A device and method for classifying natural gas hydrate types based on machine learning

Abstract

The present invention relates to a machine learning-based natural gas hydrate classification device and method. The device includes a core scanning electron microscope observation module for observing the internal microstructure of hydrate-containing cores using a scanning electron microscope to determine the specific microscopic types of hydrates within the porous medium of the core. The method comprises the following steps: determining the microscopic types of hydrates present within the core using a scanning electron microscope; establishing digital core models of both hydrate-free and hydrate-containing cores through CT scanning; extracting a pore network model based on the watershed principle, calculating characterization parameters of all pores and hydrates, and using the hydrate characterization parameters to form a hydrate microscopic characterization parameter set; and performing cluster analysis on the hydrate microscopic characterization parameter set using a self-organizing map neural network method to obtain the final classification results. The device and method of the present invention can accurately and efficiently automatically classify the microscopic types of hydrates within the core.

Description

Technical Field

[0001] The present invention relates to the field of natural gas hydrate resource evaluation and development methods, and in particular to a natural gas hydrate type classification device and method based on machine learning. Background Art

[0002] As a clean, efficient, and abundant unconventional natural gas resource, natural gas hydrates are of great significance to the sustainable development of human society. Natural gas hydrates are formed by water molecules trapping natural gas molecules under high pressure and low temperature, forming an ice-like crystalline compound. Natural gas hydrate structures are generally classified into three types: Type I, Type II, and Type H, depending on the lattice vacancies and morphology. Type II and Type H natural gas hydrates are much more stable than Type I, but Type I is the predominant natural gas hydrate.

[0003] The extraction process of natural gas hydrates is highly complex, integrating phase transitions, mass transfer, heat transfer, reservoir deformation, and multiphase flow. Hydrate microtype, a key factor influencing the hydrate extraction process, influences the seepage characteristics and saturation of fluids within porous media, and thus the flow patterns within these media. Furthermore, whether hydrates within porous media consolidate mineral particles affects the mechanical stability of the wellbore. The presence of hydrates also influences physical properties such as rock thermal conductivity and acoustic velocity, which in turn affects the detection accuracy of hydrate reservoirs. Therefore, research on hydrate microtypes can deepen our understanding of geological characteristics, such as hydrate accumulation patterns, and has significant theoretical value and engineering implications, laying the foundation for the resourceful utilization of natural gas hydrates.

[0004] The existing natural gas hydrate type identification device is mainly a method and system for identifying the occurrence state of natural gas hydrates in the sea area based on well logging data (CN112133377A). The device includes a well logging data acquisition module, a data calculation module, a model building module, a discrimination parameter calculation module, a disturbance error calculation module, and an occurrence state identification module. The device only selects discrimination parameters based on well logging data (such as shear wave velocity logging data) to determine the occurrence state of hydrates. The device does not characterize and analyze the type of hydrates from a microscopic perspective, and only identifies the hydrate type through well logging data and charts. , the reliability of its results is general, and it cannot realize the automatic discrimination of hydrate microtypes; a method, device and equipment for determining the occurrence state of natural gas hydrates (CN112304988A) mainly includes an X-CT image information measurement module, a hydrate three-dimensional data volume construction module, a hydrate sub-block identification module, a structure acquisition module, an Euler coefficient acquisition module, and an occurrence state determination module. The device only discriminates the microtype of hydrates by identifying the Euler number of hydrates, and cannot classify the hydrates in the core into types one by one. Moreover, due to the relatively single selected parameters, the accuracy of the discrimination device is general.

[0005] Currently, natural gas hydrate classification is primarily based on the type of guest molecule (e.g., CH₄, C₂H₆, CO₂), lattice structure (e.g., Type I, Type II, and Type H), and macroscopic occurrence (including dispersed, massive, and vein-like). At the microscopic level, hydrate types can be categorized as pore-filling, film-coated, and cemented, depending on the contact relationship between the hydrate and the sedimentary rock framework. Existing methods rely on visual observation using scanning electron microscopy (SEM), a cumbersome process that has not yet achieved automated classification of hydrate microtypes, nor has a recognized standard for hydrate microtype classification been established.

[0006] A patented method for hydrate occurrence morphology classification based on a pore network model (CN112151125B) statistically analyzes the hydrate shape factors and coordination numbers of three hydrate occurrence morphologies to develop a hydrate occurrence morphology classification standard. The hydrate-containing core to be classified is then scanned using CT to create a digital hydrate core. After determining the relationship between the hydrate and the core pores, the shape factors and coordination numbers of the corresponding hydrates are extracted from the digital core and compared using the hydrate occurrence morphology classification standard to determine the hydrate occurrence morphology. However, this method limits the microscopic types of hydrates to only three, and the theoretical model established is relatively simple and cannot be adjusted according to actual conditions. The judgment criteria are single and cannot automatically distinguish the microscopic types of hydrates.

[0007] A patented method and system for identifying the occurrence state of natural gas hydrates in marine areas (CN112133377A) selects identification parameters based on well logging data (such as shear wave velocity logging data) to qualitatively identify the occurrence state of hydrates. This method does not characterize and analyze hydrate types from a microscopic perspective, but only uses well logging data and charts to identify hydrate types. The results are generally unreliable and cannot automatically identify hydrate microscopic types.

[0008] A patent for a method, device, and apparatus for determining the occurrence state of natural gas hydrates (CN112304988A) identifies at least one hydrate sub-block in a three-dimensional data volume created by a CT experiment; obtains the structural category and structural evaluation value of each hydrate sub-block based on the morphological characteristics of the hydrate sub-block; obtains the Euler coefficient corresponding to the natural gas hydrate sample according to the structural type; and determines the occurrence state of the natural gas hydrate sample using the structural evaluation value and Euler coefficient. This method does not analyze the structural characteristics of hydrates in detail from a microscopic perspective, and the selected parameters are relatively simple: the microscopic type of hydrates is only determined by identifying the Euler number of the hydrate. Moreover, this method cannot classify the hydrates in the core into types one by one, so the accuracy is average and it cannot achieve automatic discrimination of the microscopic type of hydrates. Summary of the Invention

[0009] In view of the above problems existing in the prior art, the technical problem to be solved by the present invention is: how to automatically and accurately classify the microscopic types of hydrates in porous media.

[0010] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0011] A method for classifying natural gas hydrate types based on machine learning includes the following steps:

[0012] S100: Select any core without hydrate and any core with hydrate under the same environmental conditions;

[0013] S200: Scanning the hydrate-containing core using a scanning electron microscope to determine the microscopic type of hydrate currently contained in the core. The microscopic types include pore-filling type, film-coated type, and cemented type.

[0014] S300: performing CT scanning on the hydrate-containing core and the hydrate-free core respectively under high pressure and low temperature conditions to obtain slice images of the hydrate-containing core and slice images of the hydrate-free core;

[0015] S400: Use image-J software to analyze the images of hydrate-containing core slices to construct a hydrate-containing digital core model, and use image-J software to analyze the images of hydrate-free core slices to construct a hydrate-free digital core model;

[0016] S500: extracting the corresponding pore network model G1 from the hydrate-containing digital core model to obtain pore microscopic characterization parameters of the hydrate-containing digital core; extracting the corresponding pore network model G2 from the hydrate-free digital core model to obtain pore microscopic characterization parameters of the hydrate-free digital core;

[0017] S600: Comparing G1 and G2 to obtain all hydrate microscopic characterization parameters in the hydrate-containing core, wherein the hydrate microscopic characterization parameters include the coordinate system position of each hydrate, the pixel volume of each hydrate, and the pixel surface area of ​​each hydrate;

[0018] S700: Using all the hydrate microscopic characterization parameters obtained in S600, calculate and obtain a hydrate microscopic characterization parameter set. The specific steps are as follows:

[0019] S710: Screening the volume of each hydrate. If the volume of the hydrate is less than 10 cubic pixels, the hydrate is considered as image noise and is removed. If the volume of the hydrate is greater than or equal to 10 cubic pixels, the hydrate is retained.

[0020] S720: Using the existing hydrate microscopic characterization parameter calculation module to obtain a hydrate microscopic characterization parameter set, specifically:

[0021] The hydrate microscopic characterization parameter calculation module includes two parts. The first part first screens each hydrate retained in S710. The second part calculates the corresponding hydrate microscopic characterization parameters for each hydrate screened in the first part. The microscopic characterization parameters of all hydrates calculated in the second part constitute the hydrate microscopic characterization parameter set.

[0022] S800: Preset the learning rate and use the self-organizing map neural network machine learning method to calculate the hydrate microscopic characterization parameter set. The calculation result is the specific microscopic type of all hydrates retained in S720, that is, the specific microscopic type of each hydrate is classified into one of the pore-filling type, film-coated type, and cemented type.

[0023] Preferably, the method used to obtain the pore microscopic characterization parameters in S500 is the watershed method and the maximum sphere method, and the specific steps are as follows:

[0024] S510: Segmenting the core slice image using a watershed method to obtain a pore class and a rock skeleton class, wherein the core slice image includes a hydrate-containing core slice image and a hydrate-free core slice image;

[0025] S520: The maximum sphere method is used to analyze the pores and rock skeletons to obtain the microscopic characterization parameters of the pores.

[0026] Preferably, the microscopic characterization parameters of each hydrate retained and calculated in S720 include the Euler number of the hydrate, the surface area ratio of the hydrate to the pore in which it is located, and the volume ratio of the hydrate to the pore in which it is located.

[0027] Preferably, the specific process of screening each hydrate retained in S710 in the first part of S720 is as follows:

[0028] Traverse all hydrates retained in S710, compare the position of each hydrate with the pore position and pore radius of the pore where it is located, and determine whether the hydrate exists in the pore: if the hydrate does not exist in the pore, exclude the hydrate and determine the next hydrate; if the hydrate exists in the pore, retain it;

[0029] The judgment formula is as follows:

[0030]

[0031] Among them, x h 、y h 、z h Indicates the hydrate coordinate information of the hydrate-bearing core, x p 、y p 、z p represents the pore coordinate information of the core without hydrate, and R represents the pore radius of the core without hydrate.

[0032] Preferably, the specific steps of calculating the hydrate microscopic characterization parameter set using the self-organizing map neural network machine learning method in S800 are:

[0033] S810: The self-organizing map neural network includes an input layer and a competition layer;

[0034] S820: using the hydrate microscopic characterization parameter set as input data of the input layer, and performing unsupervised clustering processing on the hydrate microscopic characterization parameter set by the input layer to obtain output data of the input layer;

[0035] S830: The output data obtained in S820 is used as input data of the competition layer, and the output is the classification result of the hydrate microscopic type.

[0036] Preferably, the learning rate preset in S800 is 0.5.

[0037] A natural gas hydrate classification device based on machine learning. The natural gas hydrate classification method based on machine learning of claims 1-8 uses a natural gas hydrate classification device based on machine learning. The natural gas hydrate classification device based on machine learning includes:

[0038] Core scanning module, model generation and analysis module, hydrate microscopic characterization parameter calculation module, and machine learning hydrate type classification module;

[0039] The core scanning module is used to observe the microstructure inside the core and determine the microscopic type of hydrate contained in the core; the core scanning module is also used to perform CT scanning on the core under high pressure and low temperature conditions to obtain CT slice images;

[0040] The model generation and analysis module is used to analyze CT slice images using image-J software to establish a hydrate-containing digital core model and a hydrate-free digital core model, and extract corresponding pore network models G1 and G2 from the hydrate-containing digital core model and the hydrate-free digital core model, respectively; analyze G1 and G2 to obtain pore-related characterization parameters contained in G1 and G2, respectively;

[0041] The hydrate microscopic characterization parameter calculation module is used to compare G1 and G2 to obtain all hydrate microscopic characterization parameters in the hydrate-containing core, and calculate the hydrate microscopic characterization parameter set using all the obtained hydrate microscopic characterization parameters;

[0042] The machine learning hydrate type classification module is used to calculate the hydrate microscopic characterization parameter set, and the calculation result is the specific microscopic type of hydrate classification, which is one of the pore-filling type, the film-coated type, and the cemented type.

[0043] Compared with the prior art, the present invention has at least the following advantages:

[0044] 1. This algorithm first uses scanning electron microscopy to observe the occurrence of methane hydrates within hydrate-bearing cores and determine the number of hydrate microtypes within the cores. CT scanning is then used to obtain the true internal structure of the hydrate-bearing and hydrate-free cores, and three-dimensional digital core models of the hydrate-bearing and hydrate-free cores are established. A watershed approach is then applied to specifically delineate the rock skeleton, pores, and hydrates, extracting pore network models for the hydrate-bearing and hydrate-free cores. By comparing the differences between the two pore network models, a dataset of hydrate characterization parameters is calculated and formed. This dataset is then computed using a self-organizing map neural network algorithm, allowing for clustering of samples without prior classification results, thus achieving classification of hydrate microtypes.

[0045] 2. The method of the present invention does not require observation of each hydrate, and all hydrates in the core can be directly classified into their microscopic types through the algorithm.

[0046] 3. The effects of the device used in conjunction with this method: First, the core scanning module was used to observe the occurrence of methane hydrates within the hydrate-bearing core, clarifying the number of hydrate microtypes in the core. CT scanning was then performed to obtain the true internal structure of the core containing and not containing hydrates. Then, the model generation and analysis module was used to establish three-dimensional digital core models containing and not containing hydrates. The rock skeleton, pores, and hydrates were specifically delineated, and the pore network models of the hydrate-bearing and hydrate-free cores were extracted. The hydrate microcharacterization parameter calculation module calculated and formed a hydrate characterization parameter dataset by comparing the differences between the two pore network models. Finally, the machine learning hydrate type classification module was used. Based on the self-organizing map neural network algorithm, the samples were clustered without the need for pre-classification results, achieving the classification of hydrate microtypes.

[0047] 4. The advantage of the present invention is that it establishes a realistic three-dimensional core model containing hydrates through CT scanning. Combined with machine learning methods, it can accurately and efficiently classify the microscopic types of hydrates in porous media, filling the current gap in the research on hydrate microtypes and having important practical significance for hydrate mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a simplified structural diagram of the natural gas hydrate type classification device based on machine learning in Example 2.

[0049] Figure 2 This is the image of hydrate inside the core observed by scanning electron microscopy in Example 1.

[0050] Figure 3 This is the digital core model containing hydrate in Example 1.

[0051] Figure 4 This is the pore network model of the hydrate-free core in Example 1.

[0052] Figure 5 This is the three-dimensional model of the hydrate extracted in Example 1.

[0053] Figure 6 This is the result diagram of automatic classification of hydrate microtypes using the machine learning method in Example 1.

[0054] Figure 7 This is a simplified flow chart of the hydrate-containing core treatment process in Example 1. DETAILED DESCRIPTION

[0055] The present invention is described in further detail below.

[0056] Example 1: See Figure 7The core scanning module is used to determine the microscopic types of hydrates existing in the porous media in the core, and a digital core model containing hydrates is established through CT scanning image slices. The pore network model is extracted according to the watershed principle, and the relevant parameters of all hydrates are calculated to form a hydrate microscopic characterization parameter data set; finally, the machine learning method of the self-organizing map neural network is used to perform cluster analysis on the hydrate microscopic characterization parameter data set to obtain the division results.

[0057] A method for classifying natural gas hydrate types based on machine learning includes the following steps:

[0058] S100: Select any core without hydrate and any core with hydrate under the same environmental conditions;

[0059] S200: Scan the hydrate-containing core using a scanning electron microscope to determine the microscopic type of hydrates currently contained within the core. The microscopic types include pore-filling, film-coated, and cemented. Scanning electron microscope experiments are an existing technology that can scan the internal structure of the core and confirm the microscopic type of hydrates contained within the core based on the scanning results. Specifically, there are three microscopic types: pore-filling, film-coated, and cemented. Pore-filling hydrates are dispersed in pores and have no contact with sediments; cemented hydrates form in loose sediments; and film-coated hydrates form on the surface of rock particles and adhere to the particle surface.

[0060] S300: performing CT scanning on the hydrate-containing core and the hydrate-free core respectively under high pressure and low temperature conditions to obtain slice images of the hydrate-containing core and slice images of the hydrate-free core;

[0061] S400: using image-J software to analyze the hydrate-containing core slice image to construct a hydrate-containing digital core model, and using image-J software to analyze the hydrate-free core slice image to construct a hydrate-free digital core model. Image-J software is prior art.

[0062] S500: extracting the corresponding pore network model G1 from the hydrate-containing digital core model to obtain pore microscopic characterization parameters of the hydrate-containing digital core; extracting the corresponding pore network model G2 from the hydrate-free digital core model to obtain pore microscopic characterization parameters of the hydrate-free digital core;

[0063] The methods used to obtain the pore microscopic characterization parameters in S500 are the watershed method and the maximum sphere method. The watershed method and the maximum sphere method are existing technologies. The specific steps are as follows:

[0064] S510: Segmenting the core slice image using a watershed method to obtain a pore class and a rock skeleton class, wherein the core slice image includes a hydrate-containing core slice image and a hydrate-free core slice image; when segmenting the image, pixels with a threshold value less than 60 are classified as pores, and pixels with a threshold value greater than or equal to 60 are classified as rock skeletons;

[0065] S520: Analyze the pores and rock skeletons using the largest sphere method to obtain microscopic characterization parameters of the pores. The microscopic characterization parameters of the pores include the number of pores in the hydrate-bearing core and the hydrate-free core, the pore position (x p ,y p , z p ), pore radius (r p ) and pore volume (V p ).

[0066] S600: Comparing G1 and G2 to obtain all hydrate microscopic characterization parameters in the hydrate-bearing core, the hydrate microscopic characterization parameters including the coordinate system position of each hydrate, the pixel volume of each hydrate, and the pixel surface area of ​​each hydrate; in addition, according to calculation requirements, the hydrate microscopic characterization parameters here also include the total number of hydrate types and the shape factor of each hydrate;

[0067] S700: Using all the hydrate microscopic characterization parameters obtained in S600, calculate and obtain a hydrate microscopic characterization parameter set. The specific steps are as follows:

[0068] S710: Screening the volume of each hydrate. If the volume of the hydrate is less than 10 cubic pixels, the hydrate is considered as image noise and is removed. If the volume of the hydrate is greater than or equal to 10 cubic pixels, the hydrate is retained.

[0069] S720: Using the existing hydrate microscopic characterization parameter calculation module to obtain a hydrate microscopic characterization parameter set, specifically:

[0070] The hydrate microscopic characterization parameter calculation module includes two parts. The first part first screens each hydrate retained in S710. The second part calculates the corresponding hydrate microscopic characterization parameters for each hydrate screened in the first part. The microscopic characterization parameters of all hydrates calculated in the second part constitute the hydrate microscopic characterization parameter set.

[0071] The microscopic characterization parameters of each hydrate retained by calculation in S720 include the Euler number of the hydrate, the surface area ratio of the hydrate to the pore in which it is located, and the volume ratio of the hydrate to the pore in which it is located.

[0072] The specific process of the first part of S720 for screening each hydrate retained in S710 is as follows:

[0073] Traverse all hydrates retained in S710, compare the position of each hydrate with the pore position and pore radius of the pore where it is located, and determine whether the hydrate exists in the pore: if the hydrate does not exist in the pore, exclude the hydrate and determine the next hydrate; if the hydrate exists in the pore, retain it;

[0074] The judgment formula is as follows:

[0075]

[0076] Among them, x h 、y h 、z h Indicates the hydrate coordinate information of the hydrate-bearing core, x p 、y p 、z p represents the pore coordinate information of the core without hydrate, and R represents the pore radius of the core without hydrate.

[0077] S800: Preset the learning rate and use the self-organizing map neural network machine learning method to calculate the hydrate microscopic characterization parameter set. The self-organizing map neural network SOM machine learning method is an existing technology. The calculation result is the specific microscopic type of all hydrates retained in S720, that is, the specific microscopic type of each hydrate is classified into one of the pore-filling type, film-coated type, and cemented type.

[0078] The specific steps of calculating the hydrate microscopic characterization parameter set using the self-organizing map neural network machine learning method in S800 are:

[0079] S810: The self-organizing map neural network includes an input layer and a competition layer;

[0080] S820: using the hydrate microscopic characterization parameter set as input data of the input layer, and performing unsupervised clustering processing on the hydrate microscopic characterization parameter set by the input layer to obtain output data of the input layer;

[0081] S830: The output data obtained in S820 is used as the input data of the competition layer, and the output is the classification result of the hydrate micro-type. Here, the classification result of the type is pore-filling type, film-coated type, and cemented type.

[0082] The learning rate preset in S800 is 0.5.

[0083] Example 2: See Figure 1The model uses the core scanning module to observe and analyze the specific hydrate microtypes contained in hydrate cores and non-hydrate cores, and then performs CT scanning to obtain slice images; uses the model generation and analysis module to obtain the pore network models of hydrate-containing cores and non-hydrate-containing cores and obtain the pore network related characterization parameters; then uses the hydrate microcharacterization parameter calculation module to calculate the hydrate characterization parameter set, and finally uses the machine learning hydrate type classification module to analyze the hydrate characterization parameter set to obtain the final classification result.

[0084] A natural gas hydrate classification device based on machine learning. The natural gas hydrate classification method based on machine learning of claims 1-8 uses a natural gas hydrate classification device based on machine learning. The natural gas hydrate classification device based on machine learning includes:

[0085] Core scanning module, model generation and analysis module, hydrate microscopic characterization parameter calculation module, and machine learning hydrate type classification module;

[0086] The core scanning module includes a core scanning electron microscope observation module and a core CT scanning module;

[0087] The model generation and analysis module includes a digital core model generation and analysis module and a pore network model generation and analysis module;

[0088] The core scanning module is used to observe the microstructure inside the core and determine the microscopic type of hydrate contained in the core; the core scanning module is also used to perform CT scanning on the core under high pressure and low temperature conditions to obtain CT slice images;

[0089] The model generation and analysis module is used to analyze CT slice images using image-J software to establish a hydrate-containing digital core model and a hydrate-free digital core model, and extract corresponding pore network models G1 and G2 from the hydrate-containing digital core model and the hydrate-free digital core model, respectively; analyze G1 and G2 to obtain pore-related characterization parameters contained in G1 and G2, respectively; the pore-related characterization parameters include microscopic characterization parameters such as the number, position, radius, and volume of pores;

[0090] The hydrate microscopic characterization parameter calculation module is used to compare G1 and G2 to obtain all hydrate microscopic characterization parameters in the hydrate-containing core, and calculate the hydrate microscopic characterization parameter set using all the obtained hydrate microscopic characterization parameters. The hydrate microscopic characterization parameters include number, position, volume, surface area, shape factor, etc.

[0091] The machine learning hydrate type classification module is used to calculate the hydrate microscopic characterization parameter set, and the calculation result is the specific microscopic type of hydrate classification, which is one of the pore-filling type, the film-coated type, and the cemented type.

[0092] Figure 7 This is a block diagram of a device for classifying methane hydrate types based on machine learning according to Example 2. The device for classifying methane hydrate types based on machine learning (hereinafter referred to as "the device for classifying methane hydrate types") of the present invention.

[0093] Specifically, the core scanning electron microscope observation module is implemented according to the method described in step S200 above. The low-temperature scanning electron microscope used in the experiment is mainly a Hitachi S-3400 scanning electron microscope, with an acceleration voltage of 0.3-30 kV, a magnification of 5-300,000 times, and a maximum spatial resolution of 10 nm. It can receive secondary electron information and backscattered electrons generated by the stimulated material under test; it is configured to observe the internal microstructure of the hydrate-containing core using the scanning electron microscope to determine the specific microscopic type of the hydrate in the porous medium in the core;

[0094] The core CT scanning module is implemented according to the method described in steps S300 to S400 above. During implementation, the CT may be configured to have a resolution of 18 μm, and the size of the intercepted image is 150×150×150 pixels for performing a CT scan of the core under high pressure and low temperature conditions.

[0095] The digital core model generation and analysis module applies the obtained slice images to the image-J software to construct a digital core model;

[0096] The pore network model generation and analysis module is implemented according to the method described in step S500 above, and is configured to use the digital core model generated by the core CT scanning module. The pore network model of the digital core is extracted using the watershed method, and the pore network model is analyzed using the maximum sphere method to obtain microscopic characterization parameters such as the number, location, radius, and volume of pores.

[0097] The hydrate microscopic characterization parameter calculation module is implemented according to the method described in steps S600-S720 above. By comparing the pore network models of the hydrate-free core and the hydrate-containing core, microscopic characterization parameters such as the number, position, volume, surface area, and shape factor of the hydrate are obtained. The obtained hydrate volume is screened. Hydrates with a volume less than 10 cubic pixels are determined to be image noise and need to be removed. Then, by comparing the position and radius information of the hydrate with the pore, it is determined whether the hydrate is in the pore. If so, the microscopic characterization parameters of the hydrate (Euler number, surface area ratio of the hydrate to the pore, volume ratio of the hydrate to the pore) are calculated, and the above microscopic characterization parameters of the hydrate are written into the hydrate microscopic characterization parameter set.

[0098] The machine learning hydrate type classification module is implemented according to the method described in step S800 above, with the SOE learning rate set to 0.5 and the number of network nodes set to 100. Unsupervised clustering is performed on the hydrate microscopic characterization parameter set, and the number of categories is the number of hydrate microscopic types observed by the scanning electron microscope in S200, thereby realizing the automatic classification of hydrate microscopic types in the porous media in the core.

[0099] Data Experiment

[0100] See also Figure 2-Figure 6 To further illustrate the effectiveness of this technical approach, a detailed description of the present invention is provided using a siltstone hydrate core from the Shenhu Sea area in my country as an example. First, CT scans of both hydrate-free and hydrate-containing siltstone cores from the Shenhu Sea area were performed under high-pressure, low-temperature conditions. Digital core models of both the hydrate-free and hydrate-containing cores were constructed. These models were sized at 400×400×400 pixels and had a resolution of 3.6 microns.

[0101] Scanning electron microscopy (SEM) was used to observe the internal microstructure of hydrate cores. The authors determined the specific microscopic types of hydrates within the porous media of the siltstone cores from the Shenhu Sea area: pore-filling, film-coated, and cemented. Pore-filling hydrates are dispersed within the pores, isolated from the sediments; cemented hydrates form within loose sediments; and film-coated hydrates form on the surfaces of rock particles and adhere to them.

[0102] The watershed method was used to extract the pore network model of the digital core without hydrate and the pore network model of the digital core with hydrate. After watershed segmentation, the pixel threshold less than 60 was classified as pores, and the pixel threshold greater than or equal to 60 was classified as rock skeleton. The pores and throats in the pore network model were analyzed using the maximum sphere method, and the number and location of pores (x) in the three-dimensional siltstone core of the Shenhu sea area with and without hydrate were obtained respectively. p ,y p ,zp ), radius (r p ), volume (V p ) and other microscopic characterization parameter sets, as shown in Tables 1 and 2.

[0103] Table 1. Pore microscopic characterization parameter set of hydrate-bearing siltstone cores in the Shenhu sea area

[0104] serial number <![CDATA[Position (x p , y p , z p )]]> <![CDATA[radius (r p )]]> <![CDATA[Volume (r p )]]> 1 (1，1，2) 6.2 4.23×105 2 (1，2，1) 4.5 3.53×105 … … … …

[0105] Table 2 Pore microscopic characterization parameter set of hydrate-free siltstone cores in the Shenhu sea area

[0106] serial number <![CDATA[Position (x p , y p , z p )]]> <![CDATA[radius (r p )]]> <![CDATA[Volume (r p )]]> 1 (1，1，2) 6.9 5.73×105 2 (1，2，1) 5.3 4.23×105 … … … …

[0107] By comparing the pore network models of the cores without hydrate and those containing hydrate in the Shenhu Sea area, the number and location of hydrates in the cores (x h ,y h ,z h ), volume (V h ), surface area (S h ), shape factor, and other microscopic characterization parameters. The hydrate volume is then screened and whether the hydrate is located in pores is determined. The microscopic characterization parameters of pore hydrates (Euler number, surface area ratio of the hydrate to the pore, and volume ratio of the hydrate to the pore) are calculated and entered into the hydrate microscopic characterization parameter set, as shown in Table 3.

[0108] Table 3 Hydrate occurrence type characterization parameter dataset

[0109] Hydrate number Vr Sr Eu 1 0.763 0.832 5 2 0.656 0.711 -1 … … … …

[0110] Unsupervised clustering of the parameter sets in Table 3 was performed using a machine learning method based on a self-organizing map (SOM) neural network. The number of clusters is the number of hydrate microscopic species observed by scanning electron microscopy. The SOM network automatically learns the characteristics of each hydrate in the dataset and classifies the hydrates into three types, thus achieving an automatic classification of the microscopic types of hydrates within the porous media in the Shenhu sea core.

[0111] The results show that this method successfully classified 21,891 hydrates in the porous media of the core. Among them, the Euler number of "type 0" hydrates is mainly distributed between -10 and 0, and the volume ratio of hydrates to the pores (Vr) is between 0.8 and 1, indicating that the size of the hydrates is similar to the pore size and has good contact with the surrounding pores. It is judged to be a pore-filling hydrate.

[0112] The Euler number for "Type 1" hydrates is primarily distributed between 0 and 5, and Vr is distributed between 0 and 1. This indicates that the hydrate has little contact with other nearby pores, resulting in poor contact. Furthermore, the Vr distribution is wide, indicating that it is a membrane-coated hydrate. This is because in larger pores, membrane-coated hydrates occupy a smaller proportion of the pore space, resulting in a smaller Vr. In smaller pores, membrane-coated hydrates occupy a larger portion of the pore space, resulting in a larger Vr.

[0113] The number of "Type 2" hydrates is relatively small, the Euler number is -1 or 0, and Vr is between 0.8 and 1.0, indicating that the degree of contact between the hydrate and the rock skeleton is similar to that of other surrounding pores, and it is judged to be a cemented hydrate.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for classifying natural gas hydrate types based on machine learning, characterized by: The steps include: S100: Select any core without hydrate and any core with hydrate under the same environmental conditions; S200: Scanning the hydrate-containing core using a scanning electron microscope to determine the microscopic type of hydrate currently contained in the core. The microscopic types include pore-filling type, film-coated type, and cemented type. S300: performing CT scanning on the hydrate-containing core and the hydrate-free core respectively under high pressure and low temperature conditions to obtain slice images of the hydrate-containing core and slice images of the hydrate-free core; S400: Use image-J software to analyze the images of hydrate-containing core slices to construct a hydrate-containing digital core model, and use image-J software to analyze the images of hydrate-free core slices to construct a hydrate-free digital core model; S500: extracting the corresponding pore network model G1 from the hydrate-containing digital core model to obtain pore microscopic characterization parameters of the hydrate-containing digital core; extracting the corresponding pore network model G2 from the hydrate-free digital core model to obtain pore microscopic characterization parameters of the hydrate-free digital core; S600: Comparing G1 and G2 to obtain all hydrate microscopic characterization parameters in the hydrate-containing core, wherein the hydrate microscopic characterization parameters include the coordinate system position of each hydrate, the pixel volume of each hydrate, and the pixel surface area of ​​each hydrate; S700: Using all the hydrate microscopic characterization parameters obtained in S600, calculate and obtain a hydrate microscopic characterization parameter set. The specific steps are as follows: S710: Screening the volume of each hydrate. If the volume of the hydrate is less than 10 cubic pixels, the hydrate is considered as image noise and is removed. If the volume of the hydrate is greater than or equal to 10 cubic pixels, the hydrate is retained. S720: Using the existing hydrate microscopic characterization parameter calculation module to obtain a hydrate microscopic characterization parameter set, specifically: The hydrate microscopic characterization parameter calculation module includes two parts. The first part first screens each hydrate retained in S710. The second part calculates the corresponding hydrate microscopic characterization parameters for each hydrate screened in the first part. The microscopic characterization parameters of all hydrates calculated in the second part constitute the hydrate microscopic characterization parameter set. S800: Preset the learning rate and use the self-organizing map neural network machine learning method to calculate the hydrate microscopic characterization parameter set. The calculation result is the specific microscopic type of all hydrates retained in S720, that is, the specific microscopic type of each hydrate is classified into one of the pore-filling type, film-coated type, and cemented type.

2. The method for classifying natural gas hydrate types based on machine learning according to claim 1, wherein: The method used to obtain the pore microscopic characterization parameters in S500 is the watershed method and the maximum sphere method. The specific steps are as follows: S510: Segmenting the core slice image using a watershed method to obtain a pore class and a rock skeleton class, wherein the core slice image includes a hydrate-containing core slice image and a hydrate-free core slice image; S520: The maximum sphere method is used to analyze the pores and rock skeletons to obtain the microscopic characterization parameters of the pores.

3. The method for classifying natural gas hydrate types based on machine learning according to claim 2, wherein: The microscopic characterization parameters of each hydrate retained by calculation in S720 include the Euler number of the hydrate, the surface area ratio of the hydrate to the pore in which it is located, and the volume ratio of the hydrate to the pore in which it is located.

4. The method for classifying natural gas hydrate types based on machine learning according to claim 3, wherein: The specific process of the first part of S720 for screening each hydrate retained in S710 is as follows: Traverse all hydrates retained in S710, compare the position of each hydrate with the pore position and pore radius of the pore where it is located, and determine whether the hydrate exists in the pore: if the hydrate does not exist in the pore, exclude the hydrate and determine the next hydrate; if the hydrate exists in the pore, retain it; The judgment formula is as follows: Among them, x h 、y h 、z h Indicates the hydrate coordinate information of the hydrate-bearing core, x p 、y p 、z p represents the pore coordinate information of the core without hydrate, and R represents the pore radius of the core without hydrate.

5. The method for classifying natural gas hydrate types based on machine learning according to claim 4, characterized in that: The specific steps of calculating the hydrate microscopic characterization parameter set using the self-organizing map neural network machine learning method in S800 are: S810: The self-organizing map neural network includes an input layer and a competition layer; S820: using the hydrate microscopic characterization parameter set as input data of the input layer, and performing unsupervised clustering processing on the hydrate microscopic characterization parameter set by the input layer to obtain output data of the input layer; S830: The output data obtained in S820 is used as input data of the competition layer, and the output is the classification result of the hydrate microscopic type.

6. The method for classifying natural gas hydrate types based on machine learning according to claim 5, characterized in that: The learning rate preset in S800 is 0.

5.

7. A natural gas hydrate classification device based on machine learning, characterized in that: The method for classifying natural gas hydrate types based on machine learning according to any one of claims 1 to 6 adopts a natural gas hydrate type classification device based on machine learning, and the natural gas hydrate type classification device based on machine learning comprises: Core scanning module, model generation and analysis module, hydrate microscopic characterization parameter calculation module, and machine learning hydrate type classification module; The core scanning module is used to observe the microstructure inside the core and determine the microscopic type of hydrate contained in the core; the core scanning module is also used to perform CT scanning on the core under high pressure and low temperature conditions to obtain CT slice images; The model generation and analysis module is used to analyze CT slice images using image-J software to establish a hydrate-containing digital core model and a hydrate-free digital core model, and extract corresponding pore network models G1 and G2 from the hydrate-containing digital core model and the hydrate-free digital core model, respectively; analyze G1 and G2 to obtain pore-related characterization parameters contained in G1 and G2, respectively; The hydrate microscopic characterization parameter calculation module is used to compare G1 and G2 to obtain all hydrate microscopic characterization parameters in the hydrate-containing core, and calculate the hydrate microscopic characterization parameter set using all the obtained hydrate microscopic characterization parameters; The machine learning hydrate type classification module is used to calculate the hydrate microscopic characterization parameter set, and the calculation result is the specific microscopic type of hydrate classification, which is one of the pore-filling type, the film-coated type, and the cemented type.

Images