A method, system, device and medium for three-dimensional modeling of a concealed ore body

Abstract

This application provides a method, system, equipment, and medium for three-dimensional modeling of concealed ore bodies, relating to the field of three-dimensional geological modeling technology. The method includes: acquiring four types of multi-source data from the target area—geophysical, geochemical, geological borehole, and geological maps—and generating a standardized data volume after unifying the coordinate system; conducting multi-physics joint inversion based on the standardized data volume to construct a three-dimensional comprehensive physical property model, while simultaneously training a machine learning model using borehole samples to predict and generate a three-dimensional mineralization probability model; using a geological conceptual model as a framework, combining the soft constraints of the two types of models with the hard constraints of the borehole data, generating an initial three-dimensional ore body model through an implicit modeling algorithm; iteratively correcting the model through geophysical forward modeling and error comparison until the fitting degree requirement is met, and finally outputting the target three-dimensional ore body model. This solution effectively improves the reliability and efficiency of concealed ore body modeling.

Description

Technical Field

[0001] This application relates to the field of three-dimensional geological modeling technology, specifically to a three-dimensional modeling method, system, equipment, and medium for concealed ore bodies. Background Technology

[0002] 3D modeling of concealed ore bodies is a crucial technique in mineral exploration. Traditional methods primarily rely on sparse borehole data for spatial interpolation or manual interpretation, making them overly dependent on direct exploration data. While some methods incorporate geophysical data, they fail to achieve deep integration of multiple types of exploration data, typically only performing qualitative comparisons or simple overlay displays. This results in a significant underutilization of valuable indirect information, leading to substantial ambiguity and uncertainty in the constructed 3D ore body models. Furthermore, the modeling methods in these technologies heavily depend on the subjective delineation based on the personal experience of geological experts, which not only reduces modeling efficiency but also limits the objectivity and reliability of the models. Therefore, effectively integrating multi-source heterogeneous data and reducing subjectivity in the modeling process to improve the reliability and efficiency of concealed ore body modeling has become a pressing technical challenge in this field. Summary of the Invention

[0003] To address the aforementioned technical problems, this application provides a three-dimensional modeling method, system, equipment, and medium for concealed ore bodies.

[0004] In the first aspect, this application provides a three-dimensional modeling method for concealed ore bodies, including: acquiring multi-source data of the target area, the multi-source data including geophysical data, geochemical data, geological borehole data and geological map data, and uniformly converting all multi-source data to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume; Based on the standardized data volume, multi-physics joint inversion is performed to generate a three-dimensional comprehensive physical property model characterizing the underground physical structure; and, known mineralized point sample data and known non-mineralized point sample data are obtained from geological borehole data, multi-source data features of the locations of known mineralized point sample data and known non-mineralized point sample data are extracted from the standardized data volume, machine learning models are trained using the extracted multi-source data features, and the trained machine learning models are used to predict the three-dimensional space of the target area to generate a three-dimensional mineralization probability model. Using a geological conceptual model constructed from geological map data as the initial framework, a three-dimensional integrated physical property model and a three-dimensional mineralization probability model are used as soft constraints, and geological borehole data are used as hard constraints. An initial three-dimensional ore body model is automatically generated through an implicit modeling algorithm. Forward modeling is performed on the initial 3D ore body model to obtain its geophysical response. The geophysical response is compared with the measured values ​​of geophysical data. Based on the comparison results, the initial 3D ore body model is corrected until it meets the preset fitting degree requirements. The target 3D ore body model corresponding to the target area is then output. The target 3D ore body model includes a 3D ore body entity model and / or an internal attribute model.

[0005] By adopting the above technical solutions, multi-source data is acquired and uniformly converted into standardized data volumes. This fully utilizes heterogeneous multi-source data, avoiding reliance solely on sparse borehole data or simple combinations with geophysical data, thus reducing data ambiguity and uncertainty. Multi-physics joint inversion generates a three-dimensional comprehensive physical property model, which can characterize subsurface physical structures. Using multi-source data features to train machine learning models to generate three-dimensional mineralization probability models allows for more objective prediction of mineralization probabilities. Using a geological conceptual model as the initial framework, combined with soft and hard constraints, an initial three-dimensional ore body model is automatically generated, reducing subjectivity in the modeling process and improving modeling efficiency. Forward modeling and correction of the initial three-dimensional ore body model until the preset fitting degree requirements are met enhances the reliability and accuracy of the target three-dimensional ore body model.

[0006] Optionally, the above method also includes a dynamic update step, specifically including: Monitor whether there is any new data in the multi-source data. New data includes at least one of the following: new geological borehole data, new geochemical data, and new geophysical data. When new data is detected, the update process is automatically triggered: the new data is used as update information and incorporated into the original standardized data body. The steps of generating a three-dimensional comprehensive physical property model and a three-dimensional mineralization probability model, automatically generating an initial three-dimensional ore body model, and correcting the initial three-dimensional ore body model are repeatedly executed to complete the iterative update of the target three-dimensional ore body model.

[0007] By adopting the above technical solution, the system monitors whether there is any new data from multiple sources. When new data is added, the update process is automatically triggered. The new data is then incorporated into the original standardized data body, and the relevant steps are repeated to complete the iterative update of the target 3D ore body model. This allows for timely use of newly acquired data, making the model more closely match the actual situation, improving the reliability and accuracy of concealed ore body modeling, while reducing manual intervention and improving modeling efficiency.

[0008] Optionally, geophysical data includes gravity anomaly data and magnetic anomaly data collected on the surface of the target area according to a regular grid of measuring points; geochemical data includes the concentration data of metallic or non-metallic elements obtained by actual measurement at geological borehole core sampling points; geological borehole data includes shallow exploration borehole data and deep exploration borehole data; geological map data includes regional geological maps, stratigraphic distribution maps, and fault structure distribution maps; and geological borehole data includes borehole coordinates, borehole depth, sampling depth, inclination data, and grade.

[0009] By adopting the above technical solutions, the specific contents of geophysical data, geochemical data, geological borehole data, and geological map data are clarified, providing detailed basis for obtaining multi-source data of the target area. This helps to uniformly convert multi-source data into standardized data volumes, thereby providing a more comprehensive and detailed data foundation for constructing more accurate three-dimensional ore body models.

[0010] Optionally, all multi-source data can be uniformly transformed to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume, including: The target area is divided into regular three-dimensional grid units according to a uniform size. Various types of original point data are mapped to the corresponding three-dimensional grids through spatial matching. At the same time, the data is normalized to eliminate the differences in the dimensions of different types of data, so that all data maintain a uniform standard in spatial and numerical dimensions, and finally form a standardized three-dimensional data volume covering the entire area.

[0011] By adopting the above technical solution, multi-source data can be uniformly converted to a preset three-dimensional spatial reference coordinate system, eliminating the differences in the dimensions of different types of data, and ensuring that the data maintains a unified standard in both spatial and numerical dimensions, forming a standardized three-dimensional data volume with full coverage. This provides a unified data foundation for subsequent steps such as multi-physics joint inversion and machine learning model training, and helps to improve the reliability and efficiency of concealed ore body modeling.

[0012] Optionally, the multiphysics joint inversion specifically involves the joint inversion of gravity data and magnetic data. The inversion process uses the underground three-dimensional grid cell as the smallest computational unit. The theoretical gravity response is calculated through the density parameter of the underground three-dimensional grid cell, and the theoretical magnetic response is calculated through the magnetic susceptibility parameter. The physical property parameters are solved by combining the measured gravity data and magnetic data of the surface, and finally a three-dimensional comprehensive physical property model containing density distribution and magnetic susceptibility distribution is obtained.

[0013] By adopting the above technical solution, using the underground three-dimensional grid unit as the smallest calculation unit, the theoretical gravity response is calculated using density parameters, the theoretical magnetic response is calculated using magnetic susceptibility parameters, and the physical property parameters are solved by combining the measured gravity and magnetic data of the surface. This can result in a three-dimensional comprehensive physical property model that includes density distribution and magnetic susceptibility distribution, effectively integrating geophysical data and providing more accurate underground physical structure information for subsequent modeling of concealed ore bodies.

[0014] Optionally, multiple regularization constraints are introduced during the joint inversion of gravity data and magnetic data. These constraints include property range constraints, spatial smoothness constraints, and density-magnetic susceptibility coupling constraints. By setting the weights of the spatial smoothness regularization factor, property range regularization factor, and density-magnetic susceptibility coupling constraints respectively, the strength of each constraint is balanced to reduce the ambiguity of the joint inversion. The physical property range constraint is used to represent the pre-defined range of underground rock mass density and magnetic susceptibility values, so as to restrict the density and magnetic susceptibility calculation results of all three-dimensional mesh units to fall within the preset range; the spatial smoothness constraint is used to restrict the density and magnetic susceptibility values ​​of adjacent three-dimensional mesh units from changing abruptly; the density-magnetic susceptibility coupling constraint is used to limit the high density region to correspond to the high magnetic susceptibility region and the low density region to correspond to the low magnetic susceptibility region.

[0015] By adopting the above technical solutions, multiple regularization constraints are introduced in the joint inversion process of gravity data and magnetic data. By setting different regularization factors and weight balance constraints, the ambiguity of joint inversion can be reduced. The property range constraint can ensure that the density and magnetic susceptibility calculation results of all three-dimensional mesh elements fall within the preset range. The spatial smoothness constraint can avoid abrupt changes in the density and magnetic susceptibility values ​​of adjacent three-dimensional mesh elements. The density-magnetic susceptibility coupling constraint can limit the high-density region to correspond to the high magnetic susceptibility region and the low-density region to correspond to the low magnetic susceptibility region.

[0016] Optionally, when training the machine learning model using known mineralized point sample data and known non-mineralized point sample data, the multi-source data features of each sample include the density value, magnetic susceptibility value, and concentration value of at least one ore-forming element at the corresponding sample location. After training, the multi-source data features of each three-dimensional grid cell in the target area are input into the machine learning model to obtain the mineralization probability value corresponding to each three-dimensional grid cell. The mineralization probability values ​​of all three-dimensional grid cells constitute a three-dimensional mineralization probability model.

[0017] By adopting the above technical solution, and using the multi-source data features corresponding to known mineralized and non-mineralized point samples to train the machine learning model, the correlation between multi-source data features and mineralization can be fully explored. By inputting the multi-source data features of the three-dimensional grid cells of the target area into the model, the mineralization probability value of each cell can be obtained to form a three-dimensional mineralization probability model, which provides more accurate mineralization probability information for the modeling of concealed ore bodies, reduces the ambiguity and uncertainty of modeling, and improves the reliability and efficiency of concealed ore body modeling.

[0018] Optionally, the three-dimensional integrated physical property model and the three-dimensional mineralization probability model can be used as soft constraints, including: In the three-dimensional integrated physical property model, the spatial region where the density value is greater than the first preset threshold and the magnetic susceptibility value is greater than the second preset threshold is marked as the ore body candidate area. In the implicit modeling process, each three-dimensional mesh unit in the ore body candidate area is assigned a first gravity weight value. The magnitude of the first gravity weight value is positively correlated with the density value and magnetic susceptibility value of the corresponding three-dimensional mesh unit. The spatial region where the probability value of the three-dimensional mesh cell in the three-dimensional mineralization probability model is greater than the third preset threshold is marked as a high mineralization probability region. In the implicit modeling process, each three-dimensional mesh cell in the high mineralization probability region is assigned a second gravity weight value. The magnitude of the second gravity weight value is positively correlated with the mineralization probability value of the corresponding three-dimensional mesh cell. In the implicit modeling process, the first and second gravity weight values ​​act as spatial attraction, guiding the ore body boundary surface generated by the implicit modeling algorithm to preferentially pass through the spatial region assigned a high gravity weight value.

[0019] By adopting the above technical solution, the three-dimensional comprehensive physical property model and the three-dimensional mineralization probability model are used as soft constraints to mark the ore body candidate area and the high mineralization probability area and assign them gravity weights. This can guide the ore body boundary surface generated by the implicit modeling algorithm to preferentially pass through the area with high gravity weight value, so that the generated initial three-dimensional ore body model is more consistent with the actual ore body distribution and improves the reliability and accuracy of concealed ore body modeling.

[0020] Optionally, forward modeling is performed on the initial 3D ore body model, including: Based on the density and magnetic susceptibility values ​​of each three-dimensional grid cell in the initial three-dimensional ore body model, the forward modeling algorithm is used to calculate the theoretical gravity anomaly and theoretical magnetic anomaly values ​​generated by the initial three-dimensional ore body model at various observation points on the ground or in the air. The theoretical gravity anomaly value is compared with the measured gravity anomaly value point by point to calculate the gravity residual, and the theoretical magnetic anomaly value is compared with the measured magnetic anomaly value point by point to calculate the magnetic residual.

[0021] By adopting the above technical solution, theoretical gravity anomaly values ​​and theoretical magnetic anomaly values ​​at various observation points on the ground or in the air can be obtained for the initial three-dimensional ore body model. The residuals can be calculated by comparing them with the measured values, providing a basis for subsequent correction of the initial three-dimensional ore body model, which helps to improve the accuracy and reliability of the target three-dimensional ore body model.

[0022] Optionally, the initial 3D ore body model can be modified based on the comparison results until it meets the preset fitting requirements, including: When the root mean square error of the gravity residual or magnetic residual is greater than the corresponding preset residual threshold, the constraint strength parameter of the implicit modeling algorithm or the gravity weight of the soft constraint is adjusted, the three-dimensional ore body model is regenerated, and forward modeling and residual comparison are performed again until the root mean square error of the gravity residual and the magnetic residual are both less than or equal to the corresponding preset residual threshold.

[0023] By adopting the above technical solution, the initial three-dimensional ore body model can be corrected based on the comparison between the geophysical response and the measured values. When the root mean square error of the gravity residual or magnetic residual is greater than the corresponding preset residual threshold, the constraint strength parameter of the implicit modeling algorithm or the gravity weight of the soft constraint is adjusted, the model is regenerated and compared again, until the root mean square error of both the gravity residual and the magnetic residual is less than or equal to the corresponding preset residual threshold, thereby improving the fitting degree between the target three-dimensional ore body model and the actual situation, and enhancing the reliability and accuracy of the model.

[0024] Optionally, a geological conceptual model constructed from geological map data serves as the initial framework, including: The geological concept model includes the morphology of the stratigraphic interface within the target area and the three-dimensional surface morphology of multiple ore-controlling faults. The geological framework composed of strata and faults serves as the basic boundary for ore body modeling, limiting the distribution of ore bodies within the spatial range allowed by the geological framework, thereby constraining the overall spatial position and extension morphology of the initial three-dimensional ore body model.

[0025] By adopting the above technical solution, a geological concept model is constructed using the geological framework composed of strata and faults as the basic boundary. This can limit the distribution of ore bodies within the spatial range allowed by the geological framework, thereby constraining the overall spatial location and extension morphology of the initial three-dimensional ore body model.

[0026] Optionally, shallow exploration borehole data in the geological borehole data is used to represent the shallow strata, lithology, and shallow mineralization distribution characteristics of the target area, providing shallow geological constraints for data standardization and the initial framework of the model. Deep exploration borehole data in the geological borehole data is used to represent the deep strata and deep mineralization information of the target area. The lithology, mineralization, and element content information collected by shallow and deep boreholes together constitute the data source of known mineralized point sample data and known non-mineralized point sample data.

[0027] By adopting the above technical solutions, shallow exploration borehole data can provide the shallow strata, lithology and shallow mineralization distribution characteristics of the target area, providing shallow geological constraints for data standardization and the initial framework of the model; deep exploration borehole data can represent the deep strata and deep mineralization information of the target area, and the information collected by shallow and deep boreholes together constitutes the data source of known mineralized and non-mineralized point sample data, thereby providing comprehensive data support for subsequent modeling and improving the reliability and efficiency of concealed ore body modeling.

[0028] Optionally, the geochemical data are discrete point measured data. In the process of generating the standardized data volume, the discretely distributed element concentration measured data are extrapolated globally using a three-dimensional constrained space interpolation algorithm to obtain the element concentration value corresponding to each three-dimensional grid cell, so as to form a continuously distributed three-dimensional element concentration data volume. The three-dimensional element concentration data volume is used as a component of the standardized data volume to participate in subsequent feature extraction and model training.

[0029] By adopting the above technical solution, a three-dimensional constrained spatial interpolation algorithm is used to extrapolate discrete geochemical data across the entire domain, forming a continuously distributed three-dimensional element concentration data volume. This volume serves as a component of the standardized data volume and participates in subsequent feature extraction and model training. This solves the problem of discrete geochemical data, making the data more complete and helping to improve the reliability and efficiency of subsequent modeling.

[0030] In a second aspect of this application, a three-dimensional modeling system for concealed ore bodies is also provided, for performing the three-dimensional modeling method for concealed ore bodies according to any of the preceding claims, comprising: The acquisition unit is used to acquire multi-source data of the target area, including geophysical data, geochemical data, geological borehole data and geological map data, and to uniformly convert all multi-source data to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume. The execution unit is used to perform multi-physics joint inversion based on standardized data volume to generate a three-dimensional comprehensive physical property model characterizing the underground physical property structure; and to obtain known mineralized point sample data and known non-mineralized point sample data from geological borehole data, extract multi-source data features of the locations of known mineralized point sample data and known non-mineralized point sample data from standardized data volume, train a machine learning model using the extracted multi-source data features, and use the trained machine learning model to predict the three-dimensional space of the target area to generate a three-dimensional mineralization probability model. The generation unit is used to automatically generate an initial three-dimensional ore body model using a geological concept model constructed from geological map data as the initial framework, a three-dimensional integrated physical property model and a three-dimensional mineralization probability model as soft constraints, and geological borehole data as hard constraints. The correction unit is used to perform forward modeling on the initial three-dimensional ore body model to obtain the geophysical response of the initial three-dimensional ore body model. The geophysical response is compared with the measured values ​​of geophysical data. Based on the comparison results, the initial three-dimensional ore body model is corrected until the preset fitting degree requirement is met. The target three-dimensional ore body model corresponding to the target area is output. The target three-dimensional ore body model includes a three-dimensional ore body entity model and / or an internal attribute model.

[0031] In a third aspect of this application, an electronic device is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the program to implement the method steps of any of the above claims.

[0032] In a fourth aspect of this application, a computer-readable storage medium is also provided, which stores instructions that, when executed by a processor, perform the method steps described above.

[0033] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: 1. Acquiring multi-source data and uniformly converting it into a standardized data volume can fully utilize heterogeneous multi-source data, avoiding reliance solely on sparse borehole data or simple combinations with geophysical data, thus reducing data ambiguity and uncertainty; performing multi-physics joint inversion to generate a three-dimensional comprehensive physical property model can characterize subsurface physical structures; using multi-source data features to train machine learning models to generate three-dimensional mineralization probability models can more objectively predict mineralization probabilities; using a geological concept model as an initial framework, combined with soft and hard constraints, it automatically generates an initial three-dimensional ore body model, which can reduce subjectivity in the modeling process and improve modeling efficiency; performing forward modeling and correction on the initial three-dimensional ore body model until it meets the preset fitting requirements can improve the reliability and accuracy of the target three-dimensional ore body model; 2. Monitor whether there is any new data from multiple sources. When new data is added, the update process is automatically triggered. The new data is incorporated into the original standardized data body and the relevant steps are repeated to complete the iterative update of the target 3D ore body model. The newly acquired data can be used in a timely manner to make the model more in line with the actual situation, improve the reliability and accuracy of concealed ore body modeling, reduce manual intervention, and improve modeling efficiency. Attached Figure Description

[0034] Figure 1 This is a flowchart of a three-dimensional modeling method for concealed ore bodies provided in an embodiment of this application; Figure 2 This is a schematic diagram of a three-dimensional dynamic modeling process for a concealed ore body provided in an embodiment of this application; Figure 3 This is a structural block diagram of a three-dimensional modeling system for concealed ore bodies provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0035] Explanation of reference numerals in the attached drawings: 400 - Electronic device; 401 - Processor; 402 - Communication bus; 403 - User interface; 404 - Network interface; 405 - Memory. Detailed Implementation

[0036] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0037] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0038] In the description of the embodiments of this application, the term "multiple" means two or more. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0039] This application provides a three-dimensional modeling method for concealed ore bodies, referring to... Figure 1 , Figure 1 This is a flowchart of a three-dimensional modeling method for concealed ore bodies provided in an embodiment of this application, including the following steps: Step S101: Obtain multi-source data of the target area, including geophysical data, geochemical data, geological borehole data and geological map data, and convert all multi-source data into a preset three-dimensional spatial reference coordinate system to generate a standardized data volume; Step S102: Based on the standardized data volume, perform multi-physics joint inversion to generate a three-dimensional comprehensive physical property model characterizing the underground physical property structure; and obtain known mineralized point sample data and known non-mineralized point sample data from geological borehole data, extract multi-source data features of the locations of the known mineralized point sample data and known non-mineralized point sample data from the standardized data volume, train a machine learning model using the extracted multi-source data features, and use the trained machine learning model to predict the three-dimensional space of the target area to generate a three-dimensional mineralization probability model. Step S103: Using the geological concept model constructed from geological map data as the initial framework, the three-dimensional integrated physical property model and the three-dimensional mineralization probability model are used as soft constraints, and the geological borehole data is used as hard constraints. An initial three-dimensional ore body model is automatically generated through an implicit modeling algorithm. Step S104: Perform forward modeling on the initial three-dimensional ore body model to obtain the geophysical response of the initial three-dimensional ore body model. Compare the geophysical response with the measured values ​​of geophysical data. Based on the comparison results, correct the initial three-dimensional ore body model until it meets the preset fitting degree requirements. Output the target three-dimensional ore body model corresponding to the target area. The target three-dimensional ore body model includes a three-dimensional ore body entity model and / or an internal attribute model.

[0040] Through the above steps, multi-source data is acquired and uniformly converted into a standardized data volume. This fully utilizes heterogeneous multi-source data, avoiding reliance solely on sparse borehole data or simple combinations with geophysical data, thus reducing data ambiguity and uncertainty. Multi-physics joint inversion is performed to generate a three-dimensional comprehensive physical property model, which can characterize subsurface physical structures. Using multi-source data features to train a machine learning model to generate a three-dimensional mineralization probability model allows for more objective prediction of mineralization probabilities. Using a geological conceptual model as the initial framework, combined with soft and hard constraints, an initial three-dimensional ore body model is automatically generated, reducing subjectivity in the modeling process and improving modeling efficiency. Forward modeling and correction of the initial three-dimensional ore body model until the preset fitting degree requirements are met enhances the reliability and accuracy of the target three-dimensional ore body model.

[0041] This embodiment proposes an innovative, data-driven, and physically integrated 3D automated modeling method for concealed ore bodies. Its core principle lies in constructing a complete technical process: "standardized input of multi-source data → parallel driving of dual models → multi-constraint fusion modeling → closed-loop verification of physical response." First, by standardizing multi-source data (geological, geophysical, geochemical, etc.) from diverse sources and scales spatially and numerically, the underlying obstacle of data fusion is overcome. Next, this method innovatively constructs two core "soft constraint" models in parallel: one utilizes multi-physics joint inversion technology to extract a "3D comprehensive physical property model" characterizing the differences in the physical properties of underground rocks from geophysical data, such as obtaining the 3D distribution of underground density and magnetic susceptibility; the other uses machine learning algorithms to learn and predict a "3D mineralization probability model" across the entire space from borehole sample data and associated multi-source features—that is, learning mineralization patterns from multi-source data through machine learning and generalizing them to the entire unknown region. In the modeling phase, a macro-geological understanding (geological conceptual model) is used as the basic framework, with borehole data serving as an inviolable "hard constraint" (precise control points). Simultaneously, the aforementioned three-dimensional integrated physical property model and three-dimensional mineralization probability model are introduced as powerful "soft constraints" (trend guidance) to guide the implicit modeling algorithm to automatically and objectively generate orebody morphology. Most importantly, this method establishes a closed-loop verification and correction mechanism based on forward modeling. By comparing the model response with measured data, the model is iteratively optimized until it meets objective fit standards. That is, the geophysical response of the model is verified through forward modeling and compared with measured data, allowing for iterative correction until the accuracy requirements are met. This addresses the technical pain points of related technologies, such as over-reliance on sparse borehole data, lack of multi-source data fusion, strong model subjectivity, high model uncertainty, and lack of objective verification methods. In exploration areas with sparse boreholes and concealed ore bodies, the method in this embodiment can fully utilize indirect data such as geophysical and geochemical exploration, ultimately achieving automation, objectivity, and accuracy in the concealed ore body modeling process, significantly improving the reliability and efficiency of modeling.

[0042] In step S101 above, all raw data are unified into a single "language system." Geophysical data (gravity, magnetic methods) are two-dimensional discrete point data based on surface measurement points; geochemical data are two-dimensional point data based on borehole core sampling points; geological borehole data are three-dimensional point data containing spatial coordinates and attributes (grade, lithology); and geological map data are two-dimensional vector graphic data. These data originally had different coordinate systems, spatial resolutions, and numerical dimensions. This step maps all data to the same three-dimensional grid coordinate system through coordinate transformation and eliminates dimensional differences through normalization. For example, gravity anomalies from mGal, magnetic anomalies from nT, and elemental concentrations from ppm are all mapped to the [0,1] or [-1,1] interval, thereby generating a standardized data volume that can be uniformly processed by a computer. A standardized data volume refers to a full-domain three-dimensional structured data set after coordinate registration, gridding, and dimensional unification.

[0043] In step S102, multi-physics joint inversion refers to a numerical calculation method that combines multiple geophysical observation data to infer the physical properties of underground strata and rock masses. Gravity anomalies are determined by underground density distribution, and magnetic anomalies are determined by underground magnetic susceptibility distribution. This step constructs a unified objective function and simultaneously fits gravity and magnetic observation data to solve for the density and magnetic susceptibility of each three-dimensional grid cell. Mineralized / non-mineralized point samples are machine learning training samples divided by borehole-measured mineralized and non-mineralized locations. Multi-source data features are quantitative indicators such as physical properties, elemental content, and outliers used for model training. The model learns "what kind of multi-source data combination features represent mineralization." After training, this three-dimensional mineralization probability model is applied to each grid cell in the entire three-dimensional space to predict the probability of mineralization at each location. Step S102 relies on standardized data to calculate underground density, magnetic susceptibility, and other physical properties, and simultaneously uses measured mineralization samples to train the algorithm to predict the mineralization probability of each grid cell across the entire domain, supplementing the modeling basis from two dimensions: physical properties and mineralization patterns.

[0044] In step S103, the geological concept model is a regional basic geological framework constructed from strata, faults, and lithology; soft constraints refer to constraints that only provide weight guidance and do not force fixed results; hard constraints refer to measured benchmark conditions that must be strictly followed; the implicit modeling algorithm is an algorithm that automatically generates continuous geological bodies based on spatial interpolation and field value fitting, without the need for manual boundary delineation. The core of this step is to combine the above two types of constraints with the geological framework. The geological concept model defines the possible spatial range of the ore body (hard boundary). Based on this, the three-dimensional comprehensive physical property model tells the modeling engine that "places with high physical property anomalies are more likely to be mineralized," and the three-dimensional mineralization probability model tells the modeling engine that "places with high machine learning prediction probabilities are more likely to be mineralized." These two pieces of information serve as the soft constraint gravitational field; while known mineralization points (or borehole mineralization points) serve as precise control points that must be traversed (hard constraints). The implicit modeling algorithm transforms these constraints into a three-dimensional scalar field and automatically generates the boundary surface of the ore body by extracting the isosurface of the scalar values.

[0045] In step S104, geophysical forward modeling refers to the process of calculating theoretical surface observations based on subsurface properties. The goodness-of-fit requirement represents the error judgment standard between theoretical and measured values. This step is a verification and optimization stage, outputting a data structure for representing the target 3D ore body model corresponding to the target area. The generated 3D ore body entity model is treated as a real geological body. Based on the density and magnetic susceptibility of each grid cell within it, the gravity and magnetic anomalies it should produce on the surface are calculated using physical formulas. This calculated value is then compared point-by-point with the actual geophysical data (such as gravity and magnetic data) measured in step S101, and the root mean square error (residual) is calculated. If the residual exceeds a preset residual threshold (e.g., 0.03 mGal), it indicates that the current model is inconsistent with the observed data, requiring adjustment of modeling parameters (such as the gravity weight of soft constraints, the smoothness of implicit modeling, etc.), regeneration of the model, and further forward modeling calculation and comparison until the residual meets the requirements. The final output target 3D ore body model includes an entity model describing the ore body's shape and / or an attribute model describing the internal grade distribution.

[0046] This embodiment successfully constructed a 3D orebody model even with only sparse boreholes containing and without mineralization, whereas traditional interpolation methods fail due to insufficient data. By combining inversion and machine learning, gravity and magnetic anomalies and geochemical anomalies are incorporated as soft constraints into the modeling process, ensuring that the model not only conforms to the hard data from boreholes but also aligns with geophysical and geochemical observation data from the entire area. The modeling process is automated, requiring no manual delineation. The final output model, after forward modeling verification, shows a root mean square error (RMSE) of less than a preset residual threshold compared to measured gravity and magnetic data, guaranteeing the model's reliability. This embodiment achieves deep fusion of multi-source heterogeneous data; is algorithm-driven throughout, significantly reducing manual intervention; uses dual constraints to reduce model ambiguity; and iterative correction to improve model accuracy, balancing modeling efficiency and reliability.

[0047] In an optional embodiment, the above method further includes: step S105, a dynamic update step, such as... Figure 2 As shown, this specifically includes: monitoring whether there is any new data from multiple sources, including at least one of new geological borehole data, new geochemical data, and new geophysical data; when new data is detected, an update process is automatically triggered: the new data is used as update information and incorporated into the original standardized data body, and the steps of generating a three-dimensional comprehensive physical property model and a three-dimensional mineralization probability model, automatically generating an initial three-dimensional ore body model, and correcting the initial three-dimensional ore body model are repeatedly executed to complete the iterative update of the target three-dimensional ore body model.

[0048] In this embodiment, the system monitors whether there is any new data from multiple sources. When new data is added, the update process is automatically triggered. The new data is incorporated into the original standardized data body, and the relevant steps are repeated to complete the iterative update of the target 3D ore body model. This allows for timely use of newly acquired data, making the model more closely match the actual situation, improving the reliability and accuracy of concealed ore body modeling, while reducing manual intervention and improving modeling efficiency.

[0049] This embodiment further introduces a dynamic update and iterative optimization mechanism, upgrading the static "one-time" modeling process into a "lifecycle" model capable of responding to new data. Its core principle lies in establishing a data monitoring and automatically triggered feedback loop system. This system continuously monitors the project database, and once new exploration data (such as newly added borehole, geophysical, or geochemical sampling data) is detected, it automatically triggers a preset update workflow. This workflow seamlessly integrates the new data into the existing standardized data body, re-executing all steps such as data preprocessing, joint inversion, machine learning, implicit modeling, and forward verification, as described in steps S101 to S104, generating an updated version of the 3D orebody model. In this way, each new data entry becomes the driving force for a new round of model optimization, making the 3D orebody model no longer a static result, but a dynamic model that continuously improves itself and its accuracy as exploration work progresses. This solves the technical problems of static modeling methods in related technologies being unable to adapt to the reality of progressive, multi-stage data acquisition in exploration projects, and the technical problem of the model becoming disconnected from subsequent exploration after delivery. This embodiment endows the target 3D ore body model with the ability to self-evolve, ensuring that the model remains up-to-date and highly accurate throughout the entire exploration cycle, achieving co-evolution between the model and the exploration process. In related modeling methods, once the model is built, it is difficult to update. When new borehole data is obtained, geologists need to manually modify the ore body outlines on multiple profiles and then reconnect them to generate a surface model—a tedious and error-prone process. If the new data contradicts the original model, the adjustment workload is even greater. This embodiment solves this problem through an automated dynamic update mechanism, enabling automated dynamic model updates, adapting to a phased exploration operation mode, reducing repetitive manual labor, improving model update efficiency, and ensuring that the model remains consistent with the latest exploration data.

[0050] In an optional embodiment, the geophysical data includes gravity anomaly data and magnetic anomaly data collected on the surface of the target area according to a regular grid of measuring points; the geochemical data includes the concentration data of metallic or non-metallic elements obtained by actual measurement at geological borehole core sampling points; the geological borehole data includes shallow exploration borehole data and deep exploration borehole data; the geological map data includes regional geological maps, stratigraphic distribution maps, and fault structure distribution maps; and the geological borehole data includes borehole coordinates, borehole depth, sampling depth, inclination data, and grade.

[0051] In this embodiment, the specific contents of geophysical data, geochemical data, geological borehole data, and geological map data are clearly defined, providing detailed basis for obtaining multi-source data of the target area. This helps to uniformly convert multi-source data into standardized data volumes, thereby providing a more comprehensive and detailed data foundation for constructing a more accurate three-dimensional ore body model.

[0052] Geophysical data is limited to gravity anomaly data and magnetic anomaly data collected according to a regular grid of measurement points. This is because the joint gravity and magnetic inversion is one of the core aspects of this embodiment, and regular grid data is convenient for subsequent three-dimensional gridding processing. Regular measuring point grids are grids formed by laying measuring points at fixed intervals in surface geophysical and geochemical exploration, representing a conventional geophysical acquisition method. Geochemical data consists of measured elemental concentration values ​​from borehole cores, ensuring data authenticity and point accuracy. The content (such as borehole coordinates, trajectory, grade, etc.) and type (shallow and deep) of geological borehole data have been refined to ensure its effectiveness as a "hard constraint." Geological borehole data explicitly includes two types of boreholes: shallow and deep. Shallow / deep exploration boreholes are classified according to drilling depth, corresponding to shallow and deep geological information, respectively. For example, shallow indicates a depth less than 500 meters (or other values), and deep indicates a depth greater than or equal to 500 meters. Borehole coordinates and inclination data are essential parameters for borehole positioning and trajectory correction. Grade indicates the content of useful elements in the core, a core indicator for ore body evaluation. Geological map data includes regional geological maps, stratigraphic distribution maps, and fault structure distribution maps, which are the basic data for constructing a geological conceptual model (initial framework). This embodiment standardizes the selection criteria and parameter requirements for various types of exploration data, regulates data sources, ensures the integrity and validity of basic data, and reduces modeling errors caused by data defects from the source.

[0053] In an optional embodiment, all multi-source data are uniformly converted to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume. This includes: dividing the target area into regular three-dimensional grid units according to a uniform size; mapping various types of original point data to the corresponding three-dimensional grids through spatial matching; and normalizing the data to eliminate the differences in the dimensions of different types of data, so that all data maintain a uniform standard in spatial and numerical dimensions, ultimately forming a standardized three-dimensional data volume covering the entire domain.

[0054] In this embodiment, multi-source data can be uniformly converted to a preset three-dimensional spatial reference coordinate system, eliminating the differences in the dimensions of different types of data, and ensuring that the data maintains a unified standard in both spatial and numerical dimensions, forming a standardized three-dimensional data volume with full coverage. This provides a unified data foundation for subsequent steps such as multi-physics joint inversion and machine learning model training, and helps to improve the reliability and efficiency of concealed ore body modeling.

[0055] The three-dimensional space of the target area is discretized into regularly arranged three-dimensional units (voxels), each with fixed spatial coordinates (X, Y, Z). For example, the target area may be a three-dimensional cuboid (Xmin-Xmax, Ymin-Ymax, Zmin-Zmax), which is then cut into many small cubes according to a preset grid size (e.g., 50m × 50m × 50m). Each small cube is called a voxel. The choice of voxel size requires a trade-off: too small a size results in a huge number of grids (potentially exceeding computer memory limits), while too large a size leads to insufficient spatial resolution. It is usually determined based on a combination of exploration accuracy requirements, data density, and computer performance. Then, various types of raw data (originally discrete points, irregularly distributed, with different numerical ranges) are spatially matched and mapped to the corresponding voxels, so that each voxel acquires one or more attribute values ​​(e.g., gravity anomalies, magnetic anomalies, elemental concentrations, etc.). Some of the raw data are precise points (e.g., coordinates of borehole ore-bearing points), while others are discrete measurement points (e.g., gravity measurement points). Spatial matching means that for each voxel, it checks if any original data point falls exactly within that voxel; if so, the attribute value of that data point is assigned to that voxel; otherwise, the attribute value of that voxel needs to be inferred from the surrounding known data points using interpolation methods (such as Kriging interpolation, nearest neighbor interpolation, etc.). Different types of data have different physical units and numerical ranges. For example, gravity anomalies range from -10 to +10 mGal, magnetic anomalies range from -500 to +500 nT, and elemental concentrations range from 0 to 10000 ppm. These numerical ranges differ greatly, and direct use will lead to large numerical features dominating small numerical features in machine learning models. Normalization is a mathematical transformation that maps the original numerical values ​​to a standard interval. Commonly used normalization methods include: maximum-minimum normalization (mapping values ​​to [0,1] or [-1,1]) and Z-score normalization (subtracting the mean and dividing by the standard deviation). Normalization is performed on different types of data to eliminate differences in units. After normalization, all features have similar scales, preventing some features from dominating other features due to their large numerical range in machine learning models. This embodiment unifies heterogeneous data from different sources, formats, units, and spatial distributions into a computable framework; it generates a standardized data volume with full domain coverage and multiple attribute values ​​for each grid point, providing a unified input format for subsequent joint inversion, machine learning, and implicit modeling.

[0056] In an optional embodiment, the multiphysics joint inversion specifically refers to the joint inversion of gravity data and magnetic data. The inversion process uses the underground three-dimensional grid cell as the smallest computational unit. The theoretical gravity response is calculated through the density parameter of the underground three-dimensional grid cell, and the theoretical magnetic response is calculated through the magnetic susceptibility parameter. The physical property parameters are solved by combining the measured gravity data and magnetic data of the surface, and finally a three-dimensional comprehensive physical property model including density distribution and magnetic susceptibility distribution is obtained.

[0057] In this embodiment, the underground three-dimensional grid unit is used as the smallest calculation unit. The theoretical gravity response is calculated using density parameters and the theoretical magnetic response is calculated using magnetic susceptibility parameters. The physical property parameters are solved by combining the measured gravity and magnetic data of the surface. This can obtain a three-dimensional comprehensive physical property model that includes density distribution and magnetic susceptibility distribution. It effectively integrates geophysical data and provides more accurate underground physical structure information for subsequent modeling of concealed ore bodies.

[0058] Gravity anomalies are determined by underground density distribution, while magnetic anomalies are determined by underground magnetic susceptibility distribution. For concealed metallic ore bodies, the ore bodies typically exhibit both high density (due to the presence of heavy minerals such as chalcopyrite and pyrrhotite) and high magnetic susceptibility (due to the presence of magnetic minerals such as pyrrhotite and magnetite). Therefore, density and magnetic susceptibility are spatially correlated (both are high at the ore body and low at the surrounding rock). Joint inversion utilizes this correlation: instead of inverting density and magnetic susceptibility independently, a unified objective function is constructed to simultaneously fit gravity and magnetic observation data, ensuring that the inversion results satisfy both types of observation data and the spatial correlation constraints of density and magnetic susceptibility. In this embodiment, a three-dimensional grid is used as the smallest unit. The theoretical gravity response is calculated using grid density, and the theoretical magnetic response is calculated using magnetic susceptibility. By combining two types of measured surface data, a simultaneous solution is obtained, simultaneously yielding two sets of physical property parameters for the entire area: density and magnetic susceptibility. These are then integrated to form a three-dimensional comprehensive physical property model. Joint gravity and magnetic inversion allows for mutual constraint and complementary verification, improving the accuracy of subsurface property calculations and providing a reliable basis for subsequent orebody modeling, further reducing the ambiguity of orebody models. This addresses the severe ambiguity inherent in single geophysical inversions (using gravity or magnetic methods alone) in related technologies, where multiple different subsurface density distributions can produce the same surface gravity anomaly. This embodiment, by introducing joint constraints from both types of data, significantly narrows the space of possible solutions, resulting in more reliable density and magnetic susceptibility distributions and providing more accurate physical property constraints for subsequent orebody modeling.

[0059] In an optional embodiment, multiple regularization constraints are introduced during the joint inversion of gravity and magnetic data. These constraints include property range constraints, spatial smoothness constraints, and density-magnetic susceptibility coupling constraints. By setting weights for spatial smoothness regularization factors, property range regularization factors, and density-magnetic susceptibility coupling constraints respectively, the effectiveness of each constraint is balanced to reduce the ambiguity of the joint inversion. Property range constraints represent pre-defined ranges for the density and magnetic susceptibility of underground rock masses, ensuring that the density and magnetic susceptibility calculation results of all three-dimensional grid cells fall within these preset ranges. Spatial smoothness constraints prevent abrupt changes in the density and magnetic susceptibility values ​​of adjacent three-dimensional grid cells. Density-magnetic susceptibility coupling constraints define high-density regions as high-magnetic-susceptibility regions and low-density regions as low-magnetic-susceptibility regions.

[0060] In this embodiment, multiple regularization constraints are introduced during the joint inversion of gravity and magnetic data. By setting different regularization factors and weight balance constraints, the ambiguity of the joint inversion can be reduced. The property range constraint ensures that the density and magnetic susceptibility calculation results of all three-dimensional mesh elements fall within a preset range. The spatial smoothness constraint can prevent abrupt changes in the density and magnetic susceptibility values ​​of adjacent three-dimensional mesh elements. The density-magnetic susceptibility coupling constraint can limit high-density regions to high-magnetic susceptibility regions and low-density regions to low-magnetic susceptibility regions.

[0061] In the objective function of the inversion, in addition to requiring the fit of measured data (data terms), several penalty terms (regularization terms) reflecting prior geological knowledge are added. This embodiment introduces three types of regularization constraints and clarifies the parameter setting rules and functions of each constraint. Geophysical joint inversion is an underdetermined problem, and the solution results are not unique. Under unconstrained conditions, problems such as abrupt changes in physical property values ​​and disordered physical property matching relationships may occur, causing the model to deviate significantly from actual geological laws. This scheme sets three types of regularization conditions: physical property range constraints, spatial smoothness constraints, and density-magnetic susceptibility coupling constraints. These respectively limit the range of physical property values, restrict abrupt changes in values ​​between adjacent grids, and constrain the matching relationship between two types of physical properties. The intensity of each constraint is adjusted through dedicated factors and weights. These three types of constraints are combined into a unified objective function by setting their respective weight factors, and solved by an optimization algorithm. By setting different weight factors for each constraint term, the relationship between data fitting and geological rationality can be flexibly balanced. The three types of constraints standardize the inversion results from three dimensions: numerical range, spatial morphology, and physical property relationships, effectively reducing the ambiguity of joint inversion and ensuring that the physical property parameters obtained fully conform to the objective geological and physical laws of the region.

[0062] Joint inversion aims to find a set of densities ρ and magnetic susceptibility κ that minimizes the following objective function, for example, J=J data + λ1·J smooth + λ2·Jcouple + λ3·J bound ; among which, J data J represents the data fitting term (the difference between theoretical and measured values). smooth J is the smoothing regularization term (abrupt changes between adjacent grid cells). couple For the physical property coupling term (high density → high magnetic susceptibility), J bound λ1 is the property range constraint term; λ2 is the property relationship constraint weight (coupling weight); λ3 is the property range regularization factor (a priori geological property range). For example, λ1: 0.001~0.1; λ2: 0.2~1.5; λ3: 0.1~1.0.

[0063] Data fitting term J data : g cal,p Let g represent the theoretical gravity at the p-th measurement point (the sum of contributions from all grids). obs,p t represents the measured gravity at the p-th measuring point; cal,p Let t represent the theoretical magnetic method at the p-th measuring point. obs,p Let M represent the measured magnetic method at the p-th measuring point, where M is the total number of measuring points on the Earth's surface.

[0064] Smooth regularization term J smooth : Where j represents the adjacent grids (up, down, left, right, front, and back) of grid i; Function: The larger the value difference between adjacent grids, the larger J becomes. smooth Punishment mutation.

[0065] Property Coupling Term J couple : a and b represent the density-magnetic susceptibility linear relationship obtained from geostatistics, assuming ρ = 10·k + 2.6; the effect of this term is: deviating from this relationship, J couple To increase size, punish unreasonable combinations.

[0066] Property Range Constraint Term J bound : Where N represents the total number of 3D meshes, e.g., N=120000; i represents the i-th mesh; ρ i ρ represents the current density of the i-th grid; max Indicates the upper limit of density (e.g., 3.5 g / cm³). 3 (or other values); ρ min Indicates the lower limit of density (e.g., 2.6 g / cm³). 3 (or other values); k i k represents the current magnetic susceptibility of the i-th grid; max Indicates the upper limit of magnetic susceptibility (e.g., 0.1 SI, or other values); k min This indicates the lower limit of magnetic susceptibility (e.g., 0SI).

[0067] In an optional embodiment, when training a machine learning model using known mineralized point sample data and known non-mineralized point sample data, the multi-source data features of each sample include the density value, magnetic susceptibility value, and concentration value of at least one ore-forming element at the corresponding sample location. After training, the multi-source data features of each three-dimensional grid cell in the target area are input into the machine learning model to obtain the mineralization probability value corresponding to each three-dimensional grid cell. The mineralization probability values ​​of all three-dimensional grid cells constitute a three-dimensional mineralization probability model.

[0068] In this embodiment, the machine learning model is trained using multi-source data features corresponding to known mineralized and non-mineralized point samples. This fully explores the correlation between multi-source data features and mineralization. By inputting the multi-source data features of the target area's three-dimensional grid cells into the model, the mineralization probability values ​​of each cell can be obtained to form a three-dimensional mineralization probability model. This provides more accurate mineralization probability information for modeling concealed ore bodies, reduces the ambiguity and uncertainty of modeling, and improves the reliability and efficiency of concealed ore body modeling.

[0069] This embodiment treats mineralization as a classification problem, where each 3D grid cell is either "mineralized" (within the ore body) or "non-mineralized" (within the surrounding rock). Known mineralized points (drill hole mineralization locations) provide labels for positive samples, while known non-mineralized points (drill hole non-mineralized locations or non-mineralized drill holes) provide labels for negative samples. The feature vector of each sample consists of multi-source data values ​​for that location, including density, magnetic susceptibility, and the concentration of ore-forming elements (such as Cu, Au, As, etc.). During training, the machine learning model learns the mapping relationship between these features and labels, i.e., identifying the feature combinations of "mineralized" from multi-source data patterns. After training, the model is applied to every 3D grid cell of the entire target area. The multi-source data features of each cell are input, and the model outputs the probability value (between 0 and 1) that the cell is mineralized. This method solves the problem of how to generalize discrete, sparse drill hole mineralization information to the entire 3D space to predict the mineralization potential of non-drill hole areas. Finally, a three-dimensional mineralization probability model covering the entire space is generated, providing spatial probability guidance for implicit modeling. This allows the ore body model to be generated preferentially in areas with high probability, compensating for the lack of sparse borehole data. For example, the grid size of the exploration area is 50m×50m×20m; using 36 groups of deep and shallow exploration boreholes as the data source, 180 known mineralized point samples and 320 known non-mineralized point samples are selected (this is just one example); the machine learning model used is a random forest classification model. For each sample corresponding to the underground three-dimensional grid cell, three sets of features are extracted: ① grid density value (unit: g / cm³). 3 ); ② Mesh magnetic susceptibility value (unit: 10) -6SI); ③ Concentration values ​​of major ore-forming metal elements (unit: ppm); The three sets of values ​​are integrated into a single sample feature vector. 500 sets of samples (180 mineralized samples + 320 non-mineralized samples) and their corresponding labels are input into the random forest model. Parameters such as the number of decision trees and the maximum tree depth are set to complete the training. The model learns the inherent law of "high density, high magnetic susceptibility, and high element concentration areas corresponding to mineralized locations". All three-dimensional grid cells in the exploration area are traversed sequentially. For each grid, three types of features—density, magnetic susceptibility, and ore-forming element concentration—are extracted and input into the trained random forest model. The model outputs a mineralization probability value between 0 and 1 for each grid. According to the spatial coordinates of the three-dimensional grid, the mineralization probability values ​​of all grids are arranged in an orderly manner, finally forming a continuous three-dimensional mineralization probability model covering the entire exploration area.

[0070] Those skilled in the art will understand that the machine learning model described is not limited to the random forest classification model, but can also be other models capable of handling classification or regression problems, such as support vector machines (SVM), gradient boosting decision trees (GBDT), and deep neural networks (DNN). Similarly, the multi-source data features used for training are not limited to density, magnetic susceptibility, and elemental concentration as described herein. Depending on the target mineral type and geological background, they can also include structural features quantified from geological maps (such as distance to a fault zone), indicator variables of lithological units, and other geophysical (such as resistivity) or geochemical data (such as alteration index).

[0071] In an optional embodiment, the three-dimensional integrated physical property model and the three-dimensional mineralization probability model are used as soft constraints, including: marking the spatial regions where the density values ​​of the three-dimensional mesh cells in the three-dimensional integrated physical property model are greater than a first preset threshold and the magnetic susceptibility values ​​are greater than a second preset threshold as ore body candidate regions, and assigning a first gravity weight value to each three-dimensional mesh cell in the ore body candidate region during the implicit modeling process, the magnitude of the first gravity weight value being positively correlated with the density value and magnetic susceptibility value of the corresponding three-dimensional mesh cell; marking the spatial regions where the probability values ​​of the three-dimensional mesh cells in the three-dimensional mineralization probability model are greater than a third preset threshold as high mineralization probability regions, and assigning a second gravity weight value to each three-dimensional mesh cell in the high mineralization probability region during the implicit modeling process, the magnitude of the second gravity weight value being positively correlated with the mineralization probability value of the corresponding three-dimensional mesh cell; the first gravity weight value and the second gravity weight value act as spatial attraction during the implicit modeling process, guiding the ore body boundary surface generated by the implicit modeling algorithm to preferentially pass through the spatial regions assigned high gravity weight values.

[0072] In this embodiment, the three-dimensional comprehensive physical property model and the three-dimensional mineralization probability model are used as soft constraints to mark the ore body candidate area and the high mineralization probability area and assign them gravity weights. This can guide the ore body boundary surface generated by the implicit modeling algorithm to preferentially pass through the area with high gravity weight value, so that the generated initial three-dimensional ore body model is more consistent with the actual ore body distribution and improves the reliability and accuracy of concealed ore body modeling.

[0073] The three-dimensional integrated physical property model and the three-dimensional mineralization probability model are converted into a spatial gravitational field that can be understood by an implicit modeling engine. Specifically, voxels with density values ​​greater than a first preset threshold and magnetic susceptibility values ​​greater than a second preset threshold are selected from the three-dimensional integrated physical property model. These voxels represent high-potential areas with ore body physical property characteristics; for example, the first preset threshold (i.e., the density threshold) is 3.0 g / cm³. 3 (or other values), the second preset threshold (i.e., magnetic susceptibility threshold) is 0.05SI (or other values); voxels with probability values ​​greater than the third preset threshold are selected from the 3D mineralization probability model. These voxels represent high-probability regions predicted by machine learning. For example, the third preset threshold (i.e., probability threshold) is 0.7 (or other values). For these high-potential or high-probability voxels, a higher gravity weight value is assigned during the implicit modeling process, guiding the ore body boundary surface generated by the modeling engine to preferentially pass through these regions. The magnitude of the gravity weight value is positively correlated with the voxel's physical property parameter value or probability value: the higher the physical property anomaly and the greater the mineralization probability of the voxel, the greater the gravity. The method of this embodiment solves the problem of how to quantitatively and programmatically transmit the results of joint inversion and machine learning to the 3D modeling engine, so that the modeling process is no longer a simple data superposition display, but a true data fusion driven process; it realizes the quantitative fusion of multi-source data, so that the generated ore body model conforms to both the spatial distribution of physical property anomalies and the mineralization pattern identified by machine learning. This embodiment sets joint thresholds for density and magnetic susceptibility, as well as a mineralization probability threshold, to divide the orebody candidate area into a high-mineralization probability area. Gravitational weights are assigned to each type of area, with the weights changing positively with the physical properties and probability values. These weights create a spatial attraction, guiding the implicit algorithm to preferentially fit the orebody boundary in the high-weight areas. This scheme transforms abstract soft constraints into quantifiable and executable weight rules, clearly defining the guiding logic of physical properties and mineralization information on orebody morphology. It fully leverages the constraint effects of both types of models, further narrowing the solution range for orebody morphology and improving modeling accuracy.

[0074] In an optional embodiment, forward modeling is performed on the initial three-dimensional ore body model, including: calculating the theoretical gravity anomaly and theoretical magnetic anomaly values ​​generated at various observation points on the ground or in the air based on the density and magnetic susceptibility values ​​of each three-dimensional grid cell in the initial three-dimensional ore body model using a forward modeling algorithm; comparing the theoretical gravity anomaly values ​​with the measured gravity anomaly values ​​point by point to calculate the gravity residual; and comparing the theoretical magnetic anomaly values ​​with the measured magnetic anomaly values ​​point by point to calculate the magnetic residual.

[0075] In this embodiment, theoretical gravity anomaly values ​​and theoretical magnetic anomaly values ​​at various observation points on the ground or in the air can be obtained from the initial three-dimensional ore body model. The residuals can be calculated by comparing them with the measured values, providing a basis for subsequent correction of the initial three-dimensional ore body model and helping to improve the accuracy and reliability of the target three-dimensional ore body model.

[0076] This embodiment uses the density and magnetic susceptibility of each grid within the ore body model as a basis. It calculates the theoretical gravity and magnetic anomaly values ​​at surface / aerial observation points using a forward modeling algorithm, then compares these theoretical values ​​with field measurements point by point, calculating the gravity and magnetic residuals respectively. This scheme establishes a complete quantitative forward modeling and residual calculation system, transforming model deviations into quantifiable error indicators. This provides an objective and accurate basis for subsequent model iteration and correction, ensuring that model correction work is based on evidence. Forward modeling is the reverse process of inversion: given a known physical property model (density and magnetic susceptibility distribution), the gravity and magnetic anomaly fields generated by this model at the surface are calculated using physical formulas. The physical property model originates from the initial three-dimensional ore body model generated by implicit modeling (the ore body part has high density and high magnetic susceptibility, while the surrounding rock part has background density and background magnetic susceptibility). The theoretical gravity and magnetic anomalies obtained from the forward modeling calculation are compared point by point with the actual measured gravity and magnetic anomaly data of the mining area (obtained in the aforementioned S101 step), and the residuals are calculated. The magnitude of the residuals reflects the degree of agreement between the current orebody model and actual observation data: the smaller the residuals, the better the model can explain the observation data; the larger the residuals, the more discrepancy exists between the model and the observation data. This solves the problem of how to quantitatively assess whether the constructed 3D orebody model is consistent with the original geophysical observation data. It achieves the goal of providing quantitative evidence for model correction, ensuring that the model is no longer a product of "gut feeling" but a scientifically validated result.

[0077] In an optional embodiment, the initial three-dimensional ore body model is modified according to the comparison results until a preset fitting degree requirement is met. This includes: when the root mean square error of the gravity residual or magnetic residual is greater than the corresponding preset residual threshold, adjusting the constraint strength parameter of the implicit modeling algorithm or the gravity weight of the soft constraint, regenerating the three-dimensional ore body model, and performing forward modeling and residual comparison again until the root mean square error of both the gravity residual and the magnetic residual is less than or equal to the corresponding preset residual threshold.

[0078] In this embodiment, the initial three-dimensional ore body model can be corrected based on the comparison between the geophysical response and the measured values. When the root mean square error of the gravity residual or magnetic residual is greater than the corresponding preset residual threshold, the constraint strength parameter of the implicit modeling algorithm or the gravity weight of the soft constraint is adjusted, the model is regenerated and compared again, until the root mean square error of both the gravity residual and the magnetic residual is less than or equal to the corresponding preset residual threshold, thereby improving the fitting degree between the target three-dimensional ore body model and the actual situation, and enhancing the reliability and accuracy of the model.

[0079] First, by calculating the overall statistics of gravity and magnetic residuals (such as root mean square error), the fit between the model and the measured data is quantified into an objective evaluation index. When this index fails to meet the preset good standard (i.e., the error exceeds the threshold), the system automatically adjusts the key parameters that affect the morphology of the ore body model, namely the constraint strength parameters or the gravitational weights of soft constraints in the implicit modeling algorithm. For example, if the model response is too strong, the system will reduce the gravitational weights or constraint strength of the soft constraints, causing the scale of the ore body model generated next time to "shrink" somewhat. After adjusting the parameters, the system will re-trigger the entire process of implicit modeling and forward calculation to obtain new residuals. This cycle will repeat continuously until the root mean square error of the residuals is reduced to within an acceptable preset threshold. Specifically, a preset residual threshold is set (e.g., root mean square error of gravity < 0.03 mGal, root mean square error of magnetic force < 5 nT); the root mean square error of the current model is calculated; if it exceeds the threshold, the constraint strength parameters of the implicit modeling algorithm are adjusted (e.g., the smoothness of the ore body boundary, the gravitational weight of soft constraints, etc.), implicit modeling is rerun, a new model is generated, and forward modeling is performed again; this process is repeated until the root mean square error is less than or equal to the preset threshold. This embodiment aims to solve the technical problem of how to efficiently and objectively correct the model to approximate the real geological conditions after discovering discrepancies between the model and the actual measurements. The achieved technical effect is that it replaces the traditional inefficient manual model adjustment that relies on human experience with an automated, quantitative index-based iterative process, ensuring that the final output target model is the best-fitting solution to the measured geophysical data under the premise of satisfying all hard and soft constraints, greatly improving the objectivity and reliability of the model, and realizing the automation of the entire optimization process. Through iterative optimization, the final output target 3D ore body model achieved a good fit with the measured gravity and magnetic data, ensuring the data consistency and reliability of the model.

[0080] In an optional embodiment, a geological concept model constructed from geological map data serves as the initial framework, including: the geological concept model contains the morphology of stratigraphic interfaces and the three-dimensional surface morphology of multiple ore-controlling faults within the target area; the geological framework composed of strata and faults serves as the basic boundary for ore body modeling, limiting the distribution of ore bodies within the spatial range allowed by the geological framework, thereby constraining the overall spatial position and extension morphology of the initial three-dimensional ore body model.

[0081] In this embodiment, a geological concept model is constructed using a geological framework consisting of strata and faults as the basic boundary. This can limit the distribution of ore bodies within the spatial range allowed by the geological framework, thereby constraining the overall spatial location and extension shape of the initial three-dimensional ore body model.

[0082] A three-dimensional surface model, including stratigraphic interfaces and ore-controlling faults, is constructed using geological maps. This geological framework, formed by the combination of strata and faults, serves as a rigid boundary, restricting the distribution of ore bodies only within the space allowed by the framework. This macroscopically constrains the location and extension morphology of the ore bodies. The geological conceptual model is a simplified expression of the geological structure of the target area, including the spatial morphology of stratigraphic interfaces and the three-dimensional surfaces of fault structures. These interfaces and surfaces constitute a three-dimensional framework, defining the possible spatial range of the ore body. For example, if the ore body is known to be controlled by a fault, then the ore body can only be located within a certain zone near the fault surface (e.g., within 200 meters of the hanging wall and footwall of the fault); if the ore body is located in a specific stratigraphic position, then the ore body can only appear between the upper and lower limits of that stratigraphic interface. Using the geological conceptual model as the basic boundary for ore body modeling means that during implicit modeling, the algorithm is restricted to generating ore bodies only within the spatial range allowed by this geological framework. This paper addresses the issue that in purely data-driven modeling, the model's morphology may macroscopically violate basic geological laws. It demonstrates how to quantitatively integrate regional geological laws (strata-controlled ore deposits, tectonic ore deposits) into the 3D modeling process, preventing the generated geological model from violating basic geological common sense. This ensures that the generated orebody model conforms to geological laws, improves the model's geological rationality, and reduces invalid calculations by the algorithm in unreasonable areas (such as locations where ore bodies should not exist).

[0083] In an optional embodiment, shallow exploration borehole data in the geological borehole data is used to represent the shallow strata, lithology, and shallow mineralization distribution characteristics of the target area, providing shallow geological constraints for data standardization and the initial framework of the model. Deep exploration borehole data in the geological borehole data is used to represent the deep strata and deep mineralization information of the target area. The lithology, mineralization, and element content information collected by the shallow and deep boreholes together constitute the data source of known mineralized point sample data and known non-mineralized point sample data.

[0084] In this embodiment, shallow exploration borehole data can provide the shallow strata, lithology, and shallow mineralization distribution characteristics of the target area, providing shallow geological constraints for data standardization and the initial model framework; deep exploration borehole data can represent the deep strata and deep mineralization information of the target area, and the information collected from shallow and deep boreholes together constitutes the data source of known mineralized and non-mineralized point sample data, thereby providing comprehensive data support for subsequent modeling and improving the reliability and efficiency of concealed ore body modeling.

[0085] In mineral exploration, drilling is typically conducted in stages. Shallow boreholes (200-500 meters deep) are primarily used to understand shallow geological structures, stratigraphic distribution, and shallow mineralization characteristics; they are relatively inexpensive and deployed at a high density. Deep boreholes (500-1500 meters deep) are used to probe deep ore bodies; they are expensive and deployed in very few numbers. The two types of boreholes provide different but complementary information: shallow boreholes provide a macroscopic framework of regional geology (strata occurrence, lithological variations, tectonic location), while deep boreholes directly reveal deep mineralization information (ore location, ore grade, elemental composition). Both shallow and deep boreholes together constitute the data source for known mineralized and non-mineralized points; that is, the analysis results from both types of boreholes are used to train machine learning models. By fully utilizing drilling information at different depths, shallow geological understanding can guide the prediction of deep ore bodies. Shallow and deep boreholes together constitute the sample data source, enabling machine learning models to correlate shallow geological features (strata, lithology, geochemical anomalies) with deep mineralization results. The mineralization patterns learned by the model contain a complete information chain from shallow to deep, thus allowing predictions to be made for areas without deep boreholes. If the shallow features of a certain area are similar to those of a known mineralized area, it is inferred that there may be mineralization at deeper levels. This significantly expands the model's predictive capabilities.

[0086] In an optional embodiment, the geochemical data are discrete point measured data. During the generation of the standardized data volume, the discretely distributed element concentration measured data are extrapolated globally using a three-dimensional constrained space interpolation algorithm to obtain the element concentration value corresponding to each three-dimensional grid cell, so as to form a continuously distributed three-dimensional element concentration data volume. The three-dimensional element concentration data volume participates in subsequent feature extraction and model training as a component of the standardized data volume.

[0087] In this embodiment, a three-dimensional constrained spatial interpolation algorithm is used to extrapolate the discrete geochemical data across the entire domain, which can form a continuously distributed three-dimensional element concentration data volume. This volume serves as a component of the standardized data volume and participates in subsequent feature extraction and model training. This solves the problem of discrete geochemical data, making the data more complete and helping to improve the reliability and efficiency of subsequent modeling.

[0088] This embodiment describes how to transform discrete point-based measured data into a continuous three-dimensional elemental concentration data volume. The basic principle is that geochemical data is essentially discrete point-based data; only at borehole core sampling points are measured elemental concentration values; at locations without sampling (most voxels), the concentration is unknown. To obtain the elemental concentration value for each voxel, a spatial interpolation method is needed to extrapolate the concentration value at unknown points based on the concentration values ​​at known points. This embodiment employs a "three-dimensional constrained spatial interpolation algorithm" (such as three-dimensional kriging interpolation), which utilizes the spatial autocorrelation of elemental concentrations within a geological body—samples that are close to each other typically have similar elemental concentrations. The interpolation result is a continuous three-dimensional data volume, with an estimated elemental concentration value for each voxel. The technical problem this method aims to solve is: how to extend discrete, sparse geochemical sampling data to the entire three-dimensional space, generating a continuous three-dimensional concentration data volume that can be used for machine learning feature extraction. This forms a three-dimensional elemental concentration data volume covering the entire space, serving as an important component of the standardized data volume and providing input for subsequent machine learning feature extraction.

[0089] This application provides a method and system for three-dimensional dynamic modeling of concealed ore bodies based on multi-source heterogeneous data fusion. Specifically, it involves integrating multi-source heterogeneous data from geophysical, geochemical, and geological borehole sources for fusion processing, and realizing dynamic updating and visualization of the three-dimensional model of concealed ore bodies. The solution of this application is mainly applicable to scenarios such as deep mineral resource exploration, mine production exploration, dynamic management of mineral resource reserves, mineral prospecting prediction and target area selection, and is particularly suitable for situations with complex geological conditions, limited surface information, and where the morphology and spatial distribution of concealed ore bodies are mainly inferred from indirect exploration data.

[0090] (1) Specific structure / steps: This solution presents a method and system for 3D dynamic modeling of concealed ore bodies based on multi-source heterogeneous data fusion, with its core steps forming a closed-loop workflow. The system includes a data preprocessing module, a multi-source data fusion constraint module, a 3D geological modeling engine, and a model verification and dynamic update module.

[0091] The data preprocessing module includes geophysical data submodule, geochemical data submodule, and geological data submodule. Each submodule is responsible for standardizing, gridding, normalizing, and extracting features from the raw heterogeneous data, transforming data from different sources and scales into a unified three-dimensional spatial reference coordinate system, and generating derived data volumes (such as physical property anomalies and geochemical anomalies) that can be used for fusion.

[0092] The multi-source data fusion constraint module is connected to the data preprocessing module. This module includes a joint inversion algorithm unit and a machine learning data association unit. The joint inversion algorithm unit receives various geophysical data from the data preprocessing module and generates a three-dimensional model that comprehensively reflects the underground physical structure through a joint inversion algorithm that couples multiple physical parameters. The machine learning data association unit is trained using borehole data from known mineralization points and corresponding multi-source derived data volumes to establish a nonlinear mapping relationship model from multi-source data features to mineralization probability or ore body grade.

[0093] The 3D geological modeling engine is connected to the multi-source data fusion constraint module. The engine uses the geological concept model as the initial framework and imports the comprehensive physical property model and mineralization probability model generated by the multi-source data fusion constraint module as soft constraints. It adopts implicit modeling or stochastic simulation algorithms to automatically generate a 3D ore body entity model and its internal attribute (grade) model, under the premise of satisfying all hard data (drill holes) and soft constraints (multi-source fusion information).

[0094] The model validation and dynamic update module is connected to the 3D geological modeling engine. This module includes a fit evaluation unit and a new data triggering unit. The fit evaluation unit performs forward modeling of the initially generated 3D ore body model to calculate its geophysical response and compares it with actual observation data to evaluate the model's data fit. The new data triggering unit monitors newly entered data in the exploration database. When a new batch of valid data (such as new boreholes or geophysical profiles) is added, it automatically triggers the re-run of data preprocessing, fusion constraints, and the modeling engine to iteratively optimize and update the original model.

[0095] (2) Component selection: The key "components" of this solution refer to the algorithm and software modules. The joint inversion algorithm unit can employ a structured regularized joint inversion algorithm based on the Gauss-Newton method or the conjugate gradient method. The machine learning data association unit can utilize random forests, gradient boosting trees, or convolutional neural network models. For example, a fully connected neural network with three hidden layers can be used, where the number of nodes in the input layer corresponds to the dimension of the multi-source feature vector, and the number of nodes in the output layer is 1, representing the mineralization probability. The 3D geological modeling engine can utilize a commercial or open-source geological modeling software kernel that supports implicit modeling and co-kriging interpolation, such as implicit modeling algorithms based on radial basis functions or discrete smooth interpolation.

[0096] (3) Working process and main principles: First, various types of raw exploration data enter the data preprocessing module, and after operations such as coordinate correction, noise reduction, and gridding, a standardized three-dimensional data volume is formed.

[0097] Next, in the multi-source data fusion constraint module, on the one hand, multiple geophysical data are simultaneously inverted through a joint inversion algorithm. The principle is to construct a unified objective function, simultaneously fit multiple geophysical observation data, and introduce empirical or statistical relationships between different physical property parameters as coupling constraints, thereby reducing the ambiguity of the inversion and obtaining a comprehensive physical property model that more closely approximates the actual geological conditions. On the other hand, a machine learning model is trained using multi-source data features extracted from known mineralized borehole locations and their surrounding areas (such as density inversion values, magnetic susceptibility inversion values, elemental concentrations, etc.). This model learns the complex patterns between anomaly combinations of multi-source data and the presence of mineralization.

[0098] Then, the 3D geological modeling engine starts working. It uses the conceptual models such as the structural framework and stratigraphic interfaces provided by the geological engineer as boundary conditions, the high-value areas in the comprehensive physical property model obtained by joint inversion as soft constraints on the spatial location of the ore body, and the full-space mineralization probability volume predicted by the machine learning model as a soft trend of grade distribution. Combined with sparse borehole hard data, it uses mathematical interpolation or simulation methods to generate a 3D ore body model that conforms to geological laws and matches all observation data.

[0099] Finally, the model validation and dynamic update module performs forward modeling calculations on the generated model. For example, it calculates the gravity anomalies generated by the model and compares them with the measured gravity anomalies. If the residuals are large, it provides feedback to adjust the modeling parameters or fuse constraint weights. When new exploration data is input, the system automatically starts a new round of processing, using the new information to fine-tune or reconstruct the model, thus achieving dynamic evolution of the model.

[0100] (4) Operating procedures and precautions: S1: Data Preparation and Import: Collect and organize borehole databases, surface / airborne geophysical survey data (gravity, magnetic, electrical), rock geochemical sampling data, geological maps, and profiles for the target area. Position all data according to a unified geodetic coordinate system and import it into the system database. Note: Ensure coordinate system consistency and perform necessary diurnal and altitude corrections on the original geophysical data.

[0101] S2: Multi-source Data Preprocessing: The data preprocessing module is run to standardize various types of data. This includes combining borehole data into a 3D point set, meshing geophysical data into a 3D regular grid data volume, performing spatial interpolation on geochemical data based on engineering sampling distribution, generating 3D elemental concentration data volumes in areas with sufficient engineering sampling analysis such as boreholes and tunnels, and using constrained interpolation in sparsely sampled areas to reduce modeling difficulty. Important Notes: Mesh parameters (such as grid size) must be set appropriately based on data accuracy and exploration scale; the construction of the geochemical 3D data volume must be adapted to the interpolation method based on the engineering sampling density.

[0102] S3: Constructing a Fusion Constraint Model: Run the multi-source data fusion constraint module. First, select gravity and magnetic data, configure joint inversion parameters (such as regularization factors and property relation constraint weights), perform joint inversion, and generate a comprehensive physical property model of density and magnetic susceptibility. Second, extract training samples from known ore sections, construct and train a mineralization prediction machine learning model, and apply this model to the entire study area to generate a three-dimensional mineralization probability volume. Important Notes: Joint inversion requires a priori geological property range as constraints; machine learning model training should avoid overfitting, and cross-validation should be used to evaluate the model's generalization ability.

[0103] S4: 3D Geological Modeling: This stage focuses on multidisciplinary and multi-method collaborative modeling involving geology, geophysics, geochemistry, and drilling. Within the geological modeling engine, a preliminary 3D stratigraphic and structural framework is established based on regional geological data. Then, the integrated physical property model and mineralization probability volume generated in step S3 are imported as soft data for collaborative modeling, achieving deep fusion of multi-source heterogeneous data. The search radius and constraint strength parameters of the implicit modeling algorithm are set, and combined with borehole hard data, automatic modeling is executed to generate a preliminary 3D orebody model and internal grade model. Important Notes: The accuracy of the geological conceptual framework plays a crucial guiding role in the final model; the weights of the soft constraints need to be adjusted based on their reliability in controlling the orebody.

[0104] S5: Model Validation and Output: Using the model validation module, the preliminary orebody model's forward modeling calculations of gravity, magnetic, and electrical responses are compared with measured multiphysics anomaly data to perform comprehensive multiphysics fitting and correction, and the comprehensive fitting error is calculated. If the error is within an acceptable range, the final 3D model file (such as a mesh model or surface model) and the corresponding uncertainty assessment report are output. If the error is large, return to S4 to adjust the modeling parameters or return to S3 to adjust the fusion constraint weights and remodel. Note: Forward modeling must use the same field sources and calculation parameters as actual observations; multiphysics fitting must consider the response characteristics of different field sources to the target body.

[0105] S6: Dynamic Update: In subsequent exploration work, when new borehole data is obtained or new geophysical measurements are completed, the new data is entered into the system database. After the system's new data trigger unit recognizes the data update, it can automatically or after manual confirmation, starting from step S2 or S3, perform local or global model update calculations to generate a new version of the 3D model. Note: During dynamic updates, the potential systematic errors between new and old data should be considered, and a consistency check is required.

[0106] The following description will be provided in conjunction with specific embodiments.

[0107] First, borehole core data, high-precision ground gravity measurement data, airborne magnetic measurement data, and rock geochemical sampling and analysis data from the study area were collected. Using specialized software, Bouguer correction and topographic correction were applied to the gravity data to obtain Bouguer gravity anomaly data volumes; polarization processing was performed on the magnetic data to obtain magnetic anomaly data volumes; and, combined with the engineering sampling distribution, constrained kriging interpolation was used on the geochemical data to generate three-dimensional concentration distribution data volumes of major ore-forming elements (such as Cu and Au) in densely sampled areas, reducing the difficulty of three-dimensional modeling in sparse areas.

[0108] Secondly, a cross-gradient joint inversion model combining gravity and magnetic methods was constructed to simultaneously fit observed gravity and magnetic anomaly data, inverting to obtain a comprehensive physical property model of subsurface three-dimensional density and magnetic susceptibility distribution. Areas with overlapping high density and high magnetic susceptibility in this model were identified as potential locations of concealed rock masses or mineralized alteration zones. Simultaneously, multi-source data features (including density inversion values, magnetic anomaly values, and Cu concentration values ​​within a 100-meter radius of 50 known mineralized boreholes were selected as positive samples, and the same features from 50 non-mineralized boreholes were selected as negative samples. A random forest classification model was trained using this model to predict the mineralization favorability of the entire three-dimensional space, generating a probability distribution data volume.

[0109] Then, with the collaboration of multiple disciplines including geology, geophysics, geochemistry, and drilling as the core, initial surface models of stratigraphic interfaces and major faults are established in 3D geological modeling software (such as LeapfrogGeo) based on regional geological maps. The density anomalies obtained from the aforementioned joint inversion and the mineralization probability volumes predicted by random forests are imported as "auxiliary data" to achieve multi-source heterogeneous data fusion constraints. During modeling, the modeling algorithm is configured to, in addition to traversing known borehole points in the ore section (hard data), also be subject to spatial attraction (soft constraints) from high-value areas of density anomalies and high-value areas of mineralization probability. By adjusting the constraint strength parameters, a smooth 3D orebody shell model that simultaneously satisfies multiple data constraints is automatically generated. Furthermore, using the inverse power law of distance method, with borehole grade data as hard data and the mineralization probability volume as a soft trend, grade interpolation is performed on the interior of the orebody shell to generate a grade block model.

[0110] Subsequently, the theoretical gravity field and magnetic field of the finally constructed three-dimensional ore body-grade model were forward modeled and comprehensively corrected by multi-physics field fitting with the measured Bouguer gravity anomaly and magnetic anomaly. The results showed that the residual between the theoretical field and the measured field in the ore body area was less than 0.03 mGal, indicating that the model fits the data well.

[0111] Finally, this modeling process is encapsulated into a software system with a graphical user interface. This system has a data monitoring interface. When a user adds inclination measurement and testing data of a borehole to the database, the system prompts that the model can be updated. After the user confirms, the system automatically calls the above process, quickly recalculates and updates only the local model within the influence range of the newly added borehole, generates a revised 3D model version, and retains historical versions for comparative analysis.

[0112] This application also provides a three-dimensional modeling system for concealed ore bodies, used to execute the three-dimensional modeling method for concealed ore bodies in any of the foregoing embodiments, such as... Figure 3 As shown, the system includes: The acquisition unit is used to acquire multi-source data of the target area, including geophysical data, geochemical data, geological borehole data and geological map data, and to uniformly convert all multi-source data to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume. The execution unit is used to perform multi-physics joint inversion based on standardized data volume to generate a three-dimensional comprehensive physical property model characterizing the underground physical property structure; and to obtain known mineralized point sample data and known non-mineralized point sample data from geological borehole data, extract multi-source data features of the locations of known mineralized point sample data and known non-mineralized point sample data from standardized data volume, train a machine learning model using the extracted multi-source data features, and use the trained machine learning model to predict the three-dimensional space of the target area to generate a three-dimensional mineralization probability model. The generation unit is used to automatically generate an initial three-dimensional ore body model using a geological concept model constructed from geological map data as the initial framework, a three-dimensional integrated physical property model and a three-dimensional mineralization probability model as soft constraints, and geological borehole data as hard constraints. The correction unit is used to perform forward modeling on the initial three-dimensional ore body model to obtain the geophysical response of the initial three-dimensional ore body model. The geophysical response is compared with the measured values ​​of geophysical data. Based on the comparison results, the initial three-dimensional ore body model is corrected until the preset fitting degree requirement is met. The target three-dimensional ore body model corresponding to the target area is output. The target three-dimensional ore body model includes a three-dimensional ore body entity model and / or an internal attribute model.

[0113] It should be noted that the devices or systems provided in the above embodiments are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept. Other device or system embodiments correspond to the aforementioned method embodiments. Other technical features are described in the previous embodiments and will not be repeated here.

[0114] This application also provides a computer-readable storage medium storing instructions that, when executed, perform the steps of any of the methods described above.

[0115] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0116] This application also discloses an electronic device. For example... Figure 4 As shown, Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 400 may include: at least one processor 401, at least one network interface 404, a user interface 403, a memory 405, and at least one communication bus 402.

[0117] The communication bus 402 is used to enable communication between these components.

[0118] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0119] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0120] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the electronic device (such as a server) using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 401.

[0121] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory 405 may include a non-transitory computer-readable storage medium. The memory 405 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 405 may also be at least one storage device located remotely from the aforementioned processor 401. (Refer to...) Figure 4 The memory 405, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a three-dimensional modeling method of concealed ore bodies.

[0122] exist Figure 4In the illustrated electronic device 400, the user interface 403 is mainly used to provide an input interface for the user and acquire user input data; while the processor 401 can be used to call an application program of a three-dimensional modeling method for concealed ore bodies stored in the memory 405. When executed by one or more processors 401, the electronic device 400 performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0123] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0124] In the various embodiments provided in this application, it should be understood that the disclosed apparatus or system can be implemented in other ways. For example, the apparatus or system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0125] The above description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the disclosure herein.

[0126] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art that are not described in this disclosure.

Claims

1. A three-dimensional modeling method for concealed ore bodies, characterized in that, include: Acquire multi-source data for the target area, including geophysical data, geochemical data, geological borehole data, and geological map data, and convert all the multi-source data into a preset three-dimensional spatial reference coordinate system to generate a standardized data volume; Based on the standardized data volume, multi-physics joint inversion is performed to generate a three-dimensional comprehensive physical property model characterizing the underground physical structure; and, known mineralized point sample data and known non-mineralized point sample data are obtained from the geological borehole data, multi-source data features of the locations of the known mineralized point sample data and the known non-mineralized point sample data are extracted from the standardized data volume, a machine learning model is trained using the extracted multi-source data features, and the trained machine learning model is used to predict the three-dimensional space of the target area to generate a three-dimensional mineralization probability model. Using the geological concept model constructed from the geological map data as the initial framework, the three-dimensional integrated physical property model and the three-dimensional mineralization probability model are used as soft constraints, and the geological borehole data are used as hard constraints. An initial three-dimensional ore body model is automatically generated through an implicit modeling algorithm. Forward modeling is performed on the initial three-dimensional ore body model to obtain the geophysical response of the initial three-dimensional ore body model. The geophysical response is compared with the measured values ​​of the geophysical data. The initial three-dimensional ore body model is corrected according to the comparison results until the preset fitting degree requirement is met. The target three-dimensional ore body model corresponding to the target area is then output. The target three-dimensional ore body model includes a three-dimensional ore body entity model and / or an internal attribute model.

2. The method according to claim 1, characterized in that, The method also includes a dynamic update step, specifically including: Monitor whether there is any new data in the multi-source data, and the new data includes at least one of new geological borehole data, new geochemical data, and new geophysical data; When new data is detected, the update process is automatically triggered: the new data is used as update information and incorporated into the original standardized data body. The steps of generating a three-dimensional comprehensive physical property model and a three-dimensional mineralization probability model, automatically generating an initial three-dimensional ore body model, and correcting the initial three-dimensional ore body model are repeated to complete the iterative update of the target three-dimensional ore body model.

3. The method according to claim 1, characterized in that, All the multi-source data are uniformly transformed to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume, including: The target area is divided into regular three-dimensional grid units according to a uniform size. Various types of original point data are mapped to the corresponding three-dimensional grids through spatial matching. At the same time, the data is normalized to eliminate the differences in the dimensions of different types of data, so that all data maintain a uniform standard in spatial and numerical dimensions, and finally form a standardized three-dimensional data volume covering the entire region.

4. The method according to claim 1, characterized in that, The multiphysics joint inversion specifically involves the joint inversion of gravity data and magnetic data. The inversion process uses the underground three-dimensional grid cell as the smallest computational unit. The theoretical gravity response is calculated through the density parameter of the underground three-dimensional grid cell, and the theoretical magnetic response is calculated through the magnetic susceptibility parameter. The physical property parameters are solved by combining the measured gravity data and magnetic data of the surface, and finally the three-dimensional comprehensive physical property model containing the density distribution and magnetic susceptibility distribution is obtained.

5. The method according to claim 4, characterized in that, Multiple regularization constraints are introduced during the joint inversion of gravity data and magnetic data. These constraints include property range constraints, spatial smoothness constraints, and density-magnetic susceptibility coupling constraints. By setting weights for spatial smoothness regularization factors, property range regularization factors, and density-magnetic susceptibility coupling constraints respectively, the effectiveness of each constraint is balanced to reduce the ambiguity of the joint inversion. The physical property range constraint is used to represent the pre-defined range of underground rock mass density and magnetic susceptibility values, so as to restrict the density and magnetic susceptibility calculation results of all three-dimensional mesh units to fall within the preset range; the spatial smoothness constraint is used to restrict the density and magnetic susceptibility values ​​of adjacent three-dimensional mesh units from changing abruptly; the density-magnetic susceptibility coupling constraint is used to limit the high density region to correspond to the high magnetic susceptibility region and the low density region to correspond to the low magnetic susceptibility region.

6. The method according to claim 1, characterized in that, When training the machine learning model using the known mineralized point sample data and the known non-mineralized point sample data, the multi-source data features of each sample include the density value, magnetic susceptibility value, and concentration value of at least one ore-forming element at the corresponding sample location. After training, the multi-source data features of each three-dimensional grid cell in the target area are input into the machine learning model to obtain the mineralization probability value corresponding to each three-dimensional grid cell. The mineralization probability values ​​of all three-dimensional grid cells constitute the three-dimensional mineralization probability model.

7. The method according to claim 1, characterized in that, Using the three-dimensional integrated physical property model and the three-dimensional mineralization probability model as soft constraints, including: The spatial region where the density value is greater than the first preset threshold and the magnetic susceptibility value is greater than the second preset threshold in the three-dimensional comprehensive physical property model is marked as the ore body candidate region. In the implicit modeling process, each three-dimensional grid cell in the ore body candidate region is assigned a first gravity weight value. The magnitude of the first gravity weight value is positively correlated with the density value and magnetic susceptibility value of the corresponding three-dimensional grid cell. The spatial region where the probability value of the three-dimensional mesh cell in the three-dimensional mineralization probability model is greater than the third preset threshold is marked as a high mineralization probability region. In the implicit modeling process, a second gravity weight value is assigned to each three-dimensional mesh cell in the high mineralization probability region. The magnitude of the second gravity weight value is positively correlated with the mineralization probability value of the corresponding three-dimensional mesh cell. The first and second gravitational weight values ​​serve as spatial attraction during the implicit modeling process, guiding the ore body boundary surface generated by the implicit modeling algorithm to preferentially pass through the spatial region assigned a high gravitational weight value.

8. A three-dimensional modeling system for concealed ore bodies, characterized in that, A method for performing three-dimensional modeling of concealed ore bodies according to any one of claims 1 to 7, comprising: The acquisition unit is used to acquire multi-source data of the target area, including geophysical data, geochemical data, geological borehole data and geological map data, and to uniformly convert all the multi-source data to a preset three-dimensional spatial reference coordinate system to generate a standardized data volume. An execution unit is configured to perform multi-physics joint inversion based on the standardized data volume to generate a three-dimensional comprehensive physical property model characterizing the underground physical property structure; and to obtain known mineralized point sample data and known non-mineralized point sample data from the geological borehole data, extract multi-source data features of the locations of the known mineralized point sample data and the known non-mineralized point sample data from the standardized data volume, train a machine learning model using the extracted multi-source data features, and use the trained machine learning model to predict the three-dimensional space of the target area to generate a three-dimensional mineralization probability model. The generation unit is used to automatically generate an initial three-dimensional ore body model using the geological concept model constructed from the geological map data as the initial framework, the three-dimensional comprehensive physical property model and the three-dimensional mineralization probability model as soft constraints, and the geological borehole data as hard constraints, through an implicit modeling algorithm. The correction unit is used to perform forward modeling on the initial three-dimensional ore body model to obtain the geophysical response of the initial three-dimensional ore body model, compare the geophysical response with the measured values ​​of the geophysical data, correct the initial three-dimensional ore body model according to the comparison results until the preset fitting degree requirement is met, and output the target three-dimensional ore body model corresponding to the target area, wherein the target three-dimensional ore body model includes a three-dimensional ore body entity model and / or an internal attribute model.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a processor, perform the method as described in any one of claims 1 to 7.

Images