Seismic inversion method and system, storage medium and device

Abstract

The invention discloses a seismic inversion method and system, a storage medium and equipment, and relates to the technical field of geophysical exploration interpretation, and the method introduces various adaptive prior information through learning, and improves the interpretability and controllability of an inversion process. According to the system, based on a multi-scale feature fusion and dictionary learning method, macroscopic features in a construction model, transverse features in seismic data and vertical features in logging data are integrated together, and seismic constraint inversion is carried out. Model testing verifies that compared with a traditional inversion constraint method, the method has higher resolution and better stability. The validity of the method is also verified through actual data. In conclusion, according to the method provided by the embodiment of the invention, the existing prior information is fully mined and utilized, and the resolution, the precision and the stability of the inversion result are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of geophysical exploration and interpretation technology, and in particular to a seismic inversion method, system, storage medium and equipment. Background Art

[0002] High-resolution seismic inversion is an important technology for achieving accurate imaging of reservoir parameters and plays a vital role in predicting and describing reservoirs. Traditional methods assume that prior information follows a specific mathematical model distribution and cannot meet the requirements of fine characterization of complex reservoirs. The use of machine learning and data-driven methods in high-resolution inversion has become a research hotspot. Considering that reservoir deposition is relatively stable, reservoir parameters in the same working area have a certain consistency and horizontal continuity. Therefore, a set of adaptive feature functions can be extracted from known observation data to represent the inversion parameters. The idea of ​​data-driven prior information extraction can be realized through a sparse representation method based on dictionary learning.

[0003] In 2018, She Bin et al. first proposed post-stack impedance inversion based on logging data using single dictionary learning and sparse representation, and extended it to pre-stack AVO inversion. This method introduces adaptive high-frequency features in logging data into inversion constraints through a data-driven method, effectively improving the vertical resolution and accuracy of the inversion results. However, due to the "one-sided view" of logging data, the inversion results of this type of method have poor lateral continuity and insufficient accuracy. Prior information comes from rich geological data, high-lateral resolution seismic data, and high-vertical resolution logging data, which usually differ in scale, characteristics, and amplitude. Traditional inversion methods represented by geostatistical inversion can integrate prior information from geological, seismic, and logging data to achieve high-resolution imaging of reservoirs. However, in areas where reservoirs change dramatically and are highly heterogeneous, this method cannot accurately describe reservoir characteristics, nor can it use learning to introduce adaptive information in the inversion process, resulting in deviations in the inversion results.

[0004] Seismic inversion information can be divided into three frequency components: low frequency, medium frequency and high frequency. Low frequency information mainly comes from the initial model based on geological data, medium frequency information mainly comes from seismic data, and high frequency information mainly comes from well logging data. They correspond to the macroscopic, horizontal and vertical characteristics of the reservoir, respectively. Therefore, how to effectively introduce these multi-scale feature information to constrain the inversion process is the key to achieving high-resolution and high-precision inversion. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a seismic inversion method, system, storage medium and device, aiming to solve the problem that the prior art cannot effectively introduce multi-scale feature information to constrain the seismic inversion process, resulting in general resolution and accuracy of the seismic inversion results.

[0006] A first aspect of the present invention is to provide a seismic inversion method, the method comprising: Acquire well logging data, calculate a reflection coefficient series according to the well logging data, and calculate seismic wavelets using a least squares method based on a preset convolution model and in combination with seismic traces near the well; Extracting the seismic wavelet within the corresponding time window according to the original borehole seismic trace, and comparing the synthetic borehole seismic trace obtained by convolution of the seismic wavelet with the original borehole seismic trace; According to the seismic layer information and the geological structure interpretation information, the seismic interpretation layer is divided into multiple sublayers according to the preset geological sedimentation rules, and a geological framework structure is constructed. On the basis of the single well-seismic joint calibration, the well logging data is extrapolated and interpolated along the seismic interpretation layer using the preset interpolation algorithm to construct an initial model; The logging data, the seismic wavelet and the initial model are obtained, and the macroscopic features, lateral features and vertical features required for seismic inversion are extracted through dictionary learning and sparse matrix description, and the macroscopic features, lateral features and vertical features are incorporated into the inversion objective function using regular expressions to achieve seismic inversion.

[0007] According to one aspect of the above technical solution, the steps of extracting macroscopic features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description include: Acquire seismic data, and divide the seismic data into a plurality of seismic sample data; The seismic sample data are collected sequentially from top to bottom and from left to right to generate training samples and learn a two-dimensional feature dictionary to extract the features in the lateral direction of the reservoir; The atomic form of the two-dimensional feature dictionary is used to reflect the lateral characteristics of seismic data.

[0008] According to one aspect of the above technical solution, the objective functions of normalizing the seismic data and learning the two-dimensional feature dictionary are: ; ; In the formula, The original seismic data i The sampling point value, n is the sampling sequence length, is the earthquake dataset, is a two-dimensional feature dictionary derived from seismic data, is the regularization coefficient, A is the sparse coefficient matrix corresponding to the training sample matrix, C is the sparsity, Kis the dictionary atom length.

[0009] According to one aspect of the above technical solution, the steps of extracting macroscopic features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description include: Acquire low-frequency information and high-frequency information in the initial model and the well logging data, and perform joint dictionary learning according to the low-frequency information and the high-frequency information; The sparse reconstruction technology is used to restore the vertical high-frequency information of the entire section.

[0010] According to one aspect of the above technical solution, the objective function of sparse reconstruction is: ; In the formula, is the model data of the initial model, is the learned joint dictionary part containing low-frequency features, is the sparse reconstruction coefficient, To reconstruct the high-frequency information, is the learned part of the joint dictionary containing high-frequency features.

[0011] According to one aspect of the above technical solution, the macroscopic features, lateral features and vertical features are incorporated into the inversion objective function by using regular expressions to achieve the step of seismic inversion. The expression of the time domain seismic forward model is: ; In the formula, For earthquake channels, is the seismic wavelet, is the reflectivity, is random noise; Among them, the approximate relationship between reflectivity and impedance is expressed as: ; In the formula, and Respectively i reflectivity and impedance of the layer; The approximate relationship between the reflectivity and impedance is expressed in matrix form as follows: ; Among them, the single-channel seismic forward matrix model is expressed as: ; In the formula, d For a single-channel seismic record, m is a single-channel impedance sequence, G is the forward matrix, which represents the convolution process of wavelet, where: , , .

[0012] According to one aspect of the above technical solution, a regular expression is used to incorporate macroscopic features, lateral features and vertical features into the inversion objective function to achieve the step of seismic inversion. The multi-channel inversion objective function with regularization constraints is: ; Among them, the inversion objective function is: ; In the formula, is the earthquake dataset, is the forward modeling matrix, M are the multi-channel parameters to be inverted, is the high-frequency model obtained by joint dictionary learning and sparse reconstruction, is the initial low-frequency model built based on geological data, is the sparse reconstruction model of the intermediate inversion result of the two-dimensional feature dictionary, is the regularization coefficient, which is used to adjust the weights between different regularization constraints. is the sparse representation coefficient matrix of the joint dictionary; Among them, the first item in the inversion objective function is the decision item of the objective function, the second item is the constraint item introduced by low-frequency macroscopic features, the third item is the constraint item introduced by high-frequency vertical features, the fourth item is the constraint item introduced by medium-frequency lateral features, and the fifth item is the feature fusion constraint item.

[0013] A second aspect of the present invention is to provide a seismic inversion system, which is applied to the method described in the above technical solution, and the system comprises: The first processing module is used to obtain well logging data, calculate the reflection coefficient series according to the well logging data, and calculate the seismic wavelet by using the least square method based on a preset convolution model and in combination with the seismic trace near the well; The second processing module is used to extract the seismic wavelet within the corresponding time window according to the original well-side seismic trace, and compare the synthetic well-side seismic trace obtained by convolution of the seismic wavelet with the original well-side seismic trace; The third processing module is used to divide the seismic interpretation layer into multiple sublayers according to the preset geological sedimentation rules based on the seismic layer information and the geological structure interpretation information, build a geological framework structure, and use a preset interpolation algorithm to extrapolate and interpolate the logging data along the seismic interpretation layer on the basis of single well-seismic joint calibration to build an initial model; The fourth processing module is used to obtain the logging data, the seismic wavelet and the initial model, extract the macro features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description, and use regular expressions to incorporate the macro features, lateral features and vertical features into the inversion objective function to achieve seismic inversion.

[0014] A third aspect of the present invention is to provide a readable storage medium having a computer program stored thereon, which implements the method described in the above technical solution when executed by a processor.

[0015] The fourth aspect of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the method described in the above technical solution when executing the computer program.

[0016] Compared with the prior art, the seismic inversion method, system, storage medium and device shown in the present invention have the following beneficial effects: The method shown in the present invention introduces a variety of adaptive prior information through learning, while improving the interpretability and controllability of the inversion process. Based on multi-scale feature fusion and dictionary learning methods, the method integrates the macroscopic features in the structural model, the lateral features in the seismic data, and the vertical features in the logging data to perform seismic constrained inversion. Model testing has verified that compared with the traditional inversion constraint method, this method has higher resolution and better stability. The effectiveness of the method has also been verified by actual data. In summary, the method proposed in the present invention fully mines and utilizes the existing prior information, significantly improving the resolution, accuracy and stability of the inversion results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 It is a schematic diagram of the flow of the seismic inversion method shown in an embodiment of the present invention; Figure 2 These are three stages of multi-scale feature fusion constraint inversion based on dictionary learning in an embodiment of the present invention; Figure 3 A schematic diagram of two-dimensional feature dictionary learning and sparse reconstruction based on seismic data in an embodiment of the present invention; Figure 4 A diagram of a high-frequency information model reconstruction process based on joint dictionary learning in an embodiment of the present invention; Figure 5 Schematic diagram of a multi-scale constrained inversion process in an embodiment of the present invention; Figure 6is a schematic diagram of a Marmousi model in an embodiment of the present invention; Figure 7 Schematic diagram of different inversion results of the Malmucci model in a 6 dB Gaussian noise environment in an embodiment of the present invention; Figure 8 It is a schematic diagram of the quantitative error comparison between the impedance inversion result under 6 dB Gaussian noise and the real model in the embodiment of the present invention; Fig. 9 A schematic diagram of actual seismic profile data in an embodiment of the present invention; Fig.10 It is a schematic diagram of the inversion result of the actual post-stack seismic data profile in the embodiment of the present invention; Fig.11 It is a schematic diagram of the impedance inversion result of the blind well (W2-2132) test in the embodiment of the present invention; Fig.12 Schematic diagram of spectrum analysis of blind well inversion results in an embodiment of the present invention; Fig.13 It is a structural block diagram of the seismic inversion system shown in the embodiment of the invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. Several embodiments of the present invention are shown in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0019] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0021] Embodiment 1 See also Figure 1The first embodiment of the present invention provides a seismic inversion method, the method comprising steps S10 to S40: Step S10, acquiring well logging data, calculating a reflection coefficient series according to the well logging data, and calculating seismic wavelets using a least squares method based on a preset convolution model and in combination with seismic traces near the well.

[0022] In this embodiment, a deterministic wavelet extraction method is used to extract seismic wavelets. The deterministic wavelet extraction method is to use the reflection coefficient sequence calculated by the logging data when there is logging data, and then use the least square method based on the convolution model combined with the seismic trace near the well to estimate the seismic wavelet.

[0023] Step S20, extracting the seismic wavelet within the corresponding time window according to the original borehole seismic trace, and comparing the synthetic borehole seismic trace obtained by convolution of the seismic wavelet with the original borehole seismic trace.

[0024] In this embodiment, well seismic calibration is a bridge connecting seismic data and logging data. The method is to extract seismic wavelets within the corresponding time window range from the original well seismic traces, and then compare the synthetic well seismic traces obtained by wavelet convolution with the original well seismic traces.

[0025] Step S30, according to the seismic layer information and the geological structure interpretation information, the seismic interpretation layer is divided into multiple sub-layers according to the preset geological sedimentation rules, and a geological framework structure is constructed. On the basis of the single well-seismic joint calibration, the preset interpolation algorithm is used to extrapolate and interpolate the logging data along the seismic interpretation layer to construct an initial model.

[0026] In this embodiment, to establish the initial model, we first need to subdivide multiple layers between the interpretation layers according to certain geological sedimentary laws based on the seismic layer and geological structure interpretation information, and build a relatively fine geological framework structure. Then, under the fine geological framework model, we use appropriate mathematical interpolation methods (Kriging algorithm, etc.), and on the basis of single well-seismic joint calibration, we extrapolate and interpolate the logging data along the seismic interpretation layer to build a smooth and closed initial model. Step S40, obtaining the logging data, the seismic wavelet and the initial model, extracting the macro features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description, and incorporating the macro features, lateral features and vertical features into the inversion objective function using regular expressions to achieve seismic inversion.

[0027] In this embodiment, multi-scale feature dictionary learning, such as Figure 2The figure shows the three stages of multi-scale feature fusion constrained inversion. The first stage is the data input layer, which collects the model, seismic and logging data required for inversion; the second stage is the feature extraction layer, which extracts the macro, lateral and vertical features required for inversion through dictionary learning and sparse feature description technology; the third stage is the decision output layer, which incorporates the extracted multi-scale prior features into the inversion objective function through regularization technology.

[0028] The dictionary learning method used in this embodiment is a 2D-KSVD dictionary learning method based on seismic data. Specifically: A seismic profile is usually composed of multiple seismic traces arranged in spatial order, which contains rich two-dimensional stratigraphic distribution characteristics. Normalizing each seismic profile is equivalent to obtaining image data with an amplitude range of [0,1].

[0029] In this embodiment, based on the idea of ​​taking small samples in the image data training process, the seismic image is divided into small samples as training data, and 2D-KSVD dictionary learning is performed to extract the characteristics of the lateral direction of the reservoir. The normalization processing of seismic data and the 2D-KSVD dictionary learning objective function are as follows: (1) (2) In the formula, The original seismic data i The sampling point value, n is the sampling sequence length, is an earthquake data set, specifically, a normalized earthquake data training sample. is a two-dimensional feature dictionary derived from seismic data, is the regularization coefficient, A is the sparse coefficient matrix corresponding to the training sample matrix, C is the sparsity, K is the length of dictionary atoms. The learning process adopts the K-SVD dictionary learning algorithm, which uses greedy algorithms such as orthogonal matching pursuit (OMP) and singular value decomposition (SVD) for sparse coding and dictionary update.

[0030] Among them, the 2D-KSVD dictionary learning process based on seismic data is as follows: Figure 3 As shown, Figure 3 The red box in (a) shows the selected 10 seismic sample data. After the data is collected in small blocks by sliding the window from top to bottom and from left to right, training samples are generated and the KSVD dictionary is learned. The schematic diagram of some atoms in the dictionary is shown in Figure 3 As shown in (b) in the figure, it can be found that the atomic morphology of the two-dimensional feature dictionary can effectively reflect the lateral characteristics of seismic data ( Figure 3 (a) and Figure 3 The blocks of the same color in (b) represent the same structural features), which verifies the effectiveness and feasibility of this method in extracting seismic data features. Finally, the original sample data can be restored through sparse reconstruction technology ( Figure 3 (c) in the figure.

[0031] Compared with the prior art, the seismic inversion method shown in the present invention has the following beneficial effects: The method shown in this embodiment introduces a variety of adaptive prior information through learning, while improving the interpretability and controllability of the inversion process. Based on multi-scale feature fusion and dictionary learning methods, this method integrates the macroscopic features in the structural model, the lateral features in the seismic data, and the vertical features in the logging data to perform seismic constrained inversion. Model testing has verified that this method has higher resolution and better stability than the traditional inversion constraint method. The effectiveness of this method has also been verified by actual data. In summary, the method proposed in this embodiment fully mines and utilizes the existing prior information, significantly improving the resolution, accuracy and stability of the inversion results.

[0032] Embodiment 2 The second embodiment of the present invention also provides a seismic inversion method. The method shown in this embodiment is basically similar to the method shown in the first embodiment, except that: In this embodiment, the logging data, the seismic wavelet and the initial model are obtained, and the macroscopic features, lateral features and vertical features required for seismic inversion are extracted through dictionary learning and sparse matrix description, and the macroscopic features, lateral features and vertical features are incorporated into the inversion objective function using regular expressions to achieve the seismic inversion step. The dictionary learning adopts a joint KSVD dictionary learning method based on the model and logging data.

[0033] Specifically: Well logging data provides detailed full-band stratigraphic information, so the high-frequency information at the wellbore can be directly used for high-resolution inversion. However, due to the limited and scattered well logging data in the working area, it is difficult to directly obtain effective high-frequency information in the well-free area for high-resolution inversion. In the field of image processing, super-resolution image reconstruction algorithms can be used to restore low-resolution images to high-resolution images, and the process uses a joint dictionary learning method.

[0034] The joint dictionary is a sparse dictionary that uses samples from different datasets as training data and contains a variety of data structure features. It is widely used in super-resolution image reconstruction. For two sets of samples and with correlation, the dictionary and of samples and are found through joint dictionary learning, so that the two sets of dictionary atoms have a certain correlation.

[0035] The objective function for optimizing the training process is: (3) In the formula, and Through the dictionary and , using the common sparse factor Sparse representation, together they can be written as .

[0036] The above formula can be converted into and The dictionary learning problem in the relevant feature space is expressed as: (4) Therefore, based on the idea of ​​super-resolution image reconstruction, in this embodiment, model data and logging data are used as joint training samples, and a vertical high-frequency information reconstruction method based on joint dictionary learning is proposed.

[0037] This method uses the low-frequency and high-frequency information in the model data and logging data to perform joint dictionary learning. Figure 4 The sparse reconstruction technique shown recovers the vertical high-frequency information of the entire section, providing reliable high-frequency prior information constraints for high-resolution inversion of reservoir parameters.

[0038] Its sparse reconstruction objective function is: (5) In the formula, is the model data of the initial model, is the learned joint dictionary part containing low-frequency features, is the sparse reconstruction coefficient, To reconstruct the high-frequency information, is the learned part of the joint dictionary containing high-frequency features.

[0039] Among them, the inversion objective function is constructed. According to the convolution model proposed by Robinson, the time domain seismic forward model can be expressed as follows: (6) In the formula, For earthquake channels, is the seismic wavelet, is the reflectivity, is random noise.

[0040] According to Russell theory, the approximate relationship between reflectivity and impedance is as follows: (7) In the formula, and Respectively represent i The reflectivity and impedance of the layer.

[0041] Equation (6) can be expressed in matrix form by adding equation (7): (8) The single-channel seismic forward modeling matrix model is expressed as: (9) in, d For a single-channel seismic record, m is a single-channel impedance sequence, G is the forward matrix, representing the convolution process of wavelet.

[0042] In this embodiment, multi-channel inversion is considered, and the forward model can be expressed by the following formula: (10) in, , , .

[0043] It should be noted that seismic inversion is an optimization problem. Since it is difficult to solve, regularization methods are usually introduced to reduce the multiplicity of solutions.

[0044] The objective function of multi-channel inversion with regularization constraints is: (11) Based on the multi-scale features contained in geological data, seismic data and logging data, the low, medium and high frequency band prior information required for inversion is effectively constructed by using structural models, two-dimensional feature dictionary learning, joint dictionary learning and sparse reconstruction technology. At the same time, in order to improve the matching degree between the fusion of prior information and the decision-making process, based on the multi-channel inversion framework, the three types of prior information are integrated into the inversion decision-making process through regularization technology, and the decision-making effect of multi-frequency information is improved by adjusting the regularization coefficient.

[0045] The inversion objective function is as follows: (12) In the formula, is a seismic data set, specifically the actual seismic data used in the inversion calculation. is the forward modeling matrix, M are the multi-channel parameters to be inverted, is the high-frequency model obtained by joint dictionary learning and sparse reconstruction, is the initial low-frequency model built based on geological data, is the sparse reconstruction model of the intermediate inversion result of the two-dimensional feature dictionary, is the regularization coefficient, which is used to adjust the weights between different regularization constraints. is the sparse representation coefficient matrix of the joint dictionary; Among them, in the inversion objective function, the first item is the decision item of the objective function, the second item is the constraint item introduced by the low-frequency macroscopic features, the third item is the constraint item introduced by the high-frequency vertical features, the fourth item is the constraint item introduced by the medium-frequency lateral features, and the fifth item is the feature fusion constraint item.

[0046] The schematic diagram of the multi-scale feature fusion constraint inversion process is as follows: Figure 5 In this embodiment, the prior information of the macroscopic, lateral and vertical characteristics of the marmousi model is taken as an example, such as Figure 5 (a) Figure 5 (b) and Figure 5 As shown in (c) in Figure 5 (d) shows the process of fusion constrained inversion of multi-scale features extracted by sliding window blocks in the same area.

[0047] In order to verify the inversion effect of this method, this embodiment uses marmousi as a test model (the number of seismic traces in the model is 377, the sampling interval is 2 ms, and the number of sampling points is 350). The generated post-stack noise seismic record is as follows: Figure 6 As shown in (a) in the figure, the actual seismic impedance model is as follows Figure 6 As shown in (b) in the figure. Based on the seismic profile data, a small sliding window is used as the training sample (the window length is 50 sampling points, the window width is 3 channels, and the window sliding step is 1) for dictionary learning of the atomic representation of the two-dimensional feature dictionary. Based on the 5 randomly selected wave impedance data, small blocks are taken to generate training samples for dual joint dictionary learning. Figure 7 As shown, in this embodiment, the traditional Tikhonov inversion ( Figure 7 (a) in the sparse dictionary inversion ( Figure 7 (b) in the above), geostatistical inversion ( Figure 7 (c) in the paper) and the multi-scale feature fusion constraint method ( Figure 7 High-resolution seismic inversion results in (d).

[0048] from Figure 7It can be seen that Tikhonov inversion only introduces the prior information of the initial model, and the resolution of the inversion result is low. Sparse dictionary inversion uses learning methods to introduce prior information of well logging. The vertical resolution of the inversion result is high, but the lateral continuity is poor. Geostatistical inversion combines the prior information of modeling, logging and seismic data. The resolution of the inversion result is high, but the accuracy needs to be improved. The multi-scale feature fusion constrained seismic inversion method adopts a learning method and introduces prior information of the model, logging and seismic data. Therefore, its inversion results have higher resolution and accuracy. This method can effectively suppress the interference of horizontal noise, and the inversion results are less affected by noise interference and are closer to the real model. At the same time, Figure 8 As shown in the figure, in this embodiment, the error comparison between the inversion results of the four methods and the true model is quantitatively analyzed. It can be seen that compared with other methods, the root mean square error curve of the multi-scale feature fusion constrained inversion method ( Figure 8 (a) in the figure is smoother and closer to the true model ( Figure 8 This reflects that the multi-scale feature fusion constrained inversion result has better horizontal continuity, better correlation between adjacent trajectories, and is closer to the true model.

[0049] In order to verify the actual application effect of the method shown in this embodiment, the constrained inversion method of multi-scale feature fusion is applied to field post-stack impedance inversion (the number of seismic traces is 85, the sampling interval is 1 ms, and the number of sampling points is 130). Fig. 9 It is a profile of actual seismic data, where the sliding window length is 20, the width is 3, the sliding step is 1, and the block window slides from top to bottom and from left to right along the parallel direction of the layer to generate training data. Fig.10 This is a schematic diagram of the inversion results of the actual post-stack seismic data profile. Fig.10 (a) in the figure corresponds to the Tikhonov inversion, where Tikhonov is a person's name. Fig.10 (b) in the figure corresponds to the sparse dictionary inversion. Fig.10 (c) in the figure corresponds to the geostatistical inversion. Fig.10 (d) in the figure corresponds to the multi-scale feature fusion constraint inversion. Fig.10 It can be seen that compared with the traditional inversion results, the results of multi-scale feature fusion constraints are more consistent with the actual distribution of the reservoir, and the reservoir interface is clearer. In order to more intuitively compare the actual effect of the multi-scale feature fusion constraint inversion method, the inversion curve of Well W2-2132 is also extracted in this embodiment. Fig.11 , Fig.11 (a) in the figure corresponds to Tikhonov inversion. Fig.11 (b) in the figure corresponds to the sparse dictionary inversion. Fig.11 (c) in the figure corresponds to the geostatistical inversion. Fig.11(d) in the figure corresponds to the multi-scale feature fusion constraint inversion. Fig.11 The impedance inversion results of the blind well test show that the multi-scale feature fusion constrained inversion results are more consistent with the real well curve and contain more detailed information. The quantitative evaluation results (root mean square error and similarity) in Table 1 also show that the inversion results of the multi-scale feature fusion constrained inversion method shown in this embodiment are better than those of the traditional method.

[0050] Table 1

[0051] At the same time, through the spectrum analysis of the inversion results, such as Fig.12 , Fig.12 (a) in the figure corresponds to Tikhonov inversion. Fig.12 (b) in the figure corresponds to the sparse dictionary inversion. Fig.12 (c) in the figure corresponds to the geostatistical inversion. Fig.12 (d) in the figure corresponds to multi-scale feature fusion constrained inversion. It can be seen that the method shown in this embodiment contains richer high-frequency information and is closer to real oil well data. At the same time, it is also found in this embodiment that the inversion results based on the learning method are better than the traditional method, which is related to the adaptive ability of the inversion algorithm. In short, multi-scale feature fusion constrained inversion can effectively improve the accuracy and resolution of the inversion results of actual data.

[0052] Embodiment 3 See also Fig.13 The third embodiment of the present invention provides a seismic inversion system, which is applied to the method described in any of the above embodiments. The system includes: a first processing module 10, a second processing module 20, a third processing module 30, and a fourth processing module 40.

[0053] The first processing module 10 is used to obtain well logging data, calculate the reflection coefficient series according to the well logging data, and calculate the seismic wavelet by using the least square method based on a preset convolution model and in combination with the seismic trace near the well; The second processing module 20 is used to extract the seismic wavelet within the corresponding time window according to the original borehole seismic trace, and compare the synthetic borehole seismic trace obtained by convolution of the seismic wavelet with the original borehole seismic trace; The third processing module 30 is used to divide the seismic interpretation layer into multiple sublayers according to the preset geological sedimentation rules based on the seismic layer information and the geological structure interpretation information, build a geological framework structure, and use a preset interpolation algorithm to extrapolate and interpolate the logging data along the seismic interpretation layer on the basis of single well well-seismic joint calibration to build an initial model; The fourth processing module 40 is used to obtain the logging data, the seismic wavelet and the initial model, extract the macro features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description, and use regular expressions to incorporate the macro features, lateral features and vertical features into the inversion objective function to achieve seismic inversion.

[0054] Compared with the prior art, the seismic inversion method, system, storage medium and device shown in the present invention have the following beneficial effects: The system shown in this embodiment introduces a variety of adaptive prior information through learning, while improving the interpretability and controllability of the inversion process. Based on multi-scale feature fusion and dictionary learning methods, the system integrates the macroscopic features in the structural model, the lateral features in the seismic data, and the vertical features in the logging data to perform seismic constrained inversion. Model testing has verified that this method has higher resolution and better stability than the traditional inversion constraint method. The effectiveness of this method has also been verified by actual data. In summary, the system proposed in this embodiment fully exploits and utilizes the existing prior information, significantly improving the resolution, accuracy and stability of the inversion results.

[0055] Embodiment 4 A fourth embodiment of the present invention provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method described in any one of the above embodiments is implemented.

[0056] Embodiment 5 A fifth embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the method described in any of the above embodiments when executing the computer program.

[0057] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0058] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A seismic inversion method, characterized in that: The method comprises: Acquire well logging data, calculate a reflection coefficient series according to the well logging data, and calculate seismic wavelets using a least squares method based on a preset convolution model and in combination with seismic traces near the well; Extracting the seismic wavelet within the corresponding time window according to the original borehole seismic trace, and comparing the synthetic borehole seismic trace obtained by convolution of the seismic wavelet with the original borehole seismic trace; According to the seismic layer information and the geological structure interpretation information, the seismic interpretation layer is divided into multiple sublayers according to the preset geological sedimentation rules, and a geological framework structure is constructed. On the basis of the single well-seismic joint calibration, the well logging data is extrapolated and interpolated along the seismic interpretation layer using the preset interpolation algorithm to construct an initial model; The logging data, the seismic wavelet and the initial model are obtained, and the macroscopic features, lateral features and vertical features required for seismic inversion are extracted through dictionary learning and sparse matrix description, and the macroscopic features, lateral features and vertical features are incorporated into the inversion objective function using regular expressions to achieve seismic inversion.

2. The seismic inversion method according to claim 1, characterized in that: The steps of extracting macroscopic features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description include: Acquire seismic data, and divide the seismic data into a plurality of seismic sample data; The seismic sample data are collected sequentially from top to bottom and from left to right to generate training samples and learn a two-dimensional feature dictionary to extract the features in the lateral direction of the reservoir; The atomic form of the two-dimensional feature dictionary is used to reflect the lateral characteristics of seismic data.

3. The seismic inversion method according to claim 2, characterized in that: The objective functions of normalizing seismic data and learning the two-dimensional feature dictionary are: ； ； In the formula, The original seismic data i The sampling point value, n is the sampling sequence length, is the earthquake dataset, is a two-dimensional feature dictionary derived from seismic data, is the regularization coefficient, A is the sparse coefficient matrix corresponding to the training sample matrix, C is the sparsity, K is the dictionary atom length.

4. The seismic inversion method according to claim 1, characterized in that: The steps of extracting macroscopic features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description include: Acquire low-frequency information and high-frequency information in the initial model and the well logging data, and perform joint dictionary learning according to the low-frequency information and the high-frequency information; The sparse reconstruction technology is used to restore the vertical high-frequency information of the entire section.

5. The seismic inversion method according to claim 4, characterized in that: The objective function of sparse reconstruction is: ； In the formula, is the model data of the initial model, is the learned joint dictionary part containing low-frequency features, is the sparse reconstruction coefficient, To reconstruct the high-frequency information, is the learned part of the joint dictionary containing high-frequency features.

6. The seismic inversion method according to claim 1, characterized in that: The macroscopic features, lateral features and vertical features are incorporated into the inversion objective function using regular expressions to achieve the seismic inversion step. The expression of the time-domain seismic forward model is: ； In the formula, For earthquake channels, is the seismic wavelet, is the reflectivity, is random noise; Among them, the approximate relationship between reflectivity and impedance is expressed as: ； In the formula, and Respectively i reflectivity and impedance of the layer; The approximate relationship between the reflectivity and impedance is expressed in matrix form as follows: ； Among them, the single-channel seismic forward matrix model is expressed as: ； In the formula, d For a single-channel seismic record, m is a single-channel impedance sequence, G is the forward matrix, which represents the convolution process of wavelet, where: ， ， 。 7. The seismic inversion method according to claim 1, characterized in that: Regular expressions are used to incorporate macroscopic features, lateral features, and vertical features into the inversion objective function to achieve the steps of seismic inversion. The multi-channel inversion objective function with regularization constraints is: ； Among them, the inversion objective function is: ； In the formula, is the earthquake dataset, is the forward modeling matrix, M are the multi-channel parameters to be inverted, is the high-frequency model obtained by joint dictionary learning and sparse reconstruction, is the initial low-frequency model built based on geological data, is the sparse reconstruction model of the intermediate inversion result of the two-dimensional feature dictionary, is the regularization coefficient, which is used to adjust the weights between different regularization constraints. is the sparse representation coefficient matrix of the joint dictionary; Among them, the first item in the inversion objective function is the decision item of the objective function, the second item is the constraint item introduced by low-frequency macroscopic features, the third item is the constraint item introduced by high-frequency vertical features, the fourth item is the constraint item introduced by medium-frequency lateral features, and the fifth item is the feature fusion constraint item.

8. A seismic inversion system, characterized in that: The method applied to any one of claims 1 to 7, wherein the system comprises: The first processing module is used to obtain well logging data, calculate the reflection coefficient series according to the well logging data, and calculate the seismic wavelet by using the least square method based on a preset convolution model and in combination with the seismic trace near the well; The second processing module is used to extract the seismic wavelet within the corresponding time window according to the original well-side seismic trace, and compare the synthetic well-side seismic trace obtained by convolution of the seismic wavelet with the original well-side seismic trace; The third processing module is used to divide the seismic interpretation layer into multiple sublayers according to the preset geological sedimentation rules based on the seismic layer information and the geological structure interpretation information, build a geological framework structure, and use a preset interpolation algorithm to extrapolate and interpolate the logging data along the seismic interpretation layer on the basis of single well-seismic joint calibration to build an initial model; The fourth processing module is used to obtain the logging data, the seismic wavelet and the initial model, extract the macro features, lateral features and vertical features required for seismic inversion through dictionary learning and sparse matrix description, and use regular expressions to incorporate the macro features, lateral features and vertical features into the inversion objective function to achieve seismic inversion.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of being executed on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images