Machine learning-based automatic lithofacies identification method based on imaging gravel content

Abstract

The present invention provides a machine learning-based automatic lithofacies identification method based on imaging gravel content. The machine learning-based automatic lithofacies identification method based on imaging gravel content includes: Step 1, observing and describing the core of the cored well and conducting four-property relationship analysis; Step 2, determining the conventional and imaging logging response patterns of different lithofacies of glutenite; Step 3, through imaging image processing, finely depicting the gravel content and gravel size distribution, and quantitatively dividing the lithofacies of the entire well section; Step 4, based on the lithofacies determined by imaging logging, screening the electrical layers with stable logging characteristics as sample layers; Step 5, calibrating the conventional logging curves through imaging to determine the sensitive curves; Step 6, based on the machine learning algorithm, using the lithofacies of the sample layer as the supervised object to automatically identify the lithofacies. The machine learning-based automatic lithofacies identification method based on imaging gravel content improves the accuracy and efficiency of glutenite lithofacies identification, achieving the purpose of quantitative, fine, objective, and high-efficiency lithofacies identification.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration and development, and particularly to a machine learning-based automatic lithofacies identification method based on imaging gravel content. Background Art

[0002] In the steep slope zone of continental faulted lacustrine basins, glutenite fans of various origins are developed, which are characterized by deep burial, large thickness, large reserves, great potential, and low production degree, and have become an important potential growth point for increasing reserves and production in Shengli Oilfield at present. Glutenite fans are usually rapidly deposited near the source, vertically stacked in multiple periods, and swing and migrate zonally in the plane, with a complex sedimentation pattern. Glutenite bodies have strong heterogeneity both vertically and horizontally, with complex and diverse rock types and rapid lithofacies changes; the gravel skeleton has a great influence on the logging response, and the rock-electricity relationship is complex. At the same time, due to the low resolution of conventional logging, the response of thin interlayers is not obvious, resulting in low accuracy of identifying glutenite lithofacies using conventional logging evaluation methods, low accuracy of parameter interpretation, and difficulty in accurately identifying oil and water layers, which has become one of the bottlenecks restricting the effective production of glutenite reservoirs. Lithofacies identification is the basis for the evaluation of glutenite reservoirs. Accurate lithofacies identification is of great significance for accurately obtaining physical property parameters of glutenite reservoirs, identifying effective reservoirs, and even grasping sedimentation laws and the distribution laws of favorable reservoirs. Therefore, this patent invented a machine learning-based automatic lithofacies identification method based on imaging gravel content calculation to effectively improve the lithofacies identification accuracy and provide a reliable basis for the logging evaluation research of glutenite reservoirs.

[0003] Micro-resistivity scanning imaging logging is a response that reflects the resistivity change near the wellbore wall, with high vertical resolution (0.2 in), having an intuitive visual effect and being able to directly reflect the internal structure of glutenite bodies. Correct analysis can be made on the lithology of glutenite, the change of sedimentary grain sequence, the relative size of gravel particles, etc. from the imaging map. After calibration with core, it can reflect the wellbore wall lithology information, solving the problems of small information volume and low vertical resolution (20 - 80 cm) of conventional logging curves, and having the advantage of being more intuitive and reliable for describing reservoir characteristics than conventional logging data. However, at present, glutenite imaging logging is mostly used qualitatively. In this study, the information of imaging data is deeply mined. Based on static imaging data, through core calibration, the imaging image display tone range is used to represent different lithofacies types, bright colors represent high resistivity, dark colors represent low resistivity, and the distribution area of different components is used as the lithofacies percentage content to calculate the gravel content and achieve quantitative lithofacies division.

[0004] With the rapid development of artificial intelligence, based on big data and using machine learning algorithms as a means, the method of rapidly and automatically identifying lithology by quickly analyzing data, training and learning has become the general trend. Due to the high cost of imaging logging, only 1-2 wells in a block are used for imaging logging, and only some well sections of the coring wells are cored. Therefore, the work of identifying lithofacies in the whole area still needs to focus on the application of conventional logging curves.

[0005] In the Chinese patent application with the application number: CN201910078669.5, it involves a multi-well complex lithology intelligent identification method and system based on logging data. This method first determines the target logging data file and performs format conversion and normalization preprocessing. Then, based on the known lithology of the key coring wells in the whole area in the cored well sections, it screens and / or combines and expands the logging curve data to obtain logging curve data sensitive to lithology. Then, it tags and calibrates the logging curve data sensitive to lithology response to form a sample database, and at the same time, forms a database to be tested with the untagged logging curve data in the whole area. Furthermore, it uses the data in the sample database and combines several machine learning algorithms to perform machine learning training and then automatically establishes several lithology identification models.

[0006] In the Chinese patent application with the application number: CN201911190561.1, it involves a method, device, computer equipment and storage medium for constructing a core saturation prediction model. It obtains sample logging curves and their corresponding sample core saturations. Among them, the sample logging curves are obtained after depth correction of the original logging curves, which can improve the accuracy of the obtained sample data. During the model training process, it adopts a combination of training and testing. It divides the sample data into a training sample set and a testing sample set, trains multiple machine learning models with the training sample set, and then tests the trained machine learning models with the testing sample set. It selects the optimal trained machine learning model as the final core saturation prediction model, and the finally obtained core saturation prediction model can support the accurate prediction of the subsequent core saturation.

[0007] In the Chinese patent application with the application number: CN201910440723.6, it involves a quantitative prediction method for continental hydrocarbon source rocks based on seismic inversion and machine learning, which is used to predict the spatial distribution and organic matter content of continental hydrocarbon source rocks in a certain area. It first optimizes the elastic attributes sensitive to the lithology distinction and organic matter content of continental sedimentary strata, then trains a machine learning network that characterizes the mapping relationship between "elastic attribute - lithology" and "elastic attribute - organic matter content" in the training stage. Finally, it combines the trained machine learning network with the pre-stack elastic parameter inversion results of pre-stack seismic data to predict the spatial distribution and organic matter content of hydrocarbon source rocks.

[0008] The above existing technologies are all quite different from the present invention and cannot solve the technical problems we want to solve. Therefore, we have invented a new machine learning-based automatic lithofacies identification method for imaging gravel content. Summary of the Invention

[0009] The object of the present invention is to provide a machine learning-based automatic lithofacies identification method for glutenite based on imaging gravel content, which calculates the gravel content on the basis of imaging logging image processing, and determines the lithofacies type through training and learning by machine learning algorithms.

[0010] The object of the present invention can be achieved by the following technical measures: A machine learning-based automatic lithofacies identification method for imaging gravel content, the machine learning-based automatic lithofacies identification method for imaging gravel content includes:

[0011] Step 1: Conduct core observation and description on the cored well, and carry out four-property relationship analysis;

[0012] Step 2: Determine the conventional and imaging logging response patterns of different lithofacies of glutenite;

[0013] Step 3: Through imaging image processing, finely depict the gravel content and gravel diameter distribution, and quantitatively divide the lithofacies of the entire well section;

[0014] Step 4: Based on the lithofacies determined by imaging logging, screen the electrical layers with stable logging characteristics as sample layers;

[0015] Step 5: Determine the sensitive curves by calibrating the conventional logging curves with imaging;

[0016] Step 6: Based on machine learning algorithms, automatically identify the lithofacies with the lithofacies of the sample layer as the supervised object.

[0017] The object of the present invention can also be achieved by the following technical measures:

[0018] In Step 1, conduct core observation on the cored well with imaging logging, describe various lithologies, gravel diameters, longitudinal variations, and single-layer thicknesses, and pay special attention to the lithological characteristics corresponding to the obvious change sections in the imaging and conventional logging curves.

[0019] In Step 1, due to the systematic error between the drilling core depth and the logging depth, use the imaging image and core scanning gamma to assist in core positioning to make the rock-electricity correspondence more reasonable; conduct four-property relationship analysis in combination with logging, analysis and testing data.

[0020] In Step 2, core calibration of imaging logging is carried out. According to characteristics such as lithology and gravel size, the lithologies observed in the core are divided into several lithofacies that can be identified on the logging curves based on the logging curve response characteristics. Lithofacies refers to several lithology combinations with similar lithologies and similar logging response characteristics. The purpose is to consider the identifiability of the logging response and transform the problem of severe heterogeneity into a relatively homogeneous problem, and then determine the conventional and imaging logging response patterns of different lithofacies of glutenite.

[0021] In Step 3, imaging logging is the response of the resistivity change along the borehole wall. Bright colors represent high resistivity and dark colors represent low resistivity. Based on the static imaging data, core calibration of imaging is used to determine four components: gravel, sandy, silty, and muddy. Among them, Component 1 gravel is white spots and lumps, Component 2 sandy is bright-colored, dispersed or layered, Component 3 silty is dark-colored, dispersed, and Component 4 muddy is mainly black-layered.

[0022] In Step 3, by adjusting the tone threshold to depict the bright color, that is, the edge of the gravel, the imaging image is transformed into an image composed of areas of different tones. Different tones represent different lithofacies. The tone ranges of different lithofacies are determined through core calibration. Further, within a certain window length, the content of different lithofacies types is determined by superimposing areas of the same tone. The distribution areas of different components are used as the lithofacies percentage content, and the gravel content is calculated to achieve quantitative division of lithofacies in the whole well section.

[0023] In Step 4, based on the lithofacies determined by imaging logging and on the basis of the resolution ability of the corresponding conventional logging series, sample layers are screened.

[0024] In Step 4, the screening principles are as follows: the lithofacies are relatively homogeneous, the single-layer thickness is ≥2 m, and the characteristics of the corresponding conventional logging curves are stable.

[0025] In Step 5, through methods such as crossplot technology and principal component analysis, combined with the vertical resolution and lateral detection depth of different logging series, curves with high resolution ability for different lithofacies are selected as sensitive curves.

[0026] In Step 5, five sensitive curves are determined, namely compensated neutron logging (CNL), density logging (DEN), natural gamma ray (GR), spontaneous potential (SP), and micro-potential (RN).

[0027] In Step 6, taking the sample layers screened in Step 4 as the supervised objects and the logging values of the sensitive curves determined in Step 5 as the training objects, on the basis of normalization processing, mathematical algorithms such as self-organizing feature mapping, BP neural network, and support vector machine are used for repeated training, learning, and classification. A model with a high degree of coincidence with the core and imaging lithofacies is selected as the machine learning model for this well, and finally it is popularized and applied to other wells in the whole block to achieve automatic identification of glutenite lithofacies.

[0028] The machine learning lithofacies automatic identification method based on imaging gravel content in the present invention is a multi-scale and multi-parameter fusion gravel rock lithofacies automatic identification method based on imaging logging image processing, gravel content quantitative calculation, logging sensitivity analysis, machine learning, etc. Taking the lithofacies sample layer based on imaging identification as the supervised object for training and learning, extracting the sensitive characteristic values of conventional logging curves, and establishing a machine learning lithofacies identification model to realize the automatic identification of gravel rock lithofacies. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 FIG. is a flowchart of an embodiment of the machine learning lithofacies automatic identification method based on imaging gravel content of the present invention;

[0030] Figure 2 FIG. is a flowchart of a specific embodiment of the machine learning lithofacies automatic identification method based on imaging gravel content of the present invention;

[0031] Figure 3 FIG. is a flowchart of the machine learning lithofacies automatic identification in a specific embodiment of the present invention;

[0032] Figure 4 FIG. is a flowchart of the machine learning lithofacies automatic identification in another specific embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which the present invention belongs.

[0034] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, and / or combinations thereof.

[0035] As Figure 1 shown, Figure 1 FIG. is a flowchart of the machine learning lithofacies automatic identification method based on imaging gravel content of the present invention. The machine learning lithofacies automatic identification method based on imaging gravel content includes the following steps:

[0036] Step 1, conduct core observation and description on the cored well and carry out the analysis of the four relationships;

[0037] Step 2, calibrate the imaging logging with the core to determine the conventional and imaging logging response patterns of different lithofacies of the gravel rock;

[0038] Step 3: Through imaging and image processing, finely depict the gravel content and gravel size distribution, and quantitatively divide the lithofacies of the entire well section.

[0039] Step 4: Based on the lithofacies determined by imaging logging, screen the electrical layers with stable logging characteristics as sample layers.

[0040] Step 5: Determine the sensitive curves by calibrating conventional logging curves with imaging.

[0041] Step 6: Based on machine learning algorithms, automatically identify lithofacies with the lithofacies of the sample layer as the supervised object.

[0042] The following are several specific embodiments of applying the present invention.

[0043] Embodiment 1:

[0044] In the specific Embodiment 1 of applying the present invention, as Figure 2 shown, Figure 2 is a flowchart of a specific embodiment of the automatic lithofacies identification method of machine learning based on imaging gravel content of the present invention.

[0045] Step 1: Conduct core observation on the cored wells with imaging logging, and describe as detailed as possible various lithologies, gravel sizes, longitudinal variations, single-layer thicknesses, etc., and pay special attention to the lithological characteristics corresponding to the obvious change sections in the imaging and conventional logging curves; due to the systematic error between the drilling core depth and the logging depth, use imaging images and core scanning gamma to assist in core positioning to make the rock-electricity correspondence more reasonable; carry out the analysis of the four relationships by combining logging, analysis and testing data.

[0046] Step 2: Core-scale imaging logging, divide the lithologies observed in the core into several lithofacies that can be recognized on the logging according to the logging curve response characteristics based on characteristics such as lithology and gravel size. Lithofacies refers to several lithological combinations with similar lithologies and similar logging response characteristics. The purpose is to consider the recognizable ability of logging responses and transform the problem of severe heterogeneity into a relatively homogeneous problem, and then determine the conventional and imaging logging response patterns of different lithofacies of sandy conglomerate.

[0047] Step 3: The imaging logging is the response to the resistivity change along the wellbore wall. Bright colors represent high resistivity and dark colors represent low resistivity. Based on the static imaging data, core calibration imaging is used to determine four components, namely gravel, sandy, silty, and muddy. Among them, component 1, gravel, appears as white spots and lumps; component 2, sandy, is bright in color, either dispersed or layered; component 3, silty, is dark in color and dispersed; component 4, muddy, is mainly black and layered. By adjusting the hue threshold to depict the edge of the bright color (gravel), the imaging image is transformed into an image composed of areas of different hues. Different hues represent different lithofacies. The hue range of different lithofacies is determined through core calibration. Further, within a certain window length, the superposition of areas of the same hue is used to determine the content of different lithofacies types. The distribution areas of different components are used as the lithofacies percentage content to calculate the gravel content, realizing the quantitative division of lithofacies in the entire well section.

[0048] Step 4: Based on the lithofacies determined by imaging logging and on the basis of the resolution of the corresponding conventional logging series, sample layers are screened. The screening principles are as follows: the lithofacies is relatively homogeneous, the single-layer thickness is ≥2 m, and the characteristics of the corresponding conventional logging curves are stable.

[0049] Step 5: Through methods such as crossplot technology and principal component analysis, combined with the vertical resolution and lateral detection depth of different logging series, curves with high resolution ability for different lithofacies are selected as sensitive curves. In this example, 5 sensitive curves are determined, namely compensated neutron logging (CNL), density logging (DEN), natural gamma (GR), spontaneous potential (SP), and micro-potential (RN).

[0050] Step 6: Using the sample layers screened in Step 4 as the supervised objects and the logging values of the sensitive curves determined in Step 5 as the training objects, on the basis of normalization processing, mathematical algorithms such as self-organizing feature mapping, BP neural network, and support vector machine are used for repeated training, learning, and classification. A model with a high degree of coincidence with the core and imaging lithofacies is selected as the machine learning model for this well, and finally it is popularized and applied to other wells in the entire block to realize the automatic identification of the lithofacies of glutenite. In this example, 5 lithologies are identified: mudstone, sandstone, gravel-bearing sandstone, pebbly sandstone, and conglomerate. (See Figure 3 ).

[0051] Example 2:

[0052] In the specific implementation example 2 of applying the present invention, as Figure 2 shown, Figure 2 is a flowchart of a specific implementation example of the machine learning lithofacies automatic identification method based on imaging gravel content of the present invention.

[0053] Step 1: Conduct core observations on the cored wells with imaging logging. Try to describe in detail various lithologies, gravel diameters, vertical variations, single-layer thicknesses, etc., and particularly pay attention to the lithological characteristics corresponding to the significantly varying intervals in the imaging and conventional logging curves. Due to the systematic error between the drilling core depth and the logging depth, use imaging images and core scan gamma to assist in core positioning to make the lithology-electricity correspondence more reasonable. Carry out the analysis of the four relationships by combining data such as logging and laboratory analysis.

[0054] Step 2: Core calibration of imaging logging. According to characteristics such as lithology and gravel size, divide the lithologies observed in the core into several lithofacies that can be identified in the logging based on the logging curve response characteristics. Lithofacies refers to several lithological combinations with similar lithologies and similar logging response characteristics. The purpose is to consider the identifiability of the logging response and transform the problem of severe heterogeneity into a relatively homogeneous problem, and then determine the conventional and imaging logging response patterns of different lithofacies of sandy conglomerate.

[0055] Step 3: Imaging logging is the response to the resistivity change along the borehole wall. Bright colors represent high resistivity and dark colors represent low resistivity. Based on the static imaging data, use core calibration of imaging to determine four components, namely gravel, sandy, silty, and muddy. Among them, component 1 gravel is white spots and masses, component 2 sandy is bright and dispersed or layered, component 3 silty is dark and dispersed, and component 4 muddy is mainly black layered. By adjusting the tone threshold to depict the edge of the bright color (gravel), transform the imaging image into an image composed of areas of different tones. Different tones represent different lithofacies. Determine the tone range of different lithofacies through core calibration. Further, within a certain window length, use the superposition of areas with the same tone to determine the content of different lithofacies types. The distribution areas of different components are used as the lithofacies percentage content to calculate the gravel content and achieve the quantitative division of lithofacies in the entire well section.

[0056] Step 4: Based on the lithofacies determined by imaging logging and on the basis of the resolution ability of the corresponding conventional logging series, screen sample layers. The screening principle is: relatively homogeneous lithofacies, single-layer thickness ≥ 2m, and stable characteristics of the corresponding conventional logging curves.

[0057] Step 5: Through methods such as crossplot technology and principal component analysis, combined with the vertical resolution and lateral detection depth of different logging series, select the curves with high resolution ability for different lithofacies as sensitive curves. In this example, 4 sensitive curves are determined, namely compensated neutron logging (CNL), density logging (DEN), natural gamma (GR), and micro-potential (RN).

[0058] Step 6: Taking the sample layer screened in Step 4 as the supervised object and the logging values of the sensitive curves determined in Step 5 as the training object, on the basis of normalization processing, using mathematical algorithms such as self-organizing feature mapping, BP neural network, and support vector machine for repeated training, learning, and classification, and selecting a model with a high degree of coincidence with the core and imaging lithofacies as the machine learning model of this well, and finally promoting and applying it to other wells in the whole block to realize the automatic identification of glutenite lithofacies. (See Figure 4 )。

[0059] Example 3:

[0060] In the specific Example 3 of applying the present invention, similar to Example 1 and Example 2, the implementation process is a flowchart of a specific embodiment of the machine learning lithofacies automatic identification method based on the imaging gravel content ( Figure 2 )。

[0061] Step 1: Conduct core observation and description on the cored well and carry out four-property relationship analysis;

[0062] Step 2: Core-scale imaging logging to determine the conventional and imaging logging response patterns of different lithofacies of glutenite;

[0063] Step 3: Through imaging image processing, finely depict the gravel content and gravel diameter distribution, and quantitatively divide the lithofacies of the whole well section;

[0064] Step 4: Based on the lithofacies determined by imaging logging, screen the electrical layers with stable logging characteristics as the sample layer;

[0065] Step 5: Calibrate the conventional logging curves through imaging to determine the sensitive curves. In this Example 3, the acoustic travel time curve is more sensitive to lithology identification. Therefore, 5 sensitive curves are determined in this example, namely compensated neutron logging (CNL), density logging (DEN), natural gamma (GR), acoustic travel time (AC), and micro-potential (RN).

[0066] Step 6: Taking the sample layer screened in Step 4 as the supervised object and the logging values of the sensitive curves determined in Step 5 as the training object, on the basis of normalization processing, using mathematical algorithms such as self-organizing feature mapping, BP neural network, and support vector machine for repeated training, learning, and classification, and selecting a model with a high degree of coincidence with the core and imaging lithofacies as the machine learning model of this well, and finally promoting and applying it to other wells in the whole block to realize the automatic identification of glutenite lithofacies.

[0067] The machine learning lithofacies automatic identification method based on imaging gravel content calculation in the present invention is applied in the technical field of oilfield exploration and development, and particularly relates to a method for automatically and continuously discriminating the lithofacies of glutenite through multi-scale and multi-parameter fusion, including image logging image processing, gravel content quantitative calculation, logging sensitivity analysis, machine learning, etc. It includes six steps: four-property relationship analysis, determination of imaging logging response pattern, gravel content calculation, sample layer screening, sensitive curve analysis, and machine learning for automatic lithology identification. The invention makes full use of new logging technologies and methods, deeply excavates the information of core, imaging, and logging curves, constructs a machine learning lithology identification method for multi-scale and multi-series logging data fusion, improves the accuracy and efficiency of glutenite lithofacies identification, and achieves the purpose of quantitative, precise, objective, and highly efficient lithofacies identification.

[0068] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0069] Except for the technical features described in the specification, the rest are known technologies to those skilled in the art.

Claims

1. A machine learning-based automatic lithofacies identification method based on imaging gravel content, characterized in that, The machine learning-based automatic lithofacies identification method based on imaging gravel content includes: Step 1: Conduct core observation and description on the cored well, and carry out the analysis of the four relationships. Step 2: Determine the conventional and imaging logging response patterns of different lithofacies of glutenite. Step 3: Through imaging image processing, finely depict the gravel content and gravel size distribution, and quantitatively divide the lithofacies of the entire well section. Step 4: Based on the lithofacies determined by imaging logging, screen the electrical layers with stable logging characteristics as sample layers. Step 5: Determine the sensitive curves by calibrating the conventional logging curves with imaging. Step 6: Based on the machine learning algorithm, use the lithofacies of the sample layer as the supervised object to automatically identify the lithofacies. In Step 6, taking the sample layer screened in Step 4 as the supervised object and the logging values of the sensitive curves determined in Step 5 as the training object, on the basis of normalization processing, use self-organizing feature mapping, BP neural network, and support vector machine for repeated training, learning, and classification. Select the model with a high degree of coincidence with the core and imaging lithofacies as the machine learning model of this well, and finally promote and apply it to other wells in the entire block to realize the automatic identification of glutenite lithofacies.

2. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 1, wherein, In Step 1, conduct core observation on the cored well with imaging logging, describe various lithologies, gravel sizes, longitudinal variations, and single-layer thicknesses, and pay attention to the lithological characteristics corresponding to the obvious change sections in the imaging and conventional logging curves.

3. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 2, characterized in that, In Step 1, due to the systematic error between the drilling core depth and the logging depth, use imaging images and core scanning gamma to assist in core positioning to make the rock-electricity correspondence more reasonable; combine logging and analysis and testing to carry out the analysis of the four relationships.

4. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 1, wherein In Step 2, calibrate the imaging logging with the core. According to the lithology and gravel size, divide the lithology observed in the core into several lithofacies identified on the logging curve based on the logging curve response characteristics. Lithofacies refers to several lithological combinations with similar lithologies and similar logging response characteristics. The purpose is to consider the identifiability of the logging response and transform the problem of severe heterogeneity into a relatively homogeneous problem, and then determine the conventional and imaging logging response patterns of different lithofacies of glutenite.

5. The machine learning-based automatic lithofacies identification method based on imaging gravel content according to claim 1, wherein In Step 3, the imaging image is the response of the resistivity change along the wellbore wall. Bright colors represent high resistivity, and dark colors represent low resistivity. Based on the static imaging data, use the core to calibrate the imaging to determine the four components of gravel, sandy, silt, and mud. Among them, Component 1 gravel is white spots and lumps, Component 2 sandy is bright and dispersed or layered, Component 3 silt is dark and dispersed, and Component 4 mud is mainly black layered.

6. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 5, characterized in that In Step 3, by adjusting the hue threshold to depict the bright color, that is, the edge of the gravel, transform the imaging image into an image composed of different hue areas. Different hues represent different lithofacies. Determine the hue range of different lithofacies through core calibration. Further, within a certain window length, use the superposition of the same-hue areas to determine the content of different lithofacies types. The distribution areas of different components are used as the lithofacies percentage content, calculate the gravel content, and realize the quantitative division of the lithofacies of the entire well section.

7. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 1, characterized in that In Step 4, based on the lithofacies determined by imaging logging, screen the sample layer on the basis of the resolution ability of the corresponding conventional logging series.

8. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 7, wherein In Step 4, the screening principle is: the lithofacies is relatively homogeneous, the single-layer thickness ≥ 2m, and the corresponding conventional logging curve characteristics are stable.

9. The machine learning lithofacies automatic identification method based on imaging gravel content according to claim 1, characterized in that, In step 5, through crossplot technology and principal component analysis, combined with the vertical resolution and lateral detection depth of different logging series, curves with high resolution ability for different lithofacies are selected as sensitive curves.

10. The machine learning automatic lithofacies identification method based on imaging gravel content according to claim 9, characterized in that, In step 5, five sensitive curves are determined, namely compensated neutron log CNL, density log DEN, natural gamma ray GR, spontaneous potential SP, and micro-potential RN.

Images