Seismic wave impedance inversion method and related device
By using a bidirectional temporal convolutional neural network model, combined with residual blocks and dilated convolution, the long-range dependency and gradient problems in existing seismic impedance inversion methods are solved, achieving high-precision and efficient impedance inversion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- EARTH PULSE (NINGBO) TECHNOLOGY CO LTD
- Filing Date
- 2026-01-23
- Publication Date
- 2026-05-12
AI Technical Summary
In existing deep learning-based seismic impedance inversion methods, convolutional neural networks struggle to capture long-range dependencies, recurrent neural networks require high computational power and are prone to gradient problems, and unidirectional sequence modeling does not fully utilize bidirectional contextual information, resulting in low inversion accuracy and efficiency.
A bidirectional temporal convolutional neural network model is adopted, which expands the receptive field by combining forward and backward temporal convolutional branches with residual blocks and dilated convolution. It utilizes bidirectional contextual information from seismic data and introduces an attention mechanism to optimize the training process and improve model accuracy and efficiency.
It effectively captures long-range dependencies in seismic data, improves the accuracy and efficiency of wave impedance inversion, overcomes the dependence of traditional methods on initial models and assumptions, and achieves high-precision wave impedance inversion.
Smart Images
Figure CN122017945A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the interdisciplinary field of geophysical exploration and artificial intelligence, and specifically relates to a seismic wave impedance inversion method and related apparatus. Background Technology
[0002] Impedance inversion is a key technology in seismic exploration. Its task is to convert seismic records into impedance data, thereby revealing the characteristics of subsurface lithology and fluid distribution. Currently, traditional impedance inversion methods mainly include trace integral inversion, sparse pulse inversion, and model-based inversion. These methods eliminate the wavelet influence in seismic data, retain reflection coefficient information, and then calculate the physical parameters characterizing changes in formation properties, i.e., impedance, from the reflection coefficients. However, these methods often rely on theoretical approximations during implementation, are sensitive to initial models and assumptions, and suffer from high computational complexity and low efficiency.
[0003] In recent years, with the development of artificial intelligence technology, deep learning has been gradually introduced into seismic impedance inversion. For example, methods such as convolutional neural networks and recurrent neural networks have been initially applied and achieved certain results. However, existing deep learning-based inversion methods still have the following limitations: First, convolutional neural networks have a limited receptive field, making it difficult to effectively capture long-range dependencies in seismic data; second, recurrent neural networks require high computational power during training and are prone to gradient vanishing or gradient exploding problems; in addition, convolutional neural networks and recurrent neural networks only use a unidirectional sequence modeling approach, failing to fully utilize the bidirectional contextual information contained in seismic data, thus limiting the inversion accuracy. Summary of the Invention
[0004] To address the problems existing in the prior art, this invention provides a seismic impedance inversion method and related apparatus. The purpose is to solve the problems in existing deep learning-based inversion methods, such as the difficulty of convolutional neural networks in capturing long-range dependencies, the high computational requirements of recurrent neural networks and their susceptibility to gradient problems, and the failure of unidirectional sequence modeling to fully utilize bidirectional contextual information, which restricts the accuracy of inversion, thereby improving the accuracy and efficiency of seismic impedance inversion.
[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: According to a first aspect of the present invention, a seismic wave impedance inversion method is provided, comprising: Obtain the seismic data to be inverted; The seismic data to be inverted is input into a pre-trained bidirectional temporal convolutional neural network model to obtain the predicted wave impedance data. The bidirectional temporal convolutional neural network model includes a forward temporal convolutional branch, a backward temporal convolutional branch, and a feature merging layer. Both the forward and backward temporal convolutional branches include multiple residual blocks, and each residual block uses dilated convolution. The forward temporal convolutional branch is used to process the input seismic data sequence in forward order, and the backward temporal convolutional branch is used to process the input seismic data sequence in reverse order. The feature merging layer is used to merge the outputs of the forward and backward temporal convolutional branches.
[0006] In one possible implementation of the first aspect, the seismic data to be inverted is standardized before being input into the bidirectional temporal convolutional neural network model, the standardization process including converting the seismic data to be inverted into data conforming to a standard normal distribution.
[0007] In one possible implementation of the first aspect, the dilation rate of the dilated convolution increases exponentially with increasing network depth.
[0008] In one possible implementation of the first aspect, an attention mechanism is introduced after the feature merging layer to perform weighted processing on the merged features.
[0009] In one possible implementation of the first aspect, the bidirectional temporal convolutional neural network model is trained using training sample pairs, which include seismic data sequences and corresponding real wave impedance data, and both the seismic data sequences and the real wave impedance data are standardized.
[0010] In one possible implementation of the first aspect, the seismic data sequence and the real wave impedance data in the training sample pair are windowed with the well location as the center, and the sequence lengths are the same.
[0011] In one possible implementation of the first aspect, the training process uses an early stopping mechanism and a learning rate decay strategy.
[0012] According to a second aspect of the present invention, a seismic impedance inversion apparatus is provided, comprising: The acquisition module is used to acquire the seismic data to be inverted; The prediction module is used to input the seismic data to be inverted into a pre-trained bidirectional temporal convolutional neural network model to obtain predicted wave impedance data. The bidirectional temporal convolutional neural network model includes a forward temporal convolutional branch, a backward temporal convolutional branch, and a feature merging layer. Both the forward and backward temporal convolutional branches include multiple residual blocks, and each residual block uses dilated convolution. The forward temporal convolutional branch is used to process the input seismic data sequence in forward order, and the backward temporal convolutional branch is used to process the input seismic data sequence in reverse order. The feature merging layer is used to merge the outputs of the forward and backward temporal convolutional branches.
[0013] In one possible implementation of the second aspect, the seismic data to be inverted is standardized before being input into the bidirectional temporal convolutional neural network model, the standardization process including converting the seismic data to be inverted into data conforming to a standard normal distribution.
[0014] In one possible implementation of the second aspect, the dilation rate of the dilated convolution increases exponentially with increasing network depth.
[0015] In one possible implementation of the second aspect, an attention mechanism is introduced after the feature merging layer to perform weighted processing on the merged features.
[0016] In one possible implementation of the second aspect, the bidirectional temporal convolutional neural network model is trained using training sample pairs, which include seismic data sequences and corresponding real wave impedance data, and both the seismic data sequences and the real wave impedance data are standardized.
[0017] In one possible implementation of the second aspect, the seismic data sequence and the real wave impedance data in the training sample pair are windowed with the well location as the center, and the sequence lengths are the same.
[0018] In one possible implementation of the second aspect, the training process uses an early stopping mechanism and a learning rate decay strategy.
[0019] According to a third aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned seismic impedance inversion method.
[0020] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned seismic wave impedance inversion method.
[0021] According to a fifth aspect of the present invention, a computer program product is provided that, when executed by a processor, implements the aforementioned seismic impedance inversion method.
[0022] Compared with the prior art, the present invention has at least the following beneficial effects: This invention provides a seismic impedance inversion method that expands the model's receptive field by employing dilated convolution in the residual blocks of both the forward and backward temporal convolution branches. This effectively captures long-range dependencies in seismic data, overcoming the limitation of traditional convolutional neural networks in modeling remote geological features due to their limited receptive field. Replacing recurrent neural networks with temporal convolutional networks avoids excessive computational demands during training and eliminates gradient vanishing or exploding problems, improving model training efficiency and making it suitable for large-scale seismic data processing. By using a bidirectional sequence modeling approach, simultaneously processing both forward and reverse sequences of seismic data, the bidirectional contextual information inherent in the seismic data is fully extracted, enhancing the completeness of feature representation and thus improving the accuracy of impedance inversion. The residual block design in the model effectively alleviates the gradient degradation problem in deep network training, promoting rapid model convergence and further ensuring the reliability of the inversion results. This invention achieves high-precision impedance inversion while maintaining low computational complexity, overcoming the excessive reliance on initial models and assumptions in traditional methods.
[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0024] To more clearly illustrate the technical solutions in the specific embodiments of the present invention, the drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0025] Figure 1 This is a flowchart of a seismic wave impedance inversion method according to the present invention; Figure 2 The examples show simulated seismic data and corresponding wave impedance models. Figure 3 The learning curves of the 1D-CNN and Bidirectional TCN (BiTCN) models during the training process are shown in the example. Figure 4 The comparison results of predicted wave impedance between 1D-CNN and BiTCN in the example are shown. Figure 5 and Figure 6This example compares the wave impedance prediction performance of 1D-CNN and BiTCN on a single channel above the well. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] like Figure 1 As shown, this invention provides a seismic impedance inversion method, implemented using a pre-trained bidirectional temporal convolutional neural network model. The pre-trained bidirectional temporal convolutional neural network model can effectively capture long-range dependencies and bidirectional contextual information in seismic data, thereby improving the accuracy and efficiency of impedance inversion. The specific implementation process is as follows: Step 1: Acquire and preprocess seismic data and corresponding wave impedance data to provide standardized samples for training the bidirectional temporal convolutional neural network model. Specific sub-steps include: 1.1 Acquiring seismic data of the target layer Based on the acoustic and density logging data at the well location within the target layer, the corresponding acoustic impedance data were obtained after depth-time conversion. .
[0028] 1.2 Selecting Seismic Data Sequences at Well Locations , and the wave impedance data sequence at the well location Together, they form training sample pairs. Among them, earthquake data sequences The data structure is (1, n) timesteps , n features ), representing sequence length and characteristic number; wave impedance data sequence The data structure is (1, nt) imesteps , 1), representing the sequence length and single output channel.
[0029] 1.3 Training Sample Pairs Standardization preprocessing is performed to transform the data into a form that conforms to a standard normal distribution, thereby promoting the convergence of the bidirectional temporal convolutional neural network model.
[0030] 1.4 For pairs of training samples The learning samples are grouped, with 80% used for training the bidirectional temporal convolutional neural network model and 20% used for validating the bidirectional temporal convolutional neural network model, thus obtaining the training set and validation set.
[0031] Step 2: Construct a bidirectional temporal convolutional neural network model (bidirectional TCN model). The bidirectional TCN model includes an input layer, a bidirectional TCN layer, a feature merging layer, and an output layer. The bidirectional TCN layer consists of a forward temporal convolutional branch (forward TCN branch) and a backward temporal convolutional branch (backward TCN branch). Both the forward and backward TCN branches contain multiple residual blocks, and dilated convolutions are used in the residual blocks to expand the receptive field. The outputs of the forward and backward TCN branches are merged by the feature merging layer, and an attention mechanism is used to enhance feature extraction capabilities. Specific sub-steps include: 2.1 An input layer is constructed based on the size of the earthquake data sequence. A one-dimensional convolutional layer is added after the input layer to initially extract earthquake data features.
[0032] 2.2 Construct a forward TCN branch, which consists of multiple residual blocks. Dilated convolutions are used layer by layer in each residual block, and the dilation rate of the dilated convolutions increases exponentially with the network depth to gradually expand the receptive field.
[0033] 2.3 Constructing a reverse TCN branch: Except for the input seismic data sequence being processed in reverse order, the structure of the reverse TCN branch is the same as that of the forward TCN branch, ensuring full utilization of bidirectional contextual information.
[0034] 2.4 Merge the bidirectional features obtained from the forward TCN branch and the reverse TCN branch respectively, and generate a feature representation that integrates bidirectional contextual information through the feature merging layer.
[0035] 2.5 An attention mechanism is introduced after the feature merging layer to weight the merged features with fused bidirectional contextual information, thereby enhancing the extraction capability of key seismic features (such as thin layers and complex geological bodies). Subsequently, an output layer composed of multiple one-dimensional convolutions is constructed to further refine the information in the seismic data sequence and finally obtain the wave impedance output.
[0036] 2.6 Determine the hyperparameters of the model, such as the activation functions of each layer, the parameters of the one-dimensional convolutional layer, and the loss function, to ensure the optimization of the model structure.
[0037] Step 3: Train the bidirectional TCN model using training data, optimize model parameters through backpropagation and gradient descent algorithms, and employ network training measures to prevent overfitting. Specific sub-steps include: 3.1 Determine the optimization function (for example, the Adam optimizer) and set an appropriate learning rate to control the parameter update step size.
[0038] 3.2 Set up an early stopping mechanism and a learning rate decay strategy to monitor the performance of the validation set during training. When the performance no longer improves, terminate training early or reduce the learning rate to prevent the model from overfitting and improve the model's generalization ability.
[0039] Step 4: Input the seismic data to be inverted into the trained two-way TCN model to obtain the predicted wave impedance data. Specific sub-steps include: 4.1 Perform standardized preprocessing on unknown earthquake data in the same way as the training data to ensure consistent data distribution.
[0040] 4.2 The preprocessed seismic data is input into the trained bidirectional TCN model. The bidirectional TCN model performs inference based on the fused bidirectional features and outputs the seismic impedance prediction results for the input data.
[0041] This invention provides a seismic impedance inversion method based on a bidirectional temporal convolutional neural network (TCN) through the above-described embodiments. It utilizes both forward and backward TCN branches, along with contextual information from the seismic data, to more accurately capture the correlation between different seismic sequences. Dilated convolution and residual connections are used to expand the receptive field and suppress the gradient vanishing problem in deep networks. The introduced attention mechanism further enhances the model's ability to identify thin layers and complex geological bodies, thereby improving inversion accuracy and efficiency.
[0042] The seismic wave impedance inversion method of the present invention will be described in detail below with reference to the accompanying drawings and simulation data embodiments.
[0043] Figure 2 The simulation data used in this embodiment is shown, wherein, Figure 2 (a) in the figure represents the seismic data obtained by convolution and the addition of random noise. Figure 2 (b) in the diagram represents the actual impedance model corresponding to the seismic data. This impedance model includes six strata and typical geological structures such as a river channel, fault, and thin layers. The purpose of this invention is to... Figure 2 In the seismic data shown in (a), high-precision inversion was performed. Figure 2 The wave impedance data shown in (b) is as follows.
[0044] The specific implementation process of this embodiment includes the following four steps: Step 1: Acquire and preprocess seismic data and corresponding wave impedance data to provide standardized learning samples for the subsequent training of the bidirectional temporal convolutional neural network model.
[0045] 1.1 Acquiring seismic data of the target layer Based on the acoustic logging and density logging data at the well location within the target layer area, the corresponding acoustic impedance data were obtained after depth-time conversion. .
[0046] 1.2 A time window of length 200 is opened centered on each well location to extract the seismic data sequence at that well location. Simultaneously, extract the wave impedance data sequence at the well location within the same time range. Together they constitute the training sample pairs Since the input feature is only earthquake amplitude, the earthquake data sequence The data structure is (1, 200, 1), and the wave impedance data sequence is... The data structure is also (1, 200, 1).
[0047] 1.3 The training sample pairs are standardized and preprocessed to transform the data into a form conforming to a standard normal distribution. The transformation function is as follows:
[0048] in, instruct and The value, and These are the mean and variance of the sample data, respectively.
[0049] 1.4 The learning sample set, consisting of multiple training sample pairs, is grouped, with 80% of the samples used for model training and 20% used for model validation, thus obtaining the training set and validation set.
[0050] Step 2: The constructed bidirectional temporal convolutional neural network model includes an input layer, a bidirectional TCN layer, and an output layer. The bidirectional TCN layer consists of a forward TCN branch and a backward TCN branch. By merging bidirectional features and introducing an attention mechanism, the ability to extract key seismic features is enhanced.
[0051] 2.1 The input layer is constructed based on the size of the earthquake data sequence. After the input layer, a one-dimensional convolutional layer is added for preliminary feature extraction, with the kernel size set to 1. It should be noted that, to enhance the convergence and stability of the model, a batch normalization layer is added after each one-dimensional convolutional layer.
[0052] 2.2 Constructing the Forward TCN Branch. The forward TCN branch consists of three residual blocks connected in series. Each residual block has three one-dimensional convolutional layers as its basic structure. Simultaneously, each residual block uses dilated convolution layer by layer, and the dilation rate of the three convolutional layers within the residual block increases exponentially by 2, i.e., 2^n times per layer. 0 ,2 1 ,2 2 This gradually expands the receptive field of the model.
[0053] 2.3 Constructing the Reverse TCN Branch. The input to the reverse TCN branch is the seismic data sequence after being reversed in time order. Its network structure is exactly the same as that of the forward TCN branch to ensure that the model can make full use of the bidirectional contextual information of the seismic data.
[0054] 2.4 The feature merging layer merges the bidirectional features extracted by the forward TCN branch and the reverse TCN branch respectively, generating a feature representation that incorporates bidirectional contextual information.
[0055] 2.5 After the feature merging layer, an attention mechanism is introduced to weight the merged features, further enhancing the model's ability to extract key geological features such as thin layers and river channels. Subsequently, an output layer composed of multiple one-dimensional convolutions is constructed to further refine the information in the feature sequence and finally output the predicted wave impedance value.
[0056] 2.6 Determine the activation functions, one-dimensional convolutional layer parameters, and loss function for each layer of the model. Specifically, except for the output layer, which uses a linear activation function, all other activation functions in the model use the ReLU function. Regarding the convolutional kernel settings, except for the input and output layers where the kernel size is 1, all other one-dimensional convolutional layers have a kernel size of 3. The loss function for the model is determined to be the mean squared error loss function.
[0057] Step 3: Train the constructed bidirectional TCN model using the training data prepared in Step 1. Optimize the model parameters through backpropagation and gradient descent algorithms, and adopt network training measures to prevent overfitting.
[0058] 3.1 The optimization function of the model is determined to be the Adam optimizer, and an adaptive learning rate is set for it to effectively control the step size of parameter updates.
[0059] 3.2 During training, an early stopping mechanism and a learning rate decay strategy are implemented. By monitoring performance on the validation set in real time, training is terminated early or the learning rate is automatically reduced when performance no longer improves, thereby effectively preventing model overfitting and improving its generalization ability.
[0060] To verify the superiority of the method proposed in this invention, this embodiment selects a one-dimensional convolutional neural network (1D-CNN) as a comparison method. Figure 3 The learning curves of the two models during the training process are shown. Figure 3 In the diagram, (a) shows the learning curve of the 1D-CNN during the training process. Figure 3Figure (b) shows the learning curve of the bidirectional TCN model used in this invention during the training process. The blue curve represents the mean squared error on the training set, and the orange curve represents the mean squared error on the validation set. It can be seen that the learning curve of the bidirectional TCN model used in this invention is smoother and exhibits better convergence.
[0061] Step 4: Input the seismic data to be inverted into the trained two-way TCN model to obtain the predicted wave impedance data.
[0062] 4.1 For unknown seismic data, a standardized preprocessing method that is completely consistent with the training data is used to ensure the consistency of data distribution.
[0063] 4.2 The preprocessed seismic data is input into the trained bidirectional TCN model. The trained bidirectional TCN model performs inference calculations based on its fused bidirectional features, and finally outputs the wave impedance prediction results for the input seismic data.
[0064] Figure 4 The comparison of wave impedance prediction performance between 1D-CNN and bidirectional TCN models across the entire measurement line is shown. Figure 4 In the diagram, (a) represents the predicted wave impedance of the 1D-CNN. Figure 4 In the diagram, (b) represents the prediction impedance error of the 1D-CNN. Figure 4 In the diagram, (c) represents the predicted wave impedance of the bidirectional TCN model. Figure 4 In the figure, (d) represents the predicted impedance error of the two-way TCN model. It can be seen that the predicted impedance error of the two-way TCN model is smaller and the predicted impedance result is closer to the actual impedance. It performs better, especially at drastic changes such as stratum boundaries, and can better represent the true distribution of underground impedance.
[0065] For further analysis, Figure 5 and Figure 6 The performance of 1D-CNN and bidirectional TCN models in predicting wave impedance on a single well track is compared. The results show that the proposed method has a better ability to capture detailed changes in wave impedance, more accurate prediction of these changes, and a higher correlation coefficient between the predicted wave impedance and the actual wave impedance.
[0066] In another embodiment of the present invention, a seismic impedance inversion device is provided for implementing the above-mentioned seismic impedance inversion method, specifically including: The acquisition module is used to acquire the seismic data to be inverted.
[0067] The prediction module is used to input the seismic data to be inverted into a pre-trained bidirectional temporal convolutional neural network model to obtain predicted wave impedance data. The bidirectional temporal convolutional neural network model includes a forward temporal convolutional branch, a backward temporal convolutional branch, and a feature merging layer. Both the forward and backward temporal convolutional branches include multiple residual blocks, and each residual block uses dilated convolution. The forward temporal convolutional branch is used to process the input seismic data sequence in forward order, and the backward temporal convolutional branch is used to process the input seismic data sequence in reverse order. The feature merging layer is used to merge the outputs of the forward and backward temporal convolutional branches.
[0068] All relevant content regarding the steps involved in the aforementioned embodiment of the seismic impedance inversion method can be referenced from the functional description of the corresponding functional module of the seismic impedance inversion device in this embodiment of the invention, and will not be repeated here. The module division in this embodiment of the invention is illustrative and only represents a logical functional division. In actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a processor, exist as separate physical entities, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module.
[0069] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a seismic impedance inversion method.
[0070] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the seismic wave impedance inversion method in the above embodiments.
[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.
[0072] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0075] This invention also provides a computer program product for executing any of the aforementioned seismic impedance inversion methods. Since the computer program product provided by this invention belongs to the same inventive concept as the aforementioned seismic impedance inversion method, it possesses all the advantages of the aforementioned seismic impedance inversion method. Therefore, the beneficial effects of the computer program product provided by this invention will not be elaborated upon here.
[0076] In this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0077] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.
Claims
1. A seismic wave impedance inversion method, characterized in that, include: Obtain the seismic data to be inverted; The seismic data to be inverted is input into a pre-trained bidirectional temporal convolutional neural network model to obtain the predicted wave impedance data. The bidirectional temporal convolutional neural network model includes a forward temporal convolutional branch, a backward temporal convolutional branch, and a feature merging layer. Both the forward and backward temporal convolutional branches include multiple residual blocks, and each residual block uses dilated convolution. The forward temporal convolutional branch is used to process the input seismic data sequence in forward order, and the backward temporal convolutional branch is used to process the input seismic data sequence in reverse order. The feature merging layer is used to merge the outputs of the forward and backward temporal convolutional branches.
2. The seismic wave impedance inversion method according to claim 1, characterized in that, The seismic data to be inverted is standardized before being input into the bidirectional temporal convolutional neural network model. The standardization process includes converting the seismic data to be inverted into data that conforms to a standard normal distribution.
3. The seismic wave impedance inversion method according to claim 1, characterized in that, The dilation rate of the dilated convolution increases exponentially with increasing network depth.
4. The seismic wave impedance inversion method according to claim 1, characterized in that, An attention mechanism is introduced after the feature merging layer to perform weighted processing on the merged features.
5. The seismic wave impedance inversion method according to claim 1, characterized in that, The bidirectional temporal convolutional neural network model is obtained by training using training sample pairs, which include seismic data sequences and corresponding real wave impedance data, and both the seismic data sequences and real wave impedance data have been standardized.
6. The seismic wave impedance inversion method according to claim 5, characterized in that, The seismic data sequences and real wave impedance data in the training sample pairs are extracted by windowing with the well location as the center, and the sequence lengths are the same.
7. The seismic wave impedance inversion method according to claim 5, characterized in that, The training process uses an early stopping mechanism and a learning rate decay strategy.
8. A seismic wave impedance inversion device, characterized in that, include: The acquisition module is used to acquire the seismic data to be inverted; The prediction module is used to input the seismic data to be inverted into a pre-trained bidirectional temporal convolutional neural network model to obtain predicted wave impedance data. The bidirectional temporal convolutional neural network model includes a forward temporal convolutional branch, a backward temporal convolutional branch, and a feature merging layer. Both the forward and backward temporal convolutional branches include multiple residual blocks, and each residual block uses dilated convolution. The forward temporal convolutional branch is used to process the input seismic data sequence in forward order, and the backward temporal convolutional branch is used to process the input seismic data sequence in reverse order. The feature merging layer is used to merge the outputs of the forward and backward temporal convolutional branches.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a seismic wave impedance inversion method as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements a seismic wave impedance inversion method as described in any one of claims 1 to 7.