Seismic wave impedance inversion method, device and electronic equipment

Abstract

The application provides a seismic wave impedance inversion method and device and electronic equipment, and relates to the technical field of seismic inversion, and comprises the following steps: obtaining real seismic data of a to-be-processed work area, performing discrete processing on the real seismic data to obtain a plurality of single-channel seismic data, processing the plurality of single-channel seismic data by using a target inversion model, and obtaining a seismic wave impedance inversion result of the to-be-processed work area. The target inversion model is a model obtained by training an initial inversion model. Obviously, the method is essentially a data-driven inversion method, so the dependence on the initial model is avoided. In addition, the encoding process of the encoder module and the decoding process of the decoder module in the target inversion model are both implemented based on the mamba module. Since the mamba module provides strong feature extraction and representation capabilities for the encoder, compared with the prior art, the application can ensure the accuracy of the seismic wave impedance inversion result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of seismic inversion, and in particular to a seismic wave impedance inversion method, device and electronic equipment. BACKGROUND

[0002] Seismic inversion needs to extract geophysical properties and spatial structures from seismic images. These inversions are crucial for evaluating geological features and reservoir properties, especially in underground environments. P-wave impedance is a key elastic property that reflects reservoir characteristics and is an indispensable parameter in reservoir exploration and identification.

[0003] In the prior art, traditional inversion techniques based on models are often used, such as recursive inversion, trace integration, and geostatistical inversion methods. However, these methods highly depend on the initial model and require careful design of regularization terms. If the initial model is inaccurate, the inversion results may be biased. In addition, problems such as frequency band mismatch, forward simulation approximation, and noise pollution make it difficult to produce stable and high-resolution inversion results from field seismic data. In summary, the existing seismic wave impedance inversion method has the technical problems of strong dependence on the initial model and poor accuracy of the inversion result. SUMMARY

[0004] The present application aims to provide a seismic wave impedance inversion method, device and electronic equipment to alleviate the technical problems of strong dependence on the initial model and poor accuracy of the inversion result in the existing seismic wave impedance inversion method.

[0005] In a first aspect, the present application provides a seismic wave impedance inversion method, comprising: obtaining real seismic data of a to-be-processed work area; performing discrete processing on the real seismic data to obtain a plurality of single-channel seismic data; processing the plurality of single-channel seismic data using a target inversion model to obtain a seismic wave impedance inversion result of the to-be-processed work area; wherein the target inversion model is a model obtained by training an initial inversion model; the target inversion model comprises an encoder module, a first projection module, a decoder module and a second projection module; the encoding process of the encoder module and the decoding process of the decoder module are both realized based on a mamba module; the first projection module is used to map the output of the encoder module to the input space of the decoder module; and the second projection module is used to map the output of the decoder module to the seismic wave impedance domain.

[0006] Optionally, the encoder module and the decoder module are a mamba encoder and a mamba decoder in a trained mask autoencoder (MAE); and the training process of the MAE comprises: obtaining a first training sample set; wherein the first training sample set comprises a plurality of sample single-channel seismic data; performing block processing on a first target sample single-channel seismic data to obtain a discrete data volume; wherein the first target sample single-channel seismic data represents any sample in the first training sample set; performing random mask processing on the discrete data volume to obtain a recovery index of the masked data, an unmasked data set, and a position marker of each unmasked data in the discrete data volume; performing encoding processing on the unmasked data set by using the mamba encoder to obtain latent features of the unmasked data set; constructing input data of the mamba decoder based on the latent features, the position marker of each unmasked data in the discrete data volume, and the recovery index of the masked data; decoding the input data by using the mamba decoder to reconstruct single-channel seismic data based on a decoding result to obtain a single-channel seismic data prediction value; calculating a first loss function value based on the first target sample single-channel seismic data and the single-channel seismic data prediction value; and iteratively training the initial MAE based on the first loss function value until a preset end condition is reached to obtain a target MAE.

[0007] Optionally, the first loss function value is calculated based on the first target sample single-channel seismic data and the single-channel seismic data prediction value, comprising: selecting corresponding prediction data from the single-channel seismic data prediction value based on the position of the masked data in the first target sample single-channel seismic data; calculating the Euclidean distance between the masked data and the prediction data; and determining the Euclidean distance as the first loss function value.

[0008] Optionally, the method further comprises: obtaining a second training sample set; wherein the second training sample set comprises a plurality of second training samples, and each second training sample comprises: sample single-channel seismic data and corresponding real seismic wave impedance data; processing the second target sample single-channel seismic data by using the initial inversion model to obtain corresponding seismic wave impedance prediction data; wherein the second target sample single-channel seismic data represents any sample in the second training sample set; calculating a second loss function value based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data, and the seismic wave impedance prediction data; and iteratively training the initial inversion model based on the second loss function value to continuously adjust the model parameters of the first projection module and the second projection module until a preset iteration termination condition is reached.

[0009] Optionally, based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data, a second loss function value is calculated, including: performing forward processing on the seismic wave impedance prediction data by using a preset forward operator to obtain synthetic single-channel seismic data; calculating a residual error between the synthetic single-channel seismic data and the second target sample single-channel seismic data to obtain a reconstruction loss; calculating a mean square error between the real seismic wave impedance data and the seismic wave impedance prediction data to obtain a well loss; and calculating the second loss function value based on the reconstruction loss and the well loss.

[0010] Optionally, the second loss function is represented as: ; wherein, all represent preset weight factors, represents the reconstruction loss, represents the well loss.

[0011] Optionally, a sparsity constraint is added to a hidden layer of the encoder module.

[0012] In a second aspect, the present application provides a seismic wave impedance inversion device, including: a first acquisition module, configured to acquire real seismic data of a to-be-processed work area; a discretization module, configured to perform discretization processing on the real seismic data to obtain a plurality of single-channel seismic data; and an inversion module, configured to process the plurality of single-channel seismic data by using a target inversion model to obtain a seismic wave impedance inversion result of the to-be-processed work area; wherein the target inversion model is a model obtained by training an initial inversion model; the target inversion model includes an encoder module, a first projection module, a decoder module and a second projection module; the encoding process of the encoder module and the decoding process of the decoder module are both realized based on a mamba module; the first projection module is configured to map the output of the encoder module to the input space of the decoder module; and the second projection module is configured to map the output of the decoder module to the seismic wave impedance domain.

[0013] In a third aspect, the present application provides an electronic device, including a memory and a processor, the memory stores a computer program capable of running on the processor, and the processor realizes the seismic wave impedance inversion method of any one of the preceding embodiments when executing the computer program.

[0014] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions realize the seismic wave impedance inversion method of any one of the preceding embodiments when executed by a processor.

[0015] The present application provides a seismic wave impedance inversion method, after obtaining the real seismic data of the processing area, first, the real seismic data is discretized to obtain a plurality of single channel seismic data, and then the target inversion model is used to process the plurality of single channel seismic data to obtain the seismic wave impedance inversion result of the processing area, wherein the target inversion model is obtained by training the initial inversion model. Therefore, the present application is essentially a data-driven inversion method, which avoids the dependence on the initial model. In addition, the target inversion model includes an encoder module, a first projection module, a decoder module and a second projection module, and the encoding process of the encoder module and the decoding process of the decoder module are realized based on the mamba module. Since the mamba module has higher efficiency and scalability than the traditional self-attention mechanism when processing long sequence data, while maintaining good modeling ability, it provides strong feature extraction and representation ability for the encoder, so compared with the prior art, the present application can ensure the accuracy of the seismic wave impedance inversion result. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0017] Figure 1 A flowchart of a seismic wave impedance inversion method provided by an embodiment of the present application;

[0018] Figure 2 A network structure diagram of MAE provided by an embodiment of the present application;

[0019] Figure 3 A schematic diagram of the seismic wave impedance distribution of Marmousi2 model;

[0020] Figure 4 A schematic diagram of the seismic wave impedance distribution of Marmousi2 model; Figure 3 A schematic diagram of the synthetic seismic data calculated;

[0021] Figure 5 A schematic diagram of the difference between the prediction results of four seismic wave impedance prediction methods and the true value;

[0022] Figure 6 A schematic diagram of the difference between the prediction results of four seismic wave impedance prediction methods and the true value;

[0023] Figure 7 A functional module diagram of a seismic wave impedance inversion device provided by an embodiment of the present application;

[0024] Figure 8 A schematic diagram of an electronic device is provided for embodiments of the present application. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0027] Some embodiments of the present application will be described in detail below with reference to the drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

[0028] Embodiment one

[0029] Figure 1 A flowchart of a seismic wave impedance inversion method is provided for embodiments of the present application, as shown in Figure 1 The method specifically includes the following steps:

[0030] Step S102, real seismic data of a work area to be processed is obtained.

[0031] Step S104, the real seismic data is discretized to obtain a plurality of single-channel seismic data.

[0032] Step S106, the plurality of single-channel seismic data is processed using a target inversion model to obtain a seismic wave impedance inversion result of the work area to be processed.

[0033] The target inversion model is a model obtained by training an initial inversion model; the target inversion model includes an encoder module, a first projection module, a decoder module and a second projection module; the encoding process of the encoder module and the decoding process of the decoder module are both realized based on a mamba module; the first projection module is used to map the output of the encoder module to the input space of the decoder module; and the second projection module is used to map the output of the decoder module to the seismic wave impedance domain.

[0034] Specifically, in order to avoid the dependence of the traditional model-driven inversion method on the initial model, the embodiment of the present application applies a data-driven inversion method when performing seismic wave impedance inversion, that is, a target inversion model with single-channel seismic data as input data and corresponding seismic wave impedance data as output data is trained in advance. Considering that the input data of the target inversion model is single-channel seismic data and the single-channel seismic data is long sequence data, in order to ensure the scalability of the model and the data processing efficiency, in the embodiment of the present application, the encoder module and the decoder module in the target inversion model are both implemented based on the mamba module. The application of the mamba module can also avoid the problems of gradient disappearance and lack of effective parallelism of the target inversion model.

[0035] It is known that in the encoder-decoder structure model, the output of the encoder needs to be input into the decoder after projection. Therefore, a first projection module is arranged between the encoder module and the decoder module to map the output of the encoder module to the input space of the decoder module. In addition, after the decoder module decodes the output, the output result should also be mapped to the seismic wave impedance domain to obtain the seismic wave impedance inversion result. Therefore, a second projection module is also needed after the decoder module to realize the mapping between the data domains. In an alternative embodiment, the first projection module and the second projection module are both implemented by using a fully connected layer.

[0036] Since the input data of the target inversion model is single-channel seismic data, after obtaining the real seismic data of the processing area, the real seismic data needs to be discretized. It is known that the real seismic data is a set of seismic data recorded by multiple seismic channels. Therefore, the discretization processing is to segment the real seismic data by seismic channel, thereby obtaining multiple single-channel seismic data. Then the target inversion model is used to process the multiple single-channel seismic data, and the corresponding seismic wave impedance inversion result is obtained.

[0037] The embodiment of the present application provides a seismic wave impedance inversion method, which comprises the following steps: after obtaining real seismic data of a to-be-processed work area, the real seismic data is discretized to obtain a plurality of single-channel seismic data, and then the target inversion model is used to process the plurality of single-channel seismic data to obtain the seismic wave impedance inversion result of the to-be-processed work area, wherein the target inversion model is obtained by training the initial inversion model. Therefore, the present application is essentially a data-driven inversion method, which avoids the dependence on the initial model. In addition, the target inversion model comprises an encoder module, a first projection module, a decoder module and a second projection module, and the encoding process of the encoder module and the decoding process of the decoder module are both based on the mamba module. Since the mamba module has higher efficiency and scalability than the traditional self-attention mechanism when processing long sequence data, and can also maintain good modeling ability, the mamba module provides strong feature extraction and representation ability for the encoder, so that the present application can ensure the accuracy of the seismic wave impedance inversion result compared with the prior art.

[0038] In an optional implementation, the encoder module and the decoder module are mamba encoders and mamba decoders in a trained masked autoencoder (MAE).

[0039] The training process of the MAE comprises the following steps:

[0040] In step S201, a first training sample set is obtained, wherein the first training sample set comprises a plurality of sample single-channel seismic data.

[0041] In step S202, the first target sample single-channel seismic data in the first training sample set is processed by block processing to obtain a discrete data body.

[0042] In step S203, the discrete data body is processed by random mask processing to obtain a recovery index of the mask data, an unmasked data set and a position marker of each unmasked data in the discrete data body.

[0043] In step S204, the unmasked data set is processed by encoding using the mamba encoder to obtain the latent features of the unmasked data set.

[0044] In step S205, the input data of the mamba decoder is constructed based on the latent features, the position marker of each unmasked data in the discrete data body and the recovery index of the mask data.

[0045] In step S206, the input data is decoded using the mamba decoder to reconstruct the single-channel seismic data based on the decoding result to obtain the single-channel seismic data prediction value.

[0046] Step S207, based on the first target sample single-channel seismic data and the single-channel seismic data prediction value, a first loss function value is calculated.

[0047] Step S208, based on the first loss function value, the initial MAE is iteratively trained until a preset ending condition is reached, and a target MAE is obtained.

[0048] Generally, to obtain the target inversion model, the model parameters of the encoder module, the first projection module, the decoder module and the second projection module in the initial inversion model can be initialized first, and then the model parameters of the above four modules are iteratively updated through a large number of sample training, and the target inversion model obtained in this way can realize the inversion from single-channel seismic data to seismic wave impedance data. In order to enhance the understanding ability of the target inversion model to the seismic signal and improve the inversion precision, the embodiment of the application is different from the traditional model construction and training process. Before constructing the initial inversion model, the mask autoencoder MAE using the mamba encoder and the mamba decoder needs to be pre-trained. After the MAE reaches the preset training target, the trained mamba encoder and mamba decoder are used as the encoder module and the decoder module in the initial inversion model, respectively. Then the model parameters of the first projection module and the second projection module are continuously optimized through a large number of sample training, and then the target inversion model is obtained.

[0049] Multiple scattering and complex wave field interference in seismic signals can cause spatial blurring, resulting in a large amount of redundant information in seismic data. This process is similar to the image blurring and information redundancy phenomenon caused by light scattering and reflection in natural images. Due to this spatial redundancy, even if most of the seismic data is masked, the original data can still be recovered through the MAE.

[0050] In an optional implementation, a sparsity constraint is added to the hidden layer of the encoder module, that is, only part of the data is allowed to participate in the calculation. This mechanism prioritizes the part of the input data with the most information, while ignoring less important details, reducing redundant information, and thus achieving a more efficient and focused encoding process. This selective activation ensures that only the most relevant features are processed, further improving the efficiency and interpretability of the model.

[0051] Generally, the MAE uses the ratio of embedding dimension to patch size to maintain a similar size to the original data. However, the embodiment of the application uses a higher ratio of embedding dimension to patch size to greatly increase the information capacity and expand the seismic representation space. Figure 2 The network structure diagram of the MAE provided by the embodiment of the application is shown in Figure 2In the training of the MAE, first, the first target sample single-channel seismic data needs to be block processed, that is, discretized into a plurality of amplitude data to obtain a discrete data body. Next, a random mask operation is performed on the discrete data body to mask part of the data therein. Thus, an unmasked data set, a position marker of each unmasked data in the discrete data body, and a recovery index of the masked data are obtained. The recovery index of the masked data includes the position and the masked value of each masked data.

[0052] Next, the mamba encoder encodes the unmasked data set to capture and represent the inherent complex features and structures of the seismic data processed thereby, that is, to obtain the potential features thereof. Then, in combination with the position marker of each unmasked data in the discrete data body and the recovery index of the masked data, the input data of the mamba decoder can be constructed. The mamba decoder decodes the data input thereto, and then maps the decoding result to the seismic data domain according to the requirement of the MAE output data to obtain the single-channel seismic data prediction value, that is, to complete the reconstruction of the single-channel seismic data.

[0053] In order to quantify the training effect, after obtaining the single-channel seismic data prediction value, a first loss function value is calculated according to the single-channel seismic data prediction value and the corresponding first target sample single-channel seismic data. Then, the initial MAE is iteratively trained by using the first loss function value until a preset ending condition is reached, that is, the target MAE can be obtained. The preset ending condition is not specifically limited in the embodiment of the present application, which can be that the training reaches a specified number of rounds, or the first loss function value converges below a preset threshold.

[0054] In an optional embodiment, the step S207 of calculating the first loss function value based on the first target sample single-channel seismic data and the single-channel seismic data prediction value specifically includes the following steps:

[0055] The step S2071 screens the corresponding prediction data from the single-channel seismic data prediction value based on the position of the masked data in the first target sample single-channel seismic data.

[0056] The step S2072 calculates the Euclidean distance between the masked data and the prediction data.

[0057] The step S2073 determines the Euclidean distance as the first loss function value.

[0058] Specifically, in order to make the MAE focus on learning how to restore the mask data, and thus distinguish from the calculation method of the traditional loss function: calculate the difference between the first target sample single-channel seismic data and the single-channel seismic data prediction value. In the calculation of the first loss function value, the embodiment of the application only considers the difference between the mask data in the first target sample single-channel seismic data and the corresponding predicted data in the single-channel seismic data prediction value, specifically, the Euclidean distance between the mask data and the predicted data is taken as the first loss function value.

[0059] In an alternative embodiment, the embodiment of the application further comprises the following steps:

[0060] Step S301, a second training sample set is obtained; wherein the second training sample set includes multiple groups of second training samples, and each group of second training samples includes sample single-channel seismic data and corresponding real seismic wave impedance data.

[0061] Step S302, the second target sample single-channel seismic data is processed using the initial inversion model to obtain corresponding seismic wave impedance prediction data; wherein the second target sample single-channel seismic data represents any sample in the second training sample set.

[0062] Step S303, based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data, a second loss function value is calculated.

[0063] Step S304, based on the second loss function value, the initial inversion model is iteratively trained to continuously adjust the model parameters of the first projection module and the second projection module until a preset iteration termination condition is reached.

[0064] From the above description, it can be seen that the model parameters of the encoder module and the decoder module in the initial inversion model have been fixed, that is, the model parameters of the two are no longer adjusted. When the initial inversion model is trained, the parameters to be optimized and adjusted only include the model parameters of the first projection module and the model parameters of the second projection module.

[0065] During the model training process, the second target sample single-channel seismic data is input into the initial inversion model, and the corresponding seismic wave impedance prediction data can be output. In order to improve the credibility and accuracy of the model prediction results, when calculating the second loss function value, the embodiment of the application not only considers the prediction accuracy of the seismic wave impedance, but also calculates the synthetic seismic data using the seismic wave impedance prediction data, and further considers the difference between the synthetic seismic data and the corresponding second target sample single-channel seismic data. Then, the initial inversion model is iteratively trained using the second loss function value until a preset iteration termination condition is reached. Optionally, the preset iteration termination condition is that the correlation coefficient between the synthetic seismic data and the corresponding second target sample single-channel seismic data converges to a specified threshold or above.

[0066] In an optional implementation, the step S303 comprises the following steps of:

[0067] In step S3031, the forward processing is performed on the predicted seismic wave impedance data by using a preset forward operator to obtain the synthetic single-channel seismic data.

[0068] In step S3032, the residual error between the synthetic single-channel seismic data and the second target sample single-channel seismic data is calculated to obtain the reconstruction loss.

[0069] In step S3033, the mean square error between the real seismic wave impedance data and the predicted seismic wave impedance data is calculated to obtain the well loss.

[0070] In step S3034, the second loss function value is calculated based on the reconstruction loss and the well loss.

[0071] In the embodiment of the present application, the formula of the synthetic single-channel seismic data is: wherein, represents the synthetic single-channel seismic data, represents a wavelet matrix, represents a differential matrix, represents a column vector composed of the predicted seismic wave impedance data, that is, the preset forward operator is .

[0072] In an optional implementation, the second loss function is represented as: ; wherein, both represent preset weight factors, represents the reconstruction loss, represents the well loss.

[0073] In summary, the embodiment of the present application is a seismic wave impedance inversion method combining MAE and mamba modules. The MAE component adopts an asymmetric encoder-decoder architecture, the encoder processes the visible subset of the data block, and the lightweight decoder reconstructs the original input from the latent representation. This design minimizes the computational complexity and optimizes the accuracy. In order to effectively capture global features, the target inversion model adopts the mamba encoder and mamba decoder in MAE, which uses its self-attention mechanism to identify global correlations in seismic data. This combined method makes the embodiment of the present application have higher accuracy and computational efficiency in the seismic wave impedance inversion task.

[0074] The present application uses the open source data of Marmousi2 model for experimental verification, Figure 3This is a schematic diagram of the seismic wave impedance distribution in the Marmousi2 model, as shown below. Figure 3 As shown, the impedance values ​​exhibit significant variations around the fault in the Marmousi2 model. The Marmousi2 model is specifically designed to replicate the geological environment associated with continental drift by incorporating various typical geological features, such as temporal and spatial dimensions and significant velocity variations in fault structure. The dataset contains 13,601 records, each consisting of 2,801 samples at 1 ms intervals. The model is used to calculate the reflection coefficient, which is convolved with a 30Hz Ricker wavelet to generate... Figure 4 The synthetic seismic data shown.

[0075] In this model, 5000 records were randomly selected as sample data, with 90% allocated to the training set and 10% to the validation set. The validation set was used to evaluate network performance every 100 time intervals, facilitating model parameter tuning to minimize validation loss. The Adam optimizer was used to update network parameters, with weight decay set to 10⁻⁸ and a learning rate of 0.01. Training was performed using the PyTorch framework. For comparison, three widely used deep learning methods—RNN, LSTM, and Transformer—were evaluated. To ensure consistency in the comparative experiments, the parameters and training procedures of all models were similar.

[0076] Figure 5 The prediction results of four seismic impedance prediction methods are shown. Figure 5 Views (a)–(d) in the figure show the seismic impedance prediction results of RNN, LSTM, Transformer, and MAE-Mamba (the method provided in the embodiments of this invention), respectively. The RNN model shows the overall structural characteristics of the lithological layers, but significant fluctuations exist in certain specific areas. The LSTM model effectively preserves the long-term memory of the sequence during the cycle through a gating mechanism, exhibiting good vertical continuity. However, due to the complexity of the model, it is prone to deformation in complex lithological environments, leading to unstable results. The Transformer can better capture long-distance dependencies through its self-attention mechanism, thus achieving better global performance when processing long sequences. Its inversion results are largely consistent with the actual impedance, but there are still shortcomings in some specific features. MAE-Mamba effectively combines the advantages of RNN and Transformer, thereby improving the overall prediction accuracy and superior horizontal continuity, especially in structurally complex areas. Figure 5 The (e)-(h) views in the diagram depict the residual errors between the predictions of the four methods and the ground truth. According to... Figure 5 It can be seen that the impedance prediction error generated by MAE-Mamba is the smallest, and its value is close to the true ground impedance.

[0077] To examine the prediction performance in more detail, the 8500th trace was extracted from the impedance profile. This seismic trace is located in a complex lithology environment of the Marmousi2 model. Figure 6 The figure shows the difference between the prediction results of the four seismic wave impedance prediction methods and the true value for a single seismic trace. The results of the four methods are shown in the figure. Figure 6 It can be seen that the method has good recognition effect in the lithology change area, and the fluctuation range meets the expectation. It ensures higher consistency with the true impedance. In addition, the method has obvious advantages in global inversion, and the overall trend of the curve is clear.

[0078] Table 1 below summarizes the time required for training and prediction of different method networks, indicating that MAE-Mamba has higher computational efficiency compared with other methods involved in the experiment. Table 2 gives the R 2 , Pearson correlation coefficient (PCC) and mean square error (MSE) between the prediction results and the basic true value of the experiment. According to the data in Table 2, the MAE-Mamba method is better than other methods.

[0079] Table 1 Time required for training and prediction using Marmousi2

[0080]

[0081] Table 2 R 2 , PCC and MSE between the impedance prediction value and the true value using Marmousi2

[0082]

[0083] In addition, a series of ablation experiments are conducted to verify the effectiveness of the method. The experiments include disabling and adjusting the parameters of the MAE component to evaluate their impact on the Mamba block and the Transformer block. The performance indicators include Pearson correlation coefficient (PCC), R square value and mean square error (MSE), and Table 3 below shows the performance indicators of different experiments, indicating that the model achieves the best result when the masking ratio is 0.75. In addition, the comparison analysis with the model excluding the MAE component provides additional evidence to support the effectiveness of the method proposed in the embodiment.

[0084] Table 3 R 2 , PCC and MSE between the impedance prediction value and the true value of different ablation experiments

[0085]

[0086] Example Two

[0087] The embodiment of the present application also provides a seismic wave impedance inversion device, which is mainly used for executing the seismic wave impedance inversion method provided in the above embodiment one, and the seismic wave impedance inversion device provided by the embodiment of the present application will be specifically introduced below.

[0088] Figure 7 The function module diagram of the seismic wave impedance inversion device provided by the embodiment of the present application is shown in the figure, Figure 7 which mainly comprises: a first acquisition module 10, a discretization module 20 and an inversion module 30, wherein:

[0089] The first acquisition module 10 is used for acquiring real seismic data of a work area to be processed.

[0090] The discretization module 20 is used for discretizing the real seismic data to obtain a plurality of single-channel seismic data.

[0091] The inversion module 30 is used for processing the plurality of single-channel seismic data by using a target inversion model to obtain a seismic wave impedance inversion result of the work area to be processed.

[0092] The target inversion model is a model obtained by training an initial inversion model; the target inversion model comprises an encoder module, a first projection module, a decoder module and a second projection module; the encoding process of the encoder module and the decoding process of the decoder module are both realized based on a mamba module; the first projection module is used for mapping the output of the encoder module to the input space of the decoder module; and the second projection module is used for mapping the output of the decoder module to the seismic wave impedance domain.

[0093] The embodiment of the present application provides a seismic wave impedance inversion device, which, after acquiring real seismic data of a work area to be processed, firstly discretizes the real seismic data to obtain a plurality of single-channel seismic data, and then processes the plurality of single-channel seismic data by using a target inversion model to obtain a seismic wave impedance inversion result of the work area to be processed, wherein the target inversion model is a model obtained by training an initial inversion model. It can be known that the device of the present application essentially executes a data-driven inversion method, which avoids the dependence on the initial model. In addition, the target inversion model comprises an encoder module, a first projection module, a decoder module and a second projection module, and the encoding process of the encoder module and the decoding process of the decoder module are both realized based on a mamba module. Since the mamba module has higher efficiency and scalability in processing long sequence data compared with the traditional self-attention mechanism, and can also maintain good modeling ability, the mamba module provides strong feature extraction and representation ability for the encoder, so that the embodiment of the present application can ensure the accuracy of the seismic wave impedance inversion result compared with the prior art.

[0094] Optionally, the encoder module and the decoder module are a mamba encoder and a mamba decoder in a trained mask autoencoder (MAE); and the apparatus further comprises:

[0095] The second acquisition module is configured to acquire a first training sample set; and the first training sample set comprises a plurality of sample single-channel seismic data.

[0096] The blocking module is configured to perform blocking processing on the first target sample single-channel seismic data to obtain a discrete data volume; and the first target sample single-channel seismic data represents any sample in the first training sample set.

[0097] The masking module is configured to perform random masking processing on the discrete data volume to obtain a recovery index of masked data, an unmasked data set, and a position marker of each unmasked data in the discrete data volume.

[0098] The encoding module is configured to perform encoding processing on the unmasked data set by using the mamba encoder to obtain latent features of the unmasked data set.

[0099] The construction module is configured to construct input data of the mamba decoder based on the latent features, the position marker of each unmasked data in the discrete data volume, and the recovery index of the masked data.

[0100] The decoding and reconstruction module is configured to perform decoding on the input data by using the mamba decoder to reconstruct single-channel seismic data based on a decoding result to obtain a single-channel seismic data prediction value.

[0101] The first calculation module is configured to calculate a first loss function value based on the first target sample single-channel seismic data and the single-channel seismic data prediction value.

[0102] The first training module is configured to perform iterative training on the initial MAE based on the first loss function value until a preset end condition is reached to obtain a target MAE.

[0103] Optionally, the first calculation module is specifically configured to:

[0104] Filter out corresponding prediction data from the single-channel seismic data prediction value based on a position of the masked data in the first target sample single-channel seismic data.

[0105] Calculate an Euclidean distance between the masked data and the prediction data.

[0106] Determine the Euclidean distance as the first loss function value.

[0107] Optionally, the apparatus further comprises:

[0108] The third acquisition module is configured to acquire a second training sample set, wherein the second training sample set includes a plurality of groups of second training samples, and each group of second training samples includes sample single-channel seismic data and corresponding real seismic wave impedance data.

[0109] The processing module is configured to process the second target sample single-channel seismic data by using the initial inversion model to obtain corresponding seismic wave impedance prediction data.

[0110] The second calculation module is configured to calculate a second loss function value based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data.

[0111] The second training module is configured to iteratively train the initial inversion model based on the second loss function value to continuously adjust model parameters of the first projection module and the second projection module until a preset iteration termination condition is reached.

[0112] Optionally, the second calculation module is specifically configured to:

[0113] The second calculation module is configured to calculate a second loss function value based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data.

[0114] The second calculation module is configured to calculate a second loss function value based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data.

[0115] The second calculation module is configured to calculate a second loss function value based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data.

[0116] The second calculation module is configured to calculate a second loss function value based on the second target sample single-channel seismic data, the corresponding real seismic wave impedance data and the seismic wave impedance prediction data.

[0117] Optionally, the second loss function is represented as: ; wherein, each of a, b and c represents a preset weight factor, the reconstruction loss represents the reconstruction loss, the well loss represents the well loss.

[0118] Optionally, a sparsity constraint is added to a hidden layer of the encoder module.

[0119] Embodiment three

[0120] Referring to Figure 8 , the electronic device provided by the embodiment of the present application includes a processor 60, a memory 61, a bus 62 and a communication interface 63, the processor 60, the communication interface 63 and the memory 61 are connected through the bus 62, and the processor 60 is configured to execute executable modules stored in the memory 61, such as a computer program.

[0121] The memory 61 can include a high-speed random access memory (RAM), and can also include a non-volatile memory such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 63 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used.

[0122] The bus 62 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 8 Only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0123] The memory 61 is used to store programs, and the processor 60 executes the programs after receiving execution instructions. The method executed by the device defined by the process disclosed in any of the embodiments of the present application can be applied to the processor 60 or implemented by the processor 60.

[0124] The processor 60 can be an integrated circuit chip with signal processing capability. In implementation, each step of the above method can be completed by integrated logic circuit of hardware in the processor 60 or by instructions in the form of software. The processor 60 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. Each method, step and logic block diagram disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium in the art. The storage medium is located in the memory 61, and the processor 60 reads the information in the memory 61, and combines the hardware to complete the steps of the above method.

[0125] The computer program product of the seismic wave impedance inversion method, device and electronic equipment provided by the embodiments of the present application includes a computer readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the method described in the foregoing method embodiments. For specific implementation, please refer to the method embodiments, which will not be described here.

[0126] In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0127] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0128] It should be noted that similar reference numerals and letters refer to similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0129] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the present application is usually placed, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third", and the like are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0130] In addition, the terms "horizontal", "vertical", "overhanging", and the like do not mean that the components must be absolutely horizontal or overhanging, but can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and does not mean that the structure must be completely horizontal, but can be slightly inclined.

[0131] In the description of the present application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "arrangement", "installation", "connection", and "connection" should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication between two elements inside. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A seismic wave impedance inversion method, characterized in that, include: Obtain real seismic data for the work area to be processed; The real seismic data is discretized to obtain multiple single-channel seismic data. The target inversion model is used to process the multiple single-channel seismic data to obtain the seismic impedance inversion results of the work area to be processed; The target inversion model is obtained by training an initial inversion model. The target inversion model includes an encoder module, a first projection module, a decoder module, and a second projection module. The first projection module maps the output of the encoder module to the input space of the decoder module. The second projection module maps the output of the decoder module to the seismic impedance domain. Before constructing the initial inversion model, a masked autoencoder (MAE) using a Mamba encoder and a Mamba decoder is pre-trained. After the MAE reaches the preset training target, the trained Mamba encoder and Mamba decoder are used as the encoder module and decoder module in the initial inversion model, respectively. When training the initial inversion model, the model parameters of the encoder module and decoder module are fixed. The parameters to be optimized and adjusted include the model parameters of the first projection module and the model parameters of the second projection module.

2. The seismic wave impedance inversion method according to claim 1, characterized in that, The encoder module and the decoder module are the mamba encoder and mamba decoder in the trained masked autoencoder MAE. The training process for MAE includes: Obtain a first training sample set; wherein, the first training sample set includes: multiple sample single-channel seismic data; The single-channel seismic data of the first target sample is divided into blocks to obtain a discrete data volume; wherein, the single-channel seismic data of the first target sample represents any sample in the first training sample set; The discrete data volume is subjected to random masking to obtain the recovery index of the masked data, the set of unmasked data, and the position mark of each unmasked data in the discrete data volume; The unmasked data set is encoded using a Mamba encoder to obtain the latent features of the unmasked data set; Based on the latent features, the position marker of each unmasked data in the discrete data volume, and the recovery index of the masked data, the input data of the Mamba decoder is constructed; The input data is decoded using a Mamba decoder, and the single-channel seismic data is reconstructed based on the decoding results to obtain the predicted value of the single-channel seismic data. Based on the single-channel seismic data of the first target sample and the predicted value of the single-channel seismic data, calculate the value of the first loss function; The initial MAE is iteratively trained based on the first loss function value until the preset termination condition is met, and the target MAE is obtained.

3. The seismic wave impedance inversion method according to claim 2, characterized in that, Based on the first target sample single-channel seismic data and the predicted value of the single-channel seismic data, a first loss function value is calculated, including: Based on the position of the mask data in the single-channel seismic data of the first target sample, the corresponding prediction data is selected from the predicted values ​​of the single-channel seismic data; Calculate the Euclidean distance between the mask data and the predicted data; The Euclidean distance is determined as the value of the first loss function.

4. The seismic wave impedance inversion method according to claim 1, characterized in that, Also includes: Obtain a second training sample set; wherein, the second training sample set includes multiple sets of second training samples, each set of second training samples includes: sample single-channel seismic data and corresponding real seismic impedance data; The initial inversion model is used to process the single-channel seismic data of the second target sample to obtain the corresponding seismic impedance prediction data; wherein, the single-channel seismic data of the second target sample represents any sample in the second training sample set; Based on the second target sample single-channel seismic data, the corresponding real seismic impedance data, and the seismic impedance prediction data, the second loss function value is calculated. The initial inversion model is iteratively trained based on the second loss function value to continuously adjust the model parameters of the first projection module and the second projection module until the preset iteration termination condition is reached.

5. The seismic wave impedance inversion method according to claim 4, characterized in that, Based on the second target sample single-channel seismic data, the corresponding real seismic impedance data, and the predicted seismic impedance data, the second loss function value is calculated, including: The seismic impedance prediction data is processed by a preset forward modeling operator to obtain synthetic single-channel seismic data; The reconstruction loss is obtained by calculating the residual between the synthetic single-channel seismic data and the second target sample single-channel seismic data; The mean square error between the actual seismic impedance data and the predicted seismic impedance data is calculated to obtain the well loss. The second loss function value is calculated based on the reconstruction loss and the well loss.

6. The seismic wave impedance inversion method according to claim 5, characterized in that, The second loss function is expressed as: ;in, All represent preset weighting factors. This represents the reconstruction loss. This indicates the well loss.

7. The seismic wave impedance inversion method according to claim 1, characterized in that, Add sparsity constraints to the hidden layer of the encoder module.

8. A seismic wave impedance inversion device, characterized in that, include: The first acquisition module is used to acquire real seismic data of the work area to be processed; The discretization module is used to discretize the real seismic data to obtain multiple single-channel seismic data. The inversion module is used to process the multiple single-channel seismic data using the target inversion model to obtain the seismic impedance inversion results of the work area to be processed. The target inversion model is obtained by training an initial inversion model. The target inversion model includes an encoder module, a first projection module, a decoder module, and a second projection module. The first projection module maps the output of the encoder module to the input space of the decoder module. The second projection module maps the output of the decoder module to the seismic impedance domain. Before constructing the initial inversion model, a masked autoencoder (MAE) using a Mamba encoder and a Mamba decoder is pre-trained. After the MAE reaches the preset training target, the trained Mamba encoder and Mamba decoder are used as the encoder module and decoder module in the initial inversion model, respectively. When training the initial inversion model, the model parameters of the encoder module and decoder module are fixed. The parameters to be optimized and adjusted include the model parameters of the first projection module and the model parameters of the second projection module.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the seismic impedance inversion method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the seismic impedance inversion method according to any one of claims 1 to 7.

Images