Fracture density prediction method and device, electronic equipment and storage medium
By combining seismic data and well logging data, and using convolutional neural networks and random forest mapping models, the problem of insufficient accuracy in reservoir fracture density prediction was solved, and refined fracture density prediction was achieved.
Patent Information
- Application Number
- CN202410446739.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-15
- Publication Date
- 2025-10-21
AI Technical Summary
Existing technologies make it difficult to achieve refined predictions of reservoir fracture density. When seismic data is used alone, the prediction accuracy is insufficient and there is a lack of constraints from well logging data.
Combining seismic data and well logging data, a convolutional neural network model is used for training, and the accuracy of the initial fracture density is verified by a random forest mapping model to output the fracture density of the target area.
It achieves refined prediction of reservoir fracture density, improves prediction accuracy over a large area, and can more precisely characterize fracture density.
Smart Images
Figure CN120820982A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of crack density prediction, and in particular to a crack density prediction method, device, electronic equipment and storage medium. Background Art
[0002] In oil and gas geophysical research, fracture density is an important parameter for describing the reservoir characteristics of oil and gas reservoirs and is of great significance for the prediction of fractured reservoirs.
[0003] The main methods for seismic fracture prediction include post-stack seismic attributes, multi-component converted shear waves, and P-wave anisotropy. Qualitative fracture prediction using post-stack seismic data is the most widely used due to its simplicity and effectiveness when the required fracture prediction accuracy is relatively low. Multi-component converted shear wave technology can theoretically predict fracture density, but in practice, due to the accuracy of static corrections, it is difficult to fully decompose the P-wave and S-wave wave fields, resulting in less than ideal results. Similarly, although the accuracy of P-wave anisotropy fracture prediction technology has improved with the development of wide-azimuth seismic acquisition technology and OVT (Offset Vector Tile) processing technology, it still requires high seismic data quality and processing, making it difficult to promote and apply. These fracture prediction methods all rely solely on seismic data for fracture prediction, without the constraints of well logging data. They can only qualitatively predict the degree of fracture development and do not meet the requirements for detailed fracture prediction.
[0004] Therefore, how to achieve refined prediction of reservoir fracture density is a technical problem that needs to be solved urgently. Summary of the Invention
[0005] The present invention provides a method, device, electronic device and storage medium for predicting fracture density, which can achieve refined prediction of reservoir fracture density.
[0006] The embodiment of the present invention provides the following solutions:
[0007] In a first aspect, an embodiment of the present invention provides a method for predicting crack density, the method comprising:
[0008] Acquiring seismic data of a target area and logging data of at least one wellbore in the target area, wherein the target area is an area where fracture density is to be predicted;
[0009] The preset convolutional neural network model is trained based on the seismic data, and the initial fracture density of the wellbore is predicted based on the trained convolutional neural network model;
[0010] Verify whether the accuracy of initial fracture density meets the preset target based on well logging data;
[0011] When the accuracy reaches the preset target, the prediction results of the convolutional neural network model based on seismic data are output as the target fracture density of the target area.
[0012] In an optional embodiment, the convolutional neural network model is a nonlinear random forest mapping model; the preset convolutional neural network model is trained according to the seismic data, including:
[0013] The original training set for constructing the random forest mapping model based on earthquake data;
[0014] Perform multiple random sampling on multiple samples in the original training set to obtain multiple sub-training sets;
[0015] Multiple weak learners of the random forest mapping model are trained according to multiple sub-training sets;
[0016] The multiple weak learners that have completed training are combined into strong learners of the random forest mapping model to obtain a random forest mapping model that has completed training.
[0017] In an optional embodiment, the convolutional neural network model is a nonlinear random forest mapping model; after training the preset convolutional neural network model according to the seismic data, the method further includes:
[0018] Construct a model test set based on well logging data and core data;
[0019] Input the test samples of the model test set into the primary learner of the random forest mapping model to obtain the initial prediction results;
[0020] Input the initial prediction result into the secondary learner of the random forest mapping model to obtain the secondary prediction result;
[0021] When the deviation between the test result represented by the model test set and the secondary prediction result is less than a preset deviation threshold, it is determined that the random forest mapping model has completed training.
[0022] In an optional embodiment, verifying whether the accuracy of the initial fracture density reaches a preset target based on the well logging data includes:
[0023] The curve modeling of random forest algorithm is performed based on the well logging data to obtain the actual fracture density of the wellbore;
[0024] The prediction accuracy of the random forest mapping model is obtained based on the deviation between the actual crack density and the initial crack density;
[0025] When the prediction accuracy is greater than a preset threshold, it is determined that the accuracy of the initial crack density reaches a preset target.
[0026] In an optional embodiment, curve modeling using a random forest algorithm is performed based on well logging data to obtain the actual fracture density of the wellbore, including:
[0027] Perform data cleaning on the well logging data and construct the total porosity curve of the wellbore;
[0028] Adjust the porosity ratio of the wellbore to update the total porosity curve, where the porosity ratio is the ratio of matrix porosity to fracture porosity;
[0029] Determine whether the deviation between the adjustment result represented by the total porosity curve and the actual measurement result of the wellbore is less than a deviation threshold;
[0030] If so, the fracture porosity represented by the updated total porosity curve is determined as the actual fracture density of the wellbore.
[0031] In an optional embodiment, outputting the prediction result of the convolutional neural network model based on seismic data as a target fracture density in the target area includes:
[0032] Perform pre-stack elastic parameter inversion on seismic data to obtain the pre-stack elastic parameter volume of the target area;
[0033] The total porosity of the target area is predicted based on the prestack elastic parameter volume;
[0034] Regional total porosity and seismic data are input into the convolutional neural network model to obtain the target fracture density in the target area.
[0035] In an optional embodiment, predicting the regional total porosity of the target area based on the pre-stack elastic parameter volume includes:
[0036] Inputting the prestack elastic parameter volume into a random forest mapping model for completing a target training task, wherein the target training task is a training task for implementing a wellbore porosity prediction task;
[0037] The output of the random forest mapping model is determined as the regional total porosity of the target area.
[0038] In a second aspect, an embodiment of the present invention further provides a device for predicting crack density, the device comprising:
[0039] an acquisition module, configured to acquire seismic data of a target region and logging data of at least one wellbore in the target region, wherein the target region is a region where fracture density is to be predicted;
[0040] A training and prediction module is used to train a preset convolutional neural network model based on seismic data and predict the initial fracture density of the wellbore based on the trained convolutional neural network model;
[0041] A verification and determination module is used to verify whether the accuracy of the initial fracture density meets the preset target based on the well logging data;
[0042] The prediction output module is used to output the prediction results of the convolutional neural network model based on seismic data as the target fracture density of the target area when the accuracy reaches the preset target.
[0043] In a third aspect, an embodiment of the present invention further provides an electronic device comprising a processor and a memory, wherein the memory is coupled to the processor and stores instructions, which, when executed by the processor, enable the electronic device to perform the steps of any one of the methods in the first aspect.
[0044] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any one of the methods in the first aspect when executed by a processor.
[0045] Compared with the prior art, the method, device, electronic device, and storage medium for predicting crack density of the present invention have the following advantages:
[0046] The prediction method of the present invention obtains seismic data of the target area and logging data of at least one wellbore in the target area, trains a preset convolutional neural network model based on the seismic data, and predicts the initial fracture density of the wellbore based on the trained convolutional neural network model. When the accuracy of the initial fracture density verifies that it reaches the preset target based on the logging data, it indicates that the convolutional neural network model has been well trained and can more accurately implement fracture density prediction. The prediction result of the convolutional neural network model based on the seismic data is then output as the target fracture density of the target area. This prediction method uses the accuracy of the logging data in a small area to verify the training results of the convolutional neural network model, so that it has higher prediction accuracy in subsequent large-scale regional fracture density predictions, can more finely characterize the fracture density of the target area, and thus achieve refined prediction of reservoir fracture density. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0048] Figure 1 The process of the crack density prediction method provided by the embodiment of the present invention is as follows Figure 1 ;
[0049] Figure 2-1 The curve provided in the embodiment of the present invention is shown as follows Figure 1 ;
[0050] Figure 2-2 A second schematic diagram of a curve provided in an embodiment of the present invention;
[0051] Figure 3-1 A schematic diagram of a close-offset stacked earthquake provided by an embodiment of the present invention;
[0052] Figure 3-2 A schematic diagram of a medium-offset stacked earthquake provided by an embodiment of the present invention;
[0053] Figure 3-3 A schematic diagram of a far-offset stacked earthquake provided by an embodiment of the present invention;
[0054] Figure 4-1 A schematic diagram of a longitudinal wave inversion body provided by an embodiment of the present invention;
[0055] Figure 4-2 A schematic diagram of a shear wave inversion body provided in an embodiment of the present invention;
[0056] Figure 4-3 A schematic diagram of a density inversion volume provided by an embodiment of the present invention;
[0057] Figure 5-1 A schematic diagram of the prediction results of the total porosity of the target area provided by an embodiment of the present invention;
[0058] Figure 5-2 A schematic diagram of the prediction results of crack density in a target area provided by an embodiment of the present invention;
[0059] Figure 6 Flowchart 2 of the crack density prediction method provided by an embodiment of the present invention;
[0060] Figure 7 A schematic structural diagram of a crack density prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field fall within the scope of protection of the embodiments of the present invention.
[0062] See also Figure 1 , Figure 1The flowchart of a fracture density prediction method provided in an embodiment of the present invention can be applied to a data processing terminal to implement fracture density prediction in oil and gas exploration areas. The data processing terminal can be a computer device or a server device that can run the prediction method, and there is no specific limitation here. The prediction method includes:
[0063] S11. Acquire seismic data of a target area and logging data of at least one wellbore in the target area, wherein the target area is an area where fracture density is to be predicted.
[0064] Specifically, the target area can be determined based on the needs of oil and gas exploration or development. For example, when it is necessary to extract oil or natural gas from a reservoir in a certain area, the area can be determined as the target area. Seismic inversion can be performed on the target area to obtain seismic data; well logging data can be obtained based on the measurement results of the wells in the target area. It should be noted that the well logging data can be the data of a single wellbore, or it can include the well logging data of multiple wellbores. When the well logging data includes multiple wellbores, the test data of each wellbore is labeled so that a corresponding relationship is established between the test data of each wellbore and the label of the corresponding wellbore. After obtaining the seismic data and the well logging data of at least one wellbore, step S12 is entered.
[0065] S12. Train a preset convolutional neural network model according to the seismic data, and predict the initial fracture density of the wellbore based on the trained convolutional neural network model.
[0066] Specifically, the convolutional neural network model can be selected based on the fracture density prediction requirements, or it can be constructed from multiple sub-models to achieve fracture density prediction. Seismic data is acquired during seismic inversion and represents the target area's data volume during seismic inversion. After the convolutional neural network model is trained, the wellbore seismic data is fed into the model to predict the initial fracture density of the wellbore.
[0067] Exemplarily, the convolutional neural network model is a nonlinear random forest mapping model; training the preset convolutional neural network model based on seismic data includes:
[0068] The first step is to construct an original training set for the random forest mapping model based on seismic data. This original training set contains seismic data and corresponding fracture density for each area in the target region. Fracture density can be calculated or measured, for example, by measuring cores from wells in the target region to determine the fracture density of each wellbore.
[0069] In the second step, multiple samples in the original training set are randomly sampled multiple times to obtain multiple sub-training sets. Bootstrap sampling can be used to obtain multiple sub-training sets. For example, m samples in the original training set are randomly sampled k times, and each time a sample is collected and placed in the corresponding sub-training set, and then the sample is put back into the original training set to ensure that the sample may still be collected in other sub-training sets during the next sampling. In this way, multiple sub-training sets can be obtained, each sub-training set contains k samples, and m and k are both natural numbers greater than 2. To ensure the accuracy of the data in the sub-training set, when there are multiple wellbores in the target area, a sub-training set can be constructed based on the seismic data and fracture density of each wellbore, and multiple sub-training sets can be obtained based on the seismic data and fracture density of all wellbores.
[0070] The third step is to train multiple weak learners of the random forest mapping model based on the multiple sub-training sets. Since the random forest mapping model includes multiple weak learners, each weak learner is used to extract different data features of the seismic data and then classify and derive fracture density. Therefore, it is necessary to independently train k weak learners based on the multiple sub-training sets to ensure that each weak learner is effectively trained.
[0071] In the fourth step, the trained weak learners are combined into strong learners for the Random Forest Mapper model to obtain the fully trained Random Forest Mapper model. Based on actual needs, these k weak learners can be combined into the final strong learner using an ensemble strategy. Because each sub-training set is randomly sampled, the resulting weak learners extract different data features, thereby improving the generalization and robustness of the Random Forest Mapper model.
[0072] To ensure that the random forest mapping model can accurately predict fracture density in subsequent applications, in a specific embodiment, the convolutional neural network model is a nonlinear random forest mapping model. After training the preset convolutional neural network model based on seismic data, the method further includes:
[0073] The first step is to construct a model test set based on wellbore logging and core data. This model test set is used to verify the accuracy of fracture density predictions made by the trained random forest mapping model. A model test set can also be constructed based on actual fracture density and seismic data from each wellbore in the target area.
[0074] The second step is to input the test samples from the model test set into the primary learner of the random forest mapping model to obtain initial predictions. When training the random forest mapping model, a stacking approach can be used. This involves dividing the multiple training subsets into two parts, one for training the primary learner and the other for training the secondary learner. This stacking strategy involves not simply performing logical processing on the results of the weak learners, but rather adding an additional layer of learners. In other words, the results of the primary learner are fed back into the secondary learner as input. Therefore, to improve the accuracy of the training results of the test set, inputting the test samples into the primary learner yields initial predictions that reflect the training results of the primary learner.
[0075] The third step is to input the initial prediction results into the secondary learner of the random forest mapping model to obtain the secondary prediction results. The secondary prediction results can accurately represent the training results of the random forest mapping model, that is, the final prediction results.
[0076] In the fourth step, when the deviation between the test results represented by the model test set and the secondary prediction results is less than the preset deviation threshold, it means that the difference between the two is small and the model has achieved a good convergence effect, then it is determined that the random forest mapping model has completed training. It should be noted that the random forest mapping model usually uses a simple voting method to solve classification problems, that is, the final model output is the category with the most votes or one of them. For regression problems, a simple averaging method can be used to perform arithmetic averaging on the regression results obtained by k weak learners to obtain the final model output. Since the random forest algorithm samples each time to train the model, it has a strong generalization ability and plays a greater role in reducing the variance of the model. After the random forest mapping model completes the test, the seismic data of the wellbore is input into the random forest mapping model to obtain the initial fracture density of the wellbore. After obtaining the initial fracture density, step S13 is entered.
[0077] S13. Verify whether the accuracy of the initial fracture density reaches a preset target based on the well logging data.
[0078] Specifically, the actual fracture density of the wellbore can be calculated from well logging data. This calculation method can also be implemented based on the random forest algorithm. By comparing the actual fracture density with the initial fracture density, the difference between the two can be determined, which in turn represents the prediction accuracy of the convolutional neural network model. If the accuracy is greater than the preset accuracy threshold, it indicates that the preset accuracy target has been met. Conversely, if the accuracy is less than or equal to the accuracy threshold, the convolutional neural network model needs to be trained again until the post-training test results meet the preset target. The preset target can be set according to actual needs to meet the fracture density prediction accuracy of the target area.
[0079] Exemplarily, verifying whether the accuracy of the initial fracture density reaches a preset target based on the well logging data includes:
[0080] The first step is to perform curve modeling using the random forest algorithm based on the well logging data to determine the actual fracture density of the wellbore. The random forest algorithm is an ensemble learning method based on decision trees. By constructing and combining multiple decision trees, the model's prediction accuracy and stability are improved. Curve modeling can determine the actual fracture density of the wellbore.
[0081] In one feasible solution, obtaining the actual fracture density of the wellbore includes the following four sub-steps:
[0082] The first sub-step is to clean the logging data and construct the total porosity curve of the wellbore. Data cleaning includes missing value detection, removal of duplicate values and outliers to improve the accuracy of the logging data. Figure 2-1 Well logging data can be represented by curves, including the longitudinal wave curve (Vp), the shear wave curve (Vs), and the density curve (Rhob), and the total porosity curve (Phit) can be obtained by total porosity inversion.
[0083] The second substep involves adjusting the wellbore's porosity ratio to update the total porosity curve. The porosity ratio is the ratio of matrix porosity to fracture porosity. Rock physics modeling, using a soft porosity rock physics model, categorizes porosity into two types: matrix porosity (pores with an aspect ratio of 0.15 to 0.2) and fracture porosity (pores with an aspect ratio of 0.01 to 0.02). The total porosity curve is updated by continuously adjusting the ratio of matrix porosity to fracture porosity.
[0084] The third sub-step is to determine whether the deviation between the adjusted result represented by the total porosity curve and the actual measured result of the wellbore is less than a deviation threshold. The actual measured result of the wellbore can be obtained by measuring the fracture porosity of the wellbore core, and the deviation is determined by comparing the two.
[0085] In the fourth sub-step, when the deviation between the adjusted result and the measured result of the wellbore is less than the deviation threshold, it means that the adjusted fracture porosity is closest to the measured result, and the fracture porosity represented by the updated total porosity curve is determined as the actual fracture density of the wellbore. Figure 2-2 ,The fourth dotted line in the figure represents the crack density predicted by the ,random forest algorithm.
[0086] In the second step, the prediction accuracy of the random forest mapping model is obtained based on the deviation between the actual fracture density and the initial fracture density. Since both the actual fracture density and the initial fracture density can be represented by a curve of fracture density varying with formation depth, Figure 2-1 and 2-2In the curve, the vertical direction represents formation depth, and the horizontal fluctuations represent fluctuations in the corresponding data. A target formation thickness can be set, and the actual fracture density and initial fracture density for the same formation can be extracted based on the target formation thickness. Comparing the two yields the deviation. A larger deviation indicates lower prediction accuracy; conversely, a smaller deviation indicates higher prediction accuracy.
[0087] In the third step, if the prediction accuracy is greater than the preset threshold, it indicates that the convolutional neural network can accurately predict the crack density, and the accuracy of the initial crack density is determined to have reached the preset target. After verifying that the accuracy of the initial crack density has reached the preset target, the process proceeds to step S14.
[0088] S14. When the accuracy reaches a preset target, the prediction result of the convolutional neural network model based on the seismic data is output as the target fracture density of the target area.
[0089] Specifically, seismic data from a target region can be fed into a trained convolutional neural network model. Running the convolutional neural network model can predict the target fracture density in the target region. The target region is defined as a region occupying a certain spatial area within a stratum, which can be a geological body. Therefore, the target fracture density represents the fracture density within the stratum space, and the target fracture density is a data volume representing the fracture density in the target region.
[0090] Exemplarily, outputting the prediction result of the convolutional neural network model based on seismic data as a target fracture density in the target area includes:
[0091] The first step is to perform pre-stack elastic parameter inversion on the seismic data to obtain the pre-stack elastic parameter volume of the target area. When performing pre-stack elastic parameter inversion, the elastic impedance formula can be derived from the relationship between the Aki-Richards approximation and the reflection coefficient at any angle of incidence:
[0092]
[0093] The reflection coefficient approximation of the reflected PP wave can be further derived as:
[0094]
[0095] Where:
[0096]
[0097] Where: α0, β0, ρ0 are the means of the elastic parameters above and below the reflection interface, Δα, Δβ, Δρ are the differences in the elastic parameters above and below the reflection interface, and θ is the angle of incidence.
[0098] Based on the near-offset, mid-offset and far-offset stacked seismic and wellbore P-wave (Vp), S-wave (Vs) and density curve (Rhob) of the pre-stack gathers, the pre-stack elastic parameter inversion is carried out using equations 4-1 to 4-3 to obtain the pre-stack elastic parameter volume. Figures 3-1 to 3-3 , Figure 3-1 It is a stack of close-offset stacked earthquakes; Figure 3-2 It is a stack of medium-offset stacked earthquakes; Figure 3-3 This is the stacked volume of far-offset stacked earthquakes. Different inversion volumes are obtained after pre-stack elastic parameter inversion. Figures 4-1 to 4-3 The pre-stack elastic parameter body includes the P-wave (Vp) inversion body, the S-wave (Vs) inversion body and the density (Rhob) inversion body. The P-wave inversion body is as follows: Figure 4-1 As shown, the shear wave inversion body is as follows Figure 4-2 As shown, the density inversion volume is as follows Figure 4-3 As shown in the figure, after obtaining the pre-stack elastic parameter volume, proceed to the next step.
[0099] The second step is to predict the regional total porosity of the target area based on the prestack elastic parameter volume. This prediction can be performed using a convolutional neural network model. For example, the convolutional neural network model is first trained. When the training results meet the preset target, the prestack elastic parameter volume is input into the convolutional neural network model to determine the regional total porosity of the target area.
[0100] In one feasible solution, a nonlinear random forest mapping model is used to implement prediction, including:
[0101] The prestack elastic parameter volume is input into the random forest mapping model to complete the target training task, wherein the target training task is to implement the training task of wellbore porosity prediction; the output result of the random forest mapping model is determined as the regional total porosity of the target area. Figure 5-1 The predicted total porosity of a section in the destination region can be represented by this graph. When predicting regional total porosity, the total porosity curve and the fracture density of the wellbore can also be applied to the random forest mapping model to improve prediction accuracy. Once the regional total porosity is obtained, proceed to the next step.
[0102] In the third step, the regional total porosity and seismic data are input into the convolutional neural network model to obtain the target fracture density of the target area. The regional total porosity represents the porosity of the target area in space. As a data volume representing the porosity of the target area, the target fracture density can be restricted to improve the accuracy of the target fracture density. Please refer to Figure 5-2 , schematic diagram of the prediction results of the target crack density in the target area.
[0103] The following embodiments of the present invention will generally describe the method for predicting the target crack density in the target area. Figure 6 , Figure 6 This is the second flow chart of the prediction method, which specifically includes:
[0104] Step S601: obtaining pre-stack seismic data, and extracting seismic data of the target area based on the pre-stack seismic data.
[0105] Step S602: Perform close-offset stacking, mid-offset stacking, and far-offset stacking on the seismic data.
[0106] Step S603: performing pre-stack elastic parameter inversion on the seismic data.
[0107] Step S604: Based on the processing results of the pre-stack elastic parameter inversion, a P-wave inversion volume, a S-wave inversion volume, and a density inversion volume are obtained.
[0108] Step S611: Acquire the well logging curve of the wellbore in the target area.
[0109] Step S612: Analyze the longitudinal wave curve, the shear wave curve, and the fracture density curve of the wellbore based on the well logging curve.
[0110] Step S613: Analyze the total porosity of the wellbore based on the well logging curve.
[0111] In step S614, the longitudinal wave curve, the shear wave curve, the fracture density curve of the wellbore and the total porosity are input into the random forest mapping model.
[0112] Step S621: Acquire the well logging curve of the wellbore in the target area.
[0113] Step S622: Perform rock physics modeling on the wellbore.
[0114] Step S623: Analyze the longitudinal wave curve, the shear wave curve and the fracture density curve of the wellbore based on the well logging curve.
[0115] Step S624: Determine the fracture density of the wellbore based on the results of the rock physics modeling.
[0116] Step S625: training and verifying the random forest mapping model.
[0117] Step S615: Determine the total porosity of the target area, ie, the total porosity volume.
[0118] Step S605 : predicting the target fracture density of the target area, ie, the fracture density volume, based on the inversion volume and the total porosity volume.
[0119] The present invention adopts a random forest algorithm learning method to establish a random forest mapping model of wellbore elastic parameters, total porosity and fracture density, transforming the rock physics modeling problem into a data-driven statistical learning problem, avoiding the complex rock physics modeling and parameter adjustment process; secondly, there is usually a good correspondence between fracture density and total porosity, and the present invention uses total porosity to constrain fracture density modeling, making the fracture density prediction results interpretable and reasonable.
[0120] Based on the same technical concept as the prediction method, the embodiment of the present invention also provides a device for predicting crack density, see Figure 7 , Figure 7 Schematic diagram of the structure of the prediction device, the prediction device includes:
[0121] An acquisition module 701 is configured to acquire seismic data of a target region and logging data of at least one wellbore in the target region, wherein the target region is a region where fracture density is to be predicted;
[0122] A training prediction module 702 is configured to train a preset convolutional neural network model according to seismic data, and predict the initial fracture density of the wellbore based on the trained convolutional neural network model;
[0123] Verification and determination module 703, used to verify whether the accuracy of the initial fracture density reaches a preset target based on the well logging data;
[0124] The prediction output module 704 is used to output the prediction result of the convolutional neural network model based on the seismic data as the target fracture density of the target area when the accuracy reaches a preset target.
[0125] In an optional embodiment, the convolutional neural network model is a nonlinear random forest mapping model; the training prediction module includes:
[0126] Build a submodule for constructing the original training set of the random forest mapping model based on earthquake data;
[0127] The sampling submodule is used to perform multiple random sampling on multiple samples in the original training set to obtain multiple sub-training sets;
[0128] A training submodule, for training multiple weak learners of a random forest mapping model according to multiple sub-training sets;
[0129] The combination acquisition submodule is used to combine multiple weak learners that have completed training into a strong learner of the random forest mapping model to obtain a random forest mapping model that has completed training.
[0130] In an optional embodiment, the convolutional neural network model is a nonlinear random forest mapping model; the prediction device further includes:
[0131] A construction module for constructing a model test set based on well logging data and core data;
[0132] The first acquisition module is used to input the test samples of the model test set into the primary learner of the random forest mapping model to obtain the initial prediction results;
[0133] The second acquisition module is used to input the initial prediction result into the secondary learner of the random forest mapping model to obtain a secondary prediction result;
[0134] The determination module is used to determine that the random forest mapping model has completed training when the deviation between the test result represented by the model test set and the secondary prediction result is less than a preset deviation threshold.
[0135] In an optional embodiment, the verification and determination module includes:
[0136] The first acquisition submodule is used to perform curve modeling of the random forest algorithm based on the well logging data to obtain the actual fracture density of the wellbore;
[0137] The second acquisition submodule is used to obtain the prediction accuracy of the random forest mapping model according to the deviation between the actual crack density and the initial crack density;
[0138] The determination submodule is used to determine whether the accuracy of the initial crack density reaches a preset target when the prediction accuracy is greater than a preset threshold.
[0139] In an optional embodiment, the first obtaining submodule includes:
[0140] A construction unit is used to clean the logging data and construct the total porosity curve of the wellbore;
[0141] An updating unit, configured to adjust the porosity ratio of the wellbore to update the total porosity curve, wherein the porosity ratio is a configuration ratio of matrix porosity to fracture porosity;
[0142] a judgment unit, configured to judge whether a deviation between an adjustment result represented by a total porosity curve and an actual measurement result of the wellbore is less than a deviation threshold;
[0143] The determining unit is configured to determine the fracture porosity represented by the updated total porosity curve as the actual fracture density of the wellbore when the deviation between the adjustment result and the actual measured result of the wellbore is less than a deviation threshold.
[0144] In an optional embodiment, the prediction output module includes:
[0145] The third acquisition submodule is used to perform pre-stack elastic parameter inversion on seismic data to obtain the pre-stack elastic parameter volume of the target area;
[0146] The prediction submodule is used to predict the total porosity of the target area based on the prestack elastic parameter volume;
[0147] The fourth acquisition submodule is used to input the regional total porosity and seismic data into the convolutional neural network model to obtain the target fracture density in the target area.
[0148] In an optional embodiment, the prediction submodule includes:
[0149] An input unit, configured to input a prestack elastic parameter volume into a random forest mapping model for completing a target training task, wherein the target training task is a training task for implementing a wellbore porosity prediction task;
[0150] The determination unit is used to determine the output result of the random forest mapping model as the regional total porosity of the target area.
[0151] Based on the same technical concept as the prediction method, an embodiment of the present invention also provides an electronic device, including a processor and a memory, the memory being coupled to the processor, the memory storing instructions, and when the instructions are executed by the processor, the electronic device executes the steps of any one of the prediction methods.
[0152] Based on the same technical concept as the prediction method, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any one of the prediction methods when executed by a processor.
[0153] The technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:
[0154] The prediction method obtains seismic data from the target area and well logging data from at least one wellbore in the target area. A pre-set convolutional neural network model is trained based on the seismic data. The trained convolutional neural network model then predicts the initial fracture density of the wellbore. When the accuracy of the initial fracture density, as verified by the well logging data, reaches the preset target, the convolutional neural network model is well trained and can accurately predict fracture density. The prediction results of the convolutional neural network model based on the seismic data are then output as the target fracture density for the target area. This prediction method leverages the accuracy of well logging data in a small area to validate the training results of the convolutional neural network model, resulting in higher prediction accuracy for subsequent large-scale regional fracture density predictions. This allows for a more refined characterization of the fracture density in the target area, thereby enabling refined prediction of reservoir fracture density.
[0155] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0156] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (modules, systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0157] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0158] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0159] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0160] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
Claims
1. A method for predicting crack density, characterized in that: The method comprises: Acquiring seismic data of a target region and logging data of at least one wellbore in the target region, wherein the target region is a region where fracture density is to be predicted; Training a preset convolutional neural network model according to the seismic data, and predicting the initial fracture density of the wellbore based on the trained convolutional neural network model; Verifying whether the accuracy of the initial fracture density reaches a preset target based on the well logging data; When the accuracy reaches the preset target, the prediction result of the convolutional neural network model based on the seismic data is output as the target fracture density of the target area.
2. The method for predicting crack density according to claim 1, characterized in that: The convolutional neural network model is a nonlinear random forest mapping model; the training of the preset convolutional neural network model according to the seismic data includes: Constructing an original training set of the random forest mapping model based on the seismic data; Performing multiple random sampling on multiple samples in the original training set to obtain multiple sub-training sets; Training a plurality of weak learners of the random forest mapping model respectively according to the plurality of sub-training sets; The multiple weak learners that have completed training are combined into strong learners of the random forest mapping model to obtain a random forest mapping model that has completed training.
3. The method for predicting crack density according to claim 1, wherein: The convolutional neural network model is a nonlinear random forest mapping model; after the preset convolutional neural network model is trained according to the seismic data, the method further includes: constructing a model test set based on the well logging data and core data of the wellbore; Inputting the test samples of the model test set into the primary learner of the random forest mapping model to obtain an initial prediction result; Inputting the initial prediction result into the secondary learner of the random forest mapping model to obtain a secondary prediction result; When the deviation between the test result represented by the model test set and the secondary prediction result is less than a preset deviation threshold, it is determined that the random forest mapping model has completed training.
4. The method for predicting crack density according to claim 1, wherein: Verifying whether the accuracy of the initial fracture density reaches a preset target based on the well logging data includes: Performing curve modeling using a random forest algorithm based on the well logging data to obtain an actual fracture density of the wellbore; Obtaining prediction accuracy of the random forest mapping model according to a deviation between the actual crack density and the initial crack density; When the prediction accuracy is greater than a preset threshold, it is determined that the accuracy of the initial crack density reaches the preset target.
5. The method for predicting crack density according to claim 4, characterized in that: The performing curve modeling of a random forest algorithm based on the well logging data to obtain the actual fracture density of the wellbore includes: performing data cleaning on the well logging data and constructing a total porosity curve of the wellbore; Adjusting the porosity ratio of the wellbore to update the total porosity curve, wherein the porosity ratio is the ratio of matrix porosity to fracture porosity; determining whether a deviation between an adjustment result represented by the total porosity curve and an actual measurement result of the wellbore is less than a deviation threshold; If so, the fracture porosity represented by the updated total porosity curve is determined as the actual fracture density of the wellbore.
6. The method for predicting crack density according to claim 1, characterized in that: Outputting the prediction result of the convolutional neural network model based on the seismic data as the target fracture density of the target area includes: Performing pre-stack elastic parameter inversion on the seismic data to obtain a pre-stack elastic parameter volume of the target area; predicting the regional total porosity of the target area based on the prestack elastic parameter body; The regional total porosity and the seismic data are input into the convolutional neural network model to obtain the target fracture density of the target area.
7. The method for predicting crack density according to claim 6, characterized in that: The predicting of the regional total porosity of the target area according to the pre-stack elastic parameter body includes: Inputting the pre-stack elastic parameter volume into a random forest mapping model for completing a target training task, wherein the target training task is a training task for implementing porosity prediction of the wellbore; The output result of the random forest mapping model is determined as the regional total porosity of the target area.
8. A device for predicting crack density, characterized in that: The device comprises: an acquisition module, configured to acquire seismic data of a target region and logging data of at least one wellbore in the target region, wherein the target region is a region where fracture density is to be predicted; A training prediction module, configured to train a preset convolutional neural network model according to the seismic data, and predict the initial fracture density of the wellbore based on the trained convolutional neural network model; A verification and determination module, configured to verify whether the accuracy of the initial fracture density reaches a preset target based on the well logging data; A prediction output module is used to output the prediction result of the convolutional neural network model based on the seismic data as the target fracture density of the target area when the accuracy reaches the preset target.
9. An electronic device, characterized in that: The electronic device comprises a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.