A pre-stack seismic inversion method and device

Abstract

The present invention provides a prestack seismic inversion method and apparatus, comprising: dividing the reservoirs of a study area based on prior information to determine regional classification results; determining label information based on a stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information; classifying and labeling the label information using the regional classification results to determine labeled data with classification labels; constructing a convolutional neural network; training the convolutional neural network using the labeled data with classification labels to determine a trained convolutional neural network; and performing prestack seismic inversion based on the trained convolutional neural network to determine elastic parameter inversion results. By leveraging the powerful nonlinear problem-solving capabilities of convolutional neural networks, the present invention improves the accuracy of prestack seismic inversion and obtains more accurate elastic parameter prediction results.

Description

Technical Field

[0001] The present invention relates to the field of geophysical technology, and in particular to a pre-stack seismic inversion method and device. Background Art

[0002] This section is intended to provide a background or context to embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by inclusion in this section.

[0003] Seismic waves propagate underground and are reflected when they encounter reflecting interfaces. When seismic waves are incident non-perpendicularly on a reflecting interface, the amplitude of the reflected wave varies with the angle of incidence. By studying how the seismic wave amplitude varies with the angle of incidence, we can derive a variety of elastic parameters reflecting subsurface reservoir properties from pre-stack seismic data, which contain rich dynamic information. This provides a quantitative basis for reservoir prediction and hydrocarbon detection. Currently, the inversion algorithms studied and applied by researchers and industry are based on the Zoeppritz equation and its approximations. The Zoeppritz equation describes the energy distribution of elastic waves at elastic interfaces.

[0004] However, the application of the theoretical equation is based on the assumption that the underground medium is isotropic and horizontally layered, and that there is semi-infinite space on both sides of the interface. However, the actual underground medium distribution is complex and very different from the assumed situation. The theoretical equation cannot accurately characterize the variation of the reflection wave amplitude with the incident angle, resulting in certain errors in the application of the theoretical equation to the pre-stack AVO inversion method for elastic parameter prediction.

[0005] Therefore, the theoretical equations cannot accurately describe the propagation laws of seismic waves under complex geological conditions and the strong nonlinearity of the inversion problem.

[0006] Therefore, how to provide a new solution that can solve the above technical problems is a technical problem that needs to be solved urgently in this field. Summary of the Invention

[0007] An embodiment of the present invention provides a prestack seismic inversion method that utilizes the powerful ability of convolutional neural networks to solve nonlinear problems, thereby improving the accuracy of prestack seismic inversion and obtaining more accurate elastic parameter prediction results. The method includes:

[0008] Based on prior information, the reservoirs in the study area are divided and the regional classification results are determined;

[0009] Determine label information using a stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information.

[0010] Using the regional classification results to classify and mark the label information, and determine the label data with classification marks;

[0011] Build a convolutional neural network;

[0012] Using labeled data with classification marks to train a convolutional neural network, and determining the trained convolutional neural network;

[0013] Based on the trained convolutional neural network, pre-stack seismic inversion is performed to determine the elastic parameter inversion results.

[0014] An embodiment of the present invention further provides a pre-stack seismic inversion device, comprising:

[0015] The reservoir division module is used to divide the reservoirs in the study area according to prior information and determine the regional classification results;

[0016] a label information determination module, for determining label information based on a random geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information;

[0017] A classification and marking module is used to classify and mark label information using the regional classification results to determine label data with classification marks;

[0018] Convolutional neural network building module, used to build convolutional neural networks;

[0019] A convolutional neural network training module is used to train the convolutional neural network using labeled data with classification labels and determine the trained convolutional neural network;

[0020] The pre-stack seismic inversion module is used to perform pre-stack seismic inversion based on the trained convolutional neural network and determine the elastic parameter inversion results.

[0021] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned pre-stack seismic inversion method when executing the computer program.

[0022] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program for executing the above-mentioned pre-stack seismic inversion method.

[0023] A prestack seismic inversion method and apparatus provided by an embodiment of the present invention includes: first, dividing the reservoirs in the study area based on prior information to determine regional classification results; then, determining label information based on a stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information; then, classifying and labeling the label information using the regional classification results to determine labeled data with classification labels; then, building a convolutional neural network; then, training the convolutional neural network using the labeled data with classification labels to determine a trained convolutional neural network; and finally, performing prestack seismic inversion based on the trained convolutional neural network to determine elastic parameter inversion results. By fully utilizing prior information to constrain prestack AVO three-parameter inversion based on a convolutional neural network, the convolutional neural network no longer simply uses the convolutional neural network to construct a nonlinear relationship between prestack common reflection point gathers and the elastic parameters to be inverted. The prior information constraints can narrow the solution space of the optimization problem, thereby improving the stability and accuracy of the optimization problem. Furthermore, the labeled data obtained through the stochastic geological simulation method provides a large amount of labeled data for network training, which facilitates network model optimization, improves the network's generalization ability, and thus enhances the network's ultimate predictive capability. The embodiments of the present invention improve the accuracy of pre-stack seismic inversion and obtain more accurate elastic parameter prediction results by leveraging the powerful ability of convolutional neural networks to solve nonlinear problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0025] Figure 1 Schematic diagram of a pre-stack seismic inversion method according to an embodiment of the present invention.

[0026] Figure 2 The present invention provides a flowchart of a pre-stack seismic inversion method.

[0027] Figure 3 The figure is a schematic diagram of a label data generation process of a prestack seismic inversion method according to an embodiment of the present invention.

[0028] Figure 4 A schematic diagram of the convolutional neural network structure of a prestack seismic inversion method according to an embodiment of the present invention.

[0029] Figure 5 This is a schematic diagram of how errors change with increasing iteration times during a network training process of a prestack seismic inversion method according to an embodiment of the present invention.

[0030] Figure 6 The figure is a schematic diagram of inversion results of a pre-stack seismic inversion method according to an embodiment of the present invention.

[0031] Figure 7 A schematic diagram of a computer device for running a prestack seismic inversion method implemented in the present invention.

[0032] Figure 8 The figure is a schematic diagram of a pre-stack seismic inversion device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0034] Figure 1 FIG. 1 is a schematic diagram of a pre-stack seismic inversion method according to an embodiment of the present invention, as shown in FIG. Figure 1 As shown, an embodiment of the present invention provides a prestack seismic inversion method, which makes use of the powerful ability of convolutional neural networks to solve nonlinear problems, improves the accuracy of prestack seismic inversion, and obtains more accurate elastic parameter prediction results. The method includes:

[0035] Step S01: Divide the reservoirs in the study area according to prior information and determine the regional classification results;

[0036] Step S02: Determine label information based on a random geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information;

[0037] Step S03: using the regional classification result to classify and mark the label information, and determine the label data with the classification mark;

[0038] Step S04: building a convolutional neural network;

[0039] Step S05: training a convolutional neural network using the labeled data with classification marks, and determining a trained convolutional neural network;

[0040] Step S06: Perform pre-stack seismic inversion based on the trained convolutional neural network to determine the elastic parameter inversion results.

[0041] A prestack seismic inversion method provided by an embodiment of the present invention includes: first, dividing the reservoir of the study area based on prior information to determine a regional classification result; then, determining label information based on a stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information; then, classifying and labeling the label information using the regional classification results to determine labeled data with classification labels; next, building a convolutional neural network; and next, training the convolutional neural network using the labeled data with classification labels to determine a trained convolutional neural network; finally, performing prestack AVO (Amplitude Versus Offset) three-parameter inversion in prestack seismic inversion based on the trained convolutional neural network to determine elastic parameter inversion results. By fully utilizing prior information to constrain the prestack AVO three-parameter inversion based on the convolutional neural network, it no longer simply uses the convolutional neural network to construct a nonlinear relationship between prestack common reflection point gathers and the elastic parameters to be inverted. The prior information constraints can narrow the solution space of the optimization problem and improve the stability and accuracy of the optimization problem. Furthermore, the labeled data obtained through random geological simulation methods provides a large amount of labeled data for network training, which is beneficial for optimizing the network model and improving the network's generalization ability, thereby enhancing the network's ultimate predictive capabilities. The present invention utilizes the powerful ability of convolutional neural networks to solve nonlinear problems, thereby improving the accuracy of prestack AVO (Amplitude Versus Offset) three-parameter inversion and obtaining more accurate elastic parameter prediction results.

[0042] Combine Figure 2 The process shown specifically implements a prestack seismic inversion method provided by an embodiment of the present invention, which may include: dividing the reservoirs in the study area according to prior information to determine the regional classification results; determining label information according to a random geological simulation method constrained by rock physical information, prior information, regional classification results, and logging information; classifying and labeling the label information using the regional classification results to determine label data with classification labels; building a convolutional neural network; training the convolutional neural network using the label data with classification labels to determine the trained convolutional neural network; performing prestack seismic inversion based on the trained convolutional neural network to determine the elastic parameter inversion results.

[0043] Convolutional neural networks (CNNs), a deep learning method, can continuously optimize their network structure through training and establish nonlinear relationships between input and output. CNNs can extract high-dimensional features and automatically recognize patterns, enabling them to solve highly nonlinear problems such as image classification, image segmentation, object recognition, and natural language processing. In recent years, CNNs have also seen initial application in seismic exploration, such as fault identification, seismic phase classification, and noise processing. Given the inability of theoretical equations to accurately describe seismic wave propagation in complex geological conditions and the strong nonlinearity of the prestack AVO (Amplitude Versus Offset) three-parameter inversion problem, a CNN algorithm is proposed to construct a data-driven nonlinear relationship between prestack gathers and the elastic parameters to be inverted. This approach mitigates the inversion errors caused by the inaccuracies of mathematical equations describing seismic wave propagation in the subsurface, ultimately achieving high-precision prestack AVO three-parameter inversion. The prestack AVO (Amplitude Versus Offset) inversion method, a type of prestack seismic inversion method, can achieve more accurate inversion predictions of elastic parameters.

[0044] This embodiment of the present invention proposes a method for pre-stack Amplitude Versus Offset (AVO) inversion prediction of P-wave velocity, S-wave velocity, and density using a convolutional neural network (CNN) algorithm within a deep learning approach. Leveraging the CNN's powerful ability to construct nonlinear mapping relationships between input and output, a nonlinear mapping relationship is established between pre-stack seismic common reflection point gathers and the three parameters of P-wave velocity, S-wave velocity, and density, forming a data-driven pre-stack AVO (Amplitude Versus Offset) three-parameter inversion method.

[0045] When implementing a prestack seismic inversion method provided by an embodiment of the present invention, in one embodiment, the reservoirs in the study area are divided according to the prior information, and regional classification results are determined, including:

[0046] Based on prior information consisting of geological structure information, sedimentary information, logging data analysis results, seismic attribute analysis results and single well lithofacies information, the reservoirs in the study area are divided into favorable reservoir areas and unfavorable reservoir areas, and the regional classification results are determined; among them, the regional classification results include: the first type of favorable reservoir development area, the second type of moderately developed reservoir area, and the third type of underdeveloped reservoir area.

[0047] In the embodiment, the research target is divided into Class I reservoir favorable development areas, Class II reservoir moderate development areas, and Class III reservoir underdeveloped areas based on regional structure, sedimentary information, logging data analysis, and seismic attribute analysis as prior information, and the pre-stack common reflection point gather data are classified and marked.

[0048] Figure 3 FIG. 1 is a schematic diagram of a label data generation process of a pre-stack seismic inversion method according to an embodiment of the present invention. Figure 3 As shown, when implementing a pre-stack seismic inversion method provided by an embodiment of the present invention, in one embodiment, the aforementioned random geological simulation method based on petrophysical information, prior information, regional classification results, and well logging information constraints determines label information, including:

[0049] Analyze rock physical information and determine core measurement data;

[0050] Analyze core measurement data, combine it with well logging information, conduct rock physics model tests, and establish a rock physics model that suits the study area;

[0051] Obtain geological structure and sedimentary information from prior information, integrate geological and geophysical information, based on regional classification results, use the lithofacies probability distribution of different classification areas, and combine random geological simulation methods to obtain different types of virtual well curves;

[0052] Based on the virtual well curve, rock physics model is used to build a model and determine the simulated elastic parameters;

[0053] Using simulated elastic parameters, forward modeling of seismic wave fields is performed to determine pre-stack common reflection point gathers;

[0054] The pre-stack common reflection point gathers and simulated elastic parameters are correspondingly combined to determine the label data.

[0055] In this embodiment, the CNN algorithm, a type of deep neural network algorithm, is a supervised learning algorithm that requires a large amount of data labels to train the network and optimize network model parameters, ultimately obtaining an optimized network model. This trained and optimized network is then used for prediction. In the field of oil exploration, well data obtained through well logging can be used as labeled data. However, an exploration block has only a few wells in the early stages of exploration, so the amount of labeled data obtained is insufficient to meet the training requirements of the CNN network. This embodiment proposes a random geological simulation method that utilizes petrophysical information, geological structure and sedimentary information, and well logging information to generate labeled data.

[0056] First, analyze the rock physical information and determine the core measurement data;

[0057] By analyzing the core measurement data and combining it with the logging information of the drilled wells, including P-wave velocity, S-wave velocity, density, porosity, water saturation, mud content and other information, rock physics model tests are carried out to optimize and construct a rock physics model that is suitable for the study area.

[0058] Based on the regional classification results obtained above, the lithofacies probability distribution of different classification areas was utilized in conjunction with stochastic geological simulation methods to generate different types of virtual well curves, including lithofacies, porosity, shale content, and saturation. Finally, the rock physics model constructed and selected earlier was used to generate the corresponding simulated elastic parameters, including P-wave velocity, S-wave velocity, and density.

[0059] Finally, using the elastic parameters obtained from the simulation, wavefield forward modeling is performed to generate prestack common reflection point gathers. Each prestack common reflection point gather corresponds to a set of elastic parameter curves (P-wave velocity curves, S-wave velocity curves, and density curves), which together form a set of labeled data. Furthermore, the labeled data are classified and labeled according to the virtual well type used in the forward modeling records, so that the generated labeled data are clearly classified.

[0060] Figure 4 FIG. 1 is a schematic diagram of a convolutional neural network structure of a prestack seismic inversion method according to an embodiment of the present invention, as shown in FIG. Figure 4 As shown, when implementing a pre-stack seismic inversion method provided by an embodiment of the present invention, in one embodiment, the aforementioned construction of a convolutional neural network includes:

[0061] Construct the first convolutional layer, the second convolutional layer, and the fully connected layer; each convolutional layer includes three parts: convolution operation, activation function, and batch normalization;

[0062] According to the functional connection order, the input layer, the first convolutional layer, the second convolutional layer, the fully connected layer and the output layer are connected in sequence to build a convolutional neural network; wherein, the functional connection order includes: the input layer receives the label data with classification labels; the input layer is connected to the input of the first convolutional layer; the output of the first convolutional layer is used as the input of the second convolutional layer; the output of the second convolutional layer is connected into a one-dimensional vector as the input of the fully connected layer; the output of the fully connected layer is connected to the output layer, and the inversion result to be predicted is output.

[0063] When implementing a pre-stack seismic inversion method provided by an embodiment of the present invention, in one embodiment, a convolution layer may be constructed as follows:

[0064]

[0065] ReLU=max(0,x) (2)

[0066] Among them, BN(·) is batch normalization processing; ReLU is the activation function, * represents multiplication, which is generally expressed as star multiplication; is the output result of the convolutional layer; w k is the convolution operator; x is the input data; b k is the bias term; i is the row index of the matrix; j is the column index of the matrix.

[0067] The above-mentioned expressions for constructing the convolutional layer are for illustration only. Those skilled in the art will appreciate that, during implementation, the above formulas may be modified in a certain form and other parameters or data may be added, or other specific formulas may be provided as needed. These variations shall all fall within the scope of protection of the present invention.

[0068] The input data of the convolutional neural network are pre-stack common reflection point gathers and reservoir types, and the output is the corresponding P-wave velocity, S-wave velocity and density. The embodiment of the present invention builds a three-layer convolutional neural network model with the following structure: Figure 4 As shown in Figure 1, it includes two convolutional layers and one fully connected layer. Each convolutional layer consists of three parts: convolution operation, activation function, and batch normalization, as shown in Equation (1). The activation function uses the ReLU function, as shown in Equation (2), where BN(·) is batch normalization. The output of the second convolutional layer is concatenated into a one-dimensional vector as the input of the fully connected layer. Finally, the output of the fully connected layer is the three elastic parameters to be predicted.

[0069] The present invention provides a method for applying a convolutional neural network (CNN) algorithm in a deep learning method to perform prestack AVO inversion to predict compressional wave velocity, shear wave velocity, and density. The method utilizes CNN's powerful ability to construct a nonlinear mapping relationship between input and output to establish a nonlinear mapping relationship between prestack seismic CRP gathers and the three parameters of compressional wave velocity, shear wave velocity, and density, thereby forming a data-driven prestack AVO three-parameter inversion method. In one example of the present invention, the method mainly includes:

[0070] Based on the regional structure, sedimentary information, logging data analysis and seismic attribute analysis as prior information, the research targets are divided into Type I reservoir favorable development areas, Type II reservoir moderate development areas and Type III reservoir underdeveloped areas, and the pre-stack CRP gather data are classified and marked.

[0071] Labeled data generation is achieved using a stochastic geological simulation method constrained by rock physics, geological structure, sedimentary information, and well logging information. Core measurement data from the study area is analyzed and combined with logging data from existing wells, including P-wave velocity, S-wave velocity, density, porosity, water saturation, and shale content, to select and construct a rock physics model appropriate for the study area. Based on the regional classification results, the lithofacies probability distribution within each classification area is utilized, combined with stochastic geological simulation methods, to generate different types of virtual well curves, including lithofacies, porosity, shale content, and saturation. Finally, the previously selected rock physics model is used to generate corresponding P-wave velocity, S-wave velocity, and density. Elastic parameters derived from the simulation are used for wavefield forward modeling to generate prestack CRP gathers. Each prestack CRP gather corresponds to a set of elastic parameter curves (P-wave velocity curves, S-wave velocity curves, and density curves), which together constitute a set of labeled data. Furthermore, labeled data are classified and labeled based on the type of virtual well used in the forward modeling.

[0072] Build a CNN network: The network input data is prestack CRP gathers and reservoir type, and the output is the corresponding P-wave velocity, S-wave velocity, and density. A three-layer convolutional neural network model is constructed, consisting of two convolutional layers and one fully connected layer. Each convolutional layer includes a convolution operation, an activation function, and batch normalization. The activation function uses the ReLU function. The output of the second convolutional layer is concatenated into a one-dimensional vector, which serves as the input of the fully connected layer. The final output of the fully connected layer is the three elastic parameters to be predicted.

[0073] The embodiments of the present invention fully utilize geological and geophysical knowledge as prior information to constrain the pre-stack AVO (Amplitude Versus Offset) three-parameter inversion based on a convolutional neural network. Instead of simply applying a convolutional neural network to construct a nonlinear relationship between the pre-stack common reflection point gather and the elastic parameters to be inverted, the constraints of prior information can narrow the solution space of the optimization problem and improve the stability and accuracy of the optimization problem. In addition, the labeled data obtained through a random geological simulation method based on geological and rock physical information provides a large amount of labeled data for network training, which is beneficial to the optimization of the network model, improves the generalization ability of the network, and thus improves the network's ultimate predictive ability.

[0074] A prestack seismic inversion method provided by an embodiment of the present invention, during implementation, uses geological and geophysical information as constraints, introduces a convolutional neural network algorithm into the prestack seismic inversion task, and leverages the powerful ability of convolutional neural networks to solve nonlinear problems to improve the accuracy of prestack AVO (Amplitude Versus Offset) three-parameter inversion and obtain more accurate elastic parameter prediction results.

[0075] Figure 2 FIG. 1 is a flow chart of a pre-stack seismic inversion method according to an embodiment of the present invention, as shown in FIG. Figure 2 As shown, the embodiment of the present invention further provides a process of a pre-stack seismic inversion method, including:

[0076] Step 101: According to Figure 3 The geological structure, geological sedimentation, seismic attributes and single well lithofacies information of the study area shown in the figure are used to divide the reservoir favorable and unfavorable areas. For sandstone reservoirs, that is, the relatively high and low sand areas, they are usually divided into three categories: Class I favorable reservoir development areas, Class II moderate reservoir development areas, and Class III underdeveloped reservoir areas.

[0077] Step 102: Based on the core measurement data and well logging data, rock physics modeling experiments are conducted, and a theoretical rock physics model suitable for the target layer in the research area is selected for subsequent modeling. Then, the geological and geophysical information is integrated and stochastic geological simulation technology is used to generate virtual well data.

[0078] Step 103: Based on the well data generated in the previous step, seismic wave forward modeling is performed to obtain prestack seismic data, thereby constructing a labeled dataset. The generated labeled data are all clearly classified and belong to one of the three categories identified in step 101. They are used as a data type along with the prestack common reflection point gathers as input to the network.

[0079] Step 104: Build a CNN network, such as Figure 4 The input data is pre-stack common reflection point gather or angle gather and partition information, and the output is P-wave velocity, S-wave velocity and density. The optimization objective function of the model network decreases with the increase of the number of iterations, as shown in Figure 5 shown.

[0080] Step 105: Use the trained network to perform pre-stack AVO (Amplitude Versus Offset) three-parameter inversion to obtain P-wave velocity, S-wave velocity and density, as shown in the following example: Figure 6 The inversion results show that the predicted values ​​are roughly consistent with the true values.

[0081] Figure 7 A schematic diagram of a computer device for running a pre-stack seismic inversion method implemented in the present invention is shown in FIG. Figure 7 As shown, an embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned pre-stack seismic inversion method when executing the computer program.

[0082] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program for executing the above-mentioned pre-stack seismic inversion method.

[0083] The present invention also provides a prestack seismic inversion device, as described in the following embodiments. Since the principle of the device is similar to that of a prestack seismic inversion method, the implementation of the device can refer to the implementation of a prestack seismic inversion method, and the repeated parts will not be repeated.

[0084] Figure 8 FIG. 1 is a schematic diagram of a pre-stack seismic inversion device according to an embodiment of the present invention, Figure 8 As shown, an embodiment of the present invention further provides a pre-stack seismic inversion device, comprising:

[0085] The reservoir division module 801 is used to divide the reservoirs in the study area according to the prior information and determine the regional classification results;

[0086] The tag information determination module 802 is used to determine the tag information according to the stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information;

[0087] The classification and marking module 803 is used to classify and mark the label information using the regional classification results to determine the label data with the classification mark;

[0088] A convolutional neural network building module 804 is used to build a convolutional neural network;

[0089] A convolutional neural network training module 805 is used to train a convolutional neural network using labeled data with classification labels and determine a trained convolutional neural network;

[0090] The pre-stack seismic inversion module 806 is used to perform pre-stack AVO (Amplitude Versus Offset) three-parameter inversion based on the trained convolutional neural network to determine the elastic parameter inversion results.

[0091] When implementing a pre-stack seismic inversion device provided by an embodiment of the present invention, in one embodiment, the reservoir division module is specifically used to:

[0092] Based on prior information consisting of geological structure information, sedimentary information, logging data analysis results, seismic attribute analysis results and single well lithofacies information, the reservoirs in the study area are divided to determine the regional classification results; among them, the regional classification results include: the first type of reservoir favorable development area, the second type of reservoir moderate development area, and the third type of reservoir underdeveloped area.

[0093] When implementing a pre-stack seismic inversion device provided by an embodiment of the present invention, in one embodiment, the aforementioned tag information determination module is specifically configured to:

[0094] Analyze rock physical information and determine core measurement data;

[0095] Analyze core measurement data, combine it with well logging information, conduct rock physics model tests, and establish a rock physics model that suits the study area;

[0096] Obtain geological structure and sedimentary information from prior information, integrate geological and geophysical information, based on regional classification results, use the lithofacies probability distribution of different classification areas, and combine random geological simulation methods to obtain different types of virtual well curves;

[0097] Based on the virtual well curve, rock physics model is used to build a model and determine the simulated elastic parameters;

[0098] Using simulated elastic parameters, forward modeling of seismic wave fields is performed to determine pre-stack common reflection point gathers;

[0099] The pre-stack common reflection point gathers and simulated elastic parameters are correspondingly combined to determine the label data.

[0100] When implementing a pre-stack seismic inversion device provided by an embodiment of the present invention, in one embodiment, the aforementioned convolutional neural network building module is specifically used to:

[0101] Construct the first convolutional layer, the second convolutional layer, and the fully connected layer; each convolutional layer includes three parts: convolution operation, activation function, and batch normalization;

[0102] According to the functional connection order, the input layer, the first convolutional layer, the second convolutional layer, the fully connected layer and the output layer are connected in sequence to build a convolutional neural network; wherein, the functional connection order includes: the input layer receives the label data with classification labels; the input layer is connected to the input of the first convolutional layer; the output of the first convolutional layer is used as the input of the second convolutional layer; the output of the second convolutional layer is connected into a one-dimensional vector as the input of the fully connected layer; the output of the fully connected layer is connected to the output layer, and the inversion result to be predicted is output.

[0103] When implementing a pre-stack seismic inversion device provided by an embodiment of the present invention, in one embodiment, the aforementioned convolutional neural network building module is further used to construct a convolutional layer in the following manner:

[0104]

[0105] ReLU=max(0,x)

[0106] Among them, BN(·) is batch normalization processing; ReLU is the activation function, and * represents multiplication; is the output result of the convolutional layer; w k is the convolution operator; x is the input data; b kis the bias term; i is the row index of the matrix; j is the column index of the matrix.

[0107] In summary, the embodiments of the present invention provide a prestack seismic inversion method and apparatus, comprising: first, dividing the reservoirs of the study area based on prior information to determine regional classification results; then, determining label information based on a stochastic geological simulation method constrained by rock physical information, prior information, regional classification results, and well logging information; then, classifying and labeling the label information using the regional classification results to determine labeled data with classification labels; next, building a convolutional neural network; and further, training the convolutional neural network using the labeled data with classification labels to determine a trained convolutional neural network; finally, performing prestack AVO (Amplitude Versus Offset) three-parameter inversion based on the trained convolutional neural network to determine elastic parameter inversion results. By fully utilizing prior information to constrain the prestack AVO three-parameter inversion based on the convolutional neural network, it no longer simply applies the convolutional neural network to construct a nonlinear relationship between the prestack common reflection point gather and the elastic parameters to be inverted. The constraints of the prior information can narrow the solution space of the optimization problem, thereby improving the stability and accuracy of the optimization problem. Furthermore, the labeled data obtained through stochastic geological simulation provides a large amount of labeled data for network training, facilitating network model optimization and improving the network's generalization capabilities, thereby enhancing the network's ultimate predictive capabilities. The present invention utilizes the powerful ability of convolutional neural networks to solve nonlinear problems, improving the accuracy of prestack AVO (Amplitude Versus Offset) three-parameter inversion and obtaining more accurate elastic parameter prediction results.

[0108] This invention fully utilizes geological and geophysical knowledge as prior information to constrain the prestack AVO (Amplitude Versus Offset) three-parameter inversion based on a convolutional neural network. Rather than simply applying a convolutional neural network to construct a nonlinear relationship between the prestack common reflection point gather and the elastic parameters to be inverted, the prior information constraint can narrow the solution space for the optimization problem, thereby improving the stability and accuracy of the optimization problem. Furthermore, the labeled data obtained through a random geological simulation method based on geological and petrophysical information provides a large amount of labeled data for network training, which is beneficial for optimizing the network model, improving the network's generalization ability, and thus improving the network's ultimate predictive ability.

[0109] The present invention uses geological and geophysical information as constraints and introduces a convolutional neural network algorithm into the prestack seismic inversion task. With the help of the powerful ability of convolutional neural networks to solve nonlinear problems, the accuracy of prestack AVO (AmplitudeVersusOffset) three-parameter inversion is improved, and more accurate elastic parameter prediction results are obtained.

[0110] 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.

[0111] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (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 processor, 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.

[0112] 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.

[0113] 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.

[0114] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A pre-stack seismic inversion method, characterized in that: include: Based on prior information, the reservoirs in the study area are divided and the regional classification results are determined; Determine label information using a stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information. Using the regional classification results to classify and mark the label information, and determine the label data with classification marks; Build a convolutional neural network; Using labeled data with classification marks to train a convolutional neural network, and determining the trained convolutional neural network; Based on the trained convolutional neural network, pre-stack seismic inversion is performed to determine the elastic parameter inversion results; Among them, the label information is determined based on the stochastic geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information, including: Analyze rock physical information and determine core measurement data; Analyze core measurement data, combine it with well logging information, conduct rock physics model tests, and establish a rock physics model that suits the study area; Obtain geological structure and sedimentary information from prior information, integrate geological and geophysical information, based on regional classification results, use the lithofacies probability distribution of different classification areas, and combine random geological simulation methods to obtain different types of virtual well curves; Based on the virtual well curve, rock physics model is used to build a model and determine the simulated elastic parameters; Using simulated elastic parameters, forward modeling of seismic wave fields is performed to determine pre-stack common reflection point gathers; The pre-stack common reflection point gathers and simulated elastic parameters are correspondingly combined to determine the label data.

2. The method according to claim 1, wherein Based on prior information, the reservoirs in the study area are divided and the regional classification results are determined, including: Based on prior information consisting of geological structure information, sedimentary information, logging data analysis results, seismic attribute analysis results and single well lithofacies information, the reservoirs in the study area are divided to determine the regional classification results; among them, the regional classification results include: the first type of reservoir favorable development area, the second type of reservoir moderate development area, and the third type of reservoir underdeveloped area.

3. The method according to claim 1, wherein Build a convolutional neural network, including: Construct the first convolutional layer, the second convolutional layer, and the fully connected layer; each convolutional layer includes three parts: convolution operation, activation function, and batch normalization; According to the functional connection order, the input layer, the first convolutional layer, the second convolutional layer, the fully connected layer and the output layer are connected in sequence to build a convolutional neural network; wherein, the functional connection order includes: the input layer receives the label data with classification labels; the input layer is connected to the input of the first convolutional layer; the output of the first convolutional layer is used as the input of the second convolutional layer; the output of the second convolutional layer is connected into a one-dimensional vector as the input of the fully connected layer; the output of the fully connected layer is connected to the output layer, and the inversion result to be predicted is output.

4. The method according to claim 3, wherein Construct the convolutional layer as follows: Among them, BN(·) is batch normalization processing; ReLU is the activation function, and * represents multiplication; is the output result of the convolutional layer; w k is the convolution operator; x is the input data; b k is the bias term; i is the row index of the matrix; j is the column index of the matrix.

5. A pre-stack seismic inversion device, characterized in that: include: The reservoir division module is used to divide the reservoirs in the study area according to prior information and determine the regional classification results; a label information determination module, for determining label information based on a random geological simulation method constrained by petrophysical information, prior information, regional classification results, and well logging information; A classification and marking module is used to classify and mark label information using the regional classification results to determine label data with classification marks; Convolutional neural network building module, used to build convolutional neural networks; A convolutional neural network training module is used to train the convolutional neural network using labeled data with classification labels and determine the trained convolutional neural network; The pre-stack seismic inversion module is used to perform pre-stack seismic inversion based on the trained convolutional neural network and determine the elastic parameter inversion results; The tag information determination module is specifically used to: Analyze rock physical information and determine core measurement data; Analyze core measurement data, combine it with well logging information, conduct rock physics model tests, and establish a rock physics model that suits the study area; Obtain geological structure and sedimentary information from prior information, integrate geological and geophysical information, based on regional classification results, use the lithofacies probability distribution of different classification areas, and combine random geological simulation methods to obtain different types of virtual well curves; Based on the virtual well curve, rock physics model is used to build a model and determine the simulated elastic parameters; Using simulated elastic parameters, forward modeling of seismic wave fields is performed to determine pre-stack common reflection point gathers; The pre-stack common reflection point gathers and simulated elastic parameters are correspondingly combined to determine the label data.

6. The device according to claim 5, characterized in that Reservoir partitioning module, specifically used for: Based on prior information consisting of geological structure information, sedimentary information, logging data analysis results, seismic attribute analysis results and single well lithofacies information, the reservoirs in the study area are divided to determine the regional classification results; among them, the regional classification results include: the first type of reservoir favorable development area, the second type of reservoir moderate development area, and the third type of reservoir underdeveloped area.

7. The device according to claim 5, characterized in that Convolutional neural network building module, specifically used for: Construct the first convolutional layer, the second convolutional layer, and the fully connected layer; each convolutional layer includes three parts: convolution operation, activation function, and batch normalization; According to the functional connection order, the input layer, the first convolutional layer, the second convolutional layer, the fully connected layer and the output layer are connected in sequence to build a convolutional neural network; wherein, the functional connection order includes: the input layer receives the label data with classification labels; the input layer is connected to the input of the first convolutional layer; the output of the first convolutional layer is used as the input of the second convolutional layer; the output of the second convolutional layer is connected into a one-dimensional vector as the input of the fully connected layer; the output of the fully connected layer is connected to the output layer, and the inversion result to be predicted is output.

8. The device according to claim 7, wherein The convolutional neural network building module is also used to construct the convolutional layer in the following way: ReLU=max(0,x) Among them, BN(·) is batch normalization processing; ReLU is the activation function, and * represents multiplication; is the output result of the convolutional layer; w k is the convolution operator; x is the input data; b k is the bias term; i is the row index of the matrix; j is the column index of the matrix.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program for executing the method according to any one of claims 1 to 4.

Images