Metallization prospective area prediction system and method based on embedded knowledge reasoning

Abstract

The invention relates to the technical field of metallogenic prediction, in particular to a metallogenic prospective area prediction system and method based on embedded knowledge reasoning, and the system comprises a collection module which is used for collecting exploration data and geological map data; the data preparation module is used for cutting the exploration data into fishing net units; the knowledge graph construction and embedding module is used for constructing a knowledge graph database by utilizing the geological map data and embedding entities and relationships into a low-dimensional space; the knowledge-data fusion module is used for generating a knowledge-data fusion matrix of the corresponding fishing net unit for each fishing net unit; the model training and optimizing module is used for inputting the knowledge-data fusion matrix of the labeled fishing net unit into a deep learning network to construct a metallogenic prediction model; and the metallogenic prediction and result interpretation module is used for calculating the metallogenic probabilities of all the fishing net units in the research area by using the metallogenic prediction model. According to the method, knowledge graph embedding and exploration data are subjected to structured fusion in an input layer, so that the model not only learns a data mode, but also can understand a geological background.

Description

Technical Field

[0001] This invention relates to the field of mineralization prediction technology, specifically to a mineralization prospect prediction system and method based on embedded knowledge reasoning. Background Technology

[0002] Metallogenic prospect prediction is a core component of geological exploration, aiming to comprehensively utilize multi-source information to delineate prospecting targets and reduce exploration costs. Driven by artificial intelligence and big data technologies, metallogenic prospect prediction is undergoing a transformation from traditional models relying on expert experience to "data-driven" and "knowledge-data dual-driven" models. To meet the modern demands of deep mineral exploration and the prediction of concealed ore bodies, "data-driven" methods, represented by machine learning and deep learning, have been introduced into the field of metallogenic prediction.

[0003] In existing technologies, such as data-driven mineralization prospect prediction platforms, these platforms utilize machine learning to extract features, recognize patterns, and predict target areas from multi-source geoscientific data using chemical raster or evidence layer overlay methods. They have advantages such as high automation, strong adaptability, and ease of deployment, and are widely used in the early screening and prediction of mineral resources.

[0004] However, in the actual use of this prediction platform, during the feature engineering stage of the prediction process, it relies heavily on expert experience, is subjective, and ignores a wealth of valuable geological knowledge and metallogenic regularities contained in numerous geological maps, geological reports, and literature. The metallogenic probability distribution output by the model often lacks geological logic support and is difficult to explain "why the probability of metallogenesis in this area is high." This limitation of "emphasizing data and neglecting knowledge" causes the prediction accuracy of this platform to decrease in the prediction of complex metallogenic systems. Therefore, it is necessary to propose a metallogenic prospect prediction system and method based on embedded knowledge reasoning to solve the above problems. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a metallogenic prospect prediction system and method based on embedded knowledge reasoning. This system creates a knowledge-data fusion matrix for deep learning models, structurally aligning and fusing geological knowledge graph information with geochemical raster data. This enables the deep learning model to simultaneously learn geological patterns and data patterns, allowing it to not only learn data patterns but also understand the geological context. The prediction accuracy and AUC are superior to methods using only chemical raster or evidence layer overlay.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: a mineralization prospect prediction system based on embedded knowledge reasoning, comprising a data acquisition module, a data preparation module, a knowledge graph construction and embedding module, a knowledge-data fusion module, a model training and optimization module, and a mineralization prediction and result interpretation module.

[0007] The data acquisition module collects exploration data and geological map data within the study area, and marks positive and negative samples according to the mineral occurrence data; the geological map data includes geological entity data, geological attribute data, and geological relationship data.

[0008] The data preparation module is used to process the exploration data into evidence layers with a uniform format and spatial alignment, and to rasterize several superimposed evidence layers into several fishing net units; wherein, the fishing net unit is a multi-channel raster matrix.

[0009] The knowledge graph construction and embedding module is used to build a knowledge graph database using geological map data, and extract spatial relationship data between fishing net units and geological entity data through spatial analysis algorithms. The geological entity data and spatial relationship data are then transformed into knowledge embedding vectors through an embedding model.

[0010] The knowledge-data fusion module is used to extract the knowledge embedding vectors of the grid of the fishing net unit itself and its surrounding neighboring fishing net units, and integrate them into a knowledge embedding matrix.

[0011] The knowledge-data fusion module is also used to splice the knowledge embedding matrix with the grid of the corresponding fishing net unit according to the non-overlapping area to form the knowledge-data fusion matrix of the corresponding fishing net unit.

[0012] The model training and optimization module is used to input the knowledge-data fusion matrix of fishing net units into a deep learning network to build a mineralization prediction model.

[0013] The mineralization prediction and result interpretation module is used to calculate the mineralization probability of all fishing net units in the study area using the mineralization prediction model, generate prediction maps, and delineate target areas.

[0014] Furthermore, the data acquisition module includes geochemical element data.

[0015] Furthermore, in the knowledge graph construction and embedding module, the embedding model includes either the TransH model or the DistMult model.

[0016] Furthermore, in the knowledge-data fusion module, the knowledge-data fusion matrix includes the original exploration data characteristics and geological knowledge characteristics of the corresponding fishing net unit, as well as the geological knowledge characteristics of the surrounding adjacent fishing net units.

[0017] Furthermore, the knowledge-data fusion module also includes a tag identification unit. When the knowledge-data fusion matrix of a certain fishing net unit in the study area is located within a known mineralization area, the tag identification unit is used to mark the fishing net unit as mineralized. When a certain fishing net unit is far from the mineralization area and far from favorable mineralization structures, the tag identification unit is used to mark the fishing net unit as having a low probability of mineralization.

[0018] Furthermore, in the model training and optimization module, the deep learning network includes one or more of the following: CNN model, GNN model, and ViT model.

[0019] Furthermore, a method for predicting mineralized prospective areas based on embedded knowledge reasoning includes the following steps: Step 1, Data Preparation and Representation: Spatially align the evidence layers to generate multi-channel raster images. Cut the raster images into several fishing net units according to size. Each fishing net unit is a multi-channel raster matrix. Construct a knowledge graph. Then, using an embedding model, convert all geological entity data and spatial relationship data in the knowledge graph into knowledge graph embedding vectors.

[0020] Step 2, Knowledge-Data Fusion Matrix: For each fishing net unit to be predicted, extract the knowledge graph embedding vectors of itself and its neighbors to form a local knowledge embedding matrix.

[0021] Step 3, Deep learning modeling and training with fused features: Input the knowledge embedding matrix into the input layer of the deep learning network, perform deep learning training, and obtain the mineralization prediction model.

[0022] Step 4, Mineralization Probability Map Generation and Interpretation: The mineralization prediction model is applied to all fishing net units in the entire study area. A mineralization probability value is calculated for each fishing net unit. Then, all mineralization probability values ​​are stitched back into the original geographic space to generate a mineralization potential prediction map.

[0023] Furthermore, in step one, the knowledge graph includes geological entities such as rock masses and faults within the study area, as well as related attribute concepts and relationships, fishing net unit entities, and their spatial relationships with geological entities.

[0024] Furthermore, step four also includes analyzing and interpreting high-probability areas in the mineralization potential prediction map in conjunction with the geological background.

[0025] Furthermore, the analysis included the matching of geological background and metallogenic regularity, the consistency verification of multi-source data and prediction results, and the integrity of metallogenic conditions and the priority division of target areas.

[0026] The above approach has the following beneficial effects: 1. This solution integrates knowledge graphs with mineral exploration data in the input layer in a structured manner, enabling the model to not only learn data patterns but also understand the geological background. The prediction accuracy and AUC are superior to methods that only use chemical raster or evidence layer overlay.

[0027] 2. This approach extracts a knowledge graph from geological maps, structures key geological knowledge, and uses this knowledge for prediction through embedding and matrixing. Observations of the prediction results show that this strategy can suppress the probability of mineralization in areas "close to tectonics but with unfavorable geological backgrounds," while simultaneously identifying potential areas "with insignificant geochemical characteristics but favorable geological assemblages." Compared to data-driven approaches, this strategy plays a certain screening role.

[0028] 3. This solution addresses the problem of insufficient spatial information representation in the existing technology. By constructing an embedding matrix containing a central grid and its neighborhood, local spatial geological information is explicitly integrated into the model, effectively capturing the proximity relationships and local structures between geological entities. This enhanced representation of the geological spatial structure makes the boundaries of the final generated probability map clearer and better reflects the occurrence patterns of real ore bodies.

[0029] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0030] Figure 1 This is a structural diagram of the mineralization prospect prediction system based on embedded knowledge reasoning of the present invention.

[0031] Figure 2 This is a flowchart illustrating the steps of the mineralization prospect prediction method based on embedded knowledge reasoning of the present invention.

[0032] Figure 3 This is a flowchart illustrating the knowledge graph construction and embedding module's construction of a knowledge graph database in the mineralization prospect prediction system based on embedded knowledge reasoning, as described in this invention.

[0033] Figure 4 This is a schematic diagram of knowledge graph entities and relationships in the mineralization prospect prediction system based on embedded knowledge reasoning of the present invention.

[0034] Figure 5 This is a schematic diagram illustrating the generation of the knowledge-data fusion matrix for the corresponding fishing net unit by the knowledge-data fusion module in the mineralization prospect prediction system based on embedded knowledge reasoning of the present invention.

[0035] Figure 6 This is a mineralization prediction map generated by the mineralization prediction and result interpretation module in the mineralization prospect prediction system based on embedded knowledge reasoning of this invention. Detailed Implementation

[0036] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0038] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0039] The following detailed description illustrates the specific implementation method: Example 1: Implementation, for example, attached Figure 1 As shown: A mineralization prospect prediction system based on embedded knowledge reasoning mainly consists of the following modules: data acquisition module, data preparation module, knowledge graph construction and embedding module, knowledge-data fusion module, model training and optimization module, and mineralization prediction and result interpretation module.

[0040] The data acquisition module is mainly used to collect multi-source exploration data and geological map data of the study area; the data preparation module is mainly used to process the exploration data into a unified format and cut it into standardized fishing net units; the knowledge graph construction and embedding module is mainly used to construct a knowledge graph and convert geological entities and relationships into embedding vectors; the knowledge-data fusion module is mainly used to fuse knowledge embedding vectors and raster data into a knowledge-data fusion matrix; the model training and optimization module is mainly used to construct and optimize mineralization prediction models using deep learning networks; and the mineralization prediction and result interpretation module is mainly used to generate mineralization probability prediction maps and delineate mineral exploration target areas.

[0041] The functions of each module will be explained in detail below: The data acquisition module collects exploration data and geological map data within the study area, and marks positive and negative samples according to the mineral occurrence data. The geological map data includes geological entity data, geological attribute data, and geological relationship data. The exploration data includes geochemical element data.

[0042] The data preparation module is used to process the exploration data into evidence layers with a uniform format and spatial alignment, and to rasterize several superimposed evidence layers into several fishing net units; wherein, the fishing net unit is a multi-channel raster matrix.

[0043] Specifically, this embodiment takes the Lhasa area of ​​the Gangdise metallogenic belt in Tibet as an example. The acquisition module collects exploration data and geological map data of 39 geochemical elements in the study area, processes them into spatially aligned evidence layers, and after rasterization, divides them into 4206×2078 fishing net units of 32×32 size.

[0044] like Figure 3 and Figure 4 As shown, the knowledge graph construction and embedding module is used to construct a knowledge graph database using geological map data. It extracts spatial relationship data between fishing net units and geological entity data through spatial analysis algorithms, and transforms the geological entity data and spatial relationship data into knowledge embedding vectors through an embedding model. The embedding model includes one of the TransH model and the DistMult model.

[0045] like Figure 5 As shown, the knowledge-data fusion module is used to extract the knowledge embedding vectors of the grids of the fishing net unit itself and its neighboring fishing net units, and integrate them into a knowledge embedding matrix. The knowledge-data fusion module is also used to splice the knowledge embedding matrix with the grids of the corresponding fishing net unit according to the non-overlapping areas to form the knowledge-data fusion matrix of the corresponding fishing net unit.

[0046] In the knowledge-data fusion module, the knowledge-data fusion matrix includes the original exploration data features and local geological knowledge features of the corresponding fishing net unit. The local geological knowledge features include the geological knowledge of the fishing net unit itself and the surrounding adjacent fishing net units.

[0047] The knowledge-data fusion module also includes a tag identification unit. When the knowledge-data fusion matrix of a certain fishing net unit in the marked area is located within a known mineralization area, the tag identification unit is used to mark the fishing net unit as mineralized. When a certain fishing net unit is far from the mineralization area and far from favorable mineralization structures, the tag identification unit is used to mark the fishing net unit as a unit with low mineralization probability.

[0048] Specifically, based on the geological map of the area, geological entities such as rocks and structures, their attributes and relationships are extracted. A special entity, the "fishnet cell," is introduced from the raster map. Spatial analysis is used to extract the spatial relationships between the fishnet cell entity and the geological entities, constructing a geological map knowledge graph. In this embodiment, the TransH model, capable of handling many-to-many relationships, is used to learn an embedding for each entity. For each fishnet cell in the study area, the embedding vectors of itself and its eight neighboring raster cells are extracted to construct a 9×100-dimensional local geological knowledge graph embedding matrix. This embedding matrix is ​​then fused with 39-channel 32×32 geochemical raster data using a block matrix approach to form a 39×41×132 input tensor. The aforementioned knowledge graph embedding matrix is ​​placed in... Figure 5 In the upper left region, the corresponding multi-channel geochemical raster matrix is ​​placed in the lower right corner. The two matrices do not overlap, and the remaining regions are filled with zero values.

[0049] For each triple in the knowledge graph The TransH model minimizes the scoring function formula: (1).

[0050] The corresponding marginal loss function learns a value for each entity that can represent its geological semantics. 3D embedding vector .

[0051] For any fishing net unit Extract the knowledge embedding vectors of the fishing net cell and all fishing net cells within its spatial neighborhood (e.g., 8-neighborhood). They are organized into a local knowledge matrix according to their spatial location. It binds abstract knowledge with specific geographical locations (i.e., rasters), giving the matrix spatial knowledge and making it structurally compatible with raster data.

[0052] The model training and optimization module is used to input the knowledge-data fusion matrix of the fishing net units within the marked area into a deep learning network to construct a mineralization prediction model. In the model training and optimization module, the deep learning network includes one or more of the following: CNN model, GNN model, and ViT model. In this embodiment, a CNN model is selected as the deep learning network.

[0053] like Figure 6 As shown, the mineralization prediction and result interpretation module is used to calculate the mineralization probability of all fishing net units in the unmarked area using the mineralization prediction model, generate a prediction map, and delineate the target area.

[0054] Supervised training of the model was conducted using samples from known mining and non-mining areas. 118 fishing net units containing mining areas within the labeled region were selected as positive samples, and 122 fishing nets with low mining probability were selected as negative samples. A CNN model was used to extract features, and training employed a binary classification cross-entropy loss function, using the Adam optimizer and combining it with OneCycleLR and other learning rate strategies for efficient optimization. The best-performing CNN model was then used to predict the entire unlabeled region, generating a mining probability prediction map. The network structure includes multiple convolutional layers (for extracting spatial and interaction features), pooling layers (for dimensionality reduction and increasing the receptive field), and a final fully connected layer (for classification).

[0055] Structured fusion: The local knowledge matrix generated above With the corresponding exploration data raster matrix This involves structured stitching within a larger matrix. The knowledge-data fusion matrix can be rigorously defined using the block matrix formula: (2).

[0056] This fusion matrix serves as the unified input for subsequent CNNs. The CNN's convolutional kernels... In the fusion matrix Slide the slider on top and calculate the characteristic formula: (3).

[0057] Its local receptive field can simultaneously cover the pure knowledge region, the pure data region, and the boundary region between the two. This allows the CNN model to learn the features within each modality (such as geological background and exploration data anomaly patterns) and cross-modal interaction features (such as in which geological background a specific exploration anomaly is more indicative) in a unified computation process, in order to identify complex mineralization patterns.

[0058] The CNN using only 39 chemical raster data achieved an accuracy of 0.9 and an AUC of 0.95; the CNN with superimposed construction layers achieved an accuracy of 0.93 and an AUC of 0.94; the prediction method based on embedding inference achieved an accuracy of 0.94 and an AUC of 0.96. It can be seen that this scheme, by embedding knowledge graphs and mineral exploration data in a structured manner at the input layer, enables the model to not only learn data patterns but also understand the geological background. The prediction accuracy and AUC are better than the method using only chemical raster or evidence layer superposition.

[0059] Comparison of predicted target areas revealed that the prediction results of this method successfully corrected the high-probability areas judged solely by tectonic distance, making them more consistent with the actual geological lithology distribution. For example, the mineralization probability of some areas that are close to structures but located on non-ore-hosting rock bodies was effectively reduced. At the same time, the mineralization probability of some areas with insignificant geochemical anomalies but favorable geological combinations (such as specific lithological contact zones) was significantly increased, and new potential target areas were discovered.

[0060] This engineering example demonstrates that this method, through an innovative knowledge-data fusion scheme, successfully integrates geological knowledge into the prediction model. The prediction results not only significantly improve the evaluation indicators but also have greater geological significance, providing more accurate and reliable decision support for actual mineral exploration work.

[0061] Example 2: like Figure 2 As shown, unlike the above embodiments, a method for predicting mineralized prospective areas based on embedded knowledge reasoning includes the following steps: Step 1, Data Preparation and Representation: Spatially align the evidence layers to generate multi-channel raster images. Divide these raster images into several fishnet units according to size, with each fishnet unit representing a multi-channel raster matrix. Construct a knowledge graph. Subsequently, using an embedding model, convert all geological entity data and spatial relationship data within the knowledge graph into knowledge graph embedding vectors. The knowledge graph includes geological entities such as rock masses and faults in the study area, along with related attribute concepts and relationships, as well as fishnet unit entities and their spatial relationships with these geological entities.

[0062] Step 2, Knowledge-Data Fusion Matrix: For each fishing net unit to be predicted, extract the knowledge graph embedding vectors of itself and its neighbors to form a local knowledge embedding matrix.

[0063] Step 3, Deep learning modeling and training with fused features: Input the knowledge embedding matrix into the input layer of the deep learning network, perform deep learning training, and obtain the mineralization prediction model.

[0064] Step 4, Mineralization Probability Map Generation and Interpretation: The mineralization prediction model is applied to all unmarked fishing net units in the entire study area. A mineralization probability value is calculated for each fishing net unit. Then, all mineralization probability values ​​are stitched back into the original geographic space to generate a mineralization potential prediction map.

[0065] Step four also includes analyzing and interpreting high-probability areas in the mineralization potential prediction map in conjunction with the geological background.

[0066] Analysis of the matching between geological background and metallogenic regularity, verification of consistency between multi-source data and prediction results, and determination of the integrity of metallogenic conditions and priority of target areas.

[0067] This approach integrates knowledge graph embedding with mineral exploration data in a structured manner at the input layer. This allows the model to not only learn data patterns but also understand the geological context, resulting in higher prediction accuracy and AUC compared to methods using only chemical raster or evidence layer overlays. Furthermore, this approach extracts knowledge graphs from geological maps, structuring key geological knowledge and applying it to predictions through embedding and matrixing. Observations of the prediction results show that this strategy can suppress the probability of mineralization in areas "close to structures but with unfavorable geological backgrounds," while simultaneously identifying potential areas "with insignificant geochemical characteristics but favorable geological assemblages." Compared to data-driven approaches, this strategy plays a certain screening role.

[0068] This solution addresses the problem of insufficient spatial information representation in mineralization prediction using knowledge graphs. By constructing an embedding matrix containing a central grid and its neighborhood, local spatial geological information is explicitly integrated into the model, effectively capturing the proximity relationships and local structures between geological entities. This enhanced representation of geological spatial structure makes the boundaries of the final probability map clearer and better reflects the occurrence patterns of real ore bodies.

[0069] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A metallogenic prospective area prediction system based on embedded knowledge reasoning, characterized in that, The method comprises a collection module, a data preparation module, a knowledge graph construction and embedding module, a knowledge-data fusion module, a model training and optimization module, and a mineralization prediction and result interpretation module. The collection module collects exploration data and geological map data in the study area, and marks positive and negative samples according to mine point data; wherein the geological map data comprises geological entity data, geological attribute data and geological relationship data. The data preparation module is used for processing the exploration data into a unified format and a spatially aligned evidence layer, and dividing a plurality of superimposed evidence layer grids into a plurality of fishing net units; wherein each fishing net unit is a multi-channel grid matrix. The knowledge graph embedding module is used for constructing a knowledge graph database by using the geological map data, and extracting spatial correlation data between the fishing net units and the geological entity data by using a spatial analysis algorithm, and converting the geological entity data into knowledge embedding vectors by using an embedding model. The knowledge-data fusion module is used for extracting knowledge embedding vectors of the fishing net unit itself and the grids of the adjacent fishing net units in the neighborhood, and integrating them into a knowledge embedding matrix. The knowledge-data fusion module is also used for splicing the knowledge embedding matrix of the fishing net unit and the corresponding exploration grid data matrix according to the non-overlapping area, to form a knowledge-data fusion matrix corresponding to the fishing net unit. The model training and optimization module is used for inputting the knowledge-data fusion matrix of the fishing net unit with labels into a deep learning network, to construct a mineralization prediction model. The mineralization prediction and result interpretation module is used for calculating the mineralization probability of all fishing net units in the study area by using the mineralization prediction model, to generate a prediction map and delineate target areas.

2. The metallogenic prospective area prediction system based on embedded knowledge reasoning according to claim 1, characterized in that, In the collection module, the exploration data comprises a plurality of geochemical element data. 3.The metallogenic prospect area prediction system based on embedded knowledge reasoning according to claim 2, characterized in that, In the knowledge graph embedding module, the embedding model comprises one of a TransH model and a DistMult model.

4. The metallogenic prospect area prediction system based on embedded knowledge reasoning according to claim 3, characterized in that, In the knowledge-data fusion module, the knowledge-data fusion matrix comprises original exploration data features and geological knowledge features of the corresponding fishing net unit, and geological knowledge features of the adjacent fishing net units in the neighborhood.

5. The system for metallogenic prospective area prediction based on embedded knowledge reasoning according to claim 4, characterized in that, The knowledge-data fusion module further comprises a label identification unit, which is used for marking the fishing net unit as mineralization when the knowledge-data fusion matrix of the fishing net unit in the study area is located in a known mineralization area. The label identification unit is used for marking the fishing net unit as a low mineralization probability area when the fishing net unit is far away from the known mineralization area and the ore-controlling structure.

6. The system for metallogenic prospective area prediction based on embedded knowledge reasoning according to claim 5, characterized in that, In the model training and optimization module, the deep learning network comprises one or more of a CNN model, a GNN model and a ViT model.

7. A method for predicting metallogenic prospective area based on embedded knowledge reasoning, according to any one of the prediction system for metallogenic prospective area based on embedded knowledge reasoning as claimed in claims 1-6, characterized in that, The method comprises the following steps: Step one: data preparation and representation: spatially align the evidence layer, generate a multi-channel grid image, cut the grid image into a plurality of fishing net units according to the size, each fishing net unit is a multi-channel grid matrix, construct a knowledge graph based on the geological map data, and then convert all entities and relationships in the knowledge graph into knowledge graph embedding vectors by using an embedding model; Step two: knowledge-data fusion matrix: for each fishing net unit to be predicted, extract the knowledge graph embedding vectors of itself and its neighborhood, to form a local knowledge embedding matrix; The department knowledge embedding matrix is spliced with the exploration matrix without overlap to form a knowledge-data fusion matrix. Step three, deep learning modeling and training of fused features: input the knowledge-data fusion matrix into the input layer of the deep learning network, perform deep learning training, and obtain a metallogenic prediction model. Step four, generation and interpretation of metallogenic probability map: apply the metallogenic prediction model to all fishnet cells in the study area, calculate a metallogenic probability value for each fishnet cell, and then interpolate according to the original geographic space to generate a metallogenic potential prediction map.

8. The method for predicting mineralized prospective areas based on embedded knowledge reasoning according to claim 7, characterized in that, In step one, the knowledge graph includes geological entities such as rock bodies and faults in the study area, as well as related attribute concepts and relationships, fishnet cell entities, and their spatial relationships with geological entities.

9. The method for predicting mineralized prospective areas based on embedded knowledge reasoning according to claim 8, characterized in that, In step four, it also includes analyzing and interpreting high-probability areas in the metallogenic potential prediction map in combination with the geological background.

10. The method for predicting mineralized prospective areas based on embedded knowledge reasoning according to claim 9, characterized in that, The content of analyzing and interpreting high-probability areas in the metallogenic potential prediction map includes: geological background and metallogenic regularity matching analysis, consistency verification of multi-source data and prediction results, and target area priority division based on the completeness of metallogenic conditions.

Images