Method and device for determining seismic structure tensor feature vector and method and device for estimating stratigraphic dip angle

Abstract

The invention discloses a seismic structure tensor feature vector determination method and device and a stratigraphic dip angle estimation method and device. The seismic structure tensor feature vector determination method comprises the steps of preprocessing target seismic data; inputting the preprocessed target seismic data into a pre-trained feature vector calculation model; outputting a feature vector of the target seismic data through a feature vector calculation model; the feature vector calculation model is obtained by training in the following mode: pre-training a deep neural network model; seismic data obtained from actual data are used as training sample data, the training sample data are input into a deep neural network model obtained through pre-training, parameters of the deep neural network model obtained through pre-training are adjusted in an unsupervised training mode, and a feature vector calculation model is obtained. According to the method, the feature vectors can be quickly and efficiently calculated, the stratigraphic dip angle attributes can be calculated, and the interpretation efficiency of seismic data is integrally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic signal processing, and in particular to a method for determining a characteristic vector of a seismic structure tensor, and a method and device for estimating a stratum dip. Background Art

[0002] The seismic structure tensor originates from the image structure algorithm in the field of image processing, and has gradually become one of the basic attribute parameters widely used in seismic exploration. In the process of seismic data interpretation, the local structural eigenvector can be obtained by calculating the seismic structure tensor, and the formation dip can be further estimated. (When calculating the dip, the estimation result is determined by the relationship that the direction of the eigenvector of the structure tensor is perpendicular to the dip direction). In fact, the eigenvector based on the structure tensor is not only an important way to calculate the local dip, but also can be used for fault interpretation, unconformity analysis, guided filtering and other tasks, and a series of related seismic data interpretation methods have been developed. At present, due to the quality of the seismic data itself, noise often greatly affects the calculation results of the structure tensor. For the stable acquisition of the two-dimensional or three-dimensional image structure tensor, a series of stable algorithms based on local statistical optimization have also been developed, but they are very computationally intensive in three-dimensional calculations.

[0003] With the development of big data artificial intelligence technology, geophysicists have begun to think about how to use deep neural networks to implement traditional geophysical techniques, such as first arrival picking and fault identification, which have achieved good results and improved efficiency. The latest literature shows that by making a large number of stratigraphic dip attribute labels, efficient attribute calculation based on supervision can be achieved, which provides ideas for the rapid calculation of seismic structure tensors and related eigenvectors. However, in specific practical applications, due to the diversity of actual data in terms of bin size, sampling interval, and data main frequency, there are differences between the attribute labels of supervised training and the actual data, which will greatly affect the generalization ability of the training network parameters and have a key impact on the prediction results of the attributes. Summary of the invention

[0004] In view of the above problems, the present invention is proposed to provide a method for determining the characteristic vector of a seismic structure tensor, a method and device for estimating formation dip, which 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 method for determining an eigenvector of a seismic structure tensor, comprising:

[0006] Preprocessing the target seismic data;

[0007] Input the preprocessed target seismic data into the pre-trained feature vector calculation model;

[0008] Outputting the characteristic vector of the target seismic data through the characteristic vector calculation model;

[0009] The feature vector calculation model is trained in the following way:

[0010] Pre-training of deep neural network models;

[0011] The seismic data obtained from actual data are used as training sample data and input into the pre-trained deep neural network model. The parameters of the pre-trained deep neural network model are adjusted through unsupervised training to obtain the feature vector calculation model.

[0012] In one embodiment, the pre-training of the deep neural network model includes:

[0013] Using the synthesized seismic profile data as training sample data, inputting into the deep neural network model;

[0014] Through unsupervised training, the parameters of the neural network model are adjusted to obtain a pre-trained deep neural network model.

[0015] In one embodiment, the seismic data obtained from the actual data is used as training sample data, input into the pre-trained deep neural network model, and the parameters of the pre-trained deep neural network model are adjusted by unsupervised training to obtain the feature vector calculation model, including:

[0016] Input the training sample data into a deep neural network model; the deep neural network model uses a convolutional neural network as a basic unit and includes a plurality of residual block structures; output the estimated values ​​of each component of the feature vector u(x) at each sampling point through the neural network model;

[0017] Substitute the estimated values ​​of each component of u(x) into the preset loss function:

[0018]

[0019] Among them: e(x,y)=||g(y)-[g(y) T u(x)]u(x)||y∈Ω x ;

[0020] The above Ω x Represents the local range at the spatial position x, where x represents the position vector in space, and its components are the coordinates of each coordinate axis. y represents the neighborhood Ω x All spatial positions within the range, g(y)-[g(y) Tu(x)]u(x) is Ω x The projection of the gradient vector g(y) inside on the unit normal vector u(x);

[0021] According to the loss value calculated by the loss function, the parameters of the deep neural network model are adjusted through error back propagation;

[0022] The training sample data is two-dimensional sampling data including an inline direction and a crossline direction; or the training sample data is three-dimensional sampling data including an inline direction, a crossline direction and a time direction.

[0023] In one embodiment, the local range Ω at sample x x For a larger window length along the time direction, a smaller window length is selected along the inline or crossline direction.

[0024] In one embodiment, the target seismic data is preprocessed, including:

[0025] If the target seismic data is two-dimensional, a 3×3 template filter is performed on the target seismic data;

[0026] If the target seismic data is three-dimensional, a 3×3×3 template filter is performed on the target seismic data.

[0027] In a second aspect, an embodiment of the present invention provides a method for estimating a formation dip, comprising:

[0028] Obtaining characteristic vectors of target seismic data;

[0029] Obtaining a first component of the eigenvector in the horizontal direction and a second component of the eigenvector in the vertical direction; the horizontal direction is an inline direction or a crossline direction; the vertical direction is a time direction;

[0030] According to the lateral seismic grid spacing and vertical depth sampling, the first component and the second component of the characteristic vector are corrected to obtain a geometric vector corresponding to the characteristic vector in the actual space;

[0031] Determine the formation dip angle where the target seismic data is located according to the geometric vector;

[0032] The step of obtaining the characteristic vector of the target seismic data is achieved by the aforementioned method for determining the characteristic vector of the seismic structure tensor.

[0033] In one embodiment, the first component and the second component of the characteristic vector are corrected according to the lateral seismic grid spacing and the vertical depth sampling to obtain a geometric vector corresponding to the characteristic vector in the actual space, including:

[0034] Calculate vertical depth sampling travel time:

[0035] Calculate the geometric vector u corresponding to the eigenvector in the actual space real (x) = [u1 × O int , u2×T int ];

[0036] t sample is the sampling time interval, Vel is the reference speed; u1 is the first component of the characteristic vector in the horizontal direction, and u2 is the second component of the characteristic vector in the vertical direction; int is the lateral seismic grid spacing; T int It is the vertical depth sampling;

[0037] Determining the formation dip angle where the target seismic data is located according to the geometric vector includes:

[0038] Through the following formula Calculate the inclination of the formation where the target seismic data is located.

[0039] In a third aspect, an embodiment of the present invention provides a device for determining a characteristic vector of a seismic structure tensor, comprising:

[0040] A preprocessing module, used for preprocessing target seismic data;

[0041] A prediction module is used to input the pre-processed target seismic data into a pre-trained feature vector calculation model; and output the feature vector of the target seismic data through the feature vector calculation model;

[0042] The feature vector calculation model is trained in the following way:

[0043] Pre-training of deep neural network models;

[0044] The seismic data obtained from actual data are used as training sample data and input into the pre-trained deep neural network model. The parameters of the pre-trained deep neural network model are adjusted through unsupervised training to obtain the feature vector calculation model.

[0045] In a fourth aspect, an embodiment of the present invention provides a device for estimating a formation dip, characterized in that it includes:

[0046] A first acquisition module is used to obtain a characteristic vector of target seismic data;

[0047] A second acquisition module is used to acquire a first component of the characteristic vector in a horizontal direction and a second component of the characteristic vector in a vertical direction; the horizontal direction is an inline direction or a crossline direction; the vertical direction is a time direction;

[0048] A correction module, used for correcting the first component and the second component of the characteristic vector according to the horizontal seismic grid spacing and the vertical depth sampling, so as to obtain a geometric vector corresponding to the characteristic vector in the actual space;

[0049] A formation dip calculation module, used to determine the formation dip where the target seismic data is located according to the geometric vector;

[0050] The step of obtaining the characteristic vector of the target seismic data is achieved by the aforementioned method for determining the characteristic vector of the seismic structure tensor.

[0051] In a fifth aspect, an embodiment of the present invention provides a computing 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, it implements the method for determining the characteristic vector of the seismic structure tensor as described above, or implements the method for estimating the formation dip as described above.

[0052] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for determining the characteristic vector of the seismic structure tensor as described above is implemented, or the method for estimating the formation dip angle as described above is implemented.

[0053] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0054] The method for determining the eigenvector of the seismic structure tensor and the method and device for estimating the formation dip provided in the embodiments of the present invention obtain a eigenvector calculation model through deep neural network model training. During the training process, stable parameters are first obtained through pre-training, and then the pre-trained deep neural network model is trained by unsupervised learning using actual seismic data as samples. The pre-trained deep neural network is migrated and the parameters are fine-tuned, which enables seismic interpreters to apply it to actual seismic data with different work area parameters and data characteristics. Since there is no need to produce a large number of labels, the existing seismic data can be fully utilized. The application process has strong applicability to actual data, is simple and easy to implement, and can improve the overall efficiency of seismic data interpretation.

[0055] 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.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] 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:

[0058] Figure 1 is a flow chart of a method for determining a characteristic vector of a seismic structure tensor in an embodiment of the present invention;

[0059] Figure 2 A schematic diagram of training a deep neural network in an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of the characteristic vector of the structure tensor and the formation dip angle in a two-dimensional case in an embodiment of the present invention;

[0061] Figure 4 A flow chart of a method for estimating formation dip in an embodiment of the present invention;

[0062] Figure 5 Schematic diagram of a two-dimensional seismic data profile, seismic characteristic vectors and formation dip angle in an embodiment of the present invention;

[0063] Figure 6 This is an example of a flow chart for solving the formation dip angle in an embodiment of the present invention;

[0064] Figure 7 Schematic diagram of comparison between earthquake characteristic vector (two-dimensional) and earthquake dip attribute and actual data profile in an embodiment of the present invention;

[0065] Figure 8 It is a schematic diagram of a time slice of the formation dip calculated by a traditional method in an embodiment of the present invention and a time slice in the method for estimating the formation dip provided by the present invention;

[0066] Fig. 9 Schematic diagram of the structure of a device for determining a characteristic vector of a seismic structure tensor according to an embodiment of the present invention;

[0067] Fig.10 It is a schematic diagram of the structure of the device for estimating the formation inclination in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] 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.

[0069] In conjunction with the accompanying drawings, specific implementations of the method for determining the eigenvector of the seismic structure tensor and the method and device for estimating the formation dip angle provided by the embodiments of the present invention are described in detail below.

[0070] The embodiment of the present invention provides a method for determining the characteristic vector of the earthquake structure tensor, referring to Figure 1 As shown, the following steps are included:

[0071] S11, preprocessing the target seismic data;

[0072] S12, inputting the pre-processed target seismic data into a pre-trained feature vector calculation model;

[0073] S13, outputting the characteristic vector of the target seismic data through a characteristic vector calculation model;

[0074] The above-mentioned eigenvector calculation model is used to predict the eigenvector of target seismic data. The eigenvector calculation model can be trained in the following way:

[0075] Pre-training of deep neural network models;

[0076] The seismic data obtained from actual data are used as training sample data and input into the pre-trained deep neural network model. The parameters of the pre-trained deep neural network model are adjusted through unsupervised training to obtain the feature vector calculation model.

[0077] In the above step S11, the target seismic data is preprocessed mainly to improve the stability of the gradient vector calculation in the deep neural network model. Therefore, the data is pre-smoothed and then input into the designed network. Another way is to set a smoothing module in the deep neural network model to preprocess the target seismic data. The embodiment of the present invention does not limit which method is used.

[0078] For example, in actual data applications, users can design pre-smoothing parameters based on the quality of the target seismic data, that is, the noise level, generally a 3*3 template filter (3*3*3 template filter for three-dimensional) for pre-smoothing processing.

[0079] Furthermore, in the embodiment of the present invention, the above-mentioned deep neural network model adopts a convolutional neural network (CNN) as a basic unit, for example Figure 2 The network architecture shown includes several residual blocks (RES), for example, 3 residual blocks. If the target seismic data adopts three-dimensional data (i.e., including inline direction, crossline direction and time direction), then these 3 residual blocks are a three-dimensional CNN network architecture.

[0080] The preprocessed target seismic data is input into a pre-trained eigenvector calculation model, and the estimated values ​​of each component of the eigenvector u(x) at each sampling point are predicted by the eigenvector calculation model; when the target seismic data is two-dimensional data, the estimated values ​​of each component of u(x) are 2-channel outputs, and when the target seismic data is three-dimensional data, the estimated values ​​of each component of u(x) are 3-channel outputs.

[0081] The output of the convolutional neural network is input into the preset loss function to calculate the loss. According to the loss value calculated by the loss function, the parameters of the deep neural network model are adjusted through error back propagation.

[0082] Error back propagation is: output error -> hidden layer (layer by layer) -> input layer. Its main purpose is to distribute the error to all units in each layer by back propagating the output error, so as to obtain the error signal of each layer unit, and then correct the weight of each unit. The process of weight adjustment is also the learning and training process of the network.

[0083] The above training sample data is two-dimensional sampling data including inline direction and crossline direction; the training sample data is three-dimensional sampling data including inline direction, crossline direction and time direction.

[0084] The loss function can be defined as:

[0085]

[0086] Considering the influence of noise, in order to make the solution of seismic eigenvector (perpendicular to seismic reflection) more stable, an improved target optimization algorithm was proposed in the field of image processing, which defines

[0087] e(x,y)=||g(y)-[g(y) T u(x)]u(x)||y∈Ω x ; Formula (2)

[0088] Ω xRepresents the local range at the spatial position x, where x represents the position vector in space, and its components are the coordinates of each coordinate axis. y represents the neighborhood Ω x All spatial positions within the range, g(y)-[g(y) T u(x)]u(x) is Ω x The projection of the gradient vector g(y) inside on the unit normal vector u(x);

[0089] Reference Figure 3 As shown, by minimizing the range Ω x The projection of can obtain the normal direction, which is the above target loss function:

[0090]

[0091] In the embodiment of the present invention, a deep convolutional network is considered to fit and obtain the normal feature vector u(x), and the deep convolutional network is expressed as the following mathematical formula:

[0092]

[0093] in is the seismic data section; f θ [·] represents a deep convolutional network, and θ is a network parameter that needs to be obtained through training.

[0094]

[0095] It can be seen from the above formula (4) that the network parameters can be trained by outputting the seismic data profile to solve the target optimization function. This process does not require labeled data. Therefore, the feature vector calculation model can be obtained through the unsupervised learning training method.

[0096] Furthermore, the local range Ω at the sampling x x For a larger window length along the time direction, a smaller window length is selected along the inline or crossline direction.

[0097] Furthermore, in order to make the network optimization more stable in the presence of noise, the objective loss function can be further optimized as E ρ =E(x) / (0.5+E(x)).

[0098] In an embodiment of the present invention, the deep neural network model can obtain stable parameters through pre-training, and then use actual seismic data as samples to train the pre-trained deep neural network model using unsupervised learning. During the pre-training process, the network is also trained using unsupervised learning. At this time, seismic data with a good signal-to-noise ratio and high quality can be used, or noise-free data (such as synthetic seismic data profiles) can be used directly to pre-train the deep neural network model. The pre-trained deep neural network model is then trained in an unsupervised manner using actual seismic data (which may contain more noise) in the actual seismic data, so that the model is more likely to converge.

[0099] The process of pre-training can be similar to the process of formal training after pre-training, except that different training samples are used. The specific training process can be found in the above-mentioned model training process, which will not be repeated here.

[0100] The feature vector calculation model provided in the embodiment of the present invention adopts an unsupervised training mechanism and does not require labeled data. Therefore, the processed actual seismic data can be directly input into the network for training and fine-tuning to optimize the model parameters, so that a network model with better generalization can be obtained for the target data to achieve efficient feature vector calculation.

[0101] Furthermore, when the target seismic data is two-dimensional, it is generally necessary to calculate the inline and crossline directions. Therefore, seismic profiles can be extracted in the two directions for transfer learning fine-tuning, and then the feature vectors in the two directions can be calculated according to the profiles.

[0102] When the target seismic data is three-dimensional, considering the computing power of the graphics card memory during the actual training process, the three-dimensional data can be first divided into small three-dimensional data volumes, and then input into the three-dimensional convolutional network for training, and the parameters of the model can be fine-tuned. When predicting, the three-dimensional data is input into the network in blocks to realize model training.

[0103] The above-mentioned characteristic vector based on the structure tensor can also be widely used in tasks such as fault interpretation, unconformity analysis, guided filtering, etc. The embodiment of the present invention does not limit this.

[0104] The present invention also provides a method for estimating the formation dip angle. Figure 4 As shown, the following steps are included:

[0105] S41, obtaining a characteristic vector of target seismic data;

[0106] In this step S41, the characteristic vector of the target seismic data can be obtained according to the implementation method of the aforementioned method for determining the characteristic vector of the seismic structure tensor.

[0107] S42, obtaining a first component of the characteristic vector in the horizontal direction and a second component of the characteristic vector in the vertical direction; the horizontal direction is the inline direction or the crossline direction; the vertical direction is the time direction;

[0108] S43, correcting the first component and the second component of the characteristic vector according to the lateral seismic grid spacing and the vertical depth sampling to obtain a geometric vector corresponding to the characteristic vector in the actual space;

[0109] S44. Determine the formation inclination where the target seismic data is located based on the geometric vector.

[0110] In the above process, the characteristic vector is corrected according to the grid spacing and depth sampling interval of the actual data, and then the estimated value of the formation dip is obtained according to the geometric relationship. For time domain data, a reference velocity needs to be set to convert the time sampling interval into depth sampling (i.e. vertical depth sampling).

[0111] The feature vector predicted by the trained network can be further used to quickly obtain the formation reflection dip angle based on the geometric relationship. The feature vector obtained based on the deep neural network does not contain the geometric parameters of the target data, such as the bin size, sampling interval, etc. Therefore, in practical applications, the calculated feature vector needs to be processed and corrected according to the target data parameters so that it represents the true formation reflection surface normal vector.

[0112] Furthermore, the above S43 corrects the first component and the second component of the characteristic vector according to the horizontal seismic grid spacing and the vertical depth sampling to obtain the geometric vector corresponding to the characteristic vector in the actual space, which can be specifically implemented in the following manner:

[0113] Calculate vertical depth sampling travel time:

[0114] Calculate the geometric vector u corresponding to the eigenvector in the actual space real (x) = [u1 × O int , u2×T int ];

[0115] t sample is the sampling time interval, Vel is the reference speed; u1 is the first component of the characteristic vector in the horizontal direction, and u2 is the second component of the characteristic vector in the vertical direction; int is the lateral seismic grid spacing; T int It is the vertical depth sampling;

[0116] Determining the formation dip angle where the target seismic data is located according to the geometric vector includes:

[0117] Through the following formula Calculate the inclination of the formation where the target seismic data is located.

[0118] For details, please refer to the attached Figure 3 If the inclination angle in a certain direction of inline or crossline is calculated, the reflection surface normal vector u(x) (i.e., eigenvector) is first calculated through the network. The components of the lateral (inline or crossline direction) coordinate axis and the vertical (time direction) coordinate axis are u1 and u2, i.e., u(x) = [u1, u2]; the lateral seismic grid spacing is O int , vertical depth sampling is T int (For time domain data, such as t sample For time sampling, according to the two-way travel time principle: Vel is the reference velocity), so the eigenvector is corrected to the geometric vector u in the actual space real (x) = [u1 × O int , u2×T int ], so the inclination angle of this section direction can be estimated as

[0119] In order to make the neural network more clearly defined when predicting the inclination (calculating the arctan), when predicting the vertical u y When quantifying, constrain it to always be positive.

[0120] Reference Figure 5 As shown, Figure 5 The leftmost figure is a schematic diagram of a two-dimensional seismic data profile, the middle figure is a schematic diagram of a seismic characteristic vector (two-dimensional), and the rightmost figure is a schematic diagram of a formation dip.

[0121] An example of a flowchart for solving the formation dip angle can be found in Figure 6 As shown, from this Figure 6 It can be seen that the deep neural network is first built, then pre-trained, and then the pre-trained deep neural network is migrated and parameter fine-tuned. In the process of migration and parameter fine-tuning, the target seismic data needs to be pre-processed such as smoothing. The feature vector is predicted by the trained feature vector calculation model, and the obtained feature vector is parameter-corrected to obtain the formation inclination.

[0122] In the embodiment of the present invention, a method for determining the characteristic vector of a seismic structure tensor and a method for estimating the formation dip angle are proposed, and an efficient calculation implementation method of the formation characteristic vector based on unsupervised deep learning is proposed. The method is based on a deep convolutional neural network and combines the optimization equation of the image structure tensor to realize efficient formation characteristic vector calculation with unsupervised network optimization.

[0123] Furthermore, on the basis of obtaining the characteristic vector, the formation dip estimation is obtained through post-processing correction of the work area parameters. At the same time, the use of transfer learning and training fine-tuning mechanisms enables seismic interpreters to apply it to actual seismic data with different work area parameters and data characteristics. Although this technology is based on deep neural networks, it does not require the production of a large number of labels and can make full use of existing seismic data. It has strong applicability to actual data during application, is simple and easy to implement, and can improve the overall efficiency of seismic data interpretation. Practice has shown that this method and process can be applied to complex geological conditions and low seismic data quality, providing a powerful tool for the estimation of key attributes such as formation dip.

[0124] Taking the actual 3D seismic data dip calculation results of a certain work area as an example, refer to Figure 7 FIG. 1 is a schematic diagram showing a comparison between the seismic characteristic vector (two-dimensional) and seismic dip attributes and the actual data profiles realized by the method for determining the characteristic vector of the seismic structure tensor and the method for estimating the formation dip provided by the embodiment of the present invention. Figure 8 As shown, the figure on the left is a schematic diagram of the time slice of the formation dip calculated according to the traditional method, and the figure on the right is a schematic diagram of the time slice of the formation dip calculated according to the method for estimating the formation dip provided by an embodiment of the present invention.

[0125] from Figure 7 and Figure 8 It can be seen that the method of the present invention greatly improves the calculation efficiency compared with the traditional method in the application of actual data, and the final calculation effect is basically the same.

[0126] Based on the same inventive concept, an embodiment of the present invention also provides a device for determining the characteristic vector of a seismic structure tensor, a device for estimating the inclination of a formation, a computing device and a computer-readable storage medium. Since the principles of the problems solved by these devices are similar to the aforementioned method for determining the characteristic vector of the seismic structure tensor and the method for estimating the inclination of a formation, the implementation of the device can refer to the implementation of the aforementioned method, and the repeated parts will not be repeated.

[0127] The embodiment of the present invention provides a device for determining a characteristic vector of a seismic structure tensor, referring to Fig. 9 As shown, the determining device includes:

[0128] A preprocessing module 91 is used to preprocess the target seismic data;

[0129] The prediction module 92 is used to input the pre-processed target seismic data into a pre-trained feature vector calculation model; and output the feature vector of the target seismic data through the feature vector calculation model;

[0130] The feature vector calculation model is trained in the following way:

[0131] Pre-training of deep neural network models;

[0132] The seismic data obtained from actual data are used as training sample data and input into the pre-trained deep neural network model. The parameters of the pre-trained deep neural network model are adjusted through unsupervised training to obtain the feature vector calculation model.

[0133] The embodiment of the present invention also provides a device for estimating the formation dip angle, referring to Fig.10 As shown, including:

[0134] A first acquisition module 101 is used to obtain a characteristic vector of target seismic data;

[0135] The second acquisition module 102 is used to acquire a first component of the feature vector in a horizontal direction and a second component of the feature vector in a vertical direction; the horizontal direction is an inline direction or a crossline direction; the vertical direction is a time direction;

[0136] A correction module 103, configured to correct the first component and the second component of the characteristic vector according to the lateral seismic grid spacing and the vertical depth sampling, so as to obtain a geometric vector corresponding to the characteristic vector in the actual space;

[0137] A formation dip calculation module 104 is used to determine the formation dip where the target seismic data is located according to the geometric vector;

[0138] The step of obtaining the characteristic vector of the target seismic data is achieved by the aforementioned method for determining the characteristic vector of a seismic structure tensor.

[0139] An embodiment of the present invention also provides a computing 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, it implements the method for determining the characteristic vector of the seismic structure tensor as described above, or implements the method for estimating the formation dip as described above.

[0140] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for determining the characteristic vector of the seismic structure tensor as mentioned above, or implements the method for estimating the formation dip as mentioned above.

[0141] Regarding the device for determining the characteristic vector of the above-mentioned seismic structure tensor and the device for estimating the formation dip in the above-mentioned embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method and will not be elaborated here.

[0142] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0143] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0144] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0146] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for determining an eigenvector of a seismic structure tensor, characterized in that: include: Preprocessing the target seismic data; Input the preprocessed target seismic data into the pre-trained feature vector calculation model; Outputting the characteristic vector of the target seismic data through the characteristic vector calculation model; The feature vector calculation model is trained in the following way: Pre-training of deep neural network models; The seismic data obtained from actual data are used as training sample data and input into the pre-trained deep neural network model. The parameters of the pre-trained deep neural network model are adjusted through unsupervised training to obtain the feature vector calculation model.

2. The method according to claim 1, characterized in that The pre-training of the deep neural network model includes: Using the synthesized seismic profile data as training sample data, inputting into the deep neural network model; Through unsupervised training, the parameters of the neural network model are adjusted to obtain a pre-trained deep neural network model.

3. The method according to claim 1, characterized in that The seismic data obtained from the actual data is used as training sample data, input into the pre-trained deep neural network model, and the parameters of the pre-trained deep neural network model are adjusted by unsupervised training to obtain the feature vector calculation model, including: Input the training sample data into a deep neural network model; the deep neural network model uses a convolutional neural network as a basic unit and includes a plurality of residual block structures; output the estimated values ​​of each component of the feature vector u(x) at each spatial position x through the neural network model; Substitute the estimated values ​​of each component of u(x) into the preset loss function: where: e(x,y) = ||g(y) - [g(y) T u(x)]u(x)|| for y ∈ Ω x ; The above Ω x Represents the local range at the spatial position x, where x represents the position vector in space, and its components are the coordinates of each coordinate axis. y represents the neighborhood Ω x All spatial positions within the range, g(y)-[g(y) T u(x)]u(x) is Ω x The projection of the gradient vector g(y) inside on the unit normal vector u(x); According to the loss value calculated by the loss function, the parameters of the deep neural network model are adjusted through error back propagation; The training sample data is two-dimensional sampling data including an inline direction and a crossline direction; or the training sample data is three-dimensional sampling data including an inline direction, a crossline direction and a time direction.

4. The method according to claim 3, characterized in that The local range Ω at sample x x For a larger window length along the time direction, a smaller window length is selected along the inline or crossline direction.

5. The method according to any one of claims 1 to 4, characterized in that: Preprocess the target seismic data, including: If the target seismic data is two-dimensional, a 3×3 template filter is performed on the target seismic data; If the target seismic data is three-dimensional, a 3×3×3 template filter is performed on the target seismic data.

6. A method for estimating formation dip, characterized in that: include: Obtaining characteristic vectors of target seismic data; Obtaining a first component of the eigenvector in the horizontal direction and a second component of the eigenvector in the vertical direction; the horizontal direction is an inline direction or a crossline direction; the vertical direction is a time direction; According to the horizontal seismic grid spacing, the vertical depth sampling T int , correcting the first component and the second component of the eigenvector to obtain a geometric vector corresponding to the eigenvector in the actual space; Determine the formation dip angle where the target seismic data is located according to the geometric vector; The step of obtaining the characteristic vector of the target seismic data is achieved by a method for determining the characteristic vector of a seismic structure tensor as described in any one of claims 1-5.

7. The method according to claim 6, characterized in that According to the lateral seismic grid spacing and vertical depth sampling, the first component and the second component of the characteristic vector are corrected to obtain a geometric vector corresponding to the characteristic vector in the actual space, including: Calculate vertical depth sampling travel time: Calculate the geometric vector u corresponding to the eigenvector in the actual space real (x) = [u1 × O int , u2×T int ]; t sample is the sampling time interval, Vel is the reference speed; u1 is the first component of the characteristic vector in the horizontal direction, and u2 is the second component of the characteristic vector in the vertical direction; int is the lateral seismic grid spacing; T int It is the vertical depth sampling; Determining the formation dip angle where the target seismic data is located according to the geometric vector includes: Through the following formula Calculate the inclination of the formation where the target seismic data is located.

8. A device for determining a characteristic vector of a seismic structure tensor, characterized in that: include: A preprocessing module, used for preprocessing target seismic data; A prediction module is used to input the pre-processed target seismic data into a pre-trained feature vector calculation model; and output the feature vector of the target seismic data through the feature vector calculation model; The feature vector calculation model is trained in the following way: Pre-training of deep neural network models; The seismic data obtained from actual data are used as training sample data and input into the pre-trained deep neural network model. The parameters of the pre-trained deep neural network model are adjusted through unsupervised training to obtain the feature vector calculation model.

9. A device for estimating formation inclination, characterized in that: include: A first acquisition module is used to obtain a characteristic vector of target seismic data; A second acquisition module is used to acquire a first component of the characteristic vector in a horizontal direction and a second component of the characteristic vector in a vertical direction; the horizontal direction is an inline direction or a crossline direction; the vertical direction is a time direction; A correction module, used for correcting the first component and the second component of the characteristic vector according to the horizontal seismic grid spacing and the vertical depth sampling, so as to obtain a geometric vector corresponding to the characteristic vector in the actual space; A formation dip calculation module, used to determine the formation dip where the target seismic data is located according to the geometric vector; The step of obtaining the characteristic vector of the target seismic data is achieved by a method for determining the characteristic vector of a seismic structure tensor as described in any one of claims 1-5.

10. A computing 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 determining the characteristic vector of the seismic structure tensor as described in any one of claims 1 to 5 is implemented, or the method for estimating the formation dip as described in claim 6 or 7 is implemented.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for determining the characteristic vector of the seismic structure tensor as described in any one of claims 1 to 5, or implements the method for estimating the formation dip as described in claim 6 or 7.

Images