Method and system for predicting prospecting target area based on geological three-dimensional modeling

Abstract

The invention discloses a prospecting target area prediction method and system based on geological three-dimensional modeling, and relates to the technical field of mineral resource exploration, and the method comprises the steps: obtaining geological, geophysical, geochemical and remote sensing data of a target area, carrying out the preprocessing, generating a modeling data set, and carrying out the geological modeling; constructing a lithology classification model by combining the three-dimensional geologic model with deep learning, and extracting shallow lithology features to predict the deep lithology of the target area; screening metallogenic factors according to the prediction result of the deep lithology of the target area, the deep geophysical data and the three-dimensional geologic model, and generating a metallogenic prediction factor set; and establishing a prospecting prediction model based on the graph convolutional network, importing the metallogenic prediction factor set to obtain a metallogenic distribution prediction result, and carrying out three-dimensional prospecting target area delineation. According to the method, deep lithologic stratum prediction is carried out through multi-source geological data and geological three-dimensional modeling, and target region delineation is carried out by using the graph convolutional neural network, so that the efficiency and precision of prospecting target region prediction are improved, and the exploration cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of mineral resource exploration, and more specifically, to a method and system for predicting mineral prospecting targets based on geological three-dimensional modeling. Background Art

[0002] Target prediction is a key link in mineral resource exploration, which aims to identify potential mineral resource-rich areas through multi-source data such as geology, geophysics, and geochemistry. Accurate target prediction can significantly reduce exploration costs and improve prospecting efficiency, which is of great significance to the development and utilization of mineral resources. Traditional target prediction methods mainly rely on the experience and manual analysis of geologists, including geological mapping, geophysical exploration, geochemical exploration, etc. Although these methods are effective to a certain extent, they have limitations such as strong subjectivity, low efficiency, and limited accuracy.

[0003] With the advancement of computer technology and data science, modern prospecting target prediction technology has gradually developed in the direction of data-driven and intelligent, mainly including: geographic information system (GIS), machine learning and deep learning, big data and cloud computing, etc. And with the surge in demand for mineral resources and the deepening of exploration, prospecting has shifted from finding shallow ore bodies to finding deep hidden ore bodies. The use of three-dimensional cube models for resource prediction and evaluation of hidden ore bodies has become a hot topic in the field of mineral exploration. Therefore, how to use artificial intelligence technology and three-dimensional cube models to make high-precision predictions of hidden ore body resources is an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method and system for predicting mineral exploration target areas based on geological three-dimensional modeling. Deep lithological strata are predicted through multi-source data and geological three-dimensional modeling, and target areas are delineated using a graph convolutional neural network combined with preferred mineralization prediction factors, providing efficient and accurate technical support for mineral resource exploration.

[0005] The first aspect of the present invention provides a method for predicting a prospecting target area based on geological three-dimensional modeling, comprising the following steps: Acquire geological, geophysical, geochemical and remote sensing data of the target area, pre-process the acquired data to generate a modeling data set, perform geological modeling of the target area based on the modeling data set, and generate a three-dimensional geological model; The lithology classification model is constructed by combining the three-dimensional geological model with deep learning to extract the shallow lithology characteristics of the target area, and the deep lithology of the target area is predicted using the shallow lithology characteristics; Screen mineralization factors based on the predicted results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, and construct the best mineralization factor combination to generate a mineralization prediction factor set; A mineral exploration prediction model is established and trained based on a graph convolutional network, the mineralization prediction factor set is imported into the mineral exploration prediction model to obtain the mineralization distribution prediction result, and three-dimensional mineral exploration target area is delineated.

[0006] In this scheme, geological, geophysical, geochemical and remote sensing data are obtained for preprocessing to generate a modeling data set, and geological modeling of the target area is carried out based on the modeling data set, specifically: According to the geographic information of the target area, historical geological survey data and historical exploration engineering data in the region are obtained, geological, geophysical, geochemical and remote sensing data are screened, and the screened data are standardized, denoised and normalized; Generate a modeling data set through the preprocessed data, extract fault data according to the modeling data set, construct a three-dimensional geological entity model corresponding to the lithological strata in the target area according to the fault data, extend and trim the constructed three-dimensional geological entity model using the stratum boundary, and determine the difference range constraint to perform model interpolation; The maximum and minimum coordinates of the three-dimensional geological entity model are queried to determine the block size, the three-dimensional geological entity model is gridded using the block size, a three-dimensional geological block model is obtained, and attributes of the three-dimensional geological block model are newly created according to the data in the modeling data set; A three-dimensional geological model corresponding to the shallow stratum lithology structure of the target area is generated based on the constructed three-dimensional geological entity model and the three-dimensional geological block model.

[0007] In this solution, the lithology classification model is constructed by combining the three-dimensional geological model with deep learning to extract the shallow lithology characteristics of the target area, and the deep lithology of the target area is predicted using the shallow lithology characteristics, specifically: Slice the three-dimensional geological model in two dimensions to obtain a three-dimensional geological model cross-section with geophysical data attribute labels, and establish a training set and a validation set based on the three-dimensional geological model cross-section with geophysical data attribute labels and shallow lithology; Constructing a lithology classification model, using the training set to train the model, introducing dilation convolution combined with dilation coefficients of different sizes to convolve the three-dimensional geological model profile to obtain feature maps of different scales, matching and encoding the feature maps of different scales with stratum lithology labels, and obtaining shallow lithology characteristics under different geophysical data attribute labels; Each shallow lithology feature matrix is ​​processed by multi-head attention, and in each attention head, mask estimation is introduced to obtain attention weights, and the shallow lithology feature matrix is ​​weighted using the obtained attention weights; Multi-scale context aggregation is performed on the weighted shallow lithological features. The weighted shallow lithological features are transformed using the split standard convolution combined with different expansion coefficients through the split-conversion-merge strategy to capture contextual information of different scales. The contextual information of different scales is merged through concatenation and standard convolution. Importing the fused context information into the fully connected layer to predict the corresponding stratum lithology label, using the verification set to verify the predicted stratum lithology label, and obtaining a trained lithology classification model; The three-dimensional geological model profile corresponding to each deep geophysical data of the target area is imported into the trained lithology classification model to obtain the distribution of deep lithology in the target area.

[0008] In this plan, mineralization factors are screened based on the predicted results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, specifically: Obtain the type of mineral deposits to be explored in the target area, and use big data retrieval to obtain completed geological, geophysical, geochemical and prospecting work case data based on the mineral deposit type, and use deep lithology, deep geophysical data and three-dimensional geological models of the target area to obtain similar case data based on similarity calculation; Use the Pearson correlation coefficient to screen the ore-controlling geological conditions that meet the preset correlation standard from the acquired similar instance data, obtain the metallogenic factors in the instance data according to the screened ore-controlling geological conditions, calculate the information amount contributed by each metallogenic factor to the prospecting prediction, and preliminarily screen the metallogenic factors based on the information amount; The SVM model is trained by the initially screened metallogenic factors to perform classification prediction tasks, and several kernel functions are replaced during the training process to obtain the mean square error of the classification prediction tasks of different metallogenic factors under multiple kernel functions, and the importance scores of different metallogenic factors are generated according to the mean square error; The metallogenic factors are sorted using the importance scores, a preset number of metallogenic factors are selected according to the sorting results, and redundant metallogenic factors are eliminated through collinearity detection.

[0009] In this scheme, the optimal combination of mineralization factors is constructed to generate a mineralization prediction factor set, specifically: Obtain the metallogenic factors corresponding to the ore-controlling geological conditions, randomly generate metallogenic factor combinations to construct the initial population, use the improved dung beetle optimization algorithm to optimize the metallogenic factor combinations, and introduce chaotic mapping into the initial population to optimize the initial dung beetle position; An objective function is constructed through the classification accuracy and characteristic dimension of the mineralization factor combination, the objective function is used to positively associate with the fitness function, and the positions of dung beetle individuals in different behavior states are updated according to the fitness function; In the process of updating the position of dung beetle individuals in different behavior states, a reverse learning strategy and an adaptive nonlinear step-length control strategy are added. The reverse learning strategy obtains the reverse position of each dung beetle individual, and the chaotic population and the reverse population are combined to solve the position information of the dung beetle individual. The adaptive nonlinear step-length control strategy adaptively controls the search step length according to the number of iterations, and finely controls the position update of the dung beetle individual. And using the random coefficient in combination with the reverse learning strategy and the adaptive nonlinear step-size control strategy, when the random coefficient is less than a preset threshold, the adaptive nonlinear step-size control strategy is used to update the position, otherwise, the reverse learning strategy is used to update the position; After iterative optimization, the optimal solution output by the improved dung beetle optimization algorithm is obtained, and the best combination of mineralization factors corresponding to each ore-controlling geological condition is determined according to the optimal solution as the mineralization prediction factor set.

[0010] In this solution, a prospecting prediction model is established and trained based on a graph convolutional network, and the mineralization prediction factor set is imported into the prospecting prediction model to obtain the mineralization distribution prediction result, specifically: According to the mineralization prediction factor set, characteristic variable values ​​are extracted from the completed geological, geophysical, geochemical and prospecting work example data, the relationship between the characteristic variable values ​​of different mineralization prediction factors and mineralization distribution is obtained, a graph representation of mineralization distribution and different mineralization prediction factors is constructed, and an adjacency matrix is ​​extracted; At the same time, according to the example data, the intersection of the three-dimensional geological model of ore body distribution and the three-dimensional geological model of different metallogenic factors is combined to obtain the characteristic variable value range of different metallogenic prediction factors corresponding to the ore body distribution, and obtain the probability distribution function of different metallogenic prediction factors; Using a graph convolutional network to establish a prospecting prediction model to learn the graph representation, updating the node information according to the adjacency matrix and the state update weight matrix, and obtaining an updated node feature vector; Obtaining characteristic variable values ​​of the target area according to the mineralization prediction factor set, obtaining probability distribution values ​​corresponding to the characteristic variable values ​​of the target area through probability distribution functions of different mineralization prediction factors, and processing the adjacency matrix using the probability distribution values ​​to obtain a dynamic adjacency matrix; The node information is updated according to the dynamic adjacency matrix and the state update weight matrix to obtain the final node feature vector, and the final node feature vector is imported into the fully connected layer to obtain the mineralization distribution prediction result.

[0011] In this plan, the corresponding controlling geological conditions and mineralization laws are obtained through the types of mineral deposits in the target area, and the three-dimensional prospecting target area is delineated according to the mineralization distribution prediction results. The delineated three-dimensional prospecting target area is corrected using the controlling geological conditions and mineralization laws corresponding to the mineral deposit type, and a three-dimensional visual expression is performed.

[0012] The second aspect of the present invention provides a prospecting target area prediction system based on geological three-dimensional modeling, the system comprising a modeling data acquisition unit, a three-dimensional geological modeling unit, a deep lithology classification unit, a prospecting prediction unit and a deep prospecting target area delineation unit; The modeling data acquisition unit is responsible for acquiring geological, geophysical, geochemical and remote sensing data of the target area and performing preprocessing to generate a modeling data set; The three-dimensional geological modeling unit is responsible for modeling the shallow geological entities and attributes of the target area according to the modeling data set to obtain a three-dimensional geological model; The deep lithology classification unit is responsible for constructing a lithology classification model based on a deep learning algorithm and a three-dimensional geological model, extracting shallow lithology characteristics of the target area, and predicting the deep lithology of the target area; The prospecting prediction unit is responsible for screening mineralization factors according to the prediction results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, and constructing a mineralization prediction factor set, establishing a prospecting prediction model and training it, and importing the mineralization prediction factor set into the prospecting prediction model to obtain the mineralization distribution prediction results; The deep prospecting target area delineation unit is responsible for three-dimensional prospecting target area delineation based on the mineralization distribution prediction results to realize the visualization of the prospecting target area.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The present invention integrates multi-source data such as geology, geophysics, geochemistry, and remote sensing to construct a three-dimensional geological model, intuitively display the underground geological structure and ore body distribution, predict deep lithology and strata through multi-source data and geological three-dimensional modeling, and use graph convolutional neural network combined with preferred mineralization prediction factors to delineate target areas, reduce the traditional manual prospecting's reliance on experience, realize the automation and intelligence of prospecting target area prediction, significantly improve the efficiency and accuracy of prospecting target area prediction, and reduce exploration costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0015] Figure 1 A flow chart of a method for predicting a prospecting target area based on geological three-dimensional modeling is shown; Figure 2 A flowchart for constructing a lithology classification model to predict deep lithology characteristics of a target area is shown; Figure 3A flow chart showing the use of a prospecting prediction model to obtain mineralization distribution prediction results; Figure 4 A block diagram of a mineral prospecting target area prediction system based on geological three-dimensional modeling is shown. DETAILED DESCRIPTION

[0016] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0017] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0018] Figure 1 A flow chart of a method for predicting mineral prospecting target areas based on geological three-dimensional modeling is shown.

[0019] like Figure 1 As shown, in the first embodiment of the present invention, a method for predicting a prospecting target area based on geological three-dimensional modeling is provided, comprising: S102, obtaining geological, geophysical, geochemical and remote sensing data of the target area, preprocessing the obtained data to generate a modeling data set, performing geological modeling of the target area based on the modeling data set, and generating a three-dimensional geological model; S104, constructing a lithology classification model by combining the three-dimensional geological model with deep learning, extracting shallow lithology characteristics of the target area, and predicting deep lithology of the target area using the shallow lithology characteristics; S106, screening metallogenic factors according to the prediction results of deep lithology in the target area, deep geophysical data and three-dimensional geological model, and constructing the best metallogenic factor combination to generate a metallogenic prediction factor set; S108, establishing and training a mineral exploration prediction model based on a graph convolutional network, importing the mineralization prediction factor set into the mineral exploration prediction model to obtain a mineralization distribution prediction result, and performing three-dimensional mineral exploration target area delineation.

[0020] It should be noted that, according to the geographic information of the target area, the historical geological survey data and historical exploration engineering data in the region are obtained, including regional geological survey, gravity measurement, geomagnetic measurement, gravity and magnetic profiles, magnetotelluric sounding, phase induced polarization measurement and drilling data; the geological, geophysical, geochemical and remote sensing data are screened, and the screened data are standardized, denoised and normalized; the modeling data set is generated by the preprocessed data, and the fault data is extracted according to the modeling data set, and the fault data is extracted from MapGIS maps such as geological mineral maps, remote sensing structural interpretation maps, and electrical geological interpretation profiles. The MapGIS two-dimensional line file is converted into a GOCAD three-dimensional line string file, and the contour lines of the three-dimensional line string file are spliced ​​to form a triangular patch file to construct a three-dimensional geological solid model corresponding to the lithology and strata of the target area. The constructed three-dimensional geological entity model is extended and trimmed using the stratum boundary, and the difference range constraint is determined to perform model interpolation; the maximum and minimum coordinates of the three-dimensional geological entity model are queried to determine the block size, and the three-dimensional geological entity model is gridded using the block size to obtain the three-dimensional geological block model and create the attributes of the three-dimensional geological block model based on the data in the modeling data set; a three-dimensional geological model corresponding to the shallow stratum lithology structure of the target area is generated based on the constructed three-dimensional geological entity model and the three-dimensional geological block model.

[0021] Figure 2 A flow chart for constructing a lithology classification model to predict deep lithology characteristics of a target area is shown.

[0022] According to an embodiment of the present invention, a lithology classification model is constructed by combining the three-dimensional geological model with deep learning, the shallow lithology characteristics of the target area are extracted, and the deep lithology of the target area is predicted using the shallow lithology characteristics, specifically: S202, performing two-dimensional slicing on the three-dimensional geological model to obtain a three-dimensional geological model cross-section with geophysical data attribute labels, and establishing a training set and a validation set based on the three-dimensional geological model cross-section with geophysical data attribute labels and shallow lithology; S204, constructing a lithology classification model, using the training set to perform model training, introducing dilation convolution combined with dilation coefficients of different sizes to perform convolution on the three-dimensional geological model profile to obtain feature maps of different scales, matching and encoding the feature maps of different scales with formation lithology labels, and obtaining shallow lithology characteristics under different geophysical data attribute labels; S206, processing each shallow lithology feature matrix through multi-head attention, introducing mask estimation in each attention head to obtain attention weight, and using the obtained attention weight to weight the shallow lithology feature matrix; S208, performing multi-scale context aggregation on the weighted shallow lithological features, using a split-transform-merge strategy to perform different transformations on the weighted shallow lithological features using the split standard convolution combined with different expansion coefficients, capturing context information of different scales, and merging the context information of different scales through concatenation and standard convolution; S210, importing the fused context information into the formation lithology label corresponding to the prediction of the fully connected layer, using the verification set to verify the predicted formation lithology label, and obtaining a trained lithology classification model; S212, importing the three-dimensional geological model profile corresponding to each deep geophysical data of the target area into the trained lithology classification model to obtain the distribution of deep lithology in the target area.

[0023] It should be noted that the lithology classification model is established based on the DeeplabV3+ network architecture, and the original backbone network Xception of the model is replaced by the pre-trained lightweight MobileNetV2. In the encoder part, the dilated convolution is introduced with different dilation coefficients to convolve the 3D geological model profile to obtain feature maps of different scales, and extract and aggregate features from different angles. In the decoding part, the output of the encoder is first upsampled, and each shallow lithology feature matrix is ​​processed through multi-head attention. Mask estimation is added to the self-attention mechanism to calculate the attention weight. , , represents the number of feature items, represents the matrix dimension, Represents the mask value, represents the query vector, represents the key vector, Represents a value vector; and fuses the shallow lithology features extracted from the backbone network. In the multi-scale feature fusion, the standard convolution kernel is split into several sub-kernels, each with fewer output channels. Each sub-kernel uses different expansion coefficients to transform the shallow lithology features differently, capturing contextual information of different scales. The contextual information of different receptive fields is finally merged through concatenation and standard convolution operations, thereby effectively capturing multi-scale contextual information. Since the mineral composition of different lithologies will lead to differences in their magnetic susceptibility and apparent resistivity, different mineral compositions will also lead to different formation densities of different lithologies, resulting in differences in the regional gravity anomaly effects of the lithology. The fully connected layer of the lithology classification model is used to construct a mapping between geophysical data and lithology to achieve lithology prediction.

[0024] It should be noted that the type of mineral deposits for prospecting in the target area is obtained, and according to the type of mineral deposits, big data retrieval is used to obtain completed geological, geophysical, geochemical and prospecting work example data; the deep lithology, deep geophysical data and three-dimensional geological model of the target area are used to obtain similar example data based on similarity calculation; the Pearson correlation coefficient is used to screen the ore-controlling geological conditions that meet the preset correlation standards from the obtained similar example data, and the preferred ore-controlling geological conditions include stratigraphic conditions, structural conditions, geophysical conditions, geochemical conditions, remote sensing anomaly conditions, etc. The mineralization factors in the example data are obtained according to the screened ore-controlling geological conditions, and the preferred mineralization factors include apparent resistivity distribution characteristics, gravity distribution characteristics, fault distance field, geochemical anomaly distribution, remote sensing anomaly, geomagnetic anomaly distribution, etc. The amount of information contributed by each mineralization factor to prospecting prediction is calculated, and the mineralization factors are preliminarily screened through the information amount; the information amount model can effectively reflect the relevance of the selected mineralization factors and their classification intervals to prospecting prediction. When the information value is greater than 0, it indicates that the prospecting prediction is promoted within the classification interval of the mineralization factor; when the information value is less than 0, it indicates that the prospecting prediction is suppressed within the classification interval of the mineralization factor. The SVM model is trained by the initially screened mineralization factors to perform the classification prediction task. The SVM uses the kernel function to project the data in the existing dimension into the high-dimensional feature space, and performs the classification task through the distance of the hyperplane of the sample. Several kernel functions are replaced during the training process. Commonly used kernel functions include polynomial kernel function and RBF kernel function. The mean square error of the classification prediction task of different mineralization factors under multiple kernel functions is obtained, and the importance scores of different mineralization factors are generated according to the mean square error; the mineralization factors are sorted using the importance score, and a preset number of mineralization factors are selected according to the sorting results, and redundant mineralization factors are eliminated through collinearity detection.

[0025] The dung beetle optimization algorithm is a heuristic optimization algorithm based on the behavior of dung beetles in nature, which simulates the process of dung beetles in finding, rolling and burying dung balls. By simulating the rolling, competition, reproduction and other behaviors of dung beetles, the search and optimization of the solution space are realized. The metallogenic factors corresponding to the ore-controlling geological conditions are obtained, and the metallogenic factor combinations are randomly generated to construct the initial population. The improved dung beetle optimization algorithm is used to optimize the metallogenic factor combinations. The Tent chaotic mapping function is used to introduce chaotic mapping in the initial population to optimize the initial dung beetle positions, so that the distribution of the initial population is more uniform; according to the SVM model, the mean square error is obtained through n-fold cross-training to generate the classification accuracy corresponding to the metallogenic factor combination, and the characteristic dimensions corresponding to the metallogenic factor combination are obtained according to the sum of the dimensions of the metallogenic factors. By maximizing the classification accuracy of the metallogenic factor combination and the maximum The objective function is constructed by minimizing the characteristic dimension of the combination of mineralization factors, and the objective function is used to positively associate with the fitness function. The larger the value of the objective function, the larger the fitness value. The positions of dung beetle individuals in different behavioral states are updated according to the fitness function; in the process of updating the positions of dung beetle individuals in rolling, competing, and breeding behavioral states, a reverse learning strategy and an adaptive nonlinear step-size control strategy are added. The reverse learning strategy obtains the reverse position of each dung beetle individual, and the chaotic population and the reverse population are merged to solve the position information of the dung beetle individual, so as to further improve the diversity of the population, and the dung beetle individuals in the reverse position are updated. Expressed as , represents the upper and lower limits of the individual dimensions of dung beetles, The adaptive nonlinear step size control strategy adaptively controls the search step size according to the number of iterations. As the number of iterations increases, using a smaller step size is conducive to fine-tuning the exploration of local details and controlling the position update of the dung beetle individual. The step size in the adaptive nonlinear step size control strategy is The calculation is expressed as: , represents the original fixed step size, represents the incomplete function, , Indicates the current number of iterations and the maximum number of iterations. The random coefficient is combined with the reverse learning strategy and the adaptive nonlinear step-size control strategy. The preferred random coefficient threshold is set to 0.5. When the random coefficient is less than the preset threshold, the adaptive nonlinear step-size control strategy is used to update the position. Otherwise, the reverse learning strategy is used to update the position. The global optimal position and fitness value are recorded to determine whether the maximum number of iterations is exceeded. If exceeded, the optimization ends. After iterative optimization, the optimal solution output by the improved dung beetle optimization algorithm is obtained. According to the optimal solution, the optimal mineralization factor combination corresponding to each ore-controlling geological condition is determined as the mineralization prediction factor set.

[0026] Figure 3A flow chart of using a prospecting prediction model to obtain mineralization distribution prediction results is shown.

[0027] According to an embodiment of the present invention, a prospecting prediction model is established and trained based on a graph convolutional network, and the mineralization prediction factor set is imported into the prospecting prediction model to obtain a mineralization distribution prediction result, specifically: S302, extracting characteristic variable values ​​from completed geological, geophysical, geochemical and prospecting work example data according to the mineralization prediction factor set, obtaining the relationship between the characteristic variable values ​​of different mineralization prediction factors and mineralization distribution, constructing a graph representation of mineralization distribution and different mineralization prediction factors, and extracting an adjacency matrix; S304, simultaneously obtaining the characteristic variable value range of different metallogenic prediction factors corresponding to the ore body distribution according to the example data combined with the intersection of the three-dimensional geological model of the ore body distribution and the three-dimensional geological models of different metallogenic factors, and obtaining the probability distribution function of the different metallogenic prediction factors; S306, using a graph convolutional network to establish a prospecting prediction model to learn the graph representation, updating the node information according to the adjacency matrix and the state update weight matrix, and obtaining an updated node feature vector; S308, obtaining characteristic variable values ​​of the target area according to the metallogenic prediction factor set, obtaining probability distribution values ​​corresponding to the characteristic variable values ​​of the target area through probability distribution functions of different metallogenic prediction factors, and processing the adjacency matrix using the probability distribution values ​​to obtain a dynamic adjacency matrix; S310, updating the node information according to the dynamic adjacency matrix and the state update weight matrix, obtaining the final node feature vector, and importing the final node feature vector into the fully connected layer to obtain the mineralization distribution prediction result.

[0028] It should be noted that the relationship between the characteristic variable values ​​of different metallogenic prediction factors and the metallogenic distribution is obtained by analyzing the example data. The metallogenic prediction factors and the metallogenic distribution are used as nodes, and the obtained relationship is used as the edge structure to construct a graph representation of the metallogenic distribution and different metallogenic prediction factors. The graph representation is learned using a graph convolutional network, and the correlation information between different metallogenic prediction factors and the metallogenic distribution is obtained based on the relationship matrix and graph convolution. According to the three-dimensional geological model of different data interpolation of the instance data, the three-dimensional geological model of the ore body distribution and the three-dimensional geological model of different metallogenic factors are intersected, and the intersection area of ​​the ore body distribution is assigned 1 operation, and the assignment area is obtained from the three-dimensional geological model of different metallogenic factors to read the corresponding characteristic variable value range, and the probability distribution function of different metallogenic prediction factors is obtained according to the value distribution law of the characteristic variable value range, and the characteristic variable value of the target area is obtained according to the lithological distribution of the deep strata in the target area and the geological, geophysical, geochemical and remote sensing data. The probability distribution function of different metallogenic prediction factors is used to obtain the probability distribution value corresponding to the characteristic variable value of the target area, and the probability distribution value is used to process the adjacency matrix to obtain a dynamic adjacency matrix, so that the prospecting task in the target area has an exclusive adjacency matrix, improves the representation ability of the prospecting prediction model for the correlation between different metallogenic prediction factors and metallogenic distribution, adaptively updates the correlation information of the correlation between metallogenic distribution of different metallogenic prediction factors, and sets a dynamic adjacency matrix to reduce the overfitting risk of the model. The corresponding controlling geological conditions and mineralization laws are obtained through the types of mineral deposits in the target area. The boundaries of the prospecting target area are determined according to the results of mineralization distribution prediction, and the three-dimensional prospecting target area is delineated. With full reference to the relevant results of surface geological exploration, the controlling geological conditions and mineralization laws corresponding to the deposit type are used to correct the delineated three-dimensional prospecting target area to obtain the deep prospecting target area in the target area, and three-dimensional visualization is performed.

[0029] Figure 4 A block diagram of a mineral prospecting target area prediction system based on geological three-dimensional modeling is shown.

[0030] The second embodiment of the present invention provides a prospecting target area prediction system 4 based on geological three-dimensional modeling, which includes a modeling data acquisition unit 401, a three-dimensional geological modeling unit 402, a deep lithology classification unit 403, a prospecting prediction unit 404 and a deep prospecting target area delineation unit 405; The modeling data acquisition unit is responsible for acquiring geological, geophysical, geochemical and remote sensing data of the target area and preprocessing them to generate a modeling data set; The three-dimensional geological modeling unit is responsible for modeling the shallow geological entities and attributes of the target area according to the modeling data set to obtain a three-dimensional geological model; The deep lithology classification unit is responsible for constructing a lithology classification model based on a deep learning algorithm and a three-dimensional geological model, extracting shallow lithology characteristics of the target area, and predicting the deep lithology of the target area; The prospecting prediction unit is responsible for screening mineralization factors according to the prediction results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, and constructing a mineralization prediction factor set, establishing a prospecting prediction model and training it, and importing the mineralization prediction factor set into the prospecting prediction model to obtain the mineralization distribution prediction results; The deep prospecting target area delineation unit is responsible for three-dimensional prospecting target area delineation based on the mineralization distribution prediction results to realize the visualization of the prospecting target area.

[0031] The third embodiment of the present invention provides a computer-readable storage medium, which includes a program for a method for predicting a mineral exploration target area based on geological three-dimensional modeling. When the program for predicting a mineral exploration target area based on geological three-dimensional modeling is executed by a processor, the steps of the method for predicting a mineral exploration target area based on geological three-dimensional modeling are implemented.

[0032] In the several embodiments provided in the present application, it should be understood that the disclosed method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0033] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0034] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0035] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting mineral prospecting targets based on geological three-dimensional modeling, characterized in that: The following steps are involved: Acquire geological, geophysical, geochemical and remote sensing data of the target area, pre-process the acquired data to generate a modeling data set, perform geological modeling of the target area based on the modeling data set, and generate a three-dimensional geological model; The lithology classification model is constructed by combining the three-dimensional geological model with deep learning to extract the shallow lithology characteristics of the target area, and the deep lithology of the target area is predicted using the shallow lithology characteristics; Screen mineralization factors based on the predicted results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, and construct the best mineralization factor combination to generate a mineralization prediction factor set; A mineral exploration prediction model is established and trained based on a graph convolutional network, the mineralization prediction factor set is imported into the mineral exploration prediction model to obtain the mineralization distribution prediction result, and three-dimensional mineral exploration target area is delineated.

2. The method for predicting mineral prospecting target areas based on geological three-dimensional modeling according to claim 1, characterized in that: The acquired geological, geophysical, geochemical and remote sensing data are preprocessed to generate a modeling data set, and geological modeling of the target area is performed based on the modeling data set, specifically: According to the geographic information of the target area, historical geological survey data and historical exploration engineering data in the region are obtained, geological, geophysical, geochemical and remote sensing data are screened, and the screened data are standardized, denoised and normalized; Generate a modeling data set through the preprocessed data, extract fault data according to the modeling data set, construct a three-dimensional geological entity model corresponding to the lithological strata in the target area according to the fault data, extend and trim the constructed three-dimensional geological entity model using the stratum boundary, and determine the difference range constraint to perform model interpolation; The maximum and minimum coordinates of the three-dimensional geological entity model are queried to determine the block size, the three-dimensional geological entity model is gridded using the block size, a three-dimensional geological block model is obtained, and attributes of the three-dimensional geological block model are newly created according to the data in the modeling data set; A three-dimensional geological model corresponding to the shallow stratum lithology structure of the target area is generated based on the constructed three-dimensional geological entity model and the three-dimensional geological block model.

3. The method for predicting mineral prospecting target areas based on geological three-dimensional modeling according to claim 1, characterized in that: The lithology classification model is constructed by combining the three-dimensional geological model with deep learning, and the shallow lithology characteristics of the target area are extracted. The deep lithology of the target area is predicted using the shallow lithology characteristics, specifically: Slice the three-dimensional geological model in two dimensions to obtain a three-dimensional geological model cross-section with geophysical data attribute labels, and establish a training set and a validation set based on the three-dimensional geological model cross-section with geophysical data attribute labels and shallow lithology; Constructing a lithology classification model, using the training set to train the model, introducing dilation convolution combined with dilation coefficients of different sizes to convolve the three-dimensional geological model profile to obtain feature maps of different scales, matching and encoding the feature maps of different scales with stratum lithology labels, and obtaining shallow lithology characteristics under different geophysical data attribute labels; Each shallow lithology feature matrix is ​​processed by multi-head attention, and in each attention head, mask estimation is introduced to obtain attention weights, and the shallow lithology feature matrix is ​​weighted using the obtained attention weights; Multi-scale context aggregation is performed on the weighted shallow lithological features. The weighted shallow lithological features are transformed using the split standard convolution combined with different expansion coefficients through the split-conversion-merge strategy to capture contextual information of different scales. The contextual information of different scales is merged through concatenation and standard convolution. Importing the fused context information into the fully connected layer to predict the corresponding stratum lithology label, using the verification set to verify the predicted stratum lithology label, and obtaining a trained lithology classification model; The three-dimensional geological model profile corresponding to each deep geophysical data of the target area is imported into the trained lithology classification model to obtain the distribution of deep lithology in the target area.

4. The method for predicting mineral prospecting target areas based on geological three-dimensional modeling according to claim 1, characterized in that: The mineralization factors are screened based on the predicted results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, specifically: Obtain the type of mineral deposits to be explored in the target area, and use big data retrieval to obtain completed geological, geophysical, geochemical and prospecting work case data based on the mineral deposit type, and use deep lithology, deep geophysical data and three-dimensional geological models of the target area to obtain similar case data based on similarity calculation; Use the Pearson correlation coefficient to screen the ore-controlling geological conditions that meet the preset correlation standard from the acquired similar instance data, obtain the metallogenic factors in the instance data according to the screened ore-controlling geological conditions, calculate the information amount contributed by each metallogenic factor to the prospecting prediction, and preliminarily screen the metallogenic factors based on the information amount; The SVM model is trained by the initially screened metallogenic factors to perform classification prediction tasks, and several kernel functions are replaced during the training process to obtain the mean square error of the classification prediction tasks of different metallogenic factors under multiple kernel functions, and the importance scores of different metallogenic factors are generated according to the mean square error; The metallogenic factors are sorted using the importance scores, a preset number of metallogenic factors are selected according to the sorting results, and redundant metallogenic factors are eliminated through collinearity detection.

5. The method for predicting mineral prospecting target areas based on geological three-dimensional modeling according to claim 1, characterized in that: Construct the best combination of metallogenic factors to generate a metallogenic prediction factor set, specifically: Obtain the metallogenic factors corresponding to the ore-controlling geological conditions, randomly generate metallogenic factor combinations to construct the initial population, use the improved dung beetle optimization algorithm to optimize the metallogenic factor combinations, and introduce chaotic mapping into the initial population to optimize the initial dung beetle position; An objective function is constructed through the classification accuracy and characteristic dimension of the mineralization factor combination, the objective function is used to positively associate with the fitness function, and the positions of dung beetle individuals in different behavior states are updated according to the fitness function; In the process of updating the position of dung beetle individuals in different behavior states, a reverse learning strategy and an adaptive nonlinear step-length control strategy are added. The reverse learning strategy obtains the reverse position of each dung beetle individual, and the chaotic population and the reverse population are combined to solve the position information of the dung beetle individual. The adaptive nonlinear step-length control strategy adaptively controls the search step length according to the number of iterations, and finely controls the position update of the dung beetle individual. And using the random coefficient in combination with the reverse learning strategy and the adaptive nonlinear step-size control strategy, when the random coefficient is less than a preset threshold, the adaptive nonlinear step-size control strategy is used to update the position, otherwise, the reverse learning strategy is used to update the position; After iterative optimization, the optimal solution output by the improved dung beetle optimization algorithm is obtained, and the best combination of mineralization factors corresponding to each ore-controlling geological condition is determined according to the optimal solution as the mineralization prediction factor set.

6. The method for predicting mineral prospecting target areas based on geological three-dimensional modeling according to claim 1, characterized in that: A prospecting prediction model is established and trained based on a graph convolutional network, and the mineralization prediction factor set is imported into the prospecting prediction model to obtain the mineralization distribution prediction result, specifically: According to the mineralization prediction factor set, characteristic variable values ​​are extracted from the completed geological, geophysical, geochemical and prospecting work example data, the relationship between the characteristic variable values ​​of different mineralization prediction factors and mineralization distribution is obtained, a graph representation of mineralization distribution and different mineralization prediction factors is constructed, and an adjacency matrix is ​​extracted; At the same time, according to the example data, the intersection of the three-dimensional geological model of ore body distribution and the three-dimensional geological model of different metallogenic factors is combined to obtain the characteristic variable value range of different metallogenic prediction factors corresponding to the ore body distribution, and obtain the probability distribution function of different metallogenic prediction factors; Using a graph convolutional network to establish a prospecting prediction model to learn the graph representation, updating the node information according to the adjacency matrix and the state update weight matrix, and obtaining an updated node feature vector; Obtaining characteristic variable values ​​of the target area according to the mineralization prediction factor set, obtaining probability distribution values ​​corresponding to the characteristic variable values ​​of the target area through probability distribution functions of different mineralization prediction factors, and processing the adjacency matrix using the probability distribution values ​​to obtain a dynamic adjacency matrix; The node information is updated according to the dynamic adjacency matrix and the state update weight matrix to obtain the final node feature vector, and the final node feature vector is imported into the fully connected layer to obtain the mineralization distribution prediction result.

7. The method for predicting mineral prospecting target areas based on geological three-dimensional modeling according to claim 1, characterized in that: The corresponding controlling geological conditions and mineralization laws are obtained through the types of mineral deposits in the target area, and the three-dimensional prospecting target area is delineated according to the results of mineralization distribution prediction. The delineated three-dimensional prospecting target area is corrected using the controlling geological conditions and mineralization laws corresponding to the mineral deposit type, and a three-dimensional visual expression is performed.

8. A prospecting target area prediction system based on geological three-dimensional modeling, characterized in that: A method for predicting a prospecting target area based on geological three-dimensional modeling as described in any one of claims 1 to 7 is implemented, wherein the system comprises a modeling data acquisition unit, a three-dimensional geological modeling unit, a deep lithology classification unit, a prospecting prediction unit and a deep prospecting target area delineation unit; The modeling data acquisition unit is responsible for acquiring geological, geophysical, geochemical and remote sensing data of the target area and preprocessing them to generate a modeling data set; The three-dimensional geological modeling unit is responsible for modeling the shallow geological entities and attributes of the target area according to the modeling data set to obtain a three-dimensional geological model; The deep lithology classification unit is responsible for constructing a lithology classification model based on a deep learning algorithm and a three-dimensional geological model, extracting shallow lithology characteristics of the target area, and predicting the deep lithology of the target area; The prospecting prediction unit is responsible for screening mineralization factors according to the prediction results of deep lithology in the target area, deep geophysical data and three-dimensional geological models, and constructing a mineralization prediction factor set, establishing a prospecting prediction model and training it, and importing the mineralization prediction factor set into the prospecting prediction model to obtain the mineralization distribution prediction results; The deep prospecting target area delineation unit is responsible for three-dimensional prospecting target area delineation based on the mineralization distribution prediction results to realize the visualization of the prospecting target area.

Images