Earthquake elastic parameter inversion model establishing method and device based on multi-view learning
Through the multi-view learning method, the seismic data characteristics of different views are extracted and fused, and the problems of low convergence and multi-solvency in seismic elastic parameter in the prior art are solved, thereby achieving higher inversion accuracy and spatial continuity .
Patent Information
- Application Number
- CN202310502951.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-06
- Publication Date
- 2024-11-08
AI Technical Summary
In the seismic elastic parameter inversion, the prior art fails to fully utilize the deep network to extract the high-dimensional features of different view data, resulting in the blind well matching degree of the inversion result not high enough, and there are wavelet estimation errors and multi-solution problems.
Using a multi-view learning method, using time information and seismic information superimposed by multiple angles as multi-views, high-dimensional features are extracted through the multi-view learning module, and highly complete feature data is obtained through the high-dimensional feature fusion module. It is used to establish a seismic elastic parameter inversion model, reduce multi-solvency and improve inversion accuracy.
The accuracy and spatial continuity of seismic elastic parameters inversion are improved, the degree of coincidence of blind wells is significantly improved, the multi-solvency of inversion is reduced, and the inversion accuracy is achieved.
Smart Images

Figure CN118915144A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of geophysical technology, and in particular to a method and device for establishing a seismic elastic parameter inversion model based on multi-view learning. Background Art
[0002] In the field of exploration geophysics, seismic inversion is usually divided into two categories: post-stack inversion and pre-stack inversion according to the different seismic data used, because compared with post-stack seismic data, pre-stack seismic data covers more abundant geological information, lithology information and oil and gas content information underground. Through pre-stack seismic data, not only can we understand the wave impedance information, but also obtain a lot of other elastic parameter information, which has made great contributions to enriching the description methods of reservoirs and complex reservoirs and enhancing the ability of fluid detection. The pre-stack seismic elastic parameter inversion method (mainly referring to elastic impedance inversion) is a very important achievement in the field of oil and gas exploration in recent years and has been widely used. Elastic impedance (Elastic Impedance, EI) is a generalization of the concept of acoustic impedance (Acoustic Impedance, AI). Connolly defined elastic impedance as a function of longitudinal wave velocity, shear wave velocity, density and incident angle in 1999. It is based on the mapping relationship between the Aki-Richards approximate equation and the reflection coefficient under any incident angle. Traditional elastic parameter inversion methods usually assume a simplified linear model, while the actual geophysical process is nonlinear, which leads to large inversion errors.
[0003] With the progress and development of exploration technology and theory, the algorithms for seismic elastic parameter inversion have gradually evolved from linear and generalized linear inversion to nonlinear inversion. In recent years, some deep learning methods have also been applied to the field of seismic inversion. Feng.R et al. used convolutional networks and fully connected layers as inversion models, used the Aki-Richards approximate equation as a forward constraint, and used the error between the synthetic seismic signal and the real seismic signal to update the inversion model. The inversion results under this unsupervised learning mode are relatively reliable and have clear physical meanings. However, since the wavelet is time-space varying in the real situation, there are errors in the estimated wavelet, and the fit of the blind well is very low. In order to improve the fit of the blind well, some researchers used semi-supervised learning to train the inversion model. They not only used the Aki-Richards approximate equation to measure the error between the synthetic seismic signal and the real seismic signal, but also added logging constraints to effectively broaden the frequency band of the inversion results and improve the fit of the blind well. However, there is still the problem of wavelet estimation error. Furthermore, in order to reduce the multiple solutions of the inversion, some researchers have added new constraints when solving the optimization problem, such as adding prior low-frequency signals, seismic attribute signals, sparse reflection coefficient constraints in addition to seismic signals to the input signal, or using prior physical models to synthesize data, and then combining transfer learning to apply it to actual seismic data. Although these methods can improve the inversion effect, the reliability of the prior model needs to be guaranteed. Summary of the invention
[0004] The inventors found that the above-mentioned related methods only input the seismic signals or their attributes stacked at different angles as multi-channel data into the deep network, and did not make good use of the natural advantage of the deep network to extract high-dimensional features of different view data separately. Instead, they directly extracted common features and then directly performed downstream tasks (for example, velocity inversion), without considering the complementarity and redundancy between different observation data. In actual data, the blind well consistency of elastic inversion results is still not high enough.
[0005] In order to at least partially solve the technical problems existing in the prior art, the inventors have made the present invention, and through a specific implementation method, provide a method and device for establishing a seismic elastic parameter inversion model based on multi-view learning, using time information and seismic information superimposed at multiple different angles as multi-views, and establishing a seismic elastic parameter inversion model based on multi-view learning.
[0006] In a first aspect, an embodiment of the present invention provides a method for establishing a seismic elastic parameter inversion model based on multi-view learning, comprising:
[0007] The forward model is trained using a first sample set until the number of iterations reaches a first set number of iterations, wherein the samples in the first sample set include single-channel time information, multiple seismic information superimposed at different angles, and well logging elastic parameter data;
[0008] Using the first sample set to train an inversion model until the number of iterations reaches a second set number of iterations, the inversion model comprising a multi-view learning module, a high-dimensional feature fusion module and a multi-task learning module;
[0009] Using the first sample set and the second sample set without logging labels, with the current forward model as a constraint, the multi-task learning module in the inversion model is iteratively trained until the number of iterations reaches a third set number of iterations, and the optimal inversion model is selected according to the forward loss value in the iterative process.
[0010] In a second aspect, an embodiment of the present invention provides a method for inversion of seismic elastic parameters, comprising:
[0011] The time information of the target stratum in the study area and multiple seismic information superimposed at different angles are input into the seismic elastic parameter inversion model, and the elastic parameter data are obtained according to the output results of the model. The seismic elastic parameter inversion model is established according to the above method.
[0012] In a third aspect, an embodiment of the present invention provides a device for establishing a seismic elastic parameter inversion model based on multi-view learning, comprising:
[0013] A first iterative training module is used to train the forward model using a first sample set until the number of iterations reaches a first set number of iterations, wherein the samples in the first sample set include single-channel time information, multiple seismic information superimposed at different angles, and well logging elastic parameter data;
[0014] A second iterative training module, used for training an inversion model using the first sample set until the number of iterations reaches a second set number of iterations, the inversion model comprising a multi-view learning module, a high-dimensional feature fusion module and a multi-task learning module;
[0015] The third iterative training module is used to use the first sample set and the second sample set without logging labels, take the current forward model as a constraint, iteratively train the multi-task learning module in the inversion model, until the number of iterations reaches a third set number of iterations, and select the optimal inversion model according to the forward loss value during the iteration process.
[0016] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the method for establishing a seismic elastic parameter inversion model based on multi-view learning is implemented, or the method for inversion of seismic elastic parameters is implemented.
[0017] In a fifth aspect, an embodiment of the present disclosure provides a server, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the above-mentioned method for establishing a seismic elastic parameter inversion model based on multi-view learning, or implements the above-mentioned seismic elastic parameter inversion method.
[0018] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:
[0019] (1) The method for establishing a seismic elastic parameter inversion model based on multi-view learning provided by an embodiment of the present invention uses time information and seismic information superimposed at multiple different angles as a single view respectively. Each view is an observation result of the same object from a different perspective. There is both correlation and information complementarity between these observation results. From the perspective of multi-view learning, the high-dimensional features of data from different views are first extracted using a multi-view learning module respectively, and then fused by a high-dimensional feature fusion module to obtain highly complete feature data, which can better serve the inversion task and enable the established inversion model to have a higher seismic elastic parameter inversion accuracy.
[0020] (2) In the method for establishing a seismic elastic parameter inversion model based on multi-view learning provided in an embodiment of the present invention, the samples in the first sample set used for training the model include well logging P-wave velocity data, well logging S-wave velocity data and well logging density data, that is, the trained inversion model is used for the inversion of P-wave velocity, S-wave velocity and elastic parameters of well logging. Multi-task learning and multi-task inversion can reflect the independence and correlation between inversion tasks, thereby improving the accuracy of elastic parameter inversion.
[0021] (3) The method for establishing a seismic elastic parameter inversion model based on multi-view learning provided in an embodiment of the present invention includes learning of time information in the multi-view learning process, which can reduce the multi-solution of the inversion and improve the accuracy and spatial continuity of the inversion results.
[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.
[0023] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0025] Figure 1 This is a flow chart of a method for establishing a seismic elastic parameter inversion model in Embodiment 1 of the present invention;
[0026] Figure 2 This is a network structure diagram in Embodiment 2 of the present invention;
[0027] Figure 3 This is the inversion network diagram in the second embodiment of the present invention;
[0028] Figure 4 The cross-well profiles of the seismic signals stacked at different angles and the relative geological ages in the second embodiment of the present invention, (a) is a small-angle stacked seismic signal diagram, (b) is a medium-angle stacked seismic signal diagram, (c) is a large-angle stacked seismic signal diagram, and (d) is a relative geological age diagram;
[0029] Figure 5 The cross-sectional diagram of the low-frequency model through the well in the second embodiment of the present invention, (a) is a low-frequency longitudinal wave velocity diagram, (b) is a low-frequency shear wave velocity diagram, and (c) is a low-frequency density diagram;
[0030] Figure 6 1 is a cross-section diagram of the P-wave velocity through the well inverted in the second embodiment of the present invention, (a) is the inversion result of the unsupervised method, (b) is the inversion result of the semi-supervised method, (c) is the inversion result under the prior low-frequency constraint, and (d) is the inversion result of the method of the present invention;
[0031] Figure 7 1 is a cross-section diagram of the shear wave velocity through the well inverted in the second embodiment of the present invention, (a) is the inversion result of the unsupervised method, (b) is the inversion result of the semi-supervised method, (c) is the inversion result under the prior low-frequency constraint, and (d) is the inversion result of the method of the present invention;
[0032] Figure 8 1 is a cross-well profile of the density inverted in the second embodiment of the present invention, (a) is the inversion result of the unsupervised method, (b) is the inversion result of the semi-supervised method, (c) is the inversion result under the prior low-frequency constraint, and (d) is the inversion result of the method of the present invention;
[0033] Fig. 9 The results of the well bypass of the blind well with P-wave velocity inverted in the second embodiment of the present invention are shown in Figure 1. (a) is the inversion result of the unsupervised method, (b) is the inversion result of the semi-supervised method, (c) is the inversion result under the prior low-frequency constraint, and (d) is the inversion result of the method of the present invention.
[0034] Fig.10 The results of the well side track of the blind well of the shear wave velocity inverted in the second embodiment of the present invention, (a) is the inversion result of the unsupervised method, (b) is the inversion result of the semi-supervised method, (c) is the inversion result under the prior low-frequency constraint, and (d) is the inversion result of the method of the present invention;
[0035] Fig.11 The well bypass results of the density blind well inverted in the second embodiment of the present invention, (a) is the inversion result of the unsupervised method, (b) is the inversion result of the semi-supervised method, (c) is the inversion result under the prior low-frequency constraint, and (d) is the inversion result of the method of the present invention;
[0036] Fig.12 It is a structural schematic diagram of a device for establishing a seismic elastic parameter inversion model in an embodiment of the present invention. DETAILED DESCRIPTION
[0037] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0038] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which the present invention belongs. Although the present invention only describes preferred methods and materials, any methods and materials similar or equivalent to those described herein may also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe methods and / or materials related to the documents. In the event of conflict with any incorporated document, the content of this specification shall prevail.
[0039] In order to improve the problem of low accuracy of seismic elastic parameter inversion in the prior art, an embodiment of the present invention provides a method and device for establishing a seismic elastic parameter inversion model based on multi-view learning, which uses time information and seismic information superimposed at multiple different angles as multi-views, and improves the inversion accuracy of seismic elastic parameters based on multi-view learning.
[0040] Embodiment 1
[0041] Embodiment 1 of the present invention provides a method for establishing a seismic elastic parameter inversion model based on multi-view learning, and the process thereof is as follows Figure 1 As shown, the following steps are included:
[0042] Step S11: using the first sample set to train the forward model until the number of iterations reaches a first set number of iterations.
[0043] The samples in the first sample set include single-channel time information, multiple seismic information superimposed at different angles, and well logging elastic parameter data. Further, the time information may be relative geological age information; the multiple seismic information superimposed at different angles may include seismic information superimposed at a set small angle range, seismic information superimposed at a set medium angle range, and seismic information superimposed at a set large angle range; the well logging elastic parameter data may include well logging longitudinal wave velocity data, well logging shear wave velocity data, and well logging density data.
[0044] The forward model is iteratively trained using the first sample set. In each iteration:
[0045] Through the current forward model, the seismic information corresponding to each sample in the first sample set is forward modeled, and the forward loss value between the seismic information of each sample forward modeled and the seismic information in the sample is determined. The first forward loss value is obtained by summing them up, and the current forward model is updated according to the current first forward loss value.
[0046] The network structure of the forward model includes the Aki-Richards approximate equation and a convolution layer (which can be set to three convolution layers), so the forward model is updated in each iteration, that is, the network parameters of the convolution layer in the forward model are updated.
[0047] Step S12: using the first sample set to train the inversion model until the number of iterations reaches a second set number of iterations.
[0048] The inversion model includes a multi-view learning module, a high-dimensional feature fusion module and a multi-task learning module. The multi-view learning module is used to extract high-dimensional features of data from different views, and then the high-dimensional feature fusion module is used to fuse highly complete feature data. Then, the multi-task learning module is used to invert each elastic parameter (P-wave velocity, S-wave velocity and density).
[0049] The inversion model is iteratively trained using the first sample set. In each iteration:
[0050] Through the current inversion model, the elastic parameter data corresponding to each sample in the first sample set is inverted, the inversion loss value between the elastic parameter data inverted by each sample and the logging elastic parameter data is determined, the first inversion loss value is obtained by summing, and the current inversion model is updated according to the current first inversion loss value.
[0051] Step S13: using the first sample set and the second sample set without logging labels, taking the current forward model as a constraint, iteratively training the multi-task learning module in the inversion model until the number of iterations reaches a third set number of iterations, and selecting the optimal inversion model according to the forward loss value in the iterative process.
[0052] The specific values of the first set number of iterations, the second set number of iterations and the third set number of iterations are flexibly set based on experience, and the three may be the same or different.
[0053] Iteratively train the multi-task learning module in the inversion model. In each iteration:
[0054] Through the current inversion model, the elastic parameter data corresponding to each sample in the first sample set is inverted to determine the inversion loss value between the elastic parameter data inverted by each sample and the logging elastic parameter data, and the second inversion loss value is obtained by summing up; through the current inversion model, the elastic parameter data corresponding to each sample in the second sample set is inverted to obtain the data, added to the corresponding sample, or the corresponding data in the corresponding sample is updated; through the forward model, the seismic information corresponding to each sample in the current second sample set is forward modeled to determine the forward loss value between the seismic information forward modeled by each sample and the seismic information in the sample, and the second forward loss value is obtained by summing up, and the multi-task learning module in the current inversion model is updated according to the weighted sum of the current second inversion loss value and the second forward loss value.
[0055] During the iteration process, the weighted sum of the second inversion loss value and the second forward loss value obtained each time gradually decreases, but the inversion model is not necessarily gradually optimized. The inversion model with the smallest second forward loss value during the iteration process is determined as the optimal inversion model and used for the prediction of elastic parameter data, such as the prediction of longitudinal wave velocity, shear wave velocity and density. On this basis, elastic impedance data can also be obtained.
[0056] The method for establishing a seismic elastic parameter inversion model provided in Example 1 of the present invention uses time information and seismic information superimposed at multiple different angles as multi-views, each view is an observation result of the same object from a different perspective, and there is both correlation and information complementarity between these observation results. From the perspective of multi-view learning, the high-dimensional features of different view data are first extracted using a multi-view learning module, and then fused through a high-dimensional feature fusion module to obtain highly complete feature data, which can better serve the inversion task and enable the established model to have a higher seismic elastic parameter inversion accuracy.
[0057] The samples in the first sample set used by the training model include well logging P-wave velocity data, well logging S-wave velocity data and well logging density data, that is, the trained inversion model is used for the inversion of P-wave velocity, S-wave velocity and elastic parameters of well logging. Multi-task learning and multi-task inversion can reflect the independence and correlation between inversion tasks, thereby improving the accuracy of elastic parameter inversion.
[0058] The view learning process includes the learning of time information, which can reduce the multi-solution of inversion and improve the accuracy and spatial continuity of the inversion results.
[0059] Embodiment 2
[0060] Embodiment 2 of the present invention provides a specific application of a method for establishing a seismic elastic parameter inversion model.
[0061] The samples in the first sample set with well logging data labels contain relative geological age information R i , Earthquake information: Set the earthquake information superimposed in a small angle range Earthquake information superimposed within the set angle range And set the earthquake information of large angle range superposition Well logging elastic parameter data: Well logging P-wave velocity data Well logging shear wave velocity data and well logging density data
[0062] The samples in the second sample set without well logging labels contain relative geological age information R i* , Earthquake information: Set the earthquake information superimposed in a small angle range Earthquake information superimposed within the set angle range And set the earthquake information of large angle range superposition
[0063] The inversion model of this embodiment adopts a one-dimensional multi-view multi-task network architecture, and regards the seismic signals and relative geological year data superimposed at three different angles as data of four different views. The high-dimensional features of the data of different views are extracted using the multi-view network respectively, and then the highly complete feature data are obtained by fusion. Then, the P-wave velocity, S-wave velocity and density are inverted respectively using the multi-task learning method. In the process of training the inversion model, the Aki-Richards approximate equation is combined with three convolutional layers to reconstruct the seismic signal as an inversion loss term, and the loss term is used as the boundary condition of the optimization target to reduce the multi-solution of seismic elastic inversion and improve the blind well consistency.
[0064] See also Figure 2 and Figure 3 As shown, Figure 2 This is the overall network structure diagram including forward modeling and inversion. Figure 3 This is the network structure diagram of the inversion model.
[0065] The network training process of the present invention includes the following stages:
[0066] Phase 1: Use seismic data with well logging labels to train the forward modeling network and construct the forward modeling loss term. The network structure consists of the Aki-Richards approximate equation and three convolutional layers. Based on the Aki-Richards approximate equation, and assuming The normalized elastic impedance expression obtained according to the logging elastic parameter data is:
[0067]
[0068] In formula (1), EI it (θ) represents the normalized elastic impedance at the t-th position of the i-th sample in the first sample set, i represents the sample number, t represents the position number, and θ represents the incident angle; and They represent the mean values of the i-th sample well logging P-wave velocity data, well logging S-wave velocity data and well logging density data respectively; and They represent the logging P-wave velocity, S-wave velocity and logging density at the t-th position of the ith sample respectively.
[0069] The reflection coefficient data is obtained based on the elastic impedance data:
[0070]
[0071] In formula (2), r it (θ) is the reflection coefficient at the tth position of the i-th sample in the first sample set.
[0072] Therefore, the functional relationship between the reflection coefficient and the three elastic parameters can be obtained from (1) and (2):
[0073]
[0074] Since the wavelet in actual data is time-space variable, it is difficult to obtain seismic signals by convolving a constant wavelet with the reflection coefficient. Therefore, the convolution layer in the current forward model is convolved with the reflection coefficient data to obtain seismic information. The forward loss function is:
[0075]
[0076] In formula (4), is the forward loss, N is the number of samples in the first sample set, and W F are the network parameters of the forward model, specifically the network parameters of the three convolutional layers.
[0077] Phase 2: Use well-logging labeled seismic data and RGT data (relative geological age data) to train the inversion network. The network consists of three parts: multi-view learning, high-dimensional feature fusion, and multi-task learning. First, the seismic signals and RGT data from three different angles are input into the four-view network, and then five layers of downsampling are performed to extract high-dimensional features. Then, the high-dimensional features of the seismic signals and RGT are fused into a high-dimensional feature map. The next multi-task part is three independent sub-networks that share the same high-dimensional feature map and output high-frequency longitudinal wave velocity. High frequency shear wave velocity and high frequency density The loss function is:
[0078]
[0079] In formula (5), is the inversion loss, W B are the network parameters of the inversion model.
[0080] Phase 3: Using both well-logging labeled and unlabeled seismic data ( and RGT data R i* ), use the forward network in stage 1 as a constraint, freeze the multi-view fusion part in the inversion network, train the multi-task part in the inversion, and update W B ,At the same time, the inversion model is optimized according to the forward modeling loss.
[0081]
[0082]
[0083] Among them, L total (W B ) is the total loss, is the forward modeling loss of the unlabeled sample set after inversion and then forward modeling.
[0084] The algorithm of the network iteration process is as follows:
[0085] Input: Earthquake dataset Relative Geological Age Dataset Well logging high-frequency compressional wave velocity, well logging high-frequency shear wave velocity and well logging high-frequency density data set Well logging low-frequency longitudinal wave velocity, well logging low-frequency shear wave velocity and well logging low-frequency density data set Initial weight parameter W F ,W B , batch size N batch , initial learning rate α, number of iterations in the first stage N1, number of iterations in the second stage N2, number of iterations in the third stage N3, total number of iterations N iter, network input and output size N, parameter λ.
[0086] Output: Optimized parameter W R ,W F ,W vs ,W ρ ,W B .
[0087] Phase 1:
[0088] 1:n1←N1
[0089] 2: while n1≠0 do
[0090] 3: Random sampling from labeled data
[0091] 4: Update the network parameters W according to formula (4) F
[0092] 5:n1←n1-1
[0093] 6: end while
[0094] Phase 2:
[0095] 7:n2←N2
[0096] 8: while n2≠0 do
[0097] 9: Random sampling from labeled data
[0098] 10: Update the network parameters W according to formula (5) B
[0099] 11:n2←n2-1
[0100] 12: end while
[0101] Phase 3:
[0102] 13:n3←N3,error_s_best=100
[0103] 14:while n3≠0do
[0104] 15: Random sampling from labeled data and unlabeled data
[0105] 16: Update the network weight W according to formula (6) B
[0106] 17:n3←n3-1
[0107] 18:if error_s <error_s_best
[0108] error_s_best=error_s
[0109] Save the optimal W B
[0110] 19:end if
[0111] 20: end while
[0112] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a seismic elastic parameter inversion method, comprising:
[0113] The time information of the target stratum in the study area and multiple seismic information superimposed at different angles are input into the seismic elastic parameter inversion model, and the elastic parameter data are obtained according to the output results of the model. The seismic elastic parameter inversion model is established according to the above method.
[0114] Combine the following Figures 4 to 11 Table 1 shows the inversion effect of the embodiment of the present invention.
[0115] Figure 4 It is a cross-sectional diagram of seismic signals superimposed at three different angles, among which (a) is a small-angle superimposed seismic signal diagram, (b) is a medium-angle superimposed seismic signal diagram, (c) is a large-angle superimposed seismic signal diagram, and (d) is a relative geological age diagram. There are 1215 channels in total, the sampling time is from 2000ms to 3298ms, the sampling interval is 2ms, and the logging positions of the four channels of longitudinal wave velocity, shear wave velocity and density are the 102nd channel, the 326th channel, the 800th channel, and the 1115th channel, respectively. The 800th channel logging is selected as a blind well, and the remaining three wells are training wells. The initial learning rate is 0.0001, and the value of λ is 0.03. All data training from the first to the third stage are 1000 times, 1000 times, and 1000 times, respectively. The Adam optimizer is used for optimization. This embodiment calculates R 2 The confirmation coefficient is used to quantitatively evaluate the inversion effect of the network in actual data. Figure 5 Figure 1 is a cross-section diagram of the low-frequency model through the well, (a) is a low-frequency longitudinal wave velocity diagram, (b) is a low-frequency shear wave velocity diagram, and (c) is a low-frequency density diagram. Figures 6 to 8 The P-wave velocity profile, S-wave velocity profile and density profile inverted by different methods are shown respectively. It can be seen that the results of the three elastic parameters inverted by the embodiment of the present invention are consistent with the continuity of the seismic signal, while the inversion results of the unsupervised method, the semi-supervised method and the low-frequency constraint method all have the problem of fracture and discontinuity along the seismic signal event axis direction. Figures 9 to 11The blind well bypass results of the P-wave velocity, S-wave velocity and density inverted by different methods are shown respectively. It can be clearly seen from the blind well evaluation index in Table 1 that the P-wave velocity, S-wave velocity and density inverted by the embodiment of the present invention have a high R 2 The confirmation coefficient is the highest, that is, the degree of fit with the blind well is the highest.
[0116] Table 1 Blind well R of different methods 2 Coefficient comparison table
[0117]
[0118] Based on the inventive concept of the present invention, the embodiment of the present invention also provides a device for establishing a seismic elastic parameter inversion model based on multi-view learning, the structure of the device is as follows: Fig.12 As shown, including:
[0119] A first iterative training module 121 is used to train the forward model using a first sample set until the number of iterations reaches a first set number of iterations, wherein the samples in the first sample set include single-channel time information, multiple seismic information superimposed at different angles, and well logging elastic parameter data;
[0120] A second iterative training module 122, used for training an inversion model using the first sample set until the number of iterations reaches a second set number of iterations, the inversion model including a multi-view learning module, a high-dimensional feature fusion module and a multi-task learning module;
[0121] The third iterative training module 123 is used to use the first sample set and the second sample set without logging labels, take the current forward model as a constraint, iteratively train the multi-task learning module in the inversion model, until the number of iterations reaches a third set number of iterations, and select the optimal inversion model according to the forward loss value during the iteration process.
[0122] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0123] Based on the inventive concept of the present invention, an embodiment of the present invention also provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the above-mentioned method for establishing a seismic elastic parameter inversion model based on multi-view learning is implemented, or the above-mentioned seismic elastic parameter inversion method is implemented.
[0124] Based on the inventive concept of the present invention, an embodiment of the present invention also provides a server, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the above-mentioned method for establishing a seismic elastic parameter inversion model based on multi-view learning, or implements the above-mentioned seismic elastic parameter inversion method.
[0125] Unless otherwise specifically stated, terms such as processing, computing, calculating, determining, displaying, etc. may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, which operate and convert data represented as physical (e.g., electronic) quantities within registers or memories of a processing system into other data similarly represented as physical quantities within memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals may be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.
[0126] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.
[0127] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that an embodiment of the claimed subject matter requires more features than those set forth in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.
[0128] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein can all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. A skilled person can implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.
[0129] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.
[0130] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.
[0131] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments can be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to represent "non-exclusive or". The terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance.
Claims
1. A method for establishing a seismic elastic parameter inversion model based on multi-view learning, characterized in that: include: Using a first sample set to train a forward model until the number of iterations reaches a first set number of iterations, the samples in the first sample set comprising single-channel time information, multiple seismic information superimposed at different angles, and well logging elastic parameter data; Using the first sample set to train an inversion model until the number of iterations reaches a second set number of iterations, the inversion model comprising a multi-view learning module, a high-dimensional feature fusion module and a multi-task learning module; Using the first sample set and the second sample set without logging labels, with the current forward model as a constraint, the multi-task learning module in the inversion model is iteratively trained until the number of iterations reaches a third set number of iterations, and the optimal inversion model is selected according to the forward loss value in the iterative process.
2. The method according to claim 1, characterized in that The method of training the forward model using the first sample set comprises: The forward model is iteratively trained using the first sample set. In each iteration: Through the current forward model, the seismic information corresponding to each sample in the first sample set is forward modeled, and the forward loss value between the seismic information of each sample forward modeled and the seismic information in the sample is determined. The first forward loss value is obtained by summing them up, and the current forward model is updated according to the current first forward loss value.
3. The method according to claim 2, characterized in that The network structure of the forward model includes the Aki-Richards approximate equation and convolutional layers.
4. The method according to claim 3, characterized in that The forward modeling method obtains seismic information corresponding to each sample in the first sample set by forward modeling through the current forward model, including: For each sample in the first sample set, normalized elastic impedance data is obtained according to the logging elastic parameter data, reflection coefficient data is obtained according to the elastic impedance data, and the convolution layer in the current forward model is convolved with the reflection coefficient data to obtain seismic information.
5. The method according to claim 4, characterized in that The well logging elastic parameter data includes well logging longitudinal wave velocity data, well logging shear wave velocity data and well logging density data. The normalized elastic impedance data obtained according to the well logging elastic parameter data includes: The normalized elastic impedance data is obtained according to the logging elastic parameter data through formula (1): In formula (1), EI it (θ) represents the normalized elastic impedance at the t-th position of the i-th sample in the first sample set, i represents the sample number, t represents the position number, and θ represents the incident angle; and They represent the mean values of the i-th sample well logging P-wave velocity data, well logging S-wave velocity data and well logging density data respectively; and They represent the well logging P-wave velocity, well logging S-wave velocity and well logging density at the t-th position of the i-th sample respectively; 6. The method according to claim 1, characterized in that The using the first sample set to train the inversion model comprises: The inversion model is iteratively trained using the first sample set. In each iteration: Through the current inversion model, the elastic parameter data corresponding to each sample in the first sample set is inverted, the inversion loss value between the elastic parameter data inverted by each sample and the logging elastic parameter data is determined, the first inversion loss value is obtained by summing, and the current inversion model is updated according to the current first inversion loss value.
7. The method according to claim 1, characterized in that The multi-task learning module in the iterative training inversion model includes: Iteratively train the multi-task learning module in the inversion model. In each iteration: Using the current inversion model, inverting to obtain elastic parameter data corresponding to each sample in the first sample set, determining an inversion loss value between the inverted elastic parameter data of each sample and the logging elastic parameter data, and obtaining a second inversion loss value by summing them up; Using the current inversion model, inverting elastic parameter data corresponding to each sample in the second sample set, adding the elastic parameter data to the corresponding sample, or updating the corresponding data in the corresponding sample; Through the forward modeling, the seismic information corresponding to each sample in the current second sample set is forward modeled, and the forward loss value between the seismic information of each sample forward modeled and the seismic information in the sample is determined. The second forward loss value is obtained by summing up, and the multi-task learning module in the current inversion model is updated according to the weighted sum of the current second inversion loss value and the second forward loss value.
8. The method according to claim 7, characterized in that The step of selecting the optimal inversion model according to the forward modeling loss value in the iteration process includes: The inversion model with the smallest second forward modeling loss value during the iteration process is determined as the optimal inversion model.
9. The method according to any one of claims 1 to 8, characterized in that: The samples in the first sample set include relative geological age information, seismic information with a set small angle range superposition, seismic information with a set medium angle range superposition, seismic information with a set large angle range superposition, well logging longitudinal wave velocity data, well logging shear wave velocity data and well logging density data.
10. A method for inversion of seismic elastic parameters, characterized in that: include: The time information of the target stratum in the study area and multiple seismic information superimposed at different angles are input into the seismic elastic parameter inversion model, and the elastic parameter data are obtained according to the output results of the model. The seismic elastic parameter inversion model is established according to the method described in any one of claims 1 to 9.
11. A device for establishing a seismic elastic parameter inversion model based on multi-view learning, characterized in that: include: A first iterative training module, used for training the forward model using a first sample set until the number of iterations reaches a first set number of iterations, wherein the samples in the first sample set include single-channel time information, multiple seismic information superimposed at different angles, and well logging elastic parameter data; A second iterative training module, used for training an inversion model using the first sample set until the number of iterations reaches a second set number of iterations, the inversion model comprising a multi-view learning module, a high-dimensional feature fusion module and a multi-task learning module; The third iterative training module is used to use the first sample set and the second sample set without logging labels, take the current forward model as a constraint, iteratively train the multi-task learning module in the inversion model, until the number of iterations reaches a third set number of iterations, and select the optimal inversion model according to the forward loss value during the iteration process.
12. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, which, when executed by a processor, implement the method for establishing a seismic elastic parameter inversion model based on multi-view learning as described in any one of claims 1 to 9, or implement the seismic elastic parameter inversion method as described in claim 10.
