mineralized" and have a larger spatiotemporal scale and change more slowly (such as the regional stress field state or the fluid overpressure level).

[0042] Directly applying the standard Haken enslavement dimensionality reduction equation carries risks, namely, that purely mathematical dimensionality reduction may overlook crucial geological boundary conditions, leading to physically inaccurate extracted "slow variables." Therefore, we introduce a geological gradient constraint term, which is equivalent to embedding geological knowledge into the mathematical equation. The formal expression of the physically constrained dimensionality reduction model is as follows:

[0043] in, : Geological gradient constraint term, introducing geological constraints; Constraint coefficients, determined based on geological knowledge; Geological gradient fields, such as stratigraphic dip gradients, ensure that key geological constraints are not lost during dimensionality reduction. For example, on both sides of a major fault, even if remote sensing image features (fast variables) are similar, their geological meanings may be completely different. Constraint terms can force the model to recognize and respect this discontinuity.

[0044] Guided by this physical constraint dimensionality reduction model, a sparse learning loss function is constructed for practical calculation:

[0045] in: (Data fitting term): For the sample size, For the observed data vector, For the characteristic matrix, As a weight vector, it ensures that the data matrix is ​​observed from a high-dimensional perspective. The slow variables reconstructed from the (fast variables) are consistent with the overall trend of the original data.

[0046] (L1 regularization of sparse constraint terms): This is a sparsity coefficient (typically 0.1-0.3), which promotes weight sparsity and forces the weight vector to be sparse. A large number of elements tend to zero, thus achieving feature selection.

[0047] (L2 regularization of prior knowledge constraints): The prior weights are (0.5-0.8). As the prior weights for the regional metallogenic model, the existing regional metallogenic model knowledge ( This is introduced as a soft constraint. For example, if it is known that copper mineralization in a certain area is mainly controlled by a specific stratum, then the prior values ​​of the feature weights associated with that stratum will be higher. This balances data-driven and knowledge-driven approaches, prevents the model from deviating completely from known geological patterns, enhances the geological rationality of the results, and provides stable guidance, especially in areas with poor data quality.

[0048] (Spatial total variational smoothing term): This is a smoothing coefficient (adjusted to 0.05-0.15 depending on the scale). As a spatial difference operator, ensuring spatial continuity, the total variation penalty encourages the generation of slow variable fields to be spatially piecewise smooth, while allowing abrupt changes at boundaries (such as faults and lithological contact zones). This effectively suppresses "salt and pepper" spurious anomalies caused by noise in remote sensing images or local anomalies, making the extracted slow variables (such as the geostress gradient field) present as continuous "zones" or clear boundaries consistent with geological understanding, rather than chaotic points.

[0049] Hyperparameter selection strategy: (Sparseness): Determined through cross-validation, typically set to 0.1-0.3; (Prior constraints): Derived from regional metallogenic models, with weights ranging from 0.5 to 0.8; (Spatial smoothing): Adjusted according to the exploration scale, 0.05 for 1:50,000 mapping and 0.15 for 1:10,000 detailed exploration.

[0050] Using the sparse learning loss function we constructed, we were able to stably extract seven geological slow variables from the high-dimensional multimodal data volume, which can then be used to guide mineral exploration.

[0051]

[0052] After extracting the slow variables, they are classified according to their geological and physical implications into resistance variables that tend to hinder mineralization (such as pressure drop and temperature drop when the caprock breaks through) and driving force variables that tend to promote mineralization (such as fault activity index and elemental geochemical potential gradient). The classification is based on a theoretical understanding of the mineralization process.

[0053] Next, in order to synthesize the final "resistance control variable a" and "driving force control variable b", it is necessary to determine the relative importance (weight) of various internal slow variables.

[0054]

[0055] Specifically:

[0056] Among them: rock toughness corresponds to the geostress anomaly gradient;

[0057] Specifically:

[0058] Since the slow variable "intensity of oxygen fugacity abrupt change zone" and the slow variable "geochemical potential of ore-forming elements" are essentially identical in nature, these two variables are merged into one: .

[0059] Among them, weight Determined by ridge regression, it must meet the following conditions. and ; Normalized caprock pressure: Z-score normalized caprock breakthrough pressure; Normalized temperature drop: The temperature drop value after Z-score normalization; Normalized rock toughness: The rock toughness index after Z-score standardization; Normalized fluid overpressure: Z-score normalized fluid overpressure value; Normalized fracture activity: Z-score normalized fracture activity intensity; Normalized elemental gradient: Z-score normalized elemental concentration gradient.

[0060] Normalization employs Z-score standardization. The Z-score (also called the standard score or Z-score) is a statistical standardization method used to measure the degree to which a data point deviates from the mean of its dataset, using standard deviation as the unit. Simply put, it brings data of different dimensions and magnitudes "to the same scale" for comparison. The reference baseline is taken as the background value of the entire region. The normalized "resistance" and "driving force" become dimensionless, comparable indicators of relative intensity.

[0061] In mineral exploration (especially geochemical exploration), the Z-score is mainly used to highlight geochemical anomalies, because mineral exploration signals are often "anomalies that deviate from the regional background".

[0062] Step S103: Assign gridded values ​​to the target ore-forming region based on the resistance control variable and the driving force control variable, determine the corresponding gridded coefficient field, calculate the value of each grid point in the gridded coefficient field based on the Tom cusp catastrophe model, determine the cusp catastrophe discriminant value of the corresponding target ore-forming region, and determine the final prospecting target based on the cusp catastrophe discriminant value.

[0063] Optionally, in this embodiment, the resistance control variables and driving force control variables calculated in step S102 are mapped to real geographic space to form a computable digital field.

[0064] First, the entire target exploration area is divided into regular grid cells (e.g., 10m x 10m). Since the original geological and geophysical data observation points are non-uniform, spatial interpolation algorithms such as Kriging interpolation and inverse distance weighting are used to continuously and smoothly interpolate and extrapolate the discrete observation or computational point data for each variable (a and b) to the center point of each grid cell. This process ensures the continuous distribution of model parameters in three-dimensional space.

[0065] After the assignment is completed, two two-dimensional or three-dimensional raster data layers are formed, which are strictly aligned with the geographic coordinates of the exploration area. These are called "resistance coefficient field A" and "driving force coefficient field B," respectively. These two coefficient fields are the basic spatial data for all subsequent quantitative calculations.

[0066] For each grid point (i, j) in the gridded coefficient field, its drag value is read. and driving force value Substitute the values ​​into the discriminant formula of the Tom cusp mutation model for calculation.

[0067] Specifically, the derivation process of the Tom cusp mutation model: The standard potential function of cusp mutation (Thom, 1972) is given as follows:

[0068] in: Potential function, describing the energy landscape of the system; : State variables, corresponding to mineralization intensity in geology (such as mineralization degree or ore saturation). Two control variables (corresponding to "resistance" and "driving force" in geology, respectively). steady state conditions Given the equilibrium surface equations, describe the steady state of the system:

[0069] This is a cubic equation, and the number of roots is actually determined by the discriminant.

[0070] Next, calculate the second derivative and the singularity set condition, which are used to identify abrupt change points:

[0071] Eliminating X yields the Δ discriminant, which, when rearranged, gives the cusp mutation discriminant:

[0072] Therefore, substituting drag and driving force into the cusp mutation discriminant, we obtain:

[0073] The computation is performed in parallel across all grids in the entire region, generating a new raster data layer with a spatial resolution completely consistent with the input coefficient field—the "cusp mutation discriminant field" (Cutoff Discriminant Field). field)".

[0074] according to The sign and magnitude of the values ​​are used to perform a preliminary, theory-based classification of the geological state for each grid point: >0 Single solid root system is stable → dispersed phase, no industrial ore body / oil and gas reservoir; = 0 Critical → Small-scale / marginal oil reservoir; <0 Three real root equilibrium surface folding → First-order phase transition will inevitably occur, and industrial ore bodies / large oil and gas fields will suddenly appear "from nothing to something"; when When the system is in the range [-500, 500], it is in the critical ambiguity region and requires the activation of secondary discrimination. Time dimension verification: Calculation using multi-temporal InSAR (Discriminant value, rate of change over time, used for time-dimensional testing) If If the value is >200 / year, it is considered a dynamic evolutionary region, which may be a false anomaly. express The rate of change of the value over time, i.e., the time derivative.

[0075] Spatial correlation test: calculation Ripley's K is the spatial clustering index between negative anomaly regions and known mineral deposits. If K(r) > 1.5 (r = 5km), then the reliability is improved.

[0076] Using the above method, we obtain the cusp mutation value by solving the cusp mutation equation. Subsequently, using the cusp mutation value we obtained, we calibrate by drilling 3-5 real holes and updating dozens or hundreds of virtual holes using Bayesian methods to finally delineate the sweet spot and obtain mineralization guidance.

[0077] This example demonstrates how this embodiment uses dissipative structure theory and cusp catastrophe models as its physical core, treating the mineralization process as a catastrophe transition in an open system under an imbalance of driving forces and resistance. By constructing proxy variables for "driving forces" and "resistance" using multi-source data such as remote sensing and geophysics, the catastrophe discriminant Δ is calculated, and intelligent target area delineation is achieved using the Physics-InformedDeepONet deep learning architecture. This provides a new paradigm for the exploration of deep, concealed mineral deposits.

[0078] As described above, the intelligent mineral exploration method based on multimodal data and cybernetics models provided in this application can obtain a multimodal data volume by fusing and aligning multi-source heterogeneous data of the target mining area. Based on the Haken enslavement principle and sparse statistical learning, key geological slow variables are extracted from the data volume as system order parameters and classified as resistance and driving force control variables for gridding. Then, the Tom cusp catastrophe theory is applied to calculate the catastrophe discriminant value in the gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, the mineral exploration target area is intelligently delineated and optimized, and high-confidence exploration targets are identified, thereby improving the accuracy and efficiency of deep mineral exploration.

[0079] In one embodiment of the intelligent mineral exploration method based on multimodal data and cybernetics models of this application, it may further include the following: Step S201: Set the key intrinsic parameters controlling the evolution of the ore-forming system as slow variables, and regard the large number of original observation parameters in the multimodal data volume as fast variables to construct a dimensionality reduction model. The dimensionality reduction model is the standard Haken enslavement equation describing the evolution rate of fast variables. Step S202: Add explicit geological constraint terms to the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model. The geological gradient constraint terms are digital gradient fields determined based on prior geological knowledge, which characterize the spatial continuity prior of stratigraphic dip angle, lithological boundary or tectonic boundary, and are used to ensure that key geological constraints are not lost during the dimensionality reduction process.

[0080] Optionally, in this embodiment, during the slow variable extraction and dimensionality reduction steps, the key intrinsic parameters controlling the evolution of the ore-forming system are first designated as "slow variables," representing the core order parameters driving the macroscopic evolution of the system. Simultaneously, a large number of original observation parameters in the multimodal data volume are considered as "fast variables," which change rapidly, are numerous, and are controlled by the slow variables. Based on this framework, a dimensionality reduction model based on the standard Haken enslavement equation is constructed.

[0081]

[0082] This equation describes how the evolution rate of a fast variable is affected by a slow variable. Its core is to set the time derivative of the fast variable as a combination of a decay term and a function of the slow variable.

[0083] Subsequently, to overcome the shortcomings of purely mathematical models that may ignore actual geological constraints, the dimensionality reduction model was enhanced. Specifically, a geological gradient constraint term was explicitly added to the standard Haken enslavement equation.

[0084] The geological gradient constraint term is a digital gradient field constructed based on prior geological knowledge (such as geological maps and tectonic interpretation results), which quantitatively characterizes the spatial distribution characteristics of stratigraphic dip changes, lithological boundaries, or tectonic interfaces. During the model solution process, this constraint term participates in the calculation as a mandatory spatial continuity condition, guiding the dimensionality reduction process to consider both mathematical optimality and the degree of fit with the actual geological structure.

[0085] Through step S202, this embodiment ensures that the final slow variable extraction result is not only an effective low-dimensional representation of high-dimensional data mathematically, but also strictly controlled by the real geological spatial pattern in a physical sense. This improves the geological interpretability and reliability of the extracted order parameters and lays a data foundation for guiding the construction of sparse learning loss functions.

[0086] In one embodiment of the intelligent mineral exploration method based on multimodal data and cybernetics models of this application, it may further include the following: Step S301: Construct a data fitting term guided by the physical constraint dimensionality reduction model to measure the error of reconstructing the original observation data by slow variables; Step S302: Construct sparse constraint terms based on the physical constraint dimensionality reduction model to screen slow variables that have a key impact on mineralization and suppress redundant features; Step S303: Construct prior knowledge constraint terms based on the physical constraint dimensionality reduction model, which are used to introduce the regional mineralization experience represented by the geological gradient constraint terms in the form of weighted priors, and guide the model to converge in a more geologically reasonable direction. Step S304: Construct a spatial smoothing constraint term based on the physical constraint dimensionality reduction model to constrain the smoothness of the changes of the extracted slow variables in adjacent spatial locations, so as to suppress data noise and enhance the regional continuity of the results; Step S305: Construct a loss function based on the data fitting term, the sparse constraint term, the prior knowledge constraint term, and the spatial smoothness constraint term, and determine the corresponding sparse learning loss function by setting corresponding weight coefficients for each constraint term in the loss function through cross-validation.

[0087] Optionally, in this embodiment, a data fitting term is constructed.

[0088] The data fitting term aims to ensure that the low-dimensional slow variables extracted from the high-dimensional original observation data can effectively reconstruct the original signal, which is the foundation of model effectiveness. Specifically, the sum of squared residuals is used as the fitting term to calculate the difference between the original observation vector and the predicted vector reconstructed by the slow variables through a physically constrained model. This ensures that the dimensionality reduction process does not lose key information, and that the extracted slow variables retain the variations related to the metallogenic system in the original data to the greatest extent possible. If this condition is not met, the model will become a purely mathematical dimensionality reduction, losing its ability to interpret and reconstruct actual geological observations, resulting in a lack of reliable data foundation for subsequent abrupt change identification.

[0089] Optionally, in this embodiment, sparse constraint terms are constructed.

[0090] The sparse constraint project aims to achieve feature selection while reducing dimensionality. Specifically, it uses the L1 norm (such as Lasso) to penalize the weight coefficients of slow variables, causing the coefficients of many insignificant or redundant features to approach zero, retaining only a few slow variables that play a crucial role in the mineralization process. The effect is to improve the model's interpretability and generalization ability: on the one hand, it forces the model to focus on the most important driving factors in a geophysical sense (such as geostress and fluid pressure), conforming to the synergetics principle of "a few order parameters dominating the system"; on the other hand, by suppressing irrelevant noise features, it reduces the risk of overfitting, making the model's predictions more stable in new blocks.

[0091] Optionally, in this embodiment, prior knowledge constraints are constructed.

[0092] Prior knowledge constraints are a key step in quantifying domain experts' geological knowledge and integrating it into the machine learning process. Specifically, in the form of an L2 norm penalty term, the weight coefficients are mapped to a prior weight vector ( The prior weights, derived from regional metallogenic models, statistics of known ore deposits, or expert experience, represent the relative importance of each slow variable within a specific geological context. Their role is to guide the optimization process towards a direction that is geologically more reasonable and conforms to known laws, preventing data-driven models from producing absurd solutions that contradict basic geological principles. For example, in sedimentary rock areas, higher prior weights can be assigned to fluid overpressure; in tectonically active areas, prior constraints on fault activity can be strengthened.

[0093] Optionally, in this embodiment, a spatial smoothing constraint term is constructed.

[0094] Smoothing constraints are used to address the prevalent noise issues in spatial data such as remote sensing and geophysics, while respecting the spatial continuity of geological field variables. Specifically, a total variation or Laplace smoothing term is employed to penalize the adjacent differences of slow variables on the spatial grid. This forces the extracted slow variables (such as geostress gradients and temperature anomalies) to exhibit gradual spatial variation, filtering out spurious high-frequency abrupt changes caused by local noise, thus yielding a more continuous and stable regional anomaly field. This not only improves the readability of the resulting maps but also makes the subsequently delineated anomaly areas more geologically consistent, avoiding the creation of numerous isolated and fragmented false target areas due to data noise.

[0095] The four factors (data fitting, sparsity, prior knowledge, and spatial smoothness) are linearly combined to construct the final sparse learning loss function. Each constraint term is preceded by a weight coefficient. These hyperparameters are used to control the relative importance of the term in the overall optimization objective. They cannot be arbitrarily set and require system optimization through cross-validation. Specifically, the dataset is divided into training and validation sets. The model is trained on the training set with different hyperparameter combinations, and its performance (such as reconstruction error and predicted mineralization rate) is evaluated on the validation set. Finally, the hyperparameter combination that performs best overall on the validation set is selected. The purpose is to scientifically balance the multi-dimensional conflict between "data fitting accuracy," "model simplicity," "prior confidence," and "result smoothness," resulting in a mathematically optimal, geologically reasonable, and robust dimensionality reduction model with strong generalization ability in unknown areas. This provides reliable slow variable input for subsequent mutation detection and target area delineation.

[0096] Through step S305, this embodiment successfully constructed the corresponding sparse learning loss function under the guidance of the Haken enslavement dimensionality reduction model, laying the foundation for subsequent extraction of slow variables through the loss function.

[0097] In one embodiment of the intelligent mineral exploration method based on multimodal data and cybernetics models of this application, it may further include the following: Step S401: Perform data fusion on the multimodal data volume according to the preset data fusion algorithm to construct a surface observable proxy index. The surface observable proxy index is a surface observable representation of deep slow variables in geological data. The data fusion algorithm includes: adjusting the basic weights of the multimodal data to be fused by penalty or reward according to the cloud coverage and InSAR data point density of the multimodal data to be fused. Step S402: Perform mutual information analysis and Granger causality test on the observable proxy indicators of the land surface, and determine the indicators whose mutual information is greater than the first threshold and whose Granger causality test value is less than the second threshold, as the candidate feature set for participating in the slow variable calculation; Step S403: Solve the candidate feature set using the sparse learning loss function to determine the corresponding multiple geological slow variables.

[0098] Optionally, in this embodiment, this step is a process of data preprocessing of the multimodal data volume before applying the sparse learning loss function to extract slow variables.

[0099] First, raw multimodal data of different sources, formats, and resolutions are fused to generate a series of observable proxy indicators for the Earth's surface.

[0100] Observable proxy indicators are used to quantitatively characterize deep geological processes (i.e., "slow variables") that cannot be directly observed, using data that can be obtained at or near the surface (such as remote sensing spectra, geophysical anomalies, and geochemical element abundances).

[0101] In the process of multimodal data fusion, the key lies in introducing a dynamic weight adjustment mechanism:

[0102] in, Adjusted fusion weights; Basic weights; Cloud coverage penalty coefficient (0.02) reduces the weight of high-data-weighting-in-cloud-coverage. CloudCover: Cloud coverage level (0-1); InSAR density bonus coefficient (1.5) increases the weight of high-density InSAR data; : InSAR data point density.

[0103] The aforementioned dynamic weight adjustment mechanism adjusts the basic fusion weights in real time based on data quality and information density. Specifically, for remote sensing data with severe cloud cover, its weight is penalized by lowering it to reduce the interference of low-quality data on the results; conversely, for InSAR data with dense observation points and rich information reflecting surface deformation, its weight is rewarded by increasing it to strengthen the contribution of high-value data. Through this automated weight adjustment, information is "refined" at the data level, providing a standardized indicator dataset with lower noise and higher reliability for subsequent analysis.

[0104] Next, from the large number of proxy indicators generated in step S401, core indicators that have a strong correlation and causal driving relationship with the deep mineralization system are selected to form a candidate feature set.

[0105] Methodologically, two statistical tools, mutual information analysis and Granger causality test, were used for dual filtering.

[0106] Mutual information analysis is used to quantify the nonlinear correlation strength between each proxy indicator and the state of the mineralization system (or known mineralization), retaining indicators with strong correlation (mutual information > first threshold).

[0107] The Granger causality test further judges from the time or space sequence whether a change in one indicator "leads" to a change in another indicator or result reflecting mineral potential in a statistically significant way, retaining indicators with significant causal indication (p-value < second threshold).

[0108] This step, based on data-driven and causal logic, eliminates a large amount of irrelevant or weakly correlated information, ensuring that subsequent modeling focuses only on the key driving factors that have a substantial impact on the mineralization process.

[0109] Finally, the sparse learning loss function is applied to calculate the feature candidate set. The final output geological slow variable is a low-dimensional quantitative representation of the dynamics of complex metallogenic systems with clear physical and geological significance. This lays a solid and computable core parameter foundation for subsequent mutation discrimination and target area prediction.

[0110] Through step S403, this embodiment successfully extracts surrogate indicators from the original multimodal data, and filters the surrogate indicators based on causal driving to form a feature candidate set. By calculating the loss of the feature candidate set, the geological slow variable is obtained.

[0111] In one embodiment of the intelligent mineral exploration method based on multimodal data and cybernetics models of this application, it may further include the following: Step S501: Based on the physical role of slow variables in geological processes, classify the multiple geological slow variables into resistance variables and driving force variables respectively; Step S502: Assign prior weights to each of the resistance variables and each of the driving force variables based on the prior geological knowledge of the target mining area. Use the ridge regression method to optimize the prior weights with the goal of minimizing the prediction error, determine the corresponding weights of each of the resistance variables and each of the driving force variables, and constrain the sum of the weight coefficients to be 1. Step S503: Perform Z-score standardization on the classified drag and driving force variables to determine the corresponding normalized drag and driving force components. Step S504: Perform a weighted summation of the normalized resistance components according to the weights of the resistance variables to determine the corresponding resistance control variables; perform a weighted summation of the normalized driving force components according to the weights of the driving force variables to determine the corresponding driving force control variables.

[0112] Optionally, in this embodiment, based on the physical role of slow variables in geological processes, the multiple geological slow variables are classified into resistance variables and driving force variables respectively.

[0113] In the mineralization process, resistance variables represent those geological factors that inhibit or hinder the migration and accumulation of ore-forming materials, such as the pressure drop when the caprock breaks through (reflecting the capping capacity), the reduction of the temperature anomaly gradient (reflecting energy dissipation), and rock toughness (reflecting the ease of fracturing). Driving force variables represent factors that promote the mineralization process, such as the fluid overpressure index (reflecting the fluid transport force), the ore-controlling fault activity index (reflecting the degree of openness of tectonic channels), and the geochemical potential gradient of ore-forming elements (reflecting the direction of material migration).

[0114] This classification simplifies complex mineralization systems into the interaction of two contradictory forces: "driving" and "inhibiting," thus characterizing the mechanical basis of mineralization mutations from a physical perspective.

[0115] Optionally, in this embodiment, prior weights are assigned to each resistance variable and driving force variable based on prior geological knowledge of the target mining area.

[0116] Prior knowledge comes from regional metallogenic models, existing deposit statistical characteristics, or expert experience. Ridge regression is used to optimize the prior weights with the objective of minimizing prediction error, resulting in the weights of each variable. Ridge regression effectively mitigates multivariate collinearity by introducing an L2 norm penalty term for the weight coefficients in the loss function, ensuring stable weight estimates even when data correlation exists. The optimization process aims to minimize the prediction error of the final constructed "driving force" and "resistance" composite index for known mineralization points or mineralization intensity. Simultaneously, the constraint that the sum of all weight coefficients is 1 ensures that the relative importance of different variables can be compared and interpreted on a standardized scale, enhancing the model's interpretability and physical consistency.

[0117] Optionally, in this embodiment, the classified resistance variables and driving force variables are subjected to Z-score standardization to determine the corresponding normalized resistance components and driving force components. The normalized resistance components include normalized caprock pressure, normalized temperature drop, and normalized rock toughness; the normalized driving force components include normalized fluid overpressure, normalized fracture activity, and normalized elemental gradient.

[0118] Z-score standardization transforms original variables of different dimensions and orders of magnitude into a standard normal distribution with a mean of 0 and a standard deviation of 1 by subtracting the regional background mean and then dividing by the standard deviation. This process eliminates differences in numerical range and units between different data sources (such as remote sensing, geophysics, and geochemistry), enabling them to be weighted and fused at the same scale. More importantly, standardization based on the regional background value ensures that the calculated "driving" and "resistance" values ​​reflect the degree of anomaly relative to the regional background, directly pointing to local anomaly areas where mineralization abrupt changes may occur, thus highlighting the exploration target.

[0119] Optionally, in this embodiment, the normalized resistance components are weighted and summed according to the determined resistance variable weights to calculate the comprehensive resistance control variable; the normalized driving force components are weighted and summed according to the determined driving force variable weights to calculate the comprehensive driving force control variable.

[0120] These two control variables represent the quantitative synthesis of two forces: "inhibiting mineralization" and "promoting mineralization." Through weighted summation, the model integrates multiple observational information into two core control parameters, achieving efficient dimensionality reduction of the data. These two control variables directly correspond to the two control dimensions (a, b) in cusp catastrophe theory, paving the way for the subsequent direct application of the cusp catastrophe discriminant. The mathematical foundation for identifying mineralization mutations was laid. Ultimately, these two variables were presented in the form of a grid diagram, which visually demonstrated the spatial distribution pattern of "driving forces" and "resistance forces" within the study area, and identified potential mutation regions with strong driving forces and weak resistance forces.

[0121] Through step S504, this embodiment successfully determined the drag control variable and the driving force control variable, laying the foundation for obtaining the cusp discriminant value through the cusp catastrophe theory.

[0122] In one embodiment of the intelligent mineral exploration method based on multimodal data and cybernetics models of this application, it may further include the following: Step S601: When the cusp mutation discrimination value is greater than 0, the target mineralization area is determined to be in a stable state, indicating that there is no industrial ore body; Step S602: When the cusp mutation discrimination value is equal to 0, the target mineralization area is determined to be in a critical state, indicating that there is small-scale mineralization; Step S603: When the cusp mutation discrimination value is less than 0, it is determined that a cusp mutation has occurred in the target ore-forming area, indicating the presence of an industrial ore body; Step S604: When the cusp mutation discrimination value is within a preset critical ambiguity range, the secondary discrimination process is initiated to determine the final mineral exploration target. The secondary discrimination process includes: Initiate the time dimension verification process, use multi-temporal synthetic aperture radar interferometry data to calculate the rate of change of the cusp mutation discrimination value over time. If the absolute value of the rate of change is greater than the preset annual rate of change threshold, the region is determined to be a dynamic evolution zone, which may be a false anomaly. Initiate the spatial correlation test process, calculate the spatial clustering index of the negative anomaly area with the known mineral deposits and the cusp mutation discrimination value. If the clustering index exceeds the preset confidence threshold within the preset spatial distance, the mineral exploration confidence of the anomaly area is improved.

[0123] Optionally, in this embodiment, this step is a process of obtaining mineralization guidance through cusp discriminant values.

[0124] Specifically, the system calculates... Values ​​are automatically categorized: when When the value is greater than 0, it means that the driving and resistance systems are in a single stable equilibrium state and have not reached the critical condition for phase transition. Therefore, the ore-forming system in the target area is directly determined to be "stable" and does not have the conditions for a sudden change to form an industrial ore body. This step efficiently eliminates a large area of ​​non-prospective regions.

[0125] when When the coefficient of energy exchange is 0, the system is at a mathematically critical bifurcation point, corresponding to a "critical state" of energy and matter exchange in geology. In this state, the system is unstable and may exhibit small-scale mineralization, but it is insufficient to support large-scale industrial agglomeration. This step identifies these transition zones or marginal areas, providing a refined intermediate level for evaluating resource potential.

[0126] when When the discriminant is less than 0, it indicates that the system meets the mathematical conditions for cusp mutation (the discriminant is less than zero, and the equilibrium surface folds). Geologically, this is explained as an imbalance between driving forces and resistance leading to a "first-order phase transition," which is the critical abrupt process by which minerals suddenly aggregate from a dispersed state into an industrial ore body. This step is the core of the method, directly delineating the high-potential "sweet spot" where industrial ore bodies are most likely to exist.

[0127] Optionally, in this embodiment, because the geological interpretation of Δ values ​​near the critical value is uncertain due to data noise and model errors in actual calculations, this embodiment designs a ambiguity elimination mechanism for the "critical ambiguity interval" where Δ values ​​are close to 0. Two independent types of evidence, time dimension testing and spatial correlation testing, are introduced for secondary discrimination.

[0128] Specifically, the time dimension test utilizes multi-temporal InSAR data to calculate the rate of change of the Δ value. If the rate of change is too large, it indicates that the abnormal signal in the area is changing rapidly, which may be caused by short-term geological activities that are not mineralized (such as landslides or groundwater changes). Therefore, it is classified as a "dynamic evolution zone" or a false anomaly, effectively filtering out transient interference.

[0129] Specifically, spatial correlation testing starts with spatial statistical regularities and calculates... Spatial clustering relationship between negative anomaly areas and known mineralization points. If negative anomaly areas and known mineralization points exhibit significant spatial clustering, the credibility of the anomaly being caused by similar mineralization processes is greatly enhanced. This step utilizes the spatial regularity that "similar geological environments produce similar mineral deposits," using known information to calibrate and improve the prediction confidence of new target areas.

[0130] Through step S604, this embodiment successfully realizes the automated mapping from continuous numerical values ​​to discrete geological conclusions. On this basis, intelligent verification and decision support for uncertain areas are added, which improves the geological rationality of the prediction results and the reliability in exploration practice.

[0131] In one embodiment of the intelligent mineral exploration method based on multimodal data and cybernetics models of this application, it may further include the following: Step S701: Convert the multimodal heterogeneous data to a preset geodetic coordinate system for sub-pixel level geometric fine correction, and determine the corresponding corrected multimodal data so that the planar positioning error of the data layer is less than 1 pixel; Step S702: Set a spatial resolution benchmark, and use a downsampling algorithm to protect high-frequency information to unify the corrected multimodal data with a resolution higher than the spatial resolution benchmark to the benchmark, and use an up-interpolation algorithm based on terrain constraints to unify the corrected multimodal data with a resolution lower than the spatial resolution benchmark to the benchmark. Step S703: Match and align the vertical stratigraphic levels of each modal data with the standard geological stratigraphic level after the resolution is unified, and determine the corresponding three-dimensional aligned multimodal data volume.

[0132] Optionally, in this embodiment, all remote sensing images, geophysical exploration data, geochemical data, and geological maps from different sources are first converted to a pre-defined unified geodetic coordinate system (such as the CGCS2000 coordinate system). Based on this, ground control points or feature-point-based image automatic matching algorithms are used to perform sub-pixel-level geometric correction on various types of data. This eliminates systematic geometric distortions caused by sensor attitude, terrain undulations, projection differences, etc., ensuring precise alignment of different data layers in the planar position, ultimately ensuring that the planar positioning error of each data source is less than one pixel.

[0133] Optionally, in this embodiment, the corrected data is scaled according to the target scale of the exploration task (e.g., setting 10 meters as a unified spatial resolution benchmark). For data with an original resolution higher than the benchmark (such as high-resolution satellite imagery), a downsampling algorithm that preserves high-frequency information (such as pixel aggregation or resampling with an anti-aliasing filter) is used to process the data, reducing the amount of data while preserving as much detail and texture as possible. For data with an original resolution lower than the benchmark (such as geophysical data of certain regions), an upscaling algorithm based on terrain constraints (such as inverse distance weighted interpolation combined with a digital elevation model or kriging interpolation) is used to compensate for insufficient information and ensure that the interpolation results conform to the terrain constraints. This eliminates scale effect differences between data, enabling all variables to be calculated and compared at the same spatial granularity.

[0134] Optionally, in this embodiment, for data volumes containing depth-dimensional information (such as seismic reflection profiles, borehole columnar sections, gravity or magnetic inversion models), vertical correction is performed based on a unified elevation datum. The key operation is to precisely match and align the vertical stratigraphic levels (such as reflection interfaces and lithological boundaries) of these data with regional standard geological stratigraphic levels or fixed-depth slices. This eliminates systematic biases in depth inversion or geological interpretation using different geophysical methods, constructing a unified three-dimensional data space that is strictly aligned in the X, Y, and Z dimensions.

[0135] Through step S703, this embodiment successfully constructed a three-dimensional aligned multimodal data volume, providing a seamless and accurate data foundation for subsequent extraction of geological slow variables across depth and attributes, three-dimensional gridded calculations, and full-space catastrophe analysis.

[0136] To improve the accuracy and efficiency of deep mineral exploration, this application provides an embodiment of an intelligent mineral exploration device based on multimodal data and a cybernetics model for implementing all or part of the aforementioned intelligent mineral exploration method based on multimodal data and a cybernetics model. See [link to embodiment]. Figure 2 The intelligent mineral exploration device based on multimodal data and cybernetics models specifically includes the following components: The multi-source data processing module 10 is used to acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, perform georegistration and resolution unification on the standardized multimodal heterogeneous data, and determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. The slow variable extraction module 20 is used to construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine multiple corresponding geological slow variables. The multiple geological slow variables are classified into resistance variables and driving force variables. The weights of each resistance variable and each driving force variable are calculated according to the ridge regression algorithm, and the variables after weight calculation are normalized to determine the corresponding resistance control variables and driving force control variables. The ore body determination module 30 is used to assign gridded values ​​to the target ore-forming region based on the resistance control variable and the driving force control variable, determine the corresponding gridded coefficient field, calculate the value of each grid point of the gridded coefficient field according to the Tom cusp catastrophe model, determine the cusp catastrophe discrimination value of the corresponding target ore-forming region, and determine the final prospecting target point based on the cusp catastrophe discrimination value.

[0137] As described above, the intelligent mineral exploration device based on multimodal data and cybernetics models provided in this application can obtain a multimodal data volume by fusing and aligning multi-source heterogeneous data from the target mining area. Based on the Haken enslavement principle and sparse statistical learning, key geological slow variables are extracted from the data volume as system order parameters and classified as resistance and driving force control variables for gridding. Then, the Tom cusp catastrophe theory is applied to calculate the catastrophe discriminant value in the gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, the mineral exploration target area is intelligently delineated and optimized, and high-confidence exploration targets are identified, thereby improving the accuracy and efficiency of deep mineral exploration.

[0138] To further illustrate this solution, this application also provides a specific application example of the intelligent mineral exploration device based on multimodal data and cybernetics model, which implements the intelligent mineral exploration method based on multimodal data and cybernetics model. The example includes the following: The AI ​​implementation code in this embodiment: %% New Feature: InSAR Atmospheric Correction and PS Point Selection (Crucial!) if exist('insarMat','file') Atmospheric delay correction using PyAPS or MintPy insar_corr = atmospheric_correction(insarFile, era5Data); % Screening permanent scatterers with coherence coefficient > 0.7 psMask = coherence>0.7; stress_grad = gradient(insar_corr) . psMask; % Only PS points are involved in the calculation else warning('InSAR data is missing; DEM approximation will be used; signal-to-noise ratio is expected to be <0.1'); stress_grad = gradient(dem) 0.1; % 0.1 is the empirical weighting coefficient. end %% New Feature: Propagation of Uncertainty in Proxy Variables Delta = b^2 + (8 / 27) a.^3; Delta_std = sqrt((2 b. b_std).^2 + ((8 / 9) a.^2. a_std).^2); % First-order error propagation Key Modification: Confidence level added to target region screening. confMask = Delta_std<3000; % Standard deviation threshold target = Delta<-8000&confMask&inROI; %% New: Bayesian sequential decision making (updated after the first borehole) if exist('borehole_data', 'var') Constructing a Gaussian process surrogate model gprMdl = fitrgp([lon5', lat5'], Delta(:), 'Sigma', Delta_std(:)); % Calculate the expected information gain for the next borehole. [nextPoint, infoGain] = bayesian_optimization(gprMdl, roiPoly); fprintf('Suggested next drill location: %.4f, %.4f (expected information gain %.2f)\n', ... nextPoint(1), nextPoint(2), infoGain); end From a hardware perspective, in order to improve the accuracy and efficiency of deep mineral exploration, this application provides an embodiment of an electronic device for implementing all or part of the intelligent mineral exploration method based on multimodal data and cybernetics models. The electronic device specifically includes the following components: The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the intelligent mineral exploration method based on multimodal data and cybernetics models and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the intelligent mineral exploration method based on multimodal data and cybernetics models in the previous embodiments, and the contents of these embodiments are incorporated herein, and repeated details will not be described again.

[0139] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0140] In practical applications, parts of the intelligent mineral exploration method based on multimodal data and cybernetics models can be executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0141] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0142] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0143] In one embodiment, the intelligent mineral exploration method based on multimodal data and cybernetics models can be integrated into a central processing unit 9100. The central processing unit 9100 can be configured to perform the following control: Step S101: Acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, and perform georegistration and resolution unification on the standardized multimodal heterogeneous data to determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. Step S102: Construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine the corresponding multiple geological slow variables. Classify the multiple geological slow variables into resistance variables and driving force variables, calculate the weights of each resistance variable and each driving force variable according to the ridge regression algorithm, and normalize each variable after weight calculation to determine the corresponding resistance control variables and driving force control variables. Step S103: Assign gridded values ​​to the target ore-forming region based on the resistance control variable and the driving force control variable, determine the corresponding gridded coefficient field, calculate the value of each grid point in the gridded coefficient field based on the Tom cusp catastrophe model, determine the cusp catastrophe discriminant value of the corresponding target ore-forming region, and determine the final prospecting target based on the cusp catastrophe discriminant value.

[0144] As described above, the electronic device provided in this application embodiment obtains a multimodal data volume by fusing and aligning multi-source heterogeneous data from the target mining area. Based on the Haken enslavement principle and sparse statistical learning, it extracts key geological slow variables from the data volume as system order parameters and classifies them as resistance and driving force control variables for gridding. Then, it applies the Tom cusp catastrophe theory to calculate the catastrophe discriminant value in the gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, it intelligently delineates and optimizes the prospecting target area and high-confidence exploration targets, thereby improving the accuracy and efficiency of deep prospecting.

[0145] In another implementation, the intelligent mineral exploration method based on multimodal data and cybernetics models can be configured separately from the central processing unit 9100. For example, the intelligent mineral exploration method based on multimodal data and cybernetics models can be configured as a chip connected to the central processing unit 9100, and the functions of the intelligent mineral exploration method based on multimodal data and cybernetics models can be realized through the control of the central processing unit.

[0146] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0147] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0148] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0149] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0150] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0151] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0152] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0153] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0154] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the intelligent mineral exploration method based on multimodal data and cybernetics models, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the intelligent mineral exploration method based on multimodal data and cybernetics models, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: Step S101: Acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, and perform georegistration and resolution unification on the standardized multimodal heterogeneous data to determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. Step S102: Construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine the corresponding multiple geological slow variables. Classify the multiple geological slow variables into resistance variables and driving force variables, calculate the weights of each resistance variable and each driving force variable according to the ridge regression algorithm, and normalize each variable after weight calculation to determine the corresponding resistance control variables and driving force control variables. Step S103: Assign gridded values ​​to the target ore-forming region based on the resistance control variable and the driving force control variable, determine the corresponding gridded coefficient field, calculate the value of each grid point in the gridded coefficient field based on the Tom cusp catastrophe model, determine the cusp catastrophe discriminant value of the corresponding target ore-forming region, and determine the final prospecting target based on the cusp catastrophe discriminant value.

[0155] As described above, the computer-readable storage medium provided in this application embodiment obtains a multimodal data volume by fusing and aligning multi-source heterogeneous data from the target mining area. Based on the Haken enslavement principle and sparse statistical learning, key geological slow variables are extracted from the data volume as system order parameters and classified as resistance and driving force control variables for gridding. Then, the Tom cusp catastrophe theory is applied to calculate the catastrophe discriminant value in the gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, the prospecting target area is intelligently delineated and optimized, and high-confidence exploration targets are identified, thereby improving the accuracy and efficiency of deep mineral exploration.

[0156] Embodiments of this application also provide a computer program product capable of implementing all steps of the intelligent mineral exploration method based on multimodal data and cybernetics models, where the execution subject is a server or client, as described in the above embodiments. When executed by a processor, this computer program / instruction implements the steps of the intelligent mineral exploration method based on multimodal data and cybernetics models. For example, the computer program / instruction implements the following steps: Step S101: Acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, and perform georegistration and resolution unification on the standardized multimodal heterogeneous data to determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. Step S102: Construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine the corresponding multiple geological slow variables. Classify the multiple geological slow variables into resistance variables and driving force variables, calculate the weights of each resistance variable and each driving force variable according to the ridge regression algorithm, and normalize each variable after weight calculation to determine the corresponding resistance control variables and driving force control variables. Step S103: Assign gridded values ​​to the target ore-forming region based on the resistance control variable and the driving force control variable, determine the corresponding gridded coefficient field, calculate the value of each grid point in the gridded coefficient field based on the Tom cusp catastrophe model, determine the cusp catastrophe discriminant value of the corresponding target ore-forming region, and determine the final prospecting target based on the cusp catastrophe discriminant value.

[0157] As described above, the computer program product provided in this application integrates and aligns multi-source heterogeneous data from the target mining area to obtain a multimodal data volume. Based on the Haken enslavement principle and sparse statistical learning, it extracts key geological slow variables from the data volume as system order parameters and classifies them as resistance and driving force control variables for gridding. Then, it applies the Tom cusp catastrophe theory to calculate the catastrophe discriminant value in the gridded area. Finally, through anomaly identification, uncertainty analysis, and Bayesian updates driven by real borehole data, it intelligently delineates and optimizes the prospecting target area and high-confidence exploration targets, thereby improving the accuracy and efficiency of deep prospecting.

[0158] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0160] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0161] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0162] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A smart mineral exploration method based on multimodal data and cybernetics models, characterized in that, The method includes: Acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, and perform georegistration and resolution unification on the standardized multimodal heterogeneous data to determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. A dimensionality reduction model is constructed based on the Haken enslavement principle, and a geological gradient constraint term is introduced into the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model. A sparse learning loss function is constructed based on the physical constraint dimensionality reduction model, and the sparse learning loss function is used to extract slow variables from the multimodal data volume to determine multiple corresponding geological slow variables. These multiple geological slow variables are classified into resistance variables and driving force variables. The weights of each resistance variable and each driving force variable are calculated using the ridge regression algorithm, and the variables after weight calculation are normalized to determine the corresponding resistance control variables and driving force control variables. The target ore-forming region is gridded and assigned values ​​based on the resistance control variables and the driving force control variables to determine the corresponding gridded coefficient field. The value of each grid point in the gridded coefficient field is calculated based on the Tom cusp catastrophe model to determine the cusp catastrophe discriminant value of the target ore-forming region. The final prospecting target is determined based on the cusp catastrophe discriminant value.

2. The intelligent mineral exploration method based on multimodal data and cybernetics model according to claim 1, characterized in that, The step of constructing a dimensionality reduction model based on the Haken enslavement principle, and introducing a geological gradient constraint term into the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model, includes: The key intrinsic parameters controlling the evolution of the ore-forming system are set as slow variables, and a large number of original observation parameters in the multimodal data volume are regarded as fast variables. A dimensionality reduction model is constructed, which is the standard Haken enslavement equation describing the evolution rate of fast variables. An explicit geological constraint term is added to the dimensionality reduction model to determine the corresponding physical constraint dimensionality reduction model. The geological gradient constraint term is a digital gradient field determined based on prior geological knowledge, which characterizes the spatial continuity prior of stratigraphic dip angle, lithological boundary or tectonic boundary, and is used to ensure that key geological constraints are not lost during the dimensionality reduction process.

3. The intelligent mineral exploration method based on multimodal data and cybernetics model according to claim 1, characterized in that, The construction of the sparse learning loss function guided by the physical constraint dimensionality reduction model includes: The data fitting term is constructed based on the physical constraint dimensionality reduction model described above, which is used to measure the error of reconstructing the original observation data by slow variables; Based on the physical constraint dimensionality reduction model, sparse constraint terms are constructed to screen slow variables that have a key impact on mineralization and suppress redundant features. Guided by the physical constraint dimensionality reduction model, a prior knowledge constraint term is constructed to introduce the regional mineralization experience represented by the geological gradient constraint term in the form of weighted priors, so as to guide the model learning to converge toward a more geologically reasonable direction. Guided by the physical constraint dimensionality reduction model, a spatial smoothing constraint term is constructed to constrain the smoothness of the changes of the extracted slow variables in adjacent spatial locations, so as to suppress data noise and enhance the regional continuity of the results. A loss function is constructed based on the data fitting term, the sparse constraint term, the prior knowledge constraint term, and the spatial smoothness constraint term. Then, the corresponding weight coefficients are set for each constraint term in the loss function using cross-validation to determine the corresponding sparse learning loss function.

4. The intelligent mineral exploration method based on multimodal data and cybernetics model according to claim 1, characterized in that, The step of extracting slow variables from the multimodal data volume using the sparse learning loss function to determine multiple corresponding geological slow variables includes: The multimodal data volume is fused according to a preset data fusion algorithm to construct a surface observable proxy index. The surface observable proxy index is a surface observable representation of deep slow variables in geological data. The data fusion algorithm includes: adjusting the basic weights of the multimodal data to be fused by penalizing or rewarding them based on the cloud cover degree and InSAR data point density of the multimodal data to be fused. Mutual information analysis and Granger causality test are performed on the observable proxy indicators of the land surface to determine the indicators whose mutual information is greater than the first threshold and whose Granger causality test value is less than the second threshold, which are then used as the candidate feature set for slow variable calculation. The candidate feature set is solved by the sparse learning loss function to determine the corresponding multiple slow geological variables.

5. The intelligent mineral exploration method based on multimodal data and cybernetics model according to claim 1, characterized in that, The process involves classifying the multiple geological variables into resistance variables and driving force variables, calculating the weights of each resistance variable and each driving force variable using the ridge regression algorithm, and normalizing the variables after weight calculation to determine the corresponding resistance control variables and driving force control variables, including: Based on the physical role of slow variables in geological processes, the aforementioned geological slow variables are classified into resistance variables and driving force variables, respectively. Based on prior geological knowledge of the target mining area, prior weights are assigned to each of the aforementioned resistance variables and driving force variables. Ridge regression is used to optimize the prior weights with the goal of minimizing prediction error, thereby determining the corresponding weights of each of the aforementioned resistance variables and driving force variables, and constraining the sum of the weight coefficients to be 1. The classified drag and driving force variables are standardized using Z-score to determine the corresponding normalized drag and driving force components. The normalized resistance components are weighted and summed according to the weights of the resistance variables to determine the corresponding resistance control variables. The normalized driving force components are weighted and summed according to the weights of the driving force variables to determine the corresponding driving force control variables.

6. The intelligent mineral exploration method based on multimodal data and cybernetics model according to claim 1, characterized in that, The step of determining the final prospecting target based on the cusp mutation discriminant value includes: When the cusp mutation discrimination value is greater than 0, the target mineralization area is determined to be in a stable state, indicating that there is no industrial ore body; When the cusp mutation discrimination value is equal to 0, the target mineralization area is determined to be in a critical state, indicating that there is small-scale mineralization. When the cusp mutation discrimination value is less than 0, it is determined that a cusp mutation has occurred in the target mineralization area, indicating the presence of an industrial ore body. When the cusp mutation discrimination value is within a preset critical ambiguity range, a secondary discrimination process is initiated to determine the final mineral exploration target. The secondary discrimination process includes: Initiate the time dimension verification process, use multi-temporal synthetic aperture radar interferometry data to calculate the rate of change of the cusp mutation discrimination value over time. If the absolute value of the rate of change is greater than the preset annual rate of change threshold, the region is determined to be a dynamic evolution zone, which may be a false anomaly. Initiate the spatial correlation test process, calculate the spatial clustering index of the negative anomaly area with the known mineral deposits and the cusp mutation discrimination value. If the clustering index exceeds the preset confidence threshold within the preset spatial distance, the mineral exploration confidence of the anomaly area is improved.

7. The intelligent mineral exploration method based on multimodal data and cybernetics model according to claim 1, characterized in that, The process of georegistering and unifying the resolution of the multimodal heterogeneous data to form a three-dimensional aligned data volume includes: The multimodal heterogeneous data is converted to a preset geodetic coordinate system and subjected to sub-pixel level geometric fine correction to determine the corresponding corrected multimodal data so that the planar positioning error of the data layer is less than 1 pixel. A spatial resolution benchmark is set, and the corrected multimodal data with a resolution higher than the spatial resolution benchmark are unified to the benchmark according to the downsampling algorithm that protects high-frequency information. The corrected multimodal data with a resolution lower than the spatial resolution benchmark are unified to the benchmark according to the upinterpolation algorithm based on terrain constraints. The vertical stratigraphic levels of each modal data, after being standardized in resolution, are matched and aligned with standard geological stratigraphic levels to determine the corresponding three-dimensional aligned multimodal data volume.

8. An intelligent mineral exploration device based on multimodal data and a cybernetics model, characterized in that, The device includes: A multi-source data processing module is used to acquire multimodal heterogeneous data of the target mineralization area, standardize the multimodal heterogeneous data, perform georegistration and resolution unification on the standardized multimodal heterogeneous data, and determine the corresponding three-dimensional aligned multimodal data volume. The multimodal heterogeneous data includes satellite remote sensing data, geophysical data, geochemical data, and geological data. The slow variable extraction module is used to construct a dimensionality reduction model based on the Haken enslavement principle, introduce a geological gradient constraint term into the dimensionality reduction model, determine the corresponding physical constraint dimensionality reduction model, construct a sparse learning loss function guided by the physical constraint dimensionality reduction model, and extract slow variables from the multimodal data volume through the sparse learning loss function to determine multiple corresponding geological slow variables. The multiple geological slow variables are classified into resistance variables and driving force variables. The weights of each resistance variable and each driving force variable are calculated according to the ridge regression algorithm, and the variables after weight calculation are normalized to determine the corresponding resistance control variables and driving force control variables. The ore body determination module is used to assign gridded values ​​to the target ore-forming region based on the resistance control variables and the driving force control variables, determine the corresponding gridded coefficient field, calculate the value of each grid point of the gridded coefficient field according to the Tom cusp catastrophe model, determine the cusp catastrophe discrimination value of the corresponding target ore-forming region, and determine the final prospecting target point based on the cusp catastrophe discrimination value.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent mineral exploration method based on multimodal data and cybernetics model as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the intelligent mineral exploration method based on multimodal data and a cybernetics model as described in any one of claims 1 to 7.

Images