Three-dimensional geological modeling method, device, equipment and medium under complex fracture structure
By employing a phased deep learning modeling strategy, LSTM and Transformer models are used to reconstruct fracture bodies and stratigraphic categories, solving the inaccuracy problem of 3D geological models under complex fracture structures. This achieves 3D solid representation of fracture bodies and continuity of stratigraphic spatial relationships, thereby improving the accuracy and reliability of the model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU URBAN PLANNING & DESIGN SURVEY RES INST
- Filing Date
- 2025-12-01
- Publication Date
- 2026-05-01
AI Technical Summary
Existing three-dimensional geological models cannot fully depict the three-dimensional spatial distribution and connectivity of fracture bodies under complex fault structures, leading to stratigraphic discontinuities and logical conflicts, which affects the reliability of reservoir fine evaluation and disaster risk analysis.
A phased deep learning modeling strategy is adopted. First, the three-dimensional spatial distribution of the fault body is reconstructed using an LSTM model. Then, a complete three-dimensional geological model is generated using a Transformer model to ensure the continuity of the spatial relationship of the strata at the fault boundary.
It significantly improves the accuracy and reliability of three-dimensional geological modeling under complex fault structures, solves the problem of discontinuous strata at fault boundaries, and enhances the accuracy and geological rationality of the model.
Smart Images
Figure CN121962483A_ABST
Abstract
Description
Three-dimensional geological modeling methods, devices, equipment and media under complex fault structures Technical Field
[0001] This application relates to the fields of geological information technology and digital geological modeling technology, and in particular to a method, apparatus, equipment and medium for three-dimensional geological modeling under complex fault structures. Background Technology
[0002] When faced with stratigraphic structures containing complex faults and folds, most existing three-dimensional geological model reconstruction technologies simplify the fault structure into a two-dimensional plane and process it into a grid. This makes it impossible to fully depict the global features of the fault body, such as its three-dimensional spatial distribution and connectivity, as well as local details such as the undulations and thickness variations of the fault surface.
[0003] The direct consequence of this simplification is that the model exhibits severe stratigraphic discontinuities and logical conflicts at fault boundaries, making it difficult to accurately simulate the spatial displacement of strata caused by fault slippage. Neither multi-point statistical algorithms nor traditional neural network methods have been able to achieve dynamic and coordinated modeling of the fault as a three-dimensional geological entity and the surrounding rock strata, resulting in insufficient geological rationality of the reconstruction results and seriously affecting the reliability of reservoir fine evaluation and disaster risk analysis. Summary of the Invention
[0004] The purpose of this application is to provide a method, apparatus, equipment, and medium for three-dimensional geological modeling under complex fault structures, which can effectively improve the accuracy and reliability of three-dimensional geological modeling under complex fault structures.
[0005] To achieve the above objectives, a first aspect of this application provides a three-dimensional geological modeling method for complex fault structures, comprising: acquiring three-dimensional geological data containing fault zone identifiers and stratigraphic category labels; serializing the fault points in the three-dimensional geological data according to their geological strikes to generate a fault point coordinate sequence; inputting the fault point coordinate sequence into a pre-trained first deep learning model, outputting the probability that each spatial coordinate point belongs to a fault body, and generating a fault body reconstruction result according to a preset probability threshold; using the fault body reconstruction result as a spatial constraint, inputting the three-dimensional spatial coordinates of the area to be modeled into a pre-trained second deep learning model, outputting the stratigraphic category to which each coordinate point belongs, and generating a complete three-dimensional geological model.
[0006] Compared with existing technologies, the three-dimensional geological modeling method for complex fault structures provided in this application has the following advantages: By adopting a staged deep learning modeling strategy, a first deep learning model is first used to learn from sequential fault point data to generate a reconstruction result that accurately reflects the three-dimensional spatial distribution of the fault body, overcoming the problem of incomplete representation caused by simplifying the fault to a two-dimensional plane in traditional methods; then, the reconstruction result of the fault body is used as a spatial constraint condition input into a second deep learning model to achieve high-precision prediction of stratigraphic categories, ensuring the continuity and geological rationality of the spatial relationship of stratigraphy at the fault boundary. This method, through the synergistic effect of the two models, effectively solves the problem of inaccurate three-dimensional geological modeling under complex fault conditions, significantly improving the accuracy and reliability of the model.
[0007] In some embodiments, the step of serializing the fracture points in the three-dimensional geological data according to their geological orientation to generate a fracture point coordinate sequence includes: combining the three-dimensional spatial coordinates (x, y, z) of the fracture points and their associated physical field feature data into a 3+N dimensional feature vector, and arranging them into a time series according to the fracture orientation; this time series is the fracture point coordinate sequence containing geometric features and physical field features; wherein, N is the number of physical field features, and the physical field features include one or more of velocity, density, resistivity or seismic wave attenuation coefficient.
[0008] In some embodiments, the first deep learning model is a long short-term memory network model, whose construction and training include: the model input dimension is 3+N, N≥0; the hidden layer contains multiple LSTM units; the output layer uses the sigmoid activation function to output the probability of breakpoint assignment; the model training uses binary cross-entropy as the loss function and adopts an early stopping mechanism to prevent overfitting.
[0009] In some embodiments, when there are multiple fractures in the modeling area, the fracture reconstruction process is performed separately for each fracture, and the reconstructed fractures are superimposed and fused in the same simulation grid according to the order set by geological laws to generate the fracture reconstruction result containing multiple fractures.
[0010] In some embodiments, before inputting the three-dimensional spatial coordinates of the region to be modeled into a pre-trained second deep learning model, the method further includes: marking the spatial coordinate points identified as fracture bodies in the fracture reconstruction results as having no strata or a specific mask value in the stratigraphic category label.
[0011] In some embodiments, the second deep learning model is a Transformer model, the construction and training of which include: the model input is normalized three-dimensional spatial coordinates; the encoder part contains multiple cascaded Transformer blocks, each Transformer block contains a multi-head self-attention layer and a feedforward neural network layer; the output layer uses a softmax activation function to output the probability distribution of all preset stratum categories; the model training uses classification cross-entropy as the loss function and a learning rate decay strategy.
[0012] In some embodiments, the step of outputting the stratigraphic category of each coordinate point to generate a complete three-dimensional geological model includes: for each input coordinate, selecting the category with the highest probability in the stratigraphic category probability distribution output by the second deep learning model as the final predicted stratigraphic category of that coordinate point, and generating the complete three-dimensional geological model.
[0013] To achieve the above objectives, a second aspect of this application provides a three-dimensional geological modeling device for complex fault structures. The device includes: a processing module for acquiring three-dimensional geological data containing fault zone identifiers and stratigraphic category labels; serializing the fault points in the three-dimensional geological data according to their geological strikes to generate a fault point coordinate sequence; an input module for inputting the fault point coordinate sequence into a pre-trained first deep learning model, outputting the probability that each spatial coordinate point belongs to a fault body, and generating a fault body reconstruction result according to a preset probability threshold; and a generation module for using the fault body reconstruction result as a spatial constraint condition, inputting the three-dimensional spatial coordinates of the area to be modeled into a pre-trained second deep learning model, outputting the stratigraphic category to which each coordinate point belongs, and generating a complete three-dimensional geological model.
[0014] To achieve the above objectives, a third aspect of this application provides an electronic device, the electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method described in the first aspect.
[0015] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium comprising a stored computer program, wherein the computer program, when executed, controls the device containing the computer-readable storage medium to perform the method described in the first aspect. Attached Figure Description
[0016] Figure 1 is a flowchart of a three-dimensional geological modeling method under complex fault structures provided in an embodiment of this application; Figure 2 is another flowchart of a three-dimensional geological modeling method under complex fault structures provided in an embodiment of this application; Figure 3 is a result diagram of a three-dimensional geological modeling method under complex fault structures provided in an embodiment of this application; Figure 4 is a structural diagram of a three-dimensional geological modeling device under complex fault structures provided in an embodiment of this application; Figure 5 is a hardware structure diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0018] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.
[0019] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.
[0020] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.
[0021] In fields such as oil and gas exploration, geological hazard prediction, and resource assessment, 3D reconstruction technology of complex strata is a key foundation for achieving accurate geological interpretation and resource evaluation. As geological exploration work continues to deepen, the stratigraphic structures encountered are becoming increasingly complex, and traditional 3D reconstruction methods can no longer meet the accuracy and efficiency requirements of modern exploration.
[0022] Currently, mainstream 3D reconstruction techniques mainly rely on physical detection methods such as seismic imaging, synchronous vibration, and nuclear magnetic resonance, supplemented by geometric modeling algorithms based on mesh generation. These methods can achieve certain results in relatively simple strata; however, when complex geological structures such as faults, stratigraphic inversions, and folds are involved, the reconstruction results often deviate significantly from the actual geological features.
[0023] Ultimately, the shortcomings of existing technologies lie in their oversimplification of fracture structures. Specifically, these methods generally treat fractures as two-dimensional planar elements rather than geological entities with three-dimensional spatial distribution. This results in the inability to fully characterize the global spatial distribution features of the fracture body (such as its extent and connectivity) and its local geometry (such as the undulations and thickness variations of the fracture surface). This simplification directly leads to geometric contradictions in the strata surrounding the fracture zone: when a fault causes strata displacement, grid-based modeling methods struggle to accurately simulate the relative displacement and spatial relationships of the strata on both sides of the fault, causing discontinuities or logical conflicts in the reconstructed model at the fracture boundary.
[0024] It is worth noting that while multi-point statistical modeling methods impose constraints by pre-setting fault plane locations, they neglect the dynamic interaction between the fault body and the surrounding strata. Furthermore, traditional neural network algorithms, limited by their model architecture, treat faults as independent geometric surfaces, failing to fully consider their integrity as geological units. These shortcomings further exacerbate the distortion of stratigraphic boundaries in complex fault systems.
[0025] The aforementioned technical deficiencies not only significantly reduce the reliability of three-dimensional geological models but also directly affect the accuracy of resource assessment results, thus hindering the scientific nature of engineering decisions. In practical applications, geological models constructed based on existing technologies are insufficient to meet the needs of detailed reservoir characterization and cannot provide a reliable basis for geological hazard risk assessment, ultimately becoming a technical bottleneck restricting the improvement of exploration efficiency and engineering safety levels.
[0026] Based on this, the embodiments of this application provide a method, apparatus, equipment and medium for three-dimensional geological modeling under complex fault structures, which can effectively improve the accuracy and reliability of three-dimensional geological modeling under complex fault structures.
[0027] Please refer to Figure 1. Figure 1 is an optional flowchart of a three-dimensional geological modeling method under complex fault structures provided in the embodiments of this application. The method in Figure 1 may include, but is not limited to, steps S101 to S103.
[0028] Step S101: Obtain 3D geological data containing fault zone identifiers and stratigraphic category labels; serialize the fault points in the 3D geological data according to their geological strikes to generate a fault point coordinate sequence; Step S102: Input the fault point coordinate sequence into a pre-trained first deep learning model, output the probability that each spatial coordinate point belongs to the fault body, and generate the fault body reconstruction result according to a preset probability threshold; Step S103: Using the fault body reconstruction result as a spatial constraint, input the 3D spatial coordinates of the area to be modeled into a pre-trained second deep learning model, output the stratigraphic category to which each coordinate point belongs, and generate a complete 3D geological model.
[0029] Steps S101 to S103, as illustrated in this embodiment, employ a phased deep learning modeling strategy. First, a first deep learning model is used to learn from the sequential fault point data, generating a reconstruction result that accurately reflects the three-dimensional spatial distribution of the fault body. This overcomes the incomplete representation problem caused by simplifying the fault to a two-dimensional plane in traditional methods. Then, this fault reconstruction result is used as a spatial constraint input to a second deep learning model, achieving high-precision prediction of stratigraphic categories and ensuring the continuity and geological rationality of the spatial relationships of stratigraphic units at the fault boundaries. This method, through the synergistic effect of the two models, effectively solves the problem of inaccurate three-dimensional geological modeling under complex fault conditions, significantly improving the accuracy and reliability of the model.
[0030] In step S101 of some embodiments, the fault zone identifier can be binary label data that marks the spatial location of the fault zone, and the missing area is marked with a specific value; the stratigraphic category label can be a geological classification identifier that distinguishes different lithological strata; and the serialization process can be a data processing method that organizes discrete points into an ordered sequence according to the geological strike of the fault.
[0031] In some embodiments, the fault points in the three-dimensional geological data are serialized according to their geological orientation to generate a fault point coordinate sequence, including: combining the three-dimensional spatial coordinates (x, y, z) of the fault points and their associated physical field feature data into a 3+N dimensional feature vector, and arranging them into a time series according to the fault orientation; the time series is the fault point coordinate sequence containing geometric features and physical field features; wherein, N is the number of physical field features, and the physical field features include one or more of velocity, density, resistivity or seismic wave attenuation coefficient.
[0032] Specifically, the fracture point can be a discrete coordinate point in three-dimensional space that identifies the location of the fracture zone; the physical field characteristics can be parameters such as velocity, density, resistivity, and seismic wave attenuation coefficient obtained through geophysical exploration; the feature vector can be a multidimensional data representation composed of coordinates and physical field characteristics; and the time series can be a data sequence arranged in an orderly manner according to the fracture direction, which is a sequence of fracture point coordinates containing geometric and physical field characteristics.
[0033] After acquiring the raw 3D geological data, the labeled fault point data is extracted and sorted according to the natural extension direction of the fault structure in geological space. Specifically, the 3D coordinates (x, y, z) of each fault point are used as the basic feature, while simultaneously integrating N physical field features such as velocity, density, and resistivity to form a 3+N dimensional feature vector, where 3 represents the 3D coordinates and N represents the number of physical field features. Then, according to the natural extension direction of the fault structure in geological space, all feature vectors are arranged in an ordered manner to construct a sequence of data with spatiotemporal characteristics. The serialization processing method in this embodiment is a natural adaptation to the characteristics of geological data, enabling subsequent models to effectively capture the spatial continuity of faults and integrate multi-source heterogeneous data to achieve a more comprehensive fault body representation. Addressing the scarcity of geological data, this method enhances the expression of fault features by utilizing existing physical field data, solving the problem of incomplete representation caused by traditional methods relying solely on geometric data.
[0034] It is important to note that this application requires training two sets of deep learning models. One set is an LSTM model used for fracture reconstruction, with normalized coordinates as input and corresponding binary labeled data as output. The other set is a Transformer architecture model used to predict the spatial distribution of various strata. During data preprocessing, the input data also needs to be normalized in coordinates and binarized in attributes. The preprocessed data is then divided into training, validation, and test sets in an 8:1:1 ratio. The training set is used for parameter training of the LSTM model (fracture reconstruction) and the Transformer model (stratum filling), respectively. The validation set is used to monitor the training performance of the two models and trigger an early stopping mechanism. The test set is used to independently verify the final modeling accuracy, ensuring the independence and objectivity of model training, validation, and testing.
[0035] Please refer to Figure 2, which is another flowchart of a three-dimensional geological modeling method under complex fault structures. The specific steps are as follows: First, the initial grid model is used as the basic data carrier to enter the data processing stage. In this stage, two preprocessing operations are completed simultaneously: coordinate normalization and attribute binarization. After coordinate normalization, training dataset 2 is generated and is input into the LSTM architecture fault reconstruction model for training (the input set of data to be simulated is also connected to the model). Matching the design of the LSTM model of "input normalized coordinates and output binarized labels", the final output is a fault structure model representing the spatial morphology of the fault body. At the same time, after attribute binarization, training dataset 1 is generated and is input into the Transformer architecture stratigraphic filling model to complete the training, so that the model first learns the basic stratigraphic distribution pattern. After both models have been trained, the fault structure model output by the LSTM model is used as a spatial constraint and connected to the trained Transformer architecture stratigraphic filling model. The model predicts the spatial distribution of each stratum based on the fault constraints and finally outputs a high-precision three-dimensional geological model of the fault structure (as shown in Figure 3), thus fully implementing the phased modeling collaborative logic of "fault body reconstruction - stratigraphic filling".
[0036] In step S102 of some embodiments, the first deep learning model can be a fracture recognition network based on an LSTM architecture; the fracture reconstruction result can be predicted data reflecting the three-dimensional spatial distribution of the fracture. The serialized fracture point coordinates are input into a pre-trained LSTM model. This model has an input dimension of 3+N, with multiple LSTM units configured in the hidden layers, learning long-term spatial dependencies in the fracture point sequence through its gating mechanism. The output layer uses the sigmoid function to calculate the probability value of each coordinate point belonging to the fracture, performs binarization judgment through a preset probability threshold of 0.5, and finally generates a complete three-dimensional fracture reconstruction result, accurately depicting the spatial distribution morphology of the fracture.
[0037] In some embodiments, the first deep learning model is a long short-term memory network model, and its construction and training include: the model input dimension is 3+N, N≥0; the hidden layer contains multiple LSTM units; the output layer uses the sigmoid activation function to output the probability of breakpoint assignment; the model training uses binary cross-entropy as the loss function and adopts an early stopping mechanism to prevent overfitting.
[0038] Specifically, the LSTM unit can be a neural network unit with a memory gating mechanism, which can learn long-term dependencies in sequence data; the Sigmoid activation function can be a function that compresses the output value to the interval [0, 1], which is suitable for probability prediction; the binary cross-entropy is a loss function used for binary classification problems, which measures the difference between the predicted probability and the true label; the early stopping mechanism can be a strategy to terminate training in advance when the performance on the validation set no longer improves.
[0039] The Long Short-Term Memory (LSTM) model structure is designed as follows: Input layer: The input dimension is 3+N (N≥0), receiving serialized three-dimensional coordinate data and multi-source physical field features, which can accurately express the spatial dependencies and physical property relationships between fracture points. For example, when fusing a three-dimensional velocity structure model, the input layer will contain x, y, z coordinates and velocity values. LSTM can learn the correlation between velocity changes and fracture spatial distribution, thereby more accurately identifying fracture boundaries.
[0040] Hidden layer: Contains 128 LSTM units. Through its unique gating mechanism (input gate, forget gate, output gate), it can effectively learn the long-term spatial dependencies in the breakpoint sequence, capture the extension range, connectivity and local geometry of the break (such as the undulation of the break surface and the thickness change); at the same time, a dropout rate of 0.2 is set to prevent the model from overfitting and ensure that the model has good generalization ability.
[0041] Output layer: Contains 1 unit, uses the sigmoid activation function, maps the network output to the interval [0, 1], and outputs the probability value of the break point belonging to the break zone.
[0042] The LSTM model training parameters are as follows: Optimizer: The Adam optimizer is used with a learning rate of 0.001. This optimizer can adaptively adjust the learning rate of each parameter, effectively improving training efficiency.
[0043] Loss function: We choose binary cross-entropy, which can accurately measure the difference between the fracture markers predicted by the model and the actual fracture markers, providing a clear direction for model training optimization.
[0044] Training rounds: 100 rounds of iterative training are conducted using the training set as input, with 64 samples processed in each batch to ensure that the model can fully learn the data features.
[0045] Early stopping mechanism: Implement a strict early stopping mechanism. When the validation set loss does not improve for 10 consecutive rounds or reaches the preset loss threshold, the training process is automatically terminated to avoid the model overfitting the training data and to ensure the model's generalization performance on unknown data.
[0046] The coordinate sequence of the region to be predicted is input into a trained LSTM model. The model outputs a probability value for each coordinate point that it belongs to the fault zone. A probability threshold of 0.5 is set, and coordinate points with a predicted probability greater than 0.5 are marked as fault zone regions. The final generated fault reconstruction results present the precise location and shape in three-dimensional space. It not only includes the connectivity features of the fault network but also clearly shows the extension range and three-dimensional spatial distribution features of the fault body. This upgrades the fault structure from a traditional two-dimensional planar representation to a complete three-dimensional ontological representation, providing precise spatial constraints for subsequent stratigraphic filling.
[0047] In some embodiments, when there are multiple fractures in the modeling area, the fracture reconstruction process is performed separately for each fracture, and the reconstructed fractures are superimposed and fused in the same simulation grid according to the order set by geological laws to generate a fracture reconstruction result containing multiple fractures.
[0048] Specifically, the fracture reconstruction process can be an independent process of three-dimensional characterizing a single fracture using an LSTM model; the order of geological laws can be the processing order determined based on geological principles such as fracture formation time and cutting relationships; and the superposition and fusion can be a synthetic operation that integrates multiple fracture bodies in a unified three-dimensional mesh.
[0049] When multiple fractures exist in the modeling region, a fracture reconstruction process is first performed on each fracture individually, generating its own 3D fracture representation using an LSTM model. Then, based on the geological characteristics of the fracture system (such as formation time and cutting relationships), the processing order is determined, and the fracture bodies are sequentially superimposed onto the same 3D simulation mesh. During the superposition process, subsequent fracture bodies will cover the overlapping areas of previous fracture bodies, ultimately generating a 3D reconstruction result containing the complete fracture system.
[0050] It should be noted that when multiple faults are superimposed and merged, the final fault reconstruction result includes a set of three-dimensional spatial coordinates of all faults. During subsequent masking, all coordinate points in this set are uniformly marked as having no strata or a specific mask value to ensure that strata filling in multi-fault scenarios still adheres to fault constraints. This embodiment, through the mechanism of independent reconstruction and sequential superposition of individual faults, accurately characterizes the three-dimensional morphology of each fault in a complex fault system and their inter-fracture relationships, solving the geological rationality problem of model construction under multi-fault conditions and providing a reliable foundation for geological modeling of complex tectonic zones.
[0051] In some embodiments, before inputting the three-dimensional spatial coordinates of the region to be modeled into a pre-trained second deep learning model, the method further includes: marking the spatial coordinate points identified as fracture bodies in the fracture reconstruction results as having no strata or a specific mask value in the stratigraphic category label.
[0052] Specifically, the fracture reconstruction result can be the three-dimensional spatial distribution data of the fracture body predicted by the LSTM model; the no-stratum / specific mask value can be a special marker used to identify the fracture body area, which is excluded from the normal stratigraphic category in stratigraphic prediction; the stratigraphic category label can be a classification label system that identifies different lithological stratigraphic types.
[0053] Before inputting the 3D spatial coordinates of the region to be modeled into the second deep learning model, fracture constraint processing is required: the spatial coordinate points identified as fractures in the fracture reconstruction results are marked as no stratum (value 0) or a specific mask value in the stratigraphic category label, forming a constrained coordinate dataset: ;in, The coordinates of the constrained points are N, where N is the total number of data points and N is the coordinates of the fractured region. The value is forcibly set to 0; simultaneously, a stratigraphic category label is prepared for each coordinate point. (The values can be adjusted according to the actual situation), representing 10 different stratigraphic types.
[0054] In step S103 of some embodiments, the spatial constraint condition is to limit the spatial relationship of stratigraphic distribution based on the fracture reconstruction results; the second deep learning model can be a stratigraphic prediction network based on the Transformer architecture. Using the fracture reconstruction results as spatial constraints, the coordinate points of the fracture area are marked as stratigraphic mask values. The normalized three-dimensional spatial coordinates are input into the pre-trained Transformer model, which captures the global spatial relationship between coordinates through a multi-head self-attention mechanism and outputs the probability distribution of each coordinate point belonging to each stratigraphic category. The final stratigraphic category is selected using the maximum probability principle, generating a complete three-dimensional geological model that maintains geological continuity at the fracture boundaries.
[0055] In some embodiments, the second deep learning model is a Transformer model, the construction and training of which include: the model input is normalized three-dimensional spatial coordinates; the encoder part contains multiple cascaded Transformer blocks, each Transformer block contains a multi-head self-attention layer and a feedforward neural network layer; the output layer uses a softmax activation function to output the probability distribution of all preset stratum categories; the model training uses classification cross-entropy as the loss function and a learning rate decay strategy.
[0056] Specifically, normalization can be a process of linearly transforming three-dimensional coordinates to a uniform scale; a Transformer block can be a basic computational unit composed of self-attention layers and feedforward networks; multi-head self-attention can be multiple self-attention mechanisms running in parallel, which can capture features of different subspaces; the Softmax function can be an activation function that normalizes the output value to a probability distribution; classification cross-entropy can be a loss function for multi-classification problems, measuring the difference between the predicted probability distribution and the true distribution; and learning rate decay can be an optimization strategy that gradually reduces the learning rate as the training process progresses.
[0057] The Transformer model structure is designed as follows: Input layer: receives normalized three-dimensional spatial coordinates, with an input dimension of 3.
[0058] The encoder section consists of four cascaded Transformer blocks, each composed of a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism captures the positional relationships within the input sequence using h attention heads; its calculation formula is as follows: Where Q, K, and V are the query, key, and value matrices, respectively. is the dimension of the key vector; the output of each Transformer block is optimized through residual connections and layer normalization to enhance the training stability and feature representation capability of the model.
[0059] Output layer: Contains 10 units (corresponding to 10 stratigraphic types, which can be adjusted as needed), uses the softmax activation function, and outputs the probability distribution of all preset stratigraphic types.
[0060] The Transformer model training parameters are as follows: Loss function: The classification cross-entropy is used as the loss function to measure the difference between the predicted probability distribution and the true stratum category.
[0061] Training rounds: 150 rounds of iterative training are performed using the training set as input, with 32 samples processed in each batch.
[0062] Learning rate strategy: Reduce the learning rate by 10% every 50 rounds to gradually improve the model's convergence accuracy.
[0063] Early stopping mechanism: When the validation set loss shows no improvement for 15 consecutive rounds, training is automatically terminated to prevent overfitting.
[0064] The constrained coordinates of the region to be filled are input into the trained Transformer model to obtain the probability distribution of various stratigraphic categories for each coordinate point. The fracture reconstruction results (a set of 3D fracture spatial coordinates) output by the LSTM model are not merely used as constraints, but are transformed into input constraints for the Transformer model through masking. That is, the fracture coordinate points are first marked with no stratigraphic mask values, and then the 3D coordinate dataset containing this mask information is input into the Transformer model, enabling the model to automatically avoid fracture areas and focus on predicting the spatial distribution of normal strata during prediction. Furthermore, using the maximum probability principle, the stratigraphic category with the highest probability is selected as the final predicted stratum for that coordinate point, generating a complete 3D geological model. This model ensures the continuity of strata at the fracture boundary, accurately represents the spatial displacement of strata caused by fault displacement, and upgrades the stratigraphic model from a simple geometric distribution to a continuum model with tectonic constraints, significantly improving the geological rationality and prediction accuracy of the 3D geological model.
[0065] In some embodiments, the stratigraphic category to which each coordinate point belongs is output to generate a complete three-dimensional geological model, including: for each input coordinate, selecting the category with the highest probability in the stratigraphic category probability distribution output by the second deep learning model as the final predicted stratigraphic category of that coordinate point, and generating a complete three-dimensional geological model.
[0066] Specifically, the probability distribution can be the confidence vector of each candidate stratigraphic category output by the model; the maximum probability principle can be the decision criterion for selecting the category corresponding to the highest probability value as the prediction result.
[0067] Using the fracture reconstruction results generated in the aforementioned steps as spatial constraints, the fracture region is marked with a special mask value when inputting coordinates. The coordinate dataset containing the mask is input into the trained Transformer model to obtain the probability distribution of each coordinate point for each candidate stratigraphic category. Based on the principle of maximum probability, the stratigraphic category with the highest probability is selected as the final stratigraphic type for that coordinate point. After completing the prediction point by point, the data is integrated to generate a complete 3D geological model. This embodiment ensures the geological rationality of the stratigraphic prediction at the fracture boundary through fracture spatial constraints and probabilistic decision-making mechanisms, effectively restoring the stratigraphic displacement relationship caused by fault displacement, and generating a 3D model that conforms to geological laws. The final generated filled stratigraphic distribution ensures the continuity of the stratigraphy at the fracture boundary and accurately expresses the spatial displacement of the stratigraphy caused by fault displacement. This result not only includes the spatial distribution characteristics of the stratigraphy but also accurately reflects the disturbance of the fracture structure on the spatial relationship of the stratigraphy, upgrading the stratigraphic model from a simple geometric distribution to a continuum model with structural constraints, significantly improving the geological rationality and prediction accuracy of the 3D geological model.
[0068] This application achieves a complete representation of the fracture body from a two-dimensional plane to a three-dimensional entity through a three-stage processing method of serialization, fracture reconstruction, and stratigraphic filling, solving the problem of stratigraphic discontinuity at fracture boundaries in traditional methods. By using LSTM to learn the spatial distribution of fractures and Transformer to model global stratigraphic relationships, the accuracy and geological rationality of three-dimensional geological models under complex fracture conditions are significantly improved.
[0069] Referring to Figure 4, this application embodiment also provides a three-dimensional geological modeling device under complex fault structures, which can realize the above-mentioned three-dimensional geological modeling method under complex fault structures. The device includes: a processing module 401, used to acquire three-dimensional geological data containing fault zone identifiers and stratigraphic category labels; serialize the fault points in the three-dimensional geological data according to their geological strikes to generate a fault point coordinate sequence; an input module 402, used to input the fault point coordinate sequence into a pre-trained first deep learning model, output the probability that each spatial coordinate point belongs to the fault body, and generate a fault body reconstruction result according to a preset probability threshold; and a generation module 403, used to input the three-dimensional spatial coordinates of the area to be modeled into a pre-trained second deep learning model with the fault body reconstruction result as a spatial constraint, output the stratigraphic category to which each coordinate point belongs, and generate a complete three-dimensional geological model.
[0070] The specific implementation of the three-dimensional geological modeling device under complex fault structures is basically the same as the specific implementation of the three-dimensional geological modeling method under complex fault structures described above, and will not be repeated here.
[0071] Thirdly, this application provides an electronic device, as shown in Figure 5, which is a schematic diagram of the structure of an electronic device provided in this application.
[0072] As shown in Figure 5, the device includes: a memory 31 for storing computer programs; and a processor 32 for executing the computer programs; wherein, when the processor 32 executes the computer programs, it implements a three-dimensional geological modeling method for complex fracture structures as described in any of the above embodiments.
[0073] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 31 and executed by processor 32 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in an electronic device.
[0074] The processor 32 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.
[0075] The memory 31 can be used to store computer programs and / or modules. The processor 32 implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory 31 and calling the data stored in the memory 31. The memory 31 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 31 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0076] It should be noted that the above-mentioned electronic device includes processors and memory, but is not limited to processors and memory. Those skilled in the art will understand that the structural schematic diagram in Figure 5 is merely an example of the above-mentioned electronic device and does not constitute a limitation on the electronic device. It may include more components than shown in the figure, or combine certain components, or different components.
[0077] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed, implements the three-dimensional geological modeling method for complex fracture structures described in any of the above embodiments.
[0078] It should be understood that the implementation of all or part of the process in the above-described three-dimensional geological modeling method for complex fault structures can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the above-described three-dimensional geological modeling method for complex fault structures. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the relevant jurisdiction. For example, in some relevant jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.
[0079] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0080] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.
Claims
1. A three-dimensional geological modeling method for complex fault structures, characterized in that, include: Acquire 3D geological data including fault zone identifiers and stratigraphic category labels; The fault points in the three-dimensional geological data are serialized according to their geological strike to generate a fault point coordinate sequence. The fault point coordinate sequence is input into a pre-trained first deep learning model, which outputs the probability that each spatial coordinate point belongs to the fault body and generates a fault body reconstruction result according to a preset probability threshold. Using the fault body reconstruction result as a spatial constraint, the three-dimensional spatial coordinates of the area to be modeled are input into a pre-trained second deep learning model, which outputs the stratigraphic category to which each coordinate point belongs, and generates a complete three-dimensional geological model.
2. The three-dimensional geological modeling method under complex fault structures as described in claim 1, characterized in that, The step of serializing the fracture points in the three-dimensional geological data according to their geological orientation to generate a fracture point coordinate sequence includes: combining the three-dimensional spatial coordinates (x, y, z) of the fracture points and their associated physical field feature data into a 3+N dimensional feature vector, and arranging them into a time series according to the fracture orientation; this time series is the fracture point coordinate sequence containing geometric features and physical field features; where N is the number of physical field features, and the physical field features include one or more of velocity, density, resistivity or seismic wave attenuation coefficient.
3. The three-dimensional geological modeling method under complex fault structures as described in claim 1, characterized in that, The first deep learning model is a long short-term memory network model, and its construction and training include: the model input dimension is 3+N, N≥0; the hidden layer contains multiple LSTM units; the output layer uses the sigmoid activation function to output the probability of the breakpoint; the model training uses binary cross-entropy as the loss function and adopts an early stopping mechanism to prevent overfitting.
4. The three-dimensional geological modeling method under complex fault structures as described in claim 1, characterized in that, When there are multiple fractures in the modeling area, the fracture reconstruction process is performed separately for each fracture. The reconstructed fractures are then superimposed and fused in the same simulation grid according to the order set by geological laws to generate the fracture reconstruction result containing multiple fractures.
5. The three-dimensional geological modeling method under complex fault structures as described in claim 1, characterized in that, Before inputting the three-dimensional spatial coordinates of the area to be modeled into the pre-trained second deep learning model, the method further includes: marking the spatial coordinate points identified as fracture bodies in the fracture reconstruction results as having no strata or a specific mask value in the stratigraphic category label.
6. The three-dimensional geological modeling method under complex fault structures as described in claim 1, characterized in that, The second deep learning model is a Transformer model, and its construction and training include: the model input is normalized three-dimensional spatial coordinates; the encoder part contains multiple cascaded Transformer blocks, each Transformer block contains a multi-head self-attention layer and a feedforward neural network layer; the output layer uses the softmax activation function to output the probability distribution of all preset stratum categories; the model training uses classification cross-entropy as the loss function and adopts a learning rate decay strategy.
7. The three-dimensional geological modeling method for complex fault structures as described in claim 1, characterized in that, The step of outputting the stratigraphic category of each coordinate point to generate a complete three-dimensional geological model includes: for each input coordinate, selecting the category with the highest probability in the stratigraphic category probability distribution output by the second deep learning model as the final predicted stratigraphic category of that coordinate point, and generating the complete three-dimensional geological model.
8. A three-dimensional geological modeling device for complex fault structures, characterized in that, include: The processing module is used to acquire three-dimensional geological data containing fault zone identifiers and stratigraphic category labels; The fault points in the three-dimensional geological data are serialized according to their geological strike to generate a fault point coordinate sequence. The input module is used to input the fault point coordinate sequence into a pre-trained first deep learning model, output the probability that each spatial coordinate point belongs to the fault body, and generate the fault body reconstruction result according to a preset probability threshold. The generation module is used to input the three-dimensional spatial coordinates of the area to be modeled into a pre-trained second deep learning model with the fault body reconstruction result as a spatial constraint, output the stratigraphic category of each coordinate point, and generate a complete three-dimensional geological model.
9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the three-dimensional geological modeling method for complex fault structures as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the three-dimensional geological modeling method under complex fault structures as described in any one of claims 1 to 7.