Multi-scale seismic impedance inversion model construction method under weak well control condition, seismic impedance inversion method and related device

Abstract

The present application relates to a kind of seismic exploration technical field, it is a kind of multi-scale seismic impedance inversion model construction method for weak well control condition, seismic impedance inversion method and related device, including respectively input original seismic data and impedance label of labeled training sample to inversion model and forward model, determine corresponding impedance loss and labeled seismic reconstruction loss;Input unlabeled training sample to inversion model to obtain impedance inversion result, then input impedance inversion result to forward model to obtain unlabeled reconstruction seismic data, determine corresponding unlabeled seismic reconstruction loss;Determine corresponding total loss and update the model parameters of inversion model and forward model by back propagation.The present application extracts different dilated scale seismic reflection features by multi-scale dilated convolution, enhances horizon interface, fault boundary and thin layer detail information by wavelet guided attention module, and improves the consistency between predicted impedance and original seismic response by seismic forward reconstruction constraint.

Description

Technical Field

[0001] This invention relates to the field of seismic exploration technology, specifically a method for constructing a multi-scale seismic impedance inversion model under weak well control conditions, a seismic impedance inversion method, and related equipment. Background Technology

[0002] Seismic impedance inversion is an important reservoir prediction technique in oil and gas exploration and development. It uses seismic data, well logging data, and geological information to invert the impedance parameters of the subsurface medium, thus providing a basis for lithology identification, thin reservoir prediction, sand body distribution characterization, and evaluation of favorable oil and gas-bearing areas. Compared to conventional seismic profile interpretation, impedance has a more direct relationship with subsurface lithology and reservoir properties, therefore playing a crucial role in the detailed evaluation of oil and gas reservoirs and well location deployment.

[0003] Existing seismic impedance inversion methods mainly include: (I) Traditional post-stack inversion methods, which are mostly based on convolution models. This model, assuming vertical seismic wave incidence, horizontally layered strata, and neglecting multiples and transmission losses, approximates the seismic record as a convolution of the seismic wavelet and the subsurface reflection coefficient sequence (including noise terms). While these methods have clear physical meaning and are relatively mature in application, real-world seismic data often suffers from limited bandwidth, insufficient low-frequency information, noise interference, and inaccurate wavelet estimation. When the quality of the seismic data is poor or the well-seismic matching relationship is unsatisfactory, the inversion results are easily affected, leading to unstable impedance trends.

[0004] Furthermore, the limitations of this method become more apparent in complex structural zones and areas with thin interbedded layers. Specifically, because faults, pinch-outs, thin layers, and low-impedance anomalies are often small in scale and have rapidly changing boundaries, conventional inversion methods tend to smooth out these details, resulting in unclear interlayer interfaces, insufficiently sharp reservoir boundaries, and inadequate local anomaly responses. For oil and gas exploration, this can affect the assessment of reservoir thickness, lateral continuity, sand body connectivity, and favorable oil and gas-bearing areas.

[0005] (II) Seismic impedance inversion methods incorporating deep learning technology, such as achieving seismic impedance inversion through deep learning of a pre-defined network. These methods can learn nonlinear characteristics from seismic data, improving impedance prediction efficiency and, to some extent, addressing the lack of detail in traditional methods. However, in actual work areas, available well labels are usually limited, and well locations are unevenly distributed. Models trained entirely on label data are easily limited by the number and location of wells, resulting in insufficient prediction stability in areas far from wells. Furthermore, two key issues remain: first, ordinary convolutional structures have limited ability to express multi-scale geological features, making it difficult to simultaneously characterize thin-layer details and large-scale stratigraphic changes; second, some methods only focus on the direct mapping from seismic data to impedance labels, lacking constraints on the consistency of seismic response, leading to impedance inversion results that may not be consistent with the original seismic reflection characteristics in areas without wells. Summary of the Invention

[0006] This invention provides a method for constructing a multi-scale seismic impedance inversion model under weak well control conditions, a seismic impedance inversion method, and related devices, which overcomes the shortcomings of the prior art and can effectively solve the problem that existing impedance inversion methods have difficulty in clearly depicting information such as stratigraphic interfaces, fault boundaries, and thin-layer details.

[0007] One of the technical solutions of this invention is achieved through the following measures: a method for constructing a multi-scale seismic impedance inversion model under weak well control conditions, comprising: Based on the original earthquake data of the historical region, a labeled training sample set and an unlabeled training sample set were constructed. The original seismic data with labeled training samples and impedance labels are input into the inversion model and the forward model respectively to determine the corresponding impedance loss and labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction and outputs the corresponding impedance inversion results after splicing and fusion through upsampling. Input unlabeled training samples into the inversion model to obtain impedance inversion results, then input the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determine the corresponding unlabeled seismic reconstruction loss; The total loss is determined by weighting the impedance loss, labeled seismic reconstruction loss and unlabeled seismic reconstruction loss, and the model parameters of the inversion model and the forward model are updated by backpropagation. Repeat the above training process until the network converges, and output the trained inversion model as a multi-scale seismic impedance inversion model.

[0008] The following are further optimizations and / or improvements to the above-mentioned technical solution: The aforementioned inversion model includes a first feature extraction module, a second feature extraction module, a fusion module, and an upsampling module; The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and then fuses them to obtain fused features. Wavelet-guided attention is then introduced to obtain enhanced features. The second feature extraction module receives seismic data and extracts time-series features; The fusion module receives the enhanced features output by the first feature extraction module and the sequence features output by the second feature extraction module, and performs feature fusion after aligning the number of channels and the time dimension. The upsampling module transposes and convolves or reassembles pixels of the features output by the fusion module to restore the number of sampling points to the original seismic data, thus obtaining the impedance inversion result.

[0009] The aforementioned first feature extraction module includes: a multi-scale dilated convolution module, a fusion module, a wavelet-guided attention module, a compression module, and a feature output module; The multi-scale dilatation convolution module includes multiple parallel dilatation convolution sub-modules with different dilatation scales. Each dilatation convolution sub-module receives the original seismic data and performs time dilatation convolution at the corresponding dilatation scale to obtain seismic reflection features at different dilatation scales. The fusion module stitches and fuses the seismic reflection features at all dilatational scales output by the multi-scale dilatational convolution module; The wavelet-guided attention module determines the spatial attention weights and channel attention weights based on the fusion features output by the fusion module, and applies the spatial attention weights and channel attention weights to the fusion features output by the fusion module to obtain enhanced features; The compression module and the feature output module perform dimensionality reduction and alignment processing on the enhanced features, respectively.

[0010] The second feature extraction module mentioned above includes a bidirectional gated loop unit that receives seismic data and extracts temporal sequence features.

[0011] The total loss function is as follows: in, These are weighting coefficients, used to control the contributions of impedance loss, labeled seismic reconstruction loss, and unlabeled seismic reconstruction loss to the total loss, respectively. Among them impedance loss As shown below: in, For inversion model; These are model parameters; For the first i The original seismic data with labeled training samples; For the first i The impedance labels of the labeled training samples; N is the number of labeled training samples; Among them is the labeled seismic reconstruction loss. As shown below: in, Forward model; These are model parameters; Unlabeled seismic reconstruction loss As shown below: in, For the first j The original seismic data of M unlabeled training samples; M is the number of unlabeled training samples.

[0012] The second technical solution of the present invention is achieved through the following measures: a multi-scale seismic impedance inversion method for weak well control conditions, comprising: Acquire seismic data for the target area; Input seismic data of the target area into the multi-scale seismic impedance inversion model to obtain the corresponding impedance inversion results.

[0013] The third technical solution of the present invention is achieved through the following measures: a multi-scale seismic impedance inversion model construction device for weak well control conditions, comprising: The sample construction unit constructs labeled training sample sets and unlabeled training sample sets based on the original seismic data of the historical region; The labeled training unit takes the original seismic data and impedance labels of the labeled training samples as input to the inversion model and the forward model, respectively, and determines the corresponding impedance loss and labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction and outputs the corresponding impedance inversion results after splicing and fusion through upsampling. The unlabeled training unit inputs unlabeled training samples into the inversion model to obtain impedance inversion results, and then inputs the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determines the corresponding unlabeled seismic reconstruction loss. The model parameter update unit determines the corresponding total loss by weighting the impedance loss, labeled seismic reconstruction loss and unlabeled seismic reconstruction loss, and updates the model parameters of the inversion model and the forward model through backpropagation. The iterative unit repeats the above training process until the network converges, and outputs the trained inversion model as a multi-scale seismic impedance inversion model.

[0014] The fourth technical solution of the present invention is achieved through the following measures: a multi-scale seismic impedance inversion device for weak well control conditions, comprising: The seismic data acquisition unit acquires seismic data for the target area. The inversion unit takes the seismic data of the target area as input to the multi-scale seismic impedance inversion model and obtains the corresponding impedance inversion results.

[0015] The fifth technical solution of the present invention is achieved through the following measures: an electronic device, including a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps in the method for constructing a multi-scale seismic impedance inversion model under weak well control conditions or the multi-scale seismic impedance inversion method.

[0016] The sixth technical solution of the present invention is achieved by the following measures: a storage medium storing a computer program that can be read by a computer, the computer program being configured to execute the steps in the method for constructing a multi-scale seismic impedance inversion model under weak well control conditions or the multi-scale seismic impedance inversion method when running.

[0017] This invention utilizes a small number of labeled training samples and a large number of unlabeled training samples for model training. It employs multi-scale dilatational convolution, wavelet-guided attention, and seismic forward modeling constraints to establish a nonlinear mapping relationship between seismic data and impedance. Multi-scale dilatational convolution extracts seismic reflection features at different dilatational scales; the wavelet-guided attention module enhances information on stratigraphic interfaces, fault boundaries, and thin-layer details; and seismic forward modeling constraints improve the consistency between predicted impedance and the original seismic response. Therefore, this invention can obtain stable, continuous, and geologically plausible impedance inversion results under weak well control conditions using the constructed multi-scale seismic impedance inversion model, providing a reliable geophysical parameter basis for complex oil and gas reservoir prediction, lithological identification, and favorable area evaluation. Attached Figure Description

[0018] Appendix Figure 1 This is a schematic diagram of an implementation environment provided for an embodiment of the present invention.

[0019] Appendix Figure 2 This is a schematic diagram of a method for constructing a multi-scale seismic impedance inversion model, provided in an embodiment of the present invention.

[0020] Appendix Figure 3 The diagram shows the network structure of the inversion model and the forward model provided in the embodiments of the present invention.

[0021] Appendix Figure 4 This is a schematic diagram of another method for constructing a multi-scale seismic impedance inversion model provided in an embodiment of the present invention.

[0022] Appendix Figure 5 A schematic diagram of the Marmousi2 model seismic data recording provided in an embodiment of the present invention.

[0023] Appendix Figure 6 This is a schematic diagram of the true impedance label of the Marmousi2 model seismic data provided in an embodiment of the present invention.

[0024] Appendix Figure 7 This is a schematic diagram of the impedance inversion results of the Marmousi2 model seismic data provided in an embodiment of the present invention.

[0025] Appendix Figure 8 This is a schematic diagram of the multi-scale seismic impedance inversion method provided in an embodiment of the present invention.

[0026] Appendix Figure 9 A schematic diagram of the structure of the multi-scale seismic impedance inversion model construction device provided in an embodiment of the present invention.

[0027] Appendix Figure 10 A schematic diagram of the structure of the multi-scale seismic impedance inversion device provided in an embodiment of the present invention. Detailed Implementation

[0028] The present invention is not limited to the following embodiments, and specific implementation methods can be determined according to the technical solutions and actual conditions of the present invention.

[0029] Those skilled in the art will understand that, unless specifically stated otherwise, in the embodiments of the present invention, a "module" or "unit" refers to a computer program or part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0030] In addition, in the embodiments of the present invention, "multiple" refers to two or more, and "first" and "second" are used to distinguish descriptions and should not be construed as implying relative importance.

[0031] This invention provides a multi-scale seismic impedance inversion method for weak well control conditions, comprising: Acquire seismic data for the target area; input the seismic data of the target area into the multi-scale seismic impedance inversion model to obtain the corresponding impedance inversion results. The multi-scale seismic impedance inversion model is constructed using the multi-scale seismic impedance inversion model construction method under weak well control conditions.

[0032] The method provided in this embodiment of the invention may involve artificial intelligence (AI) technology and may be implemented based on artificial intelligence technology, such as using deep learning to train a corresponding model using samples.

[0033] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence.

[0034] Deep learning (DL) specifically refers to machine learning based on deep, pre-defined network models and methods. It has evolved from statistical machine learning and manually pre-defined network algorithms, combined with the development of modern big data and high computing power. The most important technical feature of deep learning is its ability to automatically extract features.

[0035] The aforementioned machine learning and deep learning typically include techniques such as predefined networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0036] In deep learning, the loss function is used to train a pre-defined network. Because the goal is for the network's output to be as close as possible to the actual predicted value, the loss function compares the current network's predicted value with the target value and updates the weight vector of each layer based on the difference. (There is usually an initialization process before the first update, where parameters are pre-configured for each layer in the network) until the network can predict the target value or a value very close to it. Therefore, deep learning requires pre-defining "how to compare the difference between the predicted value and the target value," which is the loss function.

[0037] As attached Figure 1 The diagram illustrates an implementation environment provided by an embodiment of the present invention. This implementation environment may include: training equipment and usage equipment.

[0038] Both the training equipment and the equipment used are computer devices; optionally, the computer device is a terminal device, such as a mobile phone, tablet computer, PC (Personal Computer) or other electronic devices; or, the computer device is a server, which can be a single server, a server cluster composed of multiple servers, or a cloud computing service center. This embodiment of the invention does not limit this.

[0039] Training equipment refers to computer equipment capable of training and learning a pre-defined network. Optionally, the training equipment has the ability to acquire a pre-defined network and train and learn it according to application requirements. For example, the training equipment acquires a pre-defined network from other devices via a network and then trains it using training samples according to application requirements, so that the pre-defined network has the ability to obtain impedance inversion results. Optionally, the training equipment has the ability to construct a pre-defined network. It can construct a pre-defined network itself according to application requirements and then train and learn it. For example, in order to obtain impedance inversion results from seismic data, the training equipment constructs a pre-defined network itself and then trains and learns it using samples according to application requirements.

[0040] The device being used refers to a computer device that has the capability to use a pre-defined network. Optionally, the device being used can obtain a pre-defined network from other devices via the network according to the application requirements. For example, if the device being used has the capability to perform impedance inversion, it can obtain a pre-defined network that has been trained and learned to obtain impedance inversion results from other devices via the network, and use the pre-defined network to perform impedance inversion.

[0041] Example 1: As shown in the attached document Figure 2 As shown in the figure, this invention discloses a method for constructing a multi-scale seismic impedance inversion model under weak well control conditions, including: Step S110: Based on the original seismic data of the historical region, construct a labeled training sample set and an unlabeled training sample set; Step S120: Input the original seismic data and impedance labels of the labeled training samples into the inversion model and the forward model respectively, and determine the corresponding impedance loss and labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction, and outputs the corresponding impedance inversion results after splicing and fusion through upsampling. Step S130: Input unlabeled training samples into the inversion model to obtain impedance inversion results, then input the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determine the corresponding unlabeled seismic reconstruction loss. Step S140: The impedance loss, labeled seismic reconstruction loss and unlabeled seismic reconstruction loss are weighted to determine the corresponding total loss, and the model parameters of the inversion model and the forward model are updated through backpropagation. Step S150: Repeat the above training process until the network converges, and output the trained inversion model as a multi-scale seismic impedance inversion model.

[0042] In step S110 above, based on the characteristic of a small number of well logging labels, labeled training sample sets and unlabeled training sample sets are established based on a small amount of seismic data with well logging impedance labels and a large amount of seismic data without well logging impedance labels, thereby reducing the model's dependence on dense well logging data.

[0043] This invention discloses a method for constructing a multi-scale seismic impedance inversion model under weak well control conditions. By establishing labeled and unlabeled training sample sets, the model's dependence on dense well logging data is reduced. Multi-scale dilatational convolution is introduced to extract seismic reflection features at different scales. Wavelet-guided attention is introduced to enhance stratigraphic interfaces, fault boundaries, and thin-layer details, thus better preserving the impedance changes of major stratigraphic interfaces, continuous high-impedance layers, and complex structural regions. This enables the model to maintain the stability and geological rationality of impedance inversion results in complex geological models. Furthermore, the use of seismic forward modeling reconstruction constraints improves the consistency between impedance inversion results and original seismic data.

[0044] Example 2: As shown in the attached document Figure 3 As shown, the embodiments of the present invention are further optimizations of the above embodiments, wherein the inversion model includes a first feature extraction module, a second feature extraction module, a fusion module, and an upsampling module; (I) The first feature extraction module includes: a multi-scale dilated convolution module, a fusion module, a wavelet-guided attention module, a compression module, and a feature output module. Specifically: The multi-scale dilatational convolution module includes multiple parallel dilatational convolutional sub-modules with different dilatational scales. Each dilatational convolutional sub-module receives the raw seismic data and performs time dilatational convolution at the corresponding dilatational scale to obtain seismic reflection features at different dilatational scales. By setting different dilatational rates, this process can expand the receptive field without significantly increasing the number of parameters, thereby simultaneously extracting thin-layer details and large-scale stratigraphic variation information.

[0045] As attached Figure 3 As shown, each dilated convolutional submodule specifically includes: (1) Depthwise convolution, within each input channel, for each spatial coordinate point On the expansion scale d Under the field of view, the corresponding seismic reflection features are extracted, and their mathematical representation is shown below: in, For each spatial coordinate point In the expansion scale d Under the field of view, extract the corresponding seismic reflection features; To expand at the scale d Below, the first in the convolution kernel n Learnable weight coefficients for each position; k The kernel length is [length]. n This refers to the index number of each weight within the convolution kernel, with a value ranging from 1 to... k ; i The trace location is a spatial dimension numerical index, specifying which seismic trace is extracted from the original seismic data X for calculation. j For time sampling location; X This is the raw earthquake data; This is the index offset; The reference anchor point is half the length of the convolution kernel, rounded up.

[0046] (2) Pointwise convolution, converting each spatial coordinate point The seismic reflection features of all channels corresponding to a given location are linearly weighted and fused, thereby recombining multiple sets of seismic reflection features extracted by deep convolution, restoring spatial consistency, and obtaining seismic reflection features at the corresponding scale.

[0047] (3) The seismic reflection features output by pointwise convolution are processed sequentially through group normalization (GN), SE module and Dropout. Among them, group normalization is used to stabilize the feature distribution under small batch training conditions; SE module generates channel weights through global feature statistics and adaptively calibrates the feature responses of different channels; Dropout randomly masks some feature responses according to a preset probability to reduce the network's dependence on specific channel combinations and suppress overfitting.

[0048] (4) Residual connection (y+res): The features output by the convolution branch are transferred to the residual connection (y+res). y Compared with the original input seismic data preserved through identity mapping res Element-wise addition is performed to preserve input information, improve gradient propagation, and enhance the stability of network training.

[0049] (5) GELU activation function: The residuals are added together and then output through the GELU activation function. This means that the output of the residual branch (convolution branch) and the identity mapping branch are first fused together and then uniformly transformed into nonlinearity. This allows the activation function to adaptively filter the effective features after mixing. Compared with "activation first and then addition", more original low-frequency seismic information can be retained.

[0050] The fusion module stitches and fuses the seismic reflection features at all dilatational scales output by the multi-scale dilatational convolution module. Its mathematical expression is as follows: in, These represent seismic reflection characteristics at different expansion scales. For feature splicing; This is for point-by-point convolution fusion.

[0051] The wavelet-guided attention module determines spatial attention weights and channel attention weights based on the fusion features output by the fusion module. These weights are then applied to the fusion features output by the fusion module to obtain enhanced features, the mathematical representation of which is shown below: in, For element-wise multiplication, To enhance features; Features of fusion; Spatial attention weights; This is used for channel attention weights. This process highlights key geological structures in seismic data, enhances information on stratigraphic interfaces, fault boundaries, and thin-layer details, while reducing the impact of noise and irrelevant responses on impedance inversion results.

[0052] As attached Figure 3 As shown, it specifically includes: The fused features output by the fusion module, after being subjected to discrete wavelet transform, yield high-frequency components in different directions, the mathematical expression of which is shown below: in, These are the high-frequency detail components in the horizontal, vertical, and diagonal directions, respectively. The high-frequency branch operation of the discrete wavelet transform indicates that the input signal is subjected to one or more discrete wavelet transforms and only its high-frequency detail components are extracted (i.e., approximate components are discarded). This operation decomposes the signal into detail information in different directions through wavelet filters; these high-frequency components contain key information such as layer interfaces, fault boundaries and thin-layer abrupt changes.

[0053] Spatial attention weights are generated by concatenating, convolving, and upsampling high-frequency components from different directions. Their mathematical expression is shown below: in, Spatial attention weights; Use the Sigmoid activation function; This is an upsampling operation. Spatial attention is used to enhance the response of structural regions such as stratigraphic interfaces and fault boundaries, while suppressing background noise and irrelevant information.

[0054] The fused features output by the fusion module are processed by weighted global average pooling to obtain channel description vectors, which are then used to generate channel attention weights through a fully connected layer and a sigmoid function. Among them, for the first c The weighted average pooling result for each channel can be expressed as: in, This is the global description of the c-th channel; H The window size represents the number of sampling points in the spatial dimension. i The trace location is a numerical index in a spatial dimension, specifying the location from the raw seismic data. X Which seismic trace should be selected for calculation, with a value range of 1 to... H ; T This represents the number of time sampling points; j Numeric index for the time dimension, with values ​​ranging from 1 to... T ; To prevent stable terms with a denominator of zero; For the fusion feature of the first c Each channel is located in Eigenvalues ​​at; To act on position Spatial attention weight coefficients.

[0055] The compression module and the feature output module perform dimensionality reduction and alignment processing on the enhanced features, respectively. The compression module and the feature output module can use adaptive pooling and convolutional preset networks, respectively.

[0056] (ii) The second feature extraction module includes a bidirectional gated loop unit that receives seismic data and extracts time-series features.

[0057] (iii) Fusion module, which receives the enhanced features output by the first feature extraction module and the sequence features output by the second feature extraction module, and performs feature fusion after aligning the number of channels and the time dimension.

[0058] (iv) Upsampling module: Transpose and convolve or reconstitute pixels of the features output by the fusion module to restore the number of sampling points of the original seismic data and obtain high-resolution impedance inversion results.

[0059] Example 3: As shown in the attached document Figure 4As shown, the embodiments of the present invention are further optimizations of the above embodiments. The forward model includes: taking the impedance sequence as input, extracting time direction features through multi-layer one-dimensional convolution, simulating the band-limited features of seismic data through seismic wavelet convolution, and outputting reconstructed seismic data. Its mathematical expression is as follows: in, This is a one-dimensional convolutional feature extraction process; * represents the wavelet convolution operator; * represents the convolution operation. This forward model is used to connect impedance prediction results with seismic response, enabling the model to focus not only on impedance label error during training, but also on whether the impedance inversion results can reconstruct the original seismic data.

[0060] Example 4: As shown in the appendix Figure 4 As shown in the figure, this invention discloses a method for constructing a multi-scale seismic impedance inversion model under weak well control conditions, including: Step S210: Based on the original seismic data of the historical region, construct a labeled training sample set and an unlabeled training sample set; This step, which involves constructing labeled and unlabeled training sample sets, includes: Based on whether or not impedance labels are present, the raw seismic data of the historical region are divided into labeled training sample sets and unlabeled training sample sets.

[0061] The labeled training sample set can be represented as: in, For the first i Labeled raw seismic data; Impedance tags obtained from well logging data; N This represents the number of labeled training samples.

[0062] The unlabeled training sample set can be represented as: in, For the first j M is the number of unlabeled raw seismic data; M is the number of unlabeled training samples.

[0063] The labeled samples are used to constrain the consistency between the impedance inversion results and the true impedance labels, while the unlabeled samples are used for training through seismic reconstruction constraints to improve the model's generalization ability to well-free regions.

[0064] Furthermore, all samples in the labeled training sample set can be standardized, as follows: The standardized raw seismic data can be represented as: Impedance label normalization can be represented as: in, and These are the standardized raw seismic data and impedance labels, respectively. These are the mean and standard deviation of the original seismic data, respectively. These represent the mean and standard deviation of the impedance labels, respectively. X This is the raw earthquake data; Y These are the impedance labels for the raw seismic data. The standardized data is used for subsequent network training.

[0065] Furthermore, to utilize the lateral continuity between adjacent seismic traces, multi-channel input samples can be constructed. This allows for the simultaneous use of longitudinal temporal variation information and lateral spatial continuity information, thereby improving the stability of impedance prediction. Specifically, taking labeled raw seismic data as an example, the multi-channel input sample construction process includes: Let the first i Earthquake records are The width of the adjacent channel window is w Then the first i Each input sample can be represented as: in, For the multi-channel input sample corresponding to the i-th target channel, w The width of the adjacent channel window can be w in this embodiment. =5, meaning that each input sample contains information about the target trace and its left and right neighboring seismic traces.

[0066] For cases where samples of adjacent channels cannot be fully obtained at the boundary location, zero-padding, boundary copying, or mirror padding methods can be used to fill in the adjacent channel window.

[0067] Step S220: Input the original seismic data of labeled training samples into the inversion model to obtain the impedance inversion result, determine the corresponding impedance loss, and input the impedance labels of the labeled training samples into the forward model to obtain the labeled reconstructed seismic data, determine the corresponding labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different dilation scales by introducing multi-scale dilation convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction and outputs the impedance inversion result after splicing and fusion. Among them impedance loss As shown below: in, For inversion model; These are model parameters; For the first i The original seismic data with labeled training samples; For the first i The impedance labels of the labeled training samples are given; N is the number of labeled training samples. This loss term ensures that the impedance inversion results are consistent with the actual logging impedance at the well point location.

[0068] Among them is the labeled seismic reconstruction loss. As shown below: in, Forward model; These are the model parameters.

[0069] This loss term is used to constrain the impedance labels to reconstruct the original seismic response after forward modeling, so that the model training considers both impedance matching and seismic response matching.

[0070] Step S230: Input unlabeled training samples into the inversion model to obtain impedance inversion results, then input the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determine the corresponding unlabeled seismic reconstruction loss.

[0071] Among them, unlabeled seismic reconstruction loss As shown below: in, For the first j The original seismic data consists of *M* unlabeled training samples; M is the number of unlabeled training samples. This loss term does not depend on well logging impedance labels, but rather on seismic forward modeling consistency constraints, allowing unlabeled seismic data to participate in model training. This alleviates the problem of sparse well data in actual oil and gas fields and improves the model's predictive stability in areas far from well points.

[0072] Step S240: The total loss is determined by weighting the impedance loss, labeled seismic reconstruction loss and unlabeled seismic reconstruction loss, and the parameters of the inversion model and forward model are updated by backpropagation. The total loss function is as follows: in, , where are weighting coefficients, used to control the contributions of impedance loss, labeled seismic reconstruction loss, and unlabeled seismic reconstruction loss to the total loss, respectively. The model minimizes Update model parameters and model parameters This allows the impedance inversion results to simultaneously satisfy both well logging constraints and seismic response constraints.

[0073] Step S250: Repeat the above training process until the network converges, and output the trained inversion model as a multi-scale seismic impedance inversion model.

[0074] Example 5: This embodiment of the invention introduces Marmousi2 geological model data for seismic impedance inversion testing, as detailed below: Obtain as attached Figure 5 The Marmousi2 model seismic data records and corresponding data are shown in the attached figure. Figure 6 The actual impedance label shown is from the attached Figure 5 The data shows that the stratigraphy in this region is significantly undulating, with strong local tectonic variations, and well-developed faults and discontinuities, making it a complex region that is difficult to process in impedance inversion. (Appendix) Figure 6 With appendix Figure 5 Correspondingly, the actual impedance distribution of the same region is displayed. It can be seen that the high and low impedance interfaces in this region are relatively clear, and the dipping strata on the left, the complex structural zone in the middle, and the continuous layered structure on the right are all relatively obvious, which can serve as an important reference for evaluating the impedance inversion results.

[0075] Input Appendix Figure 5 The Marmousi2 model seismic data shown is recorded into the multi-scale seismic impedance inversion model constructed in this invention, and the results are shown in the attached figure. Figure 7 The impedance inversion results shown are compared with the attached... Figure 6 By comparison, we can find the appendix Figure 7 The overall stratigraphic structure and impedance variation trends were restored, with the dipping strata on the left, the complex tectonic zone in the middle, and the continuous layered structure on the right all well preserved. In particular, around 2.2s to 2.7s, the prediction results preserved the continuity of the high-impedance layer and the tectonic undulation characteristics well; in the central fault and strong tectonic variation areas, the prediction results were able to reflect local impedance abrupt changes and stratigraphic deformation characteristics.

[0076] To verify the inversion effect of this invention, evaluation indicators for the impedance inversion results of the Marmousi2 model were established as shown in Table 1. The evaluation results show that the method of this invention can achieve low prediction error, and the impedance inversion results have strong consistency with the actual impedance. (See attached table for details.) Figure 6 and attached Figure 7 It can be seen that the inversion results obtained by this method generally recover the spatial variation trend of the true impedance well, and have a good ability to characterize the impedance changes of major stratigraphic interfaces, continuous high impedance layers and complex structural regions. This shows that the present invention can maintain the stability and geological rationality of the impedance inversion results in complex geological models.

[0077] Table 1 Evaluation Indicators for Impedance Inversion Results of the Marmousi2 Model .

[0078] Example 6: As shown in the appendix Figure 8 As shown, this embodiment of the invention discloses a multi-scale seismic impedance inversion method for weak well control conditions, including: Step S310: Obtain seismic data for the target area; Step S320: Input the seismic data of the target area into the multi-scale seismic impedance inversion model to obtain the corresponding impedance inversion results. The multi-scale seismic impedance inversion model is constructed using the methods described in Examples 1 to 4.

[0079] Example 7: As attached Figure 9 As shown in the figure, this invention discloses a device for constructing a multi-scale seismic impedance inversion model under weak well control conditions, characterized in that it includes: The sample construction unit constructs labeled training sample sets and unlabeled training sample sets based on the original seismic data of the historical region; The labeled training unit takes the original seismic data and impedance labels of the labeled training samples as input to the inversion model and the forward model, respectively, and determines the corresponding impedance loss and labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction and outputs the corresponding impedance inversion results after splicing and fusion through upsampling. The unlabeled training unit inputs unlabeled training samples into the inversion model to obtain impedance inversion results, and then inputs the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determines the corresponding unlabeled seismic reconstruction loss. The model parameter update unit determines the corresponding total loss by weighting the impedance loss, labeled seismic reconstruction loss and unlabeled seismic reconstruction loss, and updates the model parameters of the inversion model and the forward model through backpropagation. The iterative unit repeats the above training process until the network converges, and outputs the trained inversion model as a multi-scale seismic impedance inversion model.

[0080] Example 8: As attached Figure 10 As shown, this embodiment of the invention discloses a multi-scale seismic impedance inversion device for weak well control conditions, comprising: The seismic data acquisition unit acquires seismic data for the target area. The inversion unit takes the seismic data of the target area as input to the multi-scale seismic impedance inversion model and obtains the corresponding impedance inversion results.

[0081] Example 9: This embodiment of the invention discloses a storage medium storing a computer program that can be read by a computer. The computer program is configured to execute a method for constructing a multi-scale seismic impedance inversion model or a multi-scale seismic impedance inversion method under weak well control conditions.

[0082] The aforementioned storage media may include, but are not limited to, USB flash drives, read-only memory, portable hard drives, magnetic disks, optical disks, and other media capable of storing computer programs.

[0083] Example 10: This embodiment of the invention discloses an electronic device, including a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement a method for constructing a multi-scale seismic impedance inversion model or a multi-scale seismic impedance inversion method under weak well control conditions.

[0084] The processor described above can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. It can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The memory can include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, portable hard drives, magnetic disks, or optical disks.

[0085] 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 implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0086] 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.

[0087] 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 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] The above content is only a specific embodiment of the present invention, which has strong adaptability and implementation effect. However, the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention. Therefore, equivalent changes made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for constructing a multi-scale seismic impedance inversion model under weak well control conditions, characterized in that, include: Based on the original earthquake data of the historical region, a labeled training sample set and an unlabeled training sample set were constructed. The original seismic data with labeled training samples and impedance labels are input into the inversion model and the forward model respectively to determine the corresponding impedance loss and labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction and outputs the corresponding impedance inversion results after splicing and fusion through upsampling. Input unlabeled training samples into the inversion model to obtain impedance inversion results, then input the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determine the corresponding unlabeled seismic reconstruction loss; The total loss is determined by weighting the impedance loss, labeled seismic reconstruction loss and unlabeled seismic reconstruction loss, and the model parameters of the inversion model and the forward model are updated by backpropagation. Repeat the above training process until the network converges, and output the trained inversion model as a multi-scale seismic impedance inversion model.

2. The method for constructing a multi-scale seismic impedance inversion model under weak well control conditions according to claim 1, characterized in that, The inversion model includes a first feature extraction module, a second feature extraction module, a fusion module, and an upsampling module; The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and then fuses them to obtain fused features. Wavelet-guided attention is then introduced to obtain enhanced features. The second feature extraction module receives seismic data and extracts time-series features; The fusion module receives the enhanced features output by the first feature extraction module and the sequence features output by the second feature extraction module, and performs feature fusion after aligning the number of channels and the time dimension. The upsampling module transposes and convolves or reassembles pixels of the features output by the fusion module to restore the number of sampling points to the original seismic data, thus obtaining the impedance inversion result.

3. The method for constructing a multi-scale seismic impedance inversion model under weak well control conditions according to claim 2, characterized in that, The first feature extraction module includes: a multi-scale dilated convolution module, a fusion module, a wavelet-guided attention module, a compression module, and a feature output module; The multi-scale dilatation convolution module includes multiple parallel dilatation convolution sub-modules with different dilatation scales. Each dilatation convolution sub-module receives the original seismic data and performs time dilatation convolution at the corresponding dilatation scale to obtain seismic reflection features at different dilatation scales. The fusion module stitches and fuses the seismic reflection features at all dilatational scales output by the multi-scale dilatational convolution module; The wavelet-guided attention module determines the spatial attention weights and channel attention weights based on the fusion features output by the fusion module, and applies the spatial attention weights and channel attention weights to the fusion features output by the fusion module to obtain enhanced features; The compression module and the feature output module perform dimensionality reduction and alignment processing on the enhanced features, respectively.

4. The method for constructing a multi-scale seismic impedance inversion model under weak well control conditions according to claim 2, characterized in that, The second feature extraction module includes a bidirectional gated loop unit that receives seismic data and extracts temporal sequence features.

5. The method for constructing a multi-scale seismic impedance inversion model under weak well control conditions according to any one of claims 1 to 4, characterized in that, The total loss function is as follows: in, These are weighting coefficients, used to control the contributions of impedance loss, labeled seismic reconstruction loss, and unlabeled seismic reconstruction loss to the total loss, respectively. Among them impedance loss As shown below: in, For inversion model; These are model parameters; For the first i The original seismic data with labeled training samples; For the first i The impedance labels of the labeled training samples; N is the number of labeled training samples; Among them is the labeled seismic reconstruction loss. As shown below: in, Forward model; These are model parameters; Unlabeled seismic reconstruction loss As shown below: in, For the first j The original seismic data of M unlabeled training samples; M is the number of unlabeled training samples.

6. A multi-scale seismic impedance inversion method for weak well control conditions, characterized in that, include: Acquire seismic data for the target area; Input the seismic data of the target area into the multi-scale seismic impedance inversion model to obtain the corresponding impedance inversion results, wherein the multi-scale seismic impedance inversion model is constructed using the method described in any one of claims 1 to 5.

7. A device for constructing a multi-scale seismic impedance inversion model under weak well control conditions using the method described in any one of claims 1 to 5, characterized in that, include: The sample construction unit constructs labeled training sample sets and unlabeled training sample sets based on the original seismic data of the historical region; The labeled training unit takes the original seismic data and impedance labels of the labeled training samples as input to the inversion model and the forward model, respectively, and determines the corresponding impedance loss and labeled seismic reconstruction loss. The inversion model includes a first feature extraction module and a second feature extraction module. The first feature extraction module extracts seismic reflection features at different expansion scales by introducing multi-scale dilatational convolution and wavelet-guided attention. The second feature extraction module extracts the sequence features in the time direction and outputs the corresponding impedance inversion results after splicing and fusion through upsampling. The unlabeled training unit inputs unlabeled training samples into the inversion model to obtain impedance inversion results, and then inputs the impedance inversion results into the forward model to obtain unlabeled reconstructed seismic data, and determines the corresponding unlabeled seismic reconstruction loss. The model parameter update unit determines the corresponding total loss by weighting the impedance loss, labeled seismic reconstruction loss, and unlabeled seismic reconstruction loss, and updates the model parameters of the inversion model and the forward model through backpropagation. The iterative unit repeats the above training process until the network converges, and outputs the trained inversion model as a multi-scale seismic impedance inversion model.

8. A multi-scale seismic impedance inversion device for weak well control conditions using the method described in claim 6, characterized in that, include: The seismic data acquisition unit acquires seismic data for the target area. The inversion unit takes the seismic data of the target area as input to the multi-scale seismic impedance inversion model and obtains the corresponding impedance inversion results.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps of the method as described in any one of claims 1 to 5 or the method as described in claim 6.

10. A storage medium, characterized in that, The storage medium stores a computer program that can be read by a computer, the computer program being configured to execute the steps of the method as described in any one of claims 1 to 5 or the method as described in claim 6 when it is run.

Images