A mineral resource intelligent prediction method based on GNN and PINNs

Abstract

The application discloses a kind of mineral resources intelligent prediction method based on GNN and PINNs, it is related to intelligent geological exploration technical field, including, according to mineralization passage impedance, mineralization element node, buried structure agent node and the mineralization evidence propagation of grid to be predicted is carried out, the impedance propagation mineralization characterization data of grid to be predicted is generated, and impedance propagation mineralization characterization data is input PINNs, and the mineralization physical consistency correction of buried structure agent node, mineralization passage impedance and impedance propagation mineralization characterization data is carried out.The application is graphically associated with the initial mineralization element graph by GNN, identifies the hidden association position not directly expressed between mineralization element node, and generates buried structure agent node, so that buried fault, buried rock mass boundary and deep flow channel are converted into expressible node, to provide graph structure basis for mineralization passage impedance annotation and PINNs physical consistency correction.

Description

Technical Field

[0001] This invention relates to the field of intelligent geological exploration technology, and in particular to an intelligent prediction method for mineral resources based on GNN and PINNs. Background Technology

[0002] Mineral resource prediction has gradually evolved from single geological experience-based judgment to intelligent analysis methods driven by multi-source geoscience data. Geological maps, geophysical anomalies, geochemical elements, remote sensing alteration, known mineral deposits, and topographic information form a unified prediction basis through spatial registration, grid representation, and feature transformation. GNNs are suitable for expressing the spatial relationship between geological objects, while PINNs are suitable for embedding mineralization physical constraints. The combination of these provides a new technical path for mineral prediction under complex geological conditions.

[0003] Existing intelligent prediction of mineral resources mostly relies on the graph structure established by known mineralized objects. The node connections usually come from existing spatial adjacency or attribute similarity relationships, which makes it difficult to express the unobserved associated locations corresponding to concealed faults, concealed rock mass boundaries and deep diversion channels. Due to the lack of effective node representation of concealed structural locations, the propagation of mineralization evidence in the graph is easily limited to visible objects, affecting the physical consistency of mineralization channel identification and target area delineation. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a mineral resource intelligent prediction method based on GNN and PINNs to solve the problem that the location of hidden mineralization associations is difficult to express in multimodal geoscientific maps and participate in physical consistency propagation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an intelligent mineral resource prediction method based on GNNs and PINNs, comprising: acquiring basic data for mineralization prediction, and performing coordinate unification, spatial registration, grid division, and feature extraction to form a multimodal geoscientific feature data of the study area and the grid to be predicted; extracting mineralization-related objects based on the multimodal geoscientific feature data of the study area, setting the mineralization-related objects as mineralization element nodes, establishing node connections according to the spatial correlation of mineralization, and forming an initial mineralization element map; inputting the initial mineralization element map into a GNN, performing graph association propagation based on the node features, node connection relationships, and spatial distribution relationships of the mineralization element nodes, identifying hidden correlation positions between mineralization element nodes that are not directly expressed, generating hidden structural proxy nodes, and incorporating the hidden structural proxy nodes into the initial GNN. The metallogenic element map is used to form an extended metallogenic element map. Based on the extended metallogenic element map, the propagation obstruction analysis of metallogenic evidence is performed on the metallogenic element nodes, concealed structural proxy nodes, and node connection relationships to obtain the metallogenic channel impedance corresponding to each node connection relationship. The metallogenic channel impedance is then marked on the extended metallogenic element map to form an impedance-annotated metallogenic map. The impedance-annotated metallogenic map is input into a GNN, and metallogenic evidence propagation is performed on the metallogenic element nodes, concealed structural proxy nodes, and the grid to be predicted based on the metallogenic channel impedance. This generates impedance propagation metallogenic characterization data for the grid to be predicted. The impedance propagation metallogenic characterization data is then input into PINNs, and metallogenic physical consistency correction is performed on the concealed structural proxy nodes, metallogenic channel impedance, and impedance propagation metallogenic characterization data. Finally, metallogenic potential distribution data and mineral target area delineation data are output.

[0007] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for forming the multimodal geoscientific feature data of the study area, including the grid to be predicted and the geoscientific features, are as follows: Based on the basic data for mineralization prediction, boundary data, geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral occurrence data, and topographic data of the study area were collected. The collected data were then processed by field organization, anomaly record removal, coordinate unification, and spatial registration to obtain spatially benchmarked mineralization data. Based on spatial benchmarked mineralization data, spatial data from different sources are divided into grids, mapped to locations, extracted attributes, transformed features, and reorganized into units to form multimodal geoscientific feature data of the study area, which includes grids to be predicted and geoscientific features.

[0008] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for forming the initial ore-forming element map are as follows: Based on the multimodal geoscientific feature data of the study area, the geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence locations and spatial boundaries corresponding to each grid to be predicted are identified and grouped into similar categories to obtain a set of mineralization-related objects. Based on the set of mineralization-related objects, the spatial location, attribute characteristics and object category of each mineralization-related object are sorted out. The mineralization-related objects are set as mineralization element nodes according to the sorted object categories to obtain mineralization element node data. Using the node data of ore-forming elements, the proximity distribution, structural connections, anomalous overlaps and mineral occurrence correspondences among the nodes of ore-forming elements are correlated and organized to obtain the node connection relationships. The node data of ore-forming elements and the node connection relationships are then reorganized to form an initial ore-forming element map.

[0009] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for obtaining the mineralization element node data are as follows: Based on the set of mineralization-related objects, each mineralization-related object is assigned an object number, and the spatial location, attribute characteristics, and object category of each mineralization-related object are organized accordingly to obtain the organized data of mineralization-related objects. Based on the data of mineralization-related objects, the object number is used as the node identifier, the spatial location is used as the node location, the attribute characteristics are used as the node characteristics, and the object category is used as the node type to generate mineralization element nodes corresponding to each mineralization-related object. Based on the nodes of ore-forming elements, the node identifier, node location, node characteristics and node type are recorded in a unified manner to obtain ore-forming element node data.

[0010] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for forming the extended ore-forming element map are as follows: Based on the initial ore-forming element map, the node characteristics, node connection relationships, and spatial distribution relationships of the ore-forming element nodes are extracted. The connectivity status, adjacent distribution, and association direction between the ore-forming element nodes are also organized to obtain ore-forming element association and propagation data. The initial ore-forming element map and ore-forming element association propagation data are input into the GNN. The graph association propagation is performed on the association paths and spatial continuity states between ore-forming element nodes to obtain the set of hidden association locations that are not directly expressed by the ore-forming association objects. By utilizing the set of concealed associated locations, the location objects corresponding to concealed faults, concealed rock mass boundaries, and deep diversion channels are processed into nodes to generate concealed structural proxy nodes. The concealed structural proxy nodes are then connected and recombined with the ore-forming element nodes in the initial ore-forming element map to form an extended ore-forming element map.

[0011] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for obtaining the set of hidden correlation locations not directly expressed by mineralization correlation objects are as follows: The initial ore-forming element map and ore-forming element association propagation data are input into the GNN to perform association and propagation on the node characteristics, node connection relationships and spatial distribution relationships of the ore-forming element nodes, thereby obtaining the association path data between the ore-forming element nodes. Based on the associated path data, the spatial continuity between the nodes of the mineralization elements is identified and organized, and the spatial associated positions not directly expressed by the mineralization associated objects are extracted to obtain a set of hidden associated positions not directly expressed by the mineralization associated objects.

[0012] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for forming the impedance-labeled ore-forming map are as follows: Based on the extended metallogenic element map, metallogenic element nodes, concealed structure proxy nodes and corresponding node connection relationships are extracted. The node characteristics, spatial position relationships and connection extension states at both ends of each node connection relationship are correlated and organized to obtain node connection propagation characteristics. Based on the propagation characteristics of node connections, the propagation obstruction analysis is performed on the continuity, integrity and obstruction status of mineralization evidence in different node connection relationships, and the propagation obstruction data of mineralization evidence corresponding to each node connection relationship is obtained. By utilizing obstructed data in the propagation of mineralization evidence, a corresponding mineralization channel impedance is generated for each node connection relationship. The mineralization channel impedance is then labeled with the node connection relationships in the extended mineralization element map to form an impedance-labeled mineralization map.

[0013] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for obtaining the obstructed propagation data of mineralization evidence corresponding to the connection relationship of each node are as follows: Based on the node connection propagation characteristics, the mineralization element nodes, connection directions, spatial connection status and mineralization evidence transmission content corresponding to each node connection relationship are organized to obtain the node connection propagation organization set. Based on the node connection propagation set, the continuous transmission, complete preservation, and obstruction / interruption of mineralization evidence in the corresponding node connection relationships are analyzed to obtain the data on obstructed propagation of mineralization evidence for each node connection relationship.

[0014] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for generating impedance propagation mineralization characterization data for the grid to be predicted are as follows: Based on the impedance-labeled mineralization map, the mineralization element nodes, hidden structure proxy nodes, grids to be predicted and corresponding mineralization channel impedances are extracted. The propagation order of the connection relationship of each node is sorted and impedance constraint matching is performed to obtain impedance-constrained mineralization propagation information. Based on impedance-constrained mineralization propagation information, GNN is used to perform obstructed transmission, correlation convergence, and feature reorganization of mineralization evidence among mineralization element nodes, hidden structural proxy nodes, and grids to be predicted, generating impedance propagation mineralization characterization data for grids to be predicted.

[0015] As a preferred embodiment of the intelligent mineral resource prediction method based on GNN and PINNs described in this invention, the specific steps for outputting mineralization potential distribution data and mineral target area delineation data are as follows: Input the impedance propagation mineralization characterization data into PINNs, perform mineralization physical consistency correction, deviation marking and correlation reorganization on the impedance of concealed tectonic proxy nodes, mineralization channel impedance and impedance propagation mineralization characterization data, and obtain a set of physically corrected mineralization characterizations. Based on the physical correction mineralization characterization set, the mineralization potential of the grid to be predicted is sorted, spatial continuity is processed, and target area boundaries are extracted, outputting mineralization potential distribution data and mineral target area delineation data.

[0016] The beneficial effects of this invention are as follows: by performing graph association propagation on the initial ore-forming element map through GNN, the hidden association positions between ore-forming element nodes that are not directly expressed are identified, and hidden structural proxy nodes are generated, so that hidden faults, hidden rock mass boundaries and deep flow channels are transformed into expressible nodes, providing a graph structure basis for ore-forming channel impedance labeling and PINNs physical consistency correction. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a mineral resource intelligent prediction method based on GNN and PINNs.

[0019] Figure 2 This is a flowchart for acquiring and preprocessing mineralization prediction data.

[0020] Figure 3 A flowchart for constructing and expanding a metallogenic element map.

[0021] Figure 4 A flowchart for the propagation of mineralization evidence and target area delineation. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for intelligent prediction of mineral resources based on GNNs and PINNs, including the following steps: S1: Obtain basic data for mineralization prediction, and perform coordinate unification, spatial registration, grid division and feature extraction to form multimodal geoscientific feature data of the study area for the grid to be predicted and geoscientific features.

[0026] S1.1: Based on the basic data for mineralization prediction, collect boundary data, geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral occurrence data, and topographic data of the study area. Then, perform field sorting, anomaly record removal, coordinate unification, and spatial registration on the collected data to obtain spatially benchmarked mineralization data.

[0027] Furthermore, based on the basic data for mineralization prediction, the data processing scope is determined by the boundary data of the study area. Geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral occurrence data, and topographic data are collected in sequence, and the data source, spatial range, coordinate reference, data type, and attribute fields are recorded respectively.

[0028] When processing the collected data, the stratigraphy, lithology, and structural fields in the geological map data are organized into categorical geological features; the anomaly types and intensities in the geophysical anomaly data are organized into geophysical category features and numerical anomaly features; the element names and element contents in the geochemical element data are organized into element category features and numerical content features; the alteration types and alteration extents in the remote sensing alteration data are organized into remote sensing category features and spatial coverage features; the location and mineral type in the known mineral occurrence data are organized into point location features and mineral type category features; and the elevation and slope in the topographic data are organized into numerical topographic features. Categorical fields are converted into discrete input features using category numbers, and numerical fields are converted into dimensionless input features through dimension unification and effective range normalization. Point, line, and area spatial objects are converted into location markers, intersection lengths, coverage ratios, and object categories within the corresponding grid, respectively. Raster data is resampled according to the grid range to be predicted, and effective pixel representative values ​​and coverage ratios are extracted to form model input features that can be read by GNNs and PINNs.

[0029] After the data is processed, abnormal records are removed, including data items with missing spatial locations, missing attribute fields, coordinates exceeding the study area boundary, invalid geometric shapes, and duplicate records. Data items that spatially correspond to the study area boundary data are retained. Using the coordinate benchmark of the study area boundary data as a unified coordinate benchmark, the retained data items are transformed. The transformed geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral deposit data, and topographic data are spatially overlaid and verified with the study area boundary data to ensure that all types of data maintain positional correspondence within the same spatial range. The spatially overlaid and verified data are then spatially registered. Vector boundaries, points, linear structures, areal anomalies, and raster cells are clipped, aligned, and corrected according to the study area boundary data to obtain spatially consistent, attribute-field-complete, and usable spatial benchmark mineralization data for grid division and feature extraction.

[0030] It should be noted that the basic data for mineralization prediction is a set of data that determines the spatial range, metallogenic geological background, anomaly distribution, and known mineralization locations of the study area. It includes study area boundary data, geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral occurrence data, and topographic data. The study area boundary data defines the scope of coordinate unification, spatial registration, and grid division. The geological map data provides stratigraphic, lithological, and tectonic background information. The geophysical anomaly data is used to provide information on deep physical property anomalies. The geochemical element data is used to provide information on element enrichment. The remote sensing alteration data provides information on surface alteration distribution. The known mineral occurrence data is used to provide reference for mineralization locations. The topographic data provides spatial undulation and geomorphic constraints.

[0031] S1.2: Based on spatial benchmarked mineralization data, grid division, location mapping, attribute extraction, feature transformation and unit reorganization are performed on spatial data from different sources to form multimodal geoscientific feature data of the study area for the grid to be predicted and geoscientific features.

[0032] Furthermore, based on the spatially benchmarked mineralization data, the spatial range corresponding to the boundary data of the study area is used as the grid division range. The area covered by the boundary data of the study area is uniformly divided into grids, and each grid unit is configured with a grid number, grid center position and grid boundary range to obtain the grid to be predicted.

[0033] Based on the grid to be predicted, the geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral deposit data, and topographic data are mapped to their respective locations. Spatial data falling within the same grid boundary range are assigned to the corresponding grid to be predicted to obtain grid mapping data. Furthermore, the geological attributes, geophysical anomaly attributes, geochemical element attributes, remote sensing alteration attributes, mineral deposit location attributes, and topographic attributes within each grid to be predicted are extracted to obtain grid attribute data.

[0034] By utilizing grid attribute data, textual attributes are categorized, numerical attributes are uniformly converted, and spatial location attributes are normalized within the grid to obtain grid geoscientific feature data. Based on the grid to be predicted, grid mapping data, and grid geoscientific feature data, the geoscientific features corresponding to each grid to be predicted are reorganized into units to form multimodal geoscientific feature data of the study area for the grid to be predicted and geoscientific features.

[0035] It should be noted that when uniformly dividing the data coverage area of ​​the study area boundary into grids, the granularity of grid division is determined based on the prediction scale of the study area, the spatial data resolution, and the spatial distribution of known mineral deposits, so that each grid to be predicted can carry geological, geophysical, geochemical, remote sensing, mineral deposit, and topographic features.

[0036] Point data is written with point category, attribute value and relative position according to the grid it falls into; linear data is written with line segment length, extension direction and object category according to the intersection relationship with the grid; areal data is written with dominant category and coverage area according to the overlap ratio with the grid; raster data is resampled according to the grid range to be predicted and written with effective cell representative value and coverage ratio.

[0037] For records of the same type within the same grid, they are merged according to object category and attribute consistency. For records from different sources, they are retained as corresponding geoscientific features. Geophysical, geochemical, and remote sensing data are normalized to their effective range according to the same data layer, the same element combination, and the same band index, forming multimodal geoscientific feature data of the study area with a unified spatial location basis and a unified dimensionless expression.

[0038] S2: Based on the multimodal geoscientific feature data of the study area, extract the mineralization-related objects and set them as mineralization element nodes. Establish node connections according to the spatial correlation of mineralization to form an initial mineralization element map.

[0039] S2.1: Based on the multimodal geoscientific feature data of the study area, the geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence locations and spatial boundaries corresponding to each grid to be predicted are identified and grouped into similar categories to obtain a set of mineralization-related objects.

[0040] Furthermore, the geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral deposit locations, and spatial boundaries are read grid by grid cell according to the grid to be predicted. The adjacent grid cells sharing edges are used as the basis for determining spatial connectivity, and the connectivity of 8 neighborhoods is used as the basis for determining spatial continuity. Grids with the same category in geological attributes and sharing edges with each other are organized into geological attribute objects. Grids with the same anomaly type in geophysical anomalies and continuous connectivity within 8 neighborhoods are organized into geophysical anomaly objects. Grids with the same element combination in geochemical anomalies and continuous connectivity within 8 neighborhoods are organized into geochemical anomaly objects. Grids with the same alteration type in remote sensing alteration and continuous connectivity within 8 neighborhoods are organized into remote sensing alteration objects. The mineral deposit locations are mapped to the corresponding grid cells to be predicted according to the grid cells where the locations are located and organized into mineral deposit objects. The boundary lines, boundary surfaces, and grids that intersect, overlap, or share edges with the boundary lines or boundary surfaces corresponding to the spatial boundaries are organized into spatial boundary objects.

[0041] For geological attribute objects, geophysical anomaly objects, geochemical anomaly objects, remote sensing alteration objects, mineral occurrence objects, and spatial boundary objects, record the object location, object range, object category, and object attributes respectively. Objects with the same object category, consistent object attributes, and contiguous object ranges are grouped together, while objects with different object categories or non-contiguous object ranges are kept as independent records, thus obtaining a set of mineralization-related objects.

[0042] Specifically, the object category, object attributes, and object range of each mineralization-related object are read. Objects with the same object category and whose attribute similarity, spatial overlap ratio, or spatial spacing meets the merging conditions are grouped into the same record. The object category is used to distinguish geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence location, and spatial boundaries. The object attributes are used to calculate attribute similarity. Among them, the similarity criterion for categorical attributes is that the category number is consistent, and the similarity criterion for numerical attributes is that the normalized attribute difference is within the allowable error range of the same data source.

[0043] The object range is used to determine spatial relationships. The overlap ratio is determined by the ratio of the overlapping area of ​​two objects to the area of ​​the smaller object. When the merging ratio requirement is met, it is determined to be an overlap. When the shortest distance between the boundaries of objects is not greater than the side length of the grid to be predicted and the extension direction is consistent, it is determined to be a continuous spatial extension. When the object ranges have common edges, common points, or grid 8-neighborhoods, they are determined to be spatially connected.

[0044] For example, if two adjacent grids both belong to geochemical anomalies and have the same combination of corresponding element anomalies, and the two anomaly ranges are spatially connected, then the two geochemical anomalies will be merged into the same mineralization associated object. If one object is a geochemical anomaly and the other is a remote sensing alteration object, even if the spatial locations overlap, they will be retained as independent records because the object categories are different. If both objects are geophysical anomalies, but there is a spatial gap between the anomaly ranges, then they will not be merged and will be retained as independent records.

[0045] S2.2: Based on the set of mineralization-related objects, assign an object number to each mineralization-related object, and organize the spatial location, attribute characteristics and object category of each mineralization-related object accordingly to obtain the organized data of mineralization-related objects.

[0046] Furthermore, the object category, spatial location, and attribute characteristics of each mineralization-related object are read, the object category is converted into a category identifier, the spatial location is converted into a location identifier of the grid to be predicted, and a sequence identifier is generated according to the spatial arrangement order within the same object category. The category identifier, location identifier, and sequence identifier are combined to form the object number of each mineralization-related object.

[0047] Based on the object number, the spatial location of each mineralization-related object is organized to obtain object location records containing the grid to be predicted, the object's center location, the object's coverage area, and the relationship with adjacent objects. Based on the object number, the attribute characteristics of each mineralization-related object are organized to obtain object feature records containing geological attributes, geophysical anomaly attributes, geochemical anomaly attributes, remote sensing alteration attributes, mineral occurrence attributes, or spatial boundary attributes. Based on the object number, the object category, object location record, and object feature record are associated one by one to form an organized record corresponding to each mineralization-related object. Organized records with duplicate object numbers and identical object location and object feature records are merged to obtain organized data of mineralization-related objects.

[0048] S2.3: Organize the data based on the mineralization-related objects, using the object number as the node identifier, the spatial location as the node location, the attribute characteristics as the node characteristics, and the object category as the node type to generate mineralization element nodes corresponding to each mineralization-related object.

[0049] Furthermore, based on the data organized by the mineralization-related objects, the object number, spatial location, attribute characteristics and object category of each mineralization-related object are read one by one. The object number is written into the node identifier field so that each mineralization-related object has a unique corresponding node record. Then, the spatial location is converted into the grid number to be predicted, the grid center location and the object coverage area, and written into the node location field.

[0050] Following the order of geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence location, and spatial boundaries in the multimodal geoscientific feature data of the study area, the attribute features are arranged, completed, and formatted to form node feature fields. The object category is written into the node type field to distinguish the node attributes of different mineralization-related objects in the initial mineralization feature map.

[0051] By binding the node identifier field, node location field, node feature field, and node type field to generate ore-forming element nodes corresponding to each ore-forming associated object, and by sequentially recording all ore-forming element nodes, ore-forming element node data is obtained.

[0052] S2.4: Based on the ore-forming element nodes, the node identifier, node location, node characteristics and node type are uniformly recorded to obtain ore-forming element node data.

[0053] Furthermore, each mineralization element node is read one by one according to the object number in the data of mineralization-related objects. The object number is written into the node identifier record, the spatial location is written into the node location record, the attribute characteristics are written into the node characteristic record, and the object category is written into the node type record. Then, the field names of the node identifier record, node location record, node characteristic record, and node type record are unified, the record order is unified, and the missing content is checked.

[0054] For ore-forming element nodes with duplicate node identifiers and identical node locations, retain one node record and merge node features. For ore-forming element nodes with different node identifiers but identical node locations, retain node records according to node type. After verification, organize the node identifier records, node location records, node feature records, and node type records in the order of object number to obtain ore-forming element node data.

[0055] S2.5: Using the ore-forming element node data, the proximity distribution, structural connections, abnormal overlaps and mineral occurrence correspondences between ore-forming element nodes are correlated and organized to obtain node connection relationships. The ore-forming element node data and node connection relationships are then reorganized to form an initial ore-forming element map.

[0056] Furthermore, the node location, node characteristics, and node type of each ore-forming element node are read according to the node identifier. The ore-forming element nodes whose node locations are located in adjacent grids to be predicted or have boundary contact relationships are matched for proximity distribution to form a proximity connection record.

[0057] Based on the proximity connection record, structural connection matching is performed on ore-forming element nodes whose node types correspond to structural objects and whose node positions are in the same structural extension range, structural intersection position, or structural contact position to form a structural connection record. Based on the structural connection record, anomaly overlap matching is performed on ore-forming element nodes whose node characteristics have corresponding content of geophysical anomaly, geochemical anomaly, or remote sensing alteration and whose spatial ranges overlap, cover, or belong to the same grid to be predicted to form anomaly connection record.

[0058] Based on the abnormal connection records, mineralized element nodes whose node types correspond to known mineral points and whose node positions have a spatial correspondence with mineralized element nodes are matched to form mineral point connection records. Based on the neighbor connection records, constructed connection records, abnormal connection records, and mineral point connection records, duplicate connections and connections with the same direction are merged and sorted to obtain node connection relationships. The node connection relationships and mineralized element node data are then recombined according to the node identifiers to form an initial mineralized element map.

[0059] It should be noted that the spatial correlation of mineralization refers to the spatial correspondence between mineralization element nodes that can form node connection relationships, including proximity distribution, structural connection, anomaly overlap, and mineral occurrence correspondence. Proximity distribution represents the adjacent or contact state of mineralization element nodes in the grid to be predicted. Structural connection represents the connection of mineralization element nodes at structural extension, structural intersection, or structural contact positions. Anomaly overlap represents the intersection, coverage, or belonging to the same grid to be predicted in the spatial range of geophysical anomalies, geochemical anomalies, and remote sensing alteration. Mineral occurrence correspondence represents the positional correspondence between known mineral occurrences and mineralization element nodes.

[0060] S3: Input the initial ore-forming element map into the GNN, perform graph association propagation based on the node characteristics, node connection relationships and spatial distribution relationships of the ore-forming element nodes, identify the hidden association positions between ore-forming element nodes that are not directly expressed, generate hidden structural proxy nodes, and incorporate the hidden structural proxy nodes into the initial ore-forming element map to form an extended ore-forming element map.

[0061] S3.1: Based on the initial ore-forming element map, extract the node characteristics, node connection relationships and spatial distribution relationships of the ore-forming element nodes, and organize the connectivity status, adjacent distribution and association direction between the ore-forming element nodes to obtain ore-forming element association propagation data.

[0062] Furthermore, based on the initial ore-forming element map, taking the ore-forming element nodes and node connection relationships as the processing objects, the node identifier, node position, node characteristics and node type corresponding to each ore-forming element node are read, and the node position, node characteristics and node type are organized according to the node identifier to form a basic node record.

[0063] Read the starting node identifier, ending node identifier, connection category, and connection direction corresponding to the connection relationship of each node in the initial ore-forming element map, and match the starting node identifier and ending node identifier with the node basic record to form a node connection record. Based on the node position and node connection record, organize the spatial distribution relationship between ore-forming element nodes. Adjacent grids refer to the grids where two nodes are located have a shared edge relationship. Continuous distribution area refers to the grids where the same type of object can form a continuous connected chain through 8 neighbors, and the center distance between adjacent grids is not greater than the square root of two times the grid side length. Same spatial extension direction refers to the angle between the extension directions of two objects being 0°. Similar spatial extension direction refers to the angle between the extension directions being not greater than 30°.

[0064] Based on the judgment results, the connection relationship records for each node are recorded with direct connection markers, adjacent grid markers, continuous distribution markers, and directionally consistent markers, forming the node spatial distribution record of the GNN input.

[0065] Next, the node basic records, node connection records, and node spatial distribution records are merged according to the same ore-forming element node identifier. For each pair of ore-forming element nodes with node connection relationship, the connectivity status, adjacent distribution, and association direction are marked. The marked ore-forming element nodes, node connection relationships, connectivity status, adjacent distribution, and association direction are uniformly organized to obtain ore-forming element association propagation data for input into GNN for graph association propagation.

[0066] It should be noted that whether there is a direct connection between the two nodes is determined by whether the starting node identifier and the ending node identifier can match the same node connection record. When the grids of the two nodes have a shared edge relationship, they are determined to be adjacent grids. When the grids of the two nodes only have a point contact, they are recorded as boundary contact grids.

[0067] When determining a continuous distribution region, read the grid connection relationship between the object ranges of two nodes. If the two object ranges can form a continuous connected chain through 8 neighboring grids, and the distance between the centers of adjacent connected grids is no greater than the square root of two times the grid side length, and the shortest distance between the endpoints of the two object ranges is no greater than the side length of the grid to be predicted, then it is determined to be a continuous distribution region.

[0068] When determining the spatial extension direction, calculate the angle between the node connection direction and the main extension direction of the object range. If the angle is not greater than 10°, it is determined that the directions are consistent. If the angle is greater than 10° but not greater than 30°, it is determined that the directions are similar. If the angle is greater than 30°, it is not determined that the spatial extension directions are the same or similar.

[0069] For example, in a certain study area, fault structure node A is located in grid 10, and geochemical anomaly node B is located in grid 11. In the initial mineralization element map, there are node connection records pointing from fault structure node A to geochemical anomaly node B, and grids 10 and 11 are adjacent. At the same time, the geochemical anomaly range is continuously distributed along the fault structure direction without being interrupted by irrelevant areas. Therefore, it can be determined that fault structure node A and geochemical anomaly node B are directly connected, located in adjacent grids, in a continuous distribution area, and have similar spatial extension directions. Conversely, if geochemical anomaly node B is spatially close to fault structure node A, but there are no node connection records between them, and the extension direction of the geochemical anomaly range deviates significantly from the fault structure direction, then it is not determined to be a continuous mineralization spatial distribution relationship.

[0070] The specific training process of GNN is as follows: training samples are composed of historical exploration areas, grids containing known mineral deposits, grids of non-target areas confirmed by exploration, and their neighboring metallogenic element nodes. Multimodal geoscientific feature data, metallogenic element node data, node connection relationships, node spatial distribution records, spatial connection markers, directional continuation markers, and node feature continuation markers of the study area are read as training inputs. The corresponding relationships of known mineral deposits, structural extension relationships in existing geological interpretation maps, confirmed metallogenic association paths, hidden structural interpretation locations, and metallogenic evidence propagation results are used as training labels.

[0071] During training, the GNN transmits node features according to node connection relationships, outputs mineralization association paths, candidate hidden association locations, and node propagation features, and compares the output results with the training labels. Based on the difference results, the GNN parameters are updated, enabling the GNN to learn the correspondence between geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence locations, and spatial connection relationships.

[0072] After training, the GNN is used to identify hidden correlation locations that are not directly expressed in the current study area. After the impedance-labeled mineralization map is formed, the impedance values ​​of the mineralization channels are read, impedance constraint propagation is performed on the mineralization evidence, and impedance propagation mineralization characterization data of the grid to be predicted is generated.

[0073] S3.2: Input the initial ore-forming element map and ore-forming element association propagation data into the GNN, and perform association propagation on the node characteristics, node connection relationships and spatial distribution relationships of the ore-forming element nodes to obtain the association path data between the ore-forming element nodes.

[0074] Furthermore, based on the node identifier, node location, node characteristics, and node type in the ore-forming element node data, each ore-forming element node in the initial ore-forming element map is input and organized. The node connection relationship, connectivity status, adjacent distribution, and association direction are then bound to the corresponding ore-forming element node to obtain graph input data that can be read by the GNN.

[0075] Based on the graph input data, the GNN propagates the node features of adjacent ore-forming element nodes layer by layer along the node connection relationship. It synchronously records the node source, connection order, spatial connection status, and association direction formed during the propagation process to obtain the propagation process record between ore-forming element nodes. According to the propagation process record, the connection content that is repeated between the same ore-forming element nodes is merged, the connection content with interrupted spatial connection is removed, and the connection content with continuous transmission of node features and consistent association direction is retained. The path is then organized according to the node source, the nodes passed through, the connection order, and the terminating node to obtain the association path data between ore-forming element nodes.

[0076] S3.3: Based on the associated path data, identify and organize the spatial continuity between the nodes of the ore-forming elements, extract the spatial associated locations that are not directly expressed by the ore-forming associated objects, and obtain the set of hidden associated locations that are not directly expressed by the ore-forming associated objects.

[0077] Furthermore, based on the associated path data, the connection relationships between the starting point ore-forming element node, the ending point ore-forming element node, and the nodes traversed by the path for each associated path are numbered. The node positions of the starting point ore-forming element node, the ending point ore-forming element node, the node positions of the nodes traversed by the path, and the node features of the nodes traversed by the path are extracted according to the path number to obtain the path position data.

[0078] Based on path location data, ore-forming element nodes along the same associated path are spatially sorted according to the path propagation order and grid center coordinates. The spatial connectivity, directional continuity, and node feature continuity between adjacent ore-forming element nodes are then compared sequentially to obtain the spatial continuity status between them. Spatial connectivity is calculated based on the shortest distance between the endpoints of the object ranges and grid connectivity. When two object ranges can form a continuous connected chain through 8 neighboring grids, and the shortest distance between their endpoints is no greater than the side length of a grid to be predicted, the spatial connectivity is marked as 1; otherwise, it is marked as 0.

[0079] Directional continuation is calculated based on the angle between the main extension directions of two objects. The main extension direction is determined by the endpoint direction of the linear object, the major axis direction of the smallest bounding rectangle of the planar object, or the direction of the center line of the continuous grid. When the angle is not greater than 30°, the directional continuation is marked as 1; otherwise, it is marked as 0. Node feature continuation is calculated based on the similarity of node features. When the categorical features are consistent, the intersection ratio of element combinations is not less than 0.6, and the normalization difference of numerical features is not greater than 0.2, the node feature continuation is marked as 1; otherwise, it is marked as 0.

[0080] Based on the markings, the location segments that simultaneously satisfy spatial connection, directional continuity, and node feature continuity between the starting point ore-forming element node and the ending point ore-forming element node, and that do not have a corresponding ore-forming associated object in the ore-forming associated object set, are selected. These location segments are recorded as spatial associated locations. After merging duplicate locations, marking the path source, and organizing the spatial coordinates, a set of hidden associated locations not directly expressed by the ore-forming associated objects is obtained.

[0081] It should be noted that spatially related locations not directly expressed by mineralization-related objects refer to location segments that are not organized as independent objects in the set of mineralization-related objects, but exhibit spatial continuity, directional continuity, or continuous transmission of node characteristics in the related path data.

[0082] During the determination, the starting point ore-forming element node, the ending point ore-forming element node, and the nodes traversed by the path in the associated path data are arranged in spatial order. If there is continuous positional connection and continuous feature continuation between adjacent ore-forming element nodes, and there is no ore-forming associated object corresponding to the position segment in the ore-forming associated object set, then the position segment is recorded as a spatial associated position and included in the hidden associated position set.

[0083] The formula for determining the effective location of hidden associations is: ; in, This indicates the first in the associated path data. One candidate location segment, Indicates the first Valid determination results of hidden associated locations in candidate location segments =1 indicates that the candidate location segment enters the hidden association location set. =0 indicates that the candidate location segment will not be included in the hidden association location set. The value is 1 if all conditions within the parentheses are true, and 0 otherwise. Indicates the first The existing object representation flag for each candidate location segment is set to 1 if a corresponding mineralized associated object already exists in the set of mineralized associated objects, and 0 if no corresponding mineralized associated object exists in the set of mineralized associated objects. Indicates the first The spatial connection marker for each candidate location segment is set to 1 if the candidate location segment can form a spatial continuity with the adjacent ore-forming element nodes or concealed structural proxy nodes; otherwise, it is set to 0. Indicates the first The directional continuation flag for each candidate location segment is set to 1 when the spatial extension direction of the candidate location segment is consistent with or nearly consistent with that of its preceding and following adjacent nodes; otherwise, it is set to 0. Indicates the first The node feature continuation marker of each candidate location segment is set to 1 if there are continuously transferable geological attributes, geophysical anomalies, geochemical anomalies or remote sensing alteration content between adjacent nodes, otherwise it is set to 0.

[0084] , , , and All are binary dimensionless labels. The formula for determining the effective location of the latent association only performs conditional determination of the same type of label, and the dimensions are unified.

[0085] It should be noted that, To represent existing objects, a binary dimensionless label is used. , and These represent spatial connectivity confidence, directional continuity confidence, and node feature continuity confidence, respectively. All are dimensionless continuous values ​​between 0 and 1, used to express the degree of change from weak to strong continuity. The confidence score of the latent association location of the candidate location segment is determined by the state not expressed by existing objects and , and It was jointly determined that, when When the selection criteria for hidden associated locations are met, the candidate location segment is included in the set of hidden associated locations.

[0086] S3.4: Using the hidden associated location set, the location objects corresponding to hidden faults, hidden rock mass boundaries and deep diversion channels are processed into nodes to generate hidden structural proxy nodes. The hidden structural proxy nodes are then connected and recombined with the ore-forming element nodes in the initial ore-forming element map to form an extended ore-forming element map.

[0087] Furthermore, based on the set of hidden associated locations, the location coordinates of each hidden associated location are organized, the associated source records are recorded, and the spatial range is merged to obtain the organized data of hidden associated locations.

[0088] Based on the data organized by the hidden and related locations, the spatial morphology, extension direction, adjacent metallogenic element nodes and corresponding geological features of each location object are first extracted, and then classified in combination with geophysical anomaly extension, structural direction, lithological changes, geochemical element combination and remote sensing alteration distribution.

[0089] Location objects that are linearly and continuously distributed, consistent with known or inferred structural directions, and accompanied by continuous responses from geophysical anomaly zones, geochemical anomaly zones, or remote sensing alteration zones, are identified as location objects corresponding to concealed faults. Location objects that are boundary-shaped or planar transitional, with differences in lithology, stratigraphy, or physical properties on both sides, and corresponding to geophysical boundaries, remote sensing alteration boundaries, or geochemical anomaly boundaries, are identified as location objects corresponding to concealed rock mass boundaries. Location objects that connect multiple mineralization element nodes and form a continuous transmission relationship along structural, geophysical anomaly, geochemical enrichment, and alteration responses are identified as location objects corresponding to deep convection channels, thus obtaining classification data for concealed location objects.

[0090] Based on the classification data of hidden location objects, a proxy node identifier, proxy node location, proxy node type and proxy node characteristics are configured for each location object to complete the nodeization process and generate hidden structure proxy nodes. According to the proxy node location and proxy node characteristics of the hidden structure proxy nodes, the hidden structure proxy nodes are connected to the ore-forming element nodes in the initial ore-forming element map in terms of spatial location, node characteristics or node connection direction to obtain proxy node connection data.

[0091] By using proxy nodes to connect data, the proxy nodes of concealed structures and the data connected by the proxy nodes are incorporated into the initial ore-forming element map. The ore-forming element nodes, the proxy nodes of concealed structures, and the node connection relationships are recorded in a unified manner to form an extended ore-forming element map.

[0092] Specifically, when performing node-based processing on the set of concealed associated locations, the sequence of center points of the grid covering each location object is first extracted, and the main extension direction is obtained by least squares linear fitting. When the location object extends linearly, the fitted center line passes through no less than 3 consecutive 8-neighbor grids, and the angle between the main extension direction and the connection direction of the adjacent ore-forming element node is no greater than 30°, and it corresponds to at least two types of evidence in the geophysical anomaly zone, geochemical anomaly zone, and structural extension record, the location object is organized into the location object corresponding to the concealed fault.

[0093] For location objects distributed in a boundary shape, if there are differences in lithology, physical properties, or geochemical element combinations between adjacent grids on both sides, and the overlap ratio between the boundary line and the geophysical boundary or remote sensing alteration boundary is not less than 0.6, then they are organized into location objects corresponding to the boundaries of concealed rock masses. For location objects that connect no less than 3 ore-forming element nodes, penetrate no less than 3 different grids to be predicted, and satisfy 8-neighborhood continuous connection between adjacent grids, they are organized into location objects corresponding to deep diversion channels, and the spatial location, associated source, corresponding type, and connected objects are written into the proxy node record to form a concealed structure proxy node.

[0094] S4: Based on the extended metallogenic element map, conduct metallogenic evidence propagation obstruction analysis on metallogenic element nodes, concealed structural proxy nodes, and node connection relationships to obtain the metallogenic channel impedance corresponding to each node connection relationship, and mark the metallogenic channel impedance on the extended metallogenic element map to form an impedance-marked metallogenic map.

[0095] S4.1: Based on the extended ore-forming element map, extract the ore-forming element nodes, hidden structure proxy nodes and corresponding node connection relationships, and associate and organize the node characteristics, spatial position relationships and connection extension states at both ends of each node connection relationship to obtain the node connection propagation characteristics.

[0096] Furthermore, based on the extended ore-forming element map, the connection relationship of each node in the extended ore-forming element map is traversed, and the ore-forming element nodes or hidden structure proxy nodes corresponding to both ends of each node connection relationship are extracted. The node identifier, node type, node characteristics and node position of the two ends are read to form the node connection information. According to the node connection information, the spatial positions of the two ends are matched accordingly, and the relative position, connection direction, distance relationship and the range of the grid to be predicted between the two ends are sorted out to form the spatial position relationship.

[0097] Based on the connection direction and adjacent connection distribution of node connection relationships in the extended mineralization element map, the connection relationships of nodes are sorted out to determine whether they continue along the same mineralization association direction, whether they are connected with adjacent node connections, and whether there are any disconnection points, forming a connection extension state. The information of the two ends of the node connection, the spatial position relationship, and the connection extension state are recorded one by one according to the node connection relationship to obtain the node connection propagation characteristics.

[0098] S4.2: Based on the node connection propagation characteristics, organize the ore-forming element nodes, connection directions, spatial connection status and ore-forming evidence transmission content corresponding to each node connection relationship to obtain the node connection propagation organization set.

[0099] Furthermore, based on the node connection propagation characteristics, the two end nodes corresponding to each node connection relationship are read one by one, and the node identifier, node position, node characteristics and node type of the two end nodes are extracted accordingly. The connection direction is determined according to the connection order of the two end nodes in the extended mineralization element map.

[0100] Based on the node positions at both ends, the distance, adjacent state, overlapping state, and extended connection state between nodes are organized to form the spatial connection state corresponding to each node connection relationship. Based on the node characteristics and node types at both ends, the geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence correspondence, and hidden structural association content transmitted along the node connection relationship are merged and organized to form the metallogenic evidence transmission content corresponding to each node connection relationship. The metallogenic element nodes, connection directions, spatial connection states, and metallogenic evidence transmission content corresponding to each node connection relationship are recorded and summarized one by one according to the node connection relationship to obtain the node connection propagation collection.

[0101] S4.3: Based on the node connection propagation set, analyze the continuous transmission, complete preservation, and obstruction / interruption of mineralization evidence in the corresponding node connection relationship, and obtain the data on obstructed propagation of mineralization evidence corresponding to each node connection relationship.

[0102] Furthermore, taking the connection relationship of each node as the processing object, the corresponding mineralization element nodes, connection direction, spatial connection status and mineralization evidence transmission content are extracted. The mineralization evidence transmission content is then arranged sequentially between the starting node, intermediate connection position and ending node according to the connection direction to obtain the node connection evidence transmission sequence.

[0103] Based on the node connection evidence transmission sequence, the continuous transmission of mineralization evidence from the starting node to the ending node is checked, and the retained content of mineralization evidence at different connection positions is compared and organized. If the shortest distance between the endpoints of the corresponding object range of adjacent connection positions is greater than the side length of a grid to be predicted, and the two cannot form a continuous connected chain through 8 neighboring grids, it is marked as a spatial connection break. If the angle between the current connection direction and the previous connection direction or the main extension direction of the object is greater than 30°, it is marked as a connection direction deviation. If the geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, or concealed structural related content retained at the current connection position is less than 60% of the corresponding content at the previous connection position, it is marked as a missing transmission content. Based on the marking results of spatial connection break, connection direction deviation, and missing transmission content, the obstruction and interruption of mineralization evidence in the corresponding node connection relationship is determined.

[0104] Based on the continuous transmission status, complete preservation status, and obstruction / interruption status, a corresponding propagation obstruction record is generated for each node connection relationship. The propagation obstruction record is then associated with the corresponding node connection relationship to obtain the mineralization evidence propagation obstruction data corresponding to each node connection relationship.

[0105] S4.4: Using obstructed data from mineralization evidence propagation, generate corresponding mineralization channel impedances for each node connection relationship, and mark the mineralization channel impedances with the node connection relationships in the extended mineralization element map to form an impedance-marked mineralization map.

[0106] Furthermore, taking node connection relationships as the object of organization, the continuous transmission, complete preservation, and obstruction / interruption of the same node connection relationship in the obstructed propagation data of mineralization evidence are collected to obtain the obstruction status record corresponding to each node connection relationship.

[0107] Based on the obstructed state records corresponding to each node connection, the ore-forming channel impedance is determined according to three categories: spatial connectivity, connection direction, and transmitted content. If adjacent connection positions can be continuously connected through an 8-neighbor grid, the shortest distance between endpoints is no greater than the side length of a grid to be predicted, the angle of the connection direction is no greater than 30°, and the proportion of transmitted content retention is no less than 60%, then a low ore-forming channel impedance is generated and mapped to a dimensionless value of 0.2. If one of the three conditions is not met, but a combination of spatial disconnection and severe content loss is not formed, then a medium ore-forming channel impedance is generated and mapped to a dimensionless value of 0.5. If adjacent connection positions cannot be continuously connected through an 8-neighbor grid, or the shortest distance between endpoints is greater than the side length of a grid to be predicted and the proportion of transmitted content retention is less than 40%, or the angle of the connection direction is greater than 45°, then a high ore-forming channel impedance is generated and mapped to a dimensionless value of 0.8. The ore-forming channel impedance level and dimensionless value corresponding to each node connection relationship are written into the corresponding node connection relationship in the extended ore-forming element map to form an impedance-labeled ore-forming map.

[0108] It should be noted that the impedance of a mineralization channel characterizes the degree of obstruction encountered when mineralization evidence propagates along the node connection relationship. The impedance of a mineralization channel is determined based on the continuous transmission, complete preservation, and obstruction / interruption of the node connection relationship.

[0109] When mineralization evidence can be continuously transmitted through 8 neighboring grids in a node connection relationship, the shortest distance between endpoints is no greater than the side length of a grid to be predicted, the angle between the connection directions is no greater than 30°, and the retention rate of transmitted content is no less than 60%, the corresponding node connection relationship has low mineralization channel impedance. When there are 1 to 2 grid intervals in the mineralization evidence transmission path that do not meet the continuous connection conditions, or the retention rate of transmitted content is greater than or equal to 40% and less than 60%, it is judged as discontinuous transmission or partial content missing, and the corresponding node connection relationship has medium mineralization channel impedance. When there are 3 or more consecutive grid intervals in the mineralization evidence transmission path that do not meet the continuous connection conditions, or the angle between the connection directions is greater than 45°, or the retention rate of transmitted content is less than 40%, it is judged as an blocked state, and the corresponding node connection relationship has high mineralization channel impedance.

[0110] The impedances of low, medium, and high ore-forming channels are mapped to dimensionless values ​​of 0.2, 0.5, and 0.8, respectively, for impedance propagation calculations.

[0111] S5: Input the impedance-labeled mineralization map into GNN, propagate mineralization evidence to mineralization element nodes, concealed structural proxy nodes, and grids to be predicted based on the impedance of the mineralization channel, generate impedance propagation mineralization characterization data for the grids to be predicted, and input the impedance propagation mineralization characterization data into PINNs, perform mineralization physical consistency correction on concealed structural proxy nodes, mineralization channel impedance, and impedance propagation mineralization characterization data, and output mineralization potential distribution data and mineral target area delineation data.

[0112] S5.1: Based on the impedance-labeled mineralization map, extract the mineralization element nodes, hidden structure proxy nodes, grids to be predicted, and the corresponding mineralization channel impedances. Organize the propagation order and impedance constraint matching of the connection relationships of each node to obtain impedance-constrained mineralization propagation information.

[0113] Furthermore, the ore-forming element nodes, concealed structure proxy nodes, grids to be predicted, node connection relationships, and ore-forming channel impedances in the impedance-annotated ore-forming map are read. Each node connection relationship is processed as a unit, and the corresponding node identifiers, node types, node positions, connection start points, connection end points, and ore-forming channel impedances at both ends of the node connection relationship are recorded to form a node connection record.

[0114] Based on the node connection records, the connection source for receiving mineralization evidence for each grid to be predicted is first determined. Then, according to the connection sequence between mineralization element nodes, hidden structure proxy nodes, and grids to be predicted, the propagation order of node connection records pointing to the same grid to be predicted is sorted to obtain the mineralization evidence propagation sequence corresponding to each grid to be predicted.

[0115] Based on the mineralization evidence propagation sequence, the impedance of the mineralization channel corresponding to each node connection relationship is matched to the corresponding propagation position in the mineralization evidence propagation sequence. The propagation start point, intermediate nodes, end point and mineralization channel impedance within the same propagation sequence are continuously recorded to form impedance-constrained mineralization propagation information containing the grid to be predicted, propagation start point, intermediate nodes, end point, propagation order and mineralization channel impedance.

[0116] S5.2: Based on the impedance-constrained mineralization propagation information, the GNN is used to perform obstructed transmission, correlation convergence, and feature reorganization of mineralization evidence between mineralization element nodes, hidden structural proxy nodes, and the grid to be predicted, generating impedance propagation mineralization characterization data of the grid to be predicted.

[0117] Furthermore, based on the impedance-constrained mineralization propagation information, mineralization element nodes, concealed structural proxy nodes, grids to be predicted, node connection relationships, and mineralization channel impedance are extracted. Taking each grid to be predicted as the propagation endpoint, the mineralization element nodes and concealed structural proxy nodes that have node connection relationships with each grid to be predicted are organized to form a node propagation sequence oriented towards the grid to be predicted.

[0118] The impedance-labeled mineralization map, node propagation sequence, and mineralization channel impedance are input into the GNN. In the GNN, the node characteristics of the mineralization element nodes and the hidden structural proxy nodes are read first, and then the mineralization evidence is transmitted to the grid to be predicted along the node connection relationship.

[0119] The ore-forming channel impedance of each node connection is represented by a dimensionless value. The low, medium, and high ore-forming channel impedances correspond to 0.2, 0.5, and 0.8, respectively. The result of subtracting the ore-forming channel impedance from 1 is used as the ore-forming evidence propagation retention coefficient, that is, 0.8 for low impedance, 0.5 for medium impedance, and 0.2 for high impedance. GNN corrects the ore-forming evidence transmission intensity in the node connection relationship according to the propagation retention coefficient, so that the higher the ore-forming channel impedance, the less ore-forming evidence is retained and the lower the transmission degree, and the lower the ore-forming channel impedance, the more ore-forming evidence is retained and the higher the transmission degree, thus obtaining the impedance-constrained node propagation characteristics received by the grid to be predicted.

[0120] Based on the propagation characteristics of the impedance-constrained nodes received by the grid to be predicted, the metallogenic evidence from the metallogenic element nodes and the hidden tectonic proxy nodes is correlated and aggregated. The correlated and aggregated metallogenic evidence is then recombined with the existing geological features of the grid to be predicted to generate impedance propagation metallogenic characterization data of the grid to be predicted.

[0121] Specifically, when GNN propagates mineralization evidence based on the impedance of the mineralization channel, it retains relatively complete mineralization evidence transmission content for node connection relationships corresponding to low mineralization channel impedance, retains some continuous mineralization evidence transmission content for node connection relationships corresponding to medium mineralization channel impedance, and restricts the transmission of mineralization evidence to the grid to be predicted for node connection relationships corresponding to high mineralization channel impedance. It also associates and aggregates mineralization evidence transmitted to the same grid to be predicted by different node connection relationships, so that the grid to be predicted obtains impedance propagation mineralization characterization data corresponding to the impedance constraints.

[0122] S5.3: Input the impedance propagation mineralization characterization data into PINNs, perform mineralization physical consistency correction, deviation marking and correlation reorganization on the concealed tectonic proxy nodes, mineralization channel impedance and impedance propagation mineralization characterization data, and obtain a set of physically corrected mineralization characterizations.

[0123] Furthermore, before inputting the impedance propagation mineralization characterization data into PINNs, the corresponding predictable grids, mineralization element nodes, concealed structural proxy nodes, mineralization channel impedances, and mineralization evidence propagation sources in the impedance propagation mineralization characterization data are first sorted out to obtain the mineralization characterization data to be corrected.

[0124] Based on the mineralization characterization data to be corrected, the impedance propagation mineralization characterization data corresponding to each grid to be predicted is mapped to the concealed tectonic proxy nodes and the impedance of the mineralization channel, and input into PINNs for mineralization physical consistency correction. PINNs are used to perform consistency processing on the spatial connection relationship between concealed tectonic proxy nodes and mineralization element nodes, the correspondence between the impedance of the mineralization channel and the strength of mineralization evidence propagation, and the correspondence between the impedance propagation mineralization characterization data and the spatial continuity state of the grid to be predicted. Records that do not match the location of the concealed tectonic proxy node with the impedance propagation mineralization characterization data, do not match the impedance of the mineralization channel with the state of mineralization evidence propagation, or do not match the continuity of mineralization characterization between the grids to be predicted are marked with deviations to obtain mineralization physical deviation marking data.

[0125] Based on the physical deviation marking data of mineralization, the associated records of concealed structural proxy nodes, the labeled records of mineralization channel impedance, and the grid corresponding records of impedance propagation mineralization characterization data are associated and reorganized to ensure that the concealed structural proxy nodes, mineralization channel impedance, and impedance propagation mineralization characterization data corresponding to the same grid to be predicted are consistent, and a set of physically corrected mineralization characterizations is obtained.

[0126] Specifically, when PINNs performs mineralization physical consistency correction, it verifies the spatial connection between concealed tectonic proxy nodes and mineralization element nodes, the correspondence between the impedance of mineralization channels and the strength of mineralization evidence propagation, and the spatial continuity of impedance propagation mineralization characterization data between adjacent grids to be predicted.

[0127] If the extension direction, spatial connection, or multi-evidence correspondence of the concealed tectonic proxy node and the adjacent ore-forming element node are inconsistent, the concealed tectonic proxy node is marked with a deviation, and its connection object and node connection relationship are rematched. If the impedance level of the ore-forming channel is inconsistent with the degree of preservation of ore-forming evidence propagation, the impedance of the ore-forming channel in the corresponding node connection relationship is marked with a deviation, and the impedance level and dimensionless value are re-determined according to the low, medium, and high impedance mapping rules. If the impedance propagation ore-forming characterization data of the adjacent grid to be predicted does not satisfy the spatial continuity relationship, the characterization data of the corresponding grid is marked with a deviation, and the boundaries are re-merged or preserved according to the adjacent grids, the extension direction of the concealed tectonic structure, and the source of ore-forming evidence. The results after association and reorganization are used to update the connection record of the concealed tectonic proxy node, the impedance labeling record of the ore-forming channel, and the characterization record of the grid to be predicted, forming a physically corrected ore-forming characterization set.

[0128] The formula for the bottleneck mineralization potential due to proxy constraints is: ; in, Indicates the first One grid to be predicted, Indicates the first The mineralization potential values ​​of each grid cell to be predicted are used to form mineralization potential distribution data. Indicates pointing to the first A set of mineralization evidence propagation paths for a grid to be predicted. Represents the set of propagation paths of mineralization evidence The first in One path for the propagation of mineralization evidence. Indicates the first The set of node connections contained in the propagation path of mineralization evidence. This indicates the node connection number, which should be consistent with the node connection number in the impedance labeling of the ore-forming channel. Indicates GNN along the th The strength of mineralization evidence transmitted through node connections is a dimensionless normalized value. Indicates the first The ore-forming channel impedance corresponding to each node connection relationship is a dimensionless classification value corresponding to low, medium, or high ore-forming channel impedance. Indicates the first The bottleneck propagation value in a path of mineralization evidence propagation is determined by the connection relationship of the node with the lowest mineralization evidence transmission capacity within the path. Indicates the first The set of candidate location segments involved in the propagation path of mineralization evidence. Indicates the first The implicit correlation location validity constraint of the propagation path of mineralization evidence is set to 1 when all candidate location segments traversed by the path are valid, 0 when any candidate location segment is invalid, and merged when the path does not traverse any candidate location segment. Keep it computable. Indicates that PINNs is the first The mineralization physical consistency value of the grid output to be predicted is used to characterize the degree of consistency between the concealed tectonic proxy nodes, mineralization channel impedance, and impedance propagation mineralization characterization data.

[0129] Indicates the first The strength of mineralization evidence transmission based on node connectivity. Indicates the first The impedance of the ore-forming channel in the node connection relationship, 1 This represents the retention coefficient of mineralization evidence propagation; therefore, low impedance corresponds to a larger retention coefficient, and high impedance corresponds to a smaller retention coefficient.

[0130] It should be noted that the mineralization physical constraints embedded in PINNs include evidence propagation conservation, channel impedance correspondence, and spatial continuity constraints. For each grid to be predicted, the amount of mineralization evidence entering the grid, the amount of evidence transmitted out of the grid, and the mineralization response formed by geophysical, geochemical, and remote sensing anomalies within the grid are compared to obtain the evidence propagation residual. For each node connection relationship, the mineralization evidence transmission strength obtained by GNN is compared with the allowable propagation strength corresponding to the mineralization channel impedance to obtain the impedance constraint residual. For adjacent grids to be predicted located in the same concealed structural extension direction, the changes in their mineralization potential are compared to obtain the spatial continuity residual. Based on the above residuals, PINNs performs deviation marking and consistency correction on the concealed structural proxy nodes, mineralization channel impedance, and impedance propagation mineralization characterization data.

[0131] The specific training process of PINNs is as follows: the impedance propagation mineralization characterization data output by GNN is used as the training input, and the positions of the grid to be predicted, the positions of the hidden structural surrogate nodes, the impedance values ​​of the mineralization channel, the node propagation path, the corresponding results of known mineral points, and the structural continuity relationship confirmed by geological interpretation are read at the same time. The distribution of known mineral points, mineralization potential samples, structural extension consistency, evidence propagation conservation residuals, channel impedance residuals, and spatial continuity residuals are used as training constraints.

[0132] During training, PINNs perform physical consistency checks on the mineralization characterization output by GNN, checking whether the amount of mineralization evidence entering, exiting and grid responses correspond, whether the impedance of the mineralization channel is consistent with the degree of evidence propagation retention, and whether the mineralization characterization of adjacent grids to be predicted is continuous in the same concealed tectonic direction.

[0133] After training, PINNs are used to perform bias labeling, correlation recombination, and mineralization potential correction on the concealed tectonic proxy nodes, mineralization channel impedance, and impedance propagation mineralization characterization data of the current study area, to obtain a physically corrected mineralization characterization set.

[0134] S5.4: Based on the physical correction mineralization characterization set, perform mineralization potential sorting, spatial continuity processing and target area boundary extraction on the grid to be predicted, and output mineralization potential distribution data and mineral target area delineation data.

[0135] Furthermore, according to the grid number and spatial location of the grid to be predicted, the mineralization characterization content corresponding to each grid to be predicted in the physical correction mineralization characterization set is matched and recorded grid by grid to obtain the mineralization potential sorting result corresponding to each grid to be predicted.

[0136] Based on the results of mineralization potential analysis, adjacent grids with continuous mineralization characteristics are merged, adjacent grids with significant changes in mineralization characteristics are preserved, and spatially isolated grids lacking adjacent support are independently marked to obtain spatially continuous processing results.

[0137] Based on the spatial continuity processing results, the region contour extraction, boundary connection and boundary closure processing are performed on the continuously distributed grids to be predicted. The grids to be predicted that are continuously connected in 8 neighborhoods are classified into the same mineral potential region. The outer grid boundary of the region is used as the initial contour. The adjacent edge endpoints are connected in sequence to form a closed boundary. If there is a gap in the boundary that is no more than the side length of the grid to be predicted, it is filled in the direction of the extension of the adjacent edge. Gaps that are more than the side length of the grid to be predicted are retained as open boundaries and do not enter the target area output.

[0138] Holes are filled in isolated blank grids within the closed boundary that are surrounded by the same mineralization potential area. The jagged lines of the boundary are smoothed according to the direction of adjacent sides. Isolated areas with fewer than 4 consecutive grids or an area smaller than the sum of the areas of 4 grids to be predicted are deleted. The results of the mineralization potential of each grid to be predicted within the processed closed area are recorded in a unified manner to form mineralization potential distribution data. Areas that meet the requirements of closed boundary, number of consecutive grids and minimum area are identified as mineral target areas, and mineral target area delineation data is output.

[0139] It should be noted that when extracting the target area boundary of the grid to be predicted, the physical correction mineralization characterization set corresponding to each grid to be predicted is read according to the grid number. Grids to be predicted that are adjacent in 8 neighborhoods and have continuous mineralization characterization content are classified into the same mineralization potential area. Adjacent grids to be predicted with significant changes in mineralization characterization content are retained as area boundaries. Grids to be predicted that are spatially isolated and lack adjacent support are not included in the mineral target area. If an isolated grid has two or more types of mineralization evidence, such as known mineral occurrences, geophysical anomalies, geochemical anomalies, or remote sensing alteration, it is output separately as a mineralization grid to be verified. If it does not meet the requirement of two or more types of mineralization evidence, it is downgraded to a background grid record.

[0140] The outer edge grids of the same mineral potential region to be predicted are connected and closed. The regions that meet the requirements of continuous grid number, minimum area and closed boundary are organized into the mineral target area range, and the mineral target area delineation data is output.

[0141] In summary, this invention uses GNN to perform graph association propagation on the initial ore-forming element map, identifies hidden association locations between ore-forming element nodes that are not directly expressed, and generates hidden structural proxy nodes, so that hidden faults, hidden rock mass boundaries and deep flow channels are transformed into expressible nodes, providing a graph structure basis for ore-forming channel impedance labeling and PINNs physical consistency correction.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent prediction of mineral resources based on GNN and PINNs, characterized in that, include: Obtain basic data for mineralization prediction, and perform coordinate unification, spatial registration, grid division and feature extraction to form multimodal geoscientific feature data of the study area of ​​the grid to be predicted and geoscientific features; Based on the multimodal geoscientific feature data of the study area, mineralization-related objects were extracted and set as mineralization element nodes. Node connections were established according to the spatial correlation of mineralization to form an initial mineralization element map. The initial ore-forming element map is input into the GNN. Based on the node characteristics, node connection relationships, and spatial distribution relationships of the ore-forming element nodes, graph association propagation is performed to identify the hidden association locations between ore-forming element nodes that are not directly expressed. Hidden structural proxy nodes are generated and incorporated into the initial ore-forming element map to form an extended ore-forming element map. Based on the extended metallogenic element map, the metallogenic element nodes, concealed structural proxy nodes, and node connection relationships are analyzed for obstruction of metallogenic evidence propagation. The metallogenic channel impedance corresponding to each node connection relationship is obtained, and the metallogenic channel impedance is marked on the extended metallogenic element map to form an impedance-marked metallogenic map. The impedance-labeled mineralization map is input into the GNN. Based on the impedance of the mineralization channel, mineralization evidence is propagated to the mineralization element nodes, hidden structural proxy nodes, and the grid to be predicted, generating impedance propagation mineralization characterization data for the grid to be predicted. The impedance propagation mineralization characterization data is then input into PINNs. Mineralization physical consistency correction is performed on the hidden structural proxy nodes, mineralization channel impedance, and impedance propagation mineralization characterization data, and the output mineralization potential distribution data and mineral target area delineation data are then performed.

2. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 1, characterized in that, The specific steps for forming the multimodal geoscientific feature data of the study area, including the grid to be predicted and the geoscientific features, are as follows: Based on the basic data for mineralization prediction, boundary data, geological map data, geophysical anomaly data, geochemical element data, remote sensing alteration data, known mineral occurrence data, and topographic data of the study area were collected. The collected data were then processed by field organization, anomaly record removal, coordinate unification, and spatial registration to obtain spatially benchmarked mineralization data. Based on spatial benchmarked mineralization data, spatial data from different sources are divided into grids, mapped to locations, extracted attributes, transformed features, and reorganized into units to form multimodal geoscientific feature data of the study area, which includes grids to be predicted and geoscientific features.

3. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 1, characterized in that, The specific steps for forming the initial ore-forming element map are as follows: Based on the multimodal geoscientific feature data of the study area, the geological attributes, geophysical anomalies, geochemical anomalies, remote sensing alteration, mineral occurrence locations and spatial boundaries corresponding to each grid to be predicted are identified and grouped into similar categories to obtain a set of mineralization-related objects. Based on the set of mineralization-related objects, the spatial location, attribute characteristics and object category of each mineralization-related object are sorted out. The mineralization-related objects are set as mineralization element nodes according to the sorted object categories to obtain mineralization element node data. Using the node data of ore-forming elements, the proximity distribution, structural connections, anomalous overlaps and mineral occurrence correspondences among the nodes of ore-forming elements are correlated and organized to obtain the node connection relationships. The node data of ore-forming elements and the node connection relationships are then reorganized to form an initial ore-forming element map.

4. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 3, characterized in that, The specific steps for obtaining the mineralization element node data are as follows: Based on the set of mineralization-related objects, each mineralization-related object is assigned an object number, and the spatial location, attribute characteristics, and object category of each mineralization-related object are organized accordingly to obtain the organized data of mineralization-related objects. Based on the data of mineralization-related objects, the object number is used as the node identifier, the spatial location is used as the node location, the attribute characteristics are used as the node characteristics, and the object category is used as the node type to generate mineralization element nodes corresponding to each mineralization-related object. Based on the nodes of ore-forming elements, the node identifier, node location, node characteristics and node type are recorded in a unified manner to obtain ore-forming element node data.

5. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 1, characterized in that, The specific steps for forming the extended ore-forming element map are as follows: Based on the initial ore-forming element map, the node characteristics, node connection relationships, and spatial distribution relationships of the ore-forming element nodes are extracted. The connectivity status, adjacent distribution, and association direction between the ore-forming element nodes are also organized to obtain ore-forming element association and propagation data. The initial ore-forming element map and ore-forming element association propagation data are input into the GNN. The graph association propagation is performed on the association paths and spatial continuity states between ore-forming element nodes to obtain the set of hidden association locations that are not directly expressed by the ore-forming association objects. By utilizing the set of concealed associated locations, the location objects corresponding to concealed faults, concealed rock mass boundaries, and deep diversion channels are processed into nodes to generate concealed structural proxy nodes. The concealed structural proxy nodes are then connected and recombined with the ore-forming element nodes in the initial ore-forming element map to form an extended ore-forming element map.

6. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 5, characterized in that, The specific steps for obtaining the set of hidden association locations not directly expressed by the mineralization association objects are as follows: The initial ore-forming element map and ore-forming element association propagation data are input into the GNN to perform association and propagation on the node characteristics, node connection relationships and spatial distribution relationships of the ore-forming element nodes, thereby obtaining the association path data between the ore-forming element nodes. Based on the associated path data, the spatial continuity between the nodes of the mineralization elements is identified and organized, and the spatial associated positions not directly expressed by the mineralization associated objects are extracted to obtain a set of hidden associated positions not directly expressed by the mineralization associated objects.

7. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 1, characterized in that, The specific steps for forming the impedance-labeled ore-forming map are as follows: Based on the extended metallogenic element map, metallogenic element nodes, concealed structure proxy nodes and corresponding node connection relationships are extracted. The node characteristics, spatial position relationships and connection extension states at both ends of each node connection relationship are correlated and organized to obtain node connection propagation characteristics. Based on the propagation characteristics of node connections, the propagation obstruction analysis is performed on the continuity, integrity and obstruction status of mineralization evidence in different node connection relationships, and the propagation obstruction data of mineralization evidence corresponding to each node connection relationship is obtained. By utilizing obstructed data in the propagation of mineralization evidence, a corresponding mineralization channel impedance is generated for each node connection relationship. The mineralization channel impedance is then labeled with the node connection relationships in the extended mineralization element map to form an impedance-labeled mineralization map.

8. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 7, characterized in that, The specific steps for obtaining the data on obstructed propagation of mineralization evidence corresponding to the connection relationship of each node are as follows: Based on the node connection propagation characteristics, the mineralization element nodes, connection directions, spatial connection status and mineralization evidence transmission content corresponding to each node connection relationship are organized to obtain the node connection propagation organization set. Based on the node connection propagation set, the continuous transmission, complete preservation, and obstruction / interruption of mineralization evidence in the corresponding node connection relationships are analyzed to obtain the data on obstructed propagation of mineralization evidence for each node connection relationship.

9. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 1, characterized in that, The specific steps for generating impedance propagation mineralization characterization data for the grid to be predicted are as follows: Based on the impedance-labeled mineralization map, the mineralization element nodes, hidden structure proxy nodes, grids to be predicted and corresponding mineralization channel impedances are extracted. The propagation order of the connection relationship of each node is sorted and impedance constraint matching is performed to obtain impedance-constrained mineralization propagation information. Based on impedance-constrained mineralization propagation information, GNN is used to perform obstructed transmission, correlation convergence, and feature reorganization of mineralization evidence among mineralization element nodes, hidden structural proxy nodes, and grids to be predicted, generating impedance propagation mineralization characterization data for grids to be predicted.

10. The intelligent mineral resource prediction method based on GNN and PINNs as described in claim 9, characterized in that, The specific steps for outputting mineralization potential distribution data and mineral target area delineation data are as follows: Input the impedance propagation mineralization characterization data into PINNs, perform mineralization physical consistency correction, deviation marking and correlation reorganization on the impedance of concealed tectonic proxy nodes, mineralization channel impedance and impedance propagation mineralization characterization data, and obtain a set of physically corrected mineralization characterizations. Based on the physical correction mineralization characterization set, the mineralization potential of the grid to be predicted is sorted, spatial continuity is processed, and target area boundaries are extracted, outputting mineralization potential distribution data and mineral target area delineation data.

Images