Convolutional recurrent neural network-based dessert physical property prediction method and device, electronic equipment and medium
By constructing a sweet spot property prediction method based on convolutional recurrent neural networks, the problem of insufficient discrimination of existing models in thin interbedded reservoirs is solved, and higher prediction accuracy and generalization ability are achieved, which is applicable to reservoir parameter prediction in tight sandstone gas development.
Patent Information
- Application Number
- CN202411646617.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2026-05-19
AI Technical Summary
Existing convolutional neural network models lack sufficient discrimination in thin interbedded reservoirs and have biased predictions of physical properties at some locations. Their generalization ability is also generally poor, making it difficult to meet the reservoir parameter prediction requirements for tight sandstone gas development.
A convolutional recurrent neural network-based approach was adopted. Learning samples were constructed using a "point-to-window" model, and the structural parameters of the convolutional recurrent network model, including the convolutional structure, pooling layer combination, number of recurrent layers, number of fully connected layers, and learning rate, were determined using the controlled variable method. The model was then trained and validated, which improved the model's generalization ability and prediction accuracy.
It improved the agreement between reservoir parameter prediction results and logging curves, enhanced the ability to identify thin interbedded reservoirs, and improved the adaptability and prediction accuracy of the model in the study area. In particular, it outperformed traditional models in the identification of thin sandstone layers and effective reservoirs.
Smart Images
Figure CN122063640A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of petroleum geophysical exploration technology, and more specifically, to a method, apparatus, electronic device, and medium for predicting the properties of sweet spots based on convolutional recurrent neural networks. Background Technology
[0002] Investigations have revealed abundant tight sandstone gas resources in the Upper Paleozoic strata of the eastern margin of the Ordos Basin. The basin exhibits well-developed source rocks and high hydrocarbon generation potential, with multiple layers of tight gas reservoirs. Moderately developed vertical faults serve as excellent pathways for the migration of tight gas from the gas-generating layers to the sandstone reservoirs. Simultaneously, interlayer mudstone, interdistributary mudstone, and regional mudstone act as favorable caprocks, forming a multi-type source-reservoir-caprock combination. Utilizing 3D seismic data, well logging data, and prior geological information in this area to conduct research on reservoir sweet spot identification techniques and predict productive sweet spots in target areas is of great significance for the development of tight sandstone gas in this region.
[0003] Seismic reservoir parameter prediction techniques include impedance inversion, indirect inversion, wave equation inversion, geostatistical inversion, and artificial intelligence-based reservoir prediction. Impedance inversion is theoretically based on the Knott-Zoeppritz equation, converting interface reflection characteristics from seismic data into formation impedance information and using impedance differences in the subsurface medium for reservoir prediction. Indirect inversion, based on elasticity formulas, converts impedance inversion results into elastic parameters. In practice, empirical formulas are derived from geological, well logging, and experimental data to predict reservoir properties. Geostatistical inversion methods primarily utilize weighted averaging and statistical methods for reservoir parameter prediction. In recent years, with the development of GPUs and parallel computing, computer computing power has significantly increased. Furthermore, abundant seismic, well logging, and rock physics data can support large-scale neural network training, making neural network technology a promising area for development in geophysics.
[0004] Conventional fully connected networks possess the ability to fit arbitrary nonlinear relationships. They typically use the original seismic traces and parameter-sensitive attributes as model inputs and predict well-logging values as outputs. Convolutional neural networks (CNNs), in addition to their ability to fit nonlinear relationships, possess the local feature extraction capability that fully connected networks lack, and can fully utilize the waveform information of seismic data. Tests revealed that the CNN model has a certain ability to identify thicker sandstone layers and sweet spot reservoirs. The predicted properties of sweet spot reservoirs show a higher fit to well-logging curves than fully connected networks, and the stratigraphic profiles appear more natural. However, it lacks sufficient discrimination against thin interbedded reservoirs, and the predicted properties at some locations show bias, indicating only moderate generalization ability. Therefore, further improving the fit between reservoir parameter predictions and well-logging curves, and enhancing the ability to identify thin interbedded reservoirs, has become a key research focus.
[0005] Currently, a method for predicting the physical properties of desserts based on convolutional recurrent neural networks needs to be developed.
[0006] The information disclosed in the background section of this invention is intended only to enhance the understanding of the general background of this invention, and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art. Summary of the Invention
[0007] This invention proposes a method, device, electronic device, and medium for predicting the physical properties of desserts based on convolutional recurrent neural networks. It realizes the research on the method of predicting the physical properties of desserts using convolutional recurrent neural networks by leveraging the seismic waveform information extraction capability of convolutional networks and the seismic temporal feature mining capability of recurrent networks.
[0008] In a first aspect, embodiments of this disclosure provide a method for predicting the physical properties of desserts based on convolutional recurrent neural networks, including:
[0009] Based on the characteristics of the network model, a "point-to-window" model is adopted to construct learning samples;
[0010] A convolutional recurrent network model is established, and the model is trained using the aforementioned learning samples;
[0011] The raw seismic data is input into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
[0012] As a specific implementation of this disclosure, the "point-to-window" mode for constructing learning samples, based on the characteristics of the network model, includes:
[0013] After normalizing and performing timely deep transformation on the data, the sweet spot physical property parameter curves and their corresponding well-side seismic data are connected in a "point-to-window" manner, and the sequence samples formed in chronological order are the learning samples.
[0014] As one specific implementation of this disclosure, the learning samples include a training sample set and a verification sample set.
[0015] As a specific implementation of this disclosure, establishing a convolutional recurrent network model includes:
[0016] The control variable method is used to determine the structural parameters of the convolutional recurrent network model, including the convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the order of the learning rate.
[0017] As a specific implementation of this disclosure, the convolutional recurrent network model is trained based on the training sample set.
[0018] As a specific implementation of this disclosure, the generalization ability of the convolutional recurrent network model is tested based on the verification sample set.
[0019] Secondly, embodiments of this disclosure also provide a dessert property prediction device based on a convolutional recurrent neural network, comprising:
[0020] The sample building module constructs learning samples using a "point-to-window" mode, based on the characteristics of the network model.
[0021] The modeling module establishes a convolutional recurrent network model and trains the model using the learning samples.
[0022] The prediction module inputs the raw seismic data into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
[0023] As a specific implementation of this disclosure, the "point-to-window" mode for constructing learning samples, based on the characteristics of the network model, includes:
[0024] After normalizing and performing timely deep transformation on the data, the sweet spot physical property parameter curves and their corresponding well-side seismic data are connected in a "point-to-window" manner, and the sequence samples formed in chronological order are the learning samples.
[0025] As one specific implementation of this disclosure, the learning samples include a training sample set and a verification sample set.
[0026] As a specific implementation of this disclosure, establishing a convolutional recurrent network model includes:
[0027] The control variable method is used to determine the structural parameters of the convolutional recurrent network model, including the convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the order of the learning rate.
[0028] As a specific implementation of this disclosure, the convolutional recurrent network model is trained based on the training sample set.
[0029] As a specific implementation of this disclosure, the generalization ability of the convolutional recurrent network model is tested based on the verification sample set.
[0030] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:
[0031] Memory, which stores executable instructions;
[0032] A processor that executes the executable instructions in the memory to implement the dessert property prediction method based on a convolutional recurrent neural network.
[0033] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned dessert property prediction method based on a convolutional recurrent neural network.
[0034] The methods and apparatus of the present invention have other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description
[0035] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts.
[0036] Figure 1 A flowchart illustrating the steps of a dessert property prediction method based on a convolutional recurrent neural network according to an embodiment of the present invention is shown.
[0037] Figure 2 A schematic diagram of a convolutional recurrent neural network sample construction according to an embodiment of the present invention is shown.
[0038] Figure 3 A schematic diagram of a convolutional recurrent neural network model according to an embodiment of the present invention is shown.
[0039] Figure 4a and Figure 4b Schematic diagrams of the sandstone content prediction loss curve and porosity prediction loss curve of a convolutional recurrent neural network model according to an embodiment of the present invention are shown respectively.
[0040] Figure 5 A schematic diagram showing a comparison of the prediction results curves of sandstone content and porosity using a convolutional recurrent neural network according to an embodiment of the present invention is illustrated.
[0041] Figure 6a , Figure 6b , Figure 6c , Figure 6d The diagrams show the sandstone content prediction results of the original seismic profile, the fully connected network, the convolutional network, and the convolutional recursive network, respectively, according to an embodiment of the present invention.
[0042] Figure 7a , Figure 7b , Figure 7c , Figure 7dThe diagrams show the porosity prediction results of the original seismic profile, the fully connected network, the convolutional network, and the convolutional recursive network, respectively, according to an embodiment of the present invention.
[0043] Figure 8 A block diagram of a dessert property prediction device based on a convolutional recurrent neural network according to an embodiment of the present invention is shown.
[0044] Explanation of reference numerals in the attached figures:
[0045] 201. Sample Establishment Module; 202. Modeling Module; 203. Prediction Module. Detailed Implementation
[0046] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention are described below, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.
[0047] To facilitate understanding of the solutions and effects of the embodiments of the present invention, six specific application examples are given below. Those skilled in the art should understand that these examples are merely for the purpose of understanding the present invention, and any specific details therein are not intended to limit the present invention in any way.
[0048] Example 1
[0049] Figure 1 A flowchart illustrating the steps of a dessert property prediction method based on a convolutional recurrent neural network according to an embodiment of the present invention is shown.
[0050] like Figure 1 As shown, this dessert property prediction method based on convolutional recurrent neural networks includes:
[0051] Step 101: Based on the characteristics of the network model, construct learning samples using a "point-to-window" model;
[0052] Step 102: Establish a convolutional recurrent network model and train the model using learning samples;
[0053] Step 103: Input the raw seismic data into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
[0054] In one example, based on the characteristics of the network model, the "point-to-window" pattern is used to construct learning samples, including:
[0055] After normalizing and transforming the data, the sweet spot physical property parameter curves are connected to their corresponding well-side seismic data through a "point-to-window" method, and the sequence samples formed in chronological order are used as learning samples.
[0056] In one example, the learning samples include a training sample set and a validation sample set.
[0057] In one example, building a convolutional recurrent network model includes:
[0058] The control variable method is used to determine the structural parameters of the convolutional recurrent network model, including the convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the order of the learning rate.
[0059] In one example, a convolutional recurrent network model is trained based on a training sample set.
[0060] In one example, the generalization ability of a convolutional recurrent network model is tested against a validation sample set.
[0061] Specifically, this invention constructs a convolutional recurrent neural network model learning sample according to the sedimentary sequence, and divides the sample into a training set and a validation set. Unlike conventional networks that use multiple seismic attributes as input, the convolutional recurrent network selects raw seismic data as input. After data normalization and time-depth conversion, the sweet spot physical property parameter curves are matched with their corresponding well-side seismic data through a "point-to-window" method, and a sequence of samples is constructed in chronological order.
[0062] A convolutional recurrent neural network (CRN) model is constructed. The prediction process of the CRN involves dividing the sample set into two groups: a training sample set and a validation sample set, used for network training and model validation, respectively. The control variable method is used to determine the various structural parameters of the CRN model, including the order of convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the learning rate. This process is then used to verify the model's training effectiveness and test its generalization ability.
[0063] The structure of the convolutional recurrent neural network prediction model for dessert properties in the study area was determined through model parameter testing. Training and validation sample sets were input separately to obtain the network model and assess its adaptability. The model's accuracy and generalization ability in predicting dessert properties were evaluated using loss curves.
[0064] The constructed training samples are input into the convolutional recurrent neural network model to complete the training. The prediction results of the sweet spot properties of the training wells and validation wells are compared and analyzed with the original property curves of the wells to verify the accuracy and generalization ability of the model. After completing the single-well analysis, the profile of the property prediction results is analyzed to observe the accuracy of the property prediction of the convolutional recurrent network model and its grasp of the overall geological background of the work area.
[0065] Example 2
[0066] The present invention also provides a dessert property prediction device based on a convolutional recurrent neural network, comprising:
[0067] The sample building module constructs learning samples using a "point-to-window" mode, based on the characteristics of the network model.
[0068] The modeling module establishes a convolutional recurrent network model and trains the model using learning samples.
[0069] The prediction module inputs the raw seismic data into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
[0070] In one example, based on the characteristics of the network model, the "point-to-window" pattern is used to construct learning samples, including:
[0071] After normalizing and transforming the data, the sweet spot physical property parameter curves are connected to their corresponding well-side seismic data through a "point-to-window" method, and the sequence samples formed in chronological order are used as learning samples.
[0072] In one example, the learning samples include a training sample set and a validation sample set.
[0073] In one example, building a convolutional recurrent network model includes:
[0074] The control variable method is used to determine the structural parameters of the convolutional recurrent network model, including the convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the order of the learning rate.
[0075] In one example, a convolutional recurrent network model is trained based on a training sample set.
[0076] In one example, the generalization ability of a convolutional recurrent network model is tested against a validation sample set.
[0077] Specifically, this invention constructs a convolutional recurrent neural network model learning sample according to the sedimentary sequence, and divides the sample into a training set and a validation set. Unlike conventional networks that use multiple seismic attributes as input, the convolutional recurrent network selects raw seismic data as input. After data normalization and time-depth conversion, the sweet spot physical property parameter curves are matched with their corresponding well-side seismic data through a "point-to-window" method, and a sequence of samples is constructed in chronological order.
[0078] A convolutional recurrent neural network (CRN) model is constructed. The prediction process of the CRN involves dividing the sample set into two groups: a training sample set and a validation sample set, used for network training and model validation, respectively. The control variable method is used to determine the various structural parameters of the CRN model, including the order of convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the learning rate. This process is then used to verify the model's training effectiveness and test its generalization ability.
[0079] The structure of the convolutional recurrent neural network prediction model for dessert properties in the study area was determined through model parameter testing. Training and validation sample sets were input separately to obtain the network model and assess its adaptability. The model's accuracy and generalization ability in predicting dessert properties were evaluated using loss curves.
[0080] The constructed training samples are input into the convolutional recurrent neural network model to complete the training. The prediction results of the sweet spot properties of the training wells and validation wells are compared and analyzed with the original property curves of the wells to verify the accuracy and generalization ability of the model. After completing the single-well analysis, the profile of the property prediction results is analyzed to observe the accuracy of the property prediction of the convolutional recurrent network model and its grasp of the overall geological background of the work area.
[0081] Example 3
[0082] Taking a tight sandstone gas block in the eastern source of the Ordos Basin as an example, the prediction of sweet spot physical properties was completed according to the specific implementation steps.
[0083] Figure 2 A schematic diagram of a convolutional recurrent neural network sample construction according to an embodiment of the present invention is shown.
[0084] Training data for the convolutional recurrent network model was obtained. Unlike conventional network models that use multiple seismic attributes as input, the convolutional recurrent network uses raw seismic data as input. After data normalization and time-depth transformation, this invention uses a "point-to-window" approach for well-seismic matching and constructs sequence samples in chronological order. Figure 2 This diagram illustrates the construction of samples for a convolutional recurrent neural network. The left side shows the physical property parameter curves of the sweet spot, and the right side shows the corresponding well-side seismic data. The two are connected through a "point-to-window" method and form a sequence of samples in chronological order.
[0085] A convolutional recurrent neural network (CRN) model was constructed. After acquiring the data, seismic data normalization and a training sample set were built. The CRN prediction process divided the sample set into two groups: a training sample set and a validation sample set, used for network training and model validation, respectively. The controlled variable method was used to test the effects of the convolutional structure, convolutional layer and pooling layer combination, the number of recursive layers, the number of fully connected layers, and the learning rate on the model's training performance and generalization ability. Tables 1, 2, 3, and 4 show the impact of various test parameters on the model's training performance and generalization ability.
[0086] Table 1. Convolutional Structure Test Table
[0087]
[0088] Table 2 Recursive Structure Test Table
[0089]
[0090] Table 3 Test Table for the Number of Fully Connected Layers
[0091]
[0092]
[0093] Table 4 Learning Rate Test Table
[0094]
[0095] Figure 3 A schematic diagram of a convolutional recurrent neural network model according to an embodiment of the present invention is shown.
[0096] Based on the test results, a convolutional recurrent neural network model was finally established, consisting of two convolutional layers, two pooling layers, one LSTM recursive layer, and one fully connected layer. The ReLU activation function was selected, and the mean squared error function was used as the loss function for model training. Stochastic gradient descent was used to train the network model and update the network parameters. The learning rate was set to 0.0025, and dropout technology was used to reduce the risk of overfitting. Figure 3 A convolutional recurrent neural network model was developed to predict the physical properties of desserts in the study area.
[0097] Figure 4a and Figure 4b The diagrams show the loss curves for sandstone content prediction and porosity prediction of a convolutional recurrent neural network model according to an embodiment of the present invention. The blue curve represents the loss curve for the training sample set, and the red curve represents the loss curve for the validation sample set.
[0098] A convolutional recurrent neural network (CRN) model was trained. The structure of the CRN predictive model for the physical properties of sweet sandstone in the study area was determined through model parameter testing. The model was trained by inputting training and validation sample sets, and its adaptability was assessed. As shown in Figure 4, the CRN model is well-suited for predicting the content and porosity of sweet sandstone in the study area. It converges on both the training and validation sample sets, demonstrating high accuracy in predicting sweet sandstone properties. The accuracy of property prediction on the validation sample set is slightly lower than that on the training sample set, indicating strong generalization ability.
[0099] Figure 5 A schematic diagram showing a comparison of the prediction results curves of sandstone content and porosity using a convolutional recurrent neural network according to an embodiment of the present invention is illustrated.
[0100] Predicting sweet spot properties of tight sandstone. The constructed training samples were input into a convolutional recurrent neural network (CNN) model for training. The sweet spot property prediction results of the CNN at the training well location were compared and analyzed with the original property curves of the training well. After analyzing the training effect of the CNN sweet spot property prediction network model, the generalization ability of the network model was further verified. Figure 5 The figures show comparisons of the predicted sweet spot sandstone content from training wells with the original sandstone content logging curves, the predicted sweet spot sandstone content from verification wells with the original sandstone content logging curves, the predicted sweet spot porosity from training wells with the original porosity logging curves, and the predicted sweet spot porosity from verification wells with the original porosity logging curves. The black curve represents the original logging curve, and the red curve represents the predicted result. Observations show that the convolutional recurrent neural network model has a strong ability to identify sweet spot reservoir systems and thin sweet spot reservoirs. The predicted values from training and verification wells have a high degree of fit with the original curves, while the prediction accuracy is slightly lower in verification wells. There are a few locations where the convolutional recurrent neural network predicts sweet spot reservoirs with deviations, indicating that the model has strong generalization ability.
[0101] Figure 6a , Figure 6b , Figure 6c , Figure 6d The diagrams show the sandstone content prediction results of the original seismic profile, the fully connected network, the convolutional network, and the convolutional recursive network, respectively, according to an embodiment of the present invention.
[0102] Figure 7a , Figure 7b , Figure 7c , Figure 7d The diagrams show the porosity prediction results of the original seismic profile, the fully connected network, the convolutional network, and the convolutional recursive network, respectively, according to an embodiment of the present invention.
[0103] like Figures 6a-6d , Figures 7a-7d As shown, the convolutional recurrent neural network model has a significantly stronger ability to identify sandstone layers and effective reservoirs than the fully connected network model and the convolutional neural network model. It also has a better ability to distinguish between adjacent thin sandstone layers and effective reservoirs than the fully connected network and the convolutional neural network. The prediction results are more consistent with the original logging curves than the convolutional neural network model and significantly better than the fully connected network model. In terms of stratigraphic morphology, the lateral distribution of the three models is consistent with the geological understanding of the study area, and the convolutional recurrent neural network is better able to reflect the lateral heterogeneity of the target layer.
[0104] Example 4
[0105] Figure 8 A block diagram of a dessert property prediction device based on a convolutional recurrent neural network according to an embodiment of the present invention is shown.
[0106] like Figure 8 As shown, the dessert property prediction device based on a convolutional recurrent neural network includes:
[0107] The sample building module 201 constructs learning samples using a "point-to-window" mode based on the characteristics of the network model.
[0108] Modeling module 202: Establish a convolutional recurrent network model and train the model using learning samples;
[0109] Prediction module 203 inputs the raw seismic data into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
[0110] In one example, based on the characteristics of the network model, the "point-to-window" pattern is used to construct learning samples, including:
[0111] After normalizing and transforming the data, the sweet spot physical property parameter curves are connected to their corresponding well-side seismic data through a "point-to-window" method, and the sequence samples formed in chronological order are used as learning samples.
[0112] In one example, the learning samples include a training sample set and a validation sample set.
[0113] In one example, building a convolutional recurrent network model includes:
[0114] The control variable method is used to determine the structural parameters of the convolutional recurrent network model, including the convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the order of the learning rate.
[0115] In one example, a convolutional recurrent network model is trained based on a training sample set.
[0116] In one example, the generalization ability of a convolutional recurrent network model is tested against a validation sample set.
[0117] Example 5
[0118] This embodiment provides an electronic device, which includes: a memory storing executable instructions; and a processor that executes the executable instructions in the memory to implement the above-described method for predicting the properties of desserts based on a convolutional recurrent neural network.
[0119] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.
[0120] This memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.
[0121] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory.
[0122] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.
[0123] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.
[0124] Example 6
[0125] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned dessert property prediction method based on a convolutional recurrent neural network.
[0126] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the methods described in the foregoing embodiments of the present disclosure are performed.
[0127] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).
[0128] Those skilled in the art should understand that the above description of the embodiments of the present invention is only intended to illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any of the examples given.
[0129] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.
Claims
1. A method for predicting the physical properties of desserts based on convolutional recurrent neural networks, characterized in that, include: Based on the characteristics of the network model, a "point-to-window" model is adopted to construct learning samples; A convolutional recurrent network model is established, and the model is trained using the aforementioned learning samples; The raw seismic data is input into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
2. The dessert property prediction method based on convolutional recurrent neural networks according to claim 1, wherein, Based on the characteristics of the network model, the "point-to-window" mode is used to construct learning samples, including: After normalizing and deep transforming the data, the sweet spot physical property parameter curves and their corresponding well-side seismic data are connected in a "point-to-window" manner, and the sequence samples formed in chronological order are the learning samples.
3. The dessert property prediction method based on convolutional recurrent neural networks according to claim 2, wherein, The learning samples include a training sample set and a validation sample set.
4. The dessert property prediction method based on convolutional recurrent neural networks according to claim 1, wherein, Building a convolutional recurrent network model includes: The control variable method is used to determine the structural parameters of the convolutional recurrent network model, including the convolutional structure, the combination of convolutional and pooling layers, the number of recursive layers, the number of fully connected layers, and the order of the learning rate.
5. The dessert property prediction method based on convolutional recurrent neural networks according to claim 3, wherein, The convolutional recurrent network model is trained based on the training sample set.
6. The dessert property prediction method based on convolutional recurrent neural networks according to claim 3, wherein, The generalization ability of the convolutional recurrent network model is tested using the validation sample set.
7. A dessert property prediction device based on a convolutional recurrent neural network, characterized in that, include: The sample building module constructs learning samples using a "point-to-window" mode, based on the characteristics of the network model. The modeling module establishes a convolutional recurrent network model and trains the model using the learning samples. The prediction module inputs the raw seismic data into the trained convolutional recurrent network model to complete the prediction of the sweet spot physical properties in the study area.
8. The dessert property prediction device based on convolutional recurrent neural network according to claim 7, wherein, Based on the characteristics of the network model, the "point-to-window" mode is used to construct learning samples, including: After normalizing and deep transforming the data, the sweet spot physical property parameter curves and their corresponding well-side seismic data are connected in a "point-to-window" manner, and the sequence samples formed in chronological order are the learning samples.
9. An electronic device, characterized in that, The electronic device includes: Memory, which stores executable instructions; A processor that executes the executable instructions in the memory to implement the dessert property prediction method based on a convolutional recurrent neural network as described in any one of claims 1-6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the dessert property prediction method based on a convolutional recurrent neural network as described in any one of claims 1-6.