Method for predicting the morphology of hydraulic fracturing fractures

Abstract

This invention discloses a method for predicting the morphology of hydraulic fracturing fractures, belonging to the field of oil and gas extraction and unconventional oil and gas reservoir development technology. The method includes: acquiring target geological parameters and target engineering parameters within the target work area; inputting the target geological parameters and target engineering parameters into a trained first target prediction model to obtain target spatial parameters of the hydraulic fracturing fractures; inputting the target geological parameters, target engineering parameters, and target spatial parameters into a trained second target prediction model to obtain target morphological parameters of the hydraulic fracturing fractures; and performing parametric geometric calculations on the target spatial parameters, target morphological parameters, and the predicted number of fracture clusters N to obtain a three-dimensional spatial point cloud of the hydraulic fracturing fractures. This method solves the technical problems of low computational efficiency and insufficient prediction accuracy in related technologies when predicting the three-dimensional propagation morphology of hydraulic fracturing fractures.

Description

Technical Field

[0001] This invention relates to the field of oil and gas extraction and unconventional oil and gas reservoir development technology, and in particular to a method for predicting the morphology of hydraulic fracturing fractures. Background Technology

[0002] In recent years, as domestic oil and gas development has expanded into low-permeability, deep, and unconventional resource areas, continental shale oil, due to its abundant reserves, has become the main battleground for exploration and development. However, its unique geological conditions—high clay content and well-developed bedding structure—endow the reservoirs with extremely strong heterogeneity and anisotropy, resulting in highly complex and unpredictable hydraulic fracturing fracture propagation mechanisms. Therefore, achieving accurate evaluation and control of fracture morphology is a core technological prerequisite for the successful development of such resources.

[0003] To address this challenge, current approaches primarily rely on two types of technologies. One type is numerical simulation methods based on physical laws, namely mechanistic models. These models have evolved from early simplified two-dimensional forms to quasi-three-dimensional and full three-dimensional models capable of simulating fluid-rock interactions, further forming unconventional fracturing simulation technologies integrating geology and engineering. Their core is predicting fracture propagation by solving fluid-structure interaction equations. The other type is artificial intelligence methods, represented by machine learning and deep learning. These methods have been widely used in reservoir evaluation and are beginning to be explored for tasks such as fracturing process optimization, pressure prediction, and event recognition. Their advantages lie in their powerful data processing and efficient nonlinear fitting capabilities. However, both approaches have significant shortcomings: while mechanistic models have a solid physical foundation, their accuracy is often limited when dealing with the extremely complex heterogeneous structures of continental shale, and their massive computational load leads to inefficiency, failing to meet the needs of real-time optimization in the field; while artificial intelligence methods are efficient and flexible, their purely data-driven nature lacks physical mechanism constraints, and under real-world conditions of scarce data and high noise, the reliability and generalization ability of the models face severe challenges.

[0004] Therefore, when using the above method to predict the morphology of hydraulic fracturing fractures, there are technical problems such as low computational efficiency and insufficient prediction accuracy. Summary of the Invention

[0005] One of the technical problems to be solved by this invention is to address the issues of low computational efficiency and insufficient prediction accuracy in related technologies when predicting the three-dimensional propagation morphology of hydraulic fracturing fractures.

[0006] To address the aforementioned technical problems, in a first aspect, embodiments of the present invention provide a method for predicting the morphology of hydraulic fracturing fractures. The method includes: acquiring target geological parameters and target engineering parameters within a target work area; inputting the target geological parameters and the target engineering parameters into a trained first target prediction model to obtain target spatial parameters of the hydraulic fracturing fractures; inputting the target geological parameters, the target engineering parameters, and the target spatial parameters into a trained second target prediction model to obtain target morphological parameters of the hydraulic fracturing fractures, wherein the second target prediction model includes a classification branch and a regression branch, the classification branch being used to output the predicted number of fracture clusters N, the regression branch being used to output a sequence of morphological parameters, the length of the morphological parameter sequence corresponding to a preset maximum number of fracture clusters, and the target morphological parameters being determined from the morphological parameter sequence based on the predicted number of fracture clusters N; and performing parametric geometric calculations on the target spatial parameters, the target morphological parameters, and the predicted number of fracture clusters N to obtain a three-dimensional spatial point cloud of the hydraulic fracturing fractures.

[0007] Optionally, the first target prediction model and the second target prediction model are obtained through the following methods: acquiring multiple sets of fracture sample data, wherein each set of fracture sample data includes an input feature set and a true label set, wherein the input feature set includes sample geological parameters and sample engineering parameters, and the true label set includes true spatial parameters, true morphological parameters, and the number of true fracture clusters; based on the multiple sets of fracture sample data, constructing a first sample dataset and a second sample dataset, wherein the first sample dataset includes the input feature set and the true spatial parameters, and the second sample dataset includes the input feature set, the true spatial parameters, the true morphological parameters, and the number of true fracture clusters; training a first initial prediction model based on the first sample dataset to obtain the first target prediction model; and training a second initial prediction model based on the second sample dataset to obtain the second target prediction model.

[0008] Optionally, acquiring multiple sets of crack sample data includes: using experimental design methods to generate multiple different combinations of sample geological parameters and sample engineering parameters; performing crack propagation simulation on each parameter combination to generate a corresponding sample three-dimensional spatial point cloud; extracting features from the sample three-dimensional spatial point cloud to obtain the corresponding real spatial parameters and real morphological parameters; and constructing the multiple sets of crack sample data based on each parameter combination and the corresponding real spatial parameters and real morphological parameters.

[0009] Optionally, obtaining the corresponding true spatial parameters and true morphological parameters includes: segmenting the sample three-dimensional spatial point cloud to obtain multiple crack cluster point cloud subsets; calculating the true spatial parameters based on the multiple crack cluster point cloud subsets; and performing geometric feature calculations on each crack cluster point cloud subset to obtain the true morphological parameters of each crack cluster.

[0010] Optionally, obtaining the first target prediction model includes: inputting the sample geological parameters and sample engineering parameters from the first sample dataset into the first initial prediction model and outputting the corresponding prediction spatial parameters; calculating the mean square error between the prediction spatial parameters and the true spatial parameters obtained from the first sample dataset; iteratively adjusting the parameters of the first initial prediction model with the goal of minimizing the mean square error; and obtaining the trained model as the first target prediction model when a preset training termination condition is met.

[0011] Optionally, obtaining the second target prediction model includes the following steps, and this training is performed after the first target prediction model has been trained: inputting the sample geological parameters, sample engineering parameters, and real spatial parameters from the second sample dataset into the second initial prediction model, and outputting the corresponding regression prediction vector and classification prediction vector; calculating a regression loss term based on the regression prediction vector and the real morphological parameters after padding and masking, and calculating a classification loss term based on the classification prediction vector and the number of real crack clusters in the second sample dataset; performing a weighted summation of the regression loss term and the classification loss term to obtain a weighted combined loss function; iteratively adjusting the parameters of the second initial prediction model with the goal of minimizing the weighted combined loss function; and obtaining the trained model as the second target prediction model when a preset training termination condition is met.

[0012] Optionally, the first target prediction model predicts the target spatial parameters in the following manner: the input vector composed of the target geological parameters and the target engineering parameters is sequentially passed through multiple cascaded one-dimensional convolutional modules to extract features and obtain high-dimensional features, wherein each one-dimensional convolutional module sequentially performs one-dimensional convolution, batch normalization and nonlinear activation operations; regression analysis is performed on the high-dimensional features to obtain the target spatial parameters.

[0013] Optionally, the second target prediction model predicts the target morphological parameters in the following manner: the target geological parameters, the target engineering parameters, and the target spatial parameters are shared and encoded to obtain a shared feature vector; the shared feature vector is input into the regression branch to generate a morphological parameter prediction vector as a sequence of morphological parameters, and the shared feature vector is input into the classification branch to generate a probability vector for the number of fracture clusters; based on the probability vector for the number of fracture clusters, the predicted number of fracture clusters N is determined; according to the predicted number of fracture clusters N, the first N sets of parameter values ​​are extracted from the morphological parameter prediction vector and reconstructed into an N-row parameter matrix as the target morphological parameters, wherein each row of the N-row parameter matrix corresponds to a set of fracture cluster morphological parameters.

[0014] Optionally, obtaining the three-dimensional spatial point cloud of the hydraulic fracturing fracture includes: based on the predicted number of fracture clusters N and the spatial distribution information of each fracture cluster extracted from the target spatial parameters; extracting N sets of fracture cluster morphological parameters from the target morphological parameters according to the predicted number of fracture clusters N, wherein each set of fracture cluster morphological parameters includes at least fracture length, fracture height, dip angle, average fracture width, and sparsity; determining the center positions of N fracture clusters corresponding to the predicted number of fracture clusters N in three-dimensional space according to the spatial distribution information; generating N subsets of three-dimensional point clouds of fracture clusters based on the center positions of the N fracture clusters and the corresponding N sets of fracture cluster morphological parameters; and merging the N subsets of three-dimensional point clouds of fracture clusters to form the three-dimensional point cloud of the hydraulic fracturing fracture.

[0015] Optionally, generating N subsets of three-dimensional point clouds of crack clusters includes performing the following steps on each crack cluster: constructing a reference crack surface at the center of the crack cluster based on the crack length, crack height, and inclination angle in a set of crack cluster morphological parameters corresponding to the center position of the crack cluster; determining the scatter distribution density on the reference crack surface based on the average crack width and sparsity in the set of crack cluster morphological parameters; and generating a three-dimensional scatter set on the reference crack surface according to the scatter distribution density to obtain the subset of three-dimensional point clouds of crack clusters corresponding to the crack cluster.

[0016] Secondly, embodiments of the present invention also provide a device for predicting the morphology of hydraulic fracturing fractures. The device includes: an acquisition module for acquiring target geological parameters and target engineering parameters within a target work area; a first prediction module for inputting the target geological parameters and the target engineering parameters into a trained first target prediction model to obtain target spatial parameters of the hydraulic fracturing fractures; a second prediction module for inputting the target geological parameters, the target engineering parameters, and the target spatial parameters into a trained second target prediction model to obtain target morphological parameters of the hydraulic fracturing fractures, wherein the second target prediction model includes a classification branch and a regression branch, the classification branch outputting a predicted number of fracture clusters N, the regression branch outputting a sequence of morphological parameters, the length of which corresponds to a preset maximum number of fracture clusters, and the target morphological parameters are determined from the morphological parameter sequence based on the predicted number of fracture clusters N; and a calculation module for performing parametric geometric calculations on the target spatial parameters, the target morphological parameters, and the predicted number of fracture clusters N to obtain a three-dimensional spatial point cloud of the hydraulic fracturing fractures.

[0017] Thirdly, the present invention also provides a machine-readable storage medium storing instructions that cause a machine to perform the method described in any of the preceding claims.

[0018] Fourthly, the present invention also provides a processor for running a program, wherein the program, when run, is used to perform the method described in any of the preceding claims.

[0019] Through the above technical solution, this invention constructs an efficient fracturing design method based on a two-level cascaded neural network surrogate model, including: acquiring target geological parameters and target engineering parameters within the target work area; predicting macroscopic spatial parameters of fractures through a first target prediction model, and then predicting the geometric morphology of each fracture cluster through a second target prediction model containing both regression and classification branches; finally, generating a three-dimensional fracture point cloud based on parametric geometric reconstruction of the prediction results. This method deeply integrates physical simulation mechanisms with data-driven models. Through a two-level collaborative prediction and adaptive reconstruction mechanism, it achieves second-level efficient mapping from geological engineering parameters to three-dimensional fracture morphology while ensuring the physical rationality of the prediction results. This effectively solves the technical problems of low computational efficiency and insufficient prediction accuracy in predicting the three-dimensional propagation morphology of hydraulic fracturing fractures in related technologies.

[0020] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for predicting the morphology of hydraulic fracturing fractures provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the three-dimensional discrete lattice mechanical model provided in an embodiment of the present invention; Figure 3 This is a simulation result of the hydraulic fracturing fracture morphology based on the three-dimensional discrete lattice method provided in an embodiment of the present invention; Figure 4 This is a single-cluster crack morphology diagram based on the three-dimensional discrete lattice method simulation provided in an embodiment of the present invention; Figure 5 This is a single-cluster crack morphology diagram predicted by the neural network surrogate model provided in this embodiment of the invention; Figure 6 This is a crack prediction accuracy diagram of the neural network surrogate model provided in this embodiment of the invention; Figure 7 This is a flowchart of the hydraulic fracturing fracture morphology prediction device provided in an embodiment of the present invention. Detailed Implementation

[0022] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustrating and explaining the embodiments of the present invention, and... This is not intended to limit the embodiments of the present invention.

[0023] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0024] Example This invention provides a method for predicting the morphology of hydraulic fracturing fractures. Figure 1 This is a flowchart illustrating a method for predicting the morphology of hydraulic fracturing fractures according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S102 to S108: Step S102: Obtain the target geological parameters and target engineering parameters within the target work area; Specifically, the aforementioned target work area refers to a specific oil and gas region where hydraulic fracturing operations are planned, typically divided into well groups, blocks, or reservoir units. The target geological parameters are a series of quantitative parameters used to describe the rock mechanical properties, geostress state, and physical characteristics of the target reservoir, including but not limited to: reservoir depth, maximum and minimum horizontal principal stress, vertical stress, Poisson's ratio, elastic modulus, tensile strength, compressive strength, fracture toughness, rock density, permeability, porosity, and reservoir thickness. The target engineering parameters are a series of quantitative parameters used to describe the fracturing operation design and operating conditions, including but not limited to: fracturing fluid viscosity, fracturing fluid density, and injection rate of each perforation cluster.

[0025] Step S104: Input the target geological parameters and target engineering parameters into the trained first target prediction model to obtain the target spatial parameters of the hydraulic fracturing fracture; Specifically, the input vector, composed of standardized target geological parameters and target engineering parameters obtained from the target work area, is input into the trained first target prediction model for prediction, thereby obtaining the target spatial parameters of hydraulic fracturing fractures. The first target prediction model is specifically designed to predict the target spatial parameters of hydraulic fracturing fractures, and these target spatial parameters are features output by the model used to macroscopically describe the fracturing volume (SRV), including at least: the total number of fracture clusters (characterizing fracture complexity), and the average spacing and standard deviation between the centroids of fracture clusters (characterizing the uniformity of fracture spacing).

[0026] Step S106: Input the target geological parameters, target engineering parameters, and target spatial parameters into the trained second target prediction model to obtain the target morphological parameters of the hydraulic fracturing fractures. The second target prediction model includes a classification branch and a regression branch. The classification branch is used to output the predicted number of fracture clusters N, and the regression branch is used to output the morphological parameter sequence. The length of the morphological parameter sequence corresponds to the preset maximum number of fracture clusters. The target morphological parameters are determined from the morphological parameter sequence based on the predicted number of fracture clusters N. Specifically, based on the target geological parameters, target engineering parameters, and target spatial parameters output by the first target prediction model, these parameters are input into a trained second target prediction model for prediction, resulting in target morphological parameters of the hydraulic fracturing fractures. The second target prediction model includes regression and classification branches. Under the constraint of the target spatial parameters output by the first target prediction model, which characterize the macroscopic layout of the fractures, it collaboratively predicts the detailed geometric properties of the fracture clusters to obtain the target morphological parameters. These target morphological parameters include at least: fracture length (maximum length along the main extension direction), fracture height (extended range in the vertical direction), average fracture width, dominant dip angle, average hydraulic aperture, and sparsity.

[0027] Furthermore, in one embodiment, the first target prediction model and the second target prediction model in steps S104 and S106 above are obtained through the following steps S1-S4: Step S1: Obtain multiple sets of crack sample data. Each set of crack sample data includes an input feature set and a real label set. The input feature set includes sample geological parameters and sample engineering parameters. The real label set includes real spatial parameters, real morphological parameters, and the number of real crack clusters. In one embodiment, obtaining multiple sets of crack sample data in step S1 includes steps S11-S14: Step S11: Using experimental design methods, generate multiple sets of different combinations of sample geological parameters and sample engineering parameters; Step S12: Simulate crack propagation for each set of parameters to generate the corresponding sample three-dimensional point cloud; Step S13: Extract features from the three-dimensional point cloud of the sample to obtain the corresponding real spatial parameters and real morphological parameters; In one embodiment, the true spatial parameters and true morphological parameters in step S13 can be obtained through the following steps: segmenting the sample three-dimensional spatial point cloud to obtain multiple crack cluster point cloud subsets; calculating the true spatial parameters based on the multiple crack cluster point cloud subsets; and calculating the geometric features of each crack cluster point cloud subset to obtain the true morphological parameters of each crack cluster.

[0028] Step S14: Based on each set of parameter combinations and the corresponding real spatial parameters and real morphological parameters, construct multiple sets of crack sample data.

[0029] Step S2: Based on multiple sets of crack sample data, construct a first sample dataset and a second sample dataset. The first sample dataset contains input feature groups and real spatial parameters, and the second sample dataset contains input feature groups, real spatial parameters, real morphological parameters, and the number of real crack clusters. Specifically, in steps S1-S2 and their included steps S11-S14, Latin hypercube sampling is first used to scientifically sample within the geological engineering parameter space, generating multiple sets of input parameter combinations. Then, a three-dimensional discrete lattice method is run for a full physical simulation of each set of input parameter combinations. Among these, a geomechanical model considering the heterogeneity of the formation is established using a specific method, such as... Figure 2 As shown in Table 1, the geomechanical parameters and fracturing construction engineering parameters on which the simulation was based are also presented. Table 1 Key parameter settings for crack propagation mechanism model

[0030] Furthermore, three-dimensional discrete lattice simulation software was used to simulate fracture propagation. First, based on the reservoir thickness and the expected fracture propagation range, a three-dimensional computational domain of 200 m (length) × 200 m (width) × 60 m (height) was established. This domain was then discretized into cubic lattice elements with a side length of 1.0 m, generating a total of over 240,000 computational elements to ensure a balance between simulation accuracy and computational efficiency. The complex three-dimensional fracture morphology simulated through the above steps is shown below. Figure 3 As shown, the propagation path and fracture network structure of fractures in heterogeneous reservoirs are intuitively demonstrated.

[0031] It should be noted that, based on the above application scenario, the simulation process strictly follows the physical mechanism: during model initialization, each grid cell is assigned the rock mechanical properties listed in the table above (elastic modulus, Poisson's ratio, tensile strength, etc.), and an initial geostress field is set (σ_H=50 MPa, σ_h=45 MPa, σ_v=55 MPa). Fracturing fluid is injected through three perforation clusters (20 meters apart) located in the center of the model and on the simulated wellbore, following given engineering parameters (viscosity 5 mPa·s, single cluster discharge rate 2.29 m³ / min). During the simulation, the system dynamically couples and solves the nonlinear flow equations of the fluid in the pore-fracture network, the elastic-brittle mechanical interactions between grid cells, and the fracture propagation criterion based on the maximum circumferential stress criterion. When the combined effect of the fluid pressure and geostress on a cell causes its tensile stress to exceed its tensile strength (4.3 MPa), the cell undergoes tensile fracture, forming a new fracture surface.

[0032] In step S12, the generated point cloud is automatically parsed, and density-based (e.g., DBSCAN) or distance-based point cloud clustering algorithms are used to segment each fracture cluster, thereby extracting the true spatial parameters, including the total number of fracture clusters, the average spacing, and the standard deviation. Simultaneously, the true morphological parameters of each fracture cluster are extracted, including geometric attributes such as fracture length, fracture height, fracture width, dip angle, hydraulic aperture, and sparsity. The actual number of fracture clusters is recorded as the true classification label. For the fracturing fracture system, the overall spatial parameters (target spatial parameters, sample spatial parameters, and true spatial parameters) describe the macroscopic distribution of the fracturing fractures, including: the total number of fracture clusters, directly obtained from the clustering results, used to characterize the complexity of the fracture network; the average spacing and standard deviation of fracture clusters, obtained by calculating the Euclidean distance between the geometric centers (centroids) of each fracture cluster, used to characterize the spatial distribution density and uniformity of the fractures; and the individual morphological parameters (target morphological parameters, sample morphological parameters, and true morphological parameters) are used to accurately describe the geometric attributes of each fracture cluster.

[0033] Furthermore, the structure is organized as a two-dimensional array, where the length of the first dimension equals the number of fracture clusters, and the second dimension contains six key morphological parameters describing a single fracture cluster: fracture length, calculated along the main extension direction of the fracture through principal component analysis; fracture height, the vertical extension range of the fracture cluster; average fracture width, characterizing the aperture size of the fracture; dominant dip angle, describing the tilt angle of the fracture surface in space; average hydraulic aperture, reflecting the conductivity potential of the fracture; and sparsity, defined by calculating the ratio of the bounding box volume occupied by the fracture cluster point cloud to the effective volume of the point cloud itself, used to quantify the "abundance" or "sparseness" of the internal structure of the fracture, which is directly related to the conductivity of the fracture.

[0034] Based on this, assuming that 1200 different combinations of input parameters were generated using the method in step S2, step S13 standardizes all samples and randomly divides them into a training set (840 sets), a validation set (180 sets), and a test set (180 sets) according to a 7:1.5:1.5 ratio. This constructs a first sample dataset for training the first initial prediction model and a second sample dataset for training the second initial prediction model. The first sample dataset uses geological engineering parameters (target engineering parameters, target geological parameters) as input and target spatial parameters as output. The second dataset uses the same input as a composite output containing classification labels and morphological parameter matrices, and introduces a masking mechanism to handle the variable cluster number. Step S14 further implements data quality checks and enhancements, improving the completeness and representativeness of the dataset through anomaly removal, parameter sensitivity analysis, and sparse region supplementary sampling, laying a reliable data foundation for subsequent neural network training.

[0035] Step S3: Based on the first sample dataset, train the first initial prediction model to obtain the first target prediction model; In one embodiment, the first target prediction model in step S3 is specifically implemented through the following steps S31-S34: Step S31: Input the sample geological parameters and sample engineering parameters in the first sample dataset into the first initial prediction model, and output the corresponding prediction spatial parameters; Step S32: Calculate the mean square error between the predicted spatial parameters and the true spatial parameters obtained from the first sample dataset; Step S33: Iteratively adjust the parameters of the first initial prediction model with the goal of minimizing the mean square error; Step S34: When the preset training termination condition is met, the trained model is obtained as the first target prediction model.

[0036] In step S104, the target geological parameters and target engineering parameters are input into the first target prediction model trained by steps S31-S34 above to predict the target spatial parameters of the hydraulic fracturing fracture.

[0037] In one embodiment, the first target prediction model predicts the target spatial parameters in the following manner: the input vector composed of the target geological parameters and the target engineering parameters is sequentially passed through multiple cascaded one-dimensional convolutional modules to extract features and obtain high-dimensional features, wherein each one-dimensional convolutional module sequentially performs one-dimensional convolution, batch normalization and nonlinear activation operations; regression analysis is performed on the high-dimensional features to obtain the target spatial parameters.

[0038] Specifically, to achieve efficient prediction of the macroscopic spatial distribution of hydraulic fracturing fractures, a first-level neural network surrogate model (i.e., the first target prediction model) is first constructed and trained. The training of this model is completed in steps S31-S34 above, mainly based on the first sample dataset generated by simulation using the three-dimensional discrete lattice method and after parameterized analysis. The specific training process is as follows: The standardized sample geological parameters (including reservoir depth, maximum and minimum horizontal principal stress, vertical stress, Poisson's ratio, elastic modulus, tensile strength, compressive strength, fracture toughness, rock density, permeability, porosity, and reservoir thickness, etc.) and sample engineering parameters (including fracturing fluid viscosity, fracturing fluid density, and injection rate of each perforation cluster, etc.) in the first dataset are concatenated into an input vector and input to the first initial prediction model. This model adopts a one-dimensional convolutional module stacked structure, containing 3 to 5 convolutional layers, each followed by a batch normalization layer and a ReLU activation function. The model outputs the corresponding prediction spatial parameters, including the number of fracture clusters, average spacing, and standard deviation.

[0039] Furthermore, the mean squared error loss between the predicted spatial parameters output by the model and the real spatial parameters extracted from the point cloud simulated by the 3D discrete lattice method in the first sample dataset is calculated. Training aims to minimize the mean squared error, employing the Adam optimizer with a dynamic learning rate decay strategy. The convolutional kernel weights and fully connected layer parameters of the first initial prediction model are iteratively adjusted through the backpropagation algorithm. During this training process, the loss change is monitored using the independent validation set reserved in step S3. When the validation loss no longer decreases over several consecutive training epochs, an early stopping mechanism is triggered to prevent overfitting. After sufficient optimization, the model that has converged and has stable performance is finally saved as the first target prediction model. This model's coefficient of determination R² for the predicted spatial parameters on the test set is consistently above 0.9, meeting the accuracy requirements for replacing high-cost physical simulations. Moreover, the time for a single prediction is only in the millisecond range, improving efficiency by more than 100,000 times compared to physical simulations.

[0040] After the training process described above, the first target prediction model is applied for actual prediction in step S104: the target geological parameters and target engineering parameters obtained within the target work area are concatenated and standardized in the same order as during the training phase to form a standardized input vector. This vector is then input into the trained first target prediction model. Based on the learned high-dimensional nonlinear mapping relationship, the model performs forward propagation inference through its internal multi-layer one-dimensional convolution and fully connected structure, directly outputting the corresponding target spatial parameters for the working condition within seconds. These parameters accurately describe the macroscopic spatial distribution framework of the predicted fracture network, providing crucial input for the subsequent refined prediction of the second target prediction model and the final reconstruction of the three-dimensional fracture morphology. This method achieves a high-precision, second-level mapping from geological and engineering parameters to the macroscopic spatial distribution characteristics of fractures through a neural network surrogate model, significantly improving the iterative efficiency and intelligence level of fracturing scheme design and optimization.

[0041] Step S4: Based on the second sample dataset, train the second initial prediction model to obtain the second target prediction model.

[0042] In one embodiment, the specific steps for obtaining the second target prediction model after the first target prediction model has been trained include steps S41-S45: Step S41: Input the sample geological parameters, sample engineering parameters and real spatial parameters in the second sample dataset into the second initial prediction model, and output the corresponding regression prediction vector and classification prediction vector; Step S42: Calculate the regression loss term based on the regression prediction vector and the true morphological parameters after padding and masking, and calculate the classification loss term based on the classification prediction vector and the true number of crack clusters in the second sample dataset. Step S43: Perform a weighted summation of the regression loss term and the classification loss term to obtain the weighted combined loss function; Step S44: Iteratively adjust the parameters of the second initial prediction model with the goal of minimizing the weighted combination loss function; Step S45: When the preset training termination condition is met, the trained model is obtained as the second target prediction model.

[0043] In step S106, the target geological parameters, target engineering parameters, and target spatial parameters output by the first target prediction model are input into the second target prediction model trained by the above steps S41-S45 to predict the target morphological parameters of the hydraulic fracturing fracture.

[0044] In one embodiment, the second target prediction model predicts the target morphological parameters in the following manner: the target geological parameters, target engineering parameters, and target spatial parameters are shared and encoded to obtain a shared feature vector; the shared feature vector is input into the regression branch of the second target prediction model to generate a morphological parameter prediction vector, and the shared feature vector is input into the classification branch of the second target prediction model to generate a probability vector for the number of fracture clusters; based on the probability vector for the number of fracture clusters, the predicted number of fracture clusters N is determined; according to the predicted number of fracture clusters N, the first N sets of parameter values ​​of the morphological parameter prediction vector are reconstructed into an N-row parameter matrix as the target morphological parameters, wherein each row of the parameter matrix in the N-row parameter matrix corresponds to a set of fracture cluster morphological parameters.

[0045] Specifically, to achieve efficient and high-precision prediction of the individual geometric morphology of hydraulic fracturing fractures, this invention, after training the first target prediction model, further constructs and trains a second-level neural network surrogate model (i.e., the second target prediction model). The training of this model is based on a second sample dataset generated through simulation using the three-dimensional discrete lattice method and subjected to parameterized analysis. The specific training process is as follows: First, the standardized geological parameters, engineering parameters, and corresponding real spatial parameters of the samples in the dataset are input into the second initial prediction model. This model uses the second-level network in a two-level cascaded deep neural network architecture. It first extracts the basic high-order feature representations of the input parameters through a shared feature encoder composed of fully connected layers. Subsequently, the network structure branches into two parallel collaborative branches: the regression branch consists of several fully connected layers, outputting a flattened vector of morphological parameters of a fixed length (e.g., 60). This vector presets the maximum possible number of fracture clusters (N_max, e.g., 10) and sequentially stores all the predicted morphological parameters of these N_max "virtual" fracture clusters; the classification branch consists of a fully connected layer and a Softmax function, outputting an N_max (e.g., 10)-dimensional probability distribution vector of the number of fracture clusters, used to determine the probability that the actual number of fracture clusters is any value from 1 to N_max (e.g., 10). In the model inference process, the maximum probability predicted by the classification branch is taken as the number of crack clusters N, and the first N sets of morphological parameters are accurately extracted from the 60-dimensional output of the regression branch to reconstruct the final N×6-dimensional individual morphological parameter prediction matrix.

[0046] Furthermore, the losses between the outputs of the two branches and the true labels are calculated separately: The regression loss term (L_regression) is calculated based on the regression prediction vector output by the regression branch and the true morphological parameter matrix padded to N_max dimension and masked; the mean squared error loss is typically used. The classification loss term is calculated based on the classification prediction vector output by the classification branch and the number of true crack clusters in the second sample dataset (in terms of one...). (Hot encoding representation) Calculate the classification loss term (L_classification), typically using cross-entropy loss. Then, the two losses are weighted and combined to form a weighted combined loss function: L_total = α * L_regression + β * L_classification. A grid search on the validation set determines the hyperparameters balancing the weights of the two tasks, for example, α=1.0, β=0.5. Training aims to minimize this weighted combined loss function, employing the Adam optimizer and a dynamic learning rate strategy (such as cosine annealing), iteratively adjusting all parameters of the second initial prediction model through backpropagation. When training the regression branch, a mask corresponding to the number of true crack clusters is applied to the morphological parameter matrix labels to ensure the network learns only from valid crack cluster data. The training process is monitored using an independent validation set; an early stopping mechanism is triggered when the validation loss stops decreasing for a certain number of consecutive periods to prevent overfitting. After sufficient optimization, the finally trained second target prediction model is obtained. The model achieves an average coefficient of determination (R²) of over 0.91 for predicting individual morphological parameters on an independent test set, and its classification accuracy for the number of crack clusters reaches over 94%, meeting the requirements of a high-precision surrogate model.

[0047] After training, the model is further applied for actual prediction in step S106: the target geological parameters, target engineering parameters, and target spatial parameters output by the first target prediction model obtained within the target work area are standardized and feature-stitched using the same processing method as in the training phase, and then input into the trained second target prediction model. The model performs forward inference based on the learned bi-branch co-mapping relationship. Its classification branch outputs the predicted probability of the number of fracture clusters, thereby determining the predicted number of fracture clusters N; its regression branch simultaneously outputs a flattened prediction vector containing N_max sets of morphological parameters. Based on the determined N value, the first N sets of morphological parameters are extracted from this flattened vector and reconstructed into an N-row, M-column target morphological parameter matrix. This matrix accurately describes six key geometric attributes of each predicted fracture cluster: fracture length, fracture height, average fracture width, dominant dip angle, average hydraulic aperture, and sparsity, providing complete and accurate input for the subsequent parameterized reconstruction of the three-dimensional fracture morphology. This method, by applying a pre-trained dual-branch surrogate model, completes end-to-end high-precision prediction of the fine morphology of a variable number of fracture clusters within seconds, significantly supporting real-time intelligent optimization of fracturing schemes.

[0048] Step S108: Perform parametric geometric calculations on the target spatial parameters, target morphological parameters, and the predicted number of fracture clusters N to obtain the three-dimensional spatial point cloud of hydraulic fracturing fractures.

[0049] In one embodiment, obtaining a three-dimensional spatial point cloud of hydraulic fracturing fractures includes the following steps S51-S55: Step S51: Based on the predicted number of crack clusters N output by the classification branch of the second target prediction model, and the spatial distribution information of each crack cluster extracted from the target spatial parameters; Step S52: Based on the predicted number of crack clusters N, extract the corresponding N sets of crack cluster morphology parameters from the target morphology parameters. Each set of crack cluster morphology parameters includes at least crack length, crack height, dip angle, average crack width, and sparsity. Step S53: Based on the spatial distribution information, determine the center locations of N crack clusters in three-dimensional space that correspond to the predicted number N crack clusters; Step S54: Generate N 3D point cloud subsets of crack clusters based on the center positions of N crack clusters and the corresponding N sets of crack cluster morphology parameters; In one embodiment, step S54, generating N subsets of three-dimensional point clouds of crack clusters, includes performing the following steps on each crack cluster: constructing a reference crack surface at the center of the crack cluster based on the crack length, crack height, and inclination angle of a set of crack cluster morphological parameters corresponding to the center position of the crack cluster; determining the scatter distribution density on the reference crack surface based on the average crack width and sparsity of a set of crack cluster morphological parameters; and generating a three-dimensional scatter set on the reference crack surface according to the scatter distribution density to obtain the subset of three-dimensional point clouds of crack clusters corresponding to the crack clusters.

[0050] Step S55: Merge N subsets of the three-dimensional point cloud of fracture clusters to form a three-dimensional point cloud of hydraulic fracturing fractures.

[0051] Specifically, after obtaining the target spatial parameters (including the total number of crack clusters, average spacing and standard deviation) output by the first target prediction model and the target morphological parameter matrix (N rows × M columns, describing the detailed geometric properties of each crack cluster) output by the second target prediction model, the target spatial parameters are reverse reconstructed by a parametric geometric modeling algorithm to finally generate a visualized three-dimensional crack morphology point cloud.

[0052] The specific reconstruction process in the above steps includes the following steps: First, based on the total number N of crack clusters extracted from the target spatial parameters and the macroscopic distribution information (such as average spacing), combined with the N sets of detailed geometric parameters extracted from the target morphological parameter matrix (each set includes at least crack length, crack height, dominant dip angle, average crack width, and sparsity), the initial spatial framework of N crack clusters is determined in three-dimensional space. Second, based on the extracted N sets of morphological parameters, a parametric geometric modeling algorithm is used to reconstruct the point cloud of each crack cluster sequentially: based on the predicted crack length, crack height, dip angle, and other parameters of each crack cluster, an initial reference crack surface (e.g., an elliptical or rectangular plane) is generated at its corresponding spatial location. Next, based on the predicted average crack width and sparsity parameters of the crack cluster, three-dimensional scattered points of corresponding density are generated within the reference crack surface and its normal extension space according to a specific spatial distribution pattern (such as normal distribution or uniform distribution) to simulate the actual opening and internal structural changes of the cracks. Finally, based on the macroscopic spacing and distribution characteristics obtained from the target space parameters, the three-dimensional scattered point subsets generated by each crack cluster are reasonably arranged, oriented and merged in three-dimensional space, thereby generating a complete three-dimensional spatial point cloud model that reflects the predicted complex crack network morphology.

[0053] It should be noted that this reconstruction process achieves a closed-loop transformation from structured prediction parameters to intuitive three-dimensional geometric morphology. The three-dimensional crack morphology generated by the above reconstruction algorithm shows a high degree of geometric agreement with the simulation results of the original mechanism model based on the three-dimensional discrete lattice method. Figure 4 This invention provides a single-cluster crack morphology diagram based on a three-dimensional discrete lattice method simulation. Figure 5 This is a single-cluster crack morphology map predicted by the neural network surrogate model provided in an embodiment of the present invention, for specific comparison and combination. Figure 4 and Figure 5 The generated 3D point cloud can be directly used for visualization analysis, fracturing volume (SRV) calculation, and coupling with geological models for subsequent engineering evaluation, providing designers with intuitive and quantitative fracture propagation prediction results, and significantly improving the visualization level and decision-making efficiency of fracturing scheme design.

[0054] It should also be noted that the first and second target prediction models trained above, along with their pre- and post-processing workflows, are encapsulated into an integrated software system and embedded into the fracturing design and optimization platform to achieve an intelligent closed loop from parameter input to scheme optimization. Specifically, the system is deployed on a cloud server using a front-end and back-end separation architecture. The core back-end provides a model service with RESTful API (a standardized network service interface based on the HTTP protocol), encapsulating the complete process from parameter input and model prediction to 3D reconstruction. The front-end is an interactive graphical user interface (GUI), through which designers can intuitively input or adjust geological and engineering parameters. After a task is submitted, the system completes the entire calculation process within seconds (usually less than 1 second) and feeds back the 3D fracture model and key evaluation indicators (such as total fracture area and stimulation volume) to the front-end interface in real time, supporting rapid comparison and effect evaluation of multiple fracturing schemes.

[0055] In an optional implementation, to verify the generalization ability of the neural network surrogate model constructed in this invention under different geological conditions, it was applied to predict the parameter conditions of another work area (block B). Specifically, the model predicted all 28 clusters of cracks, and the prediction results of its neural network surrogate model (co-predicting by the first target prediction model and the second target prediction model) were quantitatively compared with the simulation results of the mechanism model based on the three-dimensional discrete lattice method. The mean absolute error (MAE) and root mean square error (RMSE) of the two models were as follows: Figure 6 As shown, comparative analysis reveals that the neural network surrogate model's predictions of formation fracture propagation agree well with the calculation results of the mechanistic model. Specifically, the MAE for all 28 predicted data points ranges from 4.52% to 8.38%, and the RMSE ranges from 6.37% to 10.35%. The fluctuation range of the two error curves across the entire data series (28 data points) remains between 1.5% and 4%, demonstrating high prediction accuracy and stability. The error distribution characteristics show that the RMSE value is generally slightly higher than the MAE, indicating a slightly larger prediction bias at some local data points, but the overall error does not exhibit abnormal fluctuations, proving the stability of the prediction system.

[0056] It should be noted that this implementation verifies that the neural network proxy model constructed in this invention can still achieve a good balance between computational efficiency (second-level prediction) and prediction accuracy (average error less than 10%) under different work area parameters. This model is suitable for lightweight application scenarios with high timeliness requirements, such as rapid iterative optimization of fracturing schemes and real-time effect prediction and monitoring.

[0057] Based on the above, embodiments of the present invention also provide a device for predicting the morphology of hydraulic fracturing fractures, such as... Figure 7As shown, the system includes: an acquisition module 707, used to acquire target geological parameters and target engineering parameters within the target work area; a first prediction module 704, connected to the acquisition module 707, used to input the target geological parameters and target engineering parameters into a trained first target prediction model to obtain target spatial parameters of the hydraulic fracturing fractures; a second prediction module 706, connected to the first prediction module 704, used to input the target geological parameters, target engineering parameters, and target spatial parameters into a trained second target prediction model to obtain target morphological parameters of the hydraulic fracturing fractures, wherein the second target prediction model includes a regression branch and a classification branch, the classification branch is used to output the predicted number of fracture clusters N, the regression branch is used to output a sequence of morphological parameters, the length of the morphological parameter sequence corresponds to a preset maximum number of fracture clusters, and the target morphological parameters are determined from the morphological parameter sequence based on the predicted number of fracture clusters N; and a calculation module 708, connected to the second prediction module 706, used to perform parametric geometric calculations on the target spatial parameters and target morphological parameters to obtain a three-dimensional spatial point cloud of the hydraulic fracturing fractures.

[0058] This invention also provides a processor for running a program, wherein the program is executed to perform: a hydraulic fracturing fracture morphology prediction method as described in any of the preceding embodiments.

[0059] The present invention also provides a computer program product, which, when executed on a data processing device, is adapted to perform the steps of the method for initializing the morphological prediction of hydraulic fracturing fractures as described in any of the above.

[0060] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0065] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0066] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0067] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0068] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method of predicting the morphology of a hydraulic fracture, characterized by, include: Obtain the target geological parameters and target engineering parameters within the target work area; The target geological parameters and the target engineering parameters are input into the trained first target prediction model to obtain the target spatial parameters of the hydraulic fracturing fracture. The target geological parameters, target engineering parameters, and target spatial parameters are input into a trained second target prediction model to obtain the target morphological parameters of the hydraulic fracturing fractures. The second target prediction model includes a classification branch and a regression branch. The classification branch is used to output the predicted number of fracture clusters N, and the regression branch is used to output a sequence of morphological parameters. The length of the morphological parameter sequence corresponds to a preset maximum number of fracture clusters. The target morphological parameters are determined from the morphological parameter sequence based on the predicted number of fracture clusters N. Parametric geometric calculations are performed on the target spatial parameters, the target morphological parameters, and the predicted number of fracture clusters N to obtain the three-dimensional spatial point cloud of the hydraulic fracturing fracture.

2. The method of claim 1, wherein, The first target prediction model and the second target prediction model are obtained in the following way: Multiple sets of crack sample data are acquired. Each set of crack sample data includes an input feature set and a real label set. The input feature set includes sample geological parameters and sample engineering parameters. The real label set includes real spatial parameters, real morphological parameters, and the number of real crack clusters. Based on the multiple sets of crack sample data, a first sample dataset and a second sample dataset are constructed. The first sample dataset includes the input feature set and the real spatial parameters, and the second sample dataset includes the input feature set, the real spatial parameters, the real morphological parameters, and the number of real crack clusters. Based on the first sample dataset, the first initial prediction model is trained to obtain the first target prediction model; Based on the second sample dataset, the second initial prediction model is trained to obtain the second target prediction model.

3. The method of claim 2, wherein, The acquisition of multiple sets of crack sample data includes: Using experimental design methods, multiple sets of different combinations of geological parameters and engineering parameters for samples were generated; Crack propagation simulation is performed for each set of parameters to generate corresponding sample 3D point clouds; Feature extraction is performed on the three-dimensional point cloud of the sample to obtain the corresponding real spatial parameters and real morphological parameters; Based on each set of parameter combinations and the corresponding real spatial parameters and real morphological parameters, the multiple sets of crack sample data are constructed.

4. The method according to claim 3, characterized in that, The process of obtaining the corresponding real spatial parameters and real morphological parameters includes: The three-dimensional point cloud of the sample is segmented to obtain multiple subsets of crack cluster point clouds; The true spatial parameters are calculated based on multiple subsets of the crack cluster point cloud. Geometric feature calculations are performed on each subset of the point cloud of the crack cluster to obtain the true morphological parameters of each crack cluster.

5. The method according to claim 2, characterized in that, The process of obtaining the first target prediction model includes: Input the geological parameters and engineering parameters of the sample in the first sample dataset into the first initial prediction model, and output the corresponding prediction spatial parameters. Calculate the mean square error between the predicted spatial parameters and the true spatial parameters obtained from the first sample dataset; With the goal of minimizing the mean square error, the parameters of the first initial prediction model are iteratively adjusted; When the preset training termination condition is met, the trained model is obtained as the first target prediction model.

6. The method according to claim 2, characterized in that, Obtaining the second target prediction model includes the following steps, and this training is performed after the first target prediction model has been trained: Input the geological parameters, engineering parameters and real spatial parameters of the second sample dataset into the second initial prediction model, and output the corresponding regression prediction vector and classification prediction vector. Based on the regression prediction vector and the true morphological parameters after padding and masking, the regression loss term is calculated, and based on the classification prediction vector and the true number of crack clusters in the second sample dataset, the classification loss term is calculated. The regression loss term and the classification loss term are weighted and summed to obtain the weighted combined loss function; With the goal of minimizing the weighted combined loss function, the parameters of the second initial prediction model are iteratively adjusted; When the preset training termination condition is met, the trained model is obtained as the second target prediction model.

7. The method according to claim 1, characterized in that, The first target prediction model predicts the target space parameters in the following way: The input vector composed of the target geological parameters and the target engineering parameters is sequentially passed through multiple cascaded one-dimensional convolutional modules to extract high-dimensional features. Each one-dimensional convolutional module sequentially performs one-dimensional convolution, batch normalization, and nonlinear activation operations. Regression analysis is performed on the high-dimensional features to obtain the target space parameters.

8. The method according to claim 1, characterized in that, The second target prediction model predicts the target morphological parameters in the following way: The target geological parameters, the target engineering parameters, and the target spatial parameters are shared and encoded to obtain a shared feature vector; The shared feature vector is input into the regression branch to generate a morphological parameter prediction vector as a morphological parameter sequence, and the shared feature vector is input into the classification branch to generate a crack cluster number probability vector. Based on the probability vector of the number of crack clusters, the predicted number of crack clusters N is determined; Based on the predicted number of crack clusters N, the first N sets of parameter values ​​are extracted from the morphological parameter prediction vector and reconstructed into an N-row parameter matrix as the target morphological parameters, wherein each row of the N-row parameter matrix corresponds to a set of crack cluster morphological parameters.

9. The method according to claim 1, characterized in that, The process of obtaining the three-dimensional point cloud of the hydraulic fracturing fracture includes: Based on the predicted number of crack clusters N and the spatial distribution information of each crack cluster extracted from the target spatial parameters; Based on the predicted number of crack clusters N, N sets of corresponding crack cluster morphology parameters are extracted from the target morphology parameters, wherein each set of crack cluster morphology parameters includes at least crack length, crack height, dip angle, average crack width, and sparsity. Based on the spatial distribution information, determine the center positions of N crack clusters in three-dimensional space that correspond to the predicted number N of crack clusters; Based on the center positions of the N crack clusters and the corresponding N sets of crack cluster morphology parameters, N three-dimensional point cloud subsets of crack clusters are generated respectively. The three-dimensional point cloud subsets of the N fracture clusters are merged to form the three-dimensional point cloud of the hydraulic fracturing fracture.

10. The method according to claim 9, characterized in that, The generation of N 3D point cloud subsets of crack clusters includes performing the following steps for each crack cluster: Based on the length, height and dip angle of the crack cluster morphology parameters corresponding to the center position of the crack cluster, a reference crack surface is constructed at the center position of the crack cluster. Based on the average crack width and sparsity in the set of crack cluster morphology parameters, the scatter distribution density on the reference crack surface is determined; Based on the scatter distribution density, a three-dimensional scatter set is generated on the reference crack surface to obtain a subset of the three-dimensional point cloud of the crack cluster corresponding to the crack cluster.

Images