A method and system for predicting sandstone-type uranium ore prospective areas based on GAT

Abstract

The application provides a method and system for predicting sandstone-type uranium mine prospective areas based on GAT, and belongs to the technical field of uranium geological research. By converting stratum unit data, sand body distribution data, structure line data, hydrogeological condition data and mineralization verification point data into graph node basic data and constructing a node feature matrix, the prediction unit not only has its own geological properties, but also has scene adaptation information for expressing the control degree. Then, by compatibility calculation of metallogenic relationship, low control area metallogenic candidate channels and transition graph units are generated and graph structure data is formed, so that the prediction processing process can express local spatial continuity relationship, sand body distribution relationship, underground water direction relationship, structure auxiliary relationship and low control area channel relationship at the same time. Then, based on the node feature aggregation processing of edge relationship features and attention coefficients, the neighborhood information contribution can be automatically adjusted according to the metallogenic relationship between the prediction units, so as to improve the identification ability of the concealed uranium metallogenic prospective area.

Description

Technical Field

[0001] This invention relates to the field of uranium geological research technology, specifically to a method and system for predicting sandstone-type uranium deposit prospect areas based on GAT. Background Technology

[0002] Sandstone-type uranium deposits are an important type of uranium resource, typically found in sandstone strata with a certain porosity and permeability. Their formation is closely related to uranium-bearing fluid migration, the development of reducing environments, favorable sand body distribution, tectonic alteration, and stratigraphic control. In sandstone-type uranium exploration, the purpose of prospective exploration area prediction is to identify favorable sections with high mineralization potential within a larger exploration area based on limited geological data, providing a basis for subsequent geological verification, drilling deployment, and resource evaluation.

[0003] Existing methods for predicting prospective uranium deposits in sandstone formations typically employ techniques such as overlaying geological expert experience, raster factor weighting, buffer analysis, local interpolation, statistical regression, and conventional machine learning classification. These methods generally transform stratigraphic conditions, sand body conditions, tectonic conditions, and hydrogeological conditions into multiple evaluation factors, which are then manually weighted, statistically fitted, or trained on samples to obtain the predicted prospective deposit distribution. In areas with good surface outcrops, high borehole control, and clear geological continuity, these methods can provide a certain degree of reference for the distribution of prospective deposits.

[0004] However, traditional prediction methods are significantly inadequate in sandstone-type uranium exploration areas covered by sand dunes and Gobi deserts, where boreholes are sparsely distributed along a limited number of working channels. These areas have strong surface cover, limited outcrop information, and field verification points and boreholes are often distributed in a zonal pattern due to accessibility, construction conditions, and topographical features. Large low-control zones exist between different working zones. Traditional planar overlay methods typically rely on spatial proximity, the strength of single factors, or weighted by human experience as primary criteria, making it difficult to express the geological connections between distant prediction units within the same sandstone body zone, or the influence of groundwater migration direction on the transmission of mineralization information. Sandstone-type uranium mineralization is not a simple process controlled by a single element, but rather the result of the spatial coordination of strata, sandstone bodies, structures, and hydrological conditions. For example, even if a prediction unit is close to a known mineralization point, its mineralization potential may not be high if it is located in an unfavorable stratum, has discontinuous sand bodies, or has a mismatch in groundwater migration direction. Conversely, even if a prediction unit is some distance from a known mineralization point, it may still have high prospective value if it is located in the same favorable stratum, in the direction of continuous sand body distribution, and if the groundwater migration relationship is consistent with the migration direction of uranium-bearing fluids.

[0005] Furthermore, traditional methods inferring prospective area boundaries in low-control zones typically rely on planar interpolation or empirical extension, easily leading to problems such as coarse boundaries, isolated local high values, and unclear prediction basis. For strategic uranium exploration, prediction results not only need to indicate potentially favorable locations but also explain the main mineralization factors supporting the prospective area, the strength of the prediction results, and whether low-control zones require priority verification. Existing methods still have shortcomings in these aspects.

[0006] Therefore, there is an urgent need for a new technology for predicting potential areas of sandstone-type uranium deposits. This technology should be able to unify stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data into graphical structure data. Furthermore, it should construct candidate channels and transitional graphical units with mineralization significance in low-control areas. This would enable the prediction process to learn the spatial, directional, and mineralization relationships between different prediction units, thereby improving the accuracy, continuity, and interpretability of predicting concealed potential areas of sandstone-type uranium deposits. Summary of the Invention

[0007] The purpose of this invention is to provide a method for predicting promising sandstone-type uranium deposits based on Gaussian Atlas (GAT). This method aims to address the problem that traditional methods for predicting promising sandstone-type uranium deposits, which are often characterized by sand dunes, Gobi deserts, and sparsely distributed boreholes along limited access routes, struggle to accurately represent the spatial coupling between strata, sand bodies, structures, and hydrology. This results in low accuracy, coarse boundary representation, and insufficient interpretability of predictions for promising areas of concealed uranium mineralization.

[0008] To achieve the above objectives, a first aspect of the present invention provides a method for predicting sandstone-type uranium deposit prospect areas based on Gaussian Atlas (GAT), the method comprising: Obtain basic geological data of the exploration area, divide the exploration area into multiple prediction units according to the boundary of the exploration area, and convert each prediction unit into corresponding basic data of map nodes; Based on the graph node basic data, the metallogenic elements of each prediction unit are numerically processed to generate a node feature matrix that includes stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value and degree of control, and a set of verification labels is generated based on the mineralization verification point data. Based on the node feature matrix and the degree of control, mineralization relationship compatibility calculation is performed on the low-control prediction unit to generate mineralization candidate channels, transition map units and updated node sets in the low-control area. Based on the updated node set, a complete edge set is constructed, consisting of spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on the complete edge set, graph structure data containing node features and edge relationship features is generated. The graph structure data is subjected to node feature aggregation processing based on edge relationship features and attention coefficients, and the graph attention processing parameters are updated based on the verification label set to generate the prospect probability, confidence level, prospect level and standardized prediction results of each prediction unit.

[0009] A second aspect of the present invention provides a system for predicting sandstone-type uranium deposit prospects based on GAT, the system being used to perform the above-described method for predicting sandstone-type uranium deposit prospects based on GAT, the system comprising: The unit division is used to obtain the geological basic data of the exploration area. Multiple prediction units are divided according to the boundary of the exploration area, and each prediction unit is converted into the corresponding basic data of the map node. The generation unit is used to perform ore-forming element numerical processing on each prediction unit based on the graph node basic data, generate a node feature matrix including stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value and control degree, and generate a set of verification labels based on mineralization verification point data. The calculation unit is used to perform mineralization relationship compatibility calculation on the low-control prediction unit based on the node feature matrix and the degree of control, and generate mineralization candidate channels, transition map units and updated node sets in the low-control area. The construction unit is used to construct a complete edge set based on the updated node set, including spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on the complete edge set, graph structure data containing node features and edge relationship features is generated. The update unit is used to perform node feature aggregation processing based on edge relationship features and attention coefficients on the graph structure data, and update the graph attention processing parameters based on the verification label set to generate the prospect probability, confidence level, prospect level and standardized prediction results of each prediction unit.

[0010] Through the above technical solution, the present invention has at least the following technical effects: First, the present invention converts stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data into unified map node basic data, avoiding the problems of scattered processing of different data and unclear factor superposition relationships in traditional methods, thus enabling the prediction of sandstone-type uranium deposit prospective areas to have a continuous data transmission foundation.

[0011] Second, the present invention constructs a node feature matrix by considering stratigraphic suitability, sand body favorableness, tectonic influence value, hydrological transport value, and degree of control. This enables the prediction unit to not only possess its own geological attributes but also scenario adaptation information for expressing the degree of control, thereby facilitating the adoption of data processing methods that are more in line with actual exploration conditions in low-control areas.

[0012] Third, the present invention generates candidate mineralization channels and transition map units in low-control areas through compatibility calculation of mineralization relationships, so that the working zones in scenarios with sand dunes, Gobi Desert coverage and sparse drilling are no longer treated as simple blank areas, but can enter the map structure calculation under the premise of geological basis.

[0013] Fourth, the present invention forms graph structure data by combining spatial adjacent edges, sand body connecting edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges, so that the prediction and processing process can simultaneously express local spatial continuity relationships, sand body distribution relationships, groundwater direction relationships, tectonic auxiliary relationships, and low-control zone channel relationships.

[0014] Fifth, the present invention uses node feature aggregation processing based on edge relationship features and attention coefficients, and utilizes the construction of a graph attention network (GAT) to make the prospect probability calculation no longer rely solely on the factor level of a single prediction unit, but can automatically adjust the contribution of neighborhood information according to the mineralization connection between prediction units, thereby improving the identification ability of concealed uranium mineralization prospect areas.

[0015] Sixth, the present invention's solution, through the output of credibility, comprehensive prospect score, prospect level, and contribution factor map, enables the prediction results to include not only the location and level of the prospect area, but also the strength of the prediction basis and the main contributing factors, which facilitates subsequent verification and deployment by surveyors.

[0016] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the steps of a method for predicting sandstone-type uranium deposit prospect areas based on GAT, provided in one embodiment of the present invention. Figure 2 This is a system structure diagram of a system for predicting sandstone-type uranium deposit prospect areas based on GAT, provided in one embodiment of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] In this embodiment of the invention, the sandstone-type uranium deposit prospect prediction method is applicable to sandstone-type uranium deposit exploration areas covered by sand dunes, Gobi deserts, and other areas where boreholes or verification points are unevenly distributed. It should be noted that the scenario of sand dune and Gobi cover with boreholes sparsely distributed along a limited number of working channels is only one of the preferred application scenarios of this invention and does not constitute a limitation on the scope of application of this invention. This embodiment of the invention is also applicable to other sandstone-type uranium deposit exploration areas with a need for predicting concealed ore bodies, uneven geological control, or complex spatial relationships among ore-forming elements.

[0020] like Figure 1 As shown, this invention provides a method for predicting sandstone-type uranium deposit prospect areas based on GAT, comprising the following steps: S10: Obtain basic geological data of the exploration area, divide the exploration area into multiple prediction units according to the boundary of the exploration area, and convert each prediction unit into corresponding basic data of the map node.

[0021] Specifically, the process involves acquiring geological baseline data and boundary data of the exploration area, and generating basic prediction units covering the exploration area based on the boundary data. The geological baseline data includes stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data. Based on the stratigraphic boundaries in the stratigraphic unit data, the sand body boundaries in the sand body distribution data, and the structural line positions in the structural line data, geological boundary correction processing is performed on the basic prediction units to generate multiple prediction units with consistent geological attributes. Each prediction unit is converted into a map node, and each map node is configured with a corresponding spatial location, unit area, and node number. The stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data are mapped to the corresponding map nodes to generate basic map node data.

[0022] In this embodiment of the invention, step S10 is used to establish the basic data objects for the entire prospective area prediction method. Specifically, firstly, stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data of the exploration area are acquired. The stratigraphic unit data is used to characterize the interpretation results related to different stratigraphic positions, lithological combinations, inter-layer contact relationships, and redox interfaces within the exploration area; the sand body distribution data is used to characterize sand body boundaries, sand body thickness, continuous length of sand bodies, sand body numbers, and main distribution direction of sand bodies; the structural line data is used to characterize major faults, secondary structural lines, structural levels, structural strikes, and the extent of structural influence; the hydrogeological condition data is used to characterize the main migration direction of groundwater, hydraulic potential distribution, hydraulic gradient, and recharge-discharge relationship; the mineralization verification point data is used to characterize known mineralization points, control points that have been verified but do not show obvious mineralization response, borehole locations, and verification results.

[0023] It is readily understood that the stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data described in the embodiments of this invention are crucial and common data types in the prediction of sandstone-type uranium deposit prospective areas. This invention does not require the use of high-density borehole data or additional complex exploration data in all embodiments. Instead, it enables limited data to participate in subsequent map structure calculations by performing unified transformation and relational expression on the aforementioned key data. This setup can adapt to actual exploration conditions where data is scarce in dune- and Gobi-covered areas and verification points are unevenly distributed.

[0024] In practical applications, the system generates basic prediction units covering the entire exploration area based on the exploration area boundary. These basic prediction units can be regular grid cells or irregular cells formed by the joint division of stratigraphic boundaries, sandbody boundaries, and structural lines. Specifically, in one executable implementation, a regular grid can be generated first based on the exploration area boundary, and then the regular grid can be locally modified using the positions of stratigraphic boundaries, sandbody boundaries, and structural lines to form multiple prediction units with consistent geological attributes. It should be noted that the consistency of geological attributes does not require the absence of any minor differences within the prediction unit, but rather that the prediction unit has an acceptable consistency in terms of the main stratigraphic positions, sandbody attributions, tectonic influences, and hydrological conditions required for subsequent calculations.

[0025] Furthermore, the size of the prediction unit can be set according to the exploration scale, sand body distribution scale, borehole spacing, and result representation accuracy. For example, in areas with large borehole spacing and wide sand body distribution, larger prediction units can be used; in areas with rapid sand body changes or dense structural lines, smaller prediction units or irregular prediction units corrected for geological boundaries can be used. The specific scale selection mentioned above should not be construed as a limitation of the present invention. As long as the exploration area can be divided into prediction units that can be used for map structure representation, it can be considered as an optional implementation of the present invention.

[0026] In this embodiment of the invention, each prediction unit is converted into a graph node, and each graph node is configured with a corresponding spatial location, unit area, and node number. Subsequently, stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data are mapped to the corresponding graph nodes. Specifically, for planar stratigraphic unit data, the stratigraphic position and lithological assemblage of the prediction unit can be determined based on the spatial overlay relationship between the prediction unit and the surface object; for sand body distribution data, the sand body number, sand body thickness, and main distribution direction can be determined based on the overlap relationship between the prediction unit and the sand body boundary or sand body zone; for structural line data, the distance from the prediction unit to the structural line can be calculated, and the directional relationship between the structural line strike and the direction of the relevant sand body in the prediction unit can be determined; for hydrogeological condition data, the main groundwater migration direction and hydraulic potential relationship can be extracted based on the location of the prediction unit; for mineralization verification point data, the verification status of the prediction unit can be determined based on whether the verification point falls within the prediction unit.

[0027] In one executable implementation, multiple prediction units can be represented as a set of graph nodes:

[0028] in, This represents the set of graph nodes corresponding to the prediction unit; Indicates the first Each prediction unit corresponds to a graph node; Indicates the prediction unit number; This indicates the total number of prediction units.

[0029] To further organize the basic geological properties of each prediction unit, the first The initial attribute set of each prediction unit can be represented as:

[0030] in, Indicates the first The initial set of attributes for each prediction unit; Indicates the first The stratigraphic attributes corresponding to each prediction unit; Indicates the first Sand body attributes corresponding to each prediction unit; Indicates the first The construction attributes corresponding to each prediction unit; Indicates the first The original hydrogeological attributes corresponding to each prediction unit.

[0031] In this embodiment of the invention, the degree of control of the prediction unit is also calculated based on the mineralization verification point data. The degree of control describes the sufficiency of verification information surrounding the prediction unit; it does not directly indicate mineralization favorability, but rather identifies areas with limited verification information and low control. The formula for calculating the degree of control is as follows:

[0032] in, Indicates the first The degree of control over each prediction unit; Indicates the first Number of mineralization verification points in the neighborhood of each prediction unit; This represents the control level balancing parameter, used to adjust the impact of the number of verification points on the control level.

[0033] It should be noted that the degree of control increases as the number of validation points in the neighborhood of the prediction unit increases; conversely, the degree of control remains low when the number of validation points is small. For scenarios with sand dunes or Gobi covering and boreholes sparsely distributed along a few work channels, this degree of control can reflect the validation differences near and between work channels, enabling subsequent methods to specifically address low-control areas.

[0034] After obtaining the initial attributes and degree of control, the first... The basic data of each graph node can be represented as follows:

[0035] in, Indicates the first Basic data for each graph node; Indicates the first Each prediction unit corresponds to a graph node; Indicates the first The initial set of attributes for each prediction unit; Indicates the first The degree of control over each prediction unit.

[0036] In this embodiment of the invention, the basic data of the map nodes output in step S10 includes not only the spatial objects of the nodes, but also the stratigraphic, sand body, structural, hydrological, and control level data required for subsequent calculations. This basic data of the map nodes unifies geological information from different spatial morphologies and data sources into the data carrier of the prediction unit.

[0037] S20: Based on the basic data of the graph nodes, perform numerical processing of mineralization elements on each prediction unit to generate a node feature matrix that includes stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value and control degree, and generate a set of verification labels based on mineralization verification point data.

[0038] Specifically, based on the stratigraphic unit data of the map node basic data, the stratigraphic position, lithological assemblage, and stratigraphic assemblage type of each prediction unit are determined, and the stratigraphic suitability of each prediction unit is generated according to the mineralization suitability corresponding to different stratigraphic assemblage types; based on the sand body distribution data of the map node basic data, the sand body thickness, sand body continuity length, and sand body distribution stability of each prediction unit are extracted, and the sand body thickness, sand body continuity length, and sand body distribution stability are converted into sand body favorableness for each prediction unit; based on the structural line data of the map node basic data, the distance from each prediction unit to the structural line, structural level, and main sand body extension are extracted. The directional relationship between the distribution direction and the strike of the structural line is established, and the structural influence value of each prediction unit is generated. Based on the hydrogeological condition data of the basic data of the graph nodes, the main migration direction of groundwater, the hydraulic potential relationship, and the spatial directional relationship between prediction units are extracted, and the hydrological migration value of each prediction unit is generated. At the same time, the control degree of each prediction unit is calculated based on the mineralization verification point data. The stratigraphic suitability, sand body favorability, structural influence value, hydrological migration value, and control degree are combined into a node feature matrix according to the prediction unit number, and a set of verification labels is generated according to whether the mineralization verification point falls into the corresponding prediction unit and the verification status of the mineralization verification point.

[0039] In this embodiment of the invention, step S20 is used to convert the basic graph node data generated in step S10 into a node feature matrix that can be used for graph attention processing, and at the same time generate a set of verification labels based on the mineralization verification point data. It should be noted that the numerical processing of mineralization elements is not a mechanical numbering of geological objects, but rather the conversion of the understanding of favorable strata, sand body connectivity, tectonic influences, and hydrological migration in the mineralization theory of sandstone-type uranium deposits into computable features, so that the subsequent data processing can reflect geological meaning.

[0040] In this embodiment of the invention, the stratigraphic position, lithological assemblage, and stratigraphic assemblage type of each prediction unit are first determined based on stratigraphic unit data. Then, stratigraphic suitability is generated according to the mineralization suitability corresponding to different stratigraphic assemblage types. Sandstone-type uranium deposits exhibit obvious stratabound characteristics. If the strata in which the prediction unit is located do not possess ore-hosting conditions, it should not be simply judged as a high-prospect area even if it is close to known mineralization points. Therefore, stratigraphic suitability is used to express the basic mineralization favorableness of the prediction unit based on stratigraphic conditions.

[0041] In one executable implementation, the formation suitability is calculated as follows:

[0042] in, Indicates the first Stratigraphic suitability of each prediction unit; Indicates the stratigraphic combination type number; Indicates the total number of stratigraphic combinations; Indicates the first Metallogenic suitability coefficients corresponding to stratigraphic combinations; Indicates the first The stratigraphic assemblage type to which each prediction unit belongs; Indicates the indicator function, when the first... The prediction unit belongs to the first... When combining similar strata, take 1; otherwise, take 0.

[0043] In this embodiment of the invention, the geological understanding represented by stratigraphic positions and lithological combinations is converted into a calculable suitability degree. Through this processing, prediction units located in favorable stratigraphic combinations receive a higher stratigraphic suitability degree, while prediction units located in unfavorable stratigraphic combinations receive a lower stratigraphic suitability degree. It should be noted that the mineralization suitability coefficient can be set based on existing geological understanding of the exploration area, regional sandstone-type uranium mineralization patterns, or expert experience; this invention does not limit its specific value.

[0044] Following this, the embodiments of the present invention further extract the sand body thickness, continuous length, and distribution stability of each prediction unit based on sand body distribution data, and convert the sand body thickness, continuous length, and distribution stability into sand body favorableness. Sand bodies are important spaces for uranium-bearing fluid migration and uranium precipitation in sandstone-type uranium deposits. Specifically, sand body thickness reflects the possible ore-bearing space within the prediction unit, continuous length reflects the connectivity of the sand body within the region, and distribution stability reflects the reliability of the sand body as a fluid channel. By combining the above information into sand body favorableness, sand body conditions can be used as node features in subsequent graph structure calculations.

[0045] The formula for calculating the favorable properties of sand bodies is as follows:

[0046] in, Indicates the first The favorable conditions of sand bodies in each prediction unit; Represents the Sigmoid function; This represents the influence coefficient corresponding to the sand body thickness; This represents the influence coefficient corresponding to the continuous length of the sand body; This represents the influence coefficient corresponding to the stability of sand body distribution; Indicates the first Standardized values ​​of sand body thickness for each prediction unit; Indicates the first Standardized value of continuous length of sand body in each prediction unit; Indicates the first Standardized values ​​of sand body distribution stability for each prediction unit.

[0047] The Sigmoid function can be represented as:

[0048] in, This represents the output value of the Sigmoid function; Indicates the function's input value; Represents the natural constant.

[0049] In this embodiment of the invention, sand body thickness, sand body continuity length, and sand body distribution stability are collectively converted into a sand body favorableness level between zero and one, facilitating their participation in subsequent nodal feature aggregation along with other mineralization elements. In practical applications, if the sand body thickness in a certain exploration area contributes more significantly to mineralization, the influence coefficient corresponding to sand body thickness can be increased; if sand body continuity is more critical to the extension of the regional prospective area, the influence coefficient corresponding to sand body continuity length can be increased. The above coefficient adjustments are optional embodiments of this invention and do not constitute a limitation on the scope of protection of this invention.

[0050] Following this, the embodiments of the present invention further extract the directional relationships between the distance from each prediction unit to the structural line, the structural level, the main distribution direction of the sand body, and the strike of the structural line based on the structural line data, and generate structural influence values. In the mineralization process of sandstone-type uranium deposits, structures are usually not a separate mineralization factor, but rather alter the mineralization favorability by affecting sand body connectivity, groundwater migration, or local reducing environments. Therefore, the structural influence values ​​in the embodiments of the present invention simultaneously consider structural distance, structural level, and directional relationships.

[0051] The formula for calculating the influence value is as follows:

[0052] in, Indicates the first The construction impact value of each prediction unit; Indicates the first The construction level coefficients corresponding to each prediction unit; Represents the natural exponential function; Indicates the first The distance of each prediction unit to the nearest construction line; This indicates the influence of the construction on distance parameters; This indicates the operation of finding the maximum value. Represents the cosine function; Indicates the first The main distribution direction of the sand body in each prediction unit; Indicates the first The direction of the construction line corresponding to each prediction unit.

[0053] In this embodiment of the invention, the structural influence generally decreases with increasing distance, and is also affected by the degree of matching between the sand body orientation and the structural orientation. When there is a good matching relationship between the main distribution direction of the sand body and the strike of the structural line, the auxiliary effect of the structure on sand body connectivity and groundwater migration is enhanced; when the directional relationship is unfavorable, the structural influence is reduced. Compared with simple structural buffer zone treatment, this embodiment of the invention can express the influence of structure on the prospect prediction of sandstone-type uranium deposits in more detail.

[0054] Subsequently, this embodiment of the invention further extracts the main migration direction of groundwater, hydraulic potential relationship, and spatial directional relationship between prediction units based on hydrogeological data, and generates hydrogeological migration values. Sandstone-type uranium mineralization is closely related to the migration of uranium elements carried by groundwater, and the groundwater migration direction can influence the spatial extension direction of mineralization information. Therefore, hydrogeological migration values ​​are used to describe the directional matching relationship between prediction units and surrounding candidate units.

[0055] The formula for calculating the hydrological transport matching degree is as follows:

[0056] in, Indicates the first The prediction unit to the first Hydrological transport matching degree of each prediction unit; Indicates by the first The prediction unit to the first Spatial orientation of each prediction unit; Indicates the first The main migration direction of groundwater corresponding to each prediction unit; Indicates the first The prediction unit and the first The hydraulic potential difference between the prediction units; This represents the scale parameter of hydraulic potential difference.

[0057] The hydrological transport value of the predicted unit can be expressed as:

[0058] in, Indicates the first Hydrological transport values ​​for each prediction unit; Indicates the first The set of candidate neighborhood units for each prediction unit; Indicates the number of units in the candidate neighborhood unit set; Indicates by the first The prediction unit to the first Hydrological transport matching degree of each prediction unit; Indicates the candidate neighborhood cell number.

[0059] It should be noted that the hydrological transport value does not simply indicate the presence of groundwater within the prediction unit, but rather expresses whether the prediction unit conforms to the groundwater transport direction and hydraulic potential relationship with surrounding units. Through this processing, the system can preferentially identify favorable prospective areas along the groundwater transport direction in low-control zones, rather than expanding solely based on planar distance.

[0060] After generating stratigraphic suitability, sand body favorability, tectonic influence, hydrological transport, and degree of control, the above features are combined into node feature vectors according to the prediction unit number:

[0061] in, Indicates the first The node feature vector of each prediction unit; Indicates the first Stratigraphic suitability of each prediction unit; Indicates the first The favorable conditions of sand bodies in each prediction unit; Indicates the first The construction impact value of each prediction unit; Indicates the first Hydrological transport values ​​for each prediction unit; Indicates the first The degree of control over each prediction unit.

[0062] The node feature matrix is ​​composed of the node feature vectors of all prediction units:

[0063] in, This represents the node feature matrix corresponding to all prediction units; This represents the node feature vector of the first prediction unit; This represents the node feature vector of the second prediction unit; Indicates the first The node feature vector of each prediction unit; This indicates the total number of prediction units.

[0064] In this embodiment of the invention, a set of verification tags is generated based on whether the mineralization verification point falls into the corresponding prediction unit and the verification status of the mineralization verification point. Specifically, if a prediction unit contains a mineralization verification point that meets the preset mineralization judgment conditions, then the prediction unit is used as a favorable sample for mineralization; if a prediction unit contains a verification point that has been verified but does not show a significant mineralization response, then the prediction unit is used as a control sample; for prediction units that have not been verified, they are not directly marked as control samples, but are used as nodes to be predicted in subsequent calculations. The set of verification tags can be represented as follows:

[0065] in, Represents the set of node indexes with verification labels; Indicates the node number; Indicates the first Validation labels for each prediction unit.

[0066] In this embodiment of the invention, it is significant that unverified prediction units are not directly used as unfavorable samples. Sand dune and Gobi desert areas contain numerous unexplored or unverified regions, which cannot be simply considered as mineral-free areas. Using unverified prediction units directly as control samples could lead to overly conservative predictions for potential areas. Therefore, this embodiment of the invention generates a set of verification labels only based on actual verification states, while allowing unverified prediction units to participate in potential probability calculations through graph-structured relationships.

[0067] S30: Based on the node feature matrix and the degree of control, perform mineralization relationship compatibility calculation on the low-control prediction unit to generate mineralization candidate channels, transition map units and updated node sets in the low-control area.

[0068] Specifically, based on the control degree of the node feature matrix and the preset control degree threshold, low-control prediction units are identified from multiple prediction units, forming a set of low-control units; based on the stratigraphic relationship and stratigraphic suitability of each prediction unit, the stratigraphic consistency compatibility between prediction units is calculated, and based on the sand body zone to which each prediction unit belongs, the difference in sand body thickness, and the relationship of sand body distribution direction, the sand body connectivity compatibility between prediction units is calculated; based on the spatial direction, the main migration direction of groundwater, and the hydraulic potential relationship between each prediction unit, the hydrological direction compatibility between prediction units is calculated; based on the distance difference between each prediction unit to the structural line and the matching relationship between the structural line strike and the connection direction of the prediction unit, the structural auxiliary compatibility between prediction units is calculated.

[0069] Based on this, mineralization candidate channel coefficients are generated between prediction units according to stratigraphic consistency compatibility, sand body connectivity compatibility, hydrological orientation compatibility, and tectonic auxiliary compatibility. Node relationships that satisfy preset channel conditions and pass through the low-control unit set are identified as mineralization candidate channels in the low-control area. Transitional map units are generated based on the spatial length of the mineralization candidate channels in the low-control area, the desired map unit spacing, the node characteristics of the prediction units at both ends of the channel, and the channel relationship characteristics. These transitional map units are added to the node sets corresponding to the original prediction units to form updated node sets, which are then transmitted to the graph structure data construction and processing.

[0070] In this embodiment of the invention, step S30 is used to establish geologically significant predictive unit relationships in low-control areas. Traditional prospective area prediction methods often rely solely on planar distances or empirical extrapolation in low-control areas, making it difficult to express potential mineralization connections within the same sandstone belt, the same groundwater migration direction, or the same tectonic influence zone. This embodiment of the invention, based on node feature matrices and control levels, first identifies low-control predictive units, then performs mineralization relationship compatibility calculations, further generating candidate mineralization channels and transition map units in the low-control area.

[0071] In this embodiment of the invention, firstly, based on the degree of control and a preset control degree threshold, low-control prediction units are identified from multiple prediction units. The determination formula for low-control prediction units is as follows:

[0072] in, Indicates the first The degree of control over each prediction unit; This indicates the threshold for the degree of control.

[0073] The set of low-level control units can be represented as:

[0074] in, Represents the set of low-level control units; Indicates the first Each prediction unit corresponds to a graph node; Indicates the first The degree of control over each prediction unit; This indicates the threshold for the degree of control.

[0075] It should be noted that a low-control prediction unit does not necessarily indicate poor mineralization conditions, but rather a lack of sufficient direct verification information around the prediction unit. In areas covered by sand dunes and Gobi deserts, where boreholes are sparsely distributed along the working channels, low-control prediction units are often located between working channels or far from existing boreholes. This embodiment of the invention identifies low-control prediction units, enabling subsequent channel construction and transition map unit generation to have clearly defined targets.

[0076] In this embodiment of the invention, the stratigraphic consistency compatibility between prediction units is further calculated based on the stratigraphic relationship and stratigraphic suitability of each prediction unit. The stratigraphic consistency compatibility is used to determine whether two prediction units are located in the same favorable stratigraphic position or have a connectable stratigraphic relationship. Its calculation formula is as follows:

[0077] in, Indicates the first The prediction unit and the first Stratigraphic consistency compatibility among prediction units; Indicates the first Stratigraphic position of each prediction unit; Indicates the first Stratigraphic position of each prediction unit; This indicates that two stratigraphic horizons are interconnected; Indicates the first Stratigraphic suitability of each prediction unit; Indicates the first Stratigraphic suitability of each prediction unit; This indicates the indicator function, which takes the value 1 when two prediction units are located at the same level or have a connectable level relationship, and takes the value 0 otherwise.

[0078] It should be noted that the spatial extension of sandstone-type uranium deposit prospective areas should conform to stratabound laws. If two prediction units lack consistency or connectivity in stratigraphy, a mineralization connection should not be forcibly established even if they are spatially close. This approach prevents candidate mineralization pathways from arbitrarily crossing unfavorable strata, thereby improving the geological rationality of prediction results in low-control areas.

[0079] In this embodiment of the invention, the sand body connectivity compatibility between prediction units is further calculated based on the relationship between the sand body zone to which each prediction unit belongs, the difference in sand body thickness, and the direction of sand body distribution. The sand body connectivity compatibility is used to express the strength of the connection between two prediction units in terms of sand body distribution. Its calculation formula is as follows:

[0080] in, Indicates the first The prediction unit and the first Sand body connectivity compatibility between prediction units; Indicates the first The sand body zone number to which each prediction unit belongs; Indicates the first The sand body zone number to which each prediction unit belongs; This indicates the indicator function, which takes a value of 1 when two prediction units belong to the same sand body zone, and a value of zero otherwise. Indicates the first The main distribution direction of the sand body in each prediction unit; Indicates the first The main distribution direction of the sand body in each prediction unit; Indicates the first Sand body thickness of each prediction unit; Indicates the first Sand body thickness of each prediction unit; This represents the scale parameter indicating the difference in sand body thickness.

[0081] It should be noted that two prediction units located within the same sand body zone, with similar main distribution directions and relatively gentle changes in sand body thickness, are more likely to have fluid migration and mineralization connections. Compared to establishing relationships solely based on spatial adjacency, this process can identify prediction units that are not directly adjacent in space but are located within the same sand body channel, making it more suitable for application scenarios with sparse boreholes in overburden areas.

[0082] In this embodiment of the invention, the hydrological direction compatibility between prediction units is further calculated based on the spatial orientation, main groundwater migration direction, and hydraulic potential relationship between each prediction unit. The hydrological direction compatibility is used to express whether the connection between two prediction units conforms to the groundwater migration direction and hydraulic potential relationship. Its calculation formula is as follows:

[0083] in, Indicates the first The prediction unit to the first Hydrological orientation compatibility of each prediction unit; Indicates by the first The prediction unit points to the first... Spatial orientation of each prediction unit; Indicates the first The main migration direction of groundwater corresponding to each prediction unit; Indicates the first Hydraulic potential values ​​for each prediction unit; Indicates the first Hydraulic potential values ​​for each prediction unit; Indicates when the first The prediction unit up to the first If a prediction unit conforms to the hydraulic potential relationship, it is set to 1; otherwise, it is set to 0. This represents the scale parameter for the difference in hydraulic potential.

[0084] It should be noted that sandstone-type uranium mineralization is closely related to groundwater migration. If the spatial orientation between two prediction units coincides with the main groundwater migration direction, and the hydraulic potential relationship supports fluid migration, then there is a high degree of hydrological directional compatibility between them. Through this treatment, mineralization connections within low-control zones can be expressed along the favorable groundwater migration direction, rather than extending in arbitrary directions.

[0085] In this embodiment of the invention, a structural auxiliary compatibility degree is also calculated based on the distance differences between each prediction unit and the structural line, as well as the matching relationship between the structural line strike and the connection direction of the prediction units. The structural auxiliary compatibility degree is used to express the auxiliary role of structural relationships in the mineralization connection between two prediction units. Its calculation formula is as follows:

[0086] in, Indicates the first The prediction unit to the first Construction of each prediction unit aids compatibility; Indicates the first The distance of each prediction unit to the nearest construction line; Indicates the first The distance of each prediction unit to the nearest construction line; Indicates the scale parameter of the construction distance difference; Indicates the first The direction of the construction line corresponding to each prediction unit; Indicates by the first The prediction unit points to the first... The spatial orientation of each prediction unit.

[0087] It should be noted that when two predicted units are close to the structural line, and their connection direction matches the structural line's strike well, the structure may have a secondary impact on sand body connectivity and groundwater migration between them. Through this processing, structural line data not only participates in calculations as node attributes but also as part of the node relationships in subsequent channel generation.

[0088] After obtaining the stratigraphic consistency compatibility, sand body connectivity compatibility, hydrological orientation compatibility, and tectonic auxiliary compatibility, the system generates ore-forming candidate channel coefficients between predicted units. The calculation formula for the ore-forming candidate channel coefficients is as follows:

[0089] in, Indicates the first The prediction unit and the first The coefficient of mineralization candidate channels between prediction units; Indicates the first The prediction unit and the first Stratigraphic consistency compatibility among prediction units; Indicates the first The prediction unit and the first Sand body connectivity compatibility between prediction units; Indicates the first The prediction unit to the first Hydrological orientation compatibility of each prediction unit; Indicates the first The prediction unit and the first Construction-aided compatibility between prediction units; This represents the auxiliary influence coefficient for construction.

[0090] This formula expresses the basic conditions for mineralization candidate channels by assuming stratigraphic consistency, sandbody connectivity, and hydrological orientation, and uses tectonic factors as auxiliary reinforcing relationships. It is easy to understand that if any one of the key conditions in stratigraphy, sandbody, and hydrology is clearly not met by two prediction units, then the node relationship should not be identified as a high-level mineralization candidate channel; if the above key conditions are simultaneously good, and the tectonic relationship has an auxiliary effect, then the node relationship can be identified as a low-control zone mineralization candidate channel.

[0091] In this embodiment of the invention, ore-forming candidate channels whose coefficients satisfy preset channel conditions and whose node relationships pass through the low-control unit set are determined as ore-forming candidate channels in the low-control area. For ore-forming candidate channels with large spatial lengths, transition map units are also generated based on the spatial length of the ore-forming candidate channel in the low-control area, the spacing between expected map units, the node characteristics of the predicted units at both ends of the channel, and the channel relationship characteristics. The formula for calculating the number of transition map units is as follows:

[0092] in, Indicates the first The prediction unit and the first The number of transition graph units that need to be generated between prediction units; Indicates rounding up; Indicates the first The prediction unit and the first The distance between the centers of each prediction unit; This indicates the desired cell spacing in the diagram.

[0093] No. The eigenvectors of each transition graph unit can be represented as:

[0094] in, Indicates the first The prediction unit and the first Between the prediction units The feature vectors of each transition graph unit; Indicates the sequence number of the transition diagram unit; Indicates the first The relative positions of each transition diagram unit on the corresponding channel; Indicates the first The node feature vector of each prediction unit; Indicates the first The node feature vector of each prediction unit; Indicates the coefficients added to the channel features; Represents the feature mapping matrix of channel relationships; It represents the channel relationship feature vector composed of mineralization candidate channel coefficient, stratigraphic consistency compatibility, sand body connectivity compatibility, hydrological orientation compatibility, and tectonic auxiliary compatibility.

[0095] It should be noted that transition map units are not new mineral occurrences, nor are they simple interpolations of mineralization results. Rather, they are used to represent geologically based connection paths within low-control areas in the map structure. The characteristics of transition map units come from the predicted units at both ends of the channel and the relationship characteristics of the channel itself, thus preserving both the geological attributes at both ends and the mineralization implications of the channel itself. In areas with dense ordinary boreholes, there are already many direct adjacency relationships between predicted units, so transition map units are not necessarily needed. However, in areas covered by sand dunes and Gobi deserts with sparse boreholes, transition map units can alleviate the problem of broken relationships between working zones.

[0096] Finally, the transition graph unit is added to the node set corresponding to the original prediction unit to form the updated node set:

[0097] in, This indicates updating the set of nodes; This represents the set of graph nodes corresponding to the original prediction unit; Represents a set of transition diagram units; This represents the union operation of sets.

[0098] In this embodiment of the invention, the updated node set includes both graph nodes corresponding to the original prediction units and transition graph units used to represent mineralization candidate channels in low-control areas. It should be noted that the transition graph units participate in subsequent graph structure calculations but are not output as final independent prospective blocks. This enhances the representation of relationships in low-control areas while ensuring that the final results still correspond to prediction units within the actual exploration area.

[0099] S40: Based on the updated node set, construct a complete edge set consisting of spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on the complete edge set, generate graph structure data containing node features and edge relationship features.

[0100] Specifically, based on the spatial contact relationships, boundary adjacency relationships, and distance relationships between nodes in the updated node set, a spatial adjacency edge set is generated; a sand body connected edge set is generated based on sand body connectivity compatibility, a hydrological direction edge set is generated based on hydrological direction compatibility, a tectonic influence edge set is generated based on tectonic auxiliary compatibility, and a mineralization candidate channel edge set is generated based on mineralization candidate channel coefficients; the spatial adjacency edge set, sand body connected edge set, hydrological direction edge set, tectonic influence edge set, and mineralization candidate channel edge set are merged to form a complete edge set; for any edge in the complete edge set, edge relationship features and edge comprehensive strength are generated based on spatial adjacency relationships, sand body connectivity compatibility, hydrological direction compatibility, tectonic auxiliary compatibility, and mineralization candidate channel coefficients; based on the updated node set, the complete edge set, the node feature matrix, the edge relationship features, and the edge comprehensive strength, graph structure data for graph attention processing is generated.

[0101] In this embodiment of the invention, step S40 is used to convert the node relationships in the updated node set into edge relationships in the graph structure, and generate graph structure data containing node features and edge relationship features. It should be noted that traditional prospective area prediction methods usually only consider the factor values ​​of the prediction unit itself or the planar adjacency relationship, while this embodiment of the invention further constructs spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges, so that multiple geological connections between prediction units can be incorporated into the graph attention processing process.

[0102] First, based on the spatial contact relationships, boundary adjacency relationships, and distance relationships between nodes in the updated node set, a set of spatial adjacency edges is generated. Spatial adjacency edges are used to express the local spatial continuity between prediction units and are the basic connection relationships in the graph structure. In practical applications, spatial adjacency edges can be generated if two prediction units share a boundary, have a spatial contact relationship, or the distance between them is less than a preset adjacency distance. It should be noted that spatial adjacency edges only express basic spatial connections and cannot fully represent the mineralization connections of sandstone-type uranium deposits; therefore, other types of edges need to be constructed further.

[0103] In one implementation, a set of sand body connectivity edges can be generated based on sand body connectivity compatibility. If the prediction units corresponding to two nodes have high sand body connectivity compatibility, a sand body connectivity edge is generated between the two nodes. Specifically, sand body connectivity edges are used to express the connections between prediction units within the same sand body zone that share the same sand body distribution direction and have a gradual change in sand body thickness. It is easy to understand that in sandstone-type uranium deposits, node relationships extending along the sand body distribution direction may be more significant for mineralization than simple spatial adjacency relationships. Therefore, sand body connectivity edges can enable subsequent processing to aggregate information along favorable sand body directions.

[0104] In one implementation, a set of hydrological directional edges can be generated based on hydrological directional compatibility. These hydrological directional edges represent the transmission relationships of mineralization information along the direction of groundwater migration. Unlike spatial adjacency edges, hydrological directional edges have directional significance, reflecting the possible connections of uranium-bearing fluids migrating from relatively high hydraulic potential areas to relatively low hydraulic potential areas. In low-control areas, hydrological directional edges can prevent the model from extending prospective information in directions inconsistent with groundwater migration, thereby improving the geological plausibility of prospective area predictions.

[0105] In one implementation, a set of structural influence edges can be generated based on structural auxiliary compatibility. Structural influence edges are used to express the auxiliary role of structural conditions in the mineralization connections between nodes. Structural influence edges can be generated if two predicted units are located in the same structural influence zone, or if the connection direction between them has a good match with the direction of the structural line. It should be noted that structural influence edges do not indicate that the structure alone determines mineralization, but rather that the structure may change the connectivity of sand bodies and the migration path of groundwater. Therefore, they should be included in the calculation as edge relationship features in subsequent graph attention processing.

[0106] In one implementation, a set of candidate mineralization channels can be generated based on the coefficients of the candidate mineralization channels. These candidate channels are primarily used to represent candidate mineralization connections within low-control areas that are supported by stratigraphy, sand bodies, hydrology, and tectonics. For the transition map units generated in step three, the system sequentially connects the channel start point, transition map unit, and channel end point according to the channel direction, creating a continuous relationship between channels in the low-control area within the map structure. This process ensures that large blank areas between operational channels no longer appear as fracture areas in the map structure.

[0107] The complete set of edges can be represented as:

[0108] in, Represents the set of all edges; Represents the set of spatially adjacent edges; Represents the set of connected edges of a sand body; Represents the set of hydrological directional edges; This indicates the construction of the set of affected edges; Represents the set of candidate mineralization pathway edges; This represents the union operation of sets.

[0109] After generating the complete edge set, for any edge in the complete edge set, an edge relationship feature vector is generated based on spatial adjacency, sand body connectivity compatibility, hydrological orientation compatibility, tectonic auxiliary compatibility, and mineralization candidate channel coefficients:

[0110] in, Indicates the first The node and the first Feature vectors of edge relationships between nodes; Indicates the first The node and the first The spatial adjacency value between nodes is set to 1 if a spatial adjacency exists, and 0 otherwise. Indicates the first The prediction unit and the first Sand body connectivity compatibility between prediction units; Indicates the first The prediction unit to the first Hydrological orientation compatibility of each prediction unit; Indicates the first The prediction unit and the first Construction-aided compatibility between prediction units; Indicates the first The prediction unit and the first The coefficient of mineralization candidate channels between prediction units.

[0111] It should be noted that the edge relationship feature vector does not only record whether two nodes are adjacent, but also records multiple mineralization relationship components. Through this processing, the subsequent graph attention process can distinguish whether an edge is mainly formed by spatial adjacency, by sand body connectivity, by hydrological direction, or by mineralization candidate channels in low-control areas, thereby improving the interpretability of the prediction results.

[0112] Furthermore, the edge comprehensive strength is calculated based on the edge relationship characteristics. The formula for calculating the edge comprehensive strength is as follows:

[0113] in, Indicates the first The node and the first The overall edge strength between nodes; Indicates the strength coefficient corresponding to spatial adjacency; This represents the strength coefficient corresponding to the connectivity compatibility of the sand body; This represents the intensity coefficient corresponding to hydrological compatibility. This represents the strength coefficient corresponding to the auxiliary compatibility of the structure; This represents the intensity coefficient corresponding to the candidate mineralization pathway coefficient.

[0114] In this embodiment of the invention, different types of edge relationships are uniformly converted into edge comprehensive strength. The higher the edge comprehensive strength, the stronger the geological connection between the two nodes, and the greater its influence should be in the subsequent node feature aggregation process. Conversely, if two nodes are spatially close but have weak sand body, hydrological, and tectonic relationships, the edge comprehensive strength is low, and its influence on subsequent calculations is correspondingly reduced.

[0115] Finally, graph structure data is generated based on the updated node set, the complete edge set, the node feature matrix, and the edge relationship feature set:

[0116] in, Represents graph structure data; This indicates updating the set of nodes; Represents the set of all edges; This indicates updating the node feature matrix corresponding to the node set; It represents the set of edge relation features composed of all edge relation feature vectors.

[0117] In this embodiment of the invention, the graph structure data is the direct input for subsequent node feature aggregation processing based on edge relationship features and attention coefficients. Through step S40, the prediction of sandstone-type uranium deposit prospect areas is transformed from the traditional planar factor superposition to graph structure calculation jointly expressed by nodes and edges, so that the conditions of the prediction unit itself and the mineralization relationship between prediction units can be uniformly processed.

[0118] S50: Perform node feature aggregation processing based on edge relationship features and attention coefficients on the graph structure data, and update the graph attention processing parameters based on the verification label set to generate the prospect probability, confidence level, prospect level and standardized prediction results of each prediction unit.

[0119] Specifically, feature transformation processing is performed on the input representation of each node in the graph structure data to generate corresponding node transformation representations; based on the node transformation representation of the current node, the node transformation representations of neighboring nodes, the edge relationship features between the current node and neighboring nodes, the ore-forming candidate channel coefficient, the degree of control, and the edge comprehensive strength, the original attention score of the neighboring nodes to the current node is generated; normalization processing is performed on the original attention scores of all neighboring nodes corresponding to the same current node to generate the attention coefficient of the neighboring nodes to the current node; based on the attention coefficient and the edge comprehensive strength, weighted aggregation processing is performed on the node transformation representation of the neighboring nodes to generate the next layer node representation of the current node; based on the final layer node representation, probability transformation is performed to generate the prospect probability corresponding to the original prediction unit; wherein, the transition graph unit participates in the node feature aggregation processing but is not output as an independent prediction block.

[0120] Based on this, the following calculations are performed: First, based on the set of verification labels and the prospective probability, a verification label error is calculated to characterize the classification differences of verified prediction units. Second, based on the set of mineralization candidate channel edges and the prospective probability, a channel consistency error is calculated to characterize the continuity of prediction results at both ends of a mineralization candidate channel. Third, based on stratigraphic suitability, sand body favorability, tectonic influence value, and hydrological migration value, a theoretical mineralization score is generated, and a ranking error is calculated based on the relative differences between theoretical mineralization scores. Fourth, based on the set of spatially adjacent edges, edge comprehensive strength, and prospective probability, a spatial edge balance error is calculated to characterize the reasonable balance of prediction results between adjacent prediction units. Fifth, based on the verification label error, channel consistency error, ranking error, and spatial edge balance error, a comprehensive training error is generated, and the graph attention processing parameters are updated based on the comprehensive training error.

[0121] Following this, based on the updated graph attention processing parameters, prospect probability calculation is performed on the graph structure data, and the prospect probability corresponding to the original prediction unit is extracted from the calculation results; the attention concentration of each prediction unit is calculated based on the attention coefficient, and the mineralization relationship support of each prediction unit is calculated based on the edge comprehensive strength; the credibility of each prediction unit is generated based on the degree of control, attention concentration, and mineralization relationship support, and a comprehensive prospect score of each prediction unit is generated based on the prospect probability and credibility; the prospect level of each prediction unit is divided based on the comprehensive prospect score, and adjacent prediction units are merged according to the prospect level, comprehensive prospect score, credibility, and edge comprehensive strength to generate prospect boundaries; the mineralization contribution ratio of each prospect block is calculated based on the attention coefficient and the edge relationship characteristics, and a standardized prediction result including the prospect boundary, prospect level, credibility map, contribution factor map, and prediction result table is output.

[0122] In this embodiment of the invention, step S50 is used to convert the graph structure data into the final prospect prediction result. This step includes node feature aggregation, graph attention processing parameter update, prospect probability calculation, confidence calculation, prospect level classification, and standardized result output. It should be noted that the node feature aggregation process in this embodiment of the invention not only uses the node's own features, but also uses edge relationship features, mineralization candidate channel coefficients, control degree, and edge comprehensive strength, so that the prospect probability calculation can reflect the mineralization relationship of sandstone-type uranium deposits.

[0123] In this embodiment of the invention, feature transformation processing is first performed on the input representation of each node in the graph structure data to generate the corresponding node transformation representation.

[0124] No. The node transformation of a layer can be represented as:

[0125] in, Indicates the first The node at the th Layer node transformation representation; Indicates the first The feature transformation matrix of the layer; Indicates the first The node at the th The layer's input representation; This indicates the sequence number of the attention processing layer in the graph.

[0126] In this embodiment of the invention, nodal features with different meanings, such as stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value, and degree of control, are transformed into a unified computational space. After feature transformation, different ore-forming elements can jointly participate in the subsequent attention coefficient calculation, avoiding the impact of different feature dimensions on computational stability.

[0127] Subsequently, based on the node transformation representation of the current node, the node transformation representation of neighboring nodes, the edge relationship characteristics between the current node and neighboring nodes, the ore-forming candidate channel coefficient, the degree of control, and the edge comprehensive strength, the original attention score of the neighboring nodes to the current node is generated. The original attention score can be expressed as:

[0128] in, Indicates the first The first in the layer The neighboring nodes of the nth pair The original attention score of each node; Represents a linear rectified function with leakage; Indicates the first The attention parameter vector of the layer; Indicates the transpose operation; This represents the vector concatenation operation; Indicates the first The node at the th Layer node transformation representation; Indicates the first The neighboring nodes at the th Layer node transformation representation; Indicates the first Layer edge relationship feature mapping matrix; Indicates the first The node and the first Feature vectors of edge relationships between nodes; Indicates the parameter affecting the candidate mineralization pathway coefficient; Indicates the first The node and the first Coefficient of ore-forming candidate channels between nodes; Indicates the first The degree of control of each node corresponding to the prediction unit; Indicates the parameters affecting the overall strength of the edge; Indicates the first The node and the first The overall edge strength between nodes.

[0129] It should be noted that the contribution of neighboring nodes to the current node depends not only on the characteristics of the two nodes but also on whether there is mineralization significance between them. Particularly in low-control areas, where the degree of control is low, the influence of the mineralization candidate channel coefficient on the original attention score is enhanced, allowing the system to make greater use of mineralization candidate channels in low-control areas for judgment. In areas with a higher degree of control, the system can rely more on existing verification information and local edge relationships for judgment. In this way, the processing can take into account the different data characteristics of areas with sufficient control and areas with low control.

[0130] Normalize the raw attention scores of all neighboring nodes corresponding to the same current node to generate the attention coefficients of the neighboring nodes to the current node:

[0131] in, Indicates the first The first in the layer The neighboring nodes of the nth pair Attention coefficient of each node; Represents the natural exponential function; Indicates the first The first in the layer The neighboring nodes of the nth pair The original attention score of each node; Indicates the first The node number within the neighborhood of each node; Indicates the first The set of neighboring nodes of a node; Indicates the first The first in the layer The neighboring nodes of the nth pair The original attention score of each node.

[0132] It should be noted that converting the contributions of different neighboring nodes to the current node into relative proportions allows for a unified comparison of spatial adjacency relationships, sand body connectivity relationships, hydrological direction relationships, tectonic influence relationships, and mineralization candidate channel relationships around the same node. Unlike manually fixed weights, the attention coefficient is generated by the combined influence of node features, edge relationship features, and validation labels, making it more suitable for handling sandstone-type uranium deposit prediction problems with complex mineralization relationships.

[0133] After obtaining the attention coefficients, a weighted aggregation process is performed on the node transformation representations of neighboring nodes based on the attention coefficients and edge synthesis strength to generate the next-level node representation of the current node:

[0134] in, Indicates the first The node at the th Layer node representation; Represents a non-linear activation function; Indicates the neighboring node number; Indicates the first The set of neighboring nodes of a node; Indicates the first The first in the layer The neighboring nodes of the nth pair Attention coefficient of each node; Indicates the first The node and the first The overall edge strength between nodes; Indicates the first The neighboring nodes at the th Layer node transformation representation.

[0135] The technical principle behind this aggregation process is that the attention coefficient represents the relative contribution of neighboring nodes, while the edge integration strength represents the strength of the geological relationship between two nodes. Together, these two factors allow neighboring nodes with strong geological relationships and greater contribution to the current node's prediction to play a larger role in the aggregation process; conversely, neighboring nodes with weak geological relationships, even if spatially close, will have a correspondingly reduced influence.

[0136] In one executable implementation, multiple sets of attention results can be generated in parallel, and then concatenated or averaged to improve computational stability. The expression for concatenating multiple sets of attention results is as follows:

[0137] in, Indicates the first The node at the th Layer node representation; Indicates the number of parallel attention results; Indicates the number of the parallel attention result; This indicates a concatenation operation; Represents a non-linear activation function; Indicates the neighboring node number; Indicates the first The set of neighboring nodes of a node; Indicates the first In the group attention calculation, the first The neighboring nodes of the nth pair Attention coefficient of each node; Indicates the first The node and the first The overall edge strength between nodes; Indicates the first In the group attention calculation, the first The neighboring nodes at the th Layer node transformation representation.

[0138] Furthermore, multiple sets of attention results can learn the contributions of neighboring nodes from different mineralization relationships. For example, one set of results can better reflect sand body connectivity, another set can better reflect hydrological direction relationships, and yet another set can better reflect tectonic auxiliary relationships. By splicing or averaging, the dependence of a single calculation result on individual validation points or local relationships can be reduced, thereby improving the stability of long-term predictions in low-control areas.

[0139] After several layers of node feature aggregation processing, a probability transformation is performed based on the final layer node representation to generate the prospect probability corresponding to the original prediction unit:

[0140] in, Indicates the first The prospect probability of each prediction unit; This indicates the normalization exponent operation; Indicates the output transformation matrix; Indicates the first Each node is in the final layer. The node representation; This represents the output bias vector; the subscript 1 indicates the probability component corresponding to the favorable mineralization category.

[0141] The normalized exponent operation can be expressed as:

[0142] in, Representing vectors In the Normalized probability on class; This represents the output score vector; Indicates the first Output score corresponding to the class; Indicates the category number; Indicates the total number of categories; This represents the natural exponential function.

[0143] It should be noted that transition graph units participate in node feature aggregation processing, but are not output as independent prediction blocks. In other words, transition graph units are used to enhance the expression of mineralization relationships and information transmission in low-control areas, and the final result still uses the original prediction units as the basic output object.

[0144] In this embodiment of the invention, the graph attention processing parameters are also updated based on the verification label set. Specifically, the verification label error is calculated based on the verification label set and the prospect probability, which is used to characterize the classification differences of the verified prediction units.

[0145] in, This indicates the error in verifying the label. Represents the set of node indexes with verification labels; Indicates the node number; The number indicates the category number, with zero representing the control category and one representing the favorable mineralization category. Indicate category Corresponding category weights; Indicates the first Each node in the category Verification label on; Indicates the first Each node belongs to the category The predicted probability; It represents logarithmic operations.

[0146] The formula for calculating category weights is as follows:

[0147] in, Indicate category Corresponding category weights; This represents the total number of nodes with verification labels; This indicates a positive number that prevents the denominator from being zero.

[0148] The aforementioned category weights are used to alleviate the problem of a limited number of favorable mineralization samples. In sandstone-type uranium deposit exploration, the number of favorable mineralization samples is usually far fewer than the control samples. Without processing, the prospect probability is prone to bias towards conservative results. By using category weights, the role of a small number of favorable mineralization samples in updating graph attention processing parameters can be enhanced.

[0149] Based on the edge set of potential mineralization channels and prospective probabilities, the channel consistency error is calculated:

[0150] in, Indicates channel consistency error; Indicates the first The node and the first The mineralization candidate channel edge formed by the nodes; Represents the set of candidate mineralization pathway edges; Indicates the first The node and the first Coefficient of ore-forming candidate channels between nodes; Indicates the first The prospect probability of each node; Indicates the first The prospect probability of each node.

[0151] This error is used to ensure reasonable continuity between the prediction results at both ends of the mineralization candidate channel. When two prediction units have a high mineralization candidate channel coefficient, they have a strong correlation in terms of strata, sand bodies, hydrology, and tectonic relationships, and their prospective probabilities should not show unfounded and drastic differences.

[0152] Based on stratigraphic suitability, sand body favorableness, tectonic influence, and hydrological transport, a theoretical mineralization score is generated:

[0153] in, Indicates the first Theoretical mineralization score for each node; The theoretical weights corresponding to formation suitability; The theoretical weights corresponding to the favorable properties of sand bodies; This represents the theoretical weight corresponding to the constructed influence value; This represents the theoretical weight corresponding to the hydrological transport value; Indicates the first Each node corresponds to the stratigraphic suitability of the prediction unit; Indicates the first The favorable conditions of sand bodies for each node in the prediction unit; Indicates the first The construction impact value of each node corresponds to the prediction unit; Indicates the first Each node corresponds to a hydrological migration value for a prediction unit.

[0154] Further, the ranking error is calculated based on the relative differences between theoretical mineralization scores:

[0155] in, Indicates sorting error; This represents the set of node pairs whose theoretical mineralization scores differ and meet preset conditions. This indicates the operation of finding the maximum value. This represents the difference in expected probabilities; Indicates the first The prospect probability of each node; Indicates the first The prospect probability of each node.

[0156] This ranking error is used to ensure that the prospect probability is consistent with the mineralization theory of sandstone-type uranium deposits. When the theoretical mineralization score of one prediction unit is significantly higher than that of another, the prospect probability should also reflect this advantage, thereby reducing the impact of uneven distribution of a small number of validation points on the prediction results of low-control areas.

[0157] Based on the spatial adjacency edge set, edge comprehensive strength, and prospect probability, the spatial edge balance error is calculated:

[0158] in, Indicates the spatial edge balance error; Indicates the first The node and the first Spatial adjacency edges formed by nodes; Represents the set of spatially adjacent edges; Indicates the first The node and the first The overall edge strength between nodes; Indicates the first The prospect probability of each node; Indicates the first The prospect probability of each node.

[0159] This error is used to improve the local spatial continuity of the prediction results, but it is adjusted by the edge integration strength. If the geological relationship between two spatially adjacent prediction units is strong, their prospect probabilities should be kept reasonably continuous; if the geological relationship between two spatially adjacent prediction units is weak, their results should not be forced to be close.

[0160] The formula for calculating the overall training error is as follows:

[0161] in, Indicates the overall training error; This indicates the error in verifying the label. Indicates channel consistency error; Indicates sorting error; Indicates the spatial edge balance error; This represents the error adjustment coefficient corresponding to the channel consistency error; This represents the error adjustment coefficient corresponding to the sorting error; This represents the error adjustment coefficient corresponding to the spatial edge balance error.

[0162] After updating the graph attention processing parameters based on the comprehensive training error, the system performs prospect probability calculation on the graph structure data and extracts the prospect probability corresponding to the original prediction unit from the calculation results. To further improve the interpretability of the results, this embodiment of the invention calculates the attention concentration of each prediction unit based on the attention coefficient. The formula for calculating the attention concentration is as follows:

[0163] in, Indicates the first Attention concentration of each prediction unit; Indicates the neighboring node number; Indicates the first The set of neighboring nodes corresponding to each prediction unit node; Indicates the first The neighboring nodes of the nth pair Attention coefficients of nodes corresponding to each prediction unit; Represents logarithmic operations; This indicates the number of neighboring nodes.

[0164] In this embodiment of the invention, attention concentration is used to express whether the basis for prediction is concentrated. When the attention of a prediction unit is mainly concentrated on a few neighboring nodes with clear mineralization relationships, it indicates that the basis for prediction of the prediction unit is relatively clear; when the attention is relatively dispersed, it indicates that the prediction result is affected by multiple relationships, and its explanatory concentration is relatively weak.

[0165] The ore-forming relationship support of each prediction unit is calculated based on the edge comprehensive strength. The formula for calculating the ore-forming relationship support is as follows:

[0166] in, Indicates the first Support of mineralization relationships for each prediction unit; Indicates the relationship with the first The set of edges connected to the nodes corresponding to each prediction unit; This indicates the number of edges in the edge set; Indicates the first The node and the first Edges between nodes; Indicates the first The node and the first The overall edge strength between nodes.

[0167] Metallogenic relationship support is used to express whether a prediction unit is in a favorable relationship network. If a prediction unit has strong sand body connectivity, hydrological direction relationship or metallogenic candidate channel relationship with its surrounding nodes, its prediction results are more regionally continuous.

[0168] Furthermore, the confidence level of each prediction unit is generated based on the degree of control, the degree of attention concentration, and the support of the mineralization relationship. The confidence level is calculated as follows:

[0169] in, Indicates the first The credibility of each prediction unit; This indicates the minimum value operation; The confidence coefficient represents the degree of control. This represents the credibility coefficient corresponding to the level of attention concentration. This represents the confidence coefficient corresponding to the support of the mineralization relationship; Indicates the first The degree of control over each prediction unit; Indicates the first Attention concentration of each prediction unit; Indicates the first Support of mineralization relationships for each prediction unit.

[0170] It is important to note that confidence level differs from prospect probability. Prospect probability indicates the likelihood that a predicted unit belongs to a favorable prospective mineralization area, while confidence level indicates the strength of the evidence supporting the prediction result. Through confidence level output, exploration personnel can distinguish between areas with "high prospect probability and sufficient evidence" and areas with "relatively high prospect probability but requiring further verification."

[0171] A comprehensive vision score is generated based on vision probability and credibility. The formula for calculating the comprehensive vision score is as follows:

[0172] in, Indicates the first The overall prospect score of each prediction unit; Indicates the first The prospect probability of each prediction unit; Indicates the first The formula is based primarily on prospective probabilities, while incorporating confidence levels for adjustment. This ensures that potential favorable areas are not completely suppressed due to insufficient confidence in low control areas, and also reflects the strength of the prediction basis when ranking prospective areas.

[0173] Based on the comprehensive vision score, the vision level of each prediction unit is divided into different levels. The vision level division can be expressed as follows:

[0174] in, Indicates the first The prospect level of each prediction unit; Indicates a first-level remote area; Indicates a secondary scenic area; Indicates a third-level remote area; Indicates the general area; Indicates the first The overall prospect score of each prediction unit; This indicates the threshold for classifying primary remote scenic areas; This indicates the threshold for classifying secondary remote areas; This indicates the threshold for classifying remote scenic areas into three levels.

[0175] In practical applications, the threshold for classifying the prospective assessment level can be a fixed threshold or a quantile threshold based on the distribution of the comprehensive prospective assessment score within the exploration area. The above threshold selection method should not be construed as a limitation of this invention.

[0176] In this embodiment of the invention, adjacent prediction units are merged based on prospect level, comprehensive prospect score, confidence level, and edge integration strength to generate prospect boundary areas. It should be noted that block merging considers not only spatial adjacency but also edge integration strength and the relationship with mineralization candidate channels. This avoids simply merging spatially adjacent but geologically weak prediction units and allows favorable prediction units located on the same mineralization candidate channel to form continuous prospect blocks.

[0177] For each prospective block, the system can further calculate the average comprehensive prospect score and average confidence level. The average comprehensive prospect score can be expressed as:

[0178] in, Indicates the first The average overall prospect score of each prospective area; Indicates the first One remote scenic area; Indicates the first Number of prediction units within each prospective area block; Indicates the first The first in the remote area One prediction unit; Indicates the first The overall prospect score of each prediction unit.

[0179] Average credibility can be expressed as:

[0180] in, Indicates the first Average credibility of each remote area block; Indicates the first One remote scenic area; Indicates the first Number of prediction units within each prospective area block; Indicates the first The first in the remote area One prediction unit; Indicates the first The credibility of each prediction unit.

[0181] To produce interpretable results, this embodiment of the invention calculates the mineralization contribution ratio of each prospective block based on the attention coefficient and edge relationship characteristics. The formula for calculating the contribution ratio is as follows:

[0182] in, Indicates the first The first of the remote scenic blocks The contribution ratio of mineralization-like relationships; Indicates the remote block number; Indicates the type of mineralization relationship; Indicates the first Distant view area; Indicates the first [unclear] within the distant view block One prediction unit; Indicates the first The set of neighboring nodes corresponding to each prediction unit node; Indicates the first The neighboring nodes of the nth pair Attention coefficients of nodes corresponding to each prediction unit; The first edge in the overall strength is represented by the second edge. Intensity components corresponding to mineralization relationships; Indicates the first The node and the first The overall edge strength between nodes.

[0183] By calculating the contribution ratio, the system can output which type of relationship—sand body connectivity, hydrological direction, tectonic influence, or mineralization candidate channel—mainly supports a prospective block. For example, if the main contribution of a prospective block comes from sand body connectivity and hydrological direction, it indicates that the prospective block is mainly distributed along favorable sand bodies and groundwater migration directions. If the main contribution of a prospective block comes from mineralization candidate channels and tectonic influence, it indicates that although the prospective block is located in a low-control zone, it is supported by both low-control zone channel relationships and tectonic auxiliary relationships.

[0184] Finally, the system outputs standardized prediction results including the prospective area boundary, prospect level, confidence map, contribution factor map, and prediction result table. The prediction result table may include block number, block area, prospect level, average comprehensive prospect score, highest comprehensive prospect score, average confidence, main contribution factors, whether it is located in a low control area, and suggested verification order. It should be noted that the above result fields are only one optional example. In practical applications, the field content can be added or adjusted according to the results submission requirements of the exploration unit. As long as it is generated based on the data processing results of this invention, it belongs to the optional implementation of this invention.

[0185] like Figure 2 As shown, this invention provides a system for predicting sandstone-type uranium deposit prospects based on GAT. The system is used to execute the aforementioned method for predicting sandstone-type uranium deposit prospects based on GAT. The system includes: The unit division is used to obtain the geological basic data of the exploration area. Multiple prediction units are divided according to the boundary of the exploration area, and each prediction unit is converted into the corresponding basic data of the map node. The generation unit is used to perform ore-forming element numerical processing on each prediction unit based on the graph node basic data, generate a node feature matrix including stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value and control degree, and generate a set of verification labels based on mineralization verification point data. The calculation unit is used to perform mineralization relationship compatibility calculation on the low-control prediction unit based on the node feature matrix and the degree of control, and generate mineralization candidate channels, transition map units and updated node sets in the low-control area. The construction unit is used to construct a complete edge set based on the updated node set, including spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on the complete edge set, graph structure data containing node features and edge relationship features is generated. The update unit is used to perform node feature aggregation processing based on edge relationship features and attention coefficients on the graph structure data, and update the graph attention processing parameters based on the verification label set to generate the prospect probability, confidence level, prospect level and standardized prediction results of each prediction unit.

[0186] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0187] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0188] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for predicting sandstone-type uranium deposit prospects based on GAT, characterized in that, The method includes: Obtain basic geological data of the exploration area, divide the exploration area into multiple prediction units according to the boundary of the exploration area, and convert each prediction unit into corresponding basic data of map nodes; Based on the graph node basic data, the metallogenic elements of each prediction unit are numerically processed to generate a node feature matrix that includes stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value and degree of control, and a set of verification labels is generated based on the mineralization verification point data. Based on the node feature matrix and the degree of control, mineralization relationship compatibility calculation is performed on the low-control prediction unit to generate mineralization candidate channels, transition map units and updated node sets in the low-control area. Based on the updated node set, a complete edge set is constructed, consisting of spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on the complete edge set, graph structure data containing node features and edge relationship features is generated. The graph structure data is subjected to node feature aggregation processing based on edge relationship features and attention coefficients, and the graph attention processing parameters are updated based on the verification label set to generate the prospect probability, confidence level, prospect level and standardized prediction results of each prediction unit.

2. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 1, characterized in that, Obtain basic geological data of the exploration area, divide the exploration area into multiple prediction units according to the boundaries of the exploration area, and convert each prediction unit into corresponding map node basic data, including: The geological baseline data and boundary data of the exploration area are acquired, and basic prediction units covering the exploration area are generated based on the boundary data. The geological baseline data includes stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data. Based on the stratigraphic boundaries in the stratigraphic unit data, the sand body boundaries in the sand body distribution data, and the structural line positions in the structural line data, geological boundary correction processing is performed on the basic prediction unit to generate multiple prediction units with consistent geological attributes. Each prediction unit is converted into a graph node, and each graph node is configured with its corresponding spatial location, unit area, and node number. Stratigraphic unit data, sand body distribution data, structural line data, hydrogeological condition data, and mineralization verification point data are mapped to corresponding map nodes to generate basic map node data.

3. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 1, characterized in that, Based on the aforementioned graph node data, ore-forming element numerical processing is performed on each prediction unit to generate a node feature matrix containing stratigraphic suitability, sand body favorability, tectonic influence values, hydrological migration values, and degree of control. Furthermore, a set of verification labels is generated based on mineralization verification point data, including: Based on the stratigraphic unit data of the graph node basic data, the stratigraphic position, lithological combination and stratigraphic combination type of each prediction unit are determined, and the stratigraphic suitability of each prediction unit is generated according to the mineralization suitability corresponding to different stratigraphic combination types. Based on the sand body distribution data of the graph node basic data, the sand body thickness, sand body continuous length and sand body distribution stability of each prediction unit are extracted, and the sand body thickness, sand body continuous length and sand body distribution stability are converted into the sand body advantage of each prediction unit. Based on the construction line data of the graph node basic data, the directional relationship between the distance from each prediction unit to the construction line, the construction level, the main distribution direction of the sand body and the direction of the construction line is extracted, and the construction influence value of each prediction unit is generated. Based on the hydrogeological conditions data of the graph node basic data, the main migration direction of groundwater, the hydraulic potential relationship and the spatial directional relationship between the prediction units are extracted, and the hydrogeological migration value of each prediction unit is generated. At the same time, the degree of control of each prediction unit is calculated based on the mineralization verification point data. The stratigraphic suitability, sand body favorableness, tectonic influence value, hydrological migration value, and degree of control are combined into a node feature matrix according to the prediction unit number. A set of verification labels is generated based on whether the mineralization verification point falls into the corresponding prediction unit and the verification status of the mineralization verification point.

4. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 1, characterized in that, Based on the node feature matrix and the degree of control, a mineralization relationship compatibility calculation is performed on the low-control prediction unit, including: Based on the control level of the node feature matrix and the preset control level threshold, low control prediction units are identified from multiple prediction units, and a set of low control units is formed. Based on the stratigraphic relationship and stratigraphic suitability of each prediction unit, the stratigraphic consistency compatibility between prediction units is calculated, and based on the sand body zone, sand body thickness difference and sand body distribution direction relationship of each prediction unit, the sand body connectivity compatibility between prediction units is calculated. Based on the spatial orientation, main groundwater migration direction and hydraulic potential relationship between each prediction unit, the hydrological orientation compatibility between prediction units is calculated. Based on the distance differences between each prediction unit and the construction line, as well as the matching relationship between the direction of the construction line and the connection direction of the prediction unit, the construction auxiliary compatibility between prediction units is calculated.

5. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 4, characterized in that, Generate mineralization candidate channels, transition map units, and updated node sets for low-control areas, including: Based on stratigraphic consistency compatibility, sand body connectivity compatibility, hydrological orientation compatibility, and tectonic auxiliary compatibility, mineralization candidate channel coefficients are generated between prediction units. The node relationships that satisfy the preset channel conditions and pass through the set of low control units are determined as the mineralization candidate channels in the low control area. Based on the spatial length of the mineralization candidate channel in the low control area, the spacing between expected map units, the node characteristics of the predicted units at both ends of the channel, and the channel relationship characteristics, a transition map unit is generated. The transition graph unit is added to the node set corresponding to the original prediction unit to form an updated node set, and the updated node set is transmitted to the graph structure data construction and processing.

6. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 1, characterized in that, Based on the updated node set, a complete edge set is constructed, consisting of spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on this complete edge set, graph structure data containing node features and edge relationship features is generated, including: Based on the spatial contact relationships, boundary adjacency relationships, and distance relationships between nodes in the updated node set, a spatial adjacency edge set is generated; A set of sand body connected edges is generated based on sand body connectivity compatibility, a set of hydrological direction edges is generated based on hydrological direction compatibility, a set of structural influence edges is generated based on structural auxiliary compatibility, and a set of mineralization candidate channel edges is generated based on mineralization candidate channel coefficient. The set of spatial adjacent edges, the set of sand body connected edges, the set of hydrological direction edges, the set of tectonic influence edges, and the set of mineralization candidate channels edges are merged to form the complete edge set; For any edge in the entire set of edges, edge relationship characteristics and edge comprehensive strength are generated based on spatial adjacency, sand body connectivity compatibility, hydrological direction compatibility, tectonic auxiliary compatibility and mineralization candidate channel coefficient. Based on the updated node set, all edge set, node feature matrix, edge relationship features, and edge comprehensive strength, graph structure data for graph attention processing is generated.

7. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 1, characterized in that, Perform node feature aggregation processing based on edge relationship features and attention coefficients on the graph structure data, including: The input representation of each node in the graph structure data is subjected to feature transformation processing to generate the corresponding node transformation representation; Based on the node transformation representation of the current node, the node transformation representation of the neighboring nodes, the edge relationship characteristics between the current node and the neighboring nodes, the ore-forming candidate channel coefficient, the degree of control and the edge comprehensive strength, the original attention score of the neighboring nodes to the current node is generated. Normalize the raw attention scores of all neighboring nodes corresponding to the same current node to generate the attention coefficients of the neighboring nodes to the current node. Based on the attention coefficient and edge synthesis strength, a weighted aggregation process is performed on the node transformation representation of the neighboring nodes to generate the next layer node representation of the current node. Based on the final layer node representation, a probability transformation is performed to generate the prospect probability corresponding to the original prediction unit; among them, the transition graph unit participates in the node feature aggregation process but is not output as an independent prediction block.

8. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 7, characterized in that, Update the graph attention processing parameters based on the set of verification labels, including: Based on the set of verification labels and the prospect probability, the verification label error used to characterize the classification differences of the verified prediction units is calculated. Based on the set of candidate mineralization channels and the prospective probability, calculate the channel consistency error used to characterize the continuity of prediction results at both ends of the candidate mineralization channel; Based on stratigraphic suitability, sand body favorability, tectonic influence value and hydrological migration value, a theoretical mineralization score is generated, and the ranking error is calculated based on the relative differences between the theoretical mineralization scores. Based on the spatial adjacent edge set, edge comprehensive strength and prospect probability, the spatial edge balance error is calculated to characterize the reasonable balance of prediction results of adjacent prediction units. A comprehensive training error is generated based on the verification label error, channel consistency error, sorting error, and spatial edge balance error, and the graph attention processing parameters are updated based on the comprehensive training error.

9. The method for predicting sandstone-type uranium deposit prospect areas based on GAT according to claim 8, characterized in that, Generate the prospect probability, confidence level, prospect level, and standardized prediction results for each prediction unit, specifically including: Based on the updated graph attention processing parameters, a prospect probability calculation is performed on the graph structure data, and the prospect probability corresponding to the original prediction unit is extracted from the calculation result. The attention concentration of each prediction unit is calculated based on the attention coefficient, and the mineralization relationship support of each prediction unit is calculated based on the edge comprehensive strength. Based on the degree of control, the degree of attention concentration, and the support of mineralization relationship, the credibility of each prediction unit is generated, and a comprehensive prospect score of each prediction unit is generated based on the prospect probability and credibility. Based on the comprehensive prospect score, the prospect level of each prediction unit is divided, and adjacent prediction units are merged according to the prospect level, comprehensive prospect score, confidence level and edge integration strength to generate the prospect boundary. The mineralization contribution ratio of each prospective area is calculated based on the attention coefficient and the edge relationship characteristics, and a standardized prediction result including the prospective area boundary, prospect level, confidence map, contribution factor map and prediction result table is output.

10. A system for predicting sandstone-type uranium deposit prospects based on GAT, characterized in that, The system is used to perform the method for predicting sandstone-type uranium deposit prospect areas based on GAT as described in any one of claims 1-9, the system comprising: The unit division is used to obtain the geological basic data of the exploration area. Multiple prediction units are divided according to the boundary of the exploration area, and each prediction unit is converted into the corresponding basic data of the map node. The generation unit is used to perform ore-forming element numerical processing on each prediction unit based on the graph node basic data, generate a node feature matrix including stratigraphic suitability, sand body favorability, tectonic influence value, hydrological migration value and control degree, and generate a set of verification labels based on mineralization verification point data. The calculation unit is used to perform mineralization relationship compatibility calculation on the low-control prediction unit based on the node feature matrix and the degree of control, and generate mineralization candidate channels, transition map units and updated node sets in the low-control area. The construction unit is used to construct a complete edge set based on the updated node set, including spatial adjacency edges, sand body connectivity edges, hydrological direction edges, tectonic influence edges, and mineralization candidate channel edges. Based on the complete edge set, graph structure data containing node features and edge relationship features is generated. The update unit is used to perform node feature aggregation processing based on edge relationship features and attention coefficients on the graph structure data, and update the graph attention processing parameters based on the verification label set to generate the prospect probability, confidence level, prospect level and standardized prediction results of each prediction unit.

Images