A manganese ore exploration system based on multi-source data

Abstract

The present application relates to a kind of manganese exploration system based on multi-source data, belong to mineral resources exploration technical field.The system constructs parameterized combination constraint field by data processing module, constructs multi-scale geological topology network and ore-forming cause skeleton by ore-forming associated feature module, and effective ore-forming characteristic subset of geological mechanism adaptation is filtered by combining constraint type parameter learning and sensitivity analysis;Core is through ore-forming potential evaluation module to construct global ore-forming favorable degree grid and three-dimensional physical-structure coupling field, realizes multi-solution mineralization configuration inversion based on prior constraint, is coupled with multi-physical field decoupling forward modeling, residual error decoupling sequencing and discrete scene set bayesian posteriori approximate inference, complete the self-checking and uncertainty quantification of mineralization prediction;Finally, through exploration target area optimization module, multi-dimensional constraint conditions are fused to construct multi-objective optimization function, and non-dominated exploration target area scheme set is obtained by solving.The present application improves the accuracy and input-output benefit of manganese deep, concealed ore body exploration.

Description

Technical Field

[0001] This invention belongs to the field of mineral resource exploration technology, specifically relating to a manganese ore exploration system based on multi-source data. Background Technology

[0002] Manganese ore is a strategic mineral resource in my country, widely used in core industrial sectors such as metallurgy, chemicals, and new energy. With the gradual depletion of shallow, exposed manganese ore resources, exploration has shifted towards deep, concealed, and complex manganese ore bodies, relying on intelligent exploration and prediction using multi-source heterogeneous data from geology, geophysical exploration, remote sensing, and drilling. Currently, multi-source data manganese ore exploration technology still suffers from many core technical deficiencies, severely restricting the accuracy of deep manganese ore exploration and the effectiveness of engineering implementation.

[0003] Traditional multi-source data normalization processing relies entirely on mathematical algorithms such as statistical filtering and mean calibration. It fails to incorporate geological constraints by considering manganese ore-specific sedimentary patterns, stratigraphic characteristics, and tectonic ore-controlling rules. This makes it impossible to distinguish between valid anomalies caused by mineralization and false anomalies resulting from environmental or instrumental interference. Consequently, it is prone to discarding valid prospecting information while retaining false data that violates geological principles. Furthermore, current technologies lack a hierarchical data diversion mechanism, resulting in all data being used indiscriminately for modeling. This leads to poor geological rationality in the basic dataset, reducing the reliability of mineralization predictions from the outset.

[0004] Existing methods for screening mineralization features rely solely on pure statistical correlation analysis to extract features, failing to construct causal logical constraints on the mineralization process. They cannot distinguish between statistical correlations and geological genetic correlations, resulting in a large number of pseudo-features without mechanistic support being mixed into the model. Furthermore, the lack of a multi-scale geological topological correlation system and a causal framework for mineralization constraints leads to parameter learning without clear geological boundaries, low accuracy in determining feature contribution, and ultimately, poor fit and weak generalization ability in the mineralization evaluation model.

[0005] Current mainstream mineralization prediction technologies are based on a single, one-way inversion prediction model, which has serious technical shortcomings: First, it can only generate a single mineralization prediction result, ignoring the multifaceted nature of geological exploration and failing to cover multiple geological interpretation schemes, easily leading to missed or misjudged ore bodies; second, it lacks a multi-physics forward modeling verification loop, with model prediction results relying solely on data fitting without physical mechanism verification, making it prone to overfitting and generating seemingly accurate but ultimately flawed predictions; third, it cannot achieve source decoupling and sensitivity analysis of residual errors, failing to accurately identify prediction deviations caused by different factors such as model errors, structural simplification, and overburden interference, lacking uncertainty quantification capabilities, and making it impossible to determine the reliability of prediction results; fourth, traditional Bayesian inference often uses a single prior model, failing to incorporate multi-scenario geological configuration characteristics, and cannot adapt to the complex mineralization geological conditions of manganese deposits.

[0006] In summary, existing manganese ore exploration technologies suffer from technical defects such as data processing without geological constraints, pseudo-correlation interference in feature screening, lack of closed-loop verification in prediction models, unquantifiable uncertainty in results, and target area layout that is divorced from engineering practice. These defects prevent the realization of high-precision, verifiable, and practical intelligent manganese ore exploration. Therefore, a manganese ore exploration system based on multi-source data, driven by geological mechanisms, and possessing dual closed-loop verification and multi-objective optimization capabilities is needed. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a manganese ore exploration system based on multi-source data. The objective of this invention can be achieved through the following technical solutions: A manganese ore exploration system based on multi-source data, comprising: The data processing module constructs a parameterized lithology-stratigraphy-structure combination constraint domain based on multi-source geological exploration data, performs multi-dimensional data consistency verification and outlier truncation, and outputs three-level data distribution. The mineralization correlation feature module constructs a multi-scale geological element topological correlation network to generate a directed acyclic causal skeleton of the mineralization process; it performs constrained parameter learning and global sensitivity analysis to screen feature subsets that have mechanistic consistency with the manganese mineralization process. The mineralization potential assessment module constructs an initial mineralization favorability spatial grid and a three-dimensional physical property-structure coupled field. Using the mineralization favorability field as a prior constraint, it performs parametric physical property inversion to generate a set of three-dimensional configurations of candidate mineralization bodies with multiple solutions. For each configuration, it conducts physical field decoupling forward simulation, calculates the residual field with the measured multi-source observation field, performs multi-source error decoupling and sensitivity ranking of the residual field, and performs Bayesian posterior approximate inference based on the discrete scene set. The exploration target area optimization module constructs an optimization objective function based on the constraints of the exploration project, and solves it to generate a set of non-dominated exploration target area schemes.

[0008] Specifically, the process of parameterizing the lithology-stratigraphy-tectonic combination constraint domain includes: The system obtains the regional sedimentary evolution time sequence, the hierarchical matching rules of fault structures, and the specific stratigraphic and lithological combination rules of manganese ore occurrence, generates system-limiting rules, and aggregates them to form the parameterized lithology-stratigraphic-tectonic combination constraint domain. Using a single sampled data point as the traversal unit, the three elements of the sample's stratigraphy, surrounding rock lithology, and surrounding structural development are sequentially matched with the parameterized lithology-stratigraphy-structural combination constraint domain, outputting a three-level data split: a preserved dataset, a suppressed dataset, and a conflict dataset.

[0009] Specifically, the process of establishing the multi-scale geological element topological association network includes: Based on spatial adjacency and the dependent relationship of material migration and enrichment, a multi-scale topological link is constructed across shallow, middle and deep layers. According to the evolution sequence of manganese ore, the evolution logic of unidirectional driving between various ore-controlling elements is obtained. Network nodes and directional correlation edges are added one by one in the order of mineralization. After the formation is completed, an editing channel for adding nodes and revising correlation relationships is opened, generating a directed acyclic causal skeleton of the mineralization process.

[0010] Specifically, the process of generating the feature subset includes: The computational boundary of constrained parameter learning is limited to the element links that have been connected in the causal framework. Within the constrained boundary, the degree of correlation between the changes of a single type of geological element and the mineralization process is calculated step by step. A sensitivity test mode of individual perturbation is adopted, and the parameter state of a type of feature is changed once to obtain the fluctuation range of the mineralization evaluation results of the whole domain. Based on the fluctuation level, the feature contribution is distinguished, and redundant pseudo-features that are correlated at the mathematical and statistical level but lack the support of sedimentary tectonic mineralization control mechanism are eliminated. The remaining elements with genetic support are summarized to generate the feature subset.

[0011] Specifically, the process of the initial mineralization favorable spatial grid and the three-dimensional property-structure coupling field includes: Based on the spatial ore-controlling mechanism of various ore-controlling elements within the feature subset, a hierarchical ore-forming potential inversion architecture is constructed according to the action sequence and influence range. For various exploration features with discrete point distribution, the point information is extended and filled into the continuous spatial unit of the whole domain. Through iterative calculation, an initial ore-forming favorable spatial grid with continuous planar distribution is generated. The morphology of stratigraphic spatial distribution, the occurrence and distribution of fault structures, and the measured physical property parameters of various geological bodies are integrated simultaneously to construct a three-dimensional physical property-structure coupling field.

[0012] Specifically, the process of generating the three-dimensional configuration set of the multi-solution candidate mineralization includes: Based on the stratigraphic distribution and property variation laws stored in the three-dimensional property-structure coupled field, the inversion prior boundary conditions are used to summarize the in-situ measured property data and delineate the property floating range corresponding to different lithological geological bodies. Under the dual constraints of inversion prior boundary conditions and property floating range, parametric property inversion is initiated round by round by alternating different property combination paths. Each set of differentiated parameter combinations corresponds to an independent mineralization spatial morphology. All spatial morphology results are summarized to generate the three-dimensional configuration set of the multi-solution candidate mineralization body.

[0013] Specifically, the specific process of the physical field decoupling forward model includes: From the set of three-dimensional configurations of the multi-solution candidate mineralization bodies, individual mineralization three-dimensional configurations are retrieved one by one. Based on the surface outcrop boundary, deep stratigraphic interface, and basin basement undulation state of the study area, the global boundary constraints of the forward modeling operation are defined. The physical mechanisms of the generation and conduction of the field sources corresponding to the magnetic field, gravity field, and resistivity field are separated. The numerical solutions of single physical fields are carried out independently in modules. The theoretical response values ​​of the single physical field are calculated in sequence over the entire domain. Then, the calculation results of the three types of physical fields are spliced ​​together under the same spatial grid framework to generate a theoretical observation field that corresponds one-to-one with the spatial location of the measured points.

[0014] Specifically, the residual decoupling and sensitivity ranking process includes: Using a unified spatial grid as a benchmark, the numerical differences between the theoretical observation field and the field measured exploration field at the same location are compared unit by unit, and all grid difference data are collected to form a complete residual field for the entire domain. Based on the regional stratigraphic cover conditions, structural simplification degree, and model parameters, three types of geological influencing factors were selected to trace the residual causes. The residuals were then split into three independent error components according to the cause category. The parameters of a single error component were modified at a time, and the fluctuation range of the overall residual field was recorded. All error components were sorted according to the residual fluctuation intensity caused by the component modification.

[0015] Specifically, the Bayesian approximate inference process includes: A single set of candidate mineralization 3D configurations is bound and encapsulated with corresponding sub-item residual data. Each binding is defined as an independent discrete scene. All corresponding scenes in the configuration set are aggregated to form a global discrete scene set. Bayesian approximate inference is carried out using the multi-solution geological information contained in the whole scene as prior material. The response characteristics of all discrete scenes are traversed, and the frequency ratio of mineralization anomalies in each spatial unit under different geological interpretation scenarios is statistically analyzed to obtain the probability distribution law of mineralization events in each spatial unit.

[0016] Specifically, the process of the non-dominated exploration target area scheme set includes: A multi-objective optimization function is constructed by integrating five constraints: mineralization probability of each spatial unit, potential geological information gain of the target area, environmental limitations of field drilling construction, project exploration investment amount, and regional ecological protection and control. Based on the multi-objective population iterative optimization logic, the borehole layout points are continuously adjusted, and invalid layout schemes with simultaneous deterioration of multiple evaluation indicators are gradually eliminated to generate the non-dominated exploration target area scheme set.

[0017] The beneficial effects of this invention are as follows: This invention constructs a parameterized combination constraint domain based on the depositional time, stratigraphic occurrence, and structural matching patterns of manganese ore, and uses geological mechanisms as prior constraints to conduct data consistency verification. Through a three-level data diversion mechanism, compliant mineralization data is retained, non-geological pseudo-anomalies are isolated, and questionable conflicting data is collected. This avoids the blind rejection of effective mineral exploration information and prevents false anomaly data from interfering with modeling. At the same time, conflicting data is retained for manual review and calibration. From the data source, the rationality, accuracy, and traceability of multi-source datasets are guaranteed, providing a reliable data foundation for subsequent mineralization modeling.

[0018] A cross-scale geological element topological correlation network and a directed acyclic causal framework for mineralization are constructed to define the geological logical boundaries of parameter learning and prevent unrelated elements from participating in the modeling. Through global sensitivity experiments with feature-by-feature perturbation, the mineralization contribution of each element is quantified, and pseudo-features with only statistical correlation and no sedimentary-tectonic mineralization control mechanism are precisely eliminated, retaining only a subset of effective features with clear mineralization causal relationships. This effectively solves the problems of feature redundancy, fitting distortion, and poor generalization in traditional models, ensuring that the mineralization evaluation model fully conforms to the actual mineralization evolution of manganese deposits.

[0019] Based on a refined ore-controlling mechanism, a hierarchical inversion architecture is constructed to generate a continuous ore-forming favorable grid and a three-dimensional property-structure coupling field, achieving integrated ore-forming characterization in two-dimensional plane and three-dimensional space. A multi-solution mineralization three-dimensional configuration set is generated based on prior boundaries and property floating ranges, aligning with the multi-solution nature of geological exploration and avoiding the one-sidedness of a single model. Through multi-physics field decoupled forward simulation, the theoretical responses of magnetic, gravitational, and resistivity fields are independently calculated, verifying the physical mechanism of the prediction model and eliminating data overfitting. The decoupling and sensitivity ranking of multi-source errors in the residual field are completed, tracing the three types of error causes—model parameters, structural simplification, and overburden interference—to clarify the core sources of prediction bias. Finally, Bayesian posterior approximate inference is conducted based on discrete scene sets, combining multiple geological scenes to quantify the mineralization probability distribution of each spatial unit, achieving precise quantification of the uncertainty of the prediction results. Attached Figure Description

[0020] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0021] Figure 1 This is an overall architecture diagram of a manganese ore exploration system based on multi-source data according to the present invention; Figure 2 This is a flowchart of the causal skeleton and feature subset generation process in this invention; Figure 3 This diagram illustrates the process of generating multiple-solution candidate configurations and Bayesian inference in this invention. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0023] Please see Figures 1-3 A manganese ore exploration system based on multi-source data, comprising: The data processing module constructs a parameterized lithology-stratigraphy-structure combination constraint domain based on multi-source geological exploration data, performs multi-dimensional data consistency verification and outlier truncation, and outputs three-level data distribution. The mineralization correlation feature module constructs a multi-scale geological element topological correlation network to generate a directed acyclic causal skeleton of the mineralization process; it performs constrained parameter learning and global sensitivity analysis to screen feature subsets that have mechanistic consistency with the manganese mineralization process. The mineralization potential assessment module constructs an initial mineralization favorability spatial grid and a three-dimensional physical property-structure coupled field. Using the mineralization favorability field as a prior constraint, it performs parametric physical property inversion to generate a set of three-dimensional configurations of candidate mineralization bodies with multiple solutions. For each configuration, it conducts physical field decoupling forward simulation, calculates the residual field with the measured multi-source observation field, performs multi-source error decoupling and sensitivity ranking of the residual field, and performs Bayesian posterior approximate inference based on the discrete scene set. The exploration target area optimization module constructs an optimization objective function based on the constraints of the exploration project, and solves it to generate a set of non-dominated exploration target area schemes.

[0024] Specifically, the process of parameterizing the lithology-stratigraphy-tectonic combination constraint domain includes: By analyzing the regional sedimentary evolution sequence, the hierarchical and matching patterns of fault structures, and the specific stratigraphic and lithological combinations of manganese deposits, these geological patterns are refined into a set of system-identifiable constraints and aggregated into parameterized constraint domains. The system uses individual sampled data as traversal units, sequentially checking whether the three elements of the sample's stratigraphy, surrounding rock lithology, and surrounding structural development fall within the constraints of the domain. Samples with all three elements matching are included in the reserved dataset. Observational data whose element combinations significantly deviate from the geological conditions of manganese deposits and can be identified as non-geological anomalies are truncated and isolated and included in the suppressed dataset. Samples whose anomaly causes cannot be determined by existing mineralization constraints, or whose anomaly locations simultaneously possess mineralization potential and unreasonable characteristics, are separately archived to form conflict datasets.

[0025] Specifically, the process of establishing the multi-scale geological element topological association network includes: Based on the spatial adjacency relationships and the dependent relationships of material migration and enrichment in the normalized dataset, a multi-scale topological link is constructed across shallow, middle and deep layers. Following the complete evolutionary sequence of manganese ore deposition and emplacement, compaction and diagenesis, tectonic fracturing and alteration, and hydrothermal superposition and enrichment, the evolutionary logic of unidirectional driving among various ore-controlling elements is sorted out. Network nodes and directional correlation edges are added one by one in the order of mineralization. After the formation is completed, an editing channel for adding nodes and revising correlation relationships is opened, and finally an iterative directed acyclic causal skeleton adapted to the regional mineralization law is formed.

[0026] Specifically, the process of generating the feature subset includes: The computational boundary of parameter learning is limited to the element links that are already connected in the causal framework, and elements outside the framework that have no logical connection are excluded from the correlation degree calculation. Within the constrained boundary, the correlation degree of single geological elements with the mineralization evolution process is calculated step by step. A sensitivity test mode of individual perturbation is adopted, changing the parameter state of only one type of feature at a time, observing the fluctuation range of the mineralization evaluation results of the whole domain, distinguishing the strength of feature contribution based on the fluctuation level, eliminating redundant pseudo-features that only have correlation at the mathematical and statistical level and lack the support of sedimentary tectonic mineralization control mechanism, and summarizing the remaining elements with genetic support to form the final feature subset.

[0027] Specifically, the process of the initial mineralization favorable spatial grid and the three-dimensional property-structure coupling field includes: The spatial ore-controlling mechanism of various ore-controlling elements within the feature subset is deconstructed. A hierarchical ore-forming potential inversion architecture is constructed according to the timing and scope of the elements' effects. For various exploration features with discrete point distribution, spatial interpolation is used to extend and fill the point information to continuous spatial units across the entire domain. Through iterative model calculations, a spatial grid of ore-forming favorability with continuous planar distribution is generated. The spatial distribution of stratigraphy, the occurrence and distribution of fault structures, and the measured physical property parameters of various geological bodies are integrated simultaneously to construct a three-dimensional physical property-structure coupling field that combines geometric morphology and physical property attributes.

[0028] Specifically, the process of generating the three-dimensional configuration set of the multi-solution candidate mineralization includes: Based on the stratigraphic distribution and property variation patterns stored in the three-dimensional property-structure coupled field, the inversion prior boundary conditions are used. In-situ measured property data obtained from field outcrop sampling and whole-hole core testing are summarized to delineate reasonable property fluctuation ranges for different lithological geological bodies. Under the dual constraints of the prior boundary and property fluctuation range, parametric property inversion is initiated round by round by alternating different property combination paths. Each set of differentiated parameter combinations corresponds to an independent mineralization spatial morphology. All spatial morphology results from the calculations are summarized to form a three-dimensional configuration set of candidate mineralization bodies with multiple geological interpretation schemes.

[0029] Specifically, the specific process of the physical field decoupling forward model includes: From the set of three-dimensional configurations of the multi-solution candidate mineralization bodies, individual mineralization three-dimensional configurations are retrieved one by one. Based on the surface outcrop boundary, deep stratigraphic interface, and basin basement undulation state of the study area, the global boundary constraints of the forward modeling operation are defined. The physical mechanisms of the generation and conduction of the field sources corresponding to the magnetic field, gravity field, and resistivity field are separated. The numerical solutions of single physical fields are carried out independently in modules. The theoretical response values ​​of the single physical field are calculated in sequence over the entire domain. Then, the calculation results of the three types of physical fields are spliced ​​together under the same spatial grid framework to generate a theoretical observation field that corresponds one-to-one with the spatial location of the measured points.

[0030] Specifically, the residual decoupling and sensitivity ranking process includes: Using a unified spatial grid as a benchmark, the numerical differences between the theoretical observation field and the field measured exploration field at the same location are compared unit by unit, and all grid difference data are collected to form a complete residual field for the entire region. Combining regional stratigraphic cover conditions, structural simplification degree, and model parameters, three types of geological influencing factors are selected to trace the residual inducing factors, and three independent error components are obtained according to the type of inducing factor. Using the controlled variable experiment method, the parameter conditions of only a single type of error component are changed at a time, and the fluctuation range of the overall residual field is recorded. The priority of all error components is completed according to the strength of residual fluctuation caused by component changes.

[0031] Specifically, the Bayesian approximate inference process includes: A single set of candidate mineralization 3D configurations is bound and encapsulated with the residual data of the score items obtained from the operation of the configuration. Each set of bound results is defined as an independent discrete scene. All corresponding scenes of the configuration set are collected to form a complete global discrete scene set. Bayesian approximate inference is carried out using the multi-solution geological information contained in the whole scene as prior material. The response characteristics of all discrete scenes are traversed, and the frequency ratio of mineralization anomalies in each spatial unit under different geological interpretation scenarios is statistically analyzed. The complete probability distribution law of mineralization events in each spatial unit is derived.

[0032] Based on the mineralization probability distribution data of each spatial unit in multiple scenarios, combined with the uncertainty index represented by information entropy, the model internal closed-loop verification is completed by comparing with the measured known mineral point data, thereby realizing the self-verification of the prediction results; combined with the cross-combination relationship between the mineralization probability of the unit and the magnitude of the uncertainty index, multi-level confidence levels are divided step by step, and at the same time, referring to the previous residual source tracing conclusions, the dominant error type that causes local prediction deviation is marked in the attached label of each grid unit.

[0033] Specifically, the process of the non-dominated exploration target area scheme set includes: By integrating five constraints—mineralization probability of each spatial unit, potential geological information gain of the target area, environmental limitations of field drilling, project exploration investment, and regional ecological protection and management—a multi-objective optimization function is constructed. Based on the multi-objective population iterative optimization logic, the borehole layout locations are continuously adjusted, and invalid layout schemes with multiple evaluation indicators deteriorating simultaneously are gradually eliminated. Finally, a set of multiple non-dominant exploration target area schemes that cannot be simultaneously optimal and mutually restrain each other are retained.

[0034] This invention provides a manganese ore exploration system based on multi-source data, covering the entire technical process of multi-source data processing, mineralization feature screening, dual-closed-loop mineralization potential assessment, and multi-target area optimization. The implementation process is as follows: 1. Implementation process of multi-source data normalization The system pre-computes the parameterized encapsulation of geological prior rules for the study area, collecting a set of sedimentary evolution time-series parameters (S1), a set of fault structure hierarchical parameters (S2), and a set of manganese ore-specific occurrence stratigraphic and lithological combination parameters (S3) across the entire study area. These three parameter sets correspond to the temporal constraints, tectonic spatial constraints, and lithological material constraints of manganese ore formation, respectively. The S1, S2, and S3 parameter sets are structurally coupled and encapsulated, unifying parameter dimensions and constraint threshold formats, and eliminating invalid prior parameters with conflicting dimensions and contradictory rules. Finally, a parameterized lithology-stratigraphic-tectonic combination constraint domain (C0) with computable and matchable characteristics is generated. Constraint domain C0 contains a complete and traversable system of rules for determining the compliant occurrence of manganese ore.

[0035] The system imports the preprocessed multi-source geological exploration sampling dataset D0. Dataset D0 integrates heterogeneous data from geological logging, geophysical and geochemical exploration, remote sensing, and drilling. All data have undergone prior spatiotemporal benchmarking and semantic unification. The system uses single sampling samples D... nAs the smallest independent traversal unit, three core matching parameters are extracted from each sample: stratigraphic attribute parameter P1, surrounding rock lithology parameter P2, and surrounding structural development parameter P3. These three parameters are simultaneously substituted into the rule-based judgment model of constraint domain C0, and multi-dimensional parallel consistency verification is performed: if P1, P2, and P3 all fall within the pre-defined compliance parameter range of constraint domain C0, the sample is determined to be geologically compliant and valid, and is included in the retained dataset Dvalid; if the combined characteristics of the three parameters completely deviate from the manganese ore occurrence geological rules of constraint domain C0, it is determined to be a non-geological pseudo-anomaly caused by instrument interference or environmental noise, and after anomaly truncation and isolation processing, it is included in the suppressed dataset Dnoise; if a single parameter is compliant, but the other parameters are anomalous, and the parameter matching results are contradictory, making it impossible to determine the anomaly's cause attribute through existing constraint rules, it is separately collected into the conflict dataset Dconf. After completing the full-domain sample traversal verification, the system outputs three levels of independent split datasets: Dvalid, Dnoise, and Dconf, achieving compliance screening of geological mechanisms in multi-source data and avoiding the technical defects of indiscriminate statistical noise reduction from the source.

[0036] 2. Implementation process of screening mineralization correlation features The system uses the compliant and preserved dataset Dvalid as its basic data source. It first extracts spatial adjacency correlation parameters R1 and material migration and enrichment dependency correlation parameters R2 from geological elements across the entire region. Based on these two types of correlation parameters, and according to three geological spatial scales (V1, V2, and V3), a cross-scale geological element topological correlation network Gnet is constructed to achieve spatial correlation coupling between geological bodies at different depths and different ore-controlling elements. The system retrieves the complete mineralization evolution time-series parameter Tall for manganese ore, covering four major mineralization stages: sedimentary emplacement, compaction and diagenesis, tectonic fracturing and alteration, and hydrothermal superposition and enrichment. Based on the time-series Tall, the system analyzes the unidirectional driving and hierarchical causal evolution logic between quality-controlling ore elements in various regions. According to the chronological order of mineralization, geological nodes (Nod) and directional related edges (Edg) are added to the topological network in sequence to clarify the role hierarchy and driving relationship of each node element. After the network is built, parameter editing channels for adding nodes, deleting related edges, and revising related weights are opened to support iterative adaptation to regional mineralization patterns. Finally, a directed acyclic causal framework (Gcas) for the mineralization process adapted to the mineralization patterns of manganese ore in the study area is generated.

[0037] The system uses the effective connectivity links of the causal framework Gcas as the absolute operational boundary to delineate the effective operational domain Fsp for constrained parameter learning. It shields redundant elements outside the framework that lack causal logic and geological correlation, preventing invalid parameters from participating in modeling calculations. Within the operational domain Fsp, it traverses various geological feature parameters Ftr step by step, calculating the correlation coefficient Ctr of each set of feature parameters dynamically changing with the mineralization evolution time series Tall, quantifying the basic correlation level between features and the mineralization process. Furthermore, a single-variable independent controlled perturbation experiment mechanism is adopted, keeping other geological parameters constant across the entire domain, and applying a small perturbation variable ΔFtr only to a single feature parameter Ftr in a single iteration, simultaneously monitoring the fluctuation range ΔScr of the overall mineralization evaluation results. Based on the magnitude of the fluctuation of ΔScr, the mineralization contribution weight Wtr of each feature parameter is calculated using normalization. The system presets a geological mechanism judgment threshold Thr, automatically identifies and removes pseudo-feature parameters that only meet the mathematical and statistical correlation and have no sedimentary-tectonic mineralization control mechanism support, accurately retains effective feature parameters with clear causal relationship of mineralization and qualified contribution weight, and finally integrates and iteratively generates a mineralization feature subset Fsel with consistent mechanism.

[0038] 3. Implementation Process of Intelligent Assessment of Mineralization Potential The first step is the construction of a comprehensive, integrated field data system. Based on the feature subset Fsel, the system decomposes and extracts the spatial ore-controlling mechanisms, temporal sequence, and influence range parameters of various ore-controlling elements. According to the hierarchical action of the elements and differences in spatial influence radii, a multi-scale coupled mineralization potential inversion model, Minv, is constructed layer by layer. Addressing the discrete point distribution characteristics of the original exploration data, an adaptive spatial interpolation iterative algorithm is used to extend, fill, and smooth the discrete point feature parameters grid by grid to a continuous spatial grid cell across the entire region. After multiple rounds of iterative convergence, an initial mineralization favorable spatial raster, Map2d, with full horizontal coverage and continuous vertical transition, is generated. Simultaneously, the system integrates the stratigraphic spatial distribution morphology parameter Stru, the fault structure occurrence parameter Fau, and the in-situ measured physical property parameters Phy of various geological bodies. This completes the accurate registration and fusion of geometric structure data and physical attribute data, constructing a three-dimensional physical property-structure coupled field, Map3d, that combines three-dimensional spatial geometry, stratigraphic relationships, and physical property distribution characteristics. This enables the digital and parameterized comprehensive characterization of geological bodies in the study area.

[0039] The second step involves iterative generation of a multi-solution three-dimensional mineralization configuration set. The system deeply extracts the stratigraphic distribution patterns, spatial gradation patterns of physical properties, and tectonic ore-controlling morphology patterns stored within the three-dimensional property-structure coupled field Map3d. These patterns are parameterized and encapsulated to construct the inversion prior boundary condition Bpri, which limits the spatial range of mineralization bodies and the baseline of basic physical properties. In-situ measured data from field outcrop sampling, whole-hole core testing, and physical property experiments in the study area are summarized. The physical property fluctuation characteristics of geological bodies with different lithologies and tectonic locations are statistically analyzed, and reasonable fluctuation ranges Rphy for physical property parameters corresponding to various geological bodies are defined. Under the dual rigid constraints of the prior boundary condition Bpri and the physical property fluctuation range Rphy, the system adopts a multi-path parameter iterative inversion strategy, alternately switching the combination schemes of physical property parameters such as density, magnetic susceptibility, and resistivity in round-by-round. Each set of differentiated parameter combinations independently initiates a parameterized physical property inversion operation, solving for a unique corresponding mineralization spatial morphology configuration Modn. The system iterates repeatedly to exhaust all compliant parameter combination paths, collects all differentiated and independently interpretable mineralization configurations, and finally forms a multi-solution candidate mineralization three-dimensional configuration set Modset covering multiple geological interpretation schemes. This conforms to the multi-solution nature of geological exploration and avoids the one-sidedness of a single model.

[0040] The third step is the decoupled forward modeling of multiple physics fields. The system sequentially retrieves a single set of three-dimensional mineralization configurations (Modn) from the multiple deconstruction sets (Modset). Combining the geological conditions of the entire study area, such as surface outcrops, deep stratigraphic interfaces, basin basement undulations, and groundwater distribution, the system parameterizes the spatial boundary constraints (Bsim) and the property stratification constraints (Lsim) for the forward modeling. Abandoning the traditional multi-field coupled mixed calculation mode, a decoupled step-by-step calculation mechanism is adopted. The magnetic field forward modeling mechanism equation (Fmag), the gravity field forward modeling mechanism equation (Fgra), and the resistivity field forward modeling mechanism equation (Fres), adapted to the geological characteristics of manganese deposits, are retrieved separately. The system independently solves the global grid theoretical response values ​​of the three types of physical fields in modules and steps, avoiding the problem of mutual interference between multiple field parameters. After all single-physics field calculations converge, the theoretical response data of the three types of physics fields are accurately spliced ​​and aligned in a unified spatial grid coordinate system. This generates a global integrated theoretical observation field Fsim that perfectly matches the spatial coordinates and grid scale of the field measurement points, ensuring that the forward modeling data and the measured data can be accurately compared and verified.

[0041] The fourth step is residual multi-source decoupling and sensitivity hierarchical ranking. Using a unified global spatial grid as a benchmark, the system compares the parameter differences between the theoretical observation field Fsim and the field measured exploration field Fobs point-to-point, collecting the difference data from all grid cells and fitting them to generate a complete, continuous, and fully covered residual field, Resall. Combining geological influence parameters such as regional stratigraphic cover thickness, structural simplification, and model parameter selection rules, residual source analysis is conducted, precisely decoupling and separating the global residual field into three independent error components: model fitting residual component Resfit, cover layer interference residual component Rescov, and geological generalization residual component Resgeo. A single-variable controlled perturbation comparison method is used, keeping the parameters of the other two error components constant, and only fine-tuning the threshold parameters of one type of residual component at a time. This monitors the overall fluctuation amplitude of the global residual field Resall and quantifies the contribution weight of each error component to the overall prediction bias. Based on the intensity of fluctuations and the magnitude of their contribution weights, the system automatically sorts the sensitivity priorities of the three types of residual components, accurately locating the core error sources that affect the accuracy of mineralization prediction.

[0042] The fifth step is Bayesian posterior approximation inference for discrete scene sets. The system uses a single mineralization 3D configuration (Modn) as its core, binding the three types of residual components (Resfit, Rescov, Resgeo) and physical field response features corresponding to this configuration, encapsulating them as a set of independent discrete geological interpretation scenes (Scase). It iterates through all configurations in the Modset, aggregating all independent scenes to form a discrete scene set (Scall) covering all geological possibilities. Using the multi-solution geological parameters, physical response parameters, and residual feature parameters contained in the scene set (Scall) as prior input, the Bayesian posterior approximation inference operation is initiated. The system iterates through the mineralization matching degree and anomalous response features of all discrete scenes cell by cell, statistically analyzing the frequency and fit probability of mineralization anomalies in each spatial cell under different geological interpretation scenarios. Through probability density fitting iteration, the system obtains the mineralization posterior probability distribution pattern (Probmap) for each cell, simultaneously quantifying the cell's mineralization uncertainty level, completing model self-verification, error source tracing, and uncertainty quantification of the prediction results, achieving dual closed-loop verification of forward and inverse models.

[0043] 4. Implementation process of generating and implementing the target area optimization plan The system extracts parameters from five core constraint dimensions: Probmap (mineralization posterior probability parameter for each spatial unit), Ginfo (potential geological information gain parameter for the target area), Dcon (constraint parameter for field drilling difficulty), Cost (limit parameter for project exploration investment), and Eco (constraint parameter for regional ecological protection and management). These five parameters are dimensionless normalized to eliminate the influence of dimensional differences. Based on multi-objective equilibrium optimization logic, a multi-dimensional optimization function, Optfun, is constructed, taking into account mineralization potential, exploration value, construction feasibility, economic cost, and ecological compliance. Relying on a multi-objective population iterative optimization algorithm, a Popinit population of borehole placement points across the entire domain is initialized. The spatial location, dip angle, and depth of the boreholes are iteratively adjusted generation by generation. After each iteration, the multi-objective performance of each scheme is compared. Invalid placement schemes with simultaneous deterioration of multiple evaluation indicators are automatically screened and eliminated, retaining the non-dominated solution where all objectives are mutually balanced and the overall performance is optimal. After multiple rounds of iterative convergence, a set of differentiated and selectable non-dominated exploration target area schemes, Solbest, is generated to realize the intelligent, engineered, and optimized dynamic deployment of exploration target areas.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A manganese ore exploration system based on multi-source data, characterized in that, include: The data processing module constructs a parameterized lithology-stratigraphy-structure combination constraint domain based on multi-source geological exploration data, performs multi-dimensional data consistency verification and outlier truncation, and outputs three-level data distribution. The mineralization correlation feature module constructs a multi-scale geological element topological correlation network to generate a directed acyclic causal skeleton of the mineralization process; it performs constrained parameter learning and global sensitivity analysis to screen feature subsets that have mechanistic consistency with the manganese mineralization process. The mineralization potential assessment module constructs an initial mineralization favorability spatial grid and a three-dimensional physical property-structure coupled field. Using the mineralization favorability field as a prior constraint, it performs parametric physical property inversion to generate a set of three-dimensional configurations of candidate mineralization bodies with multiple solutions. For each configuration, it conducts physical field decoupling forward simulation, calculates the residual field with the measured multi-source observation field, performs multi-source error decoupling and sensitivity ranking of the residual field, and performs Bayesian posterior approximate inference based on the discrete scene set. The exploration target area optimization module constructs an optimization objective function based on the constraints of the exploration project, and solves it to generate a set of non-dominated exploration target area schemes.

2. The system according to claim 1, characterized in that, The specific process of parameterizing the lithology-stratigraphy-tectonic combination constraint domain includes: The system obtains the regional sedimentary evolution time sequence, the hierarchical matching rules of fault structures, and the specific stratigraphic and lithological combination rules of manganese ore occurrence, generates system-limiting rules, and aggregates them to form the parameterized lithology-stratigraphic-tectonic combination constraint domain. Using a single sampled data point as the traversal unit, the three elements of the sample's stratigraphy, surrounding rock lithology, and surrounding structural development are sequentially matched with the parameterized lithology-stratigraphy-structural combination constraint domain, outputting a three-level data split: a preserved dataset, a suppressed dataset, and a conflict dataset.

3. The system according to claim 1, characterized in that, The specific process of establishing the multi-scale geological element topological association network includes: Based on spatial adjacency and the dependent relationship of material migration and enrichment, a multi-scale topological link is constructed across shallow, middle and deep layers. According to the evolution sequence of manganese ore, the evolution logic of unidirectional driving between various ore-controlling elements is obtained. Network nodes and directional correlation edges are added one by one in the order of mineralization. After the formation is completed, an editing channel for adding nodes and revising correlation relationships is opened, generating a directed acyclic causal skeleton of the mineralization process.

4. The system according to claim 1, characterized in that, The process of generating the feature subset includes: The computational boundary of constrained parameter learning is limited to the element links that have been connected in the causal framework. Within the constrained boundary, the degree of correlation between the changes of a single type of geological element and the mineralization process is calculated step by step. A sensitivity test mode of individual perturbation is adopted, and the parameter state of a type of feature is changed once to obtain the fluctuation range of the mineralization evaluation results of the whole domain. Based on the fluctuation level, the feature contribution is distinguished, and redundant pseudo-features that are correlated at the mathematical and statistical level but lack the support of sedimentary tectonic mineralization control mechanism are eliminated. The remaining elements with genetic support are summarized to generate the feature subset.

5. The system according to claim 1, characterized in that, The specific process of the initial mineralization favorable spatial grid and the three-dimensional property-structure coupling field includes: Based on the spatial ore-controlling mechanism of various ore-controlling elements within the feature subset, a hierarchical ore-forming potential inversion architecture is constructed according to the action sequence and influence range. For various exploration features with discrete point distribution, the point information is extended and filled into the continuous spatial unit of the whole domain. Through iterative calculation, an initial ore-forming favorable spatial grid with continuous planar distribution is generated. The morphology of stratigraphic spatial distribution, the occurrence and distribution of fault structures, and the measured physical property parameters of various geological bodies are integrated simultaneously to construct a three-dimensional physical property-structure coupling field.

6. The system according to claim 1, characterized in that, The process of generating the three-dimensional configuration set of the pleigenic candidate mineralization includes: Based on the stratigraphic distribution and property variation laws stored in the three-dimensional property-structure coupled field, the inversion prior boundary conditions are used to summarize the in-situ measured property data and delineate the property floating range corresponding to different lithological geological bodies. Under the dual constraints of inversion prior boundary conditions and property floating range, parametric property inversion is initiated round by round by alternating different property combination paths. Each set of differentiated parameter combinations corresponds to an independent mineralization spatial morphology. All spatial morphology results are summarized to generate the three-dimensional configuration set of the multi-solution candidate mineralization body.

7. The system according to claim 1, characterized in that, The specific process of the physical field decoupling forward modeling includes: From the set of three-dimensional configurations of the multi-solution candidate mineralization bodies, individual mineralization three-dimensional configurations are retrieved one by one. Based on the surface outcrop boundary, deep stratigraphic interface, and basin basement undulation state of the study area, the global boundary constraints of the forward modeling operation are defined. The physical mechanisms of the generation and conduction of the field sources corresponding to the magnetic field, gravity field, and resistivity field are separated. The numerical solutions of single physical fields are carried out independently in modules. The theoretical response values ​​of the single physical field are calculated in sequence over the entire domain. Then, the calculation results of the three types of physical fields are spliced ​​together under the same spatial grid framework to generate a theoretical observation field that corresponds one-to-one with the spatial location of the measured points.

8. The system according to claim 1, characterized in that, The specific process of residual decoupling and sensitivity ranking includes: Using a unified spatial grid as a benchmark, the numerical differences between the theoretical observation field and the field measured exploration field at the same location are compared unit by unit, and all grid difference data are collected to form a complete residual field for the entire domain. Based on the regional stratigraphic cover conditions, structural simplification degree, and model parameters, three types of geological influencing factors were selected to trace the residual causes. The residuals were then split into three independent error components according to the cause category. The parameters of a single error component were modified at a time, and the fluctuation range of the overall residual field was recorded. All error components were sorted according to the residual fluctuation intensity caused by the component modification.

9. The system according to claim 1, characterized in that, The specific process of the Bayesian approximate inference includes: A single set of candidate mineralization 3D configurations is bound and encapsulated with corresponding sub-item residual data. Each binding is defined as an independent discrete scene. All corresponding scenes in the configuration set are aggregated to form a global discrete scene set. Bayesian approximate inference is carried out using the multi-solution geological information contained in the whole scene as prior material. The response characteristics of all discrete scenes are traversed, and the frequency ratio of mineralization anomalies in each spatial unit under different geological interpretation scenarios is statistically analyzed to obtain the probability distribution law of mineralization events in each spatial unit.

10. The system according to claim 1, characterized in that, The specific process of the non-dominated exploration target area scheme set includes: A multi-objective optimization function is constructed by integrating five constraints: mineralization probability of each spatial unit, potential geological information gain of the target area, environmental limitations of field drilling construction, project exploration investment amount, and regional ecological protection and control. Based on the multi-objective population iterative optimization logic, the borehole layout points are continuously adjusted, and invalid layout schemes with simultaneous deterioration of multiple evaluation indicators are gradually eliminated to generate the non-dominated exploration target area scheme set.

Images