Deep learning-based methods, devices, equipment, and media for predicting reservoir porosity.
By constructing a deep learning model and combining seismic and drilling data, and utilizing a combination of elastic properties sensitive to fluid changes, the problem of traditional methods being unable to accurately describe complex reservoirs was solved, achieving more efficient porosity prediction and distribution analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA PETROLEUM & CHEMICAL CORP
- Filing Date
- 2022-06-27
- Publication Date
- 2026-06-30
Smart Images
Figure CN117368966B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas exploration, and more specifically, to a method, apparatus, equipment, and medium for predicting reservoir porosity based on deep learning. Background Technology
[0002] Porosity is one of the most crucial reservoir parameters for subsurface characterization, playing a vital role in understanding fluid flow characteristics, rock elasticity and mechanical behavior, and pore pressure prediction. Reservoir porosity determination can be categorized into direct measurement methods (core analysis, wellbore coring, cuttings analysis) and indirect interpretation methods (seismic, well logging). The most direct and accurate method is to obtain core samples through drilling and then measure porosity through laboratory experiments. However, this method is costly and time-consuming, not suitable for all wells, and the difficulty of obtaining porosity increases with drilling depth. Among indirect interpretation methods, well logging data currently offers the highest resolution and best continuity of geological data. When core sampling is limited, using well logging data for porosity prediction is crucial and necessary. However, like direct measurement methods, it only provides information on the vertical variation of porosity near the well site, making it difficult to obtain the porosity distribution across the entire target area. While seismic data has lower resolution, it offers wide coverage and good lateral continuity, typically providing two-dimensional or three-dimensional seismic information covering the entire target reservoir area. Seismic data inversion allows for the inference and imaging of rock properties and subsurface structures. Therefore, seismic porosity prediction methods, which combine the high vertical resolution of well logging data with the lateral continuity of seismic data, are widely used.
[0003] Over the years, based on rock physics and seismic inversion, many targeted reservoir porosity prediction methods have been developed, such as sparse pulse inversion, Bayesian inversion, and geostatistical inversion. These methods all have certain applicable conditions, with both successful and unsuccessful cases. As exploration targets become increasingly complex, traditional reservoir porosity prediction methods based on linear assumptions, such as sparse pulse inversion, can no longer adequately meet the needs for detailed reservoir characterization. Therefore, it is necessary to establish nonlinear reservoir porosity prediction methods to address the problem of complex reservoir characterization.
[0004] Therefore, it is necessary to develop a method, device, equipment, and medium for predicting reservoir porosity based on deep learning of multimodal data.
[0005] 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
[0006] This invention proposes a method, device, equipment, and medium for predicting reservoir porosity based on deep learning. It constructs a combination of elastic properties that are sensitive to fluid changes to extract more information about the fluid, predicts lithology first and then porosity, thereby achieving reservoir porosity distribution prediction and potentially improving prediction performance.
[0007] In a first aspect, embodiments of this disclosure provide a reservoir porosity prediction method based on deep learning, comprising:
[0008] Construct sample data, wherein the sample data includes elastic data, lithological data, and porosity data;
[0009] Multiple combinations of elastic properties sensitive to fluid changes are constructed and stored in the elastic data in the sample data. The expanded elastic data includes the basic elastic data and the combinations of elastic properties.
[0010] Deep learning training is performed based on the expanded elastic data, the lithological data, and the porosity data to establish an elastic-lithological model and an elastic / lithological-porosity model, respectively.
[0011] The target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity.
[0012] Preferably, constructing sample data includes:
[0013] Acquire seismic and drilling data for the target area;
[0014] The seismic and drilling data are preprocessed.
[0015] Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation.
[0016] Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained.
[0017] The elasticity data, lithology data, and porosity data are scaled to construct a training set of the sample data.
[0018] Preferably, the training set of the sample data is constructed using data from the well locations.
[0019] Preferably, establishing the elastic-lithological model includes:
[0020] At least one of the elastic data is selected as input, and the lithological data is used as output for deep learning training to obtain an elastic-lithological model.
[0021] Preferably, establishing the elastic / lithological-porosity model includes:
[0022] By selecting at least one of the elastic data and the lithological data as inputs and porosity as output, deep learning training is performed to obtain an elastic / lithological-porosity model.
[0023] Preferably, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model, and the porosity prediction includes:
[0024] The first input data in the target data is determined based on the elastic-lithological model, and the first input data is input into the elastic-lithological model to obtain the corresponding target lithological data.
[0025] The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
[0026] Preferably, the deep learning training model is a hybrid network that combines convolutional neural networks and recurrent neural networks.
[0027] As one specific implementation of this disclosure,
[0028] Secondly, embodiments of this disclosure also provide a reservoir porosity prediction device based on deep learning, comprising:
[0029] The sample construction module constructs sample data, which includes elastic data, lithological data, and porosity data.
[0030] The combined construction module constructs multiple combinations of elastic properties that are sensitive to fluid changes, and stores them in the elastic data in the sample data. The expanded elastic data includes the basic elastic data and the combinations of elastic properties.
[0031] The model building module performs deep learning training based on the expanded elastic data, the lithological data, and the porosity data to establish an elastic-lithological model and an elastic / lithological-porosity model, respectively.
[0032] The prediction module inputs the target data sequentially into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity.
[0033] Preferably, constructing sample data includes:
[0034] Acquire seismic and drilling data for the target area;
[0035] The seismic and drilling data are preprocessed.
[0036] Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation.
[0037] Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained.
[0038] The elasticity data, lithology data, and porosity data are scaled to construct a training set of the sample data.
[0039] Preferably, the training set of the sample data is constructed using data from the well locations.
[0040] Preferably, establishing the elastic-lithological model includes:
[0041] At least one of the elastic data is selected as input, and the lithological data is used as output for deep learning training to obtain an elastic-lithological model.
[0042] Preferably, establishing the elastic / lithological-porosity model includes:
[0043] By selecting at least one of the elastic data and the lithological data as inputs and porosity as output, deep learning training is performed to obtain an elastic / lithological-porosity model.
[0044] Preferably, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model, and the porosity prediction includes:
[0045] The first input data in the target data is determined based on the elastic-lithological model, and the first input data is input into the elastic-lithological model to obtain the corresponding target lithological data.
[0046] The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
[0047] Preferably, the deep learning training model is a hybrid network that combines convolutional neural networks and recurrent neural networks.
[0048] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:
[0049] Memory, which stores executable instructions;
[0050] A processor that executes the executable instructions in the memory to implement the deep learning-based reservoir porosity prediction method.
[0051] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the deep learning-based reservoir porosity prediction method.
[0052] Its beneficial effects are as follows:
[0053] (1) Construct new combinations of elastic properties that are sensitive to fluid changes to extract more information about the fluid and potentially improve prediction performance;
[0054] (2) A deep learning model is built by using a hybrid network that combines the advantages of convolutional networks and recurrent networks to further improve the accuracy of prediction;
[0055] (3) By predicting lithology first and then porosity, reservoir porosity distribution can be predicted.
[0056] 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
[0057] 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.
[0058] Figure 1 A flowchart illustrating the steps of a deep learning-based reservoir porosity prediction method according to an embodiment of the present invention is shown.
[0059] Figure 2a , Figure 2b , Figure 2c The diagrams show a comparison between the porosity predicted by using a combination of basic elastic parameters + gamma data + elastic parameters, a combination of basic elastic parameters + elastic parameters, and basic elastic parameters as input, and the actual porosity, respectively, according to an embodiment of the present invention.
[0060] Figure 3 A schematic diagram of the two-dimensional cross-sectional porosity inversion result is shown according to an embodiment of the present invention.
[0061] Figure 4 A block diagram of a deep learning-based reservoir porosity prediction device according to an embodiment of the present invention is shown.
[0062] Explanation of reference numerals in the attached figures:
[0063] 201. Sample construction module; 202. Combination construction module; 203. Model construction module; 204. Prediction module. Detailed Implementation
[0064] 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.
[0065] This invention provides a deep learning-based method for predicting reservoir porosity, comprising:
[0066] Construct sample data, which includes elasticity data, lithology data, and porosity data;
[0067] Multiple combinations of elastic properties sensitive to fluid changes are constructed and stored in the elastic data of the sample data. The expanded elastic data includes basic elastic data and combinations of elastic properties.
[0068] Deep learning training was conducted based on the expanded elasticity data, lithology data, and porosity data to establish elastic-lithology models and elastic / lithology-porosity models, respectively.
[0069] The target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity.
[0070] In one example, constructing the sample data includes:
[0071] Acquire seismic and drilling data for the target area;
[0072] Preprocessing of seismic and drilling data;
[0073] Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation.
[0074] Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained.
[0075] Scale transformations were performed on elasticity data, lithology data, and porosity data to construct a training set of sample data.
[0076] In one example, sample data is constructed using data from well locations.
[0077] In one example, establishing an elastic-lithological model includes:
[0078] By selecting at least one type of elastic data as input and using lithological data as output, deep learning training is performed to obtain an elastic-lithological model.
[0079] In one example, establishing an elastic / lithological-porosity model includes:
[0080] By selecting at least one type of elastic data and lithological data as inputs and porosity as output, deep learning training is performed to obtain an elastic / lithological-porosity model.
[0081] In one example, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity, including:
[0082] The first input data in the target data is determined based on the elastic-lithology model. The first input data is then input into the elastic-lithology model to obtain the corresponding target lithology data.
[0083] The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
[0084] In one example, the deep learning-trained model is a hybrid network that combines convolutional neural networks and recurrent neural networks.
[0085] Specifically, seismic and drilling data of the target area are acquired; the acquired seismic and drilling data are preprocessed to eliminate various random interferences and non-formation factors as much as possible, so as to reflect the properties of the formation and its pore fluids as realistically as possible; the preprocessing includes processing the acquired seismic data such as trace editing, noise reduction, and multiple suppression, and processing the acquired well logging data such as outlier removal, correction, and filtering.
[0086] Using drilling data and synthetic seismic records, target stratigraphic positions are accurately determined based on structural interpretation. Stratigraphic position determination is the foundation of structural interpretation of seismic data and seismic reservoir research, and it is also a bridge connecting well logging and seismic information. The accuracy of the determination results directly determines the accuracy of reservoir prediction.
[0087] Based on preprocessed seismic and drilling data, corresponding elasticity, lithology, and porosity data are obtained according to specific needs, resulting in sample data. Specifically, sample data acquisition involves extracting seismic attributes or performing seismic inversion on the processed seismic trace data to obtain multi-attribute data, including seismic attributes and elastic parameters. Attribute optimization is then performed through clustering or empirical selection. Elasticity data is obtained from processed drilling data, and porosity data at corresponding well locations is obtained through analysis and calculation of core samples and cuttings, followed by scaling. Specifically, the scale transformation involves converting the logging data to the time domain using a depth-time conversion. After depth-time conversion, the elasticity, lithology, and porosity data are smoothed, and then downsampled to have the same magnitude as the seismic attributes, capturing the original porosity variation trend and reducing high-frequency fluctuations.
[0088] Building upon this foundation, we employ simple mathematical operations and domain knowledge to construct new combinations of elastic properties sensitive to fluid variations. Here, we use three elastic data points—longitudinal wave (Vp), transverse wave (Vs), and density (DEN)—as the basic elastic data to construct several combinations of elastic properties: A = Vp * DEN, B = Vs * DEN, C = Vp + Vs, D = Vp - Vs, E = Vp * Vs, F = Vp / Vs, and G = Vp. 2 -2Vs 2 The equations are: A, B, C, D, E, and F, and H = Vp sin160° + Vs cos160°. Specifically, A and B are simple mathematical combinations of Vp, Vs, and DEN; C, D, E, and F are simple mathematical combinations of Vp and Vs; H and G are features constructed based on domain knowledge, where H is a fluid indicator factor, typically used to indicate fluid distribution based on seismic inversion results; and G is a combination of elastic properties constructed based on rotation angles. It should be noted that the above elastic property combinations are only simple combinations constructed for this example; in practical applications, more effective and complex combinations need to be constructed according to the actual situation.
[0089] Correlation analysis of basic elasticity data and combinations of elasticity attributes with the target variable reveals that, with the help of the construction and interaction of elasticity attribute combinations, some elasticity data with poor correlation performance can become more relevant to the target variable when combined with other elasticity data. Therefore, a reasonable combination of basic features can compensate for the shortcomings of existing features. However, it should be noted that the feature importance ranking derived from correlation analysis only reflects the simple correlation between the input features and the target data and cannot guarantee its impact on actual predictive performance.
[0090] A deep learning model is obtained by selecting at least one type of elastic data as input and lithological data as output. Alternatively, a deep learning model is obtained by selecting at least one type of elastic data and lithological data as input and porosity as output. The constructed models are then iteratively optimized using a backpropagation algorithm until the set requirements are met.
[0091] The deep learning-trained model is a hybrid network that combines the advantages of convolutional neural networks and recurrent neural networks. It extracts local morphological features of input feature attributes through convolutional networks, extracts vertical correlation features of input feature attributes through recurrent networks, and finally achieves intelligent prediction of lithology and reservoir porosity through fully connected layers.
[0092] The first input data in the target data is determined based on the elastic-lithology model. This first input data is then input into the elastic-lithology model to obtain the corresponding target lithology data. The second input data in the target data is determined based on the elastic / lithology-porosity model. This second input data, along with the target lithology data, is then input into the elastic / lithology-porosity model to obtain the corresponding porosity, thus achieving porosity prediction. After the training process, the weights and biases of the network model are determined. As long as the data distribution of lithology and reservoir parameters is similar, the trained system can be immediately applied to other regions.
[0093] The present invention also provides a reservoir porosity prediction device based on deep learning, comprising:
[0094] The sample construction module constructs sample data, which includes elastic data, lithological data, and porosity data.
[0095] The combined construction module constructs multiple combinations of elastic properties that are sensitive to fluid changes and stores them in the elastic data of the sample data. The expanded elastic data includes basic elastic data and combinations of elastic properties.
[0096] The model building module uses deep learning training based on the expanded elastic data, lithological data, and porosity data to establish elastic-lithological models and elastic / lithological-porosity models, respectively.
[0097] The prediction module inputs the target data sequentially into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity.
[0098] In one example, constructing the sample data includes:
[0099] Acquire seismic and drilling data for the target area;
[0100] Preprocessing of seismic and drilling data;
[0101] Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation.
[0102] Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained.
[0103] Scale transformations were performed on elasticity data, lithology data, and porosity data to construct a training set of sample data.
[0104] In one example, a training set of sample data is constructed using data from well locations.
[0105] In one example, establishing an elastic-lithological model includes:
[0106] By selecting at least one type of elastic data as input and using lithological data as output, deep learning training is performed to obtain an elastic-lithological model.
[0107] In one example, establishing an elastic / lithological-porosity model includes:
[0108] By selecting at least one type of elastic data and lithological data as inputs and porosity as output, deep learning training is performed to obtain an elastic / lithological-porosity model.
[0109] In one example, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity, including:
[0110] The first input data in the target data is determined based on the elastic-lithology model. The first input data is then input into the elastic-lithology model to obtain the corresponding target lithology data.
[0111] The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
[0112] In one example, the deep learning-trained model is a hybrid network that combines convolutional neural networks and recurrent neural networks.
[0113] Specifically, seismic and drilling data of the target area are acquired; the acquired seismic and drilling data are preprocessed to eliminate various random interferences and non-formation factors as much as possible, so as to reflect the properties of the formation and its pore fluids as realistically as possible; the preprocessing includes processing the acquired seismic data such as trace editing, noise reduction, and multiple suppression, and processing the acquired well logging data such as outlier removal, correction, and filtering.
[0114] Using drilling data and synthetic seismic records, target stratigraphic positions are accurately determined based on structural interpretation. Stratigraphic position determination is the foundation of structural interpretation of seismic data and seismic reservoir research, and it is also a bridge connecting well logging and seismic information. The accuracy of the determination results directly determines the accuracy of reservoir prediction.
[0115] Based on preprocessed seismic and drilling data, corresponding elasticity, lithology, and porosity data are obtained according to specific needs, resulting in sample data. Specifically, sample data acquisition involves extracting seismic attributes or performing seismic inversion on the processed seismic trace data to obtain multi-attribute data, including seismic attributes and elastic parameters. Attribute optimization is then performed through clustering or empirical selection. Elasticity data is obtained from processed drilling data, and porosity data at corresponding well locations is obtained through analysis and calculation of core samples and cuttings, followed by scaling. Specifically, the scale transformation involves converting the logging data to the time domain using a depth-time conversion. After depth-time conversion, the elasticity, lithology, and porosity data are smoothed, and then downsampled to have the same magnitude as the seismic attributes, capturing the original porosity variation trend and reducing high-frequency fluctuations.
[0116] Building upon this foundation, we employ simple mathematical operations and domain knowledge to construct new combinations of elastic properties sensitive to fluid variations. Here, we use three elastic data points—longitudinal wave (Vp), transverse wave (Vs), and density (DEN)—as the basic elastic data to construct several combinations of elastic properties: A = Vp * DEN, B = Vs * DEN, C = Vp + Vs, D = Vp - Vs, E = Vp * Vs, F = Vp / Vs, and G = Vp. 2 -2Vs 2 The equations are: A, B, C, D, E, and F, and H = Vp sin160° + Vs cos160°. Specifically, A and B are simple mathematical combinations of Vp, Vs, and DEN; C, D, E, and F are simple mathematical combinations of Vp and Vs; H and G are features constructed based on domain knowledge, where H is a fluid indicator factor, typically used to indicate fluid distribution based on seismic inversion results; and G is a combination of elastic properties constructed based on rotation angles. It should be noted that the above elastic property combinations are only simple combinations constructed for this example; in practical applications, more effective and complex combinations need to be constructed according to the actual situation.
[0117] Correlation analysis of basic elasticity data and combinations of elasticity attributes with the target variable reveals that, with the help of the construction and interaction of elasticity attribute combinations, some elasticity data with poor correlation performance can become more relevant to the target variable when combined with other elasticity data. Therefore, a reasonable combination of basic features can compensate for the shortcomings of existing features. However, it should be noted that the feature importance ranking derived from correlation analysis only reflects the simple correlation between the input features and the target data and cannot guarantee its impact on actual predictive performance.
[0118] A deep learning model is obtained by selecting at least one type of elastic data as input and lithological data as output. Alternatively, a deep learning model is obtained by selecting at least one type of elastic data and lithological data as input and porosity as output. The constructed models are then iteratively optimized using a backpropagation algorithm until the set requirements are met.
[0119] The deep learning-trained model is a hybrid network that combines the advantages of convolutional neural networks and recurrent neural networks. It extracts local morphological features of input feature attributes through convolutional networks, extracts vertical correlation features of input feature attributes through recurrent networks, and finally achieves intelligent prediction of lithology and reservoir porosity through fully connected layers.
[0120] The first input data in the target data is determined based on the elastic-lithology model. This first input data is then input into the elastic-lithology model to obtain the corresponding target lithology data. The second input data in the target data is determined based on the elastic / lithology-porosity model. This second input data, along with the target lithology data, is then input into the elastic / lithology-porosity model to obtain the corresponding porosity, thus achieving porosity prediction. After the training process, the weights and biases of the network model are determined. As long as the data distribution of lithology and reservoir parameters is similar, the trained system can be immediately applied to other regions.
[0121] The present invention also provides an electronic device, comprising: a memory storing executable instructions; and a processor executing the executable instructions in the memory to implement the above-described deep learning-based reservoir porosity prediction method.
[0122] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described deep learning-based reservoir porosity prediction method.
[0123] To facilitate understanding of the solutions and effects of the embodiments of the present invention, four 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.
[0124] Example 1
[0125] Figure 1 A flowchart illustrating the steps of the deep learning-based reservoir porosity prediction method according to the present invention is shown.
[0126] like Figure 1As shown, the deep learning-based reservoir porosity prediction method includes: Step 101, constructing sample data, which includes elastic data, lithological data, and porosity data; Step 102, constructing multiple combinations of elastic properties sensitive to fluid changes and storing them in the elastic data of the sample data, wherein the expanded elastic data includes basic elastic data and elastic property combinations; Step 103, performing deep learning training based on the expanded elastic data, lithological data, and porosity data to establish an elastic-lithological model and an elastic / lithological-porosity model, respectively; Step 104, sequentially inputting the target data into the elastic-lithological model and the elastic / lithological-porosity model to predict porosity.
[0127] This invention was applied to the prediction of reservoir porosity in a certain exploration area, which is a tight sandstone reservoir.
[0128] First, seismic and drilling data for the target area are acquired. The seismic data undergoes processing such as trace editing, denoising, and multiple suppression. The drilling data is processed by outlier removal, correction, and filtering. Using drilling data and synthetic seismic records, the target stratigraphic level is accurately determined based on structural interpretation. Seismic attributes are extracted or inverted from the processed seismic data to obtain multi-attribute data. Elastic data is obtained from the processed drilling data, and porosity data at the corresponding well locations is obtained through analysis of core samples and cuttings.
[0129] The obtained well logging elastic data, gamma data (using gamma data to replace lithological data), and porosity data are scaled and transformed. The data are then converted to the time domain using time-depth conversion. After depth-time conversion, the elastic data, gamma data, and porosity data are smoothed and then downsampled to have the same size as the seismic properties. This allows the original porosity change trend to be captured and high-frequency fluctuations to be reduced.
[0130] Building upon this foundation, we employ simple mathematical operations and domain knowledge to construct new combinations of elastic properties sensitive to fluid variations. Here, we use three elastic data points—longitudinal wave (Vp), transverse wave (Vs), and density (DEN)—as the basic elastic data to construct several combinations of elastic properties: A = Vp * DEN, B = Vs * DEN, C = Vp + Vs, D = Vp - Vs, E = Vp * Vs, F = Vp / Vs, and G = Vp. 2 -2Vs 2And H = Vp sin160° + Vs cos160°. Specifically, A and B are simple mathematical combinations of Vp, Vs, and DEN; C, D, E, and F are simple mathematical combinations of Vp and Vs; H and G are features constructed based on domain knowledge, where H is a fluid indicator factor, typically used to indicate fluid distribution based on seismic inversion results; and G is a combination of elastic properties constructed based on rotation angles. It should be noted that the above combination of elastic properties is only a simple combination constructed for this example; in practical applications, more effective and complex combinations need to be constructed according to the actual situation.
[0131] Correlation analysis of basic elastic data and elastic attribute combinations with the target variable reveals that, with the help of elastic attribute combination construction and interaction, some elastic data with poor correlation performance can become more relevant to the target variable when combined with other elastic data. Therefore, a reasonable combination of basic features can compensate for the shortcomings of existing features. However, it should be noted that the feature importance ranking derived from correlation analysis only reflects the simple correlation between the input features and the target data and cannot guarantee its impact on actual prediction performance. The final input features need to be selected based on actual testing. Based on correlation analysis and data testing, the basic elastic data Vp, Vs, DEN and elastic data combinations A, B, C, E, H are selected as inputs, with gamma as the output to construct an elastic-lithology model. Alternatively, the basic elastic data Vp, Vs, DEN and elastic data combinations B, C, E, H and gamma are selected as inputs, with porosity as the output to construct an elastic / lithology-porosity model.
[0132] In gamma (lithology) prediction: the model is trained using elastic data as input and well point gamma as output. Then, the inverted basic elastic data and corresponding elastic data combinations are used as input, and the output is the predicted gamma data. In porosity prediction: the model is trained using selected elastic and gamma data as input and well point porosity as output. Then, the predicted gamma data, inverted basic elastic data, and corresponding elastic data combinations are used as input to complete the seismic prediction of porosity spatial distribution. The constructed deep network consists of four convolutional network layers, three recurrent network layers, and one fully connected layer.
[0133] Figure 2a , Figure 2b , Figure 2cThe figures show schematic diagrams comparing the predicted porosity with the actual porosity using a combination of basic elastic parameters + gamma data + elastic parameters, a combination of basic elastic parameters + elastic parameters, and the basic elastic parameters as input, respectively, according to an embodiment of the present invention. The thick solid line in the figures represents the actual porosity, and the thin solid line represents the predicted porosity. It can be seen that the porosity prediction result using the combination of basic elastic parameters + gamma data + elastic parameters as input is closer to the actual porosity, demonstrating the effectiveness and applicability of the proposed method.
[0134] Figure 3 A schematic diagram of the two-dimensional profile porosity inversion results according to an embodiment of the present invention is shown. The black line in the figure represents the logging porosity. It can be seen that the inverted porosity is in good agreement with the logging porosity, indicating that the method can achieve good practical application results in the problem of reservoir porosity prediction.
[0135] Example 2
[0136] Figure 4 A block diagram of a deep learning-based reservoir porosity prediction device according to an embodiment of the present invention is shown.
[0137] like Figure 4 As shown, the deep learning-based reservoir porosity prediction device includes:
[0138] Sample construction module 201 constructs sample data, which includes elastic data, lithological data, and porosity data.
[0139] The combined construction module 202 constructs multiple combinations of elastic properties that are sensitive to fluid changes and stores them in the elastic data of the sample data. The expanded elastic data includes basic elastic data and combinations of elastic properties.
[0140] The model building module 203 performs deep learning training based on the expanded elastic data, lithological data, and porosity data to establish an elastic-lithological model and an elastic / lithological-porosity model, respectively.
[0141] The prediction module 204 inputs the target data into the elastic-lithology model and the elastic / lithology-porosity model in sequence to predict porosity.
[0142] As an optional approach, the construction of sample data includes:
[0143] Acquire seismic and drilling data for the target area;
[0144] Preprocessing of seismic and drilling data;
[0145] Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation.
[0146] Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained.
[0147] Scale transformations were performed on elasticity data, lithology data, and porosity data to construct a training set of sample data.
[0148] As an alternative, a training set of sample data can be constructed using data from the well locations.
[0149] As an optional approach, establishing an elastic-lithological model includes:
[0150] By selecting at least one type of elastic data as input and using lithological data as output, deep learning training is performed to obtain an elastic-lithological model.
[0151] As an optional approach, establishing an elastic / lithological-porosity model includes:
[0152] By selecting at least one type of elastic data and lithological data as inputs and porosity as output, deep learning training is performed to obtain an elastic / lithological-porosity model.
[0153] As an optional approach, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity, including:
[0154] The first input data in the target data is determined based on the elastic-lithology model. The first input data is then input into the elastic-lithology model to obtain the corresponding target lithology data.
[0155] The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
[0156] As an alternative, the model trained by deep learning is a hybrid network that combines convolutional neural networks and recurrent neural networks.
[0157] Example 3
[0158] This disclosure provides an electronic device comprising: a memory storing executable instructions; and a processor executing the executable instructions in the memory to implement the aforementioned deep learning-based reservoir porosity prediction method.
[0159] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.
[0160] 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.
[0161] 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.
[0162] 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.
[0163] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.
[0164] Example 4
[0165] This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the deep learning-based reservoir porosity prediction method.
[0166] 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.
[0167] 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).
[0168] 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.
[0169] 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 reservoir porosity based on deep learning, characterized in that, include: Construct sample data, wherein the sample data includes elastic data, lithological data, and porosity data; Multiple combinations of elastic properties sensitive to fluid changes are constructed and stored in the elastic data in the sample data, wherein the expanded elastic data includes the basic elastic data and the combinations of elastic properties; Deep learning training is performed based on the expanded elastic data, the lithological data, and the porosity data to establish an elastic-lithological model and an elastic / lithological-porosity model, respectively. The target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity; The construction of sample data includes: Acquire seismic and drilling data for the target area; The seismic and drilling data are preprocessed. Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation. Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained. Scale transformations are performed on elastic data, lithological data, and porosity data to construct a training set for the sample data. Specifically, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity, including: The first input data in the target data is determined based on the elastic-lithological model, and the first input data is input into the elastic-lithological model to obtain the corresponding target lithological data. The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
2. The reservoir porosity prediction method based on deep learning according to claim 1, wherein, The training set of the sample data is constructed using data from the well locations.
3. The reservoir porosity prediction method based on deep learning according to claim 1, wherein, Establishing the elastic-lithological model includes: At least one of the elastic data is selected as input, and the lithological data is used as output for deep learning training to obtain an elastic-lithological model.
4. The reservoir porosity prediction method based on deep learning according to claim 1, wherein, Establishing the elastic / lithological-porosity model includes: By selecting at least one of the elastic data and the lithological data as inputs and porosity as output, deep learning training is performed to obtain an elastic / lithological-porosity model.
5. The reservoir porosity prediction method based on deep learning according to claim 1, wherein, The deep learning training model is a hybrid network that combines convolutional neural networks and recurrent neural networks.
6. A reservoir porosity prediction device based on deep learning, characterized in that, include: The sample construction module constructs sample data, which includes elastic data, lithological data, and porosity data. The combined construction module constructs multiple combinations of elastic properties that are sensitive to fluid changes, and stores them in the elastic data in the sample data. The expanded elastic data includes the basic elastic data and the combinations of elastic properties. The model building module performs deep learning training based on the expanded elastic data, the lithological data, and the porosity data to establish an elastic-lithological model and an elastic / lithological-porosity model, respectively. The prediction module inputs the target data sequentially into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity; The construction of sample data includes: Acquire seismic and drilling data for the target area; The seismic and drilling data are preprocessed. Based on drilling data and synthetic seismic records, the target stratigraphic level is accurately determined according to tectonic interpretation. Based on the preprocessed seismic and drilling data, elastic data, lithological data, and porosity data are obtained. Scale transformations are performed on elastic data, lithological data, and porosity data to construct a training set for the sample data. Specifically, the target data is sequentially input into the elastic-lithology model and the elastic / lithology-porosity model to predict porosity, including: The first input data in the target data is determined based on the elastic-lithological model, and the first input data is input into the elastic-lithological model to obtain the corresponding target lithological data. The second input data in the target data is determined based on the elastic / lithology-porosity model. The second input data and the target lithology data are then input into the elastic / lithology-porosity model to obtain the corresponding porosity.
7. 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 deep learning-based reservoir porosity prediction method according to any one of claims 1-5.
8. 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 deep learning-based reservoir porosity prediction method according to any one of claims 1-5.