Intelligent lithofacies prediction method based on seismic profile configuration Hu moment

Abstract

The present application provides a kind of facies intelligent prediction method based on seismic profile configuration Hu moment, which includes: step 1, to the 3D seismic data of study area, according to the seismic waveform Hu moment attribute of each seismic data of purpose layer is extracted;Step 2, according to each trace on 3D seismic plane as a sample point, for each sample point, select its Hu invariant moment data, form the sample data set of two-dimensional space;Step 3, the sample data set is divided into label sample set and prediction sample set;Step 4, select machine learning algorithm, learn label sample set to obtain training model, then use the model to predict prediction sample set, obtain the whole area facies distribution data.The facies intelligent prediction method based on seismic profile configuration Hu moment solves the problem of facies identification, and is of great significance for improving facies prediction accuracy and realizing rapid and effective exploration.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oil and gas exploration geophysical technology, and particularly to a lithofacies intelligent prediction method based on seismic profile configuration Hu matrix. BACKGROUND

[0002] In the field of seismic exploration, lithofacies identification is an important step in oil and gas exploration, and is of great significance to the success or failure of exploration. Lithofacies is a reflection of the paleohydrodynamic conditions of the formation process of each microfacies sand body based on the structural characteristics of rocks. In complex lithologic reservoirs, due to the difference in lithofacies, there can be a large difference in porosity and permeability, and the permeability can reach 2 orders of magnitude. Therefore, it is very important to divide the lithofacies of the reservoir. According to the rock composition, structure, structure and contact relationship between particles, considering the difference in rock physical properties, the rocks are divided into conglomerate facies, sandstone facies, siltstone facies and mudstone facies.

[0003] Using electrical logging curves to qualitatively interpret sedimentary environment and lithofacies is one of the conventional means of basin sedimentary facies analysis. It is of great significance to use the electrical logging curve response mode of sedimentary environment and lithofacies for artificial intelligence simulation identification. Because, in the absence of coring, a large amount of electrical logging data can be used to determine the vertical facies sequence of a single well with the help of a computer, and the lateral change of lithofacies can be inferred, which is used to find favorable sand body distribution areas in the process of oil exploration and to deploy exploration.

[0004] In oil and gas exploration, geological modeling technology can be used to analyze the geological structure and oil and gas storage conditions of the area to be developed. For example, by studying the internal structure of the geology and the sedimentary geometry, a reservoir geological model and a lithofacies geological model can be established to represent the reservoir and reservoir model. Through the reservoir geological model and the lithofacies geological model, the three-dimensional structure of the sedimentary body in the area to be developed can be analyzed, which plays an important role in determining the oil and gas development plan of the reservoir.

[0005] In the Chinese patent application No. CN201711126988.6, a reservoir lithofacies characterization method and device are provided. The method includes: establishing a configuration unit correlation profile and a lithofacies unit correlation profile according to the distribution information of wells in the area to be analyzed, the structural unit, and the lithofacies unit; picking up a configuration unit boundary control line and a lithofacies unit interface according to the configuration unit correlation profile, the lithofacies unit correlation profile, and a three-dimensional lithofacies unit correlation profile; constructing a three-dimensional configuration unit geological model according to the configuration unit interface and the three-dimensional configuration unit geological model; and constructing a three-dimensional lithofacies unit geological model according to the lithofacies unit interface and the three-dimensional configuration unit model. The three-dimensional distribution characteristics of the configuration unit and the lithofacies unit can be accurately depicted, complex calculation and weighted estimation are not required, the occurrence of isolated lithofacies is reduced, and the accuracy of geological structure analysis is improved. The patent application only applies traditional lithofacies data for well constraint, and uses traditional seismic forward modeling, inversion, seismic attribute, and other means to optimize and combine the process.

[0006] In the Chinese patent application No. CN201811105567.X, a fracture prediction method based on lithofacies configuration is provided. The main steps include: predicting related factors affecting fracture development, wherein the related factors include lithofacies configuration, organic carbon content of source rock, curvature, distance from fault, and rock fabric; performing correlation analysis of the related factors and fracture development degree according to imaging logging fracture density statistical data and statistical values of the related factors, and determining the main geological control factors affecting fracture development; and establishing a multi-parameter fracture density calculation model based on lithofacies configuration according to the obtained main geological control factors, to predict fractures. The invention fully considers that the fracture development degree of a dense sandstone reservoir is controlled by tectonic and non-tectonic factors, takes the prediction of related factors affecting fracture development as the main line, takes the correlation of each related factor and fracture development as the key, obtains the main geological control factors, and establishes a multi-parameter fracture density calculation model, to realize fracture prediction of a dense sandstone reservoir. The purpose of the invention is to overcome the problems in the prior art that only tectonic factors are considered in the study of fracture development degree, no fracture quantitative prediction model based on geological analysis is established, the longitudinal and lateral distribution of fractures cannot be accurately predicted, the oil and gas exploration and development cannot be effectively known, and only a fracture prediction method based on lithofacies configuration is provided. The patent application solves the problem of fracture prediction of a dense sandstone reservoir by calculating and modeling based on lithofacies configuration through fracture related factors.

[0007] In the Chinese patent application No. CN202011233656.X, a sand and gravel rock facies prediction method under the control of an isochronous stratigraphic model is disclosed, which comprises the following steps: calibrating the well-seismic relationship according to well-seismic spectrum analysis, Fischer curve and coupling coefficient; obtaining intrinsic mode function components by EMD decomposition to realize reconstruction of the seismic profile; establishing a preliminary three-dimensional virtual grid; forming a node grid and constraining the basic object unit with a related threshold value; calculating the correlation index between object units by using the application value function, connecting the object units with high correlation index to obtain a stratigraphic framework, and then refining and interpolating to form a three-dimensional isochronous stratigraphic model according to the inheritance of stratigraphic deposition; obtaining isochronous stratigraphic slices according to the isochronous stratigraphic model, and predicting the sand and gravel rock facies through the isochronous stratigraphic slices.

[0008] The above prior art is quite different from the present application, and cannot solve the technical problems we want to solve. Therefore, we have invented a new rock facies intelligent prediction method based on seismic profile configuration Hu moment. SUMMARY

[0009] The purpose of the present application is to provide an estimation model between rock facies types and thickness and seismic waveform Hu moment, which greatly improves the speed and accuracy of rock facies type and thickness prediction, and a fast, efficient, reliable and practical rock facies intelligent prediction method based on seismic profile configuration Hu moment

[0010] The purpose of the present application can be achieved by the following technical measures: a rock facies intelligent prediction method based on seismic profile configuration Hu moment, which comprises the following steps: step 1, extracting the seismic waveform Hu moment attribute of each seismic data according to the purpose layer of the three-dimensional seismic data of the study area; step 2, taking each trace on the three-dimensional seismic plane as a sample point, selecting the Hu moment data of each sample point to form a sample data set in two-dimensional space; step 3, dividing the sample data set into a labeled sample set and a prediction sample set; step 4, selecting a machine learning algorithm to learn the labeled sample set to obtain a training model, and then using the model to predict the prediction sample set to obtain the rock facies distribution data of the whole area.

[0011] The purpose of the present application can also be achieved by the following technical measures:

[0012] In step 1, the determination of the purpose layer time window is based on the rock facies division sequence established by well-seismic combination; the seismic data in the time window is extracted by using the seismic interpretation horizon to determine the purpose layer time window, and the seismic waveform Hu moment attribute is calculated.

[0013] In step 1, the Hu moment attributes calculated for the seismic data in the time window include 6 absolute orthogonal invariants and 1 oblique orthogonal invariant, which are not affected by position, size and direction, and are not affected by parallel projection.

[0014] In step 2, in the three-dimensional seismic plane, the X and Y values of each point are selected as the coordinates of the sample point, and the seismic waveform Hu moment invariant attribute data formed in step 1 is taken as the characteristic value of the sample point to generate a sample data set file.

[0015] In step 3, the facies distribution data of each well in the target layer is analyzed and counted using the logging, mud logging, and core data of the known wells in the study area, and is used as a label value of the sample data near the well position to label the sample, thereby dividing the sample data set into a labeled sample set and a prediction sample set.

[0016] In step 3, in the segmentation process, in addition to labeling the samples near the well, the samples within a certain range near the well are also labeled, and this range is defined by specifying a fixed radius parameter or by a polygon range circled by a professional, and the capacity of the labeled sample set is expanded by using this range.

[0017] In step 3, the label value includes facies classification data in the target layer; the sample expansion method includes a fixed radius parameter or a polygon range circled by a professional, or a Gaussian noise or a generative adversarial network algorithm.

[0018] In step 4, based on the labeled sample set, a machine learning algorithm is used to establish an estimation model between the seismic Hu moment invariant representation parameters and the facies types and thicknesses, and a validation set of the labeled data set is used to evaluate the accuracy of the model, and finally a qualified model is used for intelligent prediction of unknown facies to obtain the facies distribution data of the whole area.

[0019] In step 4, multiple machine learning algorithms are selected to train the labeled sample set to obtain a training model, and these machine learning algorithms include multiple linear regression algorithm, nonlinear regression algorithm, ridge regression algorithm, random forest algorithm, extreme gradient boosting algorithm, support vector machine algorithm, convolutional neural network algorithm, and recurrent neural network algorithm.

[0020] In step 4, the labeled sample set is divided into a training sample set and a validation sample set during the model training process, wherein the training sample set is used for training the model, and the validation sample set is used for evaluating whether the model is qualified, and the qualified standard is comprehensively judged according to the loss function of the training sample and the fitting rate of the validation sample set.

[0021] In step 4, for the model that does not meet the evaluation standard, the parameters are automatically optimized and adjusted based on the parameters of the algorithm, and the iteration is repeated until the model meets the qualified standard.

[0022] The present application aims at the defects that an intelligent lithofacies prediction method based on seismic profile configuration has not been formed in the prior art seismic lithofacies prediction technology, and the prediction result accuracy is still low, and provides a lithofacies intelligent prediction method based on seismic waveform Hu moment.

[0023] The present application automatically establishes an estimation model between the lithofacies types and thickness and the seismic waveform Hu invariant moment based on a plurality of machine learning algorithms, greatly improves the speed and accuracy of lithofacies type and thickness prediction, is fast, efficient, reliable and practical. The present application can be applied in sandstone and gravel reservoir three-dimensional exploration, and can also be applied in target body three-dimensional exploration, solves the problem of reservoir lithofacies identification, and has important significance for improving reservoir lithofacies prediction accuracy and realizing rapid and effective exploration. The lithofacies intelligent prediction method extracts Hu invariant moment attributes of a target layer of three-dimensional seismic data as a characterization parameter to form a test sample data set; uses lithofacies classification data of a known well as a label to mark seismic trace samples within a certain distance from the well, divides the above sample set into label samples and prediction samples; based on the label sample set, uses a plurality of machine learning algorithms to automatically establish an estimation model between the seismic characterization parameter and the formation lithofacies, simultaneously uses a verification set split from the label data set to evaluate the accuracy of the model, and finally uses the qualified evaluation model to predict unknown formation lithofacies, and the prediction result has high accuracy. The present application solves the problem of lithofacies identification, and has important significance for improving lithofacies prediction accuracy and realizing rapid and effective exploration. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A flowchart of a specific embodiment of the lithofacies intelligent prediction method based on seismic profile configuration Hu moment of the present application;

[0025] Figure 2 A schematic diagram of seismic reflection characteristics and a target layer section of a research area according to an embodiment of the present application;

[0026] Figure 3 A schematic diagram of Hu moment 1 attribute plane distribution of a target layer section of a research area according to an embodiment of the present application;

[0027] Figure 4 A schematic diagram of Hu moment 2 attribute plane distribution of a target layer section of a research area according to an embodiment of the present application;

[0028] Figure 5 Fig. 3 is a schematic diagram of a planar distribution of a Hu-moment 3 attribute for a target zone of a study area according to an embodiment of the present application;

[0029] Figure 6 Fig. 4 is a schematic diagram of a planar distribution of a Hu-moment 4 attribute for a target zone of a study area according to an embodiment of the present application;

[0030] Figure 7 Fig. 5 is a schematic diagram of a planar distribution of a Hu-moment 5 attribute for a target zone of a study area according to an embodiment of the present application;

[0031] Figure 8 Fig. 6 is a schematic diagram of a planar distribution of a Hu-moment 6 attribute for a target zone of a study area according to an embodiment of the present application;

[0032] Figure 9 Fig. 7 is a schematic diagram of a planar distribution of a Hu-moment 7 attribute for a target zone of a study area according to an embodiment of the present application;

[0033] Figure 10 Fig. 8 is a schematic diagram of a planar distribution of a fusion analysis of Hu-moments 1, 2, 3, 4 attributes for a target zone of a study area according to an embodiment of the present application;

[0034] Figure 11 Fig. 9 is a schematic diagram of a planar distribution of a Hu-moment prediction of lithofacies for a target zone of a study area according to an embodiment of the present application;

[0035] Figure 12 Fig. 10 is a schematic diagram of a planar distribution of a Hu-moment 1 attribute for a target zone of a study area according to embodiment 2 of the present application;

[0036] Figure 13 Fig. 11 is a schematic diagram of a planar distribution of a Hu-moment 2 attribute for a target zone of a study area according to embodiment 2 of the present application;

[0037] Figure 14 Fig. 12 is a schematic diagram of a planar distribution of a Hu-moment 3 attribute for a target zone of a study area according to embodiment 2 of the present application;

[0038] Figure 15 Fig. 13 is a schematic diagram of a planar distribution of a Hu-moment 4 attribute for a target zone of a study area according to embodiment 2 of the present application;

[0039] Figure 16 Fig. 14 is a schematic diagram of a planar distribution of a Hu-moment 5 attribute for a target zone of a study area according to embodiment 2 of the present application;

[0040] Figure 17 Fig. 15 is a schematic diagram of a planar distribution of a Hu-moment 6 attribute for a target zone of a study area according to embodiment 2 of the present application;

[0041] Figure 18The Hu-moment 7 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 2 of the present application is shown in the figure;

[0042] Figure 19 The Hu-moment 1, 2, 3, 4 attribute fusion analysis plane distribution schematic diagram of the target layer section of the research area according to the embodiment 2 of the present application is shown in the figure;

[0043] Figure 20 The Hu-moment predicted lithofacies plane distribution schematic diagram of the target layer section of the research area according to the embodiment 2 of the present application is shown in the figure;

[0044] Figure 21 The Hu-moment 1 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0045] Figure 22 The Hu-moment 2 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0046] Figure 23 The Hu-moment 3 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0047] Figure 24 The Hu-moment 4 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0048] Figure 25 The Hu-moment 5 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0049] Figure 26 The Hu-moment 6 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0050] Figure 27 The Hu-moment 7 attribute plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0051] Figure 28 The Hu-moment 1, 2, 3, 4 attribute fusion analysis plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure;

[0052] Figure 29 The Hu-moment predicted lithofacies plane distribution schematic diagram of the target layer section of the research area according to the embodiment 3 of the present application is shown in the figure. DETAILED DESCRIPTION

[0053] In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the preferred embodiments are specifically described below with reference to the drawings.

[0054] The lithofacies intelligent prediction method based on the Hu-moment of the seismic profile configuration according to the present application comprises the following steps:

[0055] Step 1, for the 3D seismic data of the study area, the seismic waveform Hu moment attribute of each seismic data is extracted according to the target layer. Hu moment has the characteristics of keeping invariant in translation, rotation and scaling, and can well reflect the configuration of the seismic profile.

[0056] The determination of the target layer time window is based on the lithofacies division sequence established by well-seismic combination. The seismic waveform Hu invariant moment attribute is calculated for the seismic data in the time window.

[0057] The Hu moment attributes calculated for the seismic data in the time window include 6 absolute orthogonal invariants and 1 oblique orthogonal invariant, which are not affected by position, size and direction, and are not affected by parallel projection. Assuming that the image has no noise and is continuous, the invariants of moments have been proved to be effective for translation, scaling and rotation of the image. The invariants of moments have been widely applied to image pattern recognition, image registration and image reconstruction.

[0058] Step 2, taking each trace on the 3D seismic plane as a sample point, for each sample point, the Hu invariant moment data is selected to form a sample data set in two-dimensional space.

[0059] In the 3D seismic plane, the X and Y values of each point are selected as the coordinates of the sample point, and the seismic waveform Hu invariant moment attribute data formed in step 1 is taken as the characteristic value of the sample point to generate a sample data set file.

[0060] Step 3, using the logging, mud logging and core data of the known wells in the study area, the lithofacies distribution data of each well in the target layer is analyzed and counted, which is used as the label value of the sample data near the well position to label the sample, so as to divide the sample data set into a labeled sample set and a prediction sample set. In the segmentation process, in addition to labeling the samples near the well, the samples within a certain range near the well can also be labeled. This range can be defined by specifying a fixed radius parameter or by a polygon range circled by professionals. By using this range, the capacity of the labeled sample set can be expanded. Since most intelligent algorithms require that the labeled samples meet certain capacity requirements, it is necessary to expand the above-mentioned labeled sample set in the study area with few known wells.

[0061] The method of labeling and expanding the sample data near the well is described, and the label value includes but is not limited to the lithofacies classification data in the target layer. The sample expansion method includes but is not limited to a fixed radius parameter or a polygon range circled by professionals, and can also be performed by Gaussian noise and a generative adversarial network algorithm.

[0062] Step 4: Based on the labeled sample set, several machine learning algorithms are used to automatically establish an estimation model between the seismic Hu invariant moment characterization parameters and lithofacies types and thicknesses. At the same time, the accuracy of the model is evaluated using the split validation set of the labeled dataset. Finally, the qualified model is used to make intelligent predictions of unknown lithofacies and obtain lithofacies distribution data for the entire region.

[0063] Several machine learning algorithms will be selected to train the labeled sample set to obtain a training model. These algorithms include, but are not limited to, multiple linear regression, nonlinear regression, ridge regression, random forest, extreme gradient boosting, support vector machine, convolutional neural network, and recurrent neural network. During model training, the labeled sample set will be divided into a training sample set and a validation sample set. The training sample set is used to train the model, while the validation sample set is used to evaluate the model's suitability. The pass / fail criterion is a comprehensive evaluation based on the loss function of the training samples and the consistency rate with the validation sample set. For models that fail the evaluation, the parameters will be automatically optimized and adjusted based on the parameters of the algorithms used, iterating repeatedly until the model meets the pass / fail criterion.

[0064] In a specific embodiment 1 of the present invention, such as Figure 1 As shown, Figure 1 This is a flowchart of a specific embodiment of the lithofacies intelligent prediction method based on seismic profile configuration Hu moment of the present invention. At the beginning of the process, two horizons are first used as top and bottom time windows (e.g., Figure 2 The Hu moment attribute of seismic data within the time window is calculated as a feature parameter according to the seismic data trace, and a sample dataset is formed according to the trace. On this basis, the well-side data is labeled using lithofacies classification parameters statistically based on wells. Artificial intelligence algorithms are used to train and predict the sample dataset to identify the lithofacies distribution in the study area.

[0065] S101: Based on the well-seismic calibration results in the 3D seismic data, determine the top and bottom interfaces of the target layer in the study block, and use these interfaces as the target layer segment of the study block. Figure 2 This represents the reflection of the target stratum in a cross-section of a certain study area. Using the top and bottom interfaces as the extraction time window, the seismic waveform Hu moment attribute of this stratum was extracted in this embodiment.

[0066] Figures 3-9 This embodiment demonstrates the seven types of seismic waveform Hu moment attributes extracted. Figure 10 The diagram shows the planar distribution of Hu moment 1, 2, 3, and 4 attribute fusion analysis in this embodiment.

[0067] S102: Treat each trace on the three-dimensional seismic plane as a sample point, and select the Hu moment attribute of each sample point to form a sample dataset in two-dimensional space.

[0068] S103: Divide the above sample data set into a label sample set and a prediction sample set according to a certain distance beside the well, wherein the label sample set adds a column of label values, and the label values for sample labeling include but are not limited to lithofacies classification data in a target interval. For the method of label sample expansion, the method includes but is not limited to being performed by using a fixed radius parameter or a polygon range circled by a professional, and the method can also be performed by using a Gaussian noise or a generative adversarial network algorithm.

[0069] In the embodiment, the lithofacies classification is used as the label value, and the label value of each well covers the sample data within a range of 50 m beside the well.

[0070] S104: Select a machine learning algorithm to learn the label sample set to obtain a training model, and use the model to predict the prediction sample set to obtain lithofacies distribution data of the whole area.

[0071] In the embodiment, a convolutional neural network algorithm is selected to perform training and prediction. Considering that the convolutional neural network algorithm requires a large amount of training samples, the label sample set is expanded by 5 times by using a Gaussian noise expansion method while the sample expansion is performed beside the well.

[0072] Embodiment effect analysis: Figure 11 The lithofacies distribution map of the research area predicted in the embodiment is shown in the following table, and the predicted lithofacies classification result conforms to the geological deposition rule of the research area. Comparative analysis is performed on 10 verification wells, and the coincidence rate reaches 90%. The comparative table of the Hu matrix lithofacies prediction effect analysis of the target interval of the research area is shown in the following table.

[0073] Table 1 Comparative table of Hu matrix lithofacies prediction effect analysis of the target interval of the research area

[0074] Hash Lithofacies Predicted lithofacies Agreement W1 Conglomeratic sandstone Conglomeratic sandstone Yes W2 Conglomerate Conglomerate Yes W3 Conglomerate Conglomerate Yes W4 Conglomeratic sandstone Sandstone No W5 Sandstone Sandstone Yes W6 Conglomerate Conglomerate Yes W7 Shale Shale Yes W8 Conglomerate Conglomerate Yes W9 Sandstone Conglomeratic sandstone No W10 Sandstone Sandstone Yes

[0075] Embodiment 2

[0076] In the specific embodiment 2 of the application, the research area is different from other embodiments, the steps and methods used are basically the same as those of other embodiments, but in the S104 step, the support vector machine algorithm is selected to perform training and prediction in the embodiment, and only the data within 150 m beside the well is used for sample expansion of the label sample set.

[0077] Conglomerate The Hu matrix attributes of the seven types of seismic waveforms in the embodiment are shown in the following table, Conglomerate The plane distribution diagram of the Hu matrix 1, 2, 3 and 4 attribute fusion analysis in the embodiment is shown in the following figure, Yes The lithofacies distribution map of the research area predicted in the embodiment is shown in the following table, and the predicted lithofacies classification result conforms to the geological deposition rule of the research area. Comparative analysis is performed on 8 verification wells, and the coincidence rate reaches 87%. The comparative table of the Hu matrix lithofacies prediction effect analysis of the target interval of the research area is shown in the following table.

[0078] Table 2 Comparative table of effect analysis of Hu moment facies prediction in target layer section of research area in example 2

[0079] Conglomeratic sandstone Conglomeratic sandstone Yes Sandstone W1 Sandstone Yes Conglomeratic sandstone W2 Conglomeratic sandstone Yes Figures 21-27 W3 Figure 28 Figure 29 ​ W4 ​ ​ ​ W5 ​ ​ ​ W6 ​ ​ ​ W7 ​ ​ ​ W8 ​ ​ ​

[0080] Example 3:

[0081] In the specific example 3 of the application, the research area is different from other examples, the steps and methods used are basically the same as those of other examples, but in the step S104, the random forest algorithm is selected in this example to train and predict, and the well side 150m data is used for sample expansion.

[0082] ​ For the Hu moment attribute of 7 types of seismic waveforms in this example, ​ The Hu moment 1, 2, 3 and 4 attribute fusion analysis plane distribution diagram in this example is shown, ​ The facies distribution map of the research area predicted in this example, the predicted facies classification result conforms to the geological sedimentary rule of the research area, and the comparative analysis of 17 verification wells is carried out, the coincidence rate reaches 88%, and the comparative table of effect analysis of Hu moment facies prediction in target layer section of research area in this example is shown as follows:

[0083] Table 3 Comparative table of effect analysis of Hu moment facies prediction in target layer section of research area in example 2

[0084]

[0085]

[0086] The facies intelligent prediction method based on the Hu moment of seismic profile configuration of the application extracts the Hu moment attribute of the seismic waveform in the time window of the seismic data, realizes the intelligent prediction of the facies through the supervised machine learning algorithm. Compared with the prior art, the application solves the problem that the reservoir facies distribution is complex and difficult to identify by seismic, improves the prediction accuracy of the reservoir facies, and is beneficial to the three-dimensional exploration of the target body.

[0087] The above only describes the preferred embodiments of the application, and does not limit the application, any modification, equivalent replacement and improvement made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A lithofacies intelligent prediction method based on seismic profile configuration Hu moment, characterized in that, This lithofacies intelligent prediction method based on seismic profile configuration Hu moment includes: Step 1: Extract the seismic waveform Hu moment attribute of each seismic data source for the study area according to the target layer from the three-dimensional seismic data. Step 2: Take each trace on the three-dimensional seismic plane as a sample point, and select its Hu invariant moment data for each sample point to form a sample dataset in two-dimensional space. Step 3: Split the sample dataset into a labeled sample set and a predicted sample set; Step 4: Select a machine learning algorithm, learn from the labeled sample set to obtain a training model, and then use the model to predict the prediction sample set to obtain the lithofacies distribution data of the target layer in the whole area; apply the training model to the full three-dimensional prediction to obtain the three-dimensional lithofacies data of the whole area. In step 1, the determination of the target layer time window is based on the lithofacies stratigraphic sequence established by combining well and seismic analysis; the target layer time window is determined by seismic interpretation of the stratigraphic position, the seismic data within the time window is extracted, and the Hu invariant moment property of the seismic waveform is calculated. In step 1, the Hu moment properties calculated for the seismic data within the time window include 6 absolutely orthogonal invariants and 1 oblique orthogonal invariant, which are unaffected by position, magnitude, and orientation, as well as by parallel projection. In step 2, the X and Y values ​​of each point in the three-dimensional seismic plane are selected as the coordinates of the sample point. The seismic waveform Hu invariant moment attribute data formed in step 1 are used as the feature values ​​of the sample points to generate a sample dataset file. In step 3, using the logging, well logging, and core data of known wells in the study area, the lithofacies distribution data of each well in the target interval are analyzed and statistically analyzed. These data are used as the label values ​​of the sample data near the well location to label the samples, thereby dividing the sample dataset into a labeled sample set and a predicted sample set. In step 4, based on the labeled sample set, a machine learning algorithm is used to establish an estimation model between the seismic Hu invariant moment characterization parameters and lithofacies types and thicknesses. At the same time, the accuracy of the model is evaluated using the split validation set of the labeled dataset. Finally, the qualified model is used to perform intelligent prediction of three-dimensional unknown lithofacies and obtain lithofacies distribution data for the entire region. In step 4, multiple machine learning algorithms are selected to train the labeled sample set to obtain a training model. These machine learning algorithms include multiple linear regression, nonlinear regression, ridge regression, random forest, extreme gradient boosting, support vector machine, convolutional neural network, and recurrent neural network.

2. The lithofacies intelligent prediction method based on seismic profile configuration Hu moment according to claim 1, characterized in that, In step 3, during the segmentation process, in addition to labeling the samples near the well, the samples within a certain range near the well are also labeled. This range is defined by specifying a fixed radius parameter or by a polygon range delineated by professionals. This range is used to expand the capacity of the labeled sample set.

3. The lithofacies intelligent prediction method based on seismic profile configuration Hu moment according to claim 2, characterized in that, In step 3, the label values ​​include lithofacies classification data within the target stratum; the sample expansion methods include using a fixed radius parameter or a polygon range defined by professionals, or using Gaussian noise or generative adversarial network algorithms.

4. The lithofacies intelligent prediction method based on seismic profile configuration Hu moment according to claim 1, characterized in that, In step 4, during the model training process, the label sample set is divided into a training sample set and a validation sample set. The training sample set is used to train the model, and the validation sample set is used to evaluate whether the model is qualified. The qualification standard is judged comprehensively based on the loss function of the training sample set and the consistency rate of the validation sample set.

5. The lithofacies intelligent prediction method based on seismic profile configuration Hu moment according to claim 4, characterized in that, In step 4, for models that fail the evaluation, the parameters are automatically optimized and adjusted based on the parameters of the algorithm used, and the process is repeated until the model meets the qualification criteria.

Images