Thin layer structure identification method, apparatus and device, and storage medium

Abstract

The invention belongs to the technical field of petroleum geological exploration, and provides a thin layer structure identification method, device and equipment and a storage medium, and the method comprises the steps: building a thin interbed geologic model, synthesizing a seismic record based on the thin interbed geologic model, and constructing training data according to the seismic record; establishing a convolutional long short-term memory network, and training the convolutional long short-term memory network according to the training data to obtain a thin layer structure recognition model; and identifying a to-be-identified thin-layer structure by using the thin-layer structure identification model. According to the method, accurate label data can be provided by utilizing the established geologic model to perform inversion of the synthetic seismic record, so that the optimal model obtained by training can be well migrated to actual data, and high-resolution inversion of thin-layer impedance is realized. Compared with traditional impedance inversion, establishment of an initial model is avoided, horizon pickup from seismic data is not needed, and the method is more efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas geophysical exploration inversion, and in particular to a thin layer structure identification method, device, equipment and storage medium. Background Art

[0002] At present, with the continuous deepening of oil exploration and development, the development of conventional oil and gas reservoirs has entered the middle and late stages. In order to ensure the steady growth of oil field production, the exploration targets are concentrated on lithologic oil and gas reservoirs and concealed oil and gas reservoirs. These reservoirs are often thin, controlled by fractures and faults, and are difficult to explore. However, the oil and gas resources stored in them are very rich, and the exploration prospects are broad. For such oil and gas reservoirs, due to the limited vertical resolution of seismic exploration, it is difficult to distinguish the spatial distribution of reservoirs. Therefore, with the support of limited seismic logging data, thin-layer and thin-interlayer thin-layer structure identification is carried out to identify their spatial distribution morphology, which has an important impact on the exploration effect. When the reservoir is a thin sandstone interlayer with a small thickness, it is important to distinguish thin sandstone in the mudstone background, and then determine the favorable exploration and development blocks, which is of great significance for the analysis of reservoir physical properties and the deployment of wells in the later stage. In addition, the ability to correctly identify the spatial distribution of sand bodies is also one of the important tasks in the oil field exploration stage, which helps to improve the accuracy of reservoir positioning. However, in actual inversion and interpretation work, the layer tracking based on seismic data can hardly meet the requirements of the initial model for inversion, and the limited logging data with high-frequency information brings great difficulties to the high-resolution inversion of thin-layer structures. Therefore, it is of great significance to carry out seismic identification and description of thin interbedded sand bodies.

[0003] High-resolution inversion has always been an important means to improve the geological interpretation ability of seismic data. Initially, model-based inversion was the basic means to improve the vertical resolution of seismic data. Through horizon interpretation and well control modeling, researchers established a more accurate initial model and selected a suitable inversion method to improve the seismic data's ability to characterize thin-layer structures. However, this method has limited ability to improve resolution and is heavily dependent on the establishment of the initial model, and can only provide the upper limit of the frequency band range of seismic data. In actual seismic data, the seismic response characteristics of thin interlayers are often severely affected by interference effects, so it is difficult to meet the requirements of the inversion for the initial model in terms of horizon picking. This has a huge impact on the inversion results. In recent years, geostatistical inversion based on statistics and other theories can provide things beyond the seismic frequency band compared to deterministic inversion, and further improve the resolution of seismic inversion on the basis of deterministic inversion, but this method is also limited by the establishment of geological models. In order to break the dependence of inversion on the initial model, many scholars have tried to use data-driven characterization of thin-layer structures, avoiding the dependence of traditional methods on geological models. The commonly used convolutional neural networks (CNN) and generative adversarial networks (GAN) usually cut seismic data into blocks and input them as image data, ignoring the inherent connection between data. This is bound to cause large errors in the inversion results. Summary of the invention

[0004] The object of the present invention is to provide a thin layer structure identification method, device, equipment and storage medium to improve the above-mentioned problem.

[0005] In order to achieve the above objectives, the present application provides the following technical solutions:

[0006] On the one hand, an embodiment of the present application provides a thin layer structure identification method, the method comprising:

[0007] Establishing a thin interbedded geological model, synthesizing seismic records based on the thin interbedded geological model, and constructing training data based on the seismic records;

[0008] Establishing a convolutional long short-term memory network, and training the convolutional long short-term memory network according to the training data to obtain a thin layer structure recognition model;

[0009] The thin layer structure to be identified is identified using the thin layer structure identification model.

[0010] Optionally, establishing a thin interbedded geological model, synthesizing seismic records based on the thin interbedded geological model, and constructing training data according to the seismic records, including:

[0011] Establishing a thin interbedded geological model, wherein the thin interbedded geological model includes sand bodies deposited in different forms, filling speed and density into the sand bodies deposited in different forms, and obtaining a reflection coefficient model through a reflection coefficient calculation formula at vertical incidence;

[0012] The reflection coefficient model is convolved with Ricker wavelets of different main frequencies to obtain seismic records of different resolutions, and seismic records of a preset number of channels are selected from all seismic records as the training data.

[0013] Optionally, a convolutional long short-term memory network is established, and the convolutional long short-term memory network is trained according to the training data to obtain a thin layer structure recognition model, including:

[0014] Using the impedance of the thin interbedded geological model corresponding to each of the training data as label data of each of the training data;

[0015] Constructing the convolutional long short-term memory network, the convolutional long short-term memory network includes an input layer, a hidden layer, a convolutional long short-term memory layer, and an output layer; using the training data to train the convolutional long short-term memory network to obtain the thin layer structure recognition model.

[0016] Optionally, the convolutional long short-term memory network is trained using the training data to obtain the thin layer structure recognition model, including:

[0017] The training data is inputted into the input layer after adjusting the input dimension. After passing through the input layer, the input data is sent to the hidden layer through convolution calculation. The internal correlation of the seismic data is processed through the long short-term memory layer and the mapping with the impedance trace is completed at the same time. The errors of the impedance and label data obtained by inversion are continuously transmitted back, and the weight coefficients of each node of the convolutional long short-term memory network are updated to obtain the thin layer structure recognition model.

[0018] In a second aspect, an embodiment of the present application provides a thin layer structure recognition device, which includes a building module, a training module and a recognition module.

[0019] An establishment module is used to establish a thin interbedded geological model, synthesize seismic records based on the thin interbedded geological model, and construct training data according to the seismic records;

[0020] A training module, used for establishing a convolutional long short-term memory network, training the convolutional long short-term memory network according to the training data, and obtaining a thin layer structure recognition model;

[0021] The recognition module is used to recognize the thin layer structure to be recognized by using the thin layer structure recognition model.

[0022] Optionally, build modules including:

[0023] Establishing a unit for establishing a thin interbed geological model, wherein the thin interbed geological model includes sand bodies deposited in different forms, filling speed and density into the sand bodies deposited in different forms, and obtaining a reflection coefficient model through a reflection coefficient calculation formula of vertical incidence;

[0024] The selection unit is used to convolve the reflection coefficient model with Ricker wavelets of different main frequencies to obtain seismic records of different resolutions, and select seismic records of a preset number of channels from all seismic records as the training data.

[0025] Optional training modules include:

[0026] A labeling unit, used for taking the impedance of the thin interbedded geological model corresponding to each of the training data as label data of each of the training data;

[0027] A construction unit is used to construct the convolutional long short-term memory network, wherein the convolutional long short-term memory network includes an input layer, a hidden layer, a convolutional long short-term memory layer, and an output layer; the convolutional long short-term memory network is trained using the training data to obtain the thin layer structure recognition model.

[0028] Optionally, build units including:

[0029] The training unit is used to adjust the input dimension of the training data, send it to the input layer after adjustment, send the input data to the hidden layer through convolution calculation after passing through the input layer, and then process the internal association of the seismic data through the long short-term memory layer and complete the mapping with the impedance track at the same time; the errors of the impedance and label data obtained by inversion are continuously transmitted back, and the weight coefficients of each node of the convolutional long short-term memory network are updated to obtain the thin layer structure recognition model.

[0030] In a third aspect, an embodiment of the present application provides a thin layer structure recognition device, the device comprising a memory and a processor. The memory is used to store a computer program; the processor is used to implement the steps of the above-mentioned thin layer structure recognition method when executing the computer program.

[0031] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned thin layer structure identification method are implemented.

[0032] The beneficial effects of the present invention are:

[0033] 1. In the present invention, starting from the convolution model, the data connection between impedance data and seismic records is emphasized. In order to avoid the influence of inaccurate labels on the inversion results, the thin interlayer impedance geological model is directly used to synthesize seismic data. The neural network is trained using accurate impedance labels. The long short-term memory network (LSTM) is different from the convolutional neural network, and it has a stronger processing and application ability for sequence data. The convolutional long short-term memory network (cLSTM) used in the present invention is different from the conventional long short-term memory network. It uses convolution calculation to replace the fully connected layer between the input and the gating system, which can better handle the association between data. Compared with the generative adversarial network, the training efficiency is high, and the parameter adjustment is also easier. Only a small number of samples are used for network training to obtain better training effects and migration capabilities, and the recognition and prediction capabilities of thin layers are significantly improved.

[0034] 2. At present, similar patents or documents have been retrieved in the market or other public channels: high-resolution inversion method, device, medium and equipment for deep-water turbidite channel gas reservoirs (patent publication number: CN115079256A), a high-resolution seismic inversion method driven by seismic waveform (patent publication number: CN112147701A), and thin-layer high-resolution inversion method and storage medium (patent publication number: CN112444864A).

[0035] The key steps of the original method are: ① rock physics analysis and well seismic calibration in the target work area ② construction of low-frequency model under seismic waveform constraints ③ deterministic inversion based on original seismic data ④ fusion of frequency-divided data for high-resolution inversion;

[0036] Key steps of the existing method: ① Establish a thin interbedded geological model, synthesize seismic records, and produce training data and labels ② Establish a convolutional long short-term memory network ③ Train the convolutional long short-term memory network ④ Apply the convolutional long short-term memory network

[0037] The original method has problems such as complicated operation process and low reservoir inversion accuracy.

[0038] The present invention is a high-precision reservoir inversion method, which significantly improves the recognition and prediction capabilities of thin layers and is applicable to various types of lithology and stratigraphic oil and gas reservoirs. The method has broad application prospects and is of great significance to oil reservoir development.

[0039] 3. The present invention provides an efficient and accurate thin-layer structure identification method. The inversion of synthetic seismic records using the established geological model can provide accurate label data, so that the best model obtained by training can be well migrated to the actual data, and the high-resolution inversion of thin-layer impedance can be achieved. Compared with the traditional impedance inversion, the present invention avoids the establishment of the initial model, so there is no need to pick up the horizon from the seismic data, which is more efficient. From the inversion results, the high-resolution inversion results of thin-layer structures driven by data have higher resolution and lateral continuity, and the characterization of thin-layer structures is clearer and more effective. Compared with the traditional long short-term memory network, the convolutional long short-term memory network used in the present invention can better handle the spatiotemporal relationship of data, and the lateral continuity of the training results is better. Once the training is completed, the neural network can be used for the impedance inversion of actual data immediately and effectively, and the parameters required to be adjusted are also less, with high application efficiency.

[0040] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 is a schematic flow chart of a thin layer structure identification method according to an embodiment of the present invention;

[0043] Figure 2 is a schematic structural diagram of a thin layer structure identification device according to an embodiment of the present invention;

[0044] Figure 3 is a schematic structural diagram of a thin layer structure identification device described in an embodiment of the present invention;

[0045] Figure 4 is a schematic diagram of a thin interbedded geological model described in an embodiment of the present invention;

[0046] Figure 5 The relatively high-resolution seismic data (a) and the low-resolution seismic record (b) described in the embodiment of the present invention;

[0047] Figure 6 are the training data and labels described in the embodiments of the present invention;

[0048] Figure 7 is the change of the network training error as the number of iterations described in the embodiment of the present invention;

[0049] Figure 8 3 and 4. The diagrams are effect diagrams of the application of the thin layer structure identification model described in the embodiment of the present invention, which are respectively the inversion result of the seismic data synthesized by the high-frequency wavelet (a) and the inversion result of the seismic data synthesized by the low-frequency wavelet (b). DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] It should be noted that similar reference numerals or letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment provides a thin layer structure identification method, which includes step S1, step S2, step S3 and step S4.

[0054] Step S1, establishing a thin interbedded geological model, synthesizing seismic records based on the thin interbedded geological model, and constructing training data according to the seismic records;

[0055] The specific implementation steps of this step include step S11 and step S12;

[0056] Step S11, establishing a thin interbedded geological model, wherein the thin interbedded geological model includes sand bodies deposited in different forms, filling speed and density into the sand bodies deposited in different forms, and obtaining a reflection coefficient model through a reflection coefficient calculation formula of vertical incidence;

[0057] In this step, the thin interbedded geological model contains river channels, pinch-outs, lenses, wedges, and thin sand bodies formed by stable sedimentation. These sand bodies overlap each other in space, forming a complex distribution of thin interbedded sand bodies. The overall distribution is in the background of mudstone deposition, forming a favorable reservoir distribution for oil and gas storage. The established thin interbedded geological model is as follows: Figure 4 ;

[0058] Step S12: Convolve the reflection coefficient model with Ricker wavelets of different main frequencies to obtain seismic records of different resolutions, and select seismic records with a preset number of channels from all seismic records as the training data.

[0059] In this step, the obtained reflection coefficient model is convolved with Ricker wavelets of different main frequencies (30 Hz and 10 Hz Ricker wavelets are used in this embodiment) to obtain high-resolution seismic data and low-resolution seismic records, such as Figure 5 As shown in the figure, the effectiveness of the subsequent method for thin layer resolution is investigated. The seismic records synthesized by 30HZ main frequency wavelet can basically distinguish each sand body on the seismic record, while the seismic records synthesized by 10HZ main frequency wavelet have the responses of each sand body superimposed on each other due to the interference effect of the wavelet, making it difficult to distinguish. Through the inversion experiment of seismic data with two different resolutions, the method's demand for the number of labels and its ability to improve resolution are investigated.

[0060] Step S2, establishing a convolutional long short-term memory network, and training the convolutional long short-term memory network according to the training data to obtain a thin layer structure recognition model;

[0061] The specific implementation steps of this step include step S21 and step S22;

[0062] Step S21, using the impedance of the thin interbedded geological model corresponding to each of the training data as label data of each of the training data;

[0063] In this step, several tracks are selected from the seismic records synthesized by different main frequency wavelets as training data, and the impedance of the corresponding thin interbed geological model is used as label data. The standard for selecting training data is to cover the seismic response characteristics of different sand bodies as much as possible. An important feature of data prediction by neural network is the richness of data, which requires labels to cover data features as much as possible, so that the neural network model obtained by training can have better migration ability and prediction ability. For data features that the network has not seen, the neural network may make wrong judgments. However, through data enhancement, reinforcement learning and other means, the learning ability of neural network has been greatly improved. Relatively rich data can also achieve good application results. In this embodiment, a total of 5 seismic data are selected as training data, and they are divided into training set and verification set according to the ratio of 8:2 to participate in the training process of neural network. Figure 6 Shown are the training data and labels used in this embodiment.

[0064] Step S22, constructing the convolutional long short-term memory network, wherein the convolutional long short-term memory network includes an input layer, a hidden layer, a convolutional long short-term memory layer, and an output layer; using the training data to train the convolutional long short-term memory network to obtain the thin layer structure recognition model.

[0065] In this step, the network used is a convolutional long short-term memory network (cLSTM). The main component of a common long short-term memory network is a gating system, which controls the selection and flow of input information through input gates, forget gates, output gates, and cell states. Compared with convolutional neural networks, it can handle the direct intrinsic connection of data very well. Between the input layer and the hidden layer, the information processing of the conventional long short-term memory network is realized by power full connection, while the convolution calculation is used in the present invention to replace the corresponding fully connected layer, so that it can process more complex spatiotemporal information.

[0066] Meanwhile, in this step, the specific implementation steps of using the training data to train the convolutional long short-term memory network to obtain the thin layer structure recognition model include step S221;

[0067] Step S221, adjust the input dimension of the training data, send it to the input layer after adjustment, send the input data to the hidden layer through convolution calculation after passing through the input layer, and then process the correlation between the seismic data through the long short-term memory layer while completing the mapping with the impedance track; continuously back-propagate the errors of the impedance and label data obtained by inversion, update the weight coefficients of each node of the convolutional long short-term memory network, and obtain the thin layer structure recognition model.

[0068] In this step, before sending the seismic data and impedance labels to the neural network for training, the input dimension of the data needs to be adjusted. Send the data that meets the neural network input requirements to the input layer to train the convolutional long short-term memory network;

[0069] At the same time, in this step, the selected seismic track data will be scanned and divided into subsets in the convolutional long short-term memory layer according to the hyperparameter: time window length. The convolutional long short-term memory layer contains input gates, output gates, forget gates and other structures, and the initial coefficients will be determined by random assignment. The input data is first sent to the hidden layer through convolution calculation, and then the association between the seismic data is processed through the long short-term memory layer to complete the mapping with the impedance data. The errors of the inverted impedance and label data are continuously propagated back to update the weight coefficients of each node of the neural network, and the best neural network model is obtained within a limited number of training times. This model is applied to the subsequent high-resolution inversion of seismic data.

[0070] In addition to the above steps, in the training process of the neural network, the overfitting problem is usually difficult to avoid, which comes from the limited data. Taking into account the practical application problems, it is difficult for us to obtain rich labeled data. Therefore, it becomes necessary to prevent the occurrence of overfitting problems as much as possible through other means. Commonly used methods include: adding Dropout, BatchNormalization and other layers to make the network have a certain sparsity, thereby reducing the synergy between different features and making the training process of the neural network more stable, or controlling the number of iterations through the Early Stopping mechanism, and stopping the training of the neural network when the requirements are met. In this embodiment, Dropout is used to avoid the occurrence of overfitting problems;

[0071] The output layer outputs the inversion result of the neural network for the seismic data, and the final output result is infinitely close to the impedance label data. Since the initial values ​​of the weights on each layer and each node of the neural network are random, there will be deviations between the predicted results and the actual label data in the early stage of neural network training. Although this deviation is corrected to a certain extent through error backpropagation, gradient descent, etc. in the subsequent training process. However, under the support of limited data and limited training times, this correction is also limited. In order to characterize the direct error between the inversion result and the label data, the loss function: MSE (root mean square error) is used to describe this error. Figure 7 The error of network training in this embodiment changes with the number of iterations;

[0072] Step S3: using the thin layer structure recognition model to perform thin layer structure recognition.

[0073] In this step, the thin-layer structure identification model can be used on actual seismic data to perform high-resolution thin-layer structure identification. Figure 8 The following is an application effect diagram of the thin-layer structure recognition model of this embodiment. The recognition results of high-frequency seismic data and low-frequency seismic data are shown. The experimental results show that for high-frequency seismic data, the application of this network can accurately complete the recognition of thin-layer structures with less label data, and the results are clear and reliable with a certain degree of stability. For low-frequency and seismic data, the inversion results of the network with the same number of labels still provide reliable resolution, and the lateral distribution of the thin-layer structure is clear, which can meet the resolution requirements of practical applications.

[0074] Example 2

[0075] like Figure 2 As shown, this embodiment provides a thin layer structure recognition device, which includes a building module 701 , a training module 702 and a recognition module 703 .

[0076] Establishing module 701, used for establishing a thin interbedded geological model, synthesizing seismic records based on the thin interbedded geological model, and constructing training data according to the seismic records;

[0077] A training module 702 is used to establish a convolutional long short-term memory network, and train the convolutional long short-term memory network according to the training data to obtain a thin layer structure recognition model;

[0078] The identification module 703 is used to identify the thin layer structure to be identified by using the thin layer structure identification model.

[0079] In a specific implementation of the present disclosure, the establishment module 701 further includes an establishment unit 7011 and a selection unit 7012 .

[0080] Establishing unit 7011, used for establishing a thin interbed geological model, wherein the thin interbed geological model includes sand bodies deposited in different forms, filling speed and density into the sand bodies deposited in different forms, and obtaining a reflection coefficient model through a reflection coefficient calculation formula of vertical incidence;

[0081] The selection unit 7012 is used to convolve the reflection coefficient model with Ricker wavelets of different main frequencies to obtain seismic records of different resolutions, and select seismic records of a preset number of channels from all seismic records as the training data.

[0082] In a specific implementation of the present disclosure, the training module 702 further includes a marking unit 7021 and a construction unit 7022 .

[0083] The labeling unit 7021 is used to use the impedance of the thin interbed geological model corresponding to each training data as label data of each training data;

[0084] The construction unit 7022 is used to construct the convolutional long short-term memory network, which includes an input layer, a hidden layer, a convolutional long short-term memory layer, and an output layer; the convolutional long short-term memory network is trained using the training data to obtain the thin layer structure recognition model.

[0085] In a specific implementation of the present disclosure, the construction unit 7022 further includes a training unit 70221.

[0086] The training unit 70221 is used to adjust the input dimension of the training data, and send it to the input layer after adjustment. After passing through the input layer, the input data is sent to the hidden layer through convolution calculation, and then the correlation between the seismic data is processed through the long short-term memory layer, and the mapping between the seismic data and the impedance track is completed at the same time; the errors of the impedance and label data obtained by inversion are continuously back-transmitted, and the weight coefficients of each node of the convolutional long short-term memory network are updated to obtain the thin layer structure recognition model.

[0087] It should be noted that, regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0088] Example 3

[0089] Corresponding to the above method embodiment, the embodiment of the present disclosure further provides a thin layer structure recognition device. The thin layer structure recognition device described below and the thin layer structure recognition method described above can refer to each other.

[0090] Figure 3 FIG. 8 is a block diagram of a thin layer structure recognition device 800 according to an exemplary embodiment. Figure 3 As shown, the thin layer structure recognition device 800 may include: a processor 801 and a memory 802. The thin layer structure recognition device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0091] The processor 801 is used to control the overall operation of the thin layer structure recognition device 800 to complete all or part of the steps in the above-mentioned thin layer structure recognition method. The memory 802 is used to store various types of data to support the operation of the thin layer structure recognition device 800, and these data may include, for example, instructions for any application or method operating on the thin layer structure recognition device 800, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the thin layer structure identification device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0092] In an exemplary embodiment, the thin layer structure identification device 800 can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned thin layer structure identification method.

[0093] In another exemplary embodiment, a computer storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned thin layer structure recognition method are implemented. For example, the computer storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by the processor 801 of the thin layer structure recognition device 800 to complete the above-mentioned thin layer structure recognition method.

[0094] Example 4

[0095] Corresponding to the above method embodiment, the embodiment of the present disclosure further provides a storage medium. The storage medium described below and the thin layer structure identification method described above can be referred to in correspondence with each other.

[0096] A storage medium stores a computer program, which, when executed by a processor, implements the steps of the thin layer structure identification method of the above method embodiment.

[0097] The storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other storage medium that can store program codes.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A thin layer structure recognition method, characterized in that: include: Establishing a thin interbedded geological model, synthesizing seismic records based on the thin interbedded geological model, and constructing training data based on the seismic records; Establishing a convolutional long short-term memory network, and training the convolutional long short-term memory network according to the training data to obtain a thin layer structure recognition model; The thin layer structure to be identified is identified using the thin layer structure identification model.

2. The thin layer structure recognition method according to claim 1, characterized in that: Establishing a thin interbedded geological model, synthesizing seismic records based on the thin interbedded geological model, and constructing training data based on the seismic records, including: Establishing a thin interbedded geological model, wherein the thin interbedded geological model includes sand bodies deposited in different forms, filling speed and density into the sand bodies deposited in different forms, and obtaining a reflection coefficient model through a reflection coefficient calculation formula at vertical incidence; The reflection coefficient model is convolved with Ricker wavelets of different main frequencies to obtain seismic records of different resolutions, and seismic records of a preset number of channels are selected from all seismic records as the training data.

3. The thin layer structure recognition method according to claim 1, characterized in that: Establishing a convolutional long short-term memory network, training the convolutional long short-term memory network according to the training data to obtain a thin layer structure recognition model, including: Using the impedance of the thin interbedded geological model corresponding to each of the training data as label data of each of the training data; Constructing the convolutional long short-term memory network, the convolutional long short-term memory network includes an input layer, a hidden layer, a convolutional long short-term memory layer, and an output layer; using the training data to train the convolutional long short-term memory network to obtain the thin layer structure recognition model.

4. The thin layer structure recognition method according to claim 3, characterized in that: The convolutional long short-term memory network is trained using the training data to obtain the thin layer structure recognition model, including: The training data is inputted into the input layer after adjusting the input dimension. After passing through the input layer, the input data is sent to the hidden layer through convolution calculation. The internal correlation of the seismic data is processed through the long short-term memory layer and the mapping with the impedance trace is completed at the same time. The errors of the impedance and label data obtained by inversion are continuously transmitted back, and the weight coefficients of each node of the convolutional long short-term memory network are updated to obtain the thin layer structure recognition model.

5. A thin layer structure recognition device, characterized in that: include: An establishment module is used to establish a thin interbedded geological model, synthesize seismic records based on the thin interbedded geological model, and construct training data according to the seismic records; A training module, used for establishing a convolutional long short-term memory network, training the convolutional long short-term memory network according to the training data, and obtaining a thin layer structure recognition model; The recognition module is used to recognize the thin layer structure to be recognized by using the thin layer structure recognition model.

6. The thin layer structure identification device according to claim 5, characterized in that: Build modules, including: Establishing a unit for establishing a thin interbed geological model, wherein the thin interbed geological model includes sand bodies deposited in different forms, filling speed and density into the sand bodies deposited in different forms, and obtaining a reflection coefficient model through a reflection coefficient calculation formula of vertical incidence; The selection unit is used to convolve the reflection coefficient model with Ricker wavelets of different main frequencies to obtain seismic records of different resolutions, and select seismic records of a preset number of channels from all seismic records as the training data.

7. The thin layer structure identification device according to claim 5, characterized in that: Training modules include: A labeling unit, used for taking the impedance of the thin interbedded geological model corresponding to each of the training data as label data of each of the training data; A construction unit is used to construct the convolutional long short-term memory network, wherein the convolutional long short-term memory network includes an input layer, a hidden layer, a convolutional long short-term memory layer, and an output layer; the convolutional long short-term memory network is trained using the training data to obtain the thin layer structure recognition model.

8. The thin layer structure identification device according to claim 7, characterized in that: Building blocks, including: The training unit is used to adjust the input dimension of the training data, send it to the input layer after adjustment, send the input data to the hidden layer through convolution calculation after passing through the input layer, and then process the internal association of the seismic data through the long short-term memory layer and complete the mapping with the impedance track at the same time; the errors of the impedance and label data obtained by inversion are continuously transmitted back, and the weight coefficients of each node of the convolutional long short-term memory network are updated to obtain the thin layer structure recognition model.

9. A thin layer structure recognition device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the thin layer structure identification method according to any one of claims 1 to 4 when executing the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the thin layer structure identification method according to any one of claims 1 to 4 are implemented.

Images