Method, device, medium and equipment for high-resolution seismic inversion of thin-layer structures

Abstract

The present invention relates to a method, device, medium and equipment for high-resolution seismic inversion of thin-layer structures. The method comprises the following steps: constructing a thin interbedded sand body model, forward-synthesizing seismic records based on the thin interbedded sand body model to obtain seismic traces s; forming training data from the seismic traces s and impedance data z to prepare a training set {s, z} T and validation set {s,z} V ; The training set is used to train a neural network model, and the validation set is used to validate the neural network model; the neural network model is used to obtain predicted impedance, and the predicted impedance is converted into a reflection coefficient sequence; the reflection coefficient sequence is convolved with the forward wavelet to generate a seismic record, and the generated seismic record and the input seismic trace set constitute a supervisory term constraint model to drive the inversion process.

Description

Technical Field

[0001] The present invention relates to a thin-layer structure high-resolution seismic inversion method, device, medium and equipment, and belongs to the technical field of oil and gas geophysical exploration inversion. Background Art

[0002] Currently, with most domestic oilfields entering the mid-to-late stages of exploration and development, target reservoirs are increasingly concentrated in thin interbedded sand bodies and lithologic reservoirs. Although these reservoirs present significant exploration challenges, they harbor abundant oil and gas resources and offer promising prospects for exploration and development. Seismic inversion, a crucial technique for reservoir prediction during the oil and gas exploration and development phase, plays an irreplaceable role in describing the spatial distribution of underground reservoirs, characterizing the spatial structure of sand bodies, and predicting reservoir physical properties. However, the limited resolution of traditional seismic inversion methods severely restricts their application in thin-bed structure prediction. Therefore, developing high-resolution reservoir inversion methods for thin-bed and thin-interbedded reservoirs is of great practical significance for thin-bed prediction and well placement in oilfields entering the mid-to-late stages of development.

[0003] Seismic data processing for this type of reservoir often focuses on two aspects: first, directly characterizing the reservoir by improving the resolution of seismic data, and second, identifying the spatial distribution of sand bodies through inversion. Traditional impedance inversion methods rely on initial models. For reservoirs such as thin interbeds that are severely affected by interference effects, it is difficult to establish a relatively accurate initial model that meets the inversion requirements. The resolution of its inversion results is limited by the seismic frequency band and cannot meet the resolution requirements of exploration. The characterization of the sand body distribution morphology and spatial structure based on the inversion results is also very limited. In recent years, seismic inversion based on statistical theory has broken through the limitations of the seismic frequency band and further improved the resolution of the inversion results compared to deterministic inversion methods, but it is also limited by the initial geological model.

[0004] With the successful application and development of machine learning in seismic data processing and interpretation, many researchers have begun researching data-driven thin-layer structural characterization. Thanks to its powerful nonlinear mapping capabilities, the resolution of inversion results has been significantly improved. However, the development of single-data-driven impedance inversion techniques has been limited by issues such as a lack of clear physical meaning, a lack of physical interpretability, weak transferability, and a heavy reliance on labels. Summary of the Invention

[0005] In response to the above problems, the purpose of the present invention is to provide a method, device, medium and equipment for high-resolution seismic inversion of thin-layer structures, so as to greatly improve the ability of seismic inversion results to characterize thin-layer structures and improve the accuracy of prediction and description of thin-layer oil and gas reservoirs.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for high-resolution seismic inversion of thin-layer structures, comprising the steps of:

[0008] Constructing a thin interbedded sand body model, and forward-modeling a synthetic seismic record based on the thin interbedded sand body model to obtain a seismic trace s;

[0009] The well seismic trace s is used to obtain the predicted impedance using a pre-built neural network model, and the predicted impedance is converted into a reflection coefficient sequence;

[0010] The reflection coefficient sequence is convolved with the forward wavelet to generate a seismic record, and the generated seismic record and the input seismic gather constitute a supervisory term constrained neural network model to drive the inversion process.

[0011] The construction of the neural network model includes:

[0012] The seismic trace s and impedance data are used to form the training data, and the training set {s, z} is created. T and validation set {s,z} V ;

[0013] The training set {s,z} T The neural network model is trained and the validation set {s, z} is used. V The neural network model is verified.

[0014] Preferably, 70% of the training data is used as a training set, and 30% is used as a validation set.

[0015] The thin-layer structure high-resolution seismic inversion method further comprises the steps of:

[0016] Saving the network architecture and node weights of the neural network model;

[0017] Impedance inversion application analysis is carried out based on actual seismic data to test the migration generalization ability and practicality.

[0018] The neural network model is a ConvLSTM neural network model.

[0019] The neural network model is used to obtain the predicted impedance, and the predicted impedance is converted into a reflection coefficient sequence. The specific calculation formula is:

[0020]

[0021] Where, is the predicted impedance given by the neural network model, is the corresponding reflection coefficient, and i represents a certain sedimentary layer.

[0022] The objective function of the inversion process driven by the supervision term constraint neural network model is to generate seismic records by convolving the reflection coefficient sequence with the forward wavelet and combining the generated seismic records with the input seismic gathers:

[0023]

[0024] Where, is the predicted impedance given by the neural network model, is the corresponding reflection coefficient, G is the known forward wavelet matrix, and μ is the regularization parameter, which is used to balance data-driven and model constraints and adjust the inversion effect.

[0025] In a second aspect, the present invention further provides a thin-layer structure high-resolution seismic inversion device, comprising:

[0026] The first processing unit is used to construct a thin interbedded sand body model, and forward-synthesize seismic records based on the thin interbedded sand body model to obtain seismic trace s;

[0027] The second processing unit is used to obtain the predicted impedance by using the pre-built neural network model for the well seismic trace s, and convert the predicted impedance into a reflection coefficient sequence;

[0028] The third processing unit is used to convolve the reflection coefficient sequence with the forward wavelet to generate seismic records, and to form a supervisory term constrained neural network model to drive the inversion process by combining the generated seismic records with the input seismic trace gathers.

[0029] In a third aspect, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the thin-layer structure high-resolution seismic inversion method when executed by a processor.

[0030] In a fourth aspect, the present invention also 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 thin-layer structure high-resolution seismic inversion method when executing the computer program.

[0031] The present invention has the following advantages due to the adoption of the above technical solution:

[0032] The jointly driven high-resolution seismic impedance inversion method combines the advantages of data-driven to significantly improve resolution and model constraints to improve inversion accuracy to carry out experimental analysis of the spatial structure characterization of thin interbedded sand bodies. Compared with the single data-driven inversion technology, the joint drive not only improves the resolution of the inversion results, but also provides more accurate sand body characterization results, and accurately characterizes the lateral distribution and distribution range of the sand bodies. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] By reading the detailed description of the preferred embodiment below, various other advantages and benefits will become clear to those skilled in the art. The accompanying drawings are only used to illustrate the preferred embodiment and are not to be considered as limitations of the present invention. Throughout the accompanying drawings, the same reference numerals are used to represent the same components.

[0034] In the attached figure:

[0035] Figure 1 This is a flowchart for implementing a joint-driven high-resolution impedance inversion method according to the present invention;

[0036] Figure 2 This is a flowchart of the combined drive implementation of the present invention;

[0037] Figure 3 The simple thin interbedded geological model described in the present invention;

[0038] Figure 4 A synthetic seismic record corresponding to the simple thin interbedded geological model described in the present invention;

[0039] Figure 5a This is the impedance inversion result of a simple thin interbed model based on data drive in the present invention;

[0040] Figure 5b This is the impedance inversion result of the simple thin interbed model based on joint driving of the present invention;

[0041] Figure 6 is the learning curve of the ConvLSTM neural network model described in the present invention;

[0042] Figure 7 The actual seismic data described in the present invention;

[0043] Figure 8a This is the impedance inversion result of seismic data driven by the present invention;

[0044] Figure 8b This is the impedance inversion result of seismic data based on joint driving in the present invention. DETAILED DESCRIPTION

[0045] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0046] The embodiment of the present invention provides a method for high-resolution seismic inversion of thin-layer structures, comprising the steps of: constructing a thin interbedded sand body model, forward-modeling and synthesizing seismic records based on the thin interbedded sand body model to obtain a seismic trace s; forming training data from the seismic trace s and impedance data z to create a training set {s, z} T and validation set {s,z} V ; The training set is used to train a neural network model, and the validation set is used to validate the neural network model; the neural network model is used to obtain predicted impedance, and the predicted impedance is converted into a reflection coefficient sequence; the reflection coefficient sequence is convolved with the forward wavelet to generate a seismic record, and the generated seismic record and the input seismic trace set constitute a supervisory term constraint model to drive the inversion process.

[0047] Example 1:

[0048] Embodiment 1 of the present invention provides a method for high-resolution seismic inversion of thin-layer structures, comprising the steps of:

[0049] Step 1: Construct a simple thin interbedded sand body model, forward-model synthetic seismic records, and select seismic traces and corresponding impedance labels to create training data;

[0050] The simple thin interbedded sand body model is constructed as follows: Figure 3 As shown, the simple thin interbedded sand body model includes 300 seismic traces, a sampling length of 0.25s, a sampling interval of 0.5ms, and a time thickness of the sand body of about 2.5-5ms, including geological structures such as lenses and sand body pinch-outs. The sandstone velocity is 3800m / s and the background mudstone velocity is 3600m / s. Based on this model, a joint-driven high-resolution seismic impedance inversion is carried out. The specific implementation process of the present invention is as follows: Figure 1 and Figure 2 shown.

[0051] The impedance model of the thin interbedded sand body is convolved with the Ricker wavelet of 15 Hz to obtain the forward synthetic seismic record, as shown in Figure 4 As shown;

[0052] Step 2: The seismic trace s and impedance data z are used to form training data, and a training set {s, z} is created. T and validation set {s,z} V ;

[0053] Extract the 30th, 90th, 150th, 210th, and 270th seismic channels and the corresponding impedance data to form the training data, and divide them into the training set {s, z} according to the ratio of 70%:30%. T and validation set {s,z} V ;

[0054] Step 3: Train the training set to obtain a neural network model, and use the validation set to validate the neural network model;

[0055] The ConvLSTM neural network model not only has the time series modeling capability of the Long Short-Term Memory (LSTM) network, but can also extract local features of data like a Convolutional Neural Network (CNN). It has both temporal and spatial characteristics, can fully mine the implicit information in the labeled data, and can characterize the underground structure by combining the temporal and spatial characteristics of seismic data.

[0056] The basic architecture parameters of the network include: the number of network layers, the number of neurons in each layer, the size of the convolution kernel, the time step, the learning rate, the number of training times, etc. The mean squared error (MSE) is selected as the loss function to train the neural network, and the optimal network architecture and hyperparameters are determined based on the error performance of the network on the validation set. Based on the trained ConvLSTM network, a simple thin interbedded sand model impedance inversion test is carried out. The test results driven by single data are as follows: Figure 5a shown.

[0057] Step 4: Use the neural network model to obtain the predicted impedance, and convert the predicted impedance into a reflection coefficient sequence. The calculation formula is:

[0058]

[0059] Step 5: Convolve the reflection coefficient sequence with the forward wavelet to generate seismic records. The generated seismic records and the input seismic gathers constitute a supervisory term constraint model to drive the inversion process. The calculation formula is:

[0060]

[0061] In formulas (1) and (2), is the predicted impedance given by the ConvLSTM network, is the corresponding reflection coefficient, G is the known forward wavelet matrix, and μ is the regularization parameter, which is used to balance data-driven and model constraints and adjust the inversion effect.

[0062] Model constraints are introduced on the basis of single data drive to improve the inversion accuracy. Given the forward wavelet matrix, the reflection coefficient sequence obtained by the predicted impedance calculation and the seismic record generated by the wavelet convolution and the input seismic track set constitute an error term, which constrains the prediction effect of data drive and ensures that the prediction result of single data drive is within an acceptable range without large deviations. At this time, the data drive plays a role similar to the inversion operator. Under the premise that the network structure has been determined, the regularization parameter μ is adjusted through the inversion effect to balance the data drive and model constraints, thereby greatly improving the inversion resolution while ensuring the reliability and effectiveness of the inversion results. The high-resolution impedance inversion results of the joint drive are as follows: Figure 5b shown.

[0063] Step 6: Step 4: Save the network architecture, node weights, and other hyperparameters;

[0064] The training process of the ConvLSTM network model is stable and efficient. The errors of the training set and the validation set do not increase with long-term training, indicating that the network does not have problems such as overfitting. The learning curve of the ConvLSTM network model is shown in Figure 2. Figure 6 The node weights and hyperparameters of the network are saved to facilitate subsequent applications in complex scenarios such as transfer learning.

[0065] Step 7: Conduct impedance inversion application analysis based on actual seismic data to further test the migration and generalization capabilities and practicality of this method.

[0066] The experimental results of the impedance inversion experiment on a simple thin interbedded sand body model show that the inversion effect of the joint drive is significantly better than that of the single data drive. The description of the spatial structure and lateral distribution of the sand body is clearer, the boundaries of each lens are clear, and the distribution range is accurate.

[0067] In order to further consider the applicability and migration generalization ability of the present invention, the actual seismic data of an oil field in the east of China (such as Figure 7 Impedance inversion test analysis was carried out (as shown). The main target layer in the target block is shallowly buried, with a depth of about 1100-1500m. It is mainly composed of river channel deposits. In the early stage, it is in the flood season, and is mainly composed of impact fan-braided river deposits. The distribution range of sand bodies is wide and the sediment thickness is large. In the late stage, it is far away from the source, the sediment thickness gradually decreases, and meandering river deposits are the main ones. Spectral analysis of seismic data shows that its main frequency is low and the high-frequency signal decays rapidly. The ability to characterize the interbedded structure of thin sand and mudstone with strong heterogeneity based on seismic data is very limited. Therefore, single data-driven and joint-driven impedance inversion application analysis is carried out. First, the resolution is improved through inversion, and second, high frequency is restored based on data-driven, and the spatial morphology and lateral distribution of sand bodies are characterized based on the inversion results.

[0068] By comparing and analyzing the complex thin interlayer impedance inversion results of single data drive and joint drive (such as Figure 8a and Figure 8b The following conclusions can be drawn: While single-data-driven impedance inversion can significantly improve the resolution of inversion results, its lateral continuity is poor, limiting its ability to depict the lateral distribution and range of sand bodies. The combined-driven impedance inversion method improves resolution while providing more accurate inversion results, better describing the spatial structure of thin sand bodies, and enhancing the ability to characterize reservoir structure based on seismic data, providing excellent data conditions and prior knowledge for subsequent physical property analysis.

[0069] As a second embodiment, the present invention further provides a thin-layer structure high-resolution seismic inversion device, comprising:

[0070] The first processing unit is used to construct a thin interbedded sand body model, and forward-synthesize seismic records based on the thin interbedded sand body model to obtain seismic trace s;

[0071] The second processing unit is used to obtain the predicted impedance by using the pre-built neural network model for the well seismic trace s, and convert the predicted impedance into a reflection coefficient sequence;

[0072] The third processing unit is used to convolve the reflection coefficient sequence with the forward wavelet to generate seismic records, and to form a supervisory term constrained neural network model to drive the inversion process by combining the generated seismic records with the input seismic trace gathers.

[0073] As a third embodiment, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the thin-layer structure high-resolution seismic inversion method when executed by a processor.

[0074] As a fourth embodiment, the present invention also 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 thin-layer structure high-resolution seismic inversion method when executing the computer program.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A high-resolution seismic inversion method for thin-layer structures, characterized by: Including steps: Constructing a thin interbedded sand body model, and forward-modeling a synthetic seismic record based on the thin interbedded sand body model to obtain a seismic trace s; The well seismic trace s is used to obtain the predicted impedance using a pre-built neural network model, and the predicted impedance is converted into a reflection coefficient sequence. The specific calculation formula is: Where, is the predicted impedance given by the neural network model, is the corresponding reflection coefficient, i represents a certain sedimentary layer; The reflection coefficient sequence is convolved with the forward wavelet to generate seismic records. The generated seismic records and the input seismic gathers are used to form a supervisory term constrained neural network model to drive the inversion process. The objective function is: Where, is the predicted impedance given by the neural network model, is the corresponding reflection coefficient, is the known forward wavelet matrix, It is a regularization parameter used to balance data-driven and model constraints and adjust the inversion effect.

2. The thin-layer structure high-resolution seismic inversion method according to claim 1, characterized in that: The construction of the neural network model includes: Use seismic traces and impedance data to form training data and make a training set and validation set ; The training set Train to obtain a neural network model, using the validation set The neural network model is verified.

3. The thin-layer structure high-resolution seismic inversion method according to claim 2, characterized in that: 70% of the training data was used as a training set, and 30% was used as a validation set.

4. The thin-layer structure high-resolution seismic inversion method according to claim 1, characterized in that: Also includes the steps: Saving the network architecture and node weights of the neural network model; Impedance inversion application analysis is carried out based on actual seismic data to test the migration generalization ability and practicality.

5. The thin-layer structure high-resolution seismic inversion method according to claim 1, characterized in that: The neural network model is a ConvLSTM neural network model.

6. A thin-layer structure high-resolution seismic inversion device, characterized in that: include: The first processing unit is used to construct a thin interbedded sand body model, and forward-synthesize seismic records based on the thin interbedded sand body model to obtain seismic trace s; The second processing unit is used to obtain the predicted impedance using the pre-built neural network model for the well seismic trace s, and convert the predicted impedance into a reflection coefficient sequence. The specific calculation formula is: Where, is the predicted impedance given by the neural network model, is the corresponding reflection coefficient, i represents a certain sedimentary layer; The third processing unit is used to convolve the reflection coefficient sequence with the forward wavelet to generate a seismic record, and to form a supervisory term constrained neural network model to drive the inversion process by combining the generated seismic record with the input seismic gather. The objective function is: Where, is the predicted impedance given by the neural network model, is the corresponding reflection coefficient, is the known forward wavelet matrix, It is a regularization parameter used to balance data-driven and model constraints and adjust the inversion effect.

7. A computer-readable storage medium, characterized in that Computer instructions are stored, and the computer instructions are used to implement the thin-layer structure high-resolution seismic inversion method according to any one of claims 1 to 5 when executed by a processor.

8. A computer device, characterized in that: The method comprises 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 for high-resolution seismic inversion of thin-layer structures as claimed in any one of claims 1 to 5 is implemented.

Images