A 3D reconstruction and analysis system for geological structures based on multimodal data fusion

Abstract

This invention discloses a three-dimensional reconstruction and analysis system for geological structures based on multimodal data fusion, belonging to the field of geological exploration information technology. The system acquires and normalizes multi-source heterogeneous geological observation data through a preprocessing module, generating basic data and a data coverage density map. An adaptive fusion module analyzes and generates a dynamic adaptive weight field, weighting and fusing the data to obtain a primary three-dimensional geological data volume. A construction and optimization module constructs a vector-form three-dimensional spatial residual field based on the differences between the fused volume and the original data, and deduces the geometric morphology of geological interfaces. A reconstruction module ensures inter-layer consistency of the model through hierarchical bidirectional confidence propagation. A quantization module calculates uncertainty indices. Finally, a visualization module generates a three-dimensional geological scene. This invention effectively solves the challenges of conflict resolution and reliability assessment in multi-source geological data fusion, improving the accuracy of three-dimensional geological models and their engineering decision support capabilities.

Description

Technical Field

[0001] This invention relates to the field of geological exploration information technology, specifically to a three-dimensional reconstruction and analysis system for geological structures based on multimodal data fusion. Background Technology

[0002] Three-dimensional visualization of geological structures is a core technology in resource exploration and engineering geology. It aims to construct accurate three-dimensional models reflecting underground structures by integrating multi-source data such as seismic, borehole, and remote sensing data, providing crucial decision-making support for oil and gas exploration, mineral assessment, and engineering construction. In recent years, with advancements in data acquisition technology and improved computer processing capabilities, multimodal data fusion methods have gradually become an important means of enhancing model accuracy.

[0003] Existing technologies mostly employ data fusion strategies based on fixed weights or single confidence levels, integrating seismic data with sparse borehole data, or introducing data errors as a basis for weight adjustment during the fusion process. These methods can effectively construct 3D models in areas with high data quality and minimal conflicts, and possess a certain degree of spatial interpolation rationality. However, when geological structures are complex and multi-source data exhibit significant spatial conflicts or inconsistencies—for example, depth discrepancies between seismically interpreted stratigraphic levels and borehole-revealed lithological interfaces, or mismatches between the concealed structures revealed by gravity anomalies and the direction of remote sensing linear features—the fixed fusion strategies of existing methods struggle to dynamically and precisely reconcile these spatially localized data contradictions. This leads to irrational phenomena such as geometrical distortion, discontinuous fault characterization, or stratigraphic intersections in the constructed 3D geological interfaces in areas of data conflict, limiting the reliability and practicality of the model under complex geological conditions.

[0004] Therefore, how to generate a geometrically reasonable and globally consistent three-dimensional structural model under the premise of localization and multi-source conflicts in geological data remains a core technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a three-dimensional reconstruction and analysis system for geological structures based on multimodal data fusion.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] This invention discloses a three-dimensional reconstruction and analysis system for geological structures based on multimodal data fusion, comprising:

[0008] The preprocessing module is used to acquire multi-source heterogeneous geological observation data and perform spatiotemporal normalization processing to generate basic data and data coverage density maps.

[0009] An adaptive fusion module is used to perform local spatiotemporal consistency analysis on the basic data, generate a dynamic adaptive weight field, and perform weighted fusion on the geological observation data to generate a primary three-dimensional geological data volume.

[0010] The construction and optimization module is used to construct a vector-form three-dimensional spatial residual field based on the difference between the primary three-dimensional geological data volume and the geological observation data, and to deduce the geometric morphology of the geological interface and generate an optimized three-dimensional geological interface model using the three-dimensional spatial residual field as the driving force.

[0011] The reconstruction module is used to perform depth-based layering of the three-dimensional geological interface model and execute inter-layer bidirectional confidence propagation to generate a final three-dimensional geological model with inter-layer consistency and corresponding confidence distribution data.

[0012] The quantification module is used to calculate the uncertainty quantification index of the final three-dimensional geological model by fusing the three-dimensional spatial residual field, the data coverage density map, and the confidence distribution data.

[0013] The visualization module is used to visualize the final three-dimensional geological model and the uncertainty quantification index, and generate and output a three-dimensional geological scene.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] 1. This invention can dynamically adjust the contribution weight of different data sources within each spatial unit according to the consistency of the data, so that detailed features are fully preserved in areas with high data quality and good consistency, while the impact of unreliable data is automatically reduced in areas with data conflicts or sparseness through a weighting mechanism, thereby improving the overall reliability and spatial coordination of the primary three-dimensional geological data volume.

[0016] 2. This invention can quantify the differences between multi-source data into a directional residual distribution, thereby driving the global optimization of geological interfaces. It can not only effectively identify areas with concentrated data conflicts, but also automatically find the interface morphology that best fits the overall multi-source observation data while maintaining the geometric smoothness and continuity of the geological interface. This overcomes the interface abrupt changes or spatial inconsistencies that are easily generated by traditional interpolation or single-source data source fitting methods in complex tectonic regions.

[0017] 3. The three-dimensional geological scene output by this invention, which incorporates uncertainty information, enables users to intuitively identify high-risk areas in the model, providing a more reliable basis for resource assessment and engineering decision-making. Attached Figure Description

[0018] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0019] Figure 1 This is a system module connection diagram of the present invention;

[0020] Figure 2 This is a flowchart of the system modules of the present invention;

[0021] Figure 3 This is a flowchart illustrating the steps involved in constructing a three-dimensional spatial residual field according to the present invention. Detailed Implementation

[0022] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0023] In existing technologies, 3D geological structure modeling often relies on a single data source or a fixed-weight fusion method, making it difficult to effectively reconcile local spatial conflicts between heterogeneous multi-source data. Traditional methods are prone to interface geometric distortion and stratigraphic inconsistencies in areas with uneven data quality and complex geological structures. Existing systems lack spatial quantification mechanisms for data conflicts and cannot dynamically adjust the contribution weights of different data sources. Especially in fault-prone areas or when shallow and deep data are inconsistent, fixed fusion models exhibit systematic biases, making it difficult to meet the requirements of high-precision geological modeling.

[0024] To address the aforementioned issues, this study discovered an intrinsic correlation between the spatial residual distribution among multi-source data and the geometric characteristics of geological interfaces. Geometric deduction was achieved by establishing an interface optimization model based on flow conservation constraints. The vector form of the residual field can characterize both the intensity of data conflicts and indicate the direction of interface correction, while hierarchical confidence propagation effectively maintains the longitudinal consistency of the model. Therefore, a technical approach combining dynamic weight fusion, vector residual field construction, and minimum flow optimization is proposed. Further engineering verification incorporates confidence distribution and uncertainty quantification into the visualization feedback loop, forming a closed-loop system for continuous optimization of modeling accuracy.

[0025] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Example:

[0027] like Figure 1 As shown, the geological structure 3D reconstruction and analysis system based on multimodal data fusion includes:

[0028] Geological observation data are weighted and fused to generate a primary three-dimensional geological data volume;

[0029] The construction and optimization module is used to construct a vector-based three-dimensional spatial residual field based on the difference between the primary three-dimensional geological data volume and geological observation data. Driven by the three-dimensional spatial residual field, the geometric morphology of the geological interface is deduced to generate an optimized three-dimensional geological interface model.

[0030] The reconstruction module is used to perform depth-based layering of the 3D geological interface model and execute bidirectional confidence propagation between layers to generate a final 3D geological model with interlayer consistency and corresponding confidence distribution data.

[0031] The quantification module is used to calculate the uncertainty quantification index of the final three-dimensional geological model by fusing data based on the three-dimensional spatial residual field, data coverage density map and confidence distribution data.

[0032] The visualization module is used to visualize the final 3D geological model and uncertainty quantification indicators, and generate and output a 3D geological scene.

[0033] like Figure 2 As shown, the working principle of this application is as follows: The system connects to the seismic data acquisition system, borehole data entry terminal, remote sensing satellite ground receiving station, and gravity measurement instrument network through a preprocessing module deployed in the data center. The preprocessing module performs coordinate system-to-coordinate transformation and timestamp synchronization processing on the input seismic body data, borehole lithology records, remote sensing images, and gravity logging data to form a basic data array with consistent spatiotemporal reference. Simultaneously, a two-dimensional raster map representing the data distribution density is generated through a spatial distribution statistical algorithm.

[0034] The adaptive fusion module analyzes the basic data by calling a spatiotemporal consistency analysis program. A dynamic adaptive weight field is established on the three-dimensional spatial grid, and fusion weights are assigned according to the consistency of different data sources at each location. Finally, a weighted synthesis algorithm is used to output a primary three-dimensional geological data volume that integrates information from multiple sources.

[0035] The construction and optimization module calculates the spatial deviation between the primary 3D geological data volume and each original observation data using a difference detection algorithm, constructing a 3D residual field data containing magnitude and orientation information. Based on this 3D residual field data, numerical optimization methods are used to extrapolate the geometric morphology of geological interfaces, outputting an optimized 3D geological interface mesh model.

[0036] The reconstruction module divides the 3D space into multiple geological layers according to depth coordinates and propagates confidence information between layers through a two-way data transfer mechanism. This mechanism considers both the constraint effect of shallow data on the deep model and the correction effect of reliable deep observations on the shallow model, ultimately outputting a geologically sound 3D geological model and corresponding confidence distribution data.

[0037] The quantization module simultaneously reads the 3D spatial residual field, data density distribution map, and confidence matrix, and calculates the uncertainty metric value for each 3D grid cell using a multi-factor fusion algorithm. The visualization module is implemented through a graphics workstation, whose rendering pipeline couples the geological structure model with the uncertainty data for rendering. It uses color coding and transparency mapping techniques to simultaneously display geological structure and reliability information in the 3D scene, outputting 3D visualization results that can be used for geological analysis.

[0038] Through the above technical solution, this application achieves fully automated processing of three-dimensional geological structural models, from data preprocessing to output. This system can effectively integrate multi-source heterogeneous geological data, generating geometrically reasonable three-dimensional structural models with reliability assessments in complex geological areas, providing crucial technical support for mineral resource exploration and engineering geological stability evaluation.

[0039] This application further proposes that the specific steps for performing local spatiotemporal consistency analysis on basic data include:

[0040] First, analysis units are defined for the unified gridded base data. A cube with a side length of 50-200 meters is used as the standard spatiotemporal neighborhood. This size range is determined based on the statistical characteristics of seismic profile line spacing and borehole control radius. Within each spatiotemporal neighborhood, the system performs parallel calculations of consistency metrics for multi-source data.

[0041] For spatial location matching calculation, the system matches the spatial distribution of seismic reflection surfaces and borehole lithological interfaces using an iterative nearest point algorithm, and uses Hausdorff distance to measure the degree of matching between the two. In terms of directional consistency calculation, the system first extracts linear structures in remote sensing images using the Canny operator and calculates their dominant directions using the Hough transform; at the same time, it uses the Sobel operator to calculate the gradient field direction of gravity anomaly data, and the directional consistency between the two is measured by cosine similarity.

[0042] The system combines the spatial location matching and directional consistency measurement results with the preset baseline confidence levels of each data source (earthquake data: 0.8, borehole data: 0.9, gravity data: 0.7, remote sensing data: 0.6; these weights are set based on the vertical resolution and detection depth of each data source) and calculates the spatiotemporal consistency score using a weighted geometric average. Finally, the Min-Max normalization method is used to map the score to the [0,1] interval, generating a dynamic adaptive weight field.

[0043] By quantifying multi-dimensional consistency metrics using the aforementioned technical solutions, this application establishes a direct correlation between data quality and fusion weights, effectively addressing the adaptability issues of traditional fixed-weight fusion methods in areas of data conflict. Experiments show that in complex geological structures such as fault-developed zones, it improves the accuracy and reliability of multi-source data fusion.

[0044] like Figure 3 As shown, this application further proposes that the specific steps for constructing a vector-form three-dimensional spatial residual field based on the difference between the primary three-dimensional geological data volume and geological observation data include:

[0045] Heterogeneous observation data, such as seismic reflector data, borehole lithological interface data, and gravity anomaly data, were uniformly resampled into a standard 50m×50m×10m three-dimensional grid space using a spatial interpolation algorithm. The grid size of this standard 3D grid space was determined based on a comprehensive analysis of the dominant frequency of the seismic data (30-60Hz) and the average spacing of the borehole data (100-200m).

[0046] At each grid cell, the system computes in parallel the difference characteristics between multi-source observation data and the primary 3D geological data volume. A relative error metric, scalar bias, is used. The calculation formula is as follows:

[0047] ;

[0048] in, Represents the observation value of the i-th data source. This represents the estimated value of a primary three-dimensional geological data volume. is the numerical stability constant.

[0049] By using the above method of calculating relative error, the problem of scale differences caused by different physical dimensions is effectively avoided.

[0050] In vector deviation In the calculation, the system first extracts the spatial variation characteristics of each data source within its local neighborhood (3×3×3 grid) through principal component analysis, and generates a direction vector representing the dip of the geological interface. Correspondingly, the direction vector at the same location is extracted from the primary 3D geological data volume. directional deviation between the two Measured by angle difference:

[0051] ;

[0052] When the magnitude of the direction vector is less than the threshold of 0.1, the system determines that the area lacks a clear geological structure guide and automatically deviates the direction of the point. Its weight is reduced to zero.

[0053] Based on the above calculations, the system constructs a three-dimensional spatial residual field with clear physical meaning. Among these, the residual intensity... Through scalar deviation The weighted fusion yields:

[0054] ;

[0055] Weighting coefficient The system inherits the dynamic weights output by the adaptive fusion module, ensuring that observations with high data quality dominate the residual calculation. The residual direction is determined by vector synthesis of the bias angles, specifically using spherical linear interpolation to perform direction fusion on a unit sphere.

[0056] By establishing a multi-dimensional, multi-scale residual calculation system, this application achieves accurate identification and quantitative characterization of data conflict areas. The physical meaning-based residual field construction method provides a reliable driving signal for subsequent geological interface optimization and improves the modeling accuracy of complex tectonic regions.

[0057] This application further proposes that the specific steps for deriving the geometric morphology of geological interfaces and generating optimized three-dimensional geological interface models include:

[0058] The geological interface optimization problem is modeled as a minimum flow problem in a three-dimensional grid space that satisfies flow conservation constraints. Specifically, the flow conservation constraint requires that in each grid cell, the total flow into that cell must be equal to the sum of the residual source intensity within that cell and the flow out of that cell. This constraint can be formally expressed as:

[0059] ;

[0060] in, This represents the flow rate from adjacent cell j into cell i. This represents the flow rate from cell i to cell j. The residual source intensity within element i is determined by the magnitude of the residual vector of the three-dimensional spatial residual field at that element. Represents the set of neighboring units of unit i.

[0061] The objective function is solved using an iterative optimization algorithm. The objective function includes a residual flux term based on the three-dimensional spatial residual field and a smoothing penalty term based on the interface curvature. The design of the objective function comprehensively considers both the residual flux term and the smoothing penalty term based on the interface curvature, and its expression is:

[0062] ;

[0063] in, Let F be the total energy function, representing the flow field configuration across the entire grid. The first term is the residual flux term, used to measure the degree to which the flow conservation constraints are satisfied. The second term is the smoothing penalty term, used to constrain the geometric smoothness of the geological interface, through the gradient operator. Penalize flow changes between adjacent units; and These are weighting coefficients, which control the contribution ratio of each item to the overall objective. They are typically set based on data quality and geological priors. Its value is derived from the cross-validation results of historical work area data.

[0064] In the model solution process, an iterative optimization algorithm based on gradient descent is adopted, with the initial flow field set to a uniform distribution and the learning rate set to [value missing]. The maximum number of iterations is set to The flow field configuration is updated in each iteration until the objective function value changes. Below the preset change threshold The system outputs the currently optimized geological interface as the optimized 3D geological interface model. This preset change threshold is determined based on a trade-off between model convergence and computational efficiency achieved through multiple experiments. During iteration, if a region of concentrated residuals is detected and the residual vector directions are consistent (e.g., the direction consistency coefficient is higher than 0.7), the system will automatically identify and generate fault candidate surfaces and embed them as hard constraints into the flow conservation equations to enhance the model's ability to identify complex structures. When the number of iterations exceeds the limit, an alternative interpolation algorithm is activated.

[0065] This scheme models the geological interface optimization problem as a constrained minimum flow problem, and combines residual driving and smoothness constraints to achieve high-precision reconstruction of geological interfaces under multi-source heterogeneous data conditions. The entire simulation process has stable numerical performance and reproducibility under preset convergence conditions, which improves the automation level and reliability of 3D geological modeling results.

[0066] This application further proposes that, after constructing a vector-form three-dimensional spatial residual field based on the difference between the primary three-dimensional geological data volume and the geological observation data, it also includes:

[0067] Based on a three-dimensional spatial residual field, the system identifies regions of concentrated residuals. First, it performs multi-scale analysis of the residual field and employs a peak detection algorithm based on wavelet transform to identify these regions. Specifically, this is achieved by calculating the local standard deviation of the residual intensity.

[0068] ;

[0069] in This represents the residual strength value within a 7×7×7 neighborhood centered on the current grid cell. The mean value is the neighborhood value, and N=343 is the total number of neighborhood grid cells. When the residual fluctuation threshold of 0.35 is exceeded (this residual fluctuation threshold is determined based on the statistical characteristics of a large number of known fault samples, and the range of the residual fluctuation threshold is 0.2~0.5), the system marks the region as a potential fault development zone.

[0070] Within the residual set region, the system evaluates the directional consistency of the residual vector field through principal direction analysis. A directional consistency index is defined. The formula is as follows:

[0071] ;

[0072] in, This is the normalized residual direction vector. When Exceeding the directional consistency threshold At that point, the region is considered to possess significant directional consistency characteristics. Directional consistency threshold. Based on statistical analysis of data from 200 historical work areas, a typical range of 0.6 to 0.9 was set. To balance sensitivity and specificity, a preferred threshold range is 0.7 to 0.8. In this embodiment, a directional consistency threshold is used. When the directional consistency coefficient Cd is higher than this value, the directional consistency is considered to be significant.

[0073] The system, which balances sensitivity and specificity, then utilizes a region growing algorithm, starting from the seed point and moving along the main direction. Extend the function to automatically generate fault candidate surfaces within the residual set area.

[0074] When fault candidate surfaces are incorporated as hard constraints into minimum flow optimization, a distance constraint term is introduced into the objective function. The formula is as follows:

[0075] ;

[0076] in, Indicates a candidate fault surface. Let p be the distance from the grid cell to the fault plane. Control the attenuation range of the constraint effect. This is the constraint strength coefficient. This exponential term ensures that the interface geometry strictly follows the fault structure near the fault, while allowing for some flexibility in areas far from the fault.

[0077] After identifying residual concentration areas based on the three-dimensional spatial residual field, the system distinguishes between fault and non-fault anomalies through a multi-stage discrimination mechanism to ensure the geological rationality of fault candidate surfaces.

[0078] Calculate the directional consistency index of the residual vector field .when When the threshold is >0.7, the system initially identifies it as a potential fault zone. This threshold is determined based on historical work area statistics (e.g., out of 200 samples). 85% of the areas with a fault strength greater than 0.7 are true faults.

[0079] The system retrieves similar geological models from a prior geological knowledge base and performs coupled verification:

[0080] Calculate the Hausdorff distance similarity between residual abnormal patterns and known fault patterns in the knowledge base. When similarity Exceeding the second preset threshold (e.g.) When the value is greater than 0.8, the possibility of a fault is confirmed.

[0081] Verify the consistency between the residual vector direction and the regional tectonic stress field (e.g., the fault dip should be orthogonal to the principal stress direction). If the direction deviation is >30°, it is downgraded to "non-fault anomaly" (e.g., sedimentary facies transition).

[0082] Only if both conditions are met Only when S > 0.7 and S > 0.8, and passes the mechanical check, will the system confirm the region as a fault candidate surface and incorporate it as a hard constraint into the minimum flow optimization. Otherwise, it is marked as a "pending region" and the manual review interface is initiated.

[0083] This application achieves automatic reconstruction of complex fault systems by combining data-driven fault identification with interface optimization based on physical constraints. The automatic fault identification and constraint mechanism based on residual field analysis improves the automation and geometric accuracy of 3D geological modeling in complex structural regions, providing reliable technical support for engineering geological stability evaluation.

[0084] This application further proposes that, in the process of deducing the geometric morphology of geological interfaces, the following are also included:

[0085] A priori geological knowledge base was generated based on historical work area data. This knowledge base was established by integrating 3D geological models, well logging data, and seismic interpretation results from over 200 explored work areas, with a total data volume exceeding 50TB. Each geological model in the knowledge base contains three elements: structural style classification (such as thrust faults, extensional structures, strike-slip systems, etc.), geometric parameter statistics (including stratigraphic dip angle, fault density, layer thickness variation rate, etc.), and typical residual field response characteristics.

[0086] During the knowledge base training phase, the system employs deep metric learning technology, utilizing a pre-trained convolutional neural network model as the backbone network for feature extraction. In a preferred embodiment, the ResNet-50 model is used as the foundation, which has been pre-trained on the ImageNet large image dataset. The system inputs multi-source geological data, such as seismic profile slices and remote sensing image blocks, into the network to extract high-dimensional feature vectors. The embedding representation of the geological model is optimized using a triplet loss function.

[0087] ;

[0088] Where a is the anchor sample, p is the positive sample (same type construction), and n is the negative sample (different type construction). For Euclidean distance, marginal parameters Training was performed using the Adam optimizer with a learning rate of 10e-4 and a batch size of 32, for 500 epochs on a server equipped with an NVIDIA V100 GPU.

[0089] When extrapolating the geometric morphology of geological interfaces, similar geological models are retrieved from the prior geological knowledge base;

[0090] During the actual simulation, the system calculates the similarity between the residual field of the current work area and the geological models in the knowledge base in real time. An improved similarity method based on Hausdorff distance is adopted. The formula is as follows:

[0091] ;

[0092] in, For the feature point set of the current residual field, This is the feature point set of the knowledge base pattern.

[0093] The retrieved similar geological patterns are incorporated as soft constraints into the deduction process of the geometric morphology of the geological interface.

[0094] Among them, geological patterns with a similarity higher than the first preset threshold have stronger constraint weights;

[0095] When the residual anomaly pattern in the three-dimensional spatial residual field matches the known fault pattern in the prior geological knowledge base with a degree exceeding the second preset threshold, fault sensitivity analysis is automatically triggered.

[0096] Specifically, when the similarity exceeds the first threshold (For example, , When ∈ [0.7, 0.8], the system incorporates the matched geological model as a soft constraint into the objective function:

[0097] ;

[0098] in, Given the current model parameter distribution, For the prior distribution of similar geological patterns, These are the weighting coefficients calculated based on similarity. For balancing parameters, The KL divergence term ensures that the extrapolation results, while maintaining good data fit, conform to the prior constraints of geological laws.

[0099] When the detected residual abnormal pattern matches a known fault pattern more than the second threshold. (For example, At this time, the system automatically initiates fault sensitivity analysis. This analysis assesses the uncertainty of fault location through perturbation testing:

[0100] ;

[0101] in, For fault parameters, For parameter perturbation, For interface geometry response, This represents the number of samples. Based on this analysis, the system dynamically adjusts the convergence criteria and step size parameters of the optimization algorithm.

[0102] Through the above technical solution, this application improves the rationality and reliability of interface reconstruction in complex geological environments. In practical application in a deep reservoir in a western foreland basin, the system successfully identified a typical thrust-nap structure mode, guiding the optimization process to generate a geologically significant imbricate structural system that is highly consistent with the structural features revealed by subsequent drilling, effectively reducing the risk of multiple interpretations in structural analysis.

[0103] This application further proposes that the specific steps for performing depth-based layering and interlayer bidirectional confidence propagation on a three-dimensional geological interface model include:

[0104] The three-dimensional grid space is divided into multiple levels according to the geological time sequence, and a confidence distribution map is established for each level. The system first divides the three-dimensional geological model into several structural layers according to geological time boundaries based on regional stratigraphic timescales and standard borehole layering data. Typical layer thicknesses range from 50 to 200 meters; this scale is determined based on a comprehensive consideration of the frequency of regional unconformity development and the vertical resolution of seismic data. A unified confidence distribution map is used for modeling within each layer, and the initial confidence value is determined based on a comprehensive consideration of the data coverage density and quality within that layer.

[0105] Confidence propagation is performed from top to bottom and from bottom to top. In the top-down propagation, shallow information is used as a constraint to adjust the deep confidence, while in the bottom-up propagation, deep observations are used as support to correct the shallow confidence.

[0106] During the two-way confidence propagation process, the system establishes an inter-layer constraint model based on Markov random fields. The top-down propagation process is manifested as the constraint effect of the overlying strata on the underlying strata, which is achieved through conditional probability propagation:

[0107] ;

[0108] in, Z represents the geological parameters of the k-th layer, and Z is the normalization constant. This is the inter-layer transfer intensity parameter, derived from statistical analysis of the continuity of regional tectonic styles. This parameter ensures that the geometry of the stratigraphic interfaces maintains a reasonable gradient relationship in the vertical direction, avoiding abrupt changes in geological discontinuity.

[0109] Bottom-up propagation utilizes reliable deep observation data to refine shallow models, with particular emphasis on the constraints of borehole lithology data and high-quality seismic reflections. The propagation model is expressed as:

[0110] ;

[0111] in, For the observation data of the k-th layer, These are the model's predicted values. This is the observation error tolerance parameter. The system iteratively updates between layers using a confidence propagation algorithm. Each iteration includes a complete uplink and downlink propagation process until the norm of the confidence distribution at each layer is less than the convergence threshold. .

[0112] After the confidence level distribution stabilizes, the geometric properties of the 3D geological interface model are weighted and optimized based on the confidence levels of each layer to generate the final 3D geological model. The optimization objective function is as follows: Defined as:

[0113] ;

[0114] in, Let be the average confidence level of the k-th layer. Let c = 0.2 be the initial interface geometry and c be the smoothing weight. This optimization problem is solved using the preprocessed conjugate gradient method, which ensures both computational efficiency and solution stability.

[0115] This application effectively solves the common problem of inconsistency between stratigraphic time and structure in traditional 3D modeling methods by establishing a hierarchical confidence propagation mechanism under geological age constraints. It improves the geological consistency and spatial accuracy of 3D models in complex structural regions, and provides a more reliable geological framework for hydrocarbon accumulation analysis and reservoir prediction.

[0116] This application further proposes that, in performing inter-layer bidirectional confidence propagation, the physical rationality of monitoring the inter-layer geological structure is also included, specifically including:

[0117] After completing two-way confidence propagation, the system initiates an automated quality inspection process based on geomechanical principles. This inspection is based on two core principles: the principle of stratigraphic continuity requires that the same stratigraphic unit maintains the continuity of the sedimentary sequence in space, avoiding geological anomalies such as stratigraphic duplication or absence; the principle of fault cutting relationship ensures that the intersection relationship between faults and strata conforms to the basic laws of structural geology, that is, later tectonic structures should cut earlier strata, and structures of the same period should have consistent mechanical properties.

[0118] When a violation of geomechanical principles is detected in the interlayer geological structure, a proposed correction scheme is generated.

[0119] Based on the revised proposed scheme, the interlayer contact relationship is automatically adjusted while maintaining the data fit.

[0120] Among them, the principles of geomechanics include the principle of stratigraphic continuity and the principle of fault cutting relationships.

[0121] The system quantifies stratigraphic continuity by constructing an inter-layer geometric consistency index, which primarily detects the spatial abrupt change rate of stratigraphic thickness and the reasonable range of variation in stratigraphic dip angle. When the rate of change of stratigraphic thickness in the horizontal direction exceeds a threshold of 15 percent per 100 meters, the system marks it as a potential anomaly area. This threshold is determined based on statistical analysis of basin depositional rates and tectonic subsidence rates. Simultaneously, the system establishes a fault-stratigraphic intersection map, automatically identifying unreasonable phenomena such as stratigraphic faulting, duplication, or missing sections by analyzing the correspondence between stratigraphic units on both sides of the fault.

[0122] When a violation of physical plausibility is detected, the system initiates a multi-stage correction process. First, candidate correction schemes are generated based on a regional tectonic style knowledge base, making minimal necessary adjustments while maintaining the original data fit as much as possible. For stratigraphic continuity issues, the system employs a grid deformation algorithm based on elastic deformation theory, introducing smooth transition zones in anomalous areas to gradually restore the natural gradation characteristics of the strata. For fault cutting relationships, the system re-evaluates the fault attitude and activity phases, and, if necessary, locally optimizes the fault geometry to ensure a reasonable spatiotemporal relationship between the fault and the surrounding strata.

[0123] During the correction process, the system continuously monitors changes in data fitting quality. By establishing a trade-off mechanism for the objective function, it ensures that the corrected model meets the requirements of geomechanical principles without significantly reducing consistency with the original observation data. The entire correction process employs an iterative optimization strategy, re-evaluating physical rationality indicators after each iteration until all anomalies are eliminated or reduced to acceptable levels.

[0124] This technical solution enhances the scientific rigor and practicality of three-dimensional geological models by introducing an automated monitoring and correction mechanism based on geomechanical principles. In practical applications in complex tectonic zones in southern China, the system successfully detected and corrected three stratigraphic inversions and one fault relationship anomaly in the initial model, improving its structural rationality. Expert verification has shown that the corrected model better conforms to regional geological patterns, providing a more reliable scientific basis for deep oil and gas exploration and engineering geological stability assessment.

[0125] This application further proposes that the specific steps for quantifying the uncertainty index of the final three-dimensional geological model through fusion calculation include:

[0126] The local residual fluctuation index of each grid cell is extracted from the three-dimensional spatial residual field, the data coverage sparsity index of each grid cell is calculated from the gridded basic data, and the confidence index of each grid cell is extracted from the confidence distribution data.

[0127] The local residual fluctuation index, data coverage sparsity index, and confidence index are weighted and fused to generate an uncertainty quantification index.

[0128] In the calculation of local residual fluctuation index, the system establishes a 7×7×7 cubic analysis window centered on each grid cell, and characterizes the data fitting quality by calculating the coefficient of variation of the residual intensity within this window:

[0129] ;

[0130] in, The standard deviation of the residual intensity within the window. The mean, This is a numerically stable term. This indicator can effectively identify areas with low average residuals but exhibiting localized, drastic fluctuations. These areas often correspond to potential geological anomalies or data conflicts.

[0131] The data coverage sparsity index is obtained by calculating the spatial distribution characteristics of effective data points around each grid cell. The system first counts the number of sampling points for various geological observation data within a sphere centered on the target grid cell and with a radius of 200 meters. Then, combining this with data quality weights, it calculates the effective data density.

[0132] ;

[0133] in, Weighted valid data points This represents the theoretical maximum number of data points (determined based on the optimal sampling interval). This metric accurately reflects the areas of model uncertainty caused by data sparsity.

[0134] The confidence index is extracted directly from the results of hierarchical confidence propagation, but the system recalibrates it based on geological patterns. In particular, for special geological locations such as fault zones and stratigraphic pinch-out areas, the system introduces prior knowledge to correct the confidence level, ensuring the reasonableness of the index's geological significance.

[0135] The final fusion computation employs an adaptive weighting strategy based on the entropy weighting method:

[0136] ;

[0137] The weighting coefficients are dynamically determined based on the information entropy of each indicator: , Let be the standardized information entropy of the i-th indicator. This weighting method can automatically adjust the contribution of each indicator according to the actual data characteristics, avoiding bias caused by subjective weight settings.

[0138] The system also established a propagation model for uncertainty indicators. By analyzing the error propagation paths in each calculation stage, it ensures that the final uncertainty assessment accurately reflects the cumulative error throughout the entire process from data acquisition to model construction. This model specifically considers the correlation between different data sources, avoiding redundant calculations in uncertainty assessment.

[0139] Through the above technical solutions, this application has achieved a precise quantification of the reliability of three-dimensional geological models. This quantitative reliability assessment provides a key basis for exploration and development decisions, enabling geological researchers to optimize data acquisition schemes and model building strategies in a targeted manner, thereby improving the scientific nature and engineering effectiveness of geological modeling work.

[0140] This application further proposes that, after generating and outputting the 3D geological scene, in order to achieve dynamic optimization and verification of the model, the system also includes a real-time verification module, used for:

[0141] Virtual drilling sites are set in a three-dimensional geological scene. Based on an optimized three-dimensional geological interface model, the system uses a spatial interpolation algorithm to predict the complete stratigraphic sequence of each virtual drilling site, including stratigraphic interface depth, lithological combination, and distribution of physical property parameters.

[0142] Once the actual drilling operation has advanced to the corresponding location and acquired logging data, the system receives drilling lithology records, logging curves, and formation stratification data in real time through a standardized data interface. The system then initiates automatic comparative analysis, employing a dynamic time warping algorithm to calculate the degree of matching between the predicted formation sequence and the actual drilling data.

[0143] ;

[0144] in, To predict the depth of the i-th stratigraphic interface in the sequence, Let the depth be the j-th interface in the observation sequence. This algorithm identifies the optimal alignment path between two sequences. It effectively addresses sequence alignment issues caused by local variations in formation thickness, ensuring the accuracy of the correlation results.

[0145] System definition synthesis comparison deviation To quantify prediction accuracy, the formula is as follows:

[0146] ;

[0147] Where L is the total drilling depth. The number of layers was misjudged due to lithology. This represents the total number of floors. and These represent the predicted and actual formation thicknesses, with weighting coefficients, respectively. Determined based on expert experience. In this embodiment, when comparing deviations... Exceeding the third preset threshold At that time, the system automatically marks the area as a high-uncertainty anomaly zone. ∈[0.1,0.2], with a typical value such as 0.15.

[0148] For identified anomaly regions, the system initiates a multi-level local model reconstruction mechanism. First, it analyzes the spatial distribution characteristics of the deviation to determine the scope of influence and the granularity of reconstruction. Then, the system adjusts the weight allocation of the corresponding data sources in the adaptive fusion module to reduce the impact of unreliable data while enhancing the constraint strength of the 3D spatial residual field in the anomaly region. The reconstruction process prioritizes incremental calculation, recalculating only the mesh within the anomaly region and its influence range to maintain the overall stability of the model. The system also establishes a deviation cause diagnosis mechanism, automatically identifying the dominant factors causing prediction deviations by analyzing residual field patterns, data coverage, and confidence distribution, providing clear guidance for subsequent data acquisition and model optimization.

[0149] After generating and outputting the 3D geological scene, the system dynamically optimizes the model through a real-time verification module. This real-time verification module, along with the reconstruction and quantization modules, forms a closed-loop feedback mechanism. The specific interaction process includes:

[0150] When the deviation between the actual drilling data and the predicted formation sequence exceeds the third preset threshold (e.g.) When the value is 0.15 and Δ≥0.15, the real-time verification module automatically identifies the abnormal area and generates a local reconstruction instruction. This local reconstruction instruction includes the spatial coordinates of the abnormal area, the type of deviation (such as depth deviation, lithological misjudgment), and confidence degradation information.

[0151] The local reconstruction command is sent to the reconstruction module via the system's internal bus. Simultaneously, the real-time verification module packages the actual drilling data into a standardized format (including formation depth, lithology coding, and logging curves) and updates it synchronously to the preprocessing module's database, ensuring that the reconstruction is based on the latest data.

[0152] After receiving the instruction, the reconstruction module starts the incremental calculation mode:

[0153] First, based on the spatial range of the anomalous region, the weights of relevant data sources in the adaptive fusion module are dynamically adjusted (e.g., the weights of conflicting data sources are reduced); the constraint strength of the 3D spatial residual field in the anomalous region is strengthened (e.g., the residual weight coefficient β is increased to 0.5); only the mesh of the anomalous region and its buffer zone (radius 100-200 meters) is recalculated, rather than globally reconstructed, to improve efficiency.

[0154] After the partial reconstruction is completed, the system automatically pushes the new model to the visualization module to update the scene and reruns the real-time verification module for a second comparison. If the deviation still exceeds the threshold, the above process is iterated until convergence.

[0155] This application enables dynamic improvement and accuracy enhancement of the three-dimensional geological model, improves the practicality of the geological model, provides continuous optimization technical support for exploration decision-making, and effectively reduces drilling operation risks.

[0156] The following is a specific example of a three-dimensional reconstruction and analysis system for geological structures based on multimodal data fusion:

[0157] In an oil and gas exploration project in a mountainous area in western China, a three-dimensional geological modeling of a complex area with developed thrust faults is required. This area contains seismic data (dominant frequency 40Hz), sparse borehole data (average spacing 150m), gravity anomaly data (resolution 0.5km), and high-resolution remote sensing imagery (1m pixel). Significant spatial conflicts exist between the shallow (<500m) and deep (>2000m) layers in the multi-source data: the stratigraphic depth interpreted by seismic analysis deviates from the lithological interface revealed by boreholes by 30-80m; the angle between the gravity anomaly gradient direction and the strike of the linear structure in the remote sensing exceeds 45°; and traditional fixed-weight fusion models repeatedly exhibit stratigraphic overlap and fault continuity interruptions.

[0158] The system is deployed in the exploration area data center. The preprocessing module accesses multi-source data through a standardized interface: seismic data is acquired using a 24-bit seismic recorder (sampling rate 1ms), and processed with static correction and pre-stack time migration; borehole data is imported from the GeoLog database, including core logging and well logging curves (density, resistivity) for 12 exploration boreholes; gravity data is acquired using a LaCoste-Romberg gravimeter (accuracy ±0.01mGal), and processed with topographic correction and Bouguer correction; remote sensing imagery uses WorldView-3 satellite data, and is processed with orthorectification and linear structural enhancement. The preprocessing module unifies all data to the UTM projection coordinate system (WGS84 ellipsoid), generates a 50m×50m×10m three-dimensional grid data volume through Kriging interpolation, and simultaneously calculates the data coverage density map, showing that the data coverage in the deep structural area (>1500m) is only 35% of that in the shallow area.

[0159] In a three-dimensional grid space, the system uses a cube with a side length of 100m as the spatiotemporal neighborhood unit to perform parallel computation of multi-source data consistency:

[0160] The shallow fit was calculated by matching the seismic reflection surface (point set A) and the borehole lithological interface (point set B) using the iterative nearest point algorithm and the Hausdorff distance. (Normalized score 0.85), deep fit (Normalized score: 0.32).

[0161] Cosine similarity analysis was performed on the remotely sensed linear structures (extracted by the Canny operator) and the gravity gradient field (calculated by the Sobel operator) in the shallow structural region. (Directional angle <15°), deep concealed structural zone (Directional angle > 40°).

[0162] By combining the basic confidence scores of each data source (earthquake 0.8, borehole 0.9, gravity 0.7, remote sensing 0.6), the spatiotemporal consistency score is calculated by weighted geometric mean. After Min-Max normalization, the weight of deep borehole data is increased to 0.75 (shallow data 0.55), thus achieving the dominant fusion of highly reliable data in conflict areas.

[0163] After mapping the multi-source data to a 3D grid, the deviation between the observed values ​​and the fused estimates is calculated:

[0164] Scalar deviation is calculated using the relative error formula: The average scalar bias of deep mesh elements reaches 0.32 (0.15 for shallow mesh); the vector bias is calculated by extracting the direction vector through principal component analysis and then calculating the angle difference. The peak θ value near the fault zone reaches 38°.

[0165] The fusion generates a three-dimensional spatial residual field. The high residual intensity area (>0.45) is concentrated in the candidate zone of the thrust fault, and the direction vector shows that the interface dip is consistent with the regional tectonic stress field.

[0166] The interface optimization is modeled as a minimum flow problem, with the objective function being: Substituting α=1.0 and β=0.3, the gradient descent algorithm is used for iterative optimization (learning rate 0.01 (preferred range 0.001~0.1), maximum iterations 5000). When the objective function changes... The convergence is achieved, and the output optimized interface has a smooth curvature at the fault inflection point (<0.02 / m), which improves the fit with the borehole lithology interface (Hausdorff distance is reduced to <25m).

[0167] Wavelet transform multi-scale analysis of the residual field, residual fluctuation threshold. Identify residual concentration areas and calculate directional consistency index. .when At that time, along the main direction ( (Generate fault candidate surfaces as hard-constrained integration flow conservation equations, with the constraint terms as follows:) , ( This ensures that the fault interface and the residual direction are highly consistent.

[0168] The model is divided into three tectonic layers (Paleozoic, Mesozoic, and Cenozoic) according to geological time, and a Markov random field model propagation confidence score is established. When propagating from top to bottom, the overlying strata constrain the dip angle variation rate of the underlying strata to <8° / 100m; when propagating from bottom to top, deep borehole data corrects the shallow seismic interpretation bias, and the final interlayer contact relationship conforms to the principle of stratigraphic continuity (no duplication / missing phenomena).

[0169] By fusing uncertainty indicators, residual fluctuations are extracted. ), data coverage sparsity ( ) and confidence level ( Dynamic weighting using the entropy weight method The deep structural region was determined to have a U>0.65 (high uncertainty), while the shallow region had a U<0.3 (low uncertainty). The visualization module marked the high-risk areas with a red gradient.

[0170] Five virtual drilling points were set up in the 3D scene, and the predicted stratigraphic sequence was compared with the actual drilling data. Among them, the prediction deviation of well No. 2 (deep structural zone) exceeded the threshold (Δ=0.18), and the system automatically triggered local reconstruction: the weight of the seismic data in this area was adjusted to 0.45, and the residual field constraint was enhanced (β=0.5). After reconstruction, the interface depth deviation was reduced to <15m, which met the exploration accuracy requirements.

[0171] Through dynamic fusion of multimodal data and residual-driven optimization, the system has achieved high-precision construction of three-dimensional geological models in complex structural areas: it has improved the continuity of fault identification, enhanced the matching degree between interface geometry and regional tectonic stress field, and enabled the model to be continuously optimized under the feedback of drilling data through a real-time verification closed-loop mechanism, providing a reliable geological framework for subsequent oil and gas resource assessment and reducing exploration decision-making risks.

[0172] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A three-dimensional reconstruction and analysis system for geological structures based on multimodal data fusion, characterized in that: include: The preprocessing module is used to acquire multi-source heterogeneous geological observation data and perform spatiotemporal normalization processing to generate basic data and data coverage density maps. An adaptive fusion module is used to perform local spatiotemporal consistency analysis on the basic data, generate a dynamic adaptive weight field, and perform weighted fusion on the geological observation data to generate a primary three-dimensional geological data volume. The construction and optimization module is used to construct a vector-form three-dimensional spatial residual field based on the difference between the primary three-dimensional geological data volume and the geological observation data, and to deduce the geometric morphology of the geological interface and generate an optimized three-dimensional geological interface model using the three-dimensional spatial residual field as the driving force. The reconstruction module is used to perform depth-based layering of the three-dimensional geological interface model and execute inter-layer bidirectional confidence propagation to generate a final three-dimensional geological model with inter-layer consistency and corresponding confidence distribution data. The quantification module is used to calculate the uncertainty quantification index of the final three-dimensional geological model by fusing the three-dimensional spatial residual field, the data coverage density map, and the confidence distribution data. The visualization module is used to visualize the final three-dimensional geological model and the uncertainty quantification index, and generate and output a three-dimensional geological scene. The specific steps for performing local spatiotemporal consistency analysis on the basic data include: Define a spatiotemporal cube neighborhood for each grid cell in the underlying data; Within the neighborhood of the spatiotemporal cube, the spatial positional alignment between the seismic reflection surface and the borehole lithology interface, as well as the directional consistency between the gravity anomaly gradient field direction and the linear structural direction of the remote sensing image, are calculated. Based on the spatial location matching degree and the directional consistency, and combined with the preset basic confidence level of the geological observation data, a spatiotemporal consistency score is generated for each grid cell. The spatiotemporal consistency score is normalized into the dynamic adaptive weight field.

2. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 1, characterized in that: The specific steps for constructing a vector-form three-dimensional spatial residual field based on the differences between the primary three-dimensional geological data volume and the geological observation data include: The geological observation data from different sources are mapped to a unified three-dimensional grid space; In each grid cell of the three-dimensional grid space, the scalar and vector deviations between the observed values ​​from each data source of the geological observation data and the fused estimated values ​​of the primary three-dimensional geological data volume are calculated. Based on the scalar and vector deviations of all grid cells, the three-dimensional spatial residual field containing residual intensity and direction is constructed.

3. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 2, characterized in that: The specific steps for deducing the geometric morphology of geological interfaces and generating optimized three-dimensional geological interface models include: The geological interface optimization problem is modeled as a minimum flow problem that satisfies flow conservation constraints on the three-dimensional grid space. The flow conservation constraint requires that the total flow into each grid cell be equal to the sum of the residual source intensity and the outflow within that grid cell; The objective function is solved by an iterative optimization algorithm. The objective function includes a residual flux term based on the three-dimensional spatial residual field and a smoothing penalty term based on the interface curvature. When the change in the objective function value is lower than a preset change threshold, the currently optimized geological interface is output as the optimized three-dimensional geological interface model.

4. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 3, characterized in that: After constructing a vector-form three-dimensional spatial residual field based on the difference between the primary three-dimensional geological data volume and the geological observation data, the method further includes: Based on the three-dimensional spatial residual field, identify the residual concentration region; Based on the consistency of the direction of the residual vector, fault candidate surfaces are automatically generated within the residual set region; The candidate fault surface is incorporated as a hard constraint into the solution process of the minimum flow problem.

5. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 1, characterized in that: The process of deducing the geometric morphology of the geological interface also includes: A priori geological knowledge base is generated based on historical work area data. When extrapolating the geometric morphology of geological interfaces, similar geological patterns are retrieved from the prior geological knowledge base; The retrieved similar geological patterns are incorporated as soft constraints into the deduction process of the inferred geological interface geometry. Among them, geological patterns with a similarity higher than the first preset threshold have stronger constraint weights; When the degree of matching between the residual abnormal pattern in the three-dimensional spatial residual field and the known fault pattern in the prior geological knowledge base exceeds a second preset threshold, fault sensitivity analysis is automatically triggered.

6. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 3, characterized in that: The specific steps for performing depth-based stratification and inter-layer bidirectional confidence propagation on the three-dimensional geological interface model include: The three-dimensional grid space is divided into multiple levels according to the geological time sequence, and a confidence distribution map is established for each level. Confidence propagation is performed from top to bottom and from bottom to top. In the top-down propagation, shallow information is used as a constraint to adjust the deep confidence, while in the bottom-up propagation, deep observations are used as support to correct the shallow confidence. After the confidence level distribution stabilizes, the geometric properties of the three-dimensional geological interface model are weighted and optimized based on the confidence levels of each layer to generate the final three-dimensional geological model.

7. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 6, characterized in that: The implementation of inter-layer bidirectional confidence propagation also includes monitoring the physical plausibility of the inter-layer geological structure, specifically including: When a violation of geomechanical principles is detected in the interlayer geological structure, a proposed correction scheme is generated. Based on the proposed correction scheme, the interlayer contact relationship is automatically adjusted while maintaining the data fit. The geomechanical principles mentioned include the principle of stratigraphic continuity and the principle of fault cutting relationships.

8. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 1, characterized in that: The specific steps for quantifying the uncertainty index of the final three-dimensional geological model through fusion calculation include: The local residual fluctuation index of each grid cell is extracted from the three-dimensional spatial residual field, the data coverage sparsity index of each grid cell is calculated from the gridded basic data, and the confidence index of each grid cell is extracted from the confidence distribution data. The local residual fluctuation index, the data coverage sparsity index, and the confidence index are weighted and fused to generate the uncertainty quantification index.

9. The geological structure three-dimensional reconstruction and analysis system based on multimodal data fusion according to claim 1, characterized in that: After generating and outputting the 3D geological scene, the system also includes a real-time verification module for: Virtual drilling points are set in the three-dimensional geological scene; Based on the optimized three-dimensional geological interface model, the stratigraphic sequence at the virtual drilling site is predicted; When actual drilling data is received, the actual drilling data is compared with the predicted formation sequence in real time; When the comparison deviation exceeds the third preset threshold, the abnormal area is automatically identified and local model reconstruction is triggered.

Images