Three-dimensional geological model auxiliary construction method and device based on intelligent agent
By constructing a three-dimensional geological model through intelligent agents, optimizing topological relationships using AI and graph neural networks, and combining the GeoFormer-PT model and interactive platform, the problems of low efficiency, poor accuracy and insufficient interaction in existing technologies are solved, and efficient and accurate three-dimensional geological model construction and dynamic visualization are achieved.
Patent Information
- Application Number
- CN202510890874.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-30
- Publication Date
- 2025-09-30
AI Technical Summary
The existing 3D geological model construction method relies on manual operation with low efficiency and poor accuracy, slow rendering response and insufficient interactive functions. It is impossible to interact in real time and the model verification relies on professional experience, making it difficult to ensure global consistency.
An agent-based 3D geological model construction method is adopted, the AI model is used to generate the initial model, the graph neural network is combined to verify the topological relationship, the parameter field is optimized through Fourier feature mapping and GeoFormer-PT model, and the 3D software interface layer is used for real-time rendering and an interactive operation platform is introduced to achieve dynamic visualization and risk analysis.
It improves the efficiency and accuracy of model construction, realizes the dynamic visualization and risk display of three-dimensional geological models, enhances human-computer interaction functions, supports working condition simulation and data collaboration, and simplifies the model correction process.
Smart Images

Figure CN120726252A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of geological engineering technology, and in particular to a method and device for auxiliary construction of a three-dimensional geological model based on an intelligent agent. Background Art
[0002] A 3D geological model is a digital three-dimensional spatial representation constructed using computer technology by integrating multiple sources of data, including geological, geophysical, and well logging data. It is used to quantitatively characterize the geometry, attribute distribution, and topological relationships of underground geological bodies. 3D geological models can intuitively reveal complex structures, assist in identifying hidden geological patterns, support the optimization of geological development plans, reduce exploration risks, and provide geological risk warnings for underground projects.
[0003] The main process of existing 3D geological model construction methods is as follows: first, geological engineers manually draw key elements such as stratigraphic interfaces and fault lines, and generate an initial geometric model through triangulation or implicit functions. Then, they manually check stratigraphic contact relationships, repeatedly adjust control points when topological conflicts are found, and independently generate attribute fields such as lithology and porosity based on statistical methods. Finally, the model is imported into professional software for offline rendering. This method has at least the following problems: First, it relies heavily on manual labor, requiring manual drawing and verification, which makes model construction inefficient. Verification also relies on the professional experience of geological engineers, resulting in poor topological reliability. Second, statistical methods ignore global geological laws and spatial correlations, and the attribute fields are disconnected from the geometric structure, resulting in poor model accuracy. Third, offline rendering is slow to respond, cannot interact in real time, and has limited interactive functions. When defects occur in the model, the parameters or geometric framework need to be modified from scratch, which is time-consuming and difficult to ensure global consistency. Summary of the Invention
[0004] The present invention aims to solve the problems of low efficiency and accuracy, slow rendering response and few interactive functions in existing three-dimensional geological model methods, and proposes an intelligent agent-based three-dimensional geological model auxiliary construction method and device.
[0005] The technical solution adopted by the present invention to solve the above technical problems is: In a first aspect, the present invention provides an agent-based assisted construction method for a three-dimensional geological model, the method comprising: Acquiring multi-source geological data, and generating a three-dimensional generalized model base based on the multi-source geological data and an AI model; Using a graph neural network to verify and correct the stratum contact relationship in the three-dimensional generalized model base, to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules; The 3D coordinates and geological attributes in the 3D compliance model base are encoded into a high-dimensional joint vector through Fourier eigenmapping, and the high-dimensional joint vector is input into the GeoFormer-PT model. The GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field; The optimized geological parameter field is transmitted to the three-dimensional software in real time through the three-dimensional software interface layer. The three-dimensional software uses an attribute mapping engine to map the corresponding geological parameters into visual vertex attributes. The three-dimensional model is rendered in real time based on the visual vertex attributes and using CUDA accelerated parallel computing to achieve dynamic visualization of the three-dimensional geological model. determining geological risks based on the optimized geological parameter field, mapping the geological risks to the surface of the three-dimensional geological model through color coding, generating a heat map, generating geological sections in any direction using a profile tool, performing a difference analysis on the geological sections, marking high-risk areas in the heat map based on the difference analysis results, obtaining click events of the three-dimensional geological model using an interactive tool, performing a confidence analysis based on attribute data of a location corresponding to the click event, and marking the confidence analysis results in the heat map; An interactive operation platform is introduced into the three-dimensional software, and the interactive operation platform is used for working condition simulation, data collaboration and manual intervention of the model.
[0006] Furthermore, generating a three-dimensional generalized model base according to the multi-source geological data and based on the AI model includes: Convert multi-source geological data into a unified coordinate system and convert discrete geological attributes into standardized numerical codes to obtain pre-processed multi-source geological data; Select an AI model as a generative model based on geological conditions, and use an existing standard geological model as a training sample to train the generative model, wherein the AI model is a generative adversarial network model or a diffusion model; The preprocessed multi-source geological data are input into the trained generative model in the form of 3D grids or point clouds to generate a 3D generalized model base.
[0007] Furthermore, the multi-source geological data includes drilling data, drone data, geophysical data and remote sensing image data, and the discrete geological attributes include lithology and weathering degree.
[0008] Furthermore, the stratum contact relationship in the 3D generalized model base is verified and corrected using a graph neural network, including: Extracting the topological relationships between stratigraphic units from the stratigraphic interface grid of the 3D generalized model base, constructing a graph based on the topological relationships, and inputting the constructed graph into a graph neural network; The graph neural network combines geological geometric topology rules to verify whether the contact relationship in the graph conforms to geological laws. If not, the contact relationship is modified or the graph structure is adjusted, and then verified again until the contact relationship in the graph conforms to geological laws.
[0009] Furthermore, a graph is constructed based on the topological relationship, including: Each stratigraphic unit is abstracted as a node, and the contact relationship between stratigraphic units is abstracted as an edge. A graph is constructed based on the nodes and edges, and the type of contact relationship is marked on the edges.
[0010] Furthermore, the multi-head attention mechanism includes a direction-aware head, a physical constraint head, and multi-scale fusion. The direction-aware head is used to learn the directional characteristics of geological structures, the physical constraint head is used to introduce geophysical rules as prior knowledge for attention calculation, and the multi-scale fusion is used to fuse the multi-scale features extracted by hierarchical coding and output an optimized geological parameter field.
[0011] Furthermore, the working condition simulation includes: performing stability simulation on the rock mass according to engineering parameters, and obtaining a risk warning report according to the simulation results; The data collaboration includes: real-time data synchronization and collaborative decision-making for multiple terminals; The manual intervention in the model includes: receiving manually input adjustment information, correcting the three-dimensional geological model according to the adjustment information, and feeding back to the AI model through active learning for iterative optimization.
[0012] In a second aspect, the present invention provides an agent-based 3D geological model assisted construction device for implementing the agent-based 3D geological model assisted construction method according to the first aspect, the device comprising: A generalized model base generation module is used to obtain multi-source geological data and generate a three-dimensional generalized model base based on the multi-source geological data and the AI model; A compliant model base generation module is used to verify and correct the stratum contact relationship in the three-dimensional generalized model base using a graph neural network to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules; A geological parameter field optimization module is used to encode the three-dimensional coordinates and geological attributes in the three-dimensional compliance model base into a high-dimensional joint vector through Fourier eigenmapping, and input the high-dimensional joint vector into the GeoFormer-PT model. The GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field; A visualization module is used to transmit the optimized geological parameter field to the three-dimensional software in real time through the three-dimensional software interface layer. The three-dimensional software uses an attribute mapping engine to map the corresponding geological parameters into visual vertex attributes. The three-dimensional model is rendered in real time based on the visual vertex attributes and using CUDA accelerated parallel computing to achieve dynamic visualization of the three-dimensional geological model. a visualization enhancement module for determining geological risks based on the optimized geological parameter field, mapping the geological risks to the surface of the three-dimensional geological model through color coding, generating a heat map, generating geological sections in any direction using a profile tool, performing a difference analysis on the geological sections, marking high-risk areas in the heat map based on the difference analysis results, obtaining click events of the three-dimensional geological model using an interactive tool, performing a confidence analysis based on the attribute data of the location corresponding to the click event, and marking the confidence analysis results in the heat map; An interactive operating platform introduced in 3D software is used for working condition simulation, data collaboration and manual model intervention.
[0013] The beneficial effects of the present invention are as follows: the intelligent agent-based three-dimensional geological model auxiliary construction method and device provided by the present invention use AI models to construct the initial model, thereby improving the efficiency of constructing the initial model from the original data, and thus improving the model construction efficiency; using graph neural networks to automatically perform topological verification, automatically identify and repair topological errors such as stratum penetration and order reversal, improve the verification efficiency and accuracy, and thus improve the model construction efficiency and accuracy; through Fourier coding + GeoFormer global optimization, high-frequency detail capture and joint optimization of geometry and attributes are achieved, further improving the accuracy of the model; through attribute mapping engine and CUDA accelerated parallel computing, the rendering speed is improved, and dynamic visualization of the three-dimensional geological model is achieved; constructing a heat map and marking the difference analysis results and confidence analysis results can intuitively display high-risk areas and data weaknesses, thereby allowing geological engineers to intuitively understand the uncertainty and potential risks of the model. At the same time, the introduction of an interactive operation platform increases the human-computer interaction function, which is more convenient for geological engineers to simulate working conditions, collaborate on multi-terminal data, and quickly modify model defects. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 A schematic diagram of a process flow of an agent-based assisted construction method for a three-dimensional geological model provided in an embodiment; Figure 2 A schematic diagram of the generation process of a three-dimensional generalized model base provided in an embodiment; Figure 3 A schematic diagram of the process of generating a three-dimensional compliance model base provided in an embodiment; Figure 4 A schematic diagram of a process for generating an optimized geological parameter field provided in an embodiment; Figure 5 A schematic diagram of a process for embedding a 3D geological model into 3D software provided in an embodiment; Figure 6 A schematic diagram of the implementation process of the three-dimensional geological model visualization enhancement provided in the embodiment; Figure 7 A schematic diagram of the implementation process of the interactive operation platform provided in the embodiment; Figure 8 A schematic structural diagram of an agent-based three-dimensional geological model auxiliary construction device provided in an embodiment. DETAILED DESCRIPTION
[0015] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution of this embodiment will be clearly and completely described below in conjunction with the drawings in this embodiment.
[0016] Some of the processes described in the specification of the present invention and the figures above include multiple operations that appear in a specific order. However, it should be understood that these operations may not be performed in the order in which they appear herein or may be performed in parallel. The sequence numbers of the operations are merely used to distinguish between different operations and do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.
[0017] Since the construction of three-dimensional geological models is usually manually dominated and adopts independent modeling and offline rendering, the inventors have found through research that this method has at least problems such as low efficiency and accuracy, slow rendering response, and few interactive functions.
[0018] Based on this, the technical solution of the present invention is proposed. In the present invention, multi-source geological data are obtained, and a three-dimensional generalized model base is generated according to the multi-source geological data and based on an AI model; a graph neural network is used to verify and correct the stratum contact relationship in the three-dimensional generalized model base to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules; the three-dimensional coordinates and geological attributes in the three-dimensional compliant model base are encoded into a high-dimensional joint vector through Fourier feature mapping, and the high-dimensional joint vector is input into the GeoFormer-PT model, and the GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field; the optimized geological parameter field is transmitted to the three-dimensional software in real time through the three-dimensional software interface layer, and the three-dimensional software uses attribute mapping to introduce The engine maps the corresponding geological parameters into visual vertex attributes, and uses CUDA accelerated parallel computing to render the three-dimensional model in real time according to the visual vertex attributes, thereby realizing dynamic visualization of the three-dimensional geological model; determines the geological risk according to the optimized geological parameter field, maps the geological risk to the surface of the three-dimensional geological model through color coding, generates a thermal map, uses the profile tool to generate geological profiles in any direction, performs difference analysis on the geological profiles, marks high-risk areas in the thermal map according to the difference analysis results, uses the interactive tool to obtain click events of the three-dimensional geological model, performs confidence analysis according to the attribute data of the position corresponding to the click event, and marks the confidence analysis results in the thermal map; introduces an interactive operation platform in the three-dimensional software, and the interactive operation platform is used for working condition simulation, data collaboration and manual intervention of the model.
[0019] Specifically, the present invention first uses an AI model to construct an initial, rough three-dimensional generalized model base based on multi-source geological data. The three-dimensional generalized model base contains the basic geological unit geometry and initial attribute information; then, a graph neural network is used to detect the contact relationships that violate geological rules in the three-dimensional generalized model base, and a correction plan is predicted to solve the topological errors that may exist in the initial model generated by AI, thereby ensuring the geological rationality of the model; then, based on the corrected three-dimensional compliant model base, the Fourier mapping is used to enhance the input representation capability, and then the global attention mechanism of GeoFormer-PT is used to perform fine parameter optimization to obtain an optimized geological parameter field containing precise position and attribute information, thereby solving the problem that local optimization methods are difficult to handle global geological continuity and correlation; then, the three-dimensional soft The software interface layer acts as a bridge to transmit the parameter data output by GeoFormer-PT to professional 3D visualization software in real time. Through efficient data transmission and GPU-accelerated rendering, the abstract numerical data is converted into intuitive graphics, realizing dynamic visualization of the 3D geological model. Then, risk calculation and confidence assessment are carried out based on the optimized geological parameter field, and an enhanced 3D visualization model with risk heat map, high-risk area markers, cross-section views and local confidence prompts is obtained, allowing geological engineers to intuitively understand the uncertainty and potential risks of the model. Finally, an interactive operation platform is introduced in the 3D software to provide the ultimate operating environment for human-computer collaboration, realizing human-computer collaboration and model iterative optimization. The working condition simulation function applies the final optimized model to actual engineering decision-making. Data collaboration ensures the efficiency of team work.
[0020] The technical solution of this embodiment will be clearly and completely described below in conjunction with the drawings in this embodiment. Obviously, the described embodiment is only a part of the embodiments of the present invention, rather than all the embodiments.
[0021] Figure 1 A flow chart of a method for assisting the construction of a 3D geological model based on an intelligent agent is shown in FIG. Figure 1 The method includes the following steps 1 to 5.
[0022] Step 1: Acquire multi-source geological data, and generate a three-dimensional generalized model base based on the multi-source geological data and the AI model.
[0023] In this embodiment, the multi-source geological data includes drilling data, drone data, geophysical data and remote sensing image data, and the discrete geological attributes include lithology and weathering degree.
[0024] Among them, drilling data is used to provide information such as underground rock stratification, depth, and properties, and is the key to vertical control; drone data is used to provide high-precision surface topography data; geophysical data (such as seismic, gravity, and magnetic methods) is used to provide physical field information of underground structures and is used to infer geological structures; remote sensing images are used to provide rock type, structure, and other information on a large range of the surface.
[0025] See also Figure 2 In step 1, a three-dimensional generalized model base is generated according to the multi-source geological data and based on the AI model, including steps 11 to 13.
[0026] Step 11: Convert the multi-source geological data into a unified coordinate system and convert the discrete geological attributes into standardized numerical codes to obtain the preprocessed multi-source geological data.
[0027] Because data sources must maintain consistent spatial references, all multi-source geological data is converted to a unified coordinate system, such as CGCS2000. This serves as the foundation for multi-source data fusion. Discrete geological attributes are converted to numerical codes, such as converting the lithology name "basalt" to "102," to facilitate AI model processing. Similar coding is also applied to other discrete attributes, such as weathering degree (strong, moderate, and slightly weathered).
[0028] Through the above steps, multi-source, multi-format, and multi-resolution data are integrated into a unified spatial framework and numerical system, providing consistent data input for subsequent AI modeling. For example, borehole data provides precise vertical stratification but sparse horizontal information, while drone and remote sensing data provide high-resolution horizontal information, and geophysical data provides indirect information on subsurface structure, complementing each other.
[0029] Step 12: Select the AI model as the generative model based on the geological conditions, and use the existing standard geological model data as training samples to train the generative model.
[0030] The AI model is a generative adversarial network model or a diffusion model. When selecting a generative model, a generative adversarial network model (GAN) may be selected for areas with more abrupt boundaries, such as faults and intrusion boundaries; a diffusion model may be selected for gradually changing strata, such as sedimentary strata. After selecting the AI model, existing standard geological model data is used for training.
[0031] Step 13: Input the preprocessed multi-source geological data into the trained generative model in the form of 3D grids or point clouds to generate a 3D generalized model base.
[0032] After the preprocessed multi-source geological data is input into the generative model, the borehole data is used as a control point to constrain the vertical distribution of the strata, the drone data constrains the surface morphology to ensure that the generated stratum interface is consistent with the surface topography, the geophysical data provides underground structural information to help the model generate reasonable faults and folds, and the remote sensing image provides surface lithology and structural information to assist the model in horizontal layout, thereby generating an initial, rough three-dimensional generalized model base.
[0033] Step 2: Use a graph neural network to verify and correct the stratum contact relationship in the three-dimensional generalized model base to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules.
[0034] See also Figure 3 In this embodiment, step 2 includes step 21 and step 22.
[0035] Step 21: Extract the topological relationship between stratigraphic units from the stratigraphic interface grid of the 3D generalized model base, construct a graph based on the topological relationship, and input the constructed graph into the graph neural network.
[0036] Specifically, each stratigraphic unit is abstracted as a node, and the contact relationship between stratigraphic units is abstracted as an edge. A graph is constructed based on the nodes and edges, and the type of contact relationship is marked on the edge. The constructed graph is then input into the graph neural network (GNN).
[0037] Step 22: The graph neural network combines geological geometric topology rules to verify whether the contact relationship in the graph conforms to geological laws. If not, the contact relationship is modified or the graph structure is adjusted, and then verified again until the contact relationship in the graph conforms to geological laws.
[0038] Specifically, the graph neural network contains a trained graph neural network model and a geological geometric topology rule library. After receiving the graph structure, the graph neural network checks whether each edge (contact relationship) complies with the predefined geological rules (such as conformity / unconformity / fault contact rules). During the verification process, if some contact relationships are found to violate the geological rules, they are marked as conflicts, and the system attempts to automatically correct these conflicts. The correction may include modifying the edge type (for example, changing the wrong contact relationship type to the correct one) or adjusting the graph structure (for example, adding or deleting edges). After correction, it is checked again until the graph structure fully complies with the rules. If all contact relationships comply with the rules, the three-dimensional compliant model base is output. The above steps use graph neural networks to automatically perform topology verification, automatically identify and repair topology errors such as stratum penetration and order reversal, improve verification efficiency and accuracy, and thus improve the efficiency and accuracy of model construction.
[0039] Step 3: Encode the 3D coordinates and geological attributes in the 3D compliance model base into a high-dimensional joint vector through Fourier eigenmapping, and input the high-dimensional joint vector into the GeoFormer-PT model. The GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field.
[0040] See also Figure 4 In this embodiment, the multi-head attention mechanism includes a direction perception head, a physical constraint head and multi-scale fusion. The direction perception head is used to learn the directional characteristics of geological structures, the physical constraint head is used to introduce geophysical rules as prior knowledge for attention calculation, and the multi-scale fusion is used to fuse the multi-scale features extracted by hierarchical coding and output an optimized geological parameter field.
[0041] In practical applications, random Fourier features (RFF) are used to encode the three-dimensional coordinates and geological attributes in the three-dimensional compliance model base into a high-dimensional joint vector. The encoded vector dimension is usually 80 dimensions. The high-dimensional joint vector contains both high-frequency encoding of spatial position and geological attribute information to better capture high-frequency changes and periodic patterns.
[0042] The GeoFormer-PT model is a pre-trained model that integrates a multi-head attention model with geological rules. It leverages the multi-head attention mechanism to capture long-range dependencies and non-local features in geological structures. The multi-head attention mechanism splits the input high-dimensional joint vector into multiple heads, learns different representation subspaces in parallel, and then merges the results. This allows the model to simultaneously focus on information from different subspaces at different locations.
[0043] This embodiment also features two specialized attention heads: a direction-aware head and a physical constraint head. The direction-aware head specifically learns directional characteristics of geological structures, such as stratigraphic strike and dip. It enhances feature extraction that is sensitive to direction by incorporating directional constraints into the calculation of attention weights. The physical constraint head incorporates geophysical rules, such as continuity constraints and stratigraphic occurrence rules, as prior knowledge for attention calculations. This prohibits connections that violate geological laws, ensuring that model outputs conform to geological principles.
[0044] The GeoFormer-PT model employs a hierarchical encoding structure to extract features layer by layer. At shallow layers, attention is calculated using a local window (e.g., an 8×8 neighborhood) to capture local features. At deeper layers, the window is gradually expanded to the global level to incorporate global contextual information. The model integrates feature maps from different layers (at different depths) through skip connections or a feature pyramid network (FPN). This allows the model to combine shallow detail with deep semantic information, improving its ability to represent multi-scale geological structures.
[0045] The above steps utilize Fourier eigenmaps to address the difficulty in capturing high-frequency variations when directly processing 3D coordinates. The multi-head attention mechanism (particularly the direction-aware and physical constraint heads) is tailored to the characteristics of geological data (directionality and physical laws). Hierarchical encoding and multi-scale fusion ensure the gradual extraction and integration of features from local to global, enabling the model to handle complex geological structures. The resulting optimized geological parameter field provides a more accurate data foundation for subsequent geological analysis and resource assessment.
[0046] Step 4: The optimized geological parameter field is transmitted to the 3D software in real time through the 3D software interface layer. The 3D software uses an attribute mapping engine to map the corresponding geological parameters into visual vertex attributes. The 3D model is rendered in real time based on the visual vertex attributes and CUDA accelerated parallel computing is used to achieve dynamic visualization of the 3D geological model.
[0047] See also Figure 5 In practical applications, the API provided by Blender or ParaView can be used to develop a 3D software interface layer to enable real-time transmission of the optimized address parameter field to the 3D software. The 3D software then maps the corresponding geological parameters (such as rock mass stability probability) to vertex attributes. For example, a stability probability of 0 to 1 is mapped from blue (stable) to red (unstable); displacement is mapped to vertex position offsets (geometric deformation). Simultaneously, the 3D software renders the model in real time based on vertex attributes. For example, color gradients can be used to set vertex colors based on stability probability, while geometric deformations can be used to shift vertex positions based on displacement to create a deformation effect.
[0048] In the process of optimizing the transmission and rendering of geological parameter fields, the powerful parallel computing capabilities of the GPU (through CUDA programming) are utilized to perform real-time rendering calculations on 3D geological models containing a large number of vertices (millions or even hundreds of millions), realizing dynamic visualization of the 3D geological model.
[0049] Step 5: Determine geological risks based on the optimized geological parameter field, map the geological risks to the surface of the three-dimensional geological model through color coding, generate a heat map, use the profile tool to generate geological profiles in any direction, perform difference analysis on the geological profiles, mark high-risk areas in the heat map based on the difference analysis results, use the interactive tool to obtain click events of the three-dimensional geological model, perform confidence analysis based on the attribute data of the position corresponding to the click event, and mark the confidence analysis results in the heat map.
[0050] See also Figure 6In practical applications, the risk value of each area can be calculated based on the attributes (such as the location of low-strength rock layers, high-porosity zones, fault fracture zones) or geometric features (such as steep slopes and small curvature radius folds) in the optimized geological parameter field, and the preset risk calculation model or rules can be used. The calculated risk value is then intuitively mapped to the surface of the three-dimensional geological model through color coding (such as red-yellow-green) to generate a thermal map.
[0051] In addition, the profile tool can be used to cut in any direction on the three-dimensional geological model to generate geological profiles. The system can compare profiles at different locations, or compare model profiles with profiles interpreted from the original data, perform difference analysis, identify inconsistent areas, and automatically mark high-risk areas on the three-dimensional geological model based on the results of thermal maps and difference analysis.
[0052] In addition, when the user clicks on a point on the 3D geological model, the system automatically obtains the attribute data of the corresponding position of the point and performs a confidence analysis. The confidence results (such as high, medium, and low) are also marked in the thermal map or near the point through color or other visual cues.
[0053] By constructing a heat map and marking the difference analysis results and confidence analysis results, the above steps can intuitively display high-risk areas and data weaknesses, realize the visualization enhancement of the 3D geological model, and enable geological engineers to intuitively understand the uncertainty and potential risks of the model.
[0054] Step 6: Introduce an interactive operation platform into the 3D software, which is used for working condition simulation, data collaboration, and manual model intervention.
[0055] See also Figure 7 In this embodiment, the working condition simulation includes: simulating the stability of the rock mass according to the engineering parameters, and obtaining a risk warning report based on the simulation results; the data collaboration includes: real-time data synchronization and collaborative decision-making for multiple terminals; the model manual intervention includes: receiving manually input adjustment information, correcting the three-dimensional geological model according to the adjustment information, and feeding back to the AI model through active learning for iterative optimization.
[0056] The interactive operation platform mainly provides three functions: Working condition simulation: allows engineers to set different engineering scenarios (such as excavation, loading, water injection, and earthquake), call built-in or integrated numerical simulation engines (such as FEM and CFD), predict engineering responses (deformation, stress, seepage, etc.) based on the current geological model, and overlay the simulation results on the 3D model.
[0057] Data collaboration: Supports online collaborative editing and annotation of models by users on multiple terminals (such as PCs and VR glasses), management of model versions, and import of new field survey data or interpretation results.
[0058] Manual intervention in the model: You can directly modify the model geometry (such as dragging control points, adjusting the interface) or attributes (reassigning values), and feed back to the AI model through active learning for iterative optimization.
[0059] By introducing an interactive operating platform into 3D software, an ultimate operating environment for human-machine collaboration is provided, enabling human-machine collaboration and model iterative optimization. The working condition simulation function applies the final optimized model to actual engineering decision-making, and data collaboration ensures the efficiency of team work.
[0060] In summary, the intelligent agent-based 3D geological model auxiliary construction method provided in this embodiment uses an AI model to construct an initial model, which improves the efficiency of constructing the initial model from the original data, thereby improving the model construction efficiency; uses a graph neural network to automatically perform topology verification, automatically identify and repair topology errors such as stratum penetration and order reversal, improves verification efficiency and accuracy, and thereby improves the model construction efficiency and accuracy; through Fourier coding + GeoFormer global optimization, high-frequency detail capture and joint optimization of geometry and attributes are achieved, further improving the accuracy of the model; through the attribute mapping engine and CUDA accelerated parallel computing, the rendering speed is improved, and dynamic visualization of the 3D geological model is achieved; constructing a heat map and marking the difference analysis results and confidence analysis results can intuitively display high-risk areas and data weaknesses, thereby allowing geological engineers to intuitively understand the uncertainty and potential risks of the model. At the same time, the introduction of an interactive operation platform increases human-computer interaction functions, making it more convenient for geological engineers to simulate working conditions, collaborate on multi-terminal data, and quickly modify model defects.
[0061] Based on the above technical solution, this embodiment also proposes an agent-based 3D geological model auxiliary construction device for implementing the agent-based 3D geological model auxiliary construction method as described in the embodiment. Figure 8 , the device comprises: A generalized model base generation module is used to obtain multi-source geological data and generate a three-dimensional generalized model base based on the multi-source geological data and the AI model; A compliant model base generation module is used to verify and correct the stratum contact relationship in the three-dimensional generalized model base using a graph neural network to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules; A geological parameter field optimization module is used to encode the three-dimensional coordinates and geological attributes in the three-dimensional compliance model base into a high-dimensional joint vector through Fourier eigenmapping, and input the high-dimensional joint vector into the GeoFormer-PT model. The GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field; A visualization module is used to transmit the optimized geological parameter field to the three-dimensional software in real time through the three-dimensional software interface layer. The three-dimensional software uses an attribute mapping engine to map the corresponding geological parameters into visual vertex attributes. The three-dimensional model is rendered in real time based on the visual vertex attributes and using CUDA accelerated parallel computing to achieve dynamic visualization of the three-dimensional geological model. a visualization enhancement module for determining geological risks based on the optimized geological parameter field, mapping the geological risks to the surface of the three-dimensional geological model through color coding, generating a heat map, generating geological sections in any direction using a profile tool, performing a difference analysis on the geological sections, marking high-risk areas in the heat map based on the difference analysis results, obtaining click events of the three-dimensional geological model using an interactive tool, performing a confidence analysis based on the attribute data of the location corresponding to the click event, and marking the confidence analysis results in the heat map; An interactive operating platform introduced in 3D software is used for working condition simulation, data collaboration and manual model intervention.
[0062] It can be understood that since the intelligent agent-based three-dimensional geological model auxiliary construction device described in this embodiment is a device for implementing the intelligent agent-based three-dimensional geological model auxiliary construction method described in the embodiment, for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant points, please refer to the partial description of the method, which will not be repeated here.
Claims
1. An agent-based assisted construction method for a three-dimensional geological model, characterized in that: The method comprises: Acquiring multi-source geological data, and generating a three-dimensional generalized model base based on the multi-source geological data and an AI model; Using a graph neural network to verify and correct the stratum contact relationship in the three-dimensional generalized model base, to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules; The 3D coordinates and geological attributes in the 3D compliance model base are encoded into a high-dimensional joint vector through Fourier eigenmapping, and the high-dimensional joint vector is input into the GeoFormer-PT model. The GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field; The optimized geological parameter field is transmitted to the three-dimensional software in real time through the three-dimensional software interface layer. The three-dimensional software uses an attribute mapping engine to map the corresponding geological parameters into visual vertex attributes. The three-dimensional model is rendered in real time based on the visual vertex attributes and using CUDA accelerated parallel computing to achieve dynamic visualization of the three-dimensional geological model. determining geological risks based on the optimized geological parameter field, mapping the geological risks to the surface of the three-dimensional geological model through color coding, generating a heat map, generating geological sections in any direction using a profile tool, performing a difference analysis on the geological sections, marking high-risk areas in the heat map based on the difference analysis results, obtaining click events of the three-dimensional geological model using an interactive tool, performing a confidence analysis based on attribute data of a location corresponding to the click event, and marking the confidence analysis results in the heat map; An interactive operation platform is introduced into the three-dimensional software, and the interactive operation platform is used for working condition simulation, data collaboration and manual intervention of the model.
2. The agent-based assisted construction method of a 3D geological model according to claim 1, wherein: Generating a three-dimensional generalized model base according to the multi-source geological data and based on the AI model includes: Convert multi-source geological data into a unified coordinate system and convert discrete geological attributes into standardized numerical codes to obtain pre-processed multi-source geological data; Select an AI model as a generative model based on geological conditions, and use an existing standard geological model as a training sample to train the generative model, wherein the AI model is a generative adversarial network model or a diffusion model; The preprocessed multi-source geological data are input into the trained generative model in the form of 3D grids or point clouds to generate a 3D generalized model base.
3. The agent-based assisted construction method of a 3D geological model according to claim 2, wherein: The multi-source geological data includes drilling data, drone data, geophysical data and remote sensing image data, and the discrete geological attributes include lithology and weathering degree.
4. The agent-based assisted construction method of a 3D geological model according to claim 1, wherein: Verifying and correcting the stratum contact relationship in the 3D generalized model base using a graph neural network includes: Extracting the topological relationships between stratigraphic units from the stratigraphic interface grid of the 3D generalized model base, constructing a graph based on the topological relationships, and inputting the constructed graph into a graph neural network; The graph neural network combines geological geometric topology rules to verify whether the contact relationship in the graph conforms to geological laws. If not, the contact relationship is modified or the graph structure is adjusted, and then verified again until the contact relationship in the graph conforms to geological laws.
5. The agent-based assisted construction method of a 3D geological model according to claim 4, characterized in that: Constructing a graph according to the topological relationship includes: Each stratigraphic unit is abstracted as a node, and the contact relationship between stratigraphic units is abstracted as an edge. A graph is constructed based on the nodes and edges, and the type of contact relationship is marked on the edges.
6. The agent-based assisted construction method of a 3D geological model according to claim 1, wherein: The multi-head attention mechanism includes a direction-sensing head, a physical constraint head, and multi-scale fusion. The direction-sensing head is used to learn the directional characteristics of geological structures, the physical constraint head is used to introduce geophysical rules as prior knowledge for attention calculation, and the multi-scale fusion is used to fuse the multi-scale features extracted by hierarchical coding and output an optimized geological parameter field.
7. The agent-based assisted construction method of a 3D geological model according to claim 1, wherein: The working condition simulation includes: simulating the stability of the rock mass according to the engineering parameters, and obtaining a risk warning report based on the simulation results; The data collaboration includes: real-time data synchronization and collaborative decision-making for multiple terminals; The manual intervention in the model includes: receiving manually input adjustment information, correcting the three-dimensional geological model according to the adjustment information, and feeding back to the AI model through active learning for iterative optimization.
8. An agent-based 3D geological model auxiliary construction device, characterized in that: The device is used to implement the agent-based assisted construction method of a three-dimensional geological model according to any one of claims 1 to 7, comprising: A generalized model base generation module is used to obtain multi-source geological data and generate a three-dimensional generalized model base based on the multi-source geological data and the AI model; A compliant model base generation module is used to verify and correct the stratum contact relationship in the three-dimensional generalized model base using a graph neural network to obtain a three-dimensional compliant model base that conforms to geological geometric topology rules; A geological parameter field optimization module is used to encode the three-dimensional coordinates and geological attributes in the three-dimensional compliance model base into a high-dimensional joint vector through Fourier eigenmapping, and input the high-dimensional joint vector into the GeoFormer-PT model. The GeoFormer-PT model processes the high-dimensional joint vector through a multi-head attention mechanism to obtain an optimized geological parameter field; A visualization module is used to transmit the optimized geological parameter field to the three-dimensional software in real time through the three-dimensional software interface layer. The three-dimensional software uses an attribute mapping engine to map the corresponding geological parameters into visual vertex attributes. The three-dimensional model is rendered in real time based on the visual vertex attributes and using CUDA accelerated parallel computing to achieve dynamic visualization of the three-dimensional geological model. a visualization enhancement module for determining geological risks based on the optimized geological parameter field, mapping the geological risks to the surface of the three-dimensional geological model through color coding, generating a heat map, generating geological sections in any direction using a profile tool, performing a difference analysis on the geological sections, marking high-risk areas in the heat map based on the difference analysis results, obtaining click events of the three-dimensional geological model using an interactive tool, performing a confidence analysis based on the attribute data of the location corresponding to the click event, and marking the confidence analysis results in the heat map; An interactive operating platform introduced in 3D software is used for working condition simulation, data collaboration and manual model intervention.
Citation Information
Cited By
Geophysical three-dimensional visualization and geologic model construction system
CN121527337A
Oil and gas geological data association display method
CN121883750A
A method for correlating and displaying oil and gas geological data
CN121883750B
