A mineral prospecting prediction system based on multi-scale big data fusion

Abstract

The present invention provides a prospecting prediction system based on multi-scale big data fusion, the system comprising: a geological big data management module and a big data mining and analysis module, respectively for managing and mining and analyzing geological, geophysical, geochemical, remote sensing, and mineral data; a multi-factor extraction module for extracting geological elements from the mining and analysis results output by the big data mining and analysis module; a multi-scale grid calculation module for segmenting and assigning values ​​to geological elements using multi-scale multi-factor grid technology; an intelligent prospecting prediction module for compiling result maps; an intelligent prospecting evaluation module for compiling optimal decision diagrams; a decision-making support module for generating prospecting result reports; and a prospecting scheduling and command module for synchronously updating data in the prospecting scheduling and command module. By integrating a basic prospecting and prediction data set, a knowledge base, a model base, and an algorithm base, the present invention enables flexible scheduling and convenient use of various data, knowledge, models, and algorithm tools.

Description

Technical Field

[0001] The present invention relates to the technical field of mineral resource prediction, and in particular to a mineral prospecting prediction system based on multi-scale big data fusion. Background Art

[0002] The era of big data intelligence has led to the development of earth science and technology, promoted interdisciplinary and integrated innovation, and proposed four paradigms for scientific research: experimental induction, model deduction, simulation, and intensive big data discovery. Earth science is undergoing a major transformation from relatively independent disciplinary branches to a multidisciplinary integration involving information science, computer science, and other disciplines. Mineral resource prediction and evaluation based on big data and artificial intelligence technologies is currently a key area of ​​application in geosciences. The development of mineral resource prediction and evaluation is moving towards the comprehensive mapping, modeling, integration, fusion, and evaluation of multiple exploration variables.

[0003] The geoscience big data required for mineral resource prediction and evaluation encompasses multiple scales (including mineralization zones, mineral clusters, ore fields, deposits, and ore bodies), multiple factors (including geology, geophysics, geochemistry, remote sensing, and mining engineering), and multiple dimensions (including one-dimensional drill hole histograms, two-dimensional profiles / mid-section maps, and three-dimensional geological models). The advent of the big data era has provided new opportunities and opportunities for intelligent prospecting, positioning, and prediction, as well as for the quantitative prediction and evaluation of mineral resources. Geological research institutions at home and abroad are accelerating the integration and application of geoscience data across various fields of geoscience, including basic geological research, energy mineral survey and evaluation, and energy mineral safety. Therefore, it is necessary to develop a prospecting and prediction system that integrates multi-scale, multi-factor geological, geophysical, geochemical, and remote sensing big data. This system would provide a business support platform and collaborative working environment for prospecting and evaluation personnel, significantly improve the efficiency of prospecting and evaluation, establish a new paradigm for prospecting and evaluation research, and facilitate new breakthroughs in the new round of strategic initiatives for mineral exploration breakthroughs. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a mineral prospecting prediction system based on multi-scale big data fusion to overcome or at least partially solve the above-mentioned problems existing in the prior art.

[0005] To achieve the above-mentioned object of the invention, the present invention provides a mineral prospecting prediction system based on multi-scale big data fusion, the system comprising:

[0006] Geological big data management module, used to manage geological, geophysical, geochemical, remote sensing, and mineral data;

[0007] Big data mining and analysis module, used for mining and analyzing geological, geophysical, geochemical, remote sensing, and mineral data;

[0008] The multi-factor extraction module is used to extract geological elements from the mining and analysis results output by the big data mining and analysis module by integrating the attention mechanism;

[0009] Multi-scale grid calculation module, used to divide and assign values ​​to the geological elements of the multi-element extraction module using multi-scale multi-element grid technology;

[0010] The intelligent prospecting and prediction module is used to use the prospecting and prediction model to locate and quantitatively predict multi-scale mineral resources based on the geological elements that have been segmented and assigned values, and to compile the resulting maps;

[0011] Intelligent prospecting evaluation module, used to classify the prediction area and evaluate the uncertainty and risk of the result map, and compile the optimal decision map;

[0012] The decision-making support module is used to comprehensively display the optimal decision diagram through front-end plug-in-free visualization technology, and automatically generate a prospecting results report based on the optimal decision diagram based on natural language processing technology;

[0013] The prospecting scheduling and command module is used to access the optimal decision diagram and synchronously update the data in the prospecting scheduling and command module.

[0014] Furthermore, the geological big data management module is specifically used to perform:

[0015] By optimizing the node splitting strategy and spatial coverage calculation method to process the two-dimensional spatial data in geology, geophysics, geochemistry, remote sensing, and mineral resources, the distributed tile storage of geological, geophysics, geochemistry, remote sensing, and mineral resources data is optimized by dividing them according to the R-tree spatial index.

[0016] Furthermore, the big data mining and analysis module is specifically used to perform:

[0017] Geological data processing, mining and analysis: selecting the initial cluster center of geological data through cluster center initialization method according to the spatial distribution characteristics of geological data;

[0018] Geophysical data processing, mining, and analysis: using deep neural networks to learn the noise and signal characteristics in geophysical data, automatically identifying and removing noise from the data; extracting geological structure features based on graph neural networks, constructing geophysical data into a graph structure, and automatically learning the topological structure and feature information in the graph using graph convolutional networks and graph attention networks; and deep learning inversion methods based on physical information constraints, integrating geophysical gravity, magnetic, and electrical physics laws and prior knowledge into deep learning models in the form of constraints.

[0019] Geochemical data processing, mining and analysis: Based on the outlier processing method of wavelet transform and Transformer, combined with the multi-resolution analysis characteristics of wavelet transform, geochemical data is decomposed into different frequency sub-bands. The Transformer algorithm is used in different frequency sub-bands to identify weak geochemical outliers;

[0020] Remote sensing data processing, mining and analysis: wavelet transform is used to perform multi-scale decomposition in the spatial and spectral dimensions of remote sensing data, and spatiotemporal filters are used to remove noise. Time series analysis methods are introduced to dynamically suppress noise in continuous-time hyperspectral data. Generative adversarial networks are used to generate synthetic data for mineral spectral data samples. By training the generator and discriminator, mineral spectral data is generated to expand the training sample set, and the expanded sample set is used for training through the classification algorithm.

[0021] Mineral data processing, mining and analysis are used to mine and analyze the spatial correlation between ore-controlling structures, geological bodies, geophysical anomalies, geochemical anomalies and remote sensing anomalies.

[0022] Furthermore, the multi-factor extraction module is specifically configured to perform:

[0023] Establish a multi-factor extraction architecture based on streaming data processing, receive and process the mining and analysis results output by the big data mining and analysis module in real time, and automatically extract geological elements from the mining and analysis results through feature extraction based on the attention mechanism.

[0024] Furthermore, the multi-scale grid computing module is specifically used to perform:

[0025] Multi-scale grid generation: Adaptive multi-scale grid generation algorithm divides geological elements into grids and automatically adjusts the density and size of the grid according to the complexity and spatial distribution of geological elements at different scales. Dynamic scale conversion and fusion technology is used to transfer information between grids of different scales.

[0026] Multi-element grid assignment, the geological elements that have completed grid division are assigned and associated through semantic-driven element assignment and association strategies, and a semantic-driven method is introduced to assign and associate geological elements. Through the semantic model, the semantic information of different elements is assigned to the geological elements that have completed grid division.

[0027] Furthermore, the intelligent prospecting prediction module is specifically used to perform:

[0028] Knowledge graph construction, based on a multi-source heterogeneous data fusion algorithm using deep learning and knowledge graph technology, automatically identifies the features and semantic information of different types of data in the geological big data management module, uses natural language processing technology to extract prospecting-related knowledge from structured and unstructured data, and integrates them through semantic assignment and knowledge fusion technology to construct a geological knowledge graph. Through an adaptive knowledge reasoning and dynamic update mechanism, new prospecting knowledge and laws are automatically inferred based on the existing knowledge in the geological knowledge graph and the real-time input geological data. Using reinforcement learning and uncertainty reasoning technology, the reasoning strategy is dynamically adjusted, and a dynamic update mechanism for the geological knowledge graph is established to track the latest research results and exploration data in the geological field in real time and automatically update the geological knowledge graph;

[0029] The prospecting prediction model is constructed by integrating geological knowledge graphs and machine learning algorithms, combining different machine learning models, introducing transfer learning technology, meta-learning technology and automatic machine learning algorithms, and automatically selecting machine learning algorithms based on the characteristics of input geological data, problem type and target requirements. The algorithm parameters are optimized through intelligent search algorithms to build a prospecting prediction model;

[0030] Prospecting prediction: input the geological elements that have been segmented and assigned values ​​into the prospecting prediction model, conduct positioning prediction and quantitative prediction of multi-scale mineral resources, and compile the results map.

[0031] Furthermore, the intelligent prospecting evaluation module is specifically used to perform:

[0032] Determine the prediction area as the classification basis based on the quantity and category factors of potential mineral resources within the prediction area, and divide the prediction area into regions from high to low probability of mineralization;

[0033] Through multivariate statistical methods, the weights of geology, physics, chemistry, remote sensing and minerals are determined, and uncertainty and risk assessment is performed based on the weights;

[0034] Through multi-layer intelligent fusion and dynamic update algorithm, the association and logical relationship between different layers in the geological structure area are automatically identified to integrate and compile the optimal decision map. The different layers in the geological structure area include but are not limited to the basic layer, mineralization element layer, prediction element layer, minimum prediction area layer and prediction area classification and grading layer.

[0035] Furthermore, the auxiliary decision module is specifically used to perform:

[0036] Optimal decision graph visualization uses back-end grid data non-slicing technology and front-end plug-in-free visualization technology to provide optimal decision graph infinite scaling for comprehensive display and query functions;

[0037] Exploration work deployment proposal, through comprehensive analysis of prospecting prediction models, to generate a text of the next exploration work deployment proposal;

[0038] Prospecting verification project proposals, through comprehensive analysis of prospecting prediction models, to generate text for the next step of prospecting verification project proposals;

[0039] The preparation of prospecting results report is carried out through comprehensive analysis of prospecting prediction model combined with natural language processing technology to generate the text of prospecting results report.

[0040] Furthermore, the prospecting dispatching and commanding module is specifically used to execute:

[0041] Data changes in the auxiliary decision-making module are automatically detected through spatial big data caching and asynchronous transmission technology.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention proposes a mineral exploration prediction system based on multi-scale big data fusion. By integrating multi-source data such as geology, geophysics, geochemistry, remote sensing, and mineral resources, as well as unstructured information such as literature and monographs, it realizes efficient data collection, management and analysis. By combining classical algorithms with artificial intelligence technology, the geological data is deeply mined and processed, key geological elements are quickly extracted, and the unified assignment and integration of data are achieved through multi-scale grid calculation, which effectively solves the problems of multi-source data fusion and computing efficiency. By combining geological knowledge maps with machine learning and deep learning algorithms, the system can realize the precise delineation of mineralization zones, mineral clusters, and exploration target areas, as well as the optimization of deep and side blocks of mines, and provide mineral resource estimation and results map compilation functions. At the same time, the present invention can compile optimal decision diagrams, has the ability to conduct risk assessment of prediction results, provide scientific advice for exploration work deployment and prospecting verification projects, and realize the automation of report compilation. Through integrated design, the present invention realizes the flexible scheduling and convenient use of geological data, knowledge and models, significantly improves the efficiency and accuracy of mineral exploration, provides strong support for the rapid prediction and evaluation of mineral resources, and helps achieve a new round of mineral exploration breakthrough goals. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0045] Figure 1A schematic diagram of the overall structure of a mineral prospecting prediction system based on multi-scale big data fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0047] Reference Figure 1 This embodiment provides a mineral prospecting prediction system based on multi-scale big data fusion, the system comprising:

[0048] The geological big data management module is used to manage geological, geophysical, geochemical, remote sensing and mineral data.

[0049] The geological big data management module is specifically used to perform:

[0050] By optimizing the node splitting strategy and spatial coverage calculation method to process the two-dimensional spatial data in geology, geophysics, geochemistry, remote sensing, and mineral resources, the distributed tile storage of geological, geophysics, geochemistry, remote sensing, and mineral resources data is optimized by dividing them according to the R-tree spatial index.

[0051] In this embodiment, the geological big data management module is used to manage geological spatial database, geophysical spatial database, geochemical spatial database, remote sensing database, mineral deposit database, drilling database, unstructured database management and business process library.

[0052] The spatial database proposes an R-tree spatial indexing technology. By optimizing node splitting strategies (including splitting algorithms based on spatial density and data correlation) and spatial coverage range calculation methods (including adaptive boundary adjustment and elliptical coverage models), the geological big data management module can more effectively reduce data overlap, improve index construction speed and query efficiency when processing large-scale two-dimensional spatial data, and achieve better performance in high-concurrency environments. At the same time, in terms of spatial data storage and management technology, a strategy based on R-tree spatial index partitioning is adopted to optimize distributed tile storage. Through the improved tile cutting algorithm and storage strategy, data can be more evenly distributed across nodes in a distributed environment, improving data reading and writing performance while supporting efficient parallel processing and data recovery.

[0053] The geological spatial database is used to manage the 1:250,000 regional geological map spatial database, the 1:200,000 regional geological map spatial database, and the 1:50,000 regional geological map spatial database, specifically including sedimentary (volcanic) rock strata, intrusive rock rock age, intrusive rock spectrum, metamorphic rock (rock) strata, vein rock (surface), special geological bodies, informal strata, faults, tectonic deformation zones, alteration zones (surfaces), metamorphic facies zones, mineralization zones, migmatization zones, volcanic facies zones, mineral deposits (points) and other element layers.

[0054] The geophysical exploration spatial database is used to manage databases such as gravity, magnetism, electrical sounding, induced polarization electrical sounding, transient electromagnetic, magnetotelluric sounding, controlled source audio frequency geodetic sounding, ground penetrating radar, and shallow seismic; manage regional gravity databases, including measurement point data, work area information, gravity base point network data, gravity density measurement information, etc.; magnetic databases, including aeromagnetic survey information and ground magnetic survey data information, aeromagnetic base point information, aeromagnetic survey work area information, aeromagnetic work area range information, aeromagnetic data, etc.; geomagnetic base point information, ground magnetic survey work area information, magnetic survey work area range information, geomagnetic sample measurement information , ground magnetic survey data, etc.); ground electrical method database, including electrical method work area information, electrical method survey line information, DC resistivity method measurement point information, DC resistivity method observation information, DC induced polarization method measurement point information, DC induced polarization method observation data, AC induced polarization method measurement point information, AC induced polarization method observation data, artificial source frequency sounding method measurement point information, artificial source frequency sounding method observation data, transient electromagnetic sounding method measurement point information, transient electromagnetic sounding method observation data, conventional charging method measurement point information, natural electric field method measurement point information, magnetotelluric measurement point information, magnetotelluric method observation data, etc.

[0055] The geochemical spatial database is used to manage the element analysis data and related sampling information of stream sediments, rocks, and soil measurements, including geochemical exploration area information table, geochemical exploration work range table, analysis element information table, sampling location information table, regional stream sediment data table, rock measurement data table, soil measurement data table, etc., as well as spatial calculation process data and final results data such as geochemical maps, single element anomaly maps, combined element anomaly maps, and comprehensive anomaly maps.

[0056] The remote sensing database is used to manage multi-source remote sensing image data (visible light, multispectral, hyperspectral, etc.) and metadata information, as well as remote sensing interpretation results data, including annular structure maps, linear structure maps, rock identification maps, mineralization maps, alteration maps, etc.

[0057] The mineral deposit database is used to manage and record data such as the mineral deposit number, original code, mineral deposit name, X coordinate, Y coordinate, mineral type code, mineral type name, paragenetic minerals, associated minerals, number of mineral deposits, ore grade, scale, mineralization age, mineralization type, tectonic position of the mineral deposit, geological structure characteristics, ore-bearing geological bodies or ore-bearing horizons, wall rock alteration, and industrial type.

[0058] The drilling database is used to manage basic drilling information, drilling layer information, drilling measurement data, drilling detection data, drilling analysis data, etc.

[0059] Unstructured databases are used to manage various types of geological and mineral-related data, including documents, monographs, textbooks, achievement reports, and maps, and provide data support for the knowledge base management module.

[0060] The business process library is used to manage the geological, physical, chemical, remote and mineral space big data management throughout the entire life cycle of the prospecting and prediction business process, and to realize the management of the initial library, process library and results library; the business process includes the data integration stage, mining and analysis stage, prospecting and prediction stage, resource evaluation stage, decision support stage and results output stage.

[0061] The initial database is used to implement business process management and related data management in the data integration stage after the user selects the research area.

[0062] The process library is used to implement business process management and related data management for users in the mining and analysis stage, prospecting and prediction stage, resource evaluation stage, and decision-making support stage.

[0063] The results library is used to implement business process management and related data management for users during the results output stage.

[0064] The big data mining and analysis module is used to mine and analyze geological, geophysical, geochemical, remote sensing, and mineral data. In this embodiment, it is used to perform preprocessing, spatial interpolation, anomaly extraction, information extraction, and structure identification on geological, geophysical, geochemical, remote sensing, and mineral data.

[0065] The big data mining and analysis module is specifically used to perform:

[0066] Geological data processing, mining and analysis, select the initial cluster center of geological data through cluster center initialization method according to the spatial distribution characteristics of geological data.

[0067] Geological anomalies are key to mineral exploration prediction and evaluation. The algorithm library provides a large-scale distributed parallel spatial computing GIS tool algorithm for geological anomaly mining and analysis, including geological entropy analysis and tectonic influence range analysis (Buffer space analysis). Distributed parallel spatial computing optimizes the spatial clustering algorithm and proposes a cluster center initialization method. Based on the spatial distribution characteristics of the data, the initial cluster center is selected more reasonably, the number of iterations is reduced, and clustering efficiency is improved. At the same time, the data division and processing methods in the Map and Reduce stages are optimized, making data processing more balanced on the distributed cluster, reducing communication overhead, and realizing rapid clustering analysis of large-scale two-dimensional spatial data; and geological spatial data editing tools, which mainly involve the creation, modification, analysis and display of spatial data, mainly including point, line, and surface editing, attribute data editing, coordinate conversion, etc., and also provide attribute data editing tools.

[0068] Geophysical data processing, mining and analysis, using deep neural networks to learn the noise and signal characteristics in geophysical data, automatically identifying and removing noise in the data, extracting geological structure features based on graph neural networks, constructing geophysical data into a graph structure, and using graph convolutional networks and graph attention networks to automatically learn the topological structure and feature information in the graph. Based on the deep learning inversion method constrained by physical information, the geophysical gravity, magnetic, and electrical physical laws and prior knowledge are integrated into the deep learning model in the form of constraints.

[0069] Utilizing machine learning, deep learning, and other artificial intelligence algorithms provided by specialized geophysical processing algorithms and algorithm libraries, we perform format conversion, data cleaning, noise removal, outlier removal, missing value processing, correction processing, data standardization, gridding, derivative calculation, preprocessing, anomaly separation, filtering, derivative calculation, analytical extension, polarization, inversion, and quality assessment on gravity, magnetic, electrical, and seismic geophysical data, thereby improving the accuracy and effectiveness of geophysical data in intelligent prospecting prediction results. During the data preprocessing stage, a denoising technology based on deep learning was proposed. Using deep neural networks, we learn the noise and signal characteristics in geophysical data. Through comparative training on a large amount of noisy and noise-free data, the model can automatically identify and remove various types of noise in the data, such as random noise and coherent noise. Compared with traditional filtering methods, it can better preserve data details and weak signals.

[0070] During the feature extraction and analysis phase, a graph neural network-based geological structure feature extraction method was proposed. This method constructs geophysical data into a graph structure, where nodes represent data points or geological bodies, and edges represent spatial or physical property relationships between data points. Using a graph convolutional network (GCN) and a graph attention network (GAT), the topological structure and feature information in the graph are automatically learned, extracting complex geological structural features such as faults and folds, thereby improving the accuracy of geological structure identification. During the data inversion phase, a deep learning inversion method based on physical information constraints was proposed. This method incorporates geophysical gravity, magnetic, and electrical physics laws and prior knowledge into the deep learning model in the form of constraints. This allows the model to learn data features while adhering to physical laws, improving the physical rationality and reliability of the inversion results and enabling more accurate inversion of underground geological structures and physical parameters.

[0071] Geochemical data processing, mining and analysis, based on the outlier processing method of wavelet transform and Transformer, combined with the multi-resolution analysis characteristics of wavelet transform, decomposes geochemical data into different frequency sub-bands, and uses the Transformer algorithm in different frequency sub-bands to identify geochemical weak outliers.

[0072] Utilizing machine learning, deep learning, and other artificial intelligence algorithms provided by specialized geochemical processing algorithms and algorithm libraries, we preprocess, analyze, and fractalize geochemical data from stream sediments, soils, and rocks to improve the accuracy and effectiveness of geochemical data in intelligent prospecting predictions. During the anomaly analysis phase, we proposed an outlier processing method based on wavelet transforms and Transformers. By combining the multi-resolution analysis characteristics of wavelet transforms, we decompose geochemical data into different frequency subbands. Within each subband, we use the Transformer algorithm to accurately identify weak geochemical outliers, providing a more accurate data foundation for subsequent prospecting predictions.

[0073] Remote sensing data processing, mining and analysis, multi-scale decomposition of remote sensing data in spatial and spectral dimensions through wavelet transform, noise removal combined with spatiotemporal filters, introduction of time series analysis methods, dynamic noise suppression of continuous-time hyperspectral data, and use of generative adversarial networks to generate synthetic data of mineral spectral data samples. By training the generator and discriminator, mineral spectral data is generated to expand the training sample set, and the expanded sample set is used for training through the classification algorithm.

[0074] Utilizing remote sensing image processing and artificial intelligence algorithms, this approach improves the accuracy and effectiveness of remote sensing data in intelligent mineral prospecting prediction by preprocessing multi-source remote sensing data, including data dimensionality reduction and feature selection, cluster analysis, image classification and target detection, spectral feature analysis, geological structure information extraction, texture and shape feature analysis, association rule mining, and multi-source data fusion and analysis. During the data preprocessing phase, a multi-scale spatiotemporal denoising algorithm for hyperspectral data was proposed. Considering the correlation of hyperspectral data in spatial, spectral, and temporal dimensions, a multi-scale spatiotemporal denoising method was proposed. Wavelet transforms were used to perform multi-scale decomposition in spatial and spectral dimensions, combined with spatiotemporal filters to remove noise. Furthermore, time series analysis methods were introduced to dynamically suppress noise in continuous hyperspectral data, effectively preserving useful information, improving data quality, and enhancing the ability to detect weak mineral spectral signals.

[0075] In the image classification and target detection stage (lithology identification, alteration identification, mineral identification, structure extraction, etc.), a hyperspectral data classification and prediction technology was proposed, using a small sample mineral classification based on a generative adversarial network. To address the problem of the scarcity of certain mineral samples in hyperspectral data, a generative adversarial network (GAN) was used to generate synthetic data similar to real mineral samples; by training the generator and discriminator, the generator was able to generate high-quality mineral spectral data and expand the training sample set; combined with traditional classification algorithms, such as support vector machines or random forests, the expanded sample set was used for training to improve the classification accuracy of small sample mineral categories.

[0076] Mineral data processing, mining and analysis are used to mine and analyze the spatial correlation between ore-controlling structures, geological bodies, geophysical anomalies, geochemical anomalies and remote sensing anomalies.

[0077] Through the analysis and mining of mineral data, mainly including the analysis of ore deposit distribution density, mineralization anomaly information analysis, mineralization intensity coefficient analysis, mineralization complexity coefficient analysis, the spatial correlation between mineral (ization) bodies and ore-controlling structures, geological bodies, geophysical anomalies, geochemical anomalies, and remote sensing anomalies, the advantages of various data are comprehensively utilized to improve the accuracy and reliability of prospecting information.

[0078] The multi-factor extraction module is used to extract geological elements from the mining and analysis results output by the big data mining and analysis module by integrating the attention mechanism. In this embodiment, the multi-factor extraction module proposes feature extraction based on the attention mechanism, and integrates the attention mechanism into the multi-factor extraction module so that the module can automatically focus on the most important feature information for mineral exploration prediction. At the same time, a real-time dynamic extraction and update technology is proposed to establish a multi-factor extraction architecture based on streaming data processing, which can receive and process real-time data from different data sources such as geology, geophysics, geochemistry, and remote sensing. When new data flows in, the characteristic variables and characteristic values ​​of the geological elements in the mineral exploration prediction model can be updated immediately, realizing real-time dynamic extraction and monitoring of geological variables, and timely discovering potential mineral exploration clues and abnormal situations.

[0079] The multi-factor extraction module is specifically used to perform:

[0080] Establish a multi-factor extraction architecture based on streaming data processing, receive and process the mining and analysis results output by the big data mining and analysis module in real time, and automatically extract geological elements from the mining and analysis results through feature extraction based on the attention mechanism.

[0081] In this embodiment, geological, geophysical, geochemical, remote sensing, mineral and other data are extracted based on the characteristic variables and characteristic values ​​of geological elements defined in the prospecting prediction model:

[0082] Geological element extraction focuses on studying the spatial distribution patterns and mutual relationships of geological bodies such as structures, igneous rocks, strata, or mineral deposits (ore deposits and mineralized points), and extracting geological information that may be directly or indirectly related to regional mineralization;

[0083] Geophysical element extraction, including the extraction and editing of gravity, magnetic, electrical and other geophysical raw data, as well as the extraction of interpretation results data;

[0084] Geochemical element extraction, including raw data, geochemical anomaly maps and other information extraction functions;

[0085] Remote sensing element extraction, including annular structural maps, linear structural maps, rock mass, mineralized zones, alteration zones and other information extraction functions;

[0086] Mineral element extraction, including extraction based on spatial relationships, extraction based on attribute field information, etc.

[0087] The multi-scale grid computing module is used to segment and assign values ​​to the geological elements of the multi-element extraction module using multi-scale multi-element grid technology. The multi-scale grid computing module is specifically used to perform:

[0088] Multi-scale grid generation uses an adaptive multi-scale grid generation algorithm to divide geological elements into grids and automatically adjust the density and size of the grid according to the complexity and spatial distribution differences of geological elements at different scales. Through dynamic scale conversion and fusion technology, information is transmitted between grids of different scales.

[0089] In this embodiment, an adaptive multi-scale grid generation algorithm is proposed. It automatically adjusts the density and size of the grid based on the complexity and spatial distribution differences of geological features at different scales, such as mineralization zones, mineral clusters, ore fields, ore deposits, and ore bodies. For example, in areas where geological features vary dramatically and ore bodies may be present, a finer grid is automatically generated; while in areas with relatively simple geological features, a coarser grid is used. This effectively reduces the amount of computation and data storage while ensuring data accuracy. Furthermore, a dynamic scale conversion and fusion technology is designed to achieve seamless connection and information transfer between grids of different scales. When converting from the macro scale of mineralization zones to the micro scale of ore bodies, the information in the large-scale grid can be accurately mapped to the small-scale grid, while retaining key geological and geophysical information. Through this dynamic conversion and fusion, multi-scale grid data can better support comprehensive analysis from macro to micro in intelligent prospecting prediction.

[0090] Multi-element grid assignment involves assigning and associating values ​​to meshed geological elements using a semantically driven element assignment and association strategy. Using a semantic model, semantic information from different elements is assigned to meshed geological elements. Geological elements extracted by the multi-element extraction module, including geological anomalies, geophysical anomalies, geochemical anomalies, remote sensing anomalies, and mineralization bodies, are segmented and assigned values ​​at different scales to form a multi-channel data set, providing data preparation for the intelligent mineral exploration and prediction module. By proposing a semantically driven element assignment and association strategy and introducing a semantically driven approach, multiple elements are assigned values ​​and associated. By constructing semantic models from fields such as geology and geophysics, semantic information from different elements is incorporated into the grid assignment process. For example, for mineralization bodies, semantic associations are established with geological anomalies, geophysical anomalies, and other geological elements based on semantic information such as their mineral composition and genesis. These associations are then reflected in the grid assignment, further exploring potential connections between multiple elements and improving the accuracy and reliability of intelligent mineral exploration and prediction.

[0091] The intelligent prospecting prediction module is used to use the prospecting prediction model to locate and quantitatively predict multi-scale mineral resources based on the segmented and valued geological elements, and to compile the resulting maps. The construction of the prospecting prediction model in the intelligent prospecting prediction module in this embodiment relies on the relevant algorithms provided by the algorithm library, the relevant geological knowledge provided by the knowledge library, and the existing models provided by the model library. The geological knowledge map is constructed by combining the relevant algorithms in the algorithm library with the series of geological knowledge in the knowledge library. The prospecting prediction model is constructed by combining the relevant algorithms in the algorithm library, the relevant models in the model library, and the knowledge map.

[0092] The algorithm library manages key algorithms, including the GIS Algorithm Toolbox, the Mathematical Geology Algorithm Toolbox, the Geostatistical Algorithm Toolbox, the Classical Prediction Algorithm Toolbox, the Machine Learning Algorithm Toolbox, and the Deep Learning Algorithm Toolbox, and provides an extensible algorithm tool interface. A machine learning algorithm that integrates prior geological knowledge has been proposed. This combines prior geological knowledge with machine learning algorithms, incorporating geological concepts, laws, and experience as constraints or as a basis for feature engineering. This approach enables machine learning models to better understand and utilize specialized information in geological data, improving model accuracy and interpretability and avoiding model results that contradict geological common sense.

[0093] The algorithm library proposes adaptive machine learning algorithm selection and parameter optimization. Using meta-learning techniques and automated machine learning algorithms, it automatically adapts the most appropriate machine learning algorithm based on the characteristics of the input geological data, the problem type, and the target requirements. It then optimizes the algorithm parameters through intelligent search algorithms, significantly improving the efficiency and effectiveness of machine learning in geological data processing and reducing the workload and subjectivity of manual parameter adjustment. It also proposes the application of ensemble learning and transfer learning in geological prediction, combining multiple machine learning models, such as constructing an ensemble model of support vector machines, decision trees, and random forests. The prediction results of each model are synthesized through voting, averaging, and other strategies to improve the stability and accuracy of predictions. Furthermore, transfer learning technology is introduced to transfer model knowledge trained in other similar geological regions or related fields to new geological data scenarios, accelerating model training and improving the model's adaptability to new data.

[0094] To address the common challenges of limited sample size and data imbalance in geological data, a deep learning model based on data augmentation is proposed. By performing data augmentation operations such as rotation, scaling, translation, and noise addition on the prediction factor data in the prospecting prediction model, the dataset size is expanded and the data diversity is increased. Furthermore, the model structure is designed using techniques such as spatial attention mechanisms and multi-factor (geological, physical, chemical, remote sensing, and mineral features) geological attention mechanisms. This allows the model to focus more on key spatial and geological features in the data, improving its learning and generalization capabilities for geological data.

[0095] By fusing the feature representations of the two networks, joint modeling and analysis of multimodal geological data is achieved. A multimodal deep learning algorithm fusion method is proposed, combining different types of deep learning algorithms to construct a multimodal deep learning model. For example, a convolutional neural network (CNN) is used to extract spatial features from geological image data, while a recurrent neural network (RNN) or long short-term memory network (LSTM) is used to process geological process simulation data. This allows for more comprehensive mining of information in geological data, improving understanding and prediction of geological phenomena.

[0096] To explain the basis and process of predictions and help geologists understand the decision-making mechanisms of models, we have proposed the application of interpretable deep learning algorithms in the field of geology. For example, interpretable methods based on rule extraction and feature importance analysis are applied to the analysis and prediction of geological data. This not only enables models to provide accurate predictions, but also enhances trust in deep learning models, promoting their widespread application in the field of geology.

[0097] The knowledge base is used to manage prospecting knowledge graphs and large model systems, and realize functions such as prospecting knowledge question and answer, induction and summary, and report writing. In the face of data with wide sources and complex structures in the geological field, a knowledge extraction and integration algorithm for multi-source heterogeneous data fusion and a multi-source heterogeneous data fusion algorithm based on deep learning and knowledge graph technology are proposed. It can automatically identify the characteristics and semantic information of different types of data (such as geological reports, exploration data, remote sensing images, etc.), use natural language processing technology to extract prospecting-related knowledge from structured and unstructured data, and integrate this knowledge into a unified knowledge graph through semantic alignment and knowledge fusion technology, effectively solving the problems of inconsistent and incomplete information in traditional methods when processing multi-source data.

[0098] Using adaptive knowledge reasoning and dynamic update mechanisms, we have developed an adaptive knowledge reasoning engine that can automatically infer new prospecting knowledge and patterns based on existing knowledge in the knowledge graph and real-time geological data input. Using reinforcement learning and uncertainty reasoning techniques, we can dynamically adjust reasoning strategies to improve the accuracy and reliability of reasoning. Furthermore, we have established a dynamic knowledge graph update mechanism to track the latest research results and exploration data in the geological field in real time, automatically updating the knowledge graph to ensure the timeliness and accuracy of knowledge.

[0099] Using large prediction model technology, on the basis of the basic large model, the prospecting geological and mineral data set is superimposed, and after pre-training, fine-tuning, and alignment, a large model of prospecting knowledge is formed, which realizes the functions of prospecting knowledge question and answer, metallogenic geological background research and reasoning, metallogenic law research and reasoning, ore deposit model research and reasoning, prospecting model research and reasoning, prospecting prediction model establishment, prospecting prediction results summary, geological work next step suggestion compilation, prospecting prediction results report compilation, etc.

[0100] During the pre-training phase, a large-scale model pre-training technique for prospecting knowledge was proposed, integrating multimodal data fusion into the model. This innovative approach integrates multiple data types, including spatial database data from geology, geophysics, geochemistry, remote sensing, and mineral resources, as well as text reports, images, and other specialized data, into the large model for joint training. By designing a specialized multimodal fusion layer and attention mechanism, the model can fully learn the connections and complementary information between different modal data, improving its understanding and expression of prospecting knowledge. Compared to traditional single-modal pre-training methods, this technology enables the model to better handle complex geological data, enhancing its generalization and accuracy.

[0101] To address the relatively small amount of data and high annotation costs in the mineral exploration field, we proposed a model fine-tuning method based on transfer learning and domain adaptation, and another model fine-tuning method based on transfer learning and domain adaptation. First, a large-scale basic model is pre-trained using large-scale general domain data to learn a common language and knowledge representation. Then, fine-tuning is performed on a mineral exploration geological and mineral dataset. By introducing domain adaptation technology, the model can quickly adapt to the characteristics and needs of the mineral exploration field, reducing its reliance on large amounts of annotated data. A knowledge graph and logical reasoning module are introduced to generate interpretable mineral exploration conclusions and evidence by associating and reasoning the prediction results of deep learning with prior knowledge in the knowledge graph. Furthermore, a strategy for dynamically adjusting fine-tuning parameters is designed to automatically optimize the fine-tuning process based on the data distribution and model performance, improving the model's fine-tuning effect.

[0102] The model library is used to manage expert prospecting prediction models and AI prospecting prediction models: the expert prospecting prediction model management module is a prospecting prediction model summarized by geological experts based on the research and understanding of mineralization laws; the AI ​​prospecting prediction model management module is a prospecting prediction model summarized by a large model system after pre-training based on massive geological data (knowledge-driven); the intelligent prospecting prediction module generates a prospecting prediction model by extracting and integrating multi-factor spatial data of typical minerals (data-driven) combined with regional mineralization geological background and regional mineralization laws. It is the result of the integration of knowledge-driven and data-driven.

[0103] First, we proposed a classification and hierarchical management technique. Based on the source and nature of the prospecting prediction models, the model library is divided into expert prospecting prediction model management areas and AI prospecting prediction model management areas. Models in different areas are managed hierarchically based on factors such as complexity and scope of application. For expert prospecting prediction models, the hierarchies are further subdivided based on different mineralization types (such as magmatic and sedimentary); for AI prospecting prediction models, the hierarchies are based on the type of geological data used for pre-training (such as geophysical and geochemical data). A tree-like directory structure is used for storage, facilitating quick location and management of different types of models.

[0104] Secondly, a knowledge graph association technology was proposed. This technology uses knowledge graphs to link various types of knowledge from mineralization research (such as geological structure, rock characteristics, and mineralization signatures) with expert prospecting prediction models. For AI prospecting prediction models, knowledge graphs are used to connect knowledge nodes from the massive amount of geological data used for model pre-training. This technology also incorporates knowledge from massive databases related to regional mineralization geological background and regional mineralization patterns to build a more comprehensive knowledge graph association. When a model needs to be queried or invoked, the knowledge graph's relationship network allows for rapid access to relevant knowledge and information, assisting in model selection and application.

[0105] Finally, a dynamic update and adaptive adjustment technology was proposed. With the acquisition of new geological data, in-depth research on mineralization patterns, and optimization of AI algorithms, various prospecting prediction models require continuous updating. A dynamic update mechanism should be established to automatically trigger the model update process when new data or knowledge becomes available. For AI prospecting prediction models, online learning technology is used to enable real-time adjustment of model parameters based on new data. For expert prospecting prediction models, a visual editing interface is provided to facilitate geological experts to modify and improve the models based on the latest research results. Furthermore, the model library can adaptively adjust the model's application strategy and parameter settings based on information such as changes in geological conditions in different regions and feedback from prospecting practices.

[0106] The intelligent prospecting prediction module is specifically used to perform:

[0107] Knowledge graph construction, a multi-source heterogeneous data fusion algorithm based on deep learning and knowledge graph technology, automatically identifies the characteristics and semantic information of different types of data in the geological big data management module, uses natural language processing technology to extract prospecting-related knowledge from structured and unstructured data, and integrates it through semantic assignment and knowledge fusion technology to construct a geological knowledge graph. Through adaptive knowledge reasoning and dynamic update mechanism, new prospecting knowledge and laws are automatically inferred based on the existing knowledge in the geological knowledge graph and the real-time input geological data. Using reinforcement learning and uncertainty reasoning technology, the reasoning strategy is dynamically adjusted, and a dynamic update mechanism for the geological knowledge graph is established. The latest research results and exploration data in the geological field are tracked in real time, and the geological knowledge graph is automatically updated.

[0108] The mineral exploration prediction model is constructed by integrating geological knowledge graphs and machine learning algorithms, combining different machine learning models, introducing transfer learning technology, meta-learning technology and automatic machine learning algorithms, and automatically selecting machine learning algorithms based on the characteristics of the input geological data, problem types and target requirements. The algorithm parameters are optimized through intelligent search algorithms to construct a mineral exploration prediction model.

[0109] Prospecting prediction: input the geological elements that have been segmented and assigned values ​​into the prospecting prediction model, conduct positioning prediction and quantitative prediction of multi-scale mineral resources, and compile the results map.

[0110] In this embodiment, the intelligent prospecting prediction module uses the classical prediction, machine learning and deep learning algorithms in the algorithm library, as well as the multi-factor geophysical, chemical and remote sensing big data generated by the model library and multi-scale grid computing mode to perform large-scale data fusion and calculation, to achieve the functions of prospecting area delineation of mineralization belts, target area delineation of ore cluster areas, target area delineation of exploration areas, and optimization of deep edge blocks of mines, while providing the function of estimating the amount of mineral resources, and then compiling result maps, including prediction area result maps, prediction target area maps, resource amount result maps, etc.

[0111] Classic prediction algorithm module: adopts evidence weight method, information quantity method, feature analysis method, and fuzzy evidence weight method.

[0112] Weight of evidence algorithm: Prospecting prediction is carried out according to the mineral prediction business process of the weight of evidence method. First, the prediction elements defined by the expert prospecting prediction model or the AI ​​prospecting prediction model in the model library module are used to construct and preprocess the evidence layer (determine the prediction elements in the prospecting prediction model and binarize the data). Then, the weight of evidence calculation is performed (counting the number of units and the number of mineral-bearing units of each variable, calculating the positive and negative weights, and calculating the contrast coefficient). Then, a conditional independence test is performed again. Then, the posterior probability calculation and prospect area delineation are performed. Finally, the model is tested and evaluated.

[0113] Information volume method algorithm: Prospecting prediction is carried out according to the information volume method business process. First, the prediction factors defined by the expert prospecting prediction model or AI prospecting prediction model in the model library module are used to carry out the prospecting mark determination and assignment stage (prospecting mark identification, assignment calculation). Secondly, the information volume is comprehensively calculated, and then prospecting prediction is carried out to determine the favorable mineralization area. Finally, the model is verified and evaluated.

[0114] Feature analysis algorithm: Prospecting prediction is carried out according to the business process of feature analysis method. First, the prediction factors defined by the expert prospecting prediction model or AI prospecting prediction model in the model library module are used to perform feature analysis and variable extraction. Secondly, feature variables are preprocessed using data standardization. Then, a prediction model is established, model training and model evaluation are carried out, and then prospecting areas are delineated and results are verified and optimized.

[0115] Mineral resource estimation: Based on the identified prospective areas, resource estimation is carried out using the volumetric method based on the experience of prospecting experts and data parameters obtained through information technology.

[0116] In this embodiment, a multi-algorithm ensemble learning framework is constructed, organically combining classic prediction algorithms (such as weight of evidence, information, feature analysis, and fuzzy weight of evidence) with machine learning and deep learning algorithms. Within this framework, different algorithms can leverage their respective strengths, and their results can be integrated through voting mechanisms, weighted averaging, and other methods. For example, classic prediction algorithms, based on geostatistical principles, can provide relatively stable basic prediction results; machine learning algorithms (such as random forests and support vector machines) are superior in handling nonlinear relationships and complex data patterns; and deep learning algorithms (such as convolutional neural networks and recurrent neural networks) excel at uncovering deeper features in data.

[0117] By integrating the prediction results of these algorithms through an integrated learning framework, the accuracy and reliability of prospecting predictions can be effectively improved. Uncertainty quantification methods are introduced in the mineral resource estimation process. Traditional volumetric methods for resource estimation often ignore the uncertainty of geological data and the error factors in the estimation process. This new method uses techniques such as Monte Carlo simulation and Bayesian inference to perform uncertainty analysis on various parameters in the estimation process (such as the boundaries, thickness, and grade of the ore body). For example, Monte Carlo simulation is used to randomly generate parameter values ​​multiple times, and the resource volume under different parameter combinations is calculated, thereby obtaining the probability distribution range of the resource volume. This provides more comprehensive and accurate information for mineral resource assessment, enabling decision makers to fully understand the uncertainty risks of resource estimation.

[0118] The intelligent prospecting evaluation module is used to classify the prediction areas and evaluate the uncertainty and risk of the result maps, and compile the optimal decision map.

[0119] The intelligent prospecting and evaluation module is specifically used to perform:

[0120] Determine the prediction area based on the quantity and category factors of potential mineral resources within the prediction area as the classification basis, and divide the prediction area into regions from high to low probability of mineralization;

[0121] Through multivariate statistical methods, the weights of geology, physics, chemistry, remote sensing and minerals are determined, and uncertainty and risk assessment is performed based on the weights;

[0122] Through multi-layer intelligent fusion and dynamic update algorithm, the association and logical relationship between different layers in the geological structure area are automatically identified to integrate and compile the optimal decision map. The different layers in the geological structure area include but are not limited to the basic layer, mineralization element layer, prediction element layer, minimum prediction area layer and prediction area classification and grading layer.

[0123] In this embodiment, prediction areas are determined based on factors such as the quantity and type of potential mineral resources within the prediction area. Prediction areas are then optimized and classified, categorized into three categories: A, B, and C, from highest to lowest probability of mineralization. Uncertainty and risk assessments are then conducted, comprehensively considering geological, physical, chemical, and remote prediction factors to quantify these uncertainties and risks. This provides a scientific basis for mineral development decisions, ensuring investment rationality and project sustainability. For mineral prediction area optimization and classification, a multi-dimensional dynamic prediction area classification and optimization algorithm is proposed. Traditional prediction area classification and optimization relies primarily on a single or a few factors. This innovative method integrates multiple sources of geological, physical, chemical, and remote data to construct a comprehensive and sophisticated integrated evaluation index system. For example, this method not only considers the quantity and type of potential mineral resources in the prediction area, but also deeply analyzes geophysical anomaly characteristics (such as the intensity and morphology of gravity and magnetic anomalies), geochemical element composition and distribution patterns, the influence of geological structures on ore production, and geological features reflected in remote sensing imagery.

[0124] Using multivariate statistical methods such as principal component analysis and the analytic hierarchy process (AHP), we determine the weights of various indicators to achieve a comprehensive and objective evaluation of the prospecting potential of the predicted area. We have developed a dynamic, adaptive model for classifying and optimizing the predicted areas. This model automatically adjusts classification and optimization results based on new data acquired in real time (such as new exploration data and research results). Using online learning algorithms from machine learning, the prospecting prediction model continuously learns from the characteristics and patterns of new data and updates its evaluation of the predicted area. For example, when a new discovery of unique geological structures or unusual geochemical enrichment in a predicted area is made, the prospecting prediction model can quickly take this into account, reassess the prospecting potential of the predicted area, and adjust its classification and optimization level, ensuring that the prospecting evaluation is always based on the latest and most accurate information.

[0125] Compiling an Optimal Decision Map: The mineral resource prediction results map for the prediction area is constructed through a thematic base map of the geological structure of the prediction work area, a mineralization factor map, a prediction factor map, and a mineral resource prediction map. Layers include a geological structure base layer, a mineralization factor layer, a prediction factor layer, a minimum prediction area layer, and a prediction area classification and grading layer. When compiling the optimal decision map, a multi-layer intelligent fusion and dynamic update algorithm was developed. This algorithm automatically identifies the associations and logical relationships between different layers, such as the geological structure base layer, mineralization factor layer, prediction factor layer, minimum prediction area layer, and prediction area classification and grading layer, achieving efficient and accurate fusion. For example, through semantic matching and spatial location association, geological structure information is organically combined with mineralization factor information, enabling the decision map to more comprehensively reflect the relationship between geological conditions and prospecting potential. Furthermore, the algorithm supports real-time dynamic updates. When new geological data or research results are available, the corresponding layers can be rapidly updated and automatically re-fused to generate the latest optimal decision map.

[0126] The decision-making support module is used to comprehensively display the optimal decision diagram through front-end plug-in-free visualization technology, and automatically generate a prospecting results report based on the optimal decision diagram based on natural language processing technology. The decision-making support module is specifically used to perform:

[0127] Optimal decision graph visualization uses back-end grid data non-slicing technology and front-end plug-in-free visualization technology to provide optimal decision graph infinite scaling for comprehensive display and query functions;

[0128] Exploration work deployment proposal, through comprehensive analysis of prospecting prediction models, to generate a text of the next exploration work deployment proposal;

[0129] Prospecting verification project proposals, through comprehensive analysis of prospecting prediction models, to generate text for the next step of prospecting verification project proposals;

[0130] The preparation of prospecting results report is carried out through comprehensive analysis of prospecting prediction model combined with natural language processing technology to generate the text of prospecting results report.

[0131] In this embodiment, the natural language processing capabilities of the prospecting prediction model enable intelligent text generation for prospecting results reports. The prospecting prediction model can automatically organize language to generate a clearly structured and comprehensive report text based on the results of big data analysis and the conclusions of spatial comprehensive analysis and evaluation. During the text generation process, the prospecting prediction model adheres to geological and mining industry standards and report templates, accurately expressing information such as the purpose, methods, results, and conclusions of the prospecting work. For example, when compiling a prospecting prediction results report, the prospecting prediction model can, based on the analysis results of the predicted target area, elaborate on the prospecting potential, existing risks, and recommendations for the next step, providing decision makers with a comprehensive and accurate reference basis.

[0132] Back-end grid data slicing-free technology and front-end plug-in-free visualization technology provide infinitely zoomable, comprehensive display and query capabilities for prospecting areas (target areas), improving visualization loading speed and effectiveness. Large model technology is used to pre-train vertical industry large models, while integrating the system's intelligent prospecting and prediction business processes to conduct spatial comprehensive analysis and evaluation of the generated results. This allows for the deployment of further exploration work, the generation of text for prospecting verification project recommendations, and the production of text for prospecting results reports.

[0133] The prospecting scheduling and command module is used to access the optimal decision diagram and synchronously update the data in the prospecting scheduling and command module. The prospecting scheduling and command module is specifically used to execute:

[0134] Data changes in the auxiliary decision-making module are automatically detected through spatial big data caching and asynchronous transmission technology.

[0135] In this embodiment, to ensure that the prospecting and dispatching command module can obtain the latest prospecting data, a real-time data update and synchronization mechanism was developed. By establishing a real-time data connection with the decision-making support module, the optimal decision diagram data can be collected and transmitted in real time. When the data in the decision-making support module changes, the system can automatically detect and update it in a timely manner to ensure data consistency and timeliness. At the same time, spatial big data data caching and asynchronous transmission technology are used to improve the efficiency and stability of data updates, avoiding data transmission delays or interruptions that affect the normal operation of the system.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A mineral prospecting prediction system based on multi-scale big data fusion, characterized by: The system comprises: Geological big data management module, used to manage geological, geophysical, geochemical, remote sensing, and mineral data; Big data mining and analysis module, used for mining and analyzing geological, geophysical, geochemical, remote sensing, and mineral data; The multi-factor extraction module is used to extract geological elements from the mining and analysis results output by the big data mining and analysis module by integrating the attention mechanism; Multi-scale grid calculation module, used to divide and assign values ​​to the geological elements of the multi-element extraction module using multi-scale multi-element grid technology; The intelligent prospecting and prediction module is used to use the prospecting and prediction model to locate and quantitatively predict multi-scale mineral resources based on the geological elements that have been segmented and assigned values, and to compile the resulting maps; The intelligent prospecting prediction module is specifically used to perform: Knowledge graph construction, based on a multi-source heterogeneous data fusion algorithm using deep learning and knowledge graph technology, automatically identifies the features and semantic information of different types of data in the geological big data management module, uses natural language processing technology to extract prospecting-related knowledge from structured and unstructured data, and integrates them through semantic assignment and knowledge fusion technology to construct a geological knowledge graph. Through an adaptive knowledge reasoning and dynamic update mechanism, new prospecting knowledge and laws are automatically inferred based on the existing knowledge in the geological knowledge graph and the real-time input geological data. Using reinforcement learning and uncertainty reasoning technology, the reasoning strategy is dynamically adjusted, and a dynamic update mechanism for the geological knowledge graph is established to track the latest research results and exploration data in the geological field in real time and automatically update the geological knowledge graph; The prospecting prediction model is constructed by integrating geological knowledge graphs and machine learning algorithms, combining different machine learning models, introducing transfer learning technology, meta-learning technology and automatic machine learning algorithms, and automatically selecting machine learning algorithms based on the characteristics of input geological data, problem type and target requirements. The algorithm parameters are optimized through intelligent search algorithms to build a prospecting prediction model; Prospecting prediction: input the geological elements that have been segmented and assigned values ​​into the prospecting prediction model, conduct positioning prediction and quantitative prediction of mineral resources at multiple scales, and compile the results map; Intelligent prospecting evaluation module, used to classify the prediction area and evaluate the uncertainty and risk of the result map, and compile the optimal decision map; The decision-making support module is used to comprehensively display the optimal decision diagram through front-end plug-in-free visualization technology, and automatically generate a prospecting results report based on the optimal decision diagram based on natural language processing technology; The prospecting scheduling and command module is used to access the optimal decision diagram and synchronously update the data in the prospecting scheduling and command module.

2. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The geological big data management module is specifically used to perform: By optimizing the node splitting strategy and spatial coverage calculation method to process the two-dimensional spatial data in geology, geophysics, geochemistry, remote sensing, and mineral resources, the distributed tile storage of geological, geophysics, geochemistry, remote sensing, and mineral resources data is optimized by partitioning according to the R-tree spatial index.

3. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The big data mining and analysis module is specifically used to perform: Geological data processing, mining and analysis: selecting the initial cluster center of geological data through cluster center initialization method according to the spatial distribution characteristics of geological data; Geophysical data processing, mining, and analysis: using deep neural networks to learn the noise and signal characteristics in geophysical data, automatically identifying and removing noise from the data; extracting geological structure features based on graph neural networks, constructing geophysical data into a graph structure, and automatically learning the topological structure and feature information in the graph using graph convolutional networks and graph attention networks; and deep learning inversion methods based on physical information constraints, integrating geophysical gravity, magnetic, and electrical physics laws and prior knowledge into deep learning models in the form of constraints. Geochemical data processing, mining and analysis: Based on the outlier processing method of wavelet transform and Transformer, combined with the multi-resolution analysis characteristics of wavelet transform, geochemical data is decomposed into different frequency sub-bands. The Transformer algorithm is used in different frequency sub-bands to identify weak geochemical outliers; Remote sensing data processing, mining and analysis: wavelet transform is used to perform multi-scale decomposition in the spatial and spectral dimensions of remote sensing data, and spatiotemporal filters are used to remove noise. Time series analysis methods are introduced to dynamically suppress noise in continuous-time hyperspectral data. Generative adversarial networks are used to generate synthetic data for mineral spectral data samples. By training the generator and discriminator, mineral spectral data is generated to expand the training sample set, and the expanded sample set is used for training through the classification algorithm. Mineral data processing, mining and analysis are used to mine and analyze the spatial correlation between ore-controlling structures, geological bodies, geophysical anomalies, geochemical anomalies and remote sensing anomalies.

4. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The multi-factor extraction module is specifically used to perform: Establish a multi-factor extraction architecture based on streaming data processing, receive and process the mining and analysis results output by the big data mining and analysis module in real time, and automatically extract geological elements from the mining and analysis results through feature extraction based on the attention mechanism.

5. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The multi-scale grid computing module is specifically used to perform: Multi-scale grid generation: Adaptive multi-scale grid generation algorithm divides geological elements into grids and automatically adjusts the density and size of the grid according to the complexity and spatial distribution of geological elements at different scales. Dynamic scale conversion and fusion technology is used to transfer information between grids of different scales. Multi-element grid assignment, the geological elements that have completed grid division are assigned and associated through semantic-driven element assignment and association strategies, and a semantic-driven method is introduced to assign and associate geological elements. Through the semantic model, the semantic information of different elements is assigned to the geological elements that have completed grid division.

6. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The intelligent prospecting and evaluation module is specifically used to perform: Determine the prediction area based on the quantity and category factors of potential mineral resources within the prediction area as the classification basis, and divide the prediction area into regions from high to low probability of mineralization; Through multivariate statistical methods, the weights of geology, physics, chemistry, remote sensing and minerals are determined, and uncertainty and risk assessment is performed based on the weights; Through multi-layer intelligent fusion and dynamic update algorithms, the association and logical relationship between different layers in the geological structure area are automatically identified to integrate and compile the optimal decision map. The different layers in the geological structure area include basic layers, mineralization element layers, prediction element layers, minimum prediction area layers and prediction area classification and grading layers.

7. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The auxiliary decision module is specifically used to perform: Optimal decision graph visualization uses back-end grid data non-slicing technology and front-end plug-in-free visualization technology to provide optimal decision graph infinite scaling for comprehensive display and query functions; Exploration work deployment proposal, through comprehensive analysis of prospecting prediction models, to generate a text of the next exploration work deployment proposal; Prospecting verification project proposals, through comprehensive analysis of prospecting prediction models, to generate text for the next step of prospecting verification project proposals; The preparation of prospecting results report is carried out through comprehensive analysis of prospecting prediction model combined with natural language processing technology to generate the text of prospecting results report.

8. The mineral prospecting prediction system based on multi-scale big data fusion according to claim 1 is characterized in that: The prospecting dispatching and commanding module is specifically used to execute: Data changes in the auxiliary decision-making module are automatically detected through spatial big data caching and asynchronous transmission technology.

Images