A seismic phase prediction model, seismic phase prediction method and device

Abstract

The present invention discloses a method and apparatus for establishing an earthquake phase prediction model. The method comprises obtaining a plurality of first-category plane earthquake attribute data and a plurality of second-category plane earthquake attribute data that have been screened and gridded according to the same rule; screening unlabeled grids from each plane earthquake attribute data according to a set rule, extracting each first-category earthquake attribute value and each second-category earthquake attribute value of the labeled grids and the screened unlabeled grids; obtaining a first labeled sample set, a first unlabeled sample set, a second labeled sample set, and a second unlabeled sample set; and using each obtained sample set to train a selected classification model through a collaborative training method to obtain a classification model for earthquake phase prediction. The method can improve the prediction accuracy of earthquake phases and reduce the dependence on labeled samples.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration and development, and in particular to the establishment of a seismic phase prediction model, a seismic phase prediction method and a device. Background Art

[0002] With the widespread adoption of 3D seismic data and advances in geophysical processing and interpretation technologies, the use of seismic attribute data for geological interpretation and analysis can improve the accuracy of understanding subsurface geological patterns and reduce the uncertainty of reservoir prediction. However, quantification is difficult to achieve when using seismic attribute data for geological interpretation due to the multiplicity of interpretations and human subjectivity. To address this issue, many experts have proposed supervised and unsupervised learning classification algorithms.

[0003] In recent years, Bagheri et al. (2013) used a classification algorithm based on difference analysis to analyze seismic phases; Majid Bagheri & Mohammad Ali Riahi (2014) used classification algorithms such as multilayer perceptron, support vector machine, Fisher, Balson and nearest neighbor algorithms to analyze seismic phases; Dario Grana (2017) used Bayesian and expectation–maximization methods to analyze seismic phases; H. Sabeti (2009) & Ivan Sánchez Galvis (2017) used k-means clustering algorithm to analyze seismic phases; Fabio Radomille Santan & Arthur Ayres (2016) used the CNN clustering algorithm to analyze seismic phases. Neto (2014) used principal component analysis, phylogenetic tree, and k-means algorithm to perform seismic facies analysis based on seismic attribute data, and then performed sedimentary facies classification; Zhao et al. (2015) comprehensively analyzed the k-means algorithm, self-organizing map, generative topological mapping algorithm, support vector machine, Gaussian mixture model, and artificial neural network for seismic facies analysis, and proposed using unsupervised learning to determine the initial estimate of the verification classification category, and then using supervised learning for classification.

[0004] However, these methods often encounter unavoidable problems when using conventional supervised or unsupervised learning algorithms for seismic facies analysis. Supervised learning algorithms use only a small amount of labeled data for training, ignoring a large amount of unlabeled sample data. Furthermore, sufficient labeled samples are required to ensure the generalization of the training model. In oil and gas field exploration and development, the number of wells drilled is often small, so the use of supervised learning methods for seismic facies analysis and prediction generally suffers from poor generalization. Unsupervised learning algorithms use full sample data for training, highlighting the global characteristics of seismic attributes. However, the clustering decision boundaries are not constrained by well points, and the clustering results generally have a low consistency with the well points. Therefore, using clustering algorithms based on seismic attributes for reservoir predictions makes it difficult to produce geological interpretation results that are highly consistent with the well point data.

[0005] Therefore, the existing technology of using conventional supervised learning algorithms or unsupervised learning algorithms to perform seismic phase analysis has the problems of limited applicability or low prediction accuracy. Summary of the Invention

[0006] Miller and Uyar (1997) theoretically demonstrated that if a correlation can be established between the target and unlabeled sample distributions, using unlabeled samples to train a classifier can improve classification performance, thus giving rise to semi-supervised learning methods. Semi-supervised learning methods, also known as collaborative training algorithms, were originally proposed by Blum and Mitchell (1998). Generally speaking, collaborative training algorithms assume that the data feature set can be divided into two subsets, and that training on each subset yields a good classifier. The classifiers on the two feature subsets learn from each other to complete the classification task: they retrain using each other's classification results until both classifiers achieve the same classification results for the vast majority of the data.

[0007] In view of the above problems, the present invention is proposed to provide a seismic phase prediction model, a seismic phase prediction method and a device that overcome the above problems or at least partially solve the above problems, which can improve the prediction accuracy of seismic phases and reduce dependence on labeled samples.

[0008] In a first aspect, an embodiment of the present invention provides a method for establishing an earthquake phase prediction model, comprising:

[0009] Data acquisition step: obtaining a plurality of first-type plane seismic attribute data and a plurality of second-type plane seismic attribute data that are screened and gridded according to the same rule, and marking the grids in each plane seismic attribute data that match the well location information with seismic facies according to the sedimentary microfacies type of the target layer of the well;

[0010] The sample set establishment steps are as follows: unlabeled grids are screened from the seismic attribute data of each plane according to the set rules, and the first-category seismic attribute values ​​and the second-category seismic attribute values ​​of the labeled grids and the screened unlabeled grids are extracted; the seismic phase label and the first-category seismic attribute values ​​of the labeled grid are used as a labeled sample to obtain a first labeled sample set; the first-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a first unlabeled sample set; the seismic phase label and the second-category seismic attribute values ​​of the labeled grid are used as a labeled sample to obtain a second labeled sample set; the second-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a second unlabeled sample set;

[0011] Model training step: using the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set, the selected classification model is trained by a collaborative training method to obtain a classification model for earthquake phase prediction.

[0012] In a second aspect, an embodiment of the present invention provides an earthquake phase prediction method, comprising:

[0013] Multiple types of seismic attribute data are input into the classification model, and seismic phase prediction is performed based on the output results. The classification model is obtained using the above-mentioned seismic phase prediction model establishment method, and the type is consistent with the type of planar seismic attribute data screened when establishing the classification model.

[0014] In a third aspect, an embodiment of the present invention provides a device for establishing an earthquake phase prediction model, comprising:

[0015] A data acquisition module is used to obtain a plurality of first-type plane seismic attribute data and a plurality of second-type plane seismic attribute data that are screened and gridded according to the same rules. The grids in each plane seismic attribute data that match the well location information are seismically labeled according to the sedimentary microfacies type of the target layer of the well;

[0016] The sample set establishment module is used to screen unlabeled grids from the seismic attribute data of each plane according to a set rule, extract the first-category seismic attribute values ​​and the second-category seismic attribute values ​​of the labeled grid and the screened unlabeled grid; the seismic phase label of the labeled grid and the first-category seismic attribute values ​​are used as a labeled sample to obtain a first labeled sample set; the first-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a first unlabeled sample set; the seismic phase label of the labeled grid and the second-category seismic attribute values ​​are used as a labeled sample to obtain a second labeled sample set; the second-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a second unlabeled sample set;

[0017] The model training module is used to train the selected classification model by using the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set through a collaborative training method to obtain a classification model for earthquake phase prediction.

[0018] In a fourth aspect, an embodiment of the present invention provides a computer program product with earthquake phase prediction function, including a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the establishment of the above-mentioned earthquake phase prediction model is realized, or the above-mentioned earthquake phase prediction method is realized.

[0019] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0020] (1) The method for establishing an earthquake phase prediction model provided by an embodiment of the present invention obtains a plurality of first-category plane earthquake attribute data and a plurality of second-category plane earthquake attribute data that are screened and gridded according to the same rule; screens unlabeled grids from each plane earthquake attribute data according to a set rule, extracts each first-category earthquake attribute value and each second-category earthquake attribute value of the labeled grid and the screened unlabeled grid; obtains a first labeled sample set, a first unlabeled sample set, a second labeled sample set, and a second unlabeled sample set; and uses each obtained sample set to train a selected classification model through a collaborative training method to obtain a classification model for earthquake phase prediction. The classification model is trained using earthquake and well data using a collaborative training semi-supervised learning algorithm, thereby improving the correct judgment rate of the classification model; at the same time, it reduces the dependence on labeled samples and broadens its scope of application.

[0021] (2) The seismic facies prediction method provided by the embodiment of the present invention reduces the noise interference of seismic facies identification by collaboratively training a semi-supervised learning algorithm, enhances the continuity of the planar distribution of seismic facies prediction results, and significantly improves the effect of quantitative evaluation of sedimentary microfacies using seismic attributes.

[0022] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] 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:

[0024] Figure 1 This is a flow chart of the method for establishing an earthquake phase prediction model in Example 1 of the present invention;

[0025] Figure 2 for Figure 1 Specific implementation flow chart of step S11;

[0026] Figure 3for Figure 1 Specific implementation flow chart of step S13;

[0027] Figure 4 This is a flowchart of a specific implementation of the earthquake phase prediction method in the second embodiment of the present invention;

[0028] Figure 5-a Schematic diagram of thin section of tidal channel microfacies cast with grainstone;

[0029] Figure 5-b Schematic diagram of thin section of lagoonal microfacies cast with developed mudstone and weak dissolution;

[0030] Figure 5-c Schematic diagram of thin section of lagoonal microfacies cast with mudstone and cementation;

[0031] Figure 5-d Another schematic diagram of a thin section of a lagoonal microfacies cast showing the development of mudstone and cementation;

[0032] Figure 6-a Schematic diagram of the core with cross-bedding;

[0033] Figure 6-b Schematic diagram of the core with developed bioclastic particles;

[0034] Figure 6-c This is a schematic diagram of the core with developed scour surface;

[0035] Figure 6-d Schematic diagram of the core showing the development of biological wormhole structures;

[0036] Figure 7-a This is a schematic diagram of the logging facies of the tidal channel microfacies;

[0037] Figure 7-b This is a schematic diagram of the logging facies of the lagoon microfacies;

[0038] Figure 8-a This is the curve of the total correct rate of earthquake phase identification;

[0039] Figure 8-b This is a curve showing the change in the earthquake phase positive discrimination rate for tidal channels and lagoons;

[0040] Figure 9 Schematic diagram of the structure of the earthquake phase prediction model establishment device in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although 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. Rather, 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.

[0042] In order to solve the problems in the prior art that earthquake phase prediction has a strong dependence on earthquake data or the prediction results have low accuracy, the embodiments of the present invention provide an earthquake phase prediction model establishment, an earthquake phase prediction method and an apparatus, which can improve the prediction accuracy of earthquake phases and reduce dependence on labeled samples.

[0043] Example 1

[0044] The first embodiment of the present invention provides a method for establishing an earthquake phase prediction model, the process of which is as follows: Figure 1 As shown, the following steps are included:

[0045] Step S11: Data acquisition: acquiring a plurality of first-type plane seismic attribute data and a plurality of second-type plane seismic attribute data that are screened and gridded according to the same rule.

[0046] The grids in each plane seismic attribute data that match the well location information are seismically labeled according to the sedimentary microfacies type of the well's target layer.

[0047] For detailed methods of obtaining seismic attribute data of each plane, see Figure 2 As shown, the following steps may be included:

[0048] Step S111: acquiring multiple types of plane seismic attribute data at a target layer of a seismic data volume, and gridding each type of plane seismic attribute data according to the same rule.

[0049] In some embodiments, the extraction of planar seismic attribute data may include determining a target layer obtained through zero-phase wrapping constraint tracking interpretation of the target seismic event axis; converting the seismic data volume into a seismic attribute data volume of a set type, extracting seismic attributes at the target layer from the seismic attribute data volume, and obtaining planar seismic attribute data of this type.

[0050] This stratum-controlled seismic attribute extraction method uses zero-phase wrapping constraint tracking interpretation of the target seismic event to isolate the target seismic horizon from other seismic events for seismic attribute extraction. Compared to traditional methods that extract seismic attributes by tracing peaks and troughs, this stratum-controlled seismic attribute extraction method reduces the interference between stratigraphic interfaces of different periods, ensuring the accuracy of seismic attributes. It also captures global seismic attribute characteristics of the seismic event, rather than localized seismic slice attributes.

[0051] Step S112: determining a grid matching the well location in each plane seismic attribute data according to the well location information, and marking the grid with seismic facies according to the sedimentary microfacies type of the target layer of the well.

[0052] For example, if the sedimentary microfacies types of the well include water channels and non-water channels, the grid at the water channel can be marked as 1, and the grid at the non-water channel can be marked as 0.

[0053] Step S113: screening the planar seismic attribute data whose seismic attribute distribution and the marked earthquake match the set requirements.

[0054] In some embodiments, it may include, for each type of plane seismic attribute data obtained, determining multiple water channel segmentation values ​​according to its seismic attribute value, predicting the seismic phase type of each grid contained therein according to each water channel segmentation value, determining the total positive judgment rate of the predicted seismic phase type according to the marked seismic phase, and obtaining multiple total positive judgment rates; screening the plane seismic attribute data whose maximum total positive judgment rate is greater than the set total positive judgment rate threshold.

[0055] In some embodiments, it may also include, for each type of plane seismic attribute data obtained, determining multiple water channel segmentation values ​​according to its seismic attribute value, predicting the seismic phase type of each grid contained therein according to each water channel segmentation value, determining the total positive judgment rate and water channel positive judgment rate of the predicted seismic phase type according to the marked seismic phase, and obtaining multiple total positive judgment rates and water channel positive judgment rates; screening plane seismic attribute data whose maximum total positive judgment rate is greater than the set total positive judgment rate threshold and whose maximum water channel positive judgment rate is greater than the set water channel positive judgment rate threshold.

[0056] The screening of plane seismic attribute data not only considers the total correct judgment rate, but also the correct judgment rate of water channels, ensuring that the screened plane seismic attribute data can better reflect the distribution of water channels, making the final classification model have better water channel prediction ability.

[0057] Step S114: performing cluster analysis on the selected multiple planar seismic attribute data and dividing them into the first category and the second category.

[0058] In some embodiments, it may include determining the correlation coefficients between the multiple selected plane seismic attribute data through a hierarchical clustering operation; clustering the multiple selected plane seismic attribute data into two categories according to the correlation relationship, to obtain multiple first-category plane seismic attribute data and multiple second-category plane seismic attribute data.

[0059] Step S12: Sample set establishment: Filter unlabeled grids from each plane seismic attribute data according to the set rules, extract the first-category seismic attribute values ​​and the second-category seismic attribute values ​​of the labeled grid and the filtered unlabeled grid; use the seismic phase label and the first-category seismic attribute values ​​of the labeled grid as a labeled sample to obtain a first labeled sample set; use the first-category seismic attribute values ​​of the unlabeled grid as an unlabeled sample to obtain a first unlabeled sample set; use the seismic phase label and the second-category seismic attribute values ​​of the labeled grid as a labeled sample to obtain a second labeled sample set; use the second-category seismic attribute values ​​of the unlabeled grid as an unlabeled sample to obtain a second unlabeled sample set.

[0060] Step S13: Model training step: using the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set, the selected classification model is trained by a collaborative training method to obtain a classification model for earthquake phase prediction.

[0061] During the screening process of the above-mentioned plane seismic attribute data, the water channel segmentation value corresponding to the maximum total correct judgment rate of the screened plane seismic attribute data can be recorded simultaneously as the optimal water channel segmentation value.

[0062] Therefore, the specific implementation of step S13 may also include training the selected classification model through a collaborative training method using the optimal water channel segmentation value of the screened planar seismic attribute data, the first labeled sample set, the first unlabeled sample set, the second labeled sample set, and the second unlabeled sample set.

[0063] The selected classification model may be one of the following models:

[0064] Logistic regression classification model, decision tree classification model and nearest neighbor classification model.

[0065] For the specific model training process, see Figure 3 As shown, the following steps may be included:

[0066] Step S131: using the first labeled sample set to train the selected classification model to obtain a first classification model, and using the second labeled sample set to train the selected classification model to obtain a second classification model.

[0067] Specifically, the classification model trained using the first labeled sample set and the classification model trained using the second labeled sample set may be the same classification model or different classification models.

[0068] Step S132: selecting a set number of first unlabeled samples from the current first unlabeled sample set, and selecting a set number of second unlabeled samples with consistent grids from the current second unlabeled sample set.

[0069] A random selection method may be adopted. For example, when a set number of first unlabeled samples are selected from the current first unlabeled sample set, a selected identifier is set for the selected first unlabeled samples; and each selection is made only from unlabeled samples without the selected identifier.

[0070] Step S133: input the selected first unlabeled sample into the current first classification model, label the selected second unlabeled sample according to the output result, and then add it to the current second labeled sample set.

[0071] Since the grids of the selected first unlabeled sample and the second unlabeled sample are consistent, the samples with the consistent grids in the selected second unlabeled sample can be marked according to the output result.

[0072] Step S134: input the selected second unlabeled samples into the current second classification model, label the selected first unlabeled samples according to the output results, and then add them to the current first labeled sample set.

[0073] Step S133 and step S134 may be performed either one of them first, or both of them may be performed simultaneously.

[0074] Step S135: using the current first labeled sample set to train the current first classification model, and using the current second labeled sample set to train the current second classification model.

[0075] Step S136: Determine whether the prediction consistency between the current first classification model and the second classification model meets the set requirements.

[0076] If the answer in step S136 is yes, execute step S137 to end the model training; otherwise, return to execute step S132.

[0077] Step S137: End model training.

[0078] The method for establishing an earthquake phase prediction model provided in the first embodiment of the present invention obtains a plurality of first-category plane earthquake attribute data and a plurality of second-category plane earthquake attribute data that are screened and gridded according to the same rule; screens unlabeled grids from each plane earthquake attribute data according to a set rule, extracts each first-category earthquake attribute value and each second-category earthquake attribute value of the labeled grid and the screened unlabeled grid; obtains a first labeled sample set, a first unlabeled sample set, a second labeled sample set, and a second unlabeled sample set; and uses each obtained sample set to train a selected classification model through a collaborative training method to obtain a classification model for earthquake phase prediction. The classification model is trained using earthquake and well data using a collaborative training semi-supervised learning algorithm, thereby improving the correct judgment rate of the classification model; at the same time, it reduces dependence on labeled samples and broadens its scope of application.

[0079] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a method for predicting earthquake phases, comprising:

[0080] Multiple types of earthquake attribute data are input into a classification model, and earthquake phase prediction is performed based on the output results. The classification model is obtained using the above-mentioned earthquake phase prediction model establishment method, and the types are consistent with the types of planar earthquake attribute data screened when establishing the classification model.

[0081] The above-mentioned seismic facies prediction method reduces the noise interference of seismic facies identification through collaborative training of semi-supervised learning algorithms, enhances the continuity of the planar distribution of seismic facies prediction results, and significantly improves the effect of quantitative evaluation of sedimentary microfacies using seismic attributes.

[0082] In some embodiments, it may include inputting seismic attribute data consistent with the type of the first type of planar seismic attribute data into a first classification model to obtain a first output result; inputting seismic attribute data consistent with the type of the second type of planar seismic attribute data into a second classification model to obtain a second output result; and performing seismic phase prediction based on the first output result and the second output result.

[0083] The earthquake phase prediction may be performed based on only the first output result, or based on only the second output result, or with reference to both the first and second output results.

[0084] Example 2

[0085] The second embodiment of the present invention provides a specific implementation process of a seismic phase prediction method. Taking a carbonate reservoir as an example, the target layer in a certain study area mainly develops two sedimentary microfacies: tidal channel and lagoon. The specific seismic phase prediction, that is, the identification process of sedimentary microfacies, is as follows: Figure 4 As shown, the following steps are included:

[0086] The data preprocessing step includes the following steps S41 to S43:

[0087] Step S41: Use a small amount of thin sections and core data to calibrate the logging curve to identify the lithology and sedimentary microfacies, and use them as label data to identify the logging facies.

[0088] The target layer has tidal channel and lagoon microfacies. The tidal channel reservoir is mainly composed of granular limestone, which is well sorted and rounded, and the degree of bioclastic fragmentation is relatively large, reflecting strong hydrodynamic deposition. Dissolution is dominant in the tidal channel reservoir, and the pore type is mainly intergranular solution pores. In addition, secondary pores such as intragranular solution pores and mold pores are also developed (see Figure 5-a As shown). Cross-bedding ( Figure 6-a ), bioclastic particles ( Figure 6-b), tidal channel scour surface ( Figure 6-c ) and other sedimentary structures. The tidal channel reservoir shows low GR, high ILD, high AC and low density logging response characteristics ( Figure 7-a The lagoon reservoir is composed of muddy grainstone ( Figure 5-b ), granular mudstone ( Figure 5-c ) and mudstone ( Figure 5-d ) are mainly characterized by poor sorting and rounding, and a high content of mortar, reflecting a weak hydrodynamic sedimentary environment. The lagoon reservoir is mainly characterized by weak dissolution and cementation, and the pore types are mainly intergranular solution pores, mold pores and intercrystalline pores. Obvious biological wormholes can be seen on the core, indicating obvious bioturbation ( Figure 6-d The lagoon reservoir exhibits high GR, low resistivity, low acoustic time difference and high density logging response characteristics ( Figure 7-b ).

[0089] Figures 5-a to 5-d Thin-section images of tidal channels and lagoons: Figure 5-a Grainstone is visible in the middle, with strong dissolution, reflecting tidal channel microfacies; Figure 5-b In the middle, muddy grainstone is visible, weakly dissolved, and reflects lagoonal microfacies; Figure 5-c Granular mudstone is visible in the middle, indicating cementation and reflecting lagoonal microfacies; Figure 5-d Mudstone and cementation are visible in the core, reflecting the lagoon microfacies. Figures 6a to 6-d are the core facies markers of the tidal channel and lagoon microfacies: Figure 6-a Cross-bedding is visible in the middle, reflecting tidal channel microfacies; Figure 6-b Bioclastic particles can be seen in the water, reflecting the tidal channel microfacies; Figure 6-c In the middle, scour surface and tidal waterway can be seen; Figure 6-d Biological wormhole structures can be seen in it.

[0090] The identification of logging phases was completed based on the logging response characteristics of the above-mentioned tidal channel and lagoon reservoirs.

[0091] Step S42: Extracting plane seismic attribute data using the zero-phase wrapping method.

[0092] To obtain seismic attribute information for the target seismic event, the target seismic layer is isolated from other seismic events through zero-phase wrapping constraint tracking interpretation of the target seismic event. Compared with the traditional method of tracing peaks and troughs to extract seismic attributes for slices, the layer-controlled seismic attribute extraction method reduces the interference between information from different stratigraphic interfaces, ensuring the accuracy of seismic attributes. It also obtains global seismic attribute characteristics for the seismic event, rather than localized seismic slice attributes.

[0093] A total of 19 types of planar seismic attribute data of the target layer were extracted, including Maximum amplitude, Sum of amplitudes, Mean amplitude, Trace AGC, RMS amplitude, Average energy, Sum of magnitudes, Most of, Median, Average magnitude, Maximum magnitude, Sum of positive amplitudes, Struct Smooth, Sweeten, Average positive amplitude, Arclength, Minimum amplitude, Interval average Arithmetic and Extract value.

[0094] Step S43: calibrate the well logging data to planar seismic attribute data according to the geodetic coordinates, wherein the result of well point logging phase identification is used as labeled data, and the data without well points is used as unlabeled data.

[0095] A total of 257 well data were collected in the study area. Well data and seismic attribute data were linked based on their latitude and longitude coordinates, and labels and unlabeled data were annotated. Seismic attribute grid data for well-sited tidal channel sedimentary units were labeled as 1, those without tidal channel sedimentary units were labeled as 0, and those with no well-site matching seismic attribute grid data were labeled as -1. This study divided the 257 well data into training and validation sets. 70% of the wells were randomly selected as the training set for model training, and 30% of the wells were used as the validation set for model testing. Seismic attribute grid point data served as the data to be predicted.

[0096] The planar seismic attribute data optimization and grouping steps include the following steps S44 and S45:

[0097] Since collaborative training requires two sets of views, X1 and X2, and the optimization and grouping of seismic attributes are key factors in determining the effectiveness of seismic phase analysis, this work uses the sliding window method of seismic phases to perform statistics on the coincidence rate between well points and seismic attributes. For seismic attributes with higher coincidence rates, a hierarchical clustering algorithm based on correlation coefficients is used to perform cluster analysis of seismic attributes, and construct two views, X1 and X2.

[0098] Step S44: Optimizing planar seismic attribute data.

[0099] Different seismic attributes have different abilities to distinguish seismic phases. For the binary classification problem of tidal channels and lagoons, this embodiment adopts the sliding window method to perform statistical analysis of the well-seismic coincidence rate. First, the single plane seismic attribute data is divided into N equal parts (N can be 50), and the tidal channel microphase and lagoon microphase are segmented point by point and the classification accuracy is calculated. The accuracy of the seismic attribute and the well point data is determined based on the curve of the accuracy rate and the change of the seismic attribute segmentation point. The maximum value of the total accuracy rate is the matching rate between the seismic attribute and the well point data. The seismic attribute value corresponding to the maximum value of the total accuracy rate is the optimal segmentation value for seismic attribute seismic phase analysis and calculation (Figure 8a). Figure 8 is a statistical curve of the well-seismic coincidence rate of the Maximumamplitude seismic attribute sliding window method: Figure 8-a This is the curve of the total correct rate of earthquake phase identification; Figure 8-b The tidal channel and lagoon seismic phase accuracy curves are included. Curve A in Figure 8b shows the variation of the tidal channel accuracy and seismic attribute segmentation point, while Curve B shows the variation of the lagoon accuracy and seismic attribute segmentation point. The consistency of seismic attributes with wellpoint sedimentary phases is shown in Table 1. This embodiment prefers seismic attributes with an overall accuracy rate greater than 80% and a tidal channel accuracy rate greater than 30% as characteristic parameter data for seismic phase pattern recognition.

[0100] Table 1. Statistics of the coincidence rate between seismic attributes and well-point sedimentary facies

[0101]

[0102] Step S45: grouping the plane seismic attribute data using a hierarchical clustering algorithm based on correlation coefficients.

[0103] The hierarchical clustering method is a method of sequentially forming a cluster tree of data objects based on a certain similarity or dissimilarity coefficient as an indicator. Hierarchical clustering is an important method for establishing data structures between data and mining the correlation between different data points or data clusters. Conventional hierarchical clustering methods often use the distance between data points or data clusters as a clustering parameter indicator. The two points or two data clusters with the smallest distance are preferentially clustered and merged together, and clustering is performed in sequence until the conditions are met. In this embodiment, the correlation coefficient between each plane seismic attribute data is used instead of the distance as the clustering parameter indicator to perform a hierarchical clustering operation. The plane seismic attribute data with a large correlation coefficient are preferentially clustered together.

[0104] Using a hierarchical clustering algorithm based on correlation coefficients, the extracted planar seismic attribute data can be divided into two groups based on correlation coefficients. The first group includes: Maximum amplitude, Sum of amplitudes, Mean amplitude, Sum of positive amplitudes, Average positive amplitude, Minimum amplitude, Most of, and Median. The second group includes: Trace AGC, RMSamplitude, Average energy, Sum of magnitudes, Average magnitude, Maximum magnitude, Sweeten, Arclength, Interval average Arithmetic, and Extract value. These two groups of planar seismic attribute data can be used as two sets of characteristic parameters for tidal channel and lagoon seismic facies analysis and identification, and can be collaboratively trained.

[0105] The steps of collaboratively training the classification model training of the semi-supervised learning algorithm and the intelligent identification of tidal waterways are as follows: Steps S46 to S48:

[0106] Step S46: Collaborative training of the semi-supervised learning algorithm classification model training.

[0107] Co-training is an important paradigm in semi-supervised learning. It uses dual views to train two classifiers to label each other's samples to expand the training set, thereby improving learning performance with the help of unlabeled samples.

[0108] Basic principles of collaborative training algorithm:

[0109] Assume that the dataset attributes have two fully redundant views Figure 1 and 2 , let X1 and X2, then an example can be represented as (x1, x2), where x1 is the eigenvector of x in the view of X1, and x2 is the eigenvector of x in the view of X2. Assume that f is the target function in the example space X. If the label of x is l, then f(x) = f1(x1) = f2(x2) = l. A. Blum and T. Mitchell defined the so-called "compatibility", that is, for a certain distribution D on X, C1 and C2 are concept classes defined on X1 and X2 respectively. If D assigns zero probability to the example (x1, x2) that satisfies f1(x) ≠ f2(x2), then the target function f = (f1, f2) ∈ C1×C2 is said to be "compatible" with D.

[0110] The specific calculation process of the collaborative training semi-supervised learning algorithm is as follows:

[0111] 1) Use the X1 part of L to train a classifier h1;

[0112] 2) Use the X2 part of L to train a classifier h2;

[0113] 3) Use h1 to mark all elements in U', use the marking result of U' to mark U", and select p positive labels and n negative labels from U' and put them into X2;

[0114] 4) Use h2 to mark all elements in U", use the marking result of U" to mark U', and select p positive labels and n negative labels from U' and put them into X1;

[0115] 5) Use X1 to train classifier h1; use X2 to train classifier h2;

[0116] 6) Randomly select N data from U1 and add them to U', and randomly select N data from U2 and add them to U".

[0117] The loop iterates 3)-6) until the training is completed.

[0118] Step S47: Use the validation set to test the classification model.

[0119] In this second example, three classic classification algorithms, logistic regression, decision tree, and nearest neighbor algorithms, were used as classifiers for collaboratively training semi-supervised learning to identify patterns in tidal channels and lagoon microfacies. As shown in Tables 2 and 3, pattern recognition using logistic regression, decision tree, and nearest neighbor algorithms for two sets of seismic attributes revealed significant differences in the supervised learning results for the X1 and X2 seismic attributes, with a high positive detection rate. This satisfies the requirement for collaborative training, where each view learns from each other and the resulting classifier has a certain degree of accuracy.

[0120] Table 2. Statistics of seismic phase identification accuracy using supervised learning of earthquake attributes for groups X1 and X2

[0121]

[0122]

[0123] Table 3. Statistics of seismic phase identification accuracy using supervised learning of all seismic attributes and semi-supervised learning using collaborative training of grouped seismic attributes

[0124]

[0125] Step S48: using the classification model to predict the sedimentary microfacies of the target layer's non-well point corresponding grid.

[0126] The method of the above embodiment 2 is summarized as follows:

[0127] (1) The accuracy rate of seismic phase identification using a single seismic attribute using the sliding window method can reach up to 86.4%, but the accuracy rate of seismic phase identification in tidal waterways is generally low.

[0128] (2) The supervised learning and collaborative training semi-supervised learning algorithms using multiple seismic attributes based on decision tree and nearest neighbor algorithms significantly improved the accuracy of tidal channel reservoir identification.

[0129] (3) The total correct rate of multi-seismic attribute supervised learning can reach up to 95.3%, and the total correct rate after collaborative training can reach up to 96.6%.

[0130] (4) By comparing the logistic regression classifier, decision tree classifier and nearest neighbor classifier, it was found that different classifier algorithms have obvious differences in the accuracy and effect of earthquake phase identification, and the decision tree nearest neighbor algorithm classifier is better than the logistic regression classifier.

[0131] (5) Through comparative analysis of the results of tidal channel identification using supervised learning of X1 group seismic attributes, X2 group seismic attributes, and all seismic attributes and collaborative training semi-supervised learning algorithms, it was found that the use of collaborative training semi-supervised learning algorithms not only improved the positive judgment rate of seismic attribute tidal channel seismic phase identification, but also integrated the characteristics of X1 seismic attributes and X2 group seismic phase identification on the plane distribution of the seismic phase identification prediction results of the collaborative training semi-supervised learning algorithm, significantly reduced the background noise, and significantly enhanced the continuity of tidal channel seismic phase identification.

[0132] (6) After collaborative training and semi-supervised learning, the nearest neighbor classifier has the best test set performance, with a maximum positive detection rate of 88.5%, and the decision tree classifier has a maximum total positive detection rate of 96.5%.

[0133] A second embodiment of the present invention proposes a collaborative training semi-supervised learning algorithm that uses seismic and drilling data to identify tidal channel sedimentary microfacies patterns within carbonate reservoirs. Collaborative training is a semi-supervised machine learning algorithm. This algorithm divides the data feature set into two subsets, establishes a classifier on each feature subset, and retrains the classifier using the classification results of the other classifier until the two classifiers have the same classification results for the majority of the data. Taking the target layer in a study area as an example, this paper uses the phase wrapping method to extract 18 seismic attributes of the target layer, uses the sliding window method to select 12 seismic attributes, and uses a hierarchical clustering algorithm based on correlation coefficients to group the seismic attribute features. Using the two sets of seismic attributes as feature data, collaborative training semi-supervised learning is performed using logistic regression, decision tree, and nearest neighbor classifiers to classify and identify tidal channels and lagoons. The results show that the collaborative training semi-supervised learning algorithm not only improves the correctness of each classifier, but also reduces background noise and clutter, and enhances the continuity of the planar distribution of tidal channel prediction results. The collaborative training semi-supervised learning algorithm significantly improves the effectiveness of tidal channel seismic facies pattern recognition.

[0134] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a device for establishing an earthquake phase prediction model. The structure of the device is as follows: Figure 9 Shown, including:

[0135] Data acquisition module 91 is used to acquire a plurality of first-type planar seismic attribute data and a plurality of second-type planar seismic attribute data that have been screened and gridded according to the same rule. The grids in each planar seismic attribute data that match the well location information are seismically labeled according to the sedimentary microfacies type of the target layer of the well.

[0136] The sample set establishment module 92 is configured to screen unlabeled grids from the plane seismic attribute data according to a set rule, extract the first-category seismic attribute values ​​and the second-category seismic attribute values ​​of the labeled grids and the screened unlabeled grids; use the seismic phase label and the first-category seismic attribute values ​​of the labeled grid as a labeled sample to obtain a first labeled sample set; use the first-category seismic attribute values ​​of the unlabeled grid as an unlabeled sample to obtain a first unlabeled sample set; use the seismic phase label and the second-category seismic attribute values ​​of the labeled grid as a labeled sample to obtain a second labeled sample set; and use the second-category seismic attribute values ​​of the unlabeled grid as an unlabeled sample to obtain a second unlabeled sample set.

[0137] The model training module 93 is used to train the selected classification model by collaborative training using the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set to obtain a classification model for earthquake phase prediction.

[0138] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0139] Based on the inventive concept of the present invention, an embodiment of the present invention also provides a computer program product with earthquake phase prediction function, including a computer program / instruction, wherein when the computer program / instruction is executed by a processor, it implements the above-mentioned earthquake phase prediction model establishment method, or implements the above-mentioned earthquake phase prediction method.

[0140] It should be understood that the specific order or hierarchy of steps in the disclosed processes is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0141] In the foregoing detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that embodiments of the claimed subject matter require more features than are expressly recited in each claim. On the contrary, as reflected in the appended claims, the invention comprises less than all the features of any individual disclosed embodiment. The appended claims are therefore hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0142] The foregoing description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purposes of describing the above embodiments, but one of ordinary skill in the art will recognize that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to encompass all such changes, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent the term "comprising" is used in the specification or claims, the term is intended to be encompassed in a manner similar to the term "including," as explained in terms of "including," used as a transitional word in the claims. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or."

Claims

1. A method for establishing an earthquake phase prediction model, characterized in that: include: Data acquisition step: obtaining a plurality of first-type plane seismic attribute data and a plurality of second-type plane seismic attribute data that are screened and gridded according to the same rule, and performing seismic facies marking on the grids that match the well location information in each plane seismic attribute data according to the sedimentary microfacies type of the target layer of the well; The sample set establishment steps are as follows: unlabeled grids are screened from the seismic attribute data of each plane according to the set rules, and the first-category seismic attribute values ​​and the second-category seismic attribute values ​​of the labeled grids and the screened unlabeled grids are extracted; the seismic phase label and the first-category seismic attribute values ​​of the labeled grid are used as a labeled sample to obtain a first labeled sample set; the first-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a first unlabeled sample set; the seismic phase label and the second-category seismic attribute values ​​of the labeled grid are used as a labeled sample to obtain a second labeled sample set; the second-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a second unlabeled sample set; Model training step: using the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set, the selected classification model is trained by a collaborative training method to obtain a classification model for earthquake phase prediction.

2. The method according to claim 1, wherein The obtaining of the plurality of first-type plane seismic attribute data and the plurality of second-type plane seismic attribute data that are screened and gridded according to the same rule specifically includes: Acquire multiple types of plane seismic attribute data at the target layer of the seismic data volume, and grid each plane seismic attribute data according to the same rule; Determining a grid matching the well location in each plane seismic attribute data according to the well location information, and marking the grid with seismic facies according to the sedimentary microfacies type of the target layer of the well; Screening the plane seismic attribute data whose seismic attribute distribution and the marked earthquakes meet the set requirements; The selected various plane seismic attribute data are divided into the first and second categories after cluster analysis.

3. The method according to claim 2, wherein If the sedimentary microfacies types of the well include both water channels and non-water channels, the screening of plane seismic attribute data whose seismic attribute distribution and the marked seismic attributes meet the set requirements specifically includes: For each type of plane seismic attribute data obtained, multiple water channel segmentation values ​​are determined based on its seismic attribute values. Based on each water channel segmentation value, the seismic phase type of each grid contained therein is predicted. Based on the marked seismic phases, the total correct judgment rate of the predicted seismic phase type is determined to obtain multiple total correct judgment rates. Filter the plane seismic attribute data whose maximum total positive judgment rate is greater than the set total positive judgment rate threshold.

4. The method according to claim 2, wherein If the sedimentary microfacies types of the well include both water channels and non-water channels, the screening of plane seismic attribute data whose seismic attribute distribution and the marked seismic attributes meet the set requirements specifically includes: For each type of plane seismic attribute data obtained, multiple water channel segmentation values ​​are determined based on its seismic attribute values. Based on each water channel segmentation value, the seismic phase type of each grid contained therein is predicted. Based on the marked seismic phases, the total correct judgment rate of the predicted seismic phase type and the water channel correct judgment rate are determined, thereby obtaining multiple total correct judgment rates and water channel correct judgment rates. The plane seismic attribute data whose maximum total positive judgment rate is greater than the set total positive judgment rate threshold and whose maximum water channel positive judgment rate is greater than the set water channel positive judgment rate threshold are screened.

5. The method according to claim 3 or 4, wherein: After filtering the planar seismic attribute data, it also includes: The water channel segmentation value corresponding to the maximum total correct judgment rate of the screened planar seismic attribute data is obtained as the optimal water channel segmentation value.

6. The method according to claim 5, wherein The method of training the selected classification model by collaborative training using the first labeled sample set, the first unlabeled sample set, the second labeled sample set, and the second unlabeled sample set further includes: The selected classification model is trained by a collaborative training method using the optimal water channel segmentation value of the screened planar seismic attribute data, the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set.

7. The method according to claim 1, wherein The method of training the selected classification model by collaborative training using the first labeled sample set, the first unlabeled sample set, the second labeled sample set, and the second unlabeled sample set specifically includes: Using the first labeled sample set to train the selected classification model to obtain a first classification model, and using the second labeled sample set to train the selected classification model to obtain a second classification model; The following collaborative training steps are performed cyclically until the prediction accuracy of the current first classification model and the second classification model reaches the set requirements: A set number of first unlabeled samples are selected from the current first unlabeled sample set, and a set number of second unlabeled samples with consistent grids are selected from the current second unlabeled sample set; the selected first unlabeled samples are input into the current first classification model, and the selected second unlabeled samples are marked according to the output results and then added to the current second marked sample set; the selected second unlabeled samples are input into the current second classification model, and the selected first unlabeled samples are marked according to the output results and then added to the current first marked sample set; the current first classification model is trained using the current first marked sample set, and the current second classification model is trained using the current second marked sample set.

8. The method according to claim 2, wherein The method of obtaining various types of planar seismic attribute data at a target layer of a seismic data volume specifically includes: Determine the target layer obtained by zero-phase wrapping constraint tracking interpretation of the target seismic event; The seismic data volume is converted into a seismic attribute data volume of a set type, and the seismic attributes of the target layer are extracted from the seismic attribute data volume to obtain the plane seismic attribute data of the set type.

9. The method according to claim 2, wherein The screened multiple plane seismic attribute data are divided into the first category and the second category after cluster analysis, specifically including: The correlation coefficients between the selected multiple plane seismic attribute data are determined through hierarchical clustering operation; The screened multiple plane seismic attribute data are clustered into two categories according to the correlation coefficient, thereby obtaining multiple first-category plane seismic attribute data and multiple second-category plane seismic attribute data.

10. The method according to any one of claims 1 to 4 and 7 to 9, wherein The selected classification model is one of the following models: Logistic regression classification model, decision tree classification model and nearest neighbor classification model.

11. A method for predicting earthquake phases, characterized in that: include: Multiple types of seismic attribute data are input into a classification model, and seismic phase prediction is performed based on the output results. The classification model is obtained using the seismic phase prediction model establishment method described in any one of claims 1 to 10, and the type is consistent with the type of planar seismic attribute data screened when establishing the classification model.

12. The method according to claim 11, wherein The method of inputting various types of earthquake attribute data into the classification model and performing earthquake phase prediction based on the output results specifically includes: Inputting seismic attribute data consistent with the type of the first type of planar seismic attribute data into a first classification model to obtain a first output result; inputting the seismic attribute data consistent with the second type of planar seismic attribute data into the second classification model to obtain a second output result; An earthquake phase prediction is performed based on the first output result and the second output result.

13. A device for establishing an earthquake phase prediction model, characterized in that: include: A data acquisition module is used to obtain a plurality of first-type plane seismic attribute data and a plurality of second-type plane seismic attribute data that are screened and gridded according to the same rule, and the grids in each plane seismic attribute data that match the well location information are seismically labeled according to the sedimentary microfacies type of the target layer of the well; The sample set establishment module is used to screen unlabeled grids from the seismic attribute data of each plane according to a set rule, extract the first-category seismic attribute values ​​and the second-category seismic attribute values ​​of the labeled grid and the screened unlabeled grid; the seismic phase label of the labeled grid and the first-category seismic attribute values ​​are used as a labeled sample to obtain a first labeled sample set; the first-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a first unlabeled sample set; the seismic phase label of the labeled grid and the second-category seismic attribute values ​​are used as a labeled sample to obtain a second labeled sample set; the second-category seismic attribute values ​​of the unlabeled grid are used as an unlabeled sample to obtain a second unlabeled sample set; The model training module is used to train the selected classification model by using the first labeled sample set, the first unlabeled sample set, the second labeled sample set and the second unlabeled sample set through a collaborative training method to obtain a classification model for earthquake phase prediction.

14. A computer program product having earthquake phase prediction function, comprising a computer program / instructions, wherein: When the computer program / instruction is executed by a processor, the method for establishing an earthquake phase prediction model according to any one of claims 1 to 10 is implemented, or the earthquake phase prediction method according to claim 11 or 12 is implemented.

Images