Stratigraphic dip angle determination model training method and device and stratigraphic dip angle determination method and device

Abstract

The invention discloses a stratigraphic dip angle determination model training method and device and a stratigraphic dip angle determination method and device. The method comprises the following steps: obtaining label data from a pre-established sample label library to obtain a training set and a verification set; respectively inputting synthetic seismic data in label data included in the training set and the verification set into the selected deep learning network model for learning to obtain a training set loss function and a verification set loss function; and after the consistency of the training set loss function and the verification set loss function meets the requirement through multiple iterations, a trained stratigraphic dip angle determination model is obtained. And testing the stratigraphic dip angle determination model by using the test set, and after the test result meets the requirement, performing actual data application by using the stratigraphic dip angle determination model to determine the stratigraphic dip angle. According to the method, intelligent calculation of the dip angle can be realized based on the deep learning network model, the stratigraphic dip angle is determined more truly and accurately, the underground structure is reflected better, and a data foundation is laid for development of a fine exploration technology.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic exploration, and in particular to a formation dip determination model training and a formation dip determination method and device. Background Art

[0002] From a macroscopic point of view, most of the strata are inclined. The stratum dip information can reflect the stratum inclination direction and inclination angle. It is one of the important characteristic parameters of underground structures and is widely used in seismic exploration and geological interpretation. For example, the establishment of observation systems, velocity analysis and modeling, geological structure and sedimentation research, etc. all require stratum dip information. At the same time, stratum dip is also an indispensable parameter for calculating fracture properties, such as coherence and curvature. Therefore, stratum dip information is very important for seismic exploration technology. Traditional dip calculation methods include plane wave decomposition and multi-time window dip scanning. Summary of the invention

[0003] The inventors of the present application have discovered that the use of traditional inclination calculation methods to measure formation inclination, because it is achieved by applying a simplified mathematical model of the problem in practice, will result in large errors, resulting in low accuracy of the measured formation inclination data, the presence of some irrelevant interference, and the formation inclination measurement results may deviate from the true inclination characteristics of the formation. The above-mentioned problems such as insufficient measurement accuracy have restricted the development of seismic precision exploration technology, and formation inclination information is very important for seismic exploration technology. Therefore, how to measure formation inclination more accurately has become a problem that needs to be solved urgently.

[0004] In view of the above problems, the present invention is proposed to provide a method and device for intelligent calculation of formation dip angles that overcome the above problems or at least partially solve the above problems.

[0005] In a first aspect, an embodiment of the present invention provides a formation dip determination model training method, comprising:

[0006] The label data are obtained from a pre-established sample label library to obtain a training set and a validation set; the label data in the sample label library includes formation dip and corresponding synthetic seismic data; the synthetic seismic data is obtained through a pre-established formation distortion structure velocity model based on a given formation dip;

[0007] The synthetic seismic data in the labeled data included in the training set and the validation set are respectively input into the selected deep learning network model for learning, and the training set loss function and the validation set loss function are obtained according to the output formation dip prediction data and the formation dip in the labeled data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, the trained formation dip determination model is obtained.

[0008] In some optional embodiments, the above method further includes:

[0009] Pre-establish the twisted structural velocity model of the formation;

[0010] Based on a given formation dip, synthetic seismic data is obtained through a twisted structural velocity model of the formation, and a sample label library is constructed according to the formation dip and the corresponding synthetic seismic data.

[0011] In some optional embodiments, establishing a distortion structure velocity model of a formation includes:

[0012] Randomly generate horizontal velocity models of formations;

[0013] The horizontal velocity model is vertically displaced to generate a displacement field and obtain a twisted structural velocity model.

[0014] In some optional embodiments, randomly generating a horizontal velocity model includes:

[0015] A horizontal velocity model is randomly generated according to the velocity layer number, thickness of each layer and wave velocity of each layer of the stratum. The horizontal velocity model includes model depth, velocity layer number, thickness of each layer and wave velocity of each layer.

[0016] In some optional embodiments, performing vertical displacement on the horizontal velocity model to generate a displacement field and obtain a twisted structure velocity model includes:

[0017] Based on the formula The horizontal velocity model V(X,Z) is vertically displaced to generate a displacement field M(Z m );

[0018] Among them, X is the horizontal coordinate of the horizontal velocity model, and Z is the depth coordinate of the horizontal velocity model; X′ is the horizontal coordinate of the twisted structure velocity model, and Z′ is the depth coordinate of the twisted structure velocity model. m (X) is the vertical displacement;

[0019] Based on the displacement field M(Z m ), and the twisted structure velocity model is obtained: V(X′,Z′)=V(X,Z)+M(Z m ).

[0020] In some optional embodiments, synthetic seismic data is obtained through a twisted structural velocity model of the formation based on a given formation dip, and a sample label library is constructed according to the formation dip and the corresponding synthetic seismic data, including:

[0021] According to the twisted structure velocity model, the vertical displacement Z m (X), determine the formation dip: Where S(X′,Z′) represents the formation dip;

[0022] According to the pre-established tectonic velocity model V(X′, Z′) of the stratum, synthetic seismic data is obtained by the following formula: D(X′, Z′) = ρV(X′, Z′)*w, where ρ is a constant density, w is a main frequency Ricker wavelet, and D(X′, Z′) is synthetic seismic data;

[0023] The formation dip and the corresponding synthetic seismic data are used as label data of the label library.

[0024] In some optional embodiments, the synthetic seismic data in the label data included in the training set and the validation set are respectively input into the selected deep learning network model for learning, and the training set loss function and the validation set loss function are obtained according to the output formation dip prediction data and the formation dip in the label data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, a trained formation dip determination model is obtained, including:

[0025] Divide the labeled data of the training set into multiple sub-training sets and preset the number of training times for the sub-training sets;

[0026] Inputting synthetic seismic data in the label data included in a set of sub-training sets into the selected deep learning network model for learning, and after executing a preset number of training times, obtaining a sub-training set loss function based on the output formation dip prediction data and the formation dip in the label data;

[0027] The synthetic seismic records in the label data included in the validation set are input into the deep learning network model for learning, and the validation set loss function is obtained according to the output formation dip prediction data and the formation dip in the label data;

[0028] The sub-training set loss function is compared with the validation set loss function. If the difference between the two does not meet the preset consistency requirements, the next sub-training set is obtained and the process is returned to continue the step of inputting the synthetic seismic data in the label data included in a sub-training set into the selected deep learning network model for training. After multiple iterations, until the difference between the sub-training set loss function and the validation set loss function meets the preset consistency requirements, a trained formation inclination determination model is obtained.

[0029] In some optional embodiments, a deep learning network model is constructed, including

[0030] A deep learning network model is constructed based on the relationship expression between the formation dip and the synthetic seismic data: S(X′, Z′)=f(D(X′, Z′)); the deep learning network model is a Unet model, including an input layer, a convolution layer, a downsampling layer, an upsampling layer, and an output layer;

[0031] The loss function is the mean square error:

[0032] Among them, m is the number of stratum sample points, y i is the formation dip in the label data of the i-th sample point, x i is the synthetic seismic data of the ith sample point, f(x i ) is the formation dip prediction data of the i-th sample point output by the deep learning network model.

[0033] In a second aspect, an embodiment of the present invention provides a method for determining a formation dip, comprising:

[0034] Acquire three-dimensional post-stack seismic data of the formation to be predicted;

[0035] The three-dimensional post-stack seismic data is input into a trained formation dip angle determination model, and the formation dip angle of the formation to be predicted is output; the formation dip angle determination model is trained by a formation dip angle determination model training method.

[0036] In a third aspect, an embodiment of the present invention provides a formation dip determination model training device, comprising:

[0037] A data acquisition module is used to obtain label data from a pre-established sample label library to obtain a training set and a validation set;

[0038] Model building module, used to build a distortion structure velocity model of the formation and build a deep learning network model;

[0039] The model training module is used to input the synthetic seismic data in the label data included in the training set and the validation set into the selected deep learning network model for learning, and obtain the training set loss function and the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, the trained formation dip determination model is obtained.

[0040] In a fourth aspect, an embodiment of the present invention provides a device for determining a formation dip, comprising:

[0041] A data acquisition module, used for acquiring three-dimensional post-stack seismic data of the formation to be predicted;

[0042] The formation dip determination module is used to input the three-dimensional post-stack seismic data into a trained formation dip determination model and output the formation dip of the formation to be predicted; the formation dip determination model is trained by a formation dip determination model training method.

[0043] A computer storage medium stores computer executable instructions, which, when executed by a processor, can train a formation dip determination model and / or implement a formation dip determination method.

[0044] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a formation dip determination model training method and / or a formation dip determination method are implemented.

[0045] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0046] The present invention proposes a formation dip determination model training and formation dip determination method, obtains label data from a pre-established sample label library to obtain a training set and a validation set; inputs the synthetic seismic data in the label data included in the training set and the validation set into the selected deep learning network model for learning, and obtains the training set loss function and the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; after multiple iterations, the consistency of the training set loss function and the test set loss function meets the requirements, and a trained formation dip determination model is obtained. Three-dimensional post-stack seismic data of the formation to be predicted is obtained; the three-dimensional post-stack seismic data is input into the trained formation dip determination model, and the formation dip of the formation to be predicted is output, and the formation dip is intelligently determined based on the deep learning network model. The method overcomes the problems of irrelevant interference and insufficient measurement accuracy generated by conventional dip calculation methods, can realize intelligent dip calculation based on the deep learning network model, more realistically and accurately determine the formation dip, better reflect the underground structure, and lay a data foundation for the development of fine exploration technology.

[0047] 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.

[0048] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] 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:

[0050] Figure 1 This is a flow chart of a formation dip determination model training method in Embodiment 1 of the present invention;

[0051] Figure 2 This is a velocity model diagram of a local twisted structure in the first embodiment of the present invention;

[0052] Figure 3This is a schematic diagram of the formation inclination in the first embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of seismic data in Embodiment 1 of the present invention;

[0054] Figure 5 This is an example diagram of synthetic seismic data in the test set label data in the first embodiment of the present invention;

[0055] Figure 6 It is a formation dip map corresponding to the synthetic seismic data in the test set label data in the first embodiment of the present invention;

[0056] Figure 7 A formation dip angle map output by the formation dip angle determination model trained in the first embodiment of the present invention;

[0057] Figure 8 This is a schematic diagram of the structure of a formation dip determination model training device in Embodiment 1 of the present invention;

[0058] Fig. 9 This is a flow chart of a method for determining formation inclination in Embodiment 2 of the present invention;

[0059] Fig.10 This is a three-dimensional post-stack seismic data diagram in the second embodiment of the present invention;

[0060] Fig.11 It is a seismic profile diagram of sections A and B in the second embodiment of the present invention;

[0061] Fig.12 A formation dip angle map output by the formation dip angle determination model in the second embodiment of the present invention;

[0062] Fig.13 A schematic diagram of the formation dip angle obtained by plane wave decomposition in the second embodiment of the present invention;

[0063] Fig.14 A schematic diagram of the formation dip angle obtained by scanning multiple time windows in the second embodiment of the present invention;

[0064] Fig.15 This is a schematic diagram of the structure of the device for determining the formation inclination in the second embodiment of the present invention. DETAILED DESCRIPTION

[0065] 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.

[0066] In order to solve the problems of irrelevant interference and insufficient measurement accuracy in traditional formation dip calculation methods, the embodiments of the present invention provide a formation dip determination model training and formation dip determination method, which intelligently determines the formation dip based on a deep learning network model, regards dip calculation as a regression problem, and establishes a synthetic seismic data and formation dip data label library, and uses data-driven fitting of the nonlinear function relationship between seismic data and dip, thereby realizing intelligent dip calculation. This method can be applied in the field of seismic exploration, and is a device for identifying the dip attributes of seismic data volumes using a deep learning method for post-stack seismic data.

[0067] Embodiment 1

[0068] Embodiment 1 of the present invention provides a formation dip determination model training method, the process of which is as follows: Figure 1 As shown, the following steps are included:

[0069] Step S101: Obtain label data from a pre-established sample label library to obtain a training set and a validation set.

[0070] Step S102: Input the synthetic seismic data in the training set label data into the selected deep learning network model for learning, and obtain the training set loss function based on the output formation dip prediction data and the formation dip in the label data.

[0071] Step S103: Input the synthetic seismic data in the validation set label data into the selected deep learning network model for learning, and obtain the validation set loss function based on the output formation dip prediction data and the formation dip in the label data.

[0072] Step S104: Compare the training set loss function with the validation set loss function to determine whether the difference between the two meets the preset consistency requirement. If so, execute S105; if not, execute S102.

[0073] Step S105: Obtain a trained formation dip angle determination model.

[0074] The above steps S102-S105 implement the input of the synthetic seismic data in the label data included in the training set and the validation set into the selected deep learning network model for learning, and obtain the training set loss function and the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, a trained formation dip determination model is obtained.

[0075] Optionally, in the above step S101, the label data in the sample label library includes a formation dip and corresponding synthetic seismic data; the synthetic seismic data is obtained through a pre-established twisted structural velocity model of the formation based on a given formation dip;

[0076] Optionally, a distortion tectonic velocity model of the formation is pre-established, synthetic seismic data is obtained through the distortion tectonic velocity model of the formation based on a given formation dip, and a sample label library is constructed according to the formation dip and the corresponding synthetic seismic data.

[0077] Optionally, a twisted structural velocity model of the stratum is pre-established, specifically including: randomly generating a horizontal velocity model of the stratum; vertically displacing the horizontal velocity model to generate a displacement field, and obtaining the twisted structural velocity model.

[0078] Among them, randomly generating a horizontal velocity model of the stratum specifically includes: randomly generating a horizontal velocity model according to the number of velocity layers, thickness of each layer, and wave velocity of each layer of the stratum, and the horizontal velocity model includes model depth, number of velocity layers, thickness of each layer, and wave velocity of each layer. The number of layers of the velocity model is random, and the thickness of each layer is random. This random mechanism can effectively remove the influence of layer thickness and number of layers on deep learning training. For example, a two-dimensional horizontal velocity model is pre-generated as V(X,Z).

[0079] Among them, the horizontal velocity model is vertically displaced to generate a displacement field and obtain a twisted structure velocity model, which specifically includes: based on the formula The horizontal velocity model V(X,Z) is vertically displaced, and the displacement at different horizontal coordinates X is different, generating a displacement field M(Z m ), forming undulations;

[0080] Among them, X is the horizontal coordinate of the horizontal velocity model, and Z is the depth coordinate of the horizontal velocity model; X′ is the horizontal coordinate of the twisted structure velocity model, and Z′ is the depth coordinate of the twisted structure velocity model. m (X) is the vertical displacement; the displacement at different horizontal coordinates X is different, resulting in a displacement field M(Z m ), forming undulations.

[0081] Based on the displacement field M(Z m ), and the twisted structure velocity model is obtained: V(X′,Z′)=V(X,Z)+M(Z m ), when constructing the twisted structure velocity model, factors that need to be considered include velocity field, structural deformation displacement field, and formation slope caused by displacement.

[0082] For the local distortion structure velocity model, see Figure 2 As shown, Figure 2 The layers in the distorted tectonic velocity model are marked with different inclination degrees and velocities, forming an undulating structure of the strata.

[0083] In this embodiment, the vertical displacement at different positions is generated by superposition of two-dimensional Gaussian curves. Because the vertical displacement is curved, it can fit the structural deformation characteristics. The vertical displacement is generated by superposition of multiple Gaussian function curves. If the structural deformation characteristics are in three-dimensional conditions, vertical displacement needs to be generated by superposition of Gaussian surfaces to simulate complex structures. Before the vertical displacement, the stratum is a plane and the velocity model is also horizontal. After the vertical displacement is generated in the horizontal stratum, the stratum rises and falls, and the velocity model also becomes an undulating velocity model.

[0084] Optionally, synthetic seismic data is obtained based on a given formation dip through a twisted structural velocity model of the formation, specifically including:

[0085] According to the twisted structure velocity model, the vertical displacement Z m (X), determine the formation dip: where S(X′, Z′) represents the formation dip, which is also the true slope of the twisted tectonic velocity model V(X′, Z′). For an example of formation dip, see Figure 3 As shown, in Figure 3 The larger the slope, the greater the formation dip.

[0086] According to the pre-established tectonic velocity model V(X′, Z′) of the stratum, synthetic seismic data is obtained by the following formula: D(X′, Z′) = ρV(X′, Z′)*w, where ρ is a constant density, w is a dominant Ricker wavelet, and D(X′, Z′) is synthetic seismic data. Synthetic seismic recording based on the tectonic velocity model of the stratum is to convolve the random dominant Ricker wavelet with the stratum model information to produce a synthetic record. For an example of seismic data, see Figure 4 When synthesizing seismic records, the main frequency Ricker wavelet can be used, or other wavelets can be used.

[0087] In the process of determining the formation dip angle and synthesizing seismic data using the above-mentioned twisted structural velocity model V(X′, Z′), the formula for determining the formation dip angle can be As well as the formula for synthetic seismic data D(X′,Z′)=ρV(X′,Z′)*w, it can be inferred that the formation dip and the synthetic seismic data have a one-to-one correspondence with the velocity model, so there is also a correspondence between the formation dip and the synthetic seismic data. The correspondence between the two is expressed as S(X′,Z′)=f(D(X′,Z′)), where f(D(X′,Z′)) represents the nonlinear functional relationship between the two.

[0088] Optionally, using the above-mentioned formulas for calculating the formation dip and calculating synthetic seismic records, multiple formation dips and corresponding synthetic seismic records can be obtained, and a sample label library is established based on the formation dips and the corresponding synthetic seismic data. The formation dip and the corresponding synthetic seismic data are used as label data of the label library. Considering the problem that the displacement of the structural model exceeds the upper and lower boundaries, this embodiment preferably establishes a label library relative to the middle part.

[0089] In the above step S102, the synthetic seismic data in the training set label data is input into the selected deep learning network model for learning, which specifically includes:

[0090] Divide the labeled data of the training set into multiple sub-training sets and preset the number of training times for the sub-training sets;

[0091] A set of synthetic seismic data in the label data included in a sub-training set is input into a pre-built deep learning network model for learning. After a preset number of training times, the sub-training set loss function is obtained based on the output formation dip prediction data and the formation dip in the label data, and the deep learning network model at this time is used as the trained deep learning network model.

[0092] The essence of deep learning network model training is to enable the network to have the ability to fit the functional relationship between seismic data and corresponding inclination. Training requires a large amount of sample data. Whether the sample data has the training target characteristics and whether it can sufficiently express the corresponding relationship between seismic data and inclination is crucial to the training results. In this embodiment, 20,000 labels are generated by constructing a sample label library, of which 12,000 label data are used as a training set. The 12,000 labels in the training set are divided into multiple groups of sub-training sets according to the grouping rules, and the number of training sub-training sets is preset. When the training starts, the synthetic seismic data in the label data included in the first group of sub-training sets is input into the pre-constructed deep learning network model for learning. After executing the preset number of training times, a sub-training set loss function loss is obtained according to the output formation inclination prediction data and the formation inclination in the label data. After the sub-training set has learned the model, the deep learning network model at this time is used as the trained deep learning network model, and then the deep learning network model is continuously adjusted according to the sub-training set loss function to make the model converge continuously. The number of training sets can be determined as needed. After obtaining the training sets, they can be grouped to obtain sub-training sets. Alternatively, a corresponding number of label data can be obtained from the label library each time to form sub-training sets according to the number of labels in the sub-training sets.

[0093] Optionally, build a deep learning network model, including:

[0094] A deep learning network model is constructed based on the relationship expression between formation dip and synthetic seismic data: S(X′, Z′) = f(D(X′, Z′)); the deep learning network model is a Unet model, including an input layer, a convolution layer, a downsampling layer, an upsampling layer, and an output layer;

[0095] Using deep learning to calculate the dip is not essentially a classification problem, but a regression problem. The original Unet model is used in semantic segmentation because it has the ability to extract features at multiple scales, which is beneficial to improving the stability of slope or dip prediction. However, the output of the original Unet model includes an activation layer, and the activation layer mostly uses activation functions such as RELU and sigmoid functions, which can only output positive values ​​or values ​​within a limited range, such as values ​​between (-1, 1). It is not suitable for dip prediction in complex structural environments with variable dips. In order to make the output results of the model more adaptable to the changes in dip in the actual exploration environment, the original Unet model is modified, and the activation layer is no longer added to the output end, which can avoid the limitation of the output value range by the activation layer, so that the output result of the output end can meet the output digital range requirements of the formation dip prediction, and can output accurate dip values ​​to ensure that the network model is a regression model.

[0096] Optionally, the loss function of choice is mean squared error:

[0097] Among them, m is the number of stratum sample points, y i is the formation dip in the label data of the i-th sample point, x i is the synthetic seismic data of the ith sample point, f(x i ) is the formation dip prediction data of the i-th sample point output by the deep learning network model.

[0098] The loss function can also choose other error forms.

[0099] Optionally, in the above step S103, the synthetic seismic data in the validation set label data is input into the selected deep learning network model for learning, and the validation set loss function is obtained according to the output formation dip prediction data and the formation dip in the label data, which specifically includes:

[0100] The synthetic seismic data in the validation set label data is input into the deep learning network model that has been learned by the sub-training set in step S102 for learning, and the validation set loss function is obtained based on the output formation dip prediction data and the formation dip in the label data.

[0101] In this embodiment, 7000 labels out of the generated 20000 labels are used as the validation set, and the number of label data in the test set can be determined as needed and selected from the label library. After a set of sub-training sets are trained, the synthetic seismic records in the label data of the validation set are input into the trained deep learning network model for learning, and the validation set loss function Val loss is obtained based on the output formation dip prediction data and the formation dip in the label data. The role of the validation set is mainly to verify the effect of the model in the training process, adjust the model hyperparameters, and prevent overfitting.

[0102] Optionally, in the above step S104, the training set loss function is compared with the validation set loss function to determine whether the difference between the two meets the preset consistency requirements. If the difference between the two does not meet the preset consistency requirements, the next set of sub-training sets is obtained and the process is returned to continue the step of inputting the synthetic seismic data in the label data included in a set of sub-training sets into the selected deep learning network model for training. After multiple iterations, until the difference between the sub-training set loss function and the validation set loss function meets the preset consistency requirements, a trained formation inclination determination model is obtained.

[0103] If the difference between the two meets the preset consistency requirements, it means that the deep learning network model has converged well, and the model at this time can be used as a trained formation dip angle determination model. Because it is a regression model, during the training process, by observing the sub-training set loss function and the validation set loss function, if the two are close to the same, it can be judged that the model has converged well, and finally a trained formation dip angle determination model is obtained;

[0104] If the difference between the two does not meet the preset consistency requirements, it means that the deep learning network model has not yet reached the convergence requirements, and it is necessary to obtain the synthetic seismic data in the label data of the next set of sub-training sets and return to continue executing the step of inputting the synthetic seismic data in the label data included in a set of sub-training sets into the selected deep learning network model for training. After multiple iterations, until the difference between the loss function of the sub-training set and the loss function of the validation set meets the preset consistency requirements, a trained formation inclination determination model is finally obtained.

[0105] Optionally, the trained formation dip determination model obtained in the above step S105 needs to be tested to test whether the trained formation dip determination model has learned the functional relationship between the formation dip and the synthetic seismic data.

[0106] In this embodiment, 1000 labels out of the generated 20000 labels can be used as a test set, and the number of label data in the test set can be determined and selected as needed. The synthetic seismic data from the test set label data that has not participated in the learning is input into the trained formation dip determination model, and the formation dip in the test set label data is compared with the formation dip output by the trained formation dip determination model to check whether the formation dips of the two are close.

[0107] In order to further test the learning results of the model, this embodiment selects three examples for comparison. The synthetic seismic data in the test set label data is shown in Figure 5 As shown, Figure 5 The three synthetic seismic data in the test set have obvious changes in inclination, and the slope also changes with the fluctuation of the event axis; the formation inclination corresponding to the synthetic seismic data in the test set label data is as follows: Figure 6 As shown, Figure 6 The three formation dip angle data diagrams in Figure 5 The three synthetic seismic data in the figure also fully demonstrate the undulating characteristics of the strata. Figure 6 The medium slope represents the inclination degree of the formation dip; the trained formation dip determines the formation dip output by the model. Figure 7 As shown, Figure 7 The three sets of formation dip data output by the model in Figure 6 The three formation dip data correspond to the observation Figure 7 and Figure 6 It can be seen that the grayscale corresponding to the dip in the two images is almost the same, which proves that the deep learning network model has learned the functional relationship between the synthetic seismic data and the formation dip. The trained deep learning network model can be used as a model for determining the formation dip.

[0108] Based on the same inventive concept, an embodiment of the present invention further provides a formation dip determination model training device, which can be set in a computer device with computing processing capabilities. The structural schematic diagram of the device is shown in FIG. Figure 8 As shown, including:

[0109] The data acquisition module 10 is used to obtain label data from a pre-established sample label library to obtain a training set and a validation set;

[0110] A model building module 11 is used to build a distortion structure velocity model of the formation and to build a deep learning network model;

[0111] The model training module 12 is used to input the synthetic seismic data in the label data included in the training set and the validation set into the selected deep learning network model for learning, and obtain the training set loss function and the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, a trained formation dip determination model is obtained.

[0112] Regarding the formation dip determination model training 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.

[0113] The above method and device of this embodiment obtains the training set and the validation set by obtaining the label data from the pre-established sample label library; the synthetic seismic data in the label data included in the training set and the validation set are respectively input into the selected deep learning network model for learning, and the training set loss function and the validation set loss function are obtained; after multiple iterations, the consistency of the training set loss function and the test set loss function meets the requirements, and the trained formation dip determination model is obtained. The test set label data is obtained from the sample label library, and the model is tested with the test set, and finally the formation dip determination model is obtained. The formation dip determination model realizes the intelligent calculation of dip, and the formation dip is determined more accurately. This method is based on the dip calculation method of deep learning, and is driven by data. It uses data label training to find a model that is more in line with actual laws, and it has the ability to fit any mathematical function.

[0114] Embodiment 2

[0115] Embodiment 2 of the present invention provides a method for determining formation dip, the process of which is as follows: Fig. 9 As shown, the following steps are included:

[0116] Step S201: Acquire three-dimensional post-stack seismic data of the formation to be predicted.

[0117] Post-stack seismic data refers to the SGY seismic data used for seismic exploration interpretation after pre-stack gather stacking and migration. It contains information such as volume header, trace header and post-stack record.

[0118] Step S202: input the three-dimensional post-stack seismic data into the trained formation dip determination model, and output the formation dip of the formation to be predicted.

[0119] The formation dip determination model can be trained using the formation dip determination model training method provided in Example 1. Input the three-dimensional post-stack seismic data into the trained formation dip determination model to perform dip prediction, and obtain the corresponding formation dip attribute body according to the prediction result.

[0120] In this embodiment, actual data application is carried out, and a block of 3D post-stack seismic data with 301×291×161 grid points is selected. The 3D post-stack seismic data is as follows: Fig.10 As shown in Figure 2, seismic profile A and slope profile B are intercepted in the two directions of the main survey line and the connecting survey line. The seismic profiles of A and B are shown in Figure 2. Fig.11 As shown, obtain Fig.11 The three-dimensional post-stack seismic data of the seismic section shown in FIG. 1 is input into the formation dip determination model, and the formation dip of the formation of the section is output. The output formation dip is as follows: Fig.12 As shown; in order to compare the output of the formation dip determination model with the traditional dip calculation method, the plane wave decomposition and multi-time window scanning methods are used to obtain the formation dip corresponding to the above-mentioned three-dimensional seismic data. The results of the plane wave decomposition to obtain the formation dip are shown in Fig.13 As shown in the figure, the result of obtaining the formation dip angle by the multi-time window scanning method is as follows Fig.14 As shown, by comparison Fig.12 , Fig.13 as well as Fig.14 It can be seen that the gap between the plane wave decomposition and multi-time window scanning methods and the deep learning network model method in actual data application is greater, and the interference along the layer and scanning patches are more obvious. The deep learning method clearly and effectively shows the characteristics of the formation dip in both model testing and actual data testing.

[0121] Comparing the above three methods with each other, the common point is that they all show the overall fluctuation characteristics of seismic data, which also verifies the correctness of the deep learning network model method. The difference is that the deep learning network model method focuses more on the stratum fluctuation than the conventional method, is not disturbed by irrelevant factors, and clearly depicts the turning point of the stratum fluctuation conversion, which means that it reflects the underground structure more realistically and accurately.

[0122] Based on the same inventive concept, an embodiment of the present invention further provides a formation dip determination device, which can be arranged in a device capable of executing computer instructions. The structure of the device is as follows: Fig.15 As shown, including:

[0123] A data acquisition module 20, for acquiring three-dimensional post-stack seismic data of the formation to be predicted;

[0124] The formation dip determination module 21 is used to input the three-dimensional post-stack seismic data into the trained formation dip determination model and output the formation dip of the formation to be predicted;

[0125] Regarding the formation inclination determination device in the above-mentioned 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.

[0126] The above-mentioned method and device of the embodiment of the present invention obtains the three-dimensional post-stack seismic data of the formation to be predicted, inputs the three-dimensional post-stack seismic data into the trained formation dip determination model, and outputs the formation dip of the formation to be predicted. By applying specific actual data to the formation dip determination model, the formation dip of the formation to be measured is output, and at the same time, the result of the traditional dip calculation is compared with the result output by the formation dip determination model, and it is concluded that the formation dip determination model trained based on the deep learning network model is more focused on the formation fluctuation, is not disturbed by irrelevant factors, and very clearly depicts the inflection point of the formation fluctuation conversion, indicating that it reflects the underground structure more realistically and accurately.

[0127] The formation dip determination model can use a deep learning network. The essence of deep learning network training is to enable the network to have the ability to fit the functional relationship between seismic data and the corresponding dip. Training requires a large amount of sample data, that is, it is necessary to establish a sample label library containing seismic data and corresponding dip data, and whether the sample data has the training target characteristics and whether it can be sufficient to express the corresponding relationship between seismic data and dip is crucial to the training results. In this application, a label library is established by synthesizing seismic data, and a geological model is first established, that is, the twisted structural velocity model in the embodiment, and the dip is used as a known given information to guide the structural morphology of the geological model, and then the seismic data is synthesized. The synthesized seismic data corresponds to the dip information, and the dip information is accurate and known. These data are used as sample data to train the deep learning network. The trained deep learning network can accurately predict the dip of the formation to be predicted based on the synthetic seismic data of the formation to be predicted.

[0128] An embodiment of the present invention further provides a computer storage medium storing computer executable instructions. When the computer executable instructions are executed by a processor, a formation dip determination model training method and / or a formation dip determination method are implemented.

[0129] An embodiment of the present invention also provides a computer device, 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, a formation inclination determination model training method and / or a formation inclination determination method are implemented.

[0130] 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.

[0131] 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.

[0132] 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 the embodiments of the claimed subject matter require more features than are clearly stated 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.

[0133] 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.

[0134] 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.

[0135] 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.

[0136] 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 may 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," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

Claims

1. A formation dip determination model training method, characterized in that: include: Obtain labeled data from a pre-established sample label library to obtain a training set and a validation set; The label data in the sample label library includes formation dip and corresponding synthetic seismic data; the synthetic seismic data is obtained through a pre-established twisted structural velocity model of the formation based on a given formation dip; The synthetic seismic data in the labeled data included in the training set and the validation set are respectively input into the selected deep learning network model for learning, and the training set loss function and the validation set loss function are obtained according to the output formation dip prediction data and the formation dip in the labeled data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, a trained formation dip determination model is obtained.

2. The method according to claim 1, characterized in that Also includes: Pre-establish the twisted structural velocity model of the formation; Based on a given formation dip, synthetic seismic data is obtained through a twisted structural velocity model of the formation, and a sample label library is constructed according to the formation dip and the corresponding synthetic seismic data.

3. The method according to claim 2, characterized in that Establish the twisted structural velocity model of the formation, including: Randomly generate horizontal velocity models of formations; The horizontal velocity model is vertically displaced to generate a displacement field and obtain a twisted structure velocity model.

4. The method according to claim 3, characterized in that The randomly generated horizontal velocity model comprises: A horizontal velocity model is randomly generated according to the velocity layer number, thickness of each layer and wave velocity of each layer of the stratum. The horizontal velocity model includes model depth, velocity layer number, thickness of each layer and wave velocity of each layer.

5. The method according to claim 3, characterized in that The horizontal velocity model is vertically displaced to generate a displacement field, and a twisted structure velocity model is obtained, including: Based on the formula The horizontal velocity model V(X,Z) is vertically displaced to generate a displacement field M(Z m ); Among them, X is the horizontal coordinate of the horizontal velocity model, and Z is the depth coordinate of the horizontal velocity model; X′ is the horizontal coordinate of the twisted structure velocity model, and Z′ is the depth coordinate of the twisted structure velocity model. m (X) is the vertical displacement; Based on the displacement field M(Z m ), and the twisted structure velocity model is obtained: V(X′,Z′)=V(X,Z)+m(Z m ).

6. The method according to claim 2, characterized in that Based on a given formation dip, synthetic seismic data is obtained through a twisted structural velocity model of the formation, and a sample label library is constructed according to the formation dip and the corresponding synthetic seismic data, including: According to the twisted structure velocity model, the vertical displacement Z m (X), determine the formation dip: Where s(X′,Z′) represents the formation dip; According to the pre-established tectonic velocity model V(X′, Z′) of the stratum, synthetic seismic data is obtained by the following formula: D(X′, Z′) = ρV(X′, Z′)*w, where ρ is a constant density, w is a main frequency Ricker wavelet, and D(X′, Z′) is synthetic seismic data; The formation dip angle and the corresponding synthetic seismic data are used as label data of a label library.

7. The method according to claim 1, characterized in that Inputting the synthetic seismic data in the label data included in the training set and the validation set into the selected deep learning network model for learning, respectively, and obtaining the training set loss function and the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; After multiple iterations, when the consistency between the training set loss function and the test set loss function meets the requirements, the trained formation dip determination model is obtained, including: Dividing the label data of the training set into a plurality of sub-training sets and presetting the number of training times for the sub-training sets; Inputting synthetic seismic data in the label data included in a set of sub-training sets into the selected deep learning network model for learning, and after executing a preset number of training times, obtaining a sub-training set loss function based on the output formation dip prediction data and the formation dip in the label data; Inputting the synthetic seismic records in the label data included in the validation set into the deep learning network model for learning, and obtaining the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; The sub-training set loss function and the validation set loss function are compared. If the difference between the two does not meet the preset consistency requirements, the next sub-training set is obtained and the process is returned to continue the step of inputting the synthetic seismic data in the label data included in a sub-training set into the selected deep learning network model for training. After multiple iterations, until the difference between the sub-training set loss function and the validation set loss function meets the preset consistency requirements, a trained formation inclination determination model is obtained.

8. The method according to claim 7, characterized in that Build a deep learning network model, including A deep learning network model is constructed based on the relationship expression between the formation dip and the synthetic seismic data: S(X′, Z′)=f(D(X′, Z′)); the deep learning network model is a Unet model, including an input layer, a convolution layer, a downsampling layer, an upsampling layer, and an output layer; The loss function is the mean square error: Among them, m is the number of stratum sample points, y i is the formation dip in the label data of the i-th sample point, x i is the synthetic seismic data of the ith sample point, f(x i ) is the formation dip prediction data of the i-th sample point output by the deep learning network model.

9. A method for determining formation dip, characterized in that: include: Acquire three-dimensional post-stack seismic data of the formation to be predicted; The three-dimensional post-stack seismic data is input into a trained formation dip determination model to output the formation dip to be predicted; the formation dip determination model is trained by the formation dip determination model training method described in any one of claims 1-8.

10. A formation dip determination model training device, comprising: A data acquisition module is used to obtain label data from a pre-established sample label library to obtain a training set and a validation set; Model building module, used to build a distortion structure velocity model of the formation and build a deep learning network model; The model training module is used to input the synthetic seismic data in the label data included in the training set and the validation set into the selected deep learning network model for learning, and obtain the training set loss function and the validation set loss function according to the output formation dip prediction data and the formation dip in the label data; after multiple iterations, when the consistency of the training set loss function and the test set loss function meets the requirements, a trained formation dip determination model is obtained.

11. A device for determining formation inclination, characterized in that: include: A data acquisition module, used for acquiring three-dimensional post-stack seismic data of the formation to be predicted; The formation dip determination module is used to input the three-dimensional post-stack seismic data into a trained formation dip determination model and output the formation dip of the formation to be predicted; the formation dip determination model is trained by the formation dip determination model training method described in any one of claims 1-8.

12. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, which, when executed by a processor, can implement the formation dip angle determination model training method described in any one of claims 1-8 and / or the formation dip angle determination method described in claim 9.

13. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for training a formation inclination determination model described in any one of claims 1 to 8 and / or the method for determining formation inclination described in claim 9 are implemented.

Images