Seismic facies category prediction model training method and device based on expert knowledge

Abstract

The application discloses a seismic facies category prediction model training method and device based on expert knowledge, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring seismic data in a target work area; the seismic data comprises seismic profile information and seismic horizon interpretation information; wherein the seismic facies category is labeled in the seismic profile information; the seismic profile information is mapped to the seismic horizon interpretation information to obtain a low-frequency background model representing the relationship between the seismic horizon and the seismic facies category; and the low-frequency background model is used as a constraint to train a preset classification model using the seismic data as a training sample to obtain a seismic facies category prediction model. According to the embodiment of the application, the training of the model is constrained by the seismic horizon, the lateral structural change information is considered, and therefore the trained seismic facies category prediction model can automatically predict the seismic facies category, and the prediction efficiency of the seismic facies category is improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a method and apparatus for training a seismic facies category prediction model based on expert knowledge. Background Technology

[0002] Seismic facies is a comprehensive response of sedimentary bodies to various seismic signals, including amplitude, frequency, velocity, and reflection structure. It is typically described by analyzing the internal structure and external morphology of the seismic facies of each layer, supplemented by seismic parameters such as the amplitude and continuity of the seismic reflection phase axis. Improving the accuracy of seismic facies identification and enhancing oil and gas recovery are crucial tasks for oilfield development.

[0003] Traditional seismic facies analysis methods mainly involve obtaining seismic profile information, then interpreters visually observing the seismic profile information and combining it with their professional knowledge and experience to give reasonable seismic facies classification results.

[0004] However, for a certain work area, if the seismic phase of the work area is to be identified, different seismic profiles within the work area need to be interpreted. The amount of seismic profile information is very large. If it is manually interpreted by interpreters, it will consume a lot of time and be inefficient. Summary of the Invention

[0005] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method and apparatus for training a seismic facies category prediction model based on expert knowledge, so as to improve the accuracy of seismic facies category prediction.

[0006] Firstly, this application provides a method for training a seismic facies category prediction model based on expert knowledge, the method comprising:

[0007] Acquire seismic data within the target work area; the seismic data includes seismic profile information and seismic horizon interpretation information; wherein, the seismic profile information is marked with seismic facies categories;

[0008] The seismic profile information is mapped to the seismic horizon interpretation information to obtain a low-frequency background model that characterizes the relationship between seismic horizons and seismic facies categories.

[0009] Using the low-frequency background model as a constraint, the earthquake data is used as training samples to train a preset classification model, thereby obtaining an earthquake facies category prediction model.

[0010] According to the expert knowledge-based earthquake facies category prediction model training method of this application, seismic data within the target work area is acquired, and the seismic profile information is mapped to the seismic horizon interpretation information to obtain a low-frequency background model characterizing the relationship between seismic horizons and earthquake facies categories. Using the low-frequency background model as a constraint, the seismic data is used as training samples to train a preset classification model, resulting in an earthquake facies category prediction model. This embodiment of the application, by constraining the training of the earthquake facies category prediction model with seismic horizons, considers the information on lateral structural changes in earthquakes. This allows the model to expand the interpretation of earthquake facies categories in the training samples during training, thereby enabling the trained earthquake facies category prediction model to automatically predict earthquake facies categories and improving the prediction efficiency.

[0011] According to one embodiment of this application, the seismic facies category in seismic profile information is labeled in the following manner:

[0012] Obtain the sampling points on the seismic profile in the seismic profile information;

[0013] Based on the reflection characteristics of the sampling points, the seismic facies category corresponding to the reflection characteristics is matched from a preset rule set; wherein, the rule set includes the correspondence between reflection characteristics and seismic facies categories;

[0014] Mark the seismic facies category corresponding to the sampling point on the seismic profile.

[0015] According to one embodiment of this application, mapping the seismic profile information to the seismic horizon interpretation information to obtain a low-frequency background model characterizing the relationship between seismic horizons and seismic facies categories includes:

[0016] Based on the seismic horizon interpretation information, values ​​are assigned to seismic horizons to construct a scalar field characterizing the distribution of seismic horizons; wherein, the scalar values ​​of the same seismic horizon are the same in the scalar field.

[0017] The seismic profile information is mapped into the scalar field to construct a low-frequency background model.

[0018] In this embodiment, by constructing a scalar field, the seismic facies category interpretation profile can be effectively transformed from physical space to layered space, thereby constraining the network to effectively extract the complex nonlinear mapping relationship between seismic facies and seismic data. This achieves model training with a small number of sample labels, further improving the efficiency of seismic facies prediction.

[0019] According to one embodiment of this application, the step of assigning values ​​to seismic horizons based on the seismic horizon interpretation information to construct a scalar field characterizing the distribution of seismic horizons includes:

[0020] Obtain the seismic horizon distribution from the seismic horizon interpretation information;

[0021] Based on the seismic horizon distribution, the same scalar value is assigned to the same seismic horizon, thus obtaining a scalar field characterizing the seismic horizon distribution.

[0022] In this embodiment, after obtaining the seismic horizon distribution, the same scalar value is assigned to the same seismic horizon, and different scalar values ​​are assigned to different seismic horizons, so that the constructed scalar field can accurately label the seismic horizon distribution.

[0023] According to one embodiment of this application, mapping the seismic profile information to the scalar field to construct a low-frequency background model includes:

[0024] Based on the location information of sampling points on the seismic profile in the seismic profile information, the seismic facies category of the sampling points is mapped to the scalar field;

[0025] The relationship function between scalar values ​​in the scalar field and seismic phase categories is calculated by linear interpolation;

[0026] A low-frequency background model is constructed based on the aforementioned relationship function.

[0027] According to one embodiment of this application, mapping the seismic facies category of the sampling points to the scalar field based on the location information of the sampling points on the seismic profile information includes:

[0028] Through formula

[0029]

[0030] Map the seismic facies categories of the sampling points to the scalar field;

[0031] in, Let represent the scalar field, k represent the seismic profile, i represent the seismic profile number (i is a natural number greater than 0), τ represent the scalar value, w represent the width of the seismic profile, h represent the height of the seismic profile, and f represent the seismic facies category.

[0032] In this embodiment, since the stratigraphic information is contained in the scalar field, the interpolation is strictly constrained to be performed within the same stratigraphic layer indicated by the scalar value. This allows the lateral variation information of the structure to be introduced during the interpolation process, thereby enabling the robust and reliable extension of a small number of seismic facies interpretation profiles in the sample to the entire seismic work area. This results in a low-frequency background model that simultaneously follows stratigraphic interpretation and seismic facies interpretation, enabling the trained seismic facies category prediction model to automatically predict seismic facies categories and improving the prediction efficiency of seismic facies categories.

[0033] According to one embodiment of this application, constructing a low-frequency background model based on the relation function includes:

[0034] Radial basis functions are used to control the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile;

[0035] A low-frequency background model is constructed based on the horizontal variation and the relationship function.

[0036] According to one embodiment of this application, the method of using radial basis functions to control the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile includes:

[0037] Through formula

[0038]

[0039] Controlling the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile;

[0040] Among them, w i (x, y) represents the local spatial interpolation distance weight field at the point (x, y) to be interpolated on the i-th seismic profile, where the local spatial interpolation distance weight field represents the lateral variation, and (x, y) represents the position coordinates of the point to be interpolated. k y k K represents the location coordinates of the sampling point on the seismic profile. i The number of sampling points is represented by ε, and ε represents the radial parameter used to balance the importance of spatial distance correlation.

[0041] According to one embodiment of this application, constructing a low-frequency background model based on the lateral variation and the relational function includes:

[0042] Through formula

[0043]

[0044] Build a low-frequency background model;

[0045] Where q(x,y,τ,c) represents the low-frequency background model, c represents the seismic facies category at coordinates (x,y), and p i (τ,c) represents the relational function, and N represents the number of seismic profiles.

[0046] In this embodiment, the lateral variation of seismic facies within the same stratum is controlled by introducing distance-related weights through radial basis functions, thereby highlighting the influence of seismic facies interpretation with similar seismic profiles in the interpolation process, so that the low-frequency background model strictly follows the seismic interpretation horizon in the lateral direction.

[0047] According to one embodiment of this application, the step of calculating the relationship function between scalar values ​​in the scalar field and seismic phase categories through linear interpolation may include:

[0048] Linear interpolation is performed for each seismic phase category to obtain the relationship function between the scalar values ​​in the scalar field and the seismic phase category.

[0049] In this embodiment, linear interpolation is performed on each seismic phase category in the seismic profile, which avoids the introduction of non-integer floating-point values ​​during the interpolation process, thus preventing the determination of the seismic phase category.

[0050] According to one embodiment of this application, the classification model includes an encoder, a decoder, and a predictor;

[0051] The encoder is used to extract features from the input low-frequency background model and the seismic profile information through pooling layers and a residual learning module to obtain a feature vector; wherein, the residual learning module includes two consecutive convolutional layers;

[0052] The decoder is used to decode the feature vector to obtain decoded features;

[0053] The predictor is used to fuse the decoded features through the stacking of multiple convolutional layers to obtain seismic phase prediction results.

[0054] According to one embodiment of this application, the encoder and the decoder are connected via four different scale jump connections.

[0055] Secondly, this application provides a method for predicting seismic facies categories based on expert knowledge, including:

[0056] Acquire the earthquake data to be predicted for the target work area;

[0057] The earthquake data to be predicted is output to a preset earthquake facies category prediction model to obtain the prediction results output by the earthquake facies category prediction model; the prediction results characterize the earthquake facies category at each location of the earthquake profile in the earthquake data to be predicted.

[0058] The earthquake facies category prediction model is trained using the method described in the first aspect.

[0059] According to the expert knowledge-based earthquake facies prediction method provided in the embodiments of this application, the earthquake facies prediction model used in the embodiments of this application constrains the training of the earthquake facies prediction model by using earthquake horizons, taking into account the information on changes in the lateral structure of earthquakes. This allows the model to expand the interpretation of earthquake facies in the training samples during the training process, thereby enabling the trained earthquake facies prediction model to automatically predict earthquake facies and improve the prediction efficiency of earthquake facies.

[0060] Thirdly, this application provides a training device for a seismic facies category prediction model based on expert knowledge, the device comprising:

[0061] The first acquisition module is used to acquire seismic data within the target work area; the seismic data includes seismic profile information and seismic horizon interpretation information; wherein, the seismic profile information is marked with seismic facies categories;

[0062] The mapping module is used to map the seismic profile information to the seismic horizon interpretation information to obtain a low-frequency background model that characterizes the relationship between seismic horizons and seismic facies categories.

[0063] The training module is used to train a preset classification model using the earthquake data as training samples, with the low-frequency background model as a constraint, to obtain an earthquake facies category prediction model.

[0064] According to the earthquake facies category prediction model training device based on expert knowledge provided in the embodiments of this application, the earthquake facies category prediction model training is constrained by earthquake horizons, taking into account the information on earthquake lateral structural changes. This allows the model to expand the interpretation of earthquake facies categories in the training samples during the training process, thereby enabling the trained earthquake facies category prediction model to automatically predict earthquake facies categories and improve the prediction efficiency of earthquake facies categories.

[0065] Fourthly, this application provides a seismic facies category prediction device based on expert knowledge, the device comprising:

[0066] The second acquisition module is used to acquire the earthquake data to be predicted for the target work area;

[0067] The prediction module is used to output the earthquake data to be predicted to a preset earthquake facies category prediction model to obtain the prediction results output by the earthquake facies category prediction model; the prediction results characterize the earthquake facies category at each location of the earthquake profile in the earthquake data to be predicted.

[0068] The earthquake facies category prediction model is trained using the method described in the first aspect.

[0069] According to the expert knowledge-based seismic facies prediction device provided in the embodiments of this application, the seismic facies prediction model used in the embodiments of this application constrains the training of the seismic facies prediction model by using seismic horizons, taking into account the information on seismic lateral structural changes, so that the model can expand the interpretation of seismic facies in the training samples during the training process, thereby enabling the trained seismic facies prediction model to automatically predict seismic facies and improve the prediction efficiency of seismic facies.

[0070] Fifthly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first or second aspect above.

[0071] In a sixth aspect, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first or second aspect above.

[0072] In a seventh aspect, this application provides a chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the methods described in the first or second aspect above.

[0073] Eighthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first or second aspect above.

[0074] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects:

[0075] According to the expert knowledge-based earthquake facies category prediction model training method of this application, seismic data within the target work area is acquired, and the seismic profile information is mapped to the seismic horizon interpretation information to obtain a low-frequency background model characterizing the relationship between seismic horizons and earthquake facies categories. Using the low-frequency background model as a constraint, the seismic data is used as training samples to train a preset classification model, resulting in an earthquake facies category prediction model. This embodiment of the application, by constraining the training of the earthquake facies category prediction model with seismic horizons, considers the information on lateral structural changes in earthquakes. This allows the model to expand the interpretation of earthquake facies categories in the training samples during training, thereby enabling the trained earthquake facies category prediction model to automatically predict earthquake facies categories and improving the prediction efficiency.

[0076] Furthermore, in some embodiments, by constructing a scalar field, the seismic facies category interpretation profile can be effectively transformed from physical space to layered space, thereby constraining the network to effectively extract the complex nonlinear mapping relationship between seismic facies and seismic data. This achieves model training with a small number of sample labels, further improving the efficiency of seismic facies prediction.

[0077] Furthermore, in some embodiments, since the stratigraphic information is contained in the scalar field, the interpolation is strictly constrained to be performed within the same stratigraphic layer indicated by the scalar value. This allows the lateral variation information of the structure to be introduced during the interpolation process, thereby enabling the robust and reliable extension of a small number of seismic facies interpretation profiles in the sample to the entire seismic work area. This results in a low-frequency background model that simultaneously follows stratigraphic interpretation and seismic facies interpretation, enabling the trained seismic facies category prediction model to automatically predict seismic facies categories and improving the prediction efficiency of seismic facies categories.

[0078] Furthermore, in some embodiments, the lateral variation of seismic facies within the same stratum is controlled by introducing distance-related weights using radial basis functions, thereby highlighting the influence of seismic facies interpretation with similar seismic profiles in the interpolation process, so that the low-frequency background model strictly follows the seismic interpretation horizon in the lateral direction.

[0079] Furthermore, in some embodiments, linear interpolation is performed on each seismic phase category in the seismic profile, which avoids introducing non-integer floating-point values ​​during the interpolation process and thus failing to determine the seismic phase category.

[0080] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0081] The above and / or additional aspects and advantages of this application will become apparent and readily understood in conjunction with the following description of the embodiments in conjunction with the accompanying drawings, wherein:

[0082] Figure 1 This is a flowchart illustrating the training method for an earthquake facies category prediction model based on expert knowledge provided in an embodiment of this application.

[0083] Figure 2 This is a schematic diagram illustrating the seismic facies interpretation of an embodiment of this application;

[0084] Figure 3 This is a schematic diagram of a single seismic profile label according to an embodiment of this application;

[0085] Figure 4 This is a schematic diagram illustrating the cross-sectional distribution of the labels in an embodiment of this application;

[0086] Figure 5 This is a schematic diagram of the low-frequency background model construction process in an embodiment of this application;

[0087] Figure 6 This is a schematic diagram illustrating the distribution pattern of seismic facies within the same stratum defined by a spatial distance function in an embodiment of this application.

[0088] Figure 7This is a schematic diagram of a low-frequency background model for different seismic phases in an embodiment of this application;

[0089] Figure 8 This is the network architecture of the classification model in the embodiments of this application;

[0090] Figure 9 This is a schematic diagram illustrating the model training effect of an embodiment of this application;

[0091] Figure 10 This is a schematic diagram of a cross-section of the actual work area as interpreted in the embodiments of this application;

[0092] Figure 11 This is a schematic diagram illustrating the comparison between the prediction results of the well profile and the embodiments of this application, as explained by geological experts.

[0093] Figure 12 This is a flowchart illustrating the earthquake facies category prediction method based on expert knowledge provided in the embodiments of this application;

[0094] Figure 13 This is a schematic diagram of the structure of the earthquake facies category prediction model training device based on expert knowledge provided in the embodiments of this application;

[0095] Figure 14 This is a schematic diagram of the structure of the earthquake facies category prediction device based on expert knowledge provided in the embodiments of this application;

[0096] Figure 15 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0097] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0098] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0099] Traditional seismic facies analysis methods primarily involve interpreters visually observing seismic profile information after acquisition, and then combining this information with their professional knowledge and experience to arrive at a reasonable seismic facies classification. Of course, some technologies have also been developed for automatic seismic facies identification.

[0100] However, identifying seismic facies in a specific work area requires interpreting different seismic profiles within that area. The amount of seismic profile information is enormous, and both manual analysis by interpreters and current automatic seismic facies identification methods require extensive manual interpretation. Therefore, the efficiency of existing seismic facies category prediction methods is very low.

[0101] This application considers that if seismic horizon factors can be incorporated into seismic facies analysis, the complex nonlinear mapping relationship between seismic facies and seismic data can be effectively characterized. Furthermore, by training the model, the interpretation of seismic facies categories in the samples can be expanded, thereby enabling the trained seismic facies category prediction model to automatically predict seismic facies categories and improve the prediction efficiency of seismic facies categories.

[0102] The following description, in conjunction with the accompanying drawings, details the training method and apparatus for earthquake facies category prediction models based on expert knowledge provided in this application, through specific embodiments and application scenarios.

[0103] Among them, the earthquake facies category prediction model training method based on expert knowledge can be applied to the terminal, specifically executed by the hardware or software in the terminal.

[0104] The terminal includes, but is not limited to, portable communication devices such as mobile phones or tablets with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0105] The following embodiments describe a terminal including a display and a touch-sensitive surface. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, mouse, and joystick.

[0106] The earthquake facies category prediction model training method based on expert knowledge provided in this application embodiment can be executed by an electronic device or a functional module or entity in an electronic device that can implement the earthquake facies category prediction model training method based on expert knowledge. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, and wearable devices. The following uses an electronic device as the execution subject to illustrate the earthquake facies category prediction model training method based on expert knowledge provided in this application embodiment.

[0107] like Figure 1 As shown, the training method for the earthquake facies category prediction model based on expert knowledge includes steps 110, 120, and 130.

[0108] Step 110: Obtain seismic data within the target work area; the seismic data includes seismic profile information and seismic horizon interpretation information; the seismic profile information includes the seismic facies category.

[0109] Seismic data refers to information about the structure and properties of underground rocks obtained through seismic exploration techniques. This data is crucial for oil and gas exploration because it helps prospectors determine the presence, distribution, and scale of underground oil and gas resources.

[0110] The uses of seismic data include, but are not limited to: helping explorers determine the structure and properties of underground rocks, thereby identifying potential oil and gas reservoirs; providing information on underground stratigraphic structures and tectonic deformations to help explorers understand geological structures; and providing location and guidance information for oil and gas exploration, helping explorers determine the location and direction of exploration wells.

[0111] Seismic data can be obtained through seismic exploration techniques. Seismic exploration typically involves placing seismic sensors on the surface or underground, and then generating seismic waves using explosions or seismic sources. As these seismic waves pass through underground rock, they are reflected, refracted, and scattered by different types of rock. Seismic sensors record this reflection and refraction information of the seismic waves, thereby obtaining data on the structure and properties of the underground rock.

[0112] The key to interpreting seismic facies is establishing the framework of seismic facies for the work area, that is, those typical seismic facies with obvious reflection characteristics that are easy to identify and interpret. Typical seismic facies in general sedimentary basins mainly include: mound facies, lenticular bodies, infill facies, chaotic facies, high-amplitude sheet facies, and low-amplitude facies.

[0113] Seismic profile information refers to the information obtained by analyzing specific seismic profiles after acquiring a set of 3D seismic data. During the analysis of a specific seismic profile, the seismic facies category at various points along the profile can be determined based on its structural characteristics.

[0114] In this embodiment, after analyzing the obtained seismic profile information, the seismic facies categories at various locations can be marked on the seismic profile. The marking method can be labeling, for example, using numbers to label the seismic facies categories, such as label "1" representing seismic facies category 1, label "2" representing seismic facies category 2, and label "3" representing seismic facies category 3. Of course, labels can also be represented by other numbers, or by letters, symbols, or other methods or combinations thereof; this embodiment does not limit this.

[0115] Seismic horizon interpretation information is a crucial component of seismic tectonic interpretation. It is seismic horizon data obtained by tracing and interpreting seismic reflection characteristics of target horizons, such as amplitude, phase, morphology, continuity, and characteristic combinations, on the seismic data volume. In the embodiments of this application, seismic horizon interpretation information may include the identification results of seismic horizons on various seismic profiles within the seismic data.

[0116] Step 120: Map the seismic profile information to the seismic horizon interpretation information to obtain a low-frequency background model that characterizes the relationship between seismic horizons and seismic facies categories.

[0117] Seismic profile information is usually sparsely distributed in space and lacks information on changes in the lateral structure of earthquakes. If the prediction of earthquake facies is achieved solely by analyzing seismic profile information, it is difficult to effectively characterize the complex nonlinear mapping relationship between earthquake facies and seismic data.

[0118] In this embodiment of the application, by mapping seismic profile information to seismic horizon interpretation information, and using the seismic profiles and seismic horizons of the interpreted seismic facies to fit the mapping function, a low-frequency background model characterizing the relationship between seismic horizons and seismic facies categories is constructed. This low-frequency background model can be used to estimate the seismic facies categories at uninterpreted seismic profiles.

[0119] Step 130: Using the low-frequency background model as a constraint, train the preset classification model using earthquake data as training samples to obtain the earthquake phase category prediction model.

[0120] In this embodiment, the classification model can be constructed using machine learning-related algorithms. The classification model is trained using supervised learning, with the input being a low-frequency background model and seismic profile information, and the output being the seismic facies classification results on the target profile.

[0121] Specifically, fixed-scale seismic profiles can be extracted from seismic data along the directions of the seismic survey lines and longitudinal survey lines, and corresponding low-frequency background models can be constructed. Combined with the labels on the seismic facies profiles, the classification model is trained. Based on the comparison between the prediction results and the true values ​​corresponding to the labels, the loss function is calculated to provide input for backpropagation, so as to iteratively update the parameters of the classification model. After training is completed, the seismic facies category prediction model is obtained.

[0122] According to the expert knowledge-based earthquake facies category prediction model training method of this application, seismic data within the target work area is acquired, and seismic profile information is mapped to seismic horizon interpretation information to obtain a low-frequency background model characterizing the relationship between seismic horizons and earthquake facies categories. Using the low-frequency background model as a constraint, seismic data is used as training samples to train a pre-defined classification model, resulting in an earthquake facies category prediction model. This embodiment of the application, by constraining the training of the earthquake facies category prediction model with seismic horizons, takes into account information on lateral structural changes in earthquakes. This allows the model to expand the interpretation of earthquake facies categories in the training samples during training, thereby enabling the trained earthquake facies category prediction model to automatically predict earthquake facies categories and improving the prediction efficiency.

[0123] In some embodiments, seismic facies categories in seismic profile information can be labeled in the following ways:

[0124] Obtain sampling points on the seismic profile in the seismic profile information;

[0125] Based on the reflection characteristics of the sampling points, the seismic facies categories corresponding to the reflection characteristics are matched from a preset rule set; wherein, the rule set includes the correspondence between reflection characteristics and seismic facies categories;

[0126] Mark the seismic facies category corresponding to the sampling point on the seismic profile.

[0127] Specifically, reflection features are usually closely related to seismic facies categories, with different seismic facies categories corresponding to different reflection features. In this application, a correspondence between reflection features and seismic facies categories can be established in advance, forming a rule set. For example, reflection features can be stored in one column of a data table, and seismic facies categories can be stored in another column of the data table. There is a one-to-one correspondence between reflection features and seismic facies categories. After obtaining a reflection feature, the seismic facies category corresponding to that reflection feature can be obtained by looking up the table.

[0128] To illustrate with an example, such as Figure 2 As shown, after obtaining a set of 3D seismic data, specific seismic profiles are analyzed, among which... Figure 2The points in the diagram are the sampling points. On this seismic profile, the hillock-like seismic facies exhibits a flat-bottomed, convex-top structure, characterized internally by parallel to subparallel medium- to strong continuous reflection structures, chaotic reflection structures, or blank structures. The reflection structure of the hillock-like seismic facies in this profile shows that the top interface has medium to weak amplitude reflections and relatively clear boundaries. Figure 2 (Face B). Forelaport reflection configurations are reflection structures that extend towards deeper water, and their morphologies include S-shaped forelaport, oblique forelaport, S-shaped oblique composite, imbricate, and irregular forelaport reflections. In this profile, the forelaport is predominantly S-shaped (…). Figure 2 Face C in the middle. Its top and bottom strata have relatively gentle dip angles, basically parallel to the interface between the upper and lower strata, while the terminal forestrata have a steeper dip angle. Figure 2 (Middle black arrow). Another seismic phase, FaceC, is identified through the bottom of the sequence, the bioherm-like reflective bottom interface, and the basement. Figure 2 (FaceA in the image). Its reflection is less than a single cycle of waveform. On the seismic profile, FaceA corresponds to an extremely thin deposit between a wave peak and a wave trough.

[0129] by Figure 3 Taking the seismic profile shown as an example, Face B exhibits a distinct mound-like reflection on the seismic profile, Face C shows an S-shaped progradational reflection configuration, while the early progradational layer at the bottom of Face A is relatively thin. Figure 3 As shown, the labels corresponding to earthquakes are: 1 for the bottom fore-accretion layer of FaceA, 2 for FaceB, 3 for the fore-accretion layer of FaceC, and 0 for the background value.

[0130] In this example, different seismic profiles within the work area are interpreted and labeled in the same way. However, considering the long time required for actual seismic profile interpretation, interpreting a large number of seismic profiles as training data is virtually impossible. Therefore, 12 connecting lateral lines are interpreted as training data for the network within this work area. The distribution of these 12 interpreted profiles is as follows: Figure 4 As shown, the interpreted seismic profiles are relatively evenly distributed within the work area, accounting for 2.4% (12 / 500) of the total number of profiles.

[0131] In this embodiment, by introducing a set of rules that include the correspondence between reflection features and seismic facies categories, seismic facies categories can be accurately labeled on seismic profiles.

[0132] In some embodiments, mapping seismic profile information to seismic horizon interpretation information to obtain a low-frequency background model characterizing the relationship between seismic horizons and seismic facies categories may include:

[0133] Seismic horizons are assigned values ​​based on seismic horizon interpretation information to construct a scalar field characterizing the distribution of seismic horizons; in the scalar field, the scalar values ​​of the same seismic horizon are the same.

[0134] Seismic profile information is mapped into a scalar field to construct a low-frequency background model.

[0135] Specifically, a scalar field is a field that can be fully represented by its magnitude alone. Based on seismic horizon interpretation information, values ​​are assigned to seismic horizons; different horizons are assigned different values, while the same horizon has the same value. This results in a scalar field that represents the distribution of seismic horizons, and the isosurfaces of this scalar field can accurately trace seismic horizons.

[0136] By mapping seismic profile information into a scalar field, a low-frequency background model representing the relationship between seismic horizons and seismic facies categories can be obtained.

[0137] In this embodiment, by constructing a scalar field, the seismic facies category interpretation profile can be effectively transformed from physical space to layered space, thereby constraining the network to effectively extract the complex nonlinear mapping relationship between seismic facies and seismic data. This achieves model training with a small number of sample labels, further improving the efficiency of seismic facies prediction.

[0138] In some embodiments, assigning values ​​to seismic horizons based on seismic horizon interpretation information to construct a scalar field characterizing the distribution of seismic horizons includes:

[0139] Obtain the seismic horizon distribution from the seismic horizon interpretation information;

[0140] By assigning the same scalar value to the same seismic horizon based on the seismic horizon distribution, a scalar field characterizing the seismic horizon distribution is obtained.

[0141] In one example Figure 5 The value 'a' in the diagram shows the actual seismic data and the two seismic horizons superimposed on it. To construct a low-frequency background model, a scalar field was first built using seismic horizon interpretation data. Figure 5 In b), the scalar field contains information on the lateral structural changes of all layers and is used to constrain the construction of the low-frequency background model.

[0142] In this embodiment, after obtaining the seismic horizon distribution, the same scalar value is assigned to the same seismic horizon, and different scalar values ​​are assigned to different seismic horizons, so that the constructed scalar field can accurately label the seismic horizon distribution.

[0143] In some embodiments, mapping seismic profile information into a scalar field to construct a low-frequency background model includes:

[0144] Based on the location information of sampling points on the seismic profile, the seismic facies category of the sampling points is mapped to the scalar field;

[0145] The relationship function between scalar values ​​in the scalar field and seismic facies categories is calculated using linear interpolation.

[0146] A low-frequency background model is constructed based on the relational function.

[0147] Specifically, assuming there are N seismic profiles with interpreted seismic facies categories in the target exploration area (assuming their length is h and width is w), then each profile... It contains multiple sampling points, each of which is assigned a seismic facies category:

[0148]

[0149] Among them, f k，i Indicates sampling point x located on the seismic profile. k，i The seismic facies category of the location, where i represents the seismic profile number, ranging from 1 to N.

[0150] In some embodiments, it can be done through formula

[0151]

[0152] Map the seismic phase categories of the sampling points to a scalar field;

[0153] in, Let represent the scalar field, k represent the seismic profile, i represent the seismic profile number (i is a natural number greater than 0), τ represent the scalar value, w represent the width of the seismic profile, h represent the height of the seismic profile, and f represent the seismic facies category.

[0154] Since the sampling points on the seismic profile are sparsely distributed and discontinuous in space, after mapping the seismic facies categories of the sampling points to the scalar field, interpolation can be performed using linear interpolation to obtain the relationship function between the scalar values ​​in the scalar field and the seismic facies categories.

[0155] In this embodiment, since the stratigraphic information is contained in the scalar field, the interpolation is strictly constrained to be performed within the same stratigraphic layer indicated by the scalar value. This allows the lateral variation information of the structure to be introduced during the interpolation process, thereby enabling the robust and reliable extension of a small number of seismic facies interpretation profiles in the sample to the entire seismic work area. This results in a low-frequency background model that simultaneously follows stratigraphic interpretation and seismic facies interpretation, enabling the trained seismic facies category prediction model to automatically predict seismic facies categories and improving the prediction efficiency of seismic facies categories.

[0156] In some embodiments, constructing a low-frequency background model based on a relational function includes:

[0157] Radial basis functions are used to control the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile;

[0158] A low-frequency background model is constructed based on horizontal variation and relational functions.

[0159] Specifically, it can be done through formulas

[0160]

[0161] Controlling the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile;

[0162] Among them, w i (x,y) represents the local spatial interpolation distance weight field at the point (x,y) to be interpolated on the i-th seismic profile. The local spatial interpolation distance weight field represents the lateral variation, and (x,y) represents the position coordinates of the point to be interpolated. k ,y k K represents the location coordinates of the sampling point on the seismic profile. i The number of sampling points is represented by ε, and ε represents the radial parameter used to balance the importance of spatial distance correlation.

[0163] It can be done through formula

[0164]

[0165] Build a low-frequency background model;

[0166] Where q(x,y,τ,c) represents the low-frequency background model, c represents the seismic facies category at coordinates (x,y), and p i (τ,c) represents the relational function, and N represents the number of seismic profiles.

[0167] The above example will be used as an illustration. Figure 5 Figure 'c' shows the 12 seismic profiles interpreted in this seismic survey area, including 10 profiles along the lateral line and two seismic profiles along the longitudinal line. Using the aforementioned interpolation method, interpolation was performed along the seismic line and perpendicular to the seismic line under the control of a scalar field, ensuring that the low-frequency background model strictly follows the seismic interpretation horizon in the lateral direction. In addition, spatial distance functions were used in the directions along and perpendicular to the seismic line... Figure 6 a and Figure 6 b) defines the distribution pattern of seismic phases within the same stratum, and finally calculates the interpolated low-frequency background model by weighted averaging.

[0168] In this embodiment, the lateral variation of seismic facies within the same stratum is controlled by introducing distance-related weights through radial basis functions, thereby highlighting the influence of seismic facies interpretation with similar seismic profiles in the interpolation process, so that the low-frequency background model strictly follows the seismic interpretation horizon in the lateral direction.

[0169] In some embodiments, calculating the relationship function between scalar values ​​in a scalar field and seismic facies categories via linear interpolation may include:

[0170] Linear interpolation is performed for each seismic facies category to obtain the relationship function between scalar values ​​in the scalar field and seismic facies categories.

[0171] Specifically, by performing linear interpolation for each seismic facies category, a relationship function between the scalar value and each seismic facies category can be obtained. Correspondingly, based on these relationship functions, a low-frequency background model corresponding to each seismic facies category can be obtained, such as... Figure 7 As shown, a, b, c, and d represent the low-frequency background models corresponding to different earthquakes.

[0172] In this embodiment, linear interpolation is performed on each seismic phase category in the seismic profile, which avoids the introduction of non-integer floating-point values ​​during the interpolation process, thus preventing the determination of the seismic phase category.

[0173] In some embodiments, the classification model includes an encoder, a decoder, and a predictor;

[0174] The encoder is used to extract features from the input low-frequency background model and seismic profile information through pooling layers and a residual learning module to obtain feature vectors; the residual learning module includes two consecutive convolutional layers.

[0175] The decoder is used to decode the feature vector to obtain the decoded features;

[0176] The predictor is used to fuse decoded features through the stacking of multiple convolutional layers to obtain seismic facies prediction results.

[0177] The encoder and decoder are connected via four different scale jump connections.

[0178] In one example, such as Figure 8 As shown, Figure 8The network architecture used for the classification model in this example is shown below. This architecture uses a simplified encoder-decoder framework based on UNet. The network is mainly divided into three parts: an encoder, a decoder, and a predictor. The encoder part consists of delayed layers and a residual learning module, where pooling layers are used to downsample the input to expand the network's receptive field, allowing the subsequent residual learning module to systematically aggregate multi-scale structural features. In the encoder part, the input seismic data and low-frequency background model are first fed into a 5×5 convolutional layer to compute 16 locally translation-invariant feature vectors, which are then passed to a ReLU activation function and an instance normalization layer. The activated feature vectors are then fed into four connected pooling layers and a residual learning module to further extract information from different spatial scales. Each residual learning module consists of two consecutive convolutional layers with 3×3 kernels. The module is represented as a residual learning function using the local skip connections at both ends of the convolutional layers, which facilitates gradient backpropagation training of the network to improve performance and accelerate computation.

[0179] The decoder contains the same number of residual learning modules and upsampling interpolation layers as the pooling layers in the encoder. Its main function is to further extract feature patterns from the input data and restore the spatial size of the downsampled feature map, mapping the encoded output feature vector back to the input size. At each spatial scale, upsampling can be performed using simple linear interpolation, as no trainable parameters are needed to compensate for the size mismatch between the two concatenated features. The upsampled feature map is then fed into the residual modules for further feature extraction. The encoder and decoder are connected via four skip connections at different scales, which not only compensate for spatial resolution but also aggregate feature patterns across multiple scales. In the predictor part of the network, stacked convolutional layers are used to fuse all decoded features to obtain the final deposition phase prediction result. Instance normalization and activation functions are applied to the outputs of all convolutional layers except the last one.

[0180] The effectiveness of the earthquake facies category prediction model training method based on expert knowledge provided in this application is illustrated using actual earthquake data as an example. The earthquake data contains 401 (survey line direction) × 357 (longitudinal survey line direction) × 161 (time or depth axis) sampling points. The earthquake work area contains two interpreted seismic horizons, as well as 10 seismic facies interpretation profiles along the survey line direction and 2 along the longitudinal survey line direction.

[0181] In the classification model training, fixed-scale seismic profiles are extracted along the seismic lines and longitudinal lines, and corresponding low-frequency background models are constructed. These are combined with labels from a small number of seismic facies profiles to iteratively update the classification model parameters. To stabilize the training process, the input seismic data is normalized. In this normalization process, the mean and standard deviation of the entire seismic data volume are first calculated, and then the mean divided by the corresponding standard deviation is subtracted to obtain the normalized seismic data. During training, training data is input in batches, with a batch size of 8. This example uses the Adam optimizer with an adaptive step size to accelerate training, with an initial learning rate of 0.01, which is gradually decreased as predefined performance metrics cease to improve.

[0182] To evaluate the training performance of the classification model, this example uses four seismic traces from the validation dataset to estimate the corresponding impedances. The validation dataset includes three additional seismic profiles of interpreted seismic facies, which are used as blind tests and are not involved in the classification model training process, to fully demonstrate the improvement in the final predictions. Figure 9 Figures a, b, and c show the input seismic data, the manually interpreted seismic facies profile, and the model-predicted seismic facies profile, respectively. The results show that the distribution characteristics of the predicted seismic facies on the profiles are largely consistent with the interpreted labels. Due to its excellent performance on both training and validation datasets, the network is considered to have successfully learned how to predict seismic facies from the input seismic data.

[0183] To further verify the accuracy of the seismic facies category prediction model training method based on expert knowledge provided in the embodiments of this application, the inventors conducted a seismic facies prediction experiment in another large work area. A small number of samples were pre-trained for classification using the same method before well control. This example interprets a total of 10 seismic lateral lines, with the number of labels accounting for 0.25% (10 / 4000) of the total number of seismic profile traces. Figure 10 As shown. Using the method proposed in this application for training and prediction, the small sample prediction results show good performance in controlling seismic facies boundaries, and are basically consistent with the previously manually interpreted well-connected facies, such as... Figure 11 As shown. For example, Figure 11 The lower part shows the prediction results for this example. The darker lower part represents reefs, mainly distributed on one side of the platform edge, and are significantly controlled by topography. After introducing geological constraint information such as seismic horizons in this embodiment, the seismic facies are predicted through small-sample learning training. This method significantly improves the characterization of facies boundaries and seismic facies morphology, as well as the prediction efficiency, compared to traditional methods. The results are basically consistent with previous interpretations, saving a significant amount of manual interpretation time while ensuring prediction accuracy.

[0184] This application also provides a method for predicting seismic facies categories based on expert knowledge, such as... Figure 12As shown, the earthquake facies category prediction method based on expert knowledge includes steps 1210 and 1220.

[0185] Acquire the earthquake data to be predicted for the target work area;

[0186] The earthquake data to be predicted is output to the preset earthquake facies category prediction model to obtain the prediction results output by the earthquake facies category prediction model; the prediction results represent the earthquake facies category at each location of the earthquake profile in the earthquake data to be predicted.

[0187] The seismic facies category prediction model is trained using the seismic facies category prediction model training method provided in the embodiments of this application.

[0188] The seismic facies prediction model used in this application constrains the training of the seismic facies prediction model by taking into account the information on lateral structural changes in earthquakes. This allows the model to expand the interpretation of seismic facies in the training samples during the training process, thereby enabling the trained seismic facies prediction model to automatically predict seismic facies and improve the prediction efficiency of seismic facies.

[0189] The earthquake facies category prediction model training method based on expert knowledge provided in this application can be executed by an earthquake facies category prediction model training device based on expert knowledge. This application uses an example of an earthquake facies category prediction model training device based on expert knowledge executing the earthquake facies category prediction model training method to illustrate the earthquake facies category prediction model training device based on expert knowledge provided in this application.

[0190] This application also provides a training device for a seismic facies category prediction model based on expert knowledge.

[0191] like Figure 13 As shown, the training device for the earthquake facies category prediction model based on expert knowledge includes:

[0192] The first acquisition module 1310 is used to acquire seismic data within the target work area; the seismic data includes seismic profile information and seismic horizon interpretation information; wherein, the seismic profile information is marked with seismic facies categories;

[0193] The mapping module 1320 is used to map seismic profile information to seismic horizon interpretation information to obtain a low-frequency background model that characterizes the relationship between seismic horizon and seismic facies category.

[0194] Training module 1330 is used to train a preset classification model with seismic data as training samples, constrained by a low-frequency background model, to obtain a seismic facies category prediction model.

[0195] According to the earthquake facies category prediction model training device provided in the embodiments of this application, the earthquake facies category prediction model training is constrained by the earthquake horizon, taking into account the information on the transverse structural changes of earthquakes. This allows the model to expand the interpretation of earthquake facies categories in the training samples during the training process, thereby enabling the trained earthquake facies category prediction model to automatically predict earthquake facies categories and improve the prediction efficiency of earthquake facies categories.

[0196] In some embodiments, seismic facies categories in seismic profile information are labeled in the following manner:

[0197] Obtain sampling points on the seismic profile in the seismic profile information;

[0198] Based on the reflection characteristics of the sampling points, the seismic facies categories corresponding to the reflection characteristics are matched from a preset rule set; wherein, the rule set includes the correspondence between reflection characteristics and seismic facies categories;

[0199] Mark the seismic facies category corresponding to the sampling point on the seismic profile.

[0200] In some embodiments, the mapping module 1320 is further configured to:

[0201] Obtain sampling points on the seismic profile in the seismic profile information;

[0202] Based on the reflection characteristics of the sampling points, the seismic facies categories corresponding to the reflection characteristics are matched from a preset rule set; wherein, the rule set includes the correspondence between reflection characteristics and seismic facies categories;

[0203] Mark the seismic facies category corresponding to the sampling point on the seismic profile.

[0204] In some embodiments, the mapping module 1320 is further configured to:

[0205] Obtain the seismic horizon distribution from the seismic horizon interpretation information;

[0206] By assigning the same scalar value to the same seismic horizon based on the seismic horizon distribution, a scalar field characterizing the seismic horizon distribution is obtained.

[0207] In some embodiments, the mapping module 1320 is further configured to:

[0208] Based on the location information of sampling points on the seismic profile, the seismic facies category of the sampling points is mapped to the scalar field;

[0209] The relationship function between scalar values ​​in the scalar field and seismic facies categories is calculated using linear interpolation.

[0210] A low-frequency background model is constructed based on the relational function.

[0211] In some embodiments, the mapping module 1320 is further configured to:

[0212] Through formula

[0213]

[0214] Map the seismic phase categories of the sampling points to a scalar field;

[0215] in, Let represent the scalar field, k represent the seismic profile, i represent the seismic profile number (i is a natural number greater than 0), τ represent the scalar value, w represent the width of the seismic profile, h represent the height of the seismic profile, and f represent the seismic facies category.

[0216] In some embodiments, the mapping module 1320 is further configured to:

[0217] Radial basis functions are used to control the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile;

[0218] A low-frequency background model is constructed based on horizontal variation and relational functions.

[0219] In some embodiments, the mapping module 1320 is further configured to:

[0220] Through formula

[0221]

[0222] Controlling the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile;

[0223] Among them, w i (x,y) represents the local spatial interpolation distance weight field at the point (x,y) to be interpolated on the i-th seismic profile. The local spatial interpolation distance weight field represents the lateral variation, and (x,y) represents the position coordinates of the point to be interpolated. k ,y k K represents the location coordinates of the sampling point on the seismic profile. i The number of sampling points is represented by ε, and ε represents the radial parameter used to balance the importance of spatial distance correlation.

[0224] In some embodiments, the mapping module 1320 is further configured to:

[0225] Through formula

[0226]

[0227] Build a low-frequency background model;

[0228] Where q(x,y,τ,c) represents the low-frequency background model, c represents the seismic facies category at coordinates (x,y), and p i(τ,c) represents the relational function, and N represents the number of seismic profiles.

[0229] In some embodiments, the mapping module 1320 is further configured to:

[0230] Linear interpolation is performed for each seismic facies category to obtain the relationship function between scalar values ​​in the scalar field and seismic facies categories.

[0231] In some embodiments, the classification model includes an encoder, a decoder, and a predictor;

[0232] The encoder is used to extract features from the input low-frequency background model and seismic profile information through pooling layers and a residual learning module to obtain feature vectors; the residual learning module includes two consecutive convolutional layers.

[0233] The decoder is used to decode the feature vector to obtain the decoded features;

[0234] The predictor is used to fuse decoded features through the stacking of multiple convolutional layers to obtain seismic facies prediction results.

[0235] In some embodiments, the encoder and decoder are connected via four different scale jump connections.

[0236] The seismic facies category prediction method based on expert knowledge provided in this application can be executed by a seismic facies category prediction device based on expert knowledge. This application uses the example of a seismic facies category prediction device based on expert knowledge executing the method to illustrate the seismic facies category prediction device based on expert knowledge provided in this application.

[0237] This application also provides an earthquake facies category prediction device based on expert knowledge.

[0238] like Figure 14 As shown, the earthquake facies category prediction device based on expert knowledge includes:

[0239] The second acquisition module 1410 is used to acquire the earthquake data to be predicted for the target work area;

[0240] The prediction module 1420 is used to output the earthquake data to be predicted to a preset earthquake facies category prediction model to obtain the prediction results output by the earthquake facies category prediction model; the prediction results characterize the earthquake facies category at each location of the earthquake profile in the earthquake data to be predicted.

[0241] The seismic facies category prediction model is trained using the expert knowledge-based seismic facies category prediction model training method provided in the embodiments of this application.

[0242] According to the earthquake facies prediction device based on expert knowledge provided in the embodiments of this application, the earthquake facies prediction model is used to constrain the training of the earthquake facies prediction model by using earthquake horizons. It takes into account the information on changes in the lateral structure of earthquakes, so that the model can expand the interpretation of earthquake facies in the training samples during the training process. In this way, the trained earthquake facies prediction model can automatically predict earthquake facies, thereby improving the prediction efficiency of earthquake facies.

[0243] The earthquake facies category prediction model training device or the earthquake facies category prediction device based on expert knowledge in the embodiments of this application can be an electronic device or a component in an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM or self-service machine, etc. The embodiments of this application do not specifically limit it.

[0244] The earthquake facies category prediction model training device or the earthquake facies category prediction device based on expert knowledge in the embodiments of this application can be a device with an operating system. The operating system can be a Microsoft (Windows) operating system, an Android operating system, an iOS operating system, or other possible operating systems; this application embodiment does not specifically limit the specific operating system.

[0245] In some embodiments, such as Figure 15As shown, this application embodiment also provides an electronic device 1500, including a processor 1501, a memory 1502, and a computer program stored in the memory 1502 and executable on the processor 1501. When the program is executed by the processor 1501, it implements the various processes of the above-described training method for the earthquake facies category prediction model based on expert knowledge or the embodiment of the earthquake facies category prediction method based on expert knowledge, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0246] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0247] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described training method for the earthquake facies category prediction model based on expert knowledge or the embodiment of the earthquake facies category prediction method based on expert knowledge, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0248] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0249] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described expert knowledge-based seismic facies category prediction model training method or expert knowledge-based seismic facies category prediction method.

[0250] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0251] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-described expert knowledge-based seismic facies category prediction model training method or expert knowledge-based seismic facies category prediction method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0252] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0253] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0254] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0255] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0256] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0257] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for training a seismic facies category prediction model based on expert knowledge, characterized in that, include: Acquire seismic data within the target work area; The seismic data includes seismic profile information and seismic horizon interpretation information; wherein, the seismic profile information is marked with seismic facies categories; The seismic profile information is mapped to the seismic horizon interpretation information to obtain a low-frequency background model that characterizes the relationship between seismic horizons and seismic facies categories. Using the low-frequency background model as a constraint, the earthquake data is used as training samples to train a preset classification model, thereby obtaining an earthquake facies category prediction model. The process of mapping the seismic profile information to the seismic horizon interpretation information to obtain a low-frequency background model characterizing the relationship between seismic horizons and seismic facies categories includes: Based on the seismic horizon interpretation information, values ​​are assigned to seismic horizons to construct a scalar field characterizing the distribution of seismic horizons; wherein, the scalar values ​​of the same seismic horizon are the same in the scalar field; the seismic profile information is mapped to the scalar field to construct a low-frequency background model; The step of mapping the seismic profile information into the scalar field to construct a low-frequency background model includes: Based on the location information of sampling points on the seismic profile, the seismic facies category of the sampling point is mapped to the scalar field; the relationship function between the scalar value in the scalar field and the seismic facies category is calculated by linear interpolation; and a low-frequency background model is constructed based on the relationship function. The interpolation is strictly constrained to be performed within the same seismic horizon indicated by the scalar value; under the control of the scalar field, interpolation is performed along the survey line and in the direction perpendicular to the survey line, so that the low-frequency background model strictly follows the seismic horizon interpretation information in the lateral direction; the distribution pattern of the seismic phase within the same seismic horizon is defined by the spatial distance function in the direction along the survey line and in the direction perpendicular to the survey line, respectively; finally, the interpolated low-frequency background model is calculated by weighted averaging.

2. The method according to claim 1, characterized in that, Seismic facies categories in seismic profile information are labeled in the following ways: Obtain the sampling points on the seismic profile in the seismic profile information; Based on the reflection characteristics of the sampling points, the seismic facies category corresponding to the reflection characteristics is matched from a preset rule set; wherein, the rule set includes the correspondence between reflection characteristics and seismic facies categories; Mark the seismic facies category corresponding to the sampling point on the seismic profile.

3. The method according to claim 1, characterized in that, Assigning values ​​to seismic horizons based on the seismic horizon interpretation information to construct a scalar field characterizing the distribution of seismic horizons includes: Obtain the seismic horizon distribution from the seismic horizon interpretation information; Based on the seismic horizon distribution, the same scalar value is assigned to the same seismic horizon, thus obtaining a scalar field characterizing the seismic horizon distribution.

4. The method according to claim 1, characterized in that, The step of mapping the seismic facies category of the sampling points to the scalar field based on the location information of the sampling points on the seismic profile information includes: Through formula Map the seismic facies categories of the sampling points to the scalar field; in, k,i Represents a scalar field. k Indicates earthquake profile, i Indicates the seismic profile number. i For any natural number greater than 0, Represents a scalar value. w Indicates the width of the seismic profile. h Indicates the height of the seismic profile. f Indicates the earthquake phase category.

5. The method according to claim 1, characterized in that, The step of constructing a low-frequency background model based on the relation function includes: Radial basis functions are used to control the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile; A low-frequency background model is constructed based on the horizontal variation and the relationship function.

6. The method according to claim 5, characterized in that, The method of using radial basis functions to control the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile includes: Through formula Controlling the lateral variation of seismic facies categories within the same seismic horizon in a seismic profile; in, Indicates the first i Points to be interpolated on a seismic profile The local spatial interpolation distance weight field represents the lateral variation. Indicates the position coordinates of the point to be interpolated. This indicates the location coordinates of the sampling points on the seismic profile. Indicates the number of sampling points. This represents the radial parameter used to balance the relative importance of spatial distance.

7. The method according to claim 6, characterized in that, The construction of the low-frequency background model based on the lateral variation and the relationship function includes: Through formula Build a low-frequency background model; in, This represents a low-frequency background model. c Representing coordinates Earthquake phase category at the location, This represents the relational function. N Indicates the number of seismic profiles.

8. The method according to claim 1, characterized in that, The function for calculating the relationship between scalar values ​​in the scalar field and seismic facies categories through linear interpolation includes: Linear interpolation is performed for each seismic phase category to obtain the relationship function between the scalar values ​​in the scalar field and the seismic phase category.

9. The method according to claim 1, characterized in that, The classification model includes an encoder, a decoder, and a predictor; The encoder is used to extract features from the input low-frequency background model and the seismic profile information through pooling layers and a residual learning module to obtain a feature vector; wherein, the residual learning module includes two consecutive convolutional layers; The decoder is used to decode the feature vector to obtain decoded features; The predictor is used to fuse the decoded features through the stacking of multiple convolutional layers to obtain seismic phase prediction results.

10. The method according to claim 9, characterized in that, The encoder and the decoder are connected by four jump connections of different scales.

11. A method for predicting seismic facies categories based on expert knowledge, characterized in that, include: Acquire the earthquake data to be predicted for the target work area; The earthquake data to be predicted is output to a preset earthquake facies category prediction model to obtain the prediction results output by the earthquake facies category prediction model; the prediction results characterize the earthquake facies category at each location of the earthquake profile in the earthquake data to be predicted. The earthquake facies prediction model is trained using the method described in any one of claims 1-10.

12. A training device for a seismic facies category prediction model based on expert knowledge, characterized in that, A method for training an expert knowledge-based seismic facies category prediction model as described in any one of claims 1-10, comprising: The first acquisition module is used to acquire seismic data within the target work area; the seismic data includes seismic profile information and seismic horizon interpretation information; wherein, the seismic profile information is marked with seismic facies categories; The mapping module is used to map the seismic profile information to the seismic horizon interpretation information to obtain a low-frequency background model that characterizes the relationship between seismic horizons and seismic facies categories. The training module is used to train a preset classification model using the earthquake data as training samples, with the low-frequency background model as a constraint, to obtain an earthquake facies category prediction model.

13. A training device for a seismic facies category prediction model based on expert knowledge, characterized in that, include: The second acquisition module is used to acquire the earthquake data to be predicted for the target work area; The prediction module is used to output the earthquake data to be predicted to a preset earthquake facies category prediction model to obtain the prediction results output by the earthquake facies category prediction model; the prediction results characterize the earthquake facies category at each location of the earthquake profile in the earthquake data to be predicted. The earthquake facies prediction model is trained using the method described in any one of claims 1-10.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-11.

15. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-11.

16. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-11.

Images