A method for dynamically constructing coal mine geological digital twins based on multi-source heterogeneous information

Abstract

The present invention provides a method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information, including AIGC-assisted reading of geological results, end-to-end adaptive generation of structured information, construction of a primary geological body model using a generalized triangular prism algorithm, geometric smoothing and subdivision based on Kriging interpolation, dynamic updating of special geological elements, and generation of a coal mine digital twin matrix. Innovatively, based on the significant characteristics of the data-intensive paradigm of geological information, the present invention introduces AIGC-assisted reading of geological results. By assimilating big data from multi-source heterogeneous information, it achieves intelligent generation of structured information that matches the automated modeling and updating algorithm. This forms a rapid response process of information processing - primary geological body generation - geometric smoothing - encryption of abnormal geological bodies - global geological body construction. This method dynamically constructs a high-precision coal mine geological digital model using geological information from different geological survey stages, meeting the requirements of coal mine digital twins for high precision, updatability, and automation.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine digital geological model construction, and in particular to a method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information. Background Art

[0002] With the advancement of intelligent coal mining, the industry has developed a digital twin technology approach for coal mines, relying on three-dimensional geological modeling and integrating full-time mining data, achieving the transformation from cognition to application. However, the geological modeling process is plagued by numerous non-endpoint links and high maintenance costs. In practical applications, the model cannot be dynamically adjusted as geological knowledge is updated. The lag in the model construction layer affects the equivalence of the twin system. Therefore, how to quickly and dynamically construct a digital twin that can adapt to multi-time, multi-precision, and multi-format geological information has become a core issue that needs to be addressed in the full realization of intelligent coal mine construction.

[0003] 3D geological modeling technology for coal mines is relatively mature and can meet the demand for modeling accuracy. However, due to the multi-source, heterogeneous nature of ground and air data, data preprocessing requirements are high, information utilization is low, and maintenance requires professionals with both geological knowledge and computer skills. Consequently, there is still significant room for improvement in response speed and maintenance costs.

[0004] In digital twin technology, 3D geological models, as the model construction layer of digital twin technology, assume the geometric boundaries and physical structure roles of the computational layer. The accuracy of existing 3D geological modeling technology has met application requirements. However, with the advancement of excavation and exploration, the update speed of geological knowledge information has not matched the response speed of the model construction layer. This has led to a lag in the "knowledge-to-model" transformation, affecting the equivalence of the twin world and the accuracy of analytical decisions.

[0005] Dynamically constructing coal mine geological digital twins involves using "while drilling, while excavating, while mining, and while dropping" exploration technologies to acquire comprehensive geological information at different stages and combine them with various modeling algorithms to generate three-dimensional quantitative stochastic models. While these methods have achieved some progress, due to information disparities between each stage and limitations in modeling techniques, big data isomorphism of dynamic geological information from multiple sources along the "four-while" model remains elusive. Future research should promote the integration of multi-scale, multi-attribute data through fusion strategies at the data, feature, and decision levels, enabling dynamic correction and improvement of coal mine geological conditions and thus promoting the construction of coal mine digital twins.

[0006] In summary, the dynamic integration of multi-source heterogeneous geological information and the construction of coal mine digital twins are important research directions for realizing the intelligent construction of coal mines, and have important theoretical value and practical significance. Summary of the Invention

[0007] Through research, the inventors of this application have discovered that digital twin technology, by integrating three-dimensional modeling and real-time data, can significantly improve the efficiency of coal mines in risk assessment, decision support, and mining scheduling. However, existing geological modeling methods face problems such as delayed response and high maintenance costs, which limit the accuracy and practicality of digital twin systems. Therefore, how to dynamically update and integrate geological data from multiple time periods, multiple precisions, and multiple formats has become a key challenge in the intelligent construction of coal mines. To address this problem, the present invention proposes a method for dynamically constructing coal mine geological digital twins based on multi-source heterogeneous information. This method introduces AIGC (Artificial Intelligence Generated Content) technology to assist in reading geological results. By building an end-to-end structured information adaptive generation system, the data processing process is optimized. The generalized tri-prism (GTP) algorithm is used to construct a primary geological body model, and the Kriging interpolation method is combined to smoothly subdivide the geometric model. On this basis, special geological elements such as faults, collapse columns, and anticlines are dynamically updated to ultimately construct the coal mine digital twin matrix. Through big data assimilation technology, this method can automatically process multi-source heterogeneous geological information, generate high-precision digital geological models, and realize real-time updating and optimization of the models through a rapid response mechanism, meeting the requirements of coal mine digital twins in terms of high precision, updatability and automated construction, and providing strong technical support for intelligent management and decision support of coal mines.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0009] The method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information includes the following steps:

[0010] S1. AIGC assists in reading geological results: Aiming at the massive, heterogeneous, and professional coal mine geological results and maps, a pre-trained large-scale language model is used as the underlying architecture. According to different geological tasks, the output direction, depth, and accuracy of the model are finely controlled by adjusting the prompt words. The general capabilities of large-scale language models are not enough to directly solve professional problems in the field of geology. The introduction of a specialized annotated geological corpus and the introduction of transfer learning or fine-tuning enable the model to more accurately understand the terminology, formulas, and methodologies in geology, thereby improving the accuracy and efficiency of answering geological questions. A custom interface is designed to allow users to interact with the model according to task requirements, input geological data, and receive analysis results generated by the model, realizing the automated extraction of geological element information based on prior geological knowledge.

[0011] S2. End-to-end adaptive generation of structured information: Based on step S1, adaptive matching of the parametric modeling input is performed, including the JSON format output of geometric elements such as boreholes, faults, collapse columns, and anticlines. This completes the end-to-end production process from geological results and maps to geological information that can be used for digital modeling. While supporting manual input and correction of errors, an interactive interface + generative artificial intelligence is used to assist in the generation of structured data for geological modeling, achieving an information-to-model process experience without the need to access non-end links of information.

[0012] S3. Constructing a primary geological model using the generalized triangular prism algorithm: Conducting secondary development of the Unreal Engine, dividing the structured data obtained in step S2 into stratigraphic matrix data and abnormal encrypted data. The abnormal encrypted data includes faults, collapse columns, and anticlines. The stratigraphic matrix data includes point, line, and surface data of stratigraphic distribution obtained through drilling, trenching, pitting, and geophysical exploration. The stratigraphic matrix data is converted into discrete distribution data of the upper and lower surfaces of each stratigraphic layer as observation data. The Bowyer-Woston algorithm is used to construct an irregular triangulated network according to the Delaunay triangulation principle and generate ground surface data. Based on the ground surface irregular triangulated network, the ground surface data is generated. , the triangles are expanded downward along the drilling direction one by one to generate generalized triangular prism elements, and the overall description model of the stratum is constructed; the specific expansion rule is: if the three vertices of the current triangle have the same stratum number, the vertex of the new triangle expanded downward is the next point on the corresponding drill hole; if the stratum numbers are different, the vertex with the smaller number is expanded to the next point along the corresponding drill hole, and the vertex with the larger number remains unchanged; the modeling process includes extracting the initial triangle from the irregular triangulated network of the ground surface, and expanding the new triangle according to its stratum number, recording and constructing the generalized triangular prism, and gradually updating and completing the model until all triangles are processed, thus forming a primary geological body model;

[0013] S4. Geometric smoothing subdivision based on Kriging interpolation: To solve the problem of uneven changes in the stratum interface of the primary geological body model, an algorithm is established to evaluate the spatial smoothness of the generalized triangular prism model. According to the threshold control and the model's precision requirements, the area to be interpolated is obtained through iterative calculation. The Kriging interpolation method is used. Assuming that the interpolation point is z o Obeying the inherent stationary process, under the conditions of unbiased estimation and uniform distribution, calculate the optimal weight coefficient λ that minimizes the variance between the estimated value and the actual value i , calculate the stratigraphic distribution information of the interpolation point, and after multiple interpolations, make the transition between adjacent triangular prisms and boreholes smooth;

[0014] S5. Dynamic update of special geological elements: The structural data of the anomaly obtained in step S2 is converted into the constraint part of the generalized triangular prism model. Taking the fault as an example, the coal seam section is spatially fitted based on the structural data obtained in step S2. The change of the stratum height along the section or vertical section is characterized by an exponential decay function. The geological structure is displayed in an encrypted manner in the form of virtual drill holes to achieve the construction of a high-precision geological model.

[0015] S6. Generation of coal mine digital twin matrix: Utilize the large model with geological prior knowledge in step S1 to automatically extract and match appropriate lithologic descriptions from the geological description of the borehole, and use this description as an input parameter to call the interface of the drawing large model to generate a texture for rendering the coal mine geological digital twin matrix; the drawing large model generates high-quality textures that match the coal mine geological characteristics based on the lithologic descriptions, ensuring that the textures and the geological characteristics of the model accurately correspond; the generated textures and the three-dimensional geological model are submitted to the back-end system, which performs high-performance computing and data processing to ensure the accurate display of the digital twin matrix during the coal mine geological analysis, demonstration, and visualization process.

[0016] Furthermore, in step S2, end-to-end structured information is adaptively generated, and users work through a visual process interface of original input-AI-assisted extraction-manual correction-upload. Taking faults as an example, the fault information that can be extracted from geological results and maps includes strike, dip, inclination, extension length, fault distance, and fault-coal intersection. By calling the preset interface, the universal language model after fine-tuning and prompt word optimization can quickly focus on these key information and directly generate JSON format information that can be read by digital modeling software.

[0017] Furthermore, in step S4, the Kriging interpolation method is used to perform geometric smoothing for the uneven change of the stratum interface of the primary geological body model, assuming that z(x,y) is uniform in space and has the same variance σ 2 and expectation c, under the conditions of unbiased estimation Under the conditions, find the optimal solution system First, calculate the pairwise spatial correlation D of the observation data. ij and semivariance r ij =σ 2 -C ij , use Gaussian model to fit the function Fitting, calculating the spatial correlation between unknown points and known points, calculating the semivariance r between unknown points and all known points i0 , substitute into the linear equations to calculate the optimal weight coefficient λ i , by Calculate the stratigraphic distribution information of the interpolation point, evaluate the spatial smoothness of the generalized triangular prism model, and iteratively calculate the interpolation area according to the accuracy requirements;

[0018] Among them, z is the height value of the stratum distribution, (x, y) is the plane coordinate, Var(·) is the function for calculating the variance, E(·) is the function for calculating the expectation, C ij is the covariance of the i-th observation data and the j-th observation data, b0 is the nugget value, b is the arch height, f is the range, r i0 is the semivariance between the i-th observation data and the interpolation point, z i is the height value of the i-th observation data.

[0019] Furthermore, in step S5, the cross-section influence range parameter a is introduced and an exponential decay function Δh=h0e is used. -aΔL Characterizes the change of stratum height along the section or vertical section, h0 is the section distance, ΔL is the straight-line distance along the section or section dip.

[0020] Compared with the existing technology, the method of dynamically constructing coal mine geological digital twins based on multi-source heterogeneous information provided by the present invention innovatively introduces AIGC technology to assist in reading geological results. Through the big data assimilation of multi-source heterogeneous data, it realizes the intelligent generation of structured information that matches the automated modeling update algorithm. This process forms a rapid response process of "information processing-primary geological body generation-geometric smoothing-abnormal geological body encryption-global geological body construction". It can dynamically construct high-precision coal mine geological digital models based on geological data of different geological survey stages and different precisions, and make full use of geological information of all time periods. This method effectively solves the problems of high data preprocessing requirements and low information utilization in traditional geological body construction, meets the needs of coal mine digital twin model construction in terms of high precision, updatability and automation, and can be used as the core technology for coal mine digital twin matrix construction. Therefore, compared with the existing technology, it has the following advantages:

[0021] 1. Automated and efficient data processing capabilities: This method innovatively introduces AIGC technology and relies on a large-scale language model to achieve automated processing of complex, multi-source and heterogeneous coal mine geological data. This mechanism optimizes model output through prompt engineering and combines it with professional corpora in the field of geology to enhance the model's understanding of geological terminology and methods through transfer learning or fine-tuning, enabling it to accurately and quickly extract and parse massive amounts of geological reports and maps. This automated data processing method not only significantly reduces manual intervention but also significantly improves the efficiency of geological information acquisition and processing.

[0022] 2. End-to-end structured information generation process: This method achieves end-to-end automation from geological data extraction to model construction, forming a highly integrated structured information generation system. By designing a large model interface and combining it with a visual interface, users can directly extract and convert geological data such as drill holes and faults into structured formats such as JSON required for geological modeling, reducing complex intermediate processing steps.

[0023] 3. High-precision 3D modeling and spatial smoothing optimization: In terms of geological modeling, this method constructs a primary geological model based on the generalized triangular prism algorithm, capable of handling the complex structures found in coal mine geology and generating detailed geological descriptions. Simultaneously, the geological model is geometrically smoothed using the Kriging interpolation method, resulting in a high degree of continuity and smoothness across the model's stratigraphic interfaces. This two-layer modeling and optimization mechanism ensures the accuracy of the model structure, particularly excelling in handling complex stratigraphic layers and multi-layered spatial relationships, effectively enhancing the realism and practicality of the geological model.

[0024] 4. Dynamic update capability of special geological elements: In response to the variability of complex geological elements such as faults, collapse columns, and anticlines in coal mines, this method has a dynamic update mechanism. Through real-time detection of geological features and generation of structured data, combined with mathematical models such as exponential decay functions to accurately fit the changing trends of geological structures, the geological model can continuously reflect the latest geological knowledge. This dynamic update mechanism enables the coal mine digital twin to be continuously optimized and updated during the operation process.

[0025] 5. High-quality digital twin visualization: In terms of visualization, high-quality maps generated by AIGC technology are combined with the 3D geological model to ensure that the digital twin has highly consistent geological characteristics and visualization effects. This model not only provides intuitive geological data presentation but also enhances information transparency in the decision-making process. It provides strong support for coal mine geological analysis, operational demonstration, and intelligent management, significantly improving the practicality of the digital twin. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the method provided by the present invention for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information. DETAILED DESCRIPTION

[0027] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below with reference to specific illustrations.

[0028] Please refer to Figure 1 As shown, the present invention provides a method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information, comprising the following steps:

[0029] S1. AIGC assists in reading geological results: For massive, heterogeneous and professional coal mine geological results and maps, a pre-trained large language model (Large Language Models (LLMs) such as the GPT series are used as the underlying architecture. Prompt engineering is used to optimize model output. According to different geological tasks such as profile analysis, borehole extraction, and seismic data processing, the output direction, depth, and accuracy of the model can be finely controlled by adjusting prompts. The general capabilities of large-scale language models are not enough to directly solve professional problems in the field of geology. The introduction of specialized geological corpora such as annotated academic papers, technical reports, and industry standards enables the model to more accurately understand the terminology, formulas, and methodologies in geology through transfer learning or fine-tuning, thereby improving the accuracy and efficiency of answering geological problems. Custom interfaces are designed to allow users to interact with the model according to task requirements, input geological data such as exploration data, stratigraphic information, seismic data, drilling maps, and geophysical reports, obtain element information and descriptions of stratigraphic distribution and geological anomalies from boreholes, trenching, pit exploration, and geophysical exploration, and receive analysis results generated by the model to achieve automated extraction of geological element information based on geological prior knowledge.

[0030] S2. End-to-end adaptive generation of structured information: Based on step S1, adaptive matching of the parametric modeling input is performed, including the output of geometric elements such as boreholes, faults, collapse columns, and anticlines in JSON (JavaScript Object Notation) format. This enables the end-to-end production of geological results and maps into geological information that can be used for digital modeling. While supporting manual input and correction of errors, an interactive interface + generative artificial intelligence (Generative Artificial Intelligence) assists in the generation of structured data for geological modeling, achieving an information-to-model process experience without the need for non-end links such as intermediate information processing.

[0031] S3. Constructing a primary geological model using the generalized triangular prism algorithm: Secondary development of the Unreal Engine is performed, and the structured data obtained in step S2 is divided into stratigraphic matrix data and abnormal encrypted data. The abnormal encrypted data includes faults, collapse columns, anticlines, etc. The stratigraphic matrix data includes point, line, and surface data of stratigraphic distribution obtained by drilling, trenching, pit exploration, geophysical exploration, etc., and the stratigraphic matrix data is converted into discrete distribution data of the upper and lower surfaces of each stratigraphic layer as observation data; the Bowyer-Woston algorithm is used to construct an irregular triangulated network (TIN) according to the Delaunay triangulation principle and generate surface data. Based on the surface irregular triangulated network, the triangles are converted one by one. The shape is expanded downward along the drilling direction to generate a generalized triangular prism element to construct an overall description model of the stratum; the specific expansion rule is: if the three vertices of the current triangle have the same stratum number, the vertex of the new triangle expanded downward is the next point on the corresponding drill hole; if the stratum numbers are different, the vertex with the smaller number is expanded along the corresponding drill hole to the next point, and the vertex with the larger number remains unchanged; the modeling process includes extracting the initial triangle from the irregular triangulated network of the ground surface, expanding the new triangle according to its stratum number, recording and constructing the generalized triangular prism, and gradually updating and completing the model until all triangles are processed, thus forming a primary geological body model;

[0032] S4. Geometric smoothing subdivision based on Kriging interpolation: Aiming at the problem of uneven changes in the stratigraphic interface of the primary geological body model, an algorithm is established to evaluate the spatial smoothness of the generalized triangular prism model. According to the threshold control and the model's precision requirements, the area to be interpolated is obtained through iterative calculation. The Kriging interpolation method is used. It is assumed that the interpolation point in the interpolation area is z o Obeying the inherently stationary process, under the conditions of unbiased estimation and uniform distribution, calculate the optimal weight coefficient λ that minimizes the variance between the estimated value and the actual value i , calculate the stratigraphic distribution information of the interpolation point, and after multiple interpolations, make the transition between adjacent triangular prisms and boreholes smooth;

[0033] S5. Dynamic update of special geological elements (including faults, collapse columns, and anticlines): The structural data of the anomaly obtained in step S2 is converted into the constraint part of the generalized triangular prism model. Taking the fault as an example, the strike, dip, dip angle, extension length, gap, and coal-break intersection point, etc., based on the structural data obtained in step S2, are used to fit the coal seam cross section in space. An exponential decay function is used to characterize the change in stratum height along the cross section or vertical cross section. The geological structure is displayed in an encrypted manner in the form of virtual drill holes to achieve the construction of a high-precision geological model.

[0034] S6. Generation of coal mine digital twin matrix: Utilize the large model with geological prior knowledge in step S1 to automatically extract and match appropriate lithologic descriptions from the geological description of the borehole, and use this description as an input parameter to call the interface of the drawing large model to generate a texture for rendering the coal mine geological digital twin matrix; the drawing large model generates high-quality textures that match the coal mine geological characteristics based on the lithologic descriptions, and combines the high-quality textures with the high-precision three-dimensional geological model to ensure that the geological characteristics of the texture and model accurately correspond; the generated texture and the three-dimensional geological model are submitted to the back-end system, which performs high-performance computing and data processing to ensure the accurate display of the digital twin matrix during the coal mine geological analysis, demonstration, and visualization process.

[0035] As a specific embodiment, end-to-end structured information is adaptively generated in step S2, and the user works through a visual process interface of original input-AI-assisted extraction-manual correction-upload. Taking faults as an example, the fault information that can be extracted from geological results and maps includes strike, dip, inclination, extension length, fault distance, fault-coal intersection, etc., and the preset interface is called. The general language model after fine-tuning and prompt word optimization can quickly focus on these key information and directly generate JSON format information that can be read by digital modeling software.

[0036] As a specific embodiment, in step S4, the Kriging interpolation method is used to perform geometric smoothing for the uneven change of the stratum interface of the primary geological body model, assuming that z(x, y) is uniform in space and has the same variance σ 2 and expectation c (which is the expectation of z), satisfying the unbiased estimation condition Under the conditions, find the optimal solution system First, calculate the pairwise spatial correlation (Euclidean distance) of the observation data D ij and semivariance r ij =σ 2 -C ij , using the Gaussian model to fit the function Fitting, calculating the spatial correlation (Euclidean distance) between unknown points and known points, calculating the semivariance r between unknown points and all known points i0 , substitute into the linear equations to calculate the optimal weight coefficient λ i , by Calculate the stratigraphic distribution information of the interpolation point, evaluate the spatial smoothness of the generalized triangular prism model, and iteratively calculate the interpolation area according to the accuracy requirements;

[0037] Among them, z is the height value of the stratum distribution, (x, y) is the plane coordinate, Var(·) is the function for calculating the variance, E(·) is the function for calculating the expectation, C ijis the covariance of the i-th observation data and the j-th observation data, b0 is the nugget value, b is the arch height, f is the range, r i0 is the semivariance between the i-th observation data and the interpolation point, z i is the height value of the i-th observation data.

[0038] As a specific embodiment, in step S5, special geological elements (such as faults, collapse columns, and anticlines) are dynamically updated, and the abnormal body structured data obtained in step S2 is converted into the constraint part of the generalized triangular prism model; taking the fault as an example, the coal seam section is spatially fitted by obtaining structural data such as strike, dip, inclination, extension length, fault distance, and coal-fault intersection, and by introducing the section influence range parameter a, an exponential decay function Δh=h0e is used. -aΔL Characterizes the change in stratum height along a section or vertical section; where h0 is the section distance and ΔL is the straight-line distance along the section or section dip.

[0039] In order to better demonstrate the technical effect of the present invention, a field test will be conducted on the Mengcun Coal Mine in Shaanxi Province. The specific steps are as follows:

[0040] S1. Collect and organize historical geological survey data, including multi-level geological drill hole histograms with different spacing, three-dimensional geological exploration reports and maps, transient resistivity exploration results, channel wave exploration maps, etc., and upload them to the large model with the help of a predetermined visualization interface to extract geological element information; divide the stratum into 11 levels, and obtain the following effective geological elements: 43 first-level drill holes (spacing 750m), 23 second-level drill holes (spacing 100m), 698 third-level drill holes (spacing 10m), and 15 reliable faults.

[0041] S2. Based on the input requirements of the Unreal Engine modeling platform, the geological feature information was exported in JSON format. A primary model was constructed using the drill hole data and the generalized triangular prism voxel algorithm. The spatial smoothness of the model was calculated. The interpolation area was calculated based on the threshold control and accuracy requirements. Geometric smoothing was performed using the Kriging interpolation method. Two to three iterations were performed to achieve a smooth transition between adjacent triangular prisms and drill holes.

[0042] S3. Based on the geometric element data of the fault anomaly, fit the coal seam section in space, set the section influence range parameter a, and use the exponential decay function Δh=h0e -aΔL Describe the changes in stratum height along the cross section or vertical section, inversely calculate the stratum distribution on both sides of the fault, and display the geological structure in an in-depth manner through virtual drilling, ultimately achieving the construction of a high-precision geological model.

[0043] S4. Lithologic information is automatically extracted from the borehole geological description and used as input to generate a base map for the coal mine geological digital twin. This map is matched to the coal mine geological characteristics and combined with the 3D geological model to ensure consistency. Finally, the map and 3D model are submitted to the backend system, providing the foundation for the subsequent construction of the digital twin system.

[0044] Compared with the existing technology, the method of dynamically constructing coal mine geological digital twins based on multi-source heterogeneous information provided by the present invention innovatively introduces AIGC technology to assist in reading geological results. Through the big data assimilation of multi-source heterogeneous data, it realizes the intelligent generation of structured information that matches the automated modeling update algorithm. This process forms a rapid response process of "information processing-primary geological body generation-geometric smoothing-abnormal geological body encryption-global geological body construction". It can dynamically construct high-precision coal mine geological digital models based on geological data of different geological survey stages and different precisions, and make full use of geological information of all time periods. This method effectively solves the problems of high data preprocessing requirements and low information utilization in traditional geological body construction, meets the needs of coal mine digital twin model construction in terms of high precision, updatability and automation, and can be used as the core technology for coal mine digital twin matrix construction. Therefore, compared with the existing technology, it has the following advantages:

[0045] 1. Automated and efficient data processing capabilities: This method innovatively introduces AIGC technology and relies on a large-scale language model to achieve automated processing of complex, multi-source and heterogeneous coal mine geological data. This mechanism optimizes model output through prompt engineering and combines it with professional corpora in the field of geology to enhance the model's understanding of geological terminology and methods through transfer learning or fine-tuning, enabling it to accurately and quickly extract and parse massive amounts of geological reports and maps. This automated data processing method not only significantly reduces manual intervention but also significantly improves the efficiency of geological information acquisition and processing.

[0046] 2. End-to-end structured information generation process: This method achieves end-to-end automation from geological data extraction to model construction, forming a highly integrated structured information generation system. By designing a large model interface and combining it with a visual interface, users can directly extract and convert geological data such as drill holes and faults into structured formats such as JSON required for geological modeling, reducing complex intermediate processing steps.

[0047] 3. High-precision 3D modeling and spatial smoothing optimization: In terms of geological modeling, this method constructs a primary geological model based on the generalized triangular prism algorithm, capable of handling the complex structures found in coal mine geology and generating detailed geological descriptions. Simultaneously, the geological model is geometrically smoothed using the Kriging interpolation method, resulting in a high degree of continuity and smoothness across the model's stratigraphic interfaces. This two-layer modeling and optimization mechanism ensures the accuracy of the model structure, particularly excelling in handling complex stratigraphic layers and multi-layered spatial relationships, effectively enhancing the realism and practicality of the geological model.

[0048] 4. Dynamic update capability of special geological elements: In response to the variability of complex geological elements such as faults, collapse columns, and anticlines in coal mines, this method has a dynamic update mechanism. Through real-time detection of geological features and generation of structured data, combined with mathematical models such as exponential decay functions to accurately fit the changing trends of geological structures, the geological model can continuously reflect the latest geological knowledge. This dynamic update mechanism enables the coal mine digital twin to be continuously optimized and updated during the operation process.

[0049] 5. High-quality digital twin visualization: In terms of visualization, high-quality maps generated by AIGC technology are combined with the 3D geological model to ensure that the digital twin has highly consistent geological characteristics and visualization effects. This model not only provides intuitive geological data presentation but also enhances information transparency in the decision-making process. It provides strong support for coal mine geological analysis, operational demonstration, and intelligent management, significantly improving the practicality of the digital twin.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information, characterized in that: The following steps are involved: S1. AIGC assists in reading geological results: Aiming to address the massive, heterogeneous, and specialized coal mine geological results and maps, a pre-trained large-scale language model is used as the underlying architecture. The output direction, depth, and accuracy of the model are finely controlled by adjusting prompt words according to different geological tasks. The general capabilities of large-scale language models are insufficient to directly solve specialized problems in the field of geology. By introducing a specialized annotated geological corpus and, through transfer learning or fine-tuning, enabling the model to more accurately understand geological terminology, formulas, and methodologies, the accuracy and efficiency of answering geological questions are improved. Design a custom interface that allows users to interact with the model according to task requirements, input geological data, and receive analysis results generated by the model, thereby achieving automated extraction of geological element information based on prior geological knowledge; S2. End-to-end adaptive generation of structured information: Based on step S1, adaptive matching of the parametric modeling input is performed, including the JSON format output of geometric elements such as boreholes, faults, collapse columns, and anticlines. This completes the end-to-end production process from geological results and maps to geological information that can be used for digital modeling. While supporting manual input and correction of errors, an interactive interface + generative artificial intelligence is used to assist in the generation of structured data for geological modeling, achieving an information-to-model process experience without the need to access non-end links of information. S3. Constructing a primary geological model using the generalized triangular prism algorithm: Conducting secondary development of the Unreal Engine, dividing the structured data obtained in step S2 into stratigraphic matrix data and abnormal encrypted data. The abnormal encrypted data includes faults, collapse columns, and anticlines. The stratigraphic matrix data includes point, line, and surface data of stratigraphic distribution obtained through drilling, trenching, pitting, and geophysical exploration. The stratigraphic matrix data is converted into discrete distribution data of the upper and lower surfaces of each stratigraphic layer as observation data. The Bowyer-Woston algorithm is used to construct an irregular triangulated network according to the Delaunay triangulation principle and generate ground surface data. Based on the ground surface irregular triangulated network, the ground surface data is generated. , the triangles are expanded downward along the drilling direction one by one to generate generalized triangular prism elements, and the overall description model of the stratum is constructed; the specific expansion rule is: if the three vertices of the current triangle have the same stratum number, the vertex of the new triangle expanded downward is the next point on the corresponding drill hole; if the stratum numbers are different, the vertex with the smaller number is expanded to the next point along the corresponding drill hole, and the vertex with the larger number remains unchanged; the modeling process includes extracting the initial triangle from the irregular triangulated network of the ground surface, and expanding the new triangle according to its stratum number, recording and constructing the generalized triangular prism, and gradually updating and completing the model until all triangles are processed, thus forming a primary geological body model; S4. Geometric smoothing subdivision based on Kriging interpolation: To solve the problem of uneven changes in the stratum interface of the primary geological body model, an algorithm is established to evaluate the spatial smoothness of the generalized triangular prism model. According to the threshold control and the model's precision requirements, the area to be interpolated is obtained through iterative calculation. The Kriging interpolation method is used, assuming that the interpolation point is z o Obeying the inherent stationary process, under the conditions of unbiased estimation and uniform distribution, calculate the optimal weight coefficient λ that minimizes the variance between the estimated value and the actual value i , calculate the stratigraphic distribution information of the interpolation point, and after multiple interpolations, make the transition between adjacent triangular prisms and boreholes smooth; S5. Dynamic update of special geological elements: The structural data of the anomaly obtained in step S2 is converted into the constraint part of the generalized triangular prism model. Taking the fault as an example, the coal seam section is spatially fitted based on the structural data obtained in step S2. The change of the stratum height along the section or vertical section is characterized by an exponential decay function. The geological structure is displayed in an encrypted manner in the form of virtual drill holes to achieve the construction of a high-precision geological model. S6. Generation of coal mine digital twin matrix: Utilize the large model with geological prior knowledge in step S1 to automatically extract and match appropriate lithologic descriptions from the geological description of the borehole, and use this description as an input parameter to call the interface of the drawing large model to generate a texture for rendering the coal mine geological digital twin matrix; the drawing large model generates high-quality textures that match the coal mine geological characteristics based on the lithologic descriptions, ensuring that the textures and the geological characteristics of the model accurately correspond; the generated textures and the three-dimensional geological model are submitted to the back-end system, which performs high-performance computing and data processing to ensure the accurate display of the digital twin matrix during the coal mine geological analysis, demonstration, and visualization process.

2. The method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information according to claim 1, characterized in that: In step S2, end-to-end structured information is adaptively generated. Users work through a visual process interface of original input - AI-assisted extraction - manual correction - upload. Taking faults as an example, the fault information that can be extracted from geological results and maps includes strike, dip, inclination, extension length, fault distance, and fault-coal intersection. By calling the preset interface, the universal language model after fine-tuning and prompt word optimization can quickly focus on these key information and directly generate JSON format information that can be read by digital modeling software.

3. The method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information according to claim 1, characterized in that: In step S4, the Kriging interpolation method is used to perform geometric smoothing on the uneven interface of the primary geological model, assuming that z(x,y) is uniform in space and has the same variance σ 2 and expectation c, under the conditions of unbiased estimation Under the conditions, find the optimal solution system First, calculate the pairwise spatial correlation D of the observation data. ij and semivariance r ij =σ 2 -C ij , using the Gaussian model to fit the function Fitting, calculating the spatial correlation between unknown points and known points, calculating the semivariance r between unknown points and all known points i0 , substitute into the linear equations to calculate the optimal weight coefficient λ i , by Calculate the stratigraphic distribution information of the interpolation point, evaluate the spatial smoothness of the generalized triangular prism model, and iteratively calculate the interpolation area according to the accuracy requirements; Among them, z is the height value of the stratum distribution, (x, y) is the plane coordinate, Var(·) is the function for calculating the variance, E(·) is the function for calculating the expectation, C ij is the covariance of the i-th observation data and the j-th observation data, b0 is the nugget value, b is the arch height, f is the range, r i0 is the semivariance between the i-th observation data and the interpolation point, z i is the height value of the i-th observation data.

4. The method for dynamically constructing a coal mine geological digital twin based on multi-source heterogeneous information according to claim 1, characterized in that: In step S5, the cross-section influence range parameter a is introduced and an exponential decay function Δh=h0e is used. -aΔL Characterizes the change of stratum height along the section or vertical section, h0 is the section distance, ΔL is the straight-line distance along the section or section dip.

Images