Intelligent modeling method and system for three-dimensional geological structure based on deep learning

Abstract

The application discloses a three-dimensional geological structure intelligent modeling method and system based on deep learning, relates to the technical field of geological structure modeling, and comprises the following steps: data feature extraction and mapping are completed through gravity-magnetic collaborative inversion deep network to generate a geological physical parameter field; topological relationship construction and coding conversion are completed by using a space topological constraint three-dimensional coding model to obtain a topological feature matrix; then, a stratigraphic interface nonlinear smoothing fitting algorithm is used to process interface feature points to obtain continuous stratigraphic interface data; subsequently, a multi-scale rock mass integrated intelligent modeling simulation platform is used to complete rock mass layering, parameter calibration and generate a preliminary model. Through the collaborative cooperation of the deep network, topological coding, nonlinear fitting and integrated simulation platform, the application improves the geological topological expression, stratigraphic fitting and overall modeling precision, reduces manual operation, and adapts to intelligent modeling work under complex geological scenes.

Description

Technical Field

[0001] This invention relates to the field of geological structure modeling technology, and in particular to a method and system for intelligent modeling of three-dimensional geological structures based on deep learning. Background Technology

[0002] Three-dimensional geological structure modeling is a core technology in geological research and resource exploration. It accurately presents the spatial distribution, geometric morphology, and physical properties of geological bodies through digital means, providing scientific support for engineering design and resource evaluation. As mineral resource exploration extends to deeper levels and the demand for engineering in complex geological environments increases, traditional modeling methods are no longer sufficient to meet the requirements of high precision and efficiency. Gravity and geomagnetic data, as important data sources reflecting the characteristics of underground geological bodies, are crucial for overcoming technological bottlenecks through their collaborative utilization and intelligent inversion. Existing technologies mainly achieve the construction of three-dimensional geological structures through geological data acquisition, preprocessing, interpolation calculation, and geometric modeling. Typically, discrete geological data is first obtained through drilling and geophysical exploration, spatial interpolation methods are used to fill data gaps, geometric modeling techniques are then used to construct the surface or volumetric model of the geological body, and finally, visualization techniques are used to present the modeling results. Some technologies attempt to combine single gravity or geomagnetic data for inversion analysis, use traditional mathematical models to describe the relationship between geological body parameters, optimize the modeling effect by manually setting modeling parameters and manually correcting model deviations, rely on the experience and judgment of geologists to complete data integration and model adjustment, and the modeling process is mainly based on step-by-step processing, lacking a collaborative linkage mechanism between data and algorithms.

[0003] Existing technologies have two main drawbacks: First, the data utilization and inversion accuracy are insufficient. Traditional methods often process gravity or geomagnetic data separately, failing to fully explore the correlation characteristics between the two types of data. Furthermore, they are poorly adaptable to complex geological conditions during the inversion process, making it difficult to accurately characterize the spatial topological relationships of geological bodies, resulting in deviations between the modeling results and the actual geological conditions. Second, the modeling efficiency and intelligence level are low. They rely on manual intervention to set key parameters, lack efficient and smoothing methods for fitting stratigraphic interfaces, and lack an integrated modeling platform with multi-algorithm collaboration. This makes it impossible to achieve automated linkage of data processing, model building, and iterative optimization. When faced with large-scale data or complex geological structures, the modeling cycle is long and it is difficult to ensure model consistency. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides a method and system for intelligent modeling of three-dimensional geological structures based on deep learning.

[0005] The technical solution adopted in this invention is a deep learning-based intelligent modeling method for three-dimensional geological structures, comprising the following steps: S1, acquiring gravity observation data and geomagnetic observation data, and performing feature extraction and correlation mapping on the data through a gravity-geomagnetic collaborative inversion deep network to generate an initial geophysical parameter field; S2, based on a spatial topological constraint three-dimensional coding model, constructing and encoding the topological relationships of the initial geophysical parameter field to form a multi-dimensional topological feature matrix; S3, using a nonlinear smoothing fitting algorithm for stratigraphic interfaces to fit the stratigraphic interface feature points in the multi-dimensional topological feature matrix to obtain continuous and smooth stratigraphic interface characterization data; S4 S5. The stratigraphic interface characterization data is input into the multi-scale integrated intelligent modeling and simulation platform for rock mass, and the rock mass structure is divided into layers and the parameters are calibrated to generate a preliminary three-dimensional rock mass structure model. S6. The preliminary three-dimensional rock mass structure model is iteratively optimized by gravity and geomagnetism collaborative inversion depth network, and the topological relationship deviation is corrected by combining the spatial topological constraint three-dimensional coding model. The interface accuracy is adjusted by using the stratigraphic interface nonlinear smoothing fitting algorithm. S7. Based on the optimized three-dimensional rock mass structure data, the three-dimensional geological structure is intelligently modeled by the multi-scale integrated intelligent modeling and simulation platform for rock mass, and a complete three-dimensional geological structure model including rock mass parameters, stratigraphic interfaces and topological relationships is output.

[0006] This system is a deep learning-based intelligent modeling system for 3D geological structures. It utilizes a deep learning-based intelligent modeling method for 3D geological structures and includes: a gravity and geomagnetic data acquisition and preprocessing unit, used to acquire gravity and geomagnetic observation data and perform data filtering, establishing a data transmission connection with a gravity and geomagnetic co-inversion deep network computing unit; a gravity and geomagnetic co-inversion deep network computing unit, used to extract features and perform correlation mapping on the acquired observation data, generating an initial geophysical parameter field, and establishing bidirectional data interaction with a spatial topology-constrained 3D encoding model processing unit and an iterative optimization unit; and a spatial topology-constrained 3D encoding model processing unit, used to construct and encode the topological relationships of the initial geophysical parameter field, outputting a multi-dimensional topological feature matrix, and nonlinearly mapping the stratigraphic interface. The nonlinear smoothing fitting algorithm operation unit establishes a data transmission connection; the nonlinear smoothing fitting algorithm operation unit for stratigraphic interfaces is used to fit the stratigraphic interface feature points in the topological feature matrix, generate stratigraphic interface characterization data, and establish a data transmission connection with the multi-scale rock mass integrated intelligent modeling and simulation unit; the multi-scale rock mass integrated intelligent modeling and simulation unit is used to receive stratigraphic interface characterization data and perform rock mass structure layering, parameter calibration, and model construction, and establishes data interaction with the iterative optimization unit and the three-dimensional geological structure output unit respectively; the three-dimensional geological structure output unit is used to receive the optimized three-dimensional rock mass structure data, integrate rock mass parameters, stratigraphic interfaces, and topological relationship information, and output a complete three-dimensional geological structure model. All units synchronize data transmission and work collaboratively through a high-speed data bus.

[0007] Beneficial Effects: This invention proposes a deep learning-based intelligent modeling method and system for 3D geological structures. It utilizes gravity and geomagnetism collaborative inversion technology to deeply mine the correlation features between two types of data. A multi-layer network architecture and dynamic weight adjustment enhance the accuracy of data inversion, solving the modeling bias problem caused by traditional single data processing. A spatial topological constraint coding model constructs precise correlations between geological body units, and a combination of adjacency and distance matrix fusion achieves quantitative representation of topological relationships, overcoming the fuzzy topological description shortcomings of traditional methods. A nonlinear smoothing fitting algorithm for stratigraphic interfaces is used to accurately fit feature points. The fitting parameters are determined through optimization strategies and repeatedly verified and adjusted to ensure the continuity and smoothness of stratigraphic interfaces, overcoming the problems of low efficiency and poor accuracy of traditional fitting methods. Simultaneously, a multi-scale integrated intelligent modeling and simulation platform for rock masses is equipped with parallel computing hardware, supporting efficient processing of large-scale data. Combined with adaptive mesh generation, it achieves accurate modeling of complex geological bodies. With a multi-unit collaborative system architecture, it achieves fully automated linkage from data acquisition, feature extraction, model construction, iterative optimization to result output, significantly reducing manual intervention and improving modeling efficiency and consistency, providing reliable technical support for deep resource exploration, complex geological engineering, and other scenarios. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of method step S2 of the present invention; Figure 3 This is a flowchart of method step S3 of the present invention; Figure 4 This is a flowchart of method step S4 of the present invention; Figure 5 This is a flowchart of step S5 of the method of the present invention; Figure 6 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0009] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0010] like Figure 1As shown, the intelligent modeling method for three-dimensional geological structures based on deep learning includes the following steps: S1, acquiring gravity and geomagnetic observation data, and performing feature extraction and correlation mapping on the data through a gravity-geomagnetic collaborative inversion deep network to generate an initial geophysical parameter field; S2, based on a spatial topological constraint three-dimensional coding model, constructing and encoding the topological relationships of the initial geophysical parameter field to form a multi-dimensional topological feature matrix; S3, using a nonlinear smoothing fitting algorithm for stratigraphic interfaces to fit the stratigraphic interface feature points in the multi-dimensional topological feature matrix to obtain continuous and smooth stratigraphic interface characterization data; S4, [the method is described in the original text, but the provided text is incomplete and requires further context to translate accurately]. The surface characterization data is input into the multi-scale integrated intelligent modeling and simulation platform for rock mass, which performs rock mass structure layering and parameter calibration to generate a preliminary three-dimensional rock mass structure model; S5, the preliminary three-dimensional rock mass structure model is iteratively optimized by gravity and geomagnetism collaborative inversion depth network, and the topological relationship deviation is corrected by combining spatial topological constraint three-dimensional coding model, and the interface accuracy is adjusted by using nonlinear smooth fitting algorithm of stratigraphic interface; S6, based on the optimized three-dimensional rock mass structure data, the three-dimensional geological structure is intelligently modeled by the multi-scale integrated intelligent modeling and simulation platform for rock mass, and a complete three-dimensional geological structure model including rock mass parameters, stratigraphic interfaces and topological relationships is output.

[0011] Step S1 completes the acquisition of observational data and the generation of initial geophysical parameter fields, providing basic data support for subsequent modeling. Specifically, gravity and geomagnetic observation data for the target area are first collected using specialized geophysical equipment. The number of samples collected in a single batch is set to 2048 sets to ensure data coverage of every key area of ​​the target exploration region. After collection, the two types of raw observational data are directly input into the gravity-geomagnetic co-inversion depth network. This network adopts a fixed architecture of 3 input layers, 12 hidden layers, and 2 output layers. The hidden layers use a structure combining residual connections and an attention mechanism, where the channel attention weights of the attention mechanism are dynamically adjusted through a specific activation function. During network operation, following a pre-defined computational logic, gravity and geomagnetic observation data are first extracted independently to uncover key information reflecting the physical properties of underground geological bodies, such as density and magnetism. Then, a cross-dimensional correlation mapping algorithm is used to deeply fuse the features of the two types of data, eliminating data redundancy and enhancing effective information. Finally, an initial geophysical parameter field that can preliminarily reflect the distribution of physical properties of underground geological bodies is generated. This parameter field includes core information such as the density distribution and magnetic distribution of geological bodies, providing a data foundation for subsequent topological relationship construction and stratigraphic interface fitting. The entire process requires no manual intervention and achieves efficient data processing through the network's parallel computing capabilities.

[0012] Step S2 constructs the topological relationships between geological body units using a spatial topological constraint 3D coding model and completes the coding conversion, providing structured data support for stratigraphic interface fitting. During implementation, all geological body unit data from the initial geophysical parameter field are first called, and each geological body unit is accurately calibrated spatially. The center coordinates and boundary coordinates of each unit are extracted, and a complete unit coordinate dataset is established. Based on this dataset, the spatial positional relationships between different geological body units are analyzed one by one using a fusion of adjacency and distance matrices. The topological type of each unit—whether it belongs to an adjacent unit, an inclusive unit, or a separate unit—is clearly determined, forming a preliminary topological relationship determination result. Subsequently, a topological association matrix is ​​constructed based on this determination result, according to... For matrix elements of the same topological type, corresponding quantitative values ​​are assigned to achieve digital representation of topological relationships. Then, the topological correlation matrix is ​​fused with the core data in the initial geophysical parameter field element by element to ensure deep binding of topological relationship information and physical attribute information. The fused data is then mapped to a preset 64-dimensional feature space through an encoding conversion algorithm. During the mapping process, key feature information is retained and data dimensionality is compressed, ultimately generating a multi-dimensional topological feature matrix. This matrix includes both the physical attribute parameters of geological bodies and the topological correlation information between units, providing structured, high-dimensional data support for subsequent stratigraphic interface feature point screening and fitting. The entire process strictly follows spatial topological constraint rules to ensure the accuracy of topological relationship representation.

[0013] Step S3 achieves accurate fitting of formation interface feature points through a nonlinear smoothing fitting algorithm, obtaining continuous and smooth formation interface data to provide key interface information for rock mass structure modeling. Specifically, feature point data of the formation interface is first extracted from the multi-dimensional topological feature matrix using a feature filtering algorithm. The focus is on extracting the spatial coordinate parameters and corresponding physical property parameters of each feature point, forming a complete feature point parameter dataset. Then, statistical analysis is used to identify outliers in this dataset, filtering out feature points that deviate from the normal distribution range by more than three standard deviations and removing them to avoid invalid data affecting fitting accuracy. Next, the filtered valid feature point parameter dataset is input into the formation interface nonlinear smoothing fitting algorithm. The initial fitting coefficients, decay coefficients, and weight coefficients of the algorithm are determined using a particle swarm optimization strategy. The optimization iteration count was 100. The fitting calculation process was initiated. During the fitting process, the algorithm used a nonlinear fitting function to fit the feature points point by point based on the spatial distribution pattern and physical property correlation of the feature points. The fitting accuracy of key areas was enhanced by dynamically adjusting the weight coefficients of each feature point. After the fitting was completed, the interface continuity of the fitting results was verified, and the fitting error between adjacent feature points was calculated. If the error exceeded the preset threshold, the fitting coefficients were readjusted and the fitting operation was restarted until the fitting error of adjacent feature points was less than the preset threshold. Finally, continuous and smooth stratigraphic interface characterization data was obtained. This data can accurately reflect the spatial morphology and distribution pattern of the stratigraphic interface, providing an accurate interface basis for the subsequent stratification of rock mass structure. The entire process was optimized and verified in multiple rounds to ensure the accuracy and continuity of the stratigraphic interface fitting.

[0014] Step S4 utilizes a multi-scale integrated intelligent modeling and simulation platform to complete the stratification and preliminary modeling of the rock mass structure, establishing a basic framework for the three-dimensional geological structure. During implementation, continuous and smooth stratigraphic interface characterization data is first imported into the multi-scale integrated intelligent modeling and simulation platform via a high-speed data transmission protocol. This platform is equipped with eight parallel computing GPUs, supporting parallel processing of 2048 data samples per batch, demonstrating efficient data processing and modeling capabilities. Subsequently, the platform sets the criteria for stratifying the rock mass structure, including differences in stratigraphic interface depth, gradient changes in physical property parameters, and topological correlation types, ensuring the scientific and rational nature of the stratification. Based on these criteria, the platform uses an automatic stratification algorithm to divide the geological body within the target area layer by layer, classifying the rock mass units corresponding to different strata according to their spatial location and attribute characteristics, forming a stratified rock mass structure dataset containing information on each layer of rock mass units. Then, parameter calibration is performed on this dataset, invoking gravity and geomagnetism protocols. Using the geophysical parameters output by the inversion depth network as a calibration benchmark, the density, magnetism, and other physical property parameters of each rock mass unit are adjusted one by one to make the parameter values ​​more consistent with the actual geological conditions. After calibration, the layered rock mass structure data and stratigraphic interface characterization data are spatially aligned and fused to integrate the rock mass layering information, physical property parameters, and stratigraphic interface information. A preliminary three-dimensional rock mass structure model is constructed in the platform using the voxel modeling method. During the modeling process, an adaptive meshing strategy is adopted, and the number of mesh units is dynamically adjusted according to the complexity of the geological body. The mesh unit size is enlarged in simple geological areas and reduced in complex geological areas to ensure a balance between model accuracy and computational efficiency. The final generated preliminary three-dimensional rock mass structure model can fully present the layered structure of the rock mass, the morphology of the stratigraphic interface, and the physical property parameters of each unit, providing a basic model framework for subsequent iterative optimization. The entire process achieves efficient and accurate preliminary modeling through the platform's parallel computing capabilities and intelligent modeling algorithms.

[0015] Step S5 improves the accuracy of the 3D rock mass structure model through multi-technology collaborative iterative optimization and corrects deviations in the preliminary model. In this process, the preliminary 3D rock mass structure model is first input into a gravity-geomagnetic collaborative inversion depth network. The network calculates the corresponding gravity-geomagnetic response data based on the rock mass physical property parameters and spatial distribution in the model. This response data is then compared point-by-point with the original observation data, and the difference between the two is calculated to generate a deviation matrix. The deviation matrix identifies areas in the model that do not match the actual observation data. Subsequently, based on the magnitude and distribution of deviations in the deviation matrix, a spatial topological constraint 3D coding model is used to correct the topological relationships in the preliminary model. The spatial positions and association methods of geological units in areas with large deviations are adjusted to make the topological relationships between units more consistent with actual geological structures. Finally, a nonlinear smoothing fitting algorithm for stratigraphic interfaces is used to correct deviations exceeding a certain value in the deviation matrix. The stratigraphic interface corresponding to the region with a preset threshold is subjected to secondary fitting. The interface morphology and feature point distribution of the region are re-optimized to improve the fitting accuracy of the local interface. After completing one round of deviation calculation, topology correction and interface optimization, the corrected model is input into the gravity-geomagnetic co-inversion depth network again, and the above deviation calculation, topology correction and interface optimization process is repeated. In each iteration, the optimization parameters are dynamically adjusted until the overall deviation between the gravity-geomagnetic response data output by the model and the original observation data is less than the preset threshold. The preset threshold is set to 5% of the mean of the original observation data. Finally, the optimized three-dimensional rock mass structure data is obtained. This data is highly consistent with the actual geological conditions in terms of physical property parameters, topological relationships and stratigraphic interface morphology, which greatly improves the reliability and accuracy of the model. The entire iterative optimization process systematically corrects various deviations of the preliminary model through multi-technology collaboration, ensuring the quality of the final model.

[0016] Step S6 completes the construction and output of a complete three-dimensional geological structure model through a multi-scale integrated intelligent modeling and simulation platform for rock mass, achieving the modeling goal. Specifically, the optimized three-dimensional rock mass structure data is first imported into the platform. The platform first verifies the integrity of the input data, confirming that it includes the physical property parameters, topological association information, and stratigraphic interface characterization data of all rock mass units. Then, it initiates the intelligent modeling process for the three-dimensional geological structure. During modeling, the platform employs adaptive mesh generation technology, dynamically adjusting the number of mesh units according to the complexity of the geological body. The number of mesh units in complex geological structures can reach up to millions, while in simple geological areas, the number is controlled to below one hundred thousand, ensuring both modeling accuracy and computational efficiency. Subsequently, the platform integrates and processes the optimized three-dimensional rock mass structure data, arranging rock mass parameters, stratigraphic interfaces, and topological relationship information according to spatial coordinates. By binding data lines together, a unified 3D data model is constructed. During the model building process, the platform uses multi-threaded parallel computing to complete the rendering and generation of the model, ensuring the spatial continuity and data consistency of the model. After generation, the model undergoes comprehensive quality testing, including multiple indicators such as the rationality of rock mass unit parameters, the smoothness of stratigraphic interfaces, and the accuracy of topological relationships. Once all test indicators meet the preset standards, the platform outputs the complete 3D geological structure model in a standard data format. The output data includes physical property parameter files for each rock mass unit, spatial coordinate files for stratigraphic interfaces, and topological relationship description files. This model can comprehensively and accurately present the 3D geological structure characteristics of the target area, providing reliable digital support for subsequent geological analysis, resource exploration, engineering design, and other work. The entire process achieves fully automated processing from data input to model output, ensuring the accuracy and practicality of the modeling results.

[0017] Preferably, the expression for the gravity-geomagnetic co-inversion depth network is: Where G is the gravity observation data matrix, M is the geomagnetic observation data matrix, and W is the... g W is the weight matrix for gravity data. m This is the weight matrix for geomagnetic data. Here, x is the gradient operator, and x is the spatial coordinate vector. σ W is the activation function. f Let b be the feature mapping weight matrix. f Given the bias vector; the optimized formula for the inversion parameters is calculated as follows: ,in For the optimal geophysical parameter field, It is the L2 norm. λ is the regularization coefficient, tr is the trace operator, and R is the smoothing constraint matrix.

[0018] Specifically, the gravity-geomagnetic co-inversion deep network, based on the physical correlation between gravity and geomagnetic data, first clarifies the complementarity of the two types of data in reflecting the physical properties of underground geological bodies. It then mines the spatial variation characteristics of the data through gradient operations, dynamically allocates the contribution of the two types of data using a weight matrix, and combines this with an activation function to achieve nonlinear mapping of features. Finally, it integrates the features of the entire data domain through integral operations to form an initial expression that can characterize the distribution of geophysical parameters. This construction logic is adopted because gravity and geomagnetic data correspond to the density and magnetic properties of geological bodies, respectively. Their co-inversion improves the comprehensiveness of parameter inversion, gradient operations enhance the spatial variation information of the data, the weight matrix can adapt to the reliability differences of data in different regions, and the activation function solves the problem that linear models struggle to fit complex geological features. The element values ​​of the gravity data weight matrix and the geomagnetic data weight matrix were determined through iterative optimization using training samples, with initial values ​​controlled between 0.1 and 0.9. The feature mapping weight matrix dimension was set to 256×128, the initial value of the bias vector elements was set to 0.01, and the regularization coefficient was set to 0.001. The smoothing constraint matrix was constructed using an identity matrix combined with a spatial distance attenuation factor. In implementation, the collected gravity and geomagnetic data were first input into the network. Feature extraction and association mapping were completed through an initial expression. Then, based on minimizing the error between the inversion result and the ideal parameter field using an optimization formula, a smoothing constraint was introduced through a trace operator to avoid overfitting. Finally, the optimal geophysical parameter field was output, providing an accurate physical parameter foundation for subsequent topology modeling.

[0019] Preferably, the expression for the spatial topological constraint three-dimensional encoding model is: ,in The topological feature matrix, N For the number of geological bodies, M The number of feature points for each geological body. For elements of the topological correlation matrix, This is a topological mapping function. and The first i The and the first j The parameter vector of each feature point The elements of the encoding transformation matrix are used; the encoding optimization formula is calculated as follows: ,in To optimize the topological feature matrix, This is the topological encoding weight matrix. This is the encoding bias vector.

[0020] Specifically, the spatial topological constraint 3D coding model focuses on the spatial positional relationships between geological body units. It first traverses all geological bodies and feature points through a double summation operation, quantifies the correlation strength between units using topological correlation matrix elements, and transforms physical parameters and spatial positional relationships into calculable numerical relationships through a topological relationship mapping function. Then, it combines this with a coding transformation matrix to complete the initial adjustment of feature dimensions, forming the core expression of the topological feature matrix. The topological relationships (adjacent, included, separated) of geological body units in a 3D geological structure directly affect modeling accuracy. The double summation comprehensively covers all unit combinations, the topological correlation matrix and mapping function enable the digital representation of topological relationships, and the coding transformation matrix provides an adaptive dimension for subsequent feature processing. The number of geological bodies is determined based on the exploration range of the target area, with no less than 50 geological bodies per area. The number of feature points for each geological body is set to 100 to 200. The topological correlation matrix elements take values ​​of 0, 1, and 2, corresponding to separated, adjacent, and included relationships, respectively. The coding transformation matrix dimension is set to 64×64, the topological coding weight matrix dimension is 64×32, and the bias vector element value range is -0.05 to 0.05. During implementation, unit coordinates and attribute information are first extracted based on the initial geophysical parameter field. The topological and attribute features of all units are integrated through a double summation operation to generate an initial topological feature matrix. Then, the features are optimized by introducing encoding weights and biases in an optimized manner, and the matrix is ​​mapped to a preset 64-dimensional feature space. Finally, the optimized multi-dimensional topological feature matrix is ​​output, providing structured topological data support for stratigraphic interface fitting.

[0021] Preferably, the expression for the nonlinear smoothing fitting algorithm for the formation interface is: in Formation interface function, The coordinate vector of the interface feature points. The total number of feature points. These are the fitting coefficients. The attenuation coefficient is... These are the weighting coefficients. For the first The coordinate vector of each feature point It is an exponential function.

[0022] Specifically, the nonlinear smoothing fitting algorithm for formation interfaces is based on the continuous and smooth physical characteristics of formation interfaces. It integrates the contributions of all feature points to the interface using a multiplication operation, constructs the spatial attenuation effect of feature points through an exponential function, and dynamically adjusts the influence of each feature point by combining the fitting coefficient, attenuation coefficient, and weight coefficient. Simultaneously, gradient operations are introduced to enhance the spatial continuity of the interface, forming a nonlinear fitting function. The spatial morphology of the formation interface is determined by a large number of feature points. The multiplication operation integrates the constraints of all feature points, the exponential function simulates the physical law of feature point influence attenuation with increasing distance, and the gradient operation ensures the smooth continuity of the interface, avoiding abrupt changes or breaks. The total number of feature points is determined based on the complexity of the formation interface, with no less than 300 feature points per interface. The fitting coefficient is determined through 100 iterations using a particle swarm optimization strategy, ranging from 0.3 to 0.8. The attenuation coefficient ranges from 0.01 to 0.05, and the weight coefficient is assigned based on the reliability of the feature points, with a weight coefficient of no less than 0.7 for highly reliable feature points. During implementation, effective stratigraphic interface feature points are first selected from the topological feature matrix. After removing outliers, the initial values ​​of each parameter are determined. The spatial and attribute information of all feature points are integrated through multiplication operations. The interface morphology is constrained by exponential functions and gradient operations. The coefficients are dynamically adjusted to ensure that the fitting results meet the continuity requirements. The error verification of adjacent feature points ensures that the fitted stratigraphic interface characterization data is continuous and smooth, accurately reflects the actual spatial morphology of the stratigraphy, and provides reliable interface data for rock mass structure modeling.

[0023] Preferably, the gravity-geomagnetic collaborative inversion deep network adopts a network architecture of 3 input layers, 12 hidden layers, and 2 output layers. The hidden layers adopt a structure combining residual connections and an attention mechanism, wherein the channel attention weights of the attention mechanism are dynamically adjusted by the Sigmoid function. The encoding dimension of the spatial topological constraint 3D encoding model is set to 64 dimensions, and the topological relationship is constructed by fusing adjacency matrices and distance matrices. The fitting coefficients of the nonlinear smoothing fitting algorithm for the stratigraphic interface are determined by a particle swarm optimization strategy, and the number of optimization iterations is set to 100. The multi-scale rock mass integrated intelligent modeling and simulation platform is equipped with GPU cluster hardware, including 8 parallel computing GPUs, supporting parallel processing of 2048 data samples per batch. The mesh division accuracy of the 3D geological structure intelligent modeling is set to adaptive mesh, and the number of mesh units is dynamically adjusted according to the complexity of the geological body.

[0024] Preferred, such as Figure 2As shown, S2 includes the following sub-steps: S21, spatial coordinate calibration of each geological body unit in the initial geophysical parameter field, extraction of unit center coordinates and boundary coordinates, and establishment of unit coordinate dataset; S22, based on the unit coordinate dataset, analysis of the spatial positional relationship between different geological body units, determination of the topological types of adjacent units, including units and separated units, and formation of preliminary topological relationship determination results; S23, based on the preliminary topological relationship determination results, construction of a topological correlation matrix, with matrix elements assigned corresponding values ​​according to the topological type, to perform quantitative characterization of topological relationships; S24, fusion of the topological correlation matrix and the geophysical parameter field data, mapping the fused data to a feature space of a preset dimension through an encoding conversion algorithm, and generating a multi-dimensional topological feature matrix.

[0025] Specifically, the topological relationship construction and encoding conversion process in step S2 achieves precise transformation from geological body unit data to a multi-dimensional topological feature matrix through four sub-steps, providing structured and digital topological relationship support for subsequent stratigraphic interface fitting. The implementation first performs S21, calibrating the spatial coordinates of all geological body units in the initial geophysical parameter field one by one, accurately extracting the center and boundary coordinates of each unit, and establishing a complete unit coordinate dataset in unit number order to ensure that no spatial location information of each unit is omitted. Then, in step S22, based on the constructed unit coordinate dataset, the spatial location association between any two geological body units is analyzed one by one through spatial distance calculation and positional relationship determination algorithms, clearly determining the topological type of each unit as an adjacent unit, including unit, or separate unit, forming a comprehensive preliminary topological relationship determination result. Finally, step S23 is carried out, constructing the topological relationship matrix based on the preliminary determination result. The matrix assigns corresponding quantitative values ​​to different topological types, with separation relationships assigned a value of 0, adjacency relationships assigned a value of 1, and inclusion relationships assigned a value of 2, thus realizing the digital and matrix representation of topological relationships. Finally, S24 is executed to fuse the topological correlation matrix with the core data such as density and magnetism in the initial geophysical parameter field element by element according to the unit number. The fused data is mapped to the preset 64-dimensional feature space through an encoding conversion algorithm. During the mapping process, key feature information is retained and data dimensionality is compressed, ultimately generating a multi-dimensional topological feature matrix. This matrix includes both the physical properties of geological bodies and spatial topological relationships, laying the data foundation for subsequent stratigraphic interface feature point screening and fitting processing.

[0026] Preferred, such as Figure 3As shown, step S3 includes the following sub-steps: S31, selecting formation interface feature point data from the multi-dimensional topological feature matrix, extracting the coordinate parameters and physical property parameters of the feature points to form a feature point parameter dataset; S32, identifying outliers in the feature point parameter dataset, filtering out feature points that deviate from the normal distribution range through statistical analysis, and removing invalid data to ensure fitting accuracy; S33, inputting the filtered feature point parameter dataset into the formation interface nonlinear smoothing fitting algorithm, setting the initial fitting coefficients and iteration parameters of the algorithm, and starting the fitting calculation process; S34, verifying the interface continuity of the fitting calculation results, adjusting the fitting coefficients through fitting error analysis of adjacent feature points until continuous and smooth formation interface characterization data is obtained.

[0027] Specifically, step S3, the nonlinear smoothing fitting process of the stratigraphic interface, ensures the continuous smoothness of the stratigraphic interface characterization data, providing a reliable interface basis for rock mass structure modeling. During implementation, S31 is performed first, using feature recognition and filtering algorithms to accurately extract feature point data that characterizes the morphology of the stratigraphic interface from the multi-dimensional topological feature matrix. The focus is on extracting the three-dimensional spatial coordinate parameters and corresponding physical property parameters of each feature point, organizing them into a feature point parameter dataset according to spatial location. Then, S32 is executed, using statistical analysis methods to identify outliers in the feature point parameter dataset. The mean and standard deviation of the dataset are calculated, and feature points deviating from the normal distribution range by more than three standard deviations are identified as outliers and removed to avoid invalid data negatively impacting fitting accuracy. Finally, S33 is performed, inputting the filtered valid feature point parameter dataset into the nonlinear smoothing fitting process of the stratigraphic interface. The smooth fitting algorithm uses a particle swarm optimization strategy to determine the initial fitting coefficients, decay coefficients, and weight coefficients. The optimization iteration count is set to 100. The fitting calculation process is then initiated, and the algorithm performs nonlinear fitting based on the spatial distribution of feature points and the correlation of physical properties. Finally, step S34 is executed to verify the interface continuity of the fitting calculation results. The spatial distance error and attribute parameter error between adjacent feature points are calculated. If the error exceeds a preset threshold, the algorithm returns to step S33 to readjust the fitting coefficients and restart the fitting operation. This process is repeated iteratively until the fitting error of adjacent feature points is less than the preset threshold, ultimately yielding continuous and smooth stratigraphic interface characterization data that accurately reflects the actual spatial morphology of the stratigraphic interface.

[0028] Preferred, such as Figure 4As shown, step S4 includes the following sub-steps: S41, importing stratigraphic interface characterization data into a multi-scale integrated intelligent modeling and simulation platform for rock mass, and setting the criteria for dividing the rock mass structure into layers, including stratigraphic interface depth, differences in physical properties, and topological relationships; S42, based on the division criteria, the platform automatically performs layering processing on the geological body, classifying the rock mass units corresponding to different strata to form a layered rock mass structure dataset; S43, performing parameter calibration on the layered rock mass structure dataset, and adjusting the physical property parameters of each rock mass unit by combining the geophysical parameters output by the gravity-geomagnetic collaborative inversion depth network; S44, fusing the calibrated layered rock mass structure data with the stratigraphic interface characterization data to construct a preliminary three-dimensional rock mass structure model including rock mass layering information and interface information.

[0029] Specifically, step S4, the preliminary modeling process of the three-dimensional rock mass structure, achieves rock mass stratification, parameter calibration, and preliminary modeling through systematic operations, establishing the basic framework of the three-dimensional geological structure and providing an initial model for iterative optimization. During implementation, step S41 is executed first, importing continuous and smooth stratigraphic interface characterization data into a multi-scale integrated intelligent modeling and simulation platform for rock masses via a high-speed data transmission protocol. This platform is equipped with eight parallel computing GPUs, supporting parallel processing of 2048 data samples per batch. The platform clearly defines the criteria for dividing the rock mass structure into layers, including differences in stratigraphic interface depth, gradient changes in physical property parameters, and topological correlation types, ensuring the scientific validity of the stratification. Then, step S42 is performed. Based on the defined stratification criteria, the platform uses an automatic stratification algorithm to divide the geological bodies within the target area layer by layer. Following the order of stratigraphic interface depth from shallow to deep, the rock mass units corresponding to different strata are classified and organized, forming a layered rock mass structure dataset including unit number, spatial location, and physical properties. Next, step S43 is performed to calibrate the parameters of the layered rock mass structure dataset. The geophysical parameters output by the gravity-geomagnetic co-inversion depth network are used as the calibration benchmark. The density, magnetism, and other physical property parameters of each rock mass unit are adjusted one by one to make the parameter values ​​more consistent with the actual geological conditions. Finally, step S44 is executed to align the calibrated layered rock mass structure data with the stratigraphic interface characterization data in spatial coordinates. The rock mass layering information, physical property parameters, and stratigraphic interface information are integrated through a data fusion algorithm. A preliminary three-dimensional rock mass structure model is constructed in the platform using the voxel modeling method. An adaptive meshing strategy is adopted during the modeling process, and the number of mesh units is dynamically adjusted according to the complexity of the geological body. Finally, a preliminary model that can fully present the layered rock mass structure, stratigraphic interface morphology, and unit physical properties is generated.

[0030] Preferred, such as Figure 5As shown, step S5 includes the following sub-steps: S51, inputting the preliminary three-dimensional rock mass structure model into the gravity-geomagnetic collaborative inversion depth network, calculating the deviation between the gravity-geomagnetic response data output by the model and the observation data, and generating a deviation matrix; S52, based on the deviation matrix, correcting the topological relationships in the preliminary three-dimensional rock mass structure model through a spatial topological constraint three-dimensional coding model, and adjusting the spatial position and association mode of the geological units; S53, using a nonlinear smoothing fitting algorithm for stratigraphic interfaces, performing secondary fitting on the stratigraphic interfaces in areas with large deviations, and optimizing the interface morphology and accuracy; S54, repeating the deviation calculation, topology correction, and interface optimization process until the deviation between the gravity-geomagnetic response data output by the model and the observation data meets a preset threshold, thereby obtaining the optimized three-dimensional rock mass structure data.

[0031] Specifically, the model iteration and optimization process in step S5 achieves gradual improvement in model accuracy through multi-technology collaboration, corrects deviations in the preliminary model, and ensures the accuracy and reliability of the final modeling results. During implementation, S51 is executed first, inputting the preliminary 3D rock mass structure model into a gravity-geomagnetic collaborative inversion depth network. The network calculates the corresponding gravity-geomagnetic response data based on the rock mass physical property parameters and spatial distribution in the model. This response data is then compared point-by-point with the original gravity and geomagnetic observation data, calculating the difference between the two and generating a deviation matrix based on spatial location. The deviation matrix identifies areas in the model that do not match the actual observation data. Next, in step S52, based on the magnitude and spatial distribution of deviations in the deviation matrix, a spatial topological constraint 3D coding model is invoked to correct the topological relationships in the preliminary model. For areas with large deviations, the spatial location of the corresponding geological units and the association methods between units are adjusted to make the topological relationships between units more consistent with actual geological structural patterns. Then, step S53 is carried out. Using a nonlinear smoothing fitting algorithm for stratigraphic interfaces, a secondary fitting is performed on the stratigraphic interfaces corresponding to areas where the deviation value in the deviation matrix exceeds a preset threshold. The distribution of interface feature points and the interface morphology in this area are re-optimized to improve the fitting accuracy of local interfaces. Finally, S54 is executed, and the model after secondary correction is input into the gravity-geomagnetic co-inversion depth network again. The above deviation calculation, topology correction and interface optimization process is repeated. In each iteration, the optimization parameters are dynamically adjusted according to the deviation changes. The preset deviation threshold is 5% of the mean of the original observation data. The iteration stops when the overall deviation between the gravity-geomagnetic response data output by the model and the original observation data is less than the threshold. The optimized three-dimensional rock mass structure data is then output. This data is highly consistent with the actual geological conditions in terms of physical property parameters, topological relationships and stratigraphic interface morphology.

[0032] like Figure 6As shown, a deep learning-based intelligent modeling system for 3D geological structures is applied. This system includes: a gravity and geomagnetic data acquisition and preprocessing unit, used to acquire gravity and geomagnetic observation data and perform data filtering, establishing a data transmission connection with a gravity and geomagnetic co-inversion deep network computing unit, which can use existing equipment such as relative gravimeters, three-component in-well magnetometers, and GPS positioning modules; a gravity and geomagnetic co-inversion deep network computing unit, used to extract features and perform correlation mapping on the acquired observation data, generating an initial geophysical parameter field, and establishing bidirectional data interaction with a spatial topology-constrained 3D encoding model processing unit and an iterative optimization unit; and a spatial topology-constrained 3D encoding model processing unit, used to construct and encode the topological relationships of the initial geophysical parameter field, and output the data. The system generates a multi-dimensional topological feature matrix and establishes a data transmission connection with the nonlinear smoothing fitting algorithm unit for stratigraphic interfaces. This unit then performs fitting processing on the stratigraphic interface feature points in the topological feature matrix, generating stratigraphic interface characterization data, and establishes a data transmission connection with the multi-scale integrated intelligent modeling and simulation unit for rock masses. The multi-scale integrated intelligent modeling and simulation unit receives the stratigraphic interface characterization data and performs rock mass structure layering, parameter calibration, and model construction, establishing data interaction with the iterative optimization unit and the 3D geological structure output unit, respectively. The 3D geological structure output unit receives the optimized 3D rock mass structure data, integrates rock mass parameters, stratigraphic interfaces, and topological relationship information, and outputs a complete 3D geological structure model. All units synchronously transmit data and collaborate via a high-speed data bus. The gravity and geomagnetism co-inversion deep network computing unit, the spatial topology constraint 3D coding model processing unit, the nonlinear smoothing fitting algorithm operation unit for stratigraphic interfaces, the multi-scale rock mass integrated intelligent modeling and simulation unit, and the 3D geological structure output unit are all mounted on the main computer. During specific calculations, multiple deep learning servers, GPU clusters, solid-state drives, and other hardware facilities can be set up; at the same time, relevant software is installed on the hardware to run.

[0033] This invention presents a deep learning-based intelligent modeling method and system for 3D geological structures. It employs a gravity-geomagnetic collaborative inversion architecture to simultaneously mine the correlation features between the two types of data. Multi-layer networks and dynamic weight adjustments enhance the accuracy of data inversion. Furthermore, spatial topological constraint coding technology is combined to quantify the spatial relationships between geological body units through matrix fusion, representing complex topological relationships and overcoming the limitations of traditional single-data processing and ambiguous topological descriptions. Through nonlinear smoothing fitting techniques at stratigraphic interfaces, fitting parameters are determined through optimization strategies and repeatedly verified and adjusted to ensure the continuity and smoothness of stratigraphic interfaces. Coupled with the parallel computing capabilities of a multi-scale integrated modeling and simulation platform, it achieves efficient processing of large-scale data and adaptive mesh generation, improving the accuracy of complex geological body modeling.

[0034] This method and system address the problems of insufficient data utilization and inversion accuracy in traditional methods. It deeply integrates gravity and geomagnetic data through a collaborative inversion network, and combines topological constraint coding to achieve accurate mapping between geological parameters and spatial relationships, significantly reducing modeling bias. To address the shortcomings of low modeling efficiency and insufficient intelligence, a multi-unit collaborative system architecture is constructed, integrating a fully automated linkage mechanism for data acquisition, feature extraction, model building, and iterative optimization. Through GPU cluster parallel processing and particle swarm optimization strategies, manual intervention is reduced, shortening the modeling cycle. Simultaneously, residual connections and attention mechanisms are used to optimize network performance, and multiple rounds of iterative optimization correct model biases, ensuring the consistency and reliability of modeling results and providing an efficient and accurate technical solution for complex geological scenarios.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A deep learning-based intelligent modeling method for three-dimensional geological structures, characterized in that, Includes the following steps: S1. Acquire gravity and geomagnetic observation data, and extract and map features from the data using a gravity-geomagnetic co-inversion depth network to generate an initial geophysical parameter field. S2. Based on a spatial topological constraint 3D coding model, construct and encode the topological relationships of the initial geophysical parameter field to form a multi-dimensional topological feature matrix. S3. Use a nonlinear smoothing fitting algorithm for stratigraphic interfaces to fit the stratigraphic interface feature points in the multi-dimensional topological feature matrix, obtaining continuous and smooth stratigraphic interface characterization data. S4. Input the stratigraphic interface characterization data into a multi-scale integrated intelligent modeling and simulation platform for rock mass layering and parameter calibration, generating a preliminary 3D rock mass structure model. S5. Iteratively optimize the preliminary 3D rock mass structure model using a gravity-geomagnetic co-inversion depth network, correct topological relationship deviations using a spatial topological constraint 3D coding model, and adjust interface accuracy using a nonlinear smoothing fitting algorithm for stratigraphic interfaces. S6, based on optimized rock mass 3D structure data, completes 3D geological structure intelligent modeling through a multi-scale integrated intelligent modeling and simulation platform for rock mass, and outputs a complete 3D geological structure model including rock mass parameters, stratigraphic interfaces and topological relationships.

2. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, The expression for the gravity-geomagnetic co-inversion depth network is: Where G is the gravity observation data matrix, M is the geomagnetic observation data matrix, and W is the... g W is the weight matrix for gravity data. m This is the weight matrix for geomagnetic data. Here, x is the gradient operator, and x is the spatial coordinate vector. σ W is the activation function. f Let b be the feature mapping weight matrix. f Here is the bias vector; the optimized formula for the inversion parameters is calculated as follows: ,in For the optimal geophysical parameter field, It is the L2 norm. λ is the regularization coefficient, tr is the trace operator, and R is the smoothing constraint matrix.

3. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, The expression for the spatial topological constraint three-dimensional encoding model is: ,in The topological feature matrix, For the number of geological bodies, The number of feature points for each geological body. For elements of the topological correlation matrix, This is a topological mapping function. and The first The and the first The parameter vector of each feature point The elements of the encoding transformation matrix are used; the encoding optimization formula is calculated as follows: ,in To optimize the topological feature matrix, The topological encoding weight matrix, This is the encoding bias vector.

4. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, The expression for the nonlinear smoothing fitting algorithm for the formation interface is: in Formation interface function, The coordinate vector of the interface feature points. The total number of feature points. These are the fitting coefficients. The attenuation coefficient is... These are the weighting coefficients. For the first The coordinate vector of each feature point It is an exponential function.

5. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, The gravity-geomagnetic co-inversion deep network adopts a network architecture of 3 input layers, 12 hidden layers, and 2 output layers. The hidden layers use a structure combining residual connections and an attention mechanism, where the channel attention weights of the attention mechanism are dynamically adjusted using the Sigmoid function. The encoding dimension of the spatial topological constraint 3D encoding model is set to 64 dimensions, and the topological relationship is constructed using a fusion of adjacency matrix and distance matrix. The fitting coefficients of the nonlinear smoothing fitting algorithm for stratigraphic interfaces are determined by a particle swarm optimization strategy, and the number of optimization iterations is set to 100. The mesh generation accuracy of the intelligent modeling of 3D geological structures is set to adaptive mesh, and the number of mesh cells is dynamically adjusted according to the complexity of the geological body.

6. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, S2 includes the following sub-steps: S21, spatial coordinate calibration of each geological body unit in the initial geophysical parameter field, extraction of unit center coordinates and boundary coordinates, and establishment of unit coordinate dataset; S22, based on the unit coordinate dataset, analysis of the spatial positional relationship between different geological body units, determination of the topological type of adjacent units, including units and separate units, and formation of preliminary topological relationship determination results; S23, based on the preliminary topological relationship determination results, construction of a topological correlation matrix, with matrix elements assigned corresponding values ​​according to the topological type, to perform quantitative characterization of topological relationships; S24, fusion of the topological correlation matrix and the geophysical parameter field data, mapping the fused data to a feature space of a preset dimension through an encoding conversion algorithm, and generating a multi-dimensional topological feature matrix.

7. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, S3 includes the following sub-steps: S31, selecting formation interface feature point data from the multi-dimensional topological feature matrix, extracting the coordinate parameters and physical property parameters of the feature points to form a feature point parameter dataset; S32, identifying outliers in the feature point parameter dataset, filtering out feature points that deviate from the normal distribution range through statistical analysis, and removing invalid data to ensure fitting accuracy; S33, inputting the filtered feature point parameter dataset into the formation interface nonlinear smoothing fitting algorithm, setting the initial fitting coefficients and iteration parameters of the algorithm, and starting the fitting calculation process; S34, verifying the interface continuity of the fitting calculation results, adjusting the fitting coefficients through fitting error analysis of adjacent feature points until continuous and smooth formation interface characterization data is obtained.

8. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, S4 includes the following sub-steps: S41, importing stratigraphic interface characterization data into a multi-scale integrated intelligent modeling and simulation platform for rock mass, and setting the criteria for dividing the rock mass structure into layers, including stratigraphic interface depth, differences in physical properties, and topological relationships; S42, based on the criteria, the platform automatically performs layering processing on the geological body, classifying the rock mass units corresponding to different strata to form a layered rock mass structure dataset; S43, performing parameter calibration on the layered rock mass structure dataset, and adjusting the physical property parameters of each rock mass unit in combination with the geophysical parameters output by the gravity-geomagnetic collaborative inversion depth network. S44. The calibrated layered rock mass structure data and stratigraphic interface characterization data are fused to construct a preliminary three-dimensional rock mass structure model that includes rock mass layering information and interface information.

9. The intelligent modeling method for three-dimensional geological structures based on deep learning according to claim 1, characterized in that, S5 includes the following sub-steps: S51, inputting the preliminary three-dimensional rock mass structure model into the gravity-geomagnetic collaborative inversion depth network, calculating the deviation between the gravity-geomagnetic response data output by the model and the observation data, and generating a deviation matrix; S52, based on the deviation matrix, correcting the topological relationships in the preliminary three-dimensional rock mass structure model through a spatial topological constraint three-dimensional coding model, and adjusting the spatial position and association mode of the geological body units; S53, using a nonlinear smoothing fitting algorithm for stratigraphic interfaces, performing secondary fitting on stratigraphic interfaces in areas with large deviations, and optimizing the interface morphology and accuracy; S54, repeat the deviation calculation, topology correction and interface optimization process until the deviation between the gravity and geomagnetic response data output by the model and the observation data meets the preset threshold, and obtain the optimized three-dimensional rock mass structure data.

10. A deep learning-based intelligent modeling system for three-dimensional geological structures, characterized in that: This system is applied to the deep learning-based intelligent modeling method for three-dimensional geological structures as described in claim 1, comprising: a gravity and geomagnetic data acquisition and preprocessing unit, used to acquire gravity observation data and geomagnetic observation data and perform data filtering, and establish a data transmission connection with the gravity and geomagnetic co-inversion deep network computing unit; a gravity and geomagnetic co-inversion deep network computing unit, used to extract features and perform correlation mapping on the acquired observation data, generate an initial geophysical parameter field, and establish bidirectional data interaction with the spatial topology-constrained three-dimensional coding model processing unit and the iterative optimization unit respectively; and a spatial topology-constrained three-dimensional coding model processing unit, used to construct and encode the topological relationships of the initial geophysical parameter field, output a multi-dimensional topological feature matrix, and perform nonlinear smooth fitting calculation with the stratigraphic interface. The algorithm calculation unit establishes a data transmission connection; the nonlinear smoothing fitting algorithm calculation unit for stratigraphic interfaces is used to fit the stratigraphic interface feature points in the topological feature matrix, generate stratigraphic interface characterization data, and establish a data transmission connection with the multi-scale rock mass integrated intelligent modeling and simulation unit; the multi-scale rock mass integrated intelligent modeling and simulation unit is used to receive stratigraphic interface characterization data and perform rock mass structure layering, parameter calibration, and model construction, and establishes data interaction with the iterative optimization unit and the three-dimensional geological structure output unit respectively; the three-dimensional geological structure output unit is used to receive the optimized three-dimensional rock mass structure data, integrate rock mass parameters, stratigraphic interfaces, and topological relationship information, and output a complete three-dimensional geological structure model. All units synchronize data transmission and work collaboratively through a high-speed data bus.

Images