WO2022011894A1

WO2022011894A1 - Convolutional neural network-based modeling method and device for pore network model

Info

Publication number: WO2022011894A1
Application number: PCT/CN2020/126213
Authority: WO
Inventors: 王猛; 刘海波; 唐雁冰; 徐大年; 杨玉卿; 杨鑫; 刘志杰; 张志强; 李闽; 张国栋
Original assignee: 中海油田服务股份有限公司
Priority date: 2020-07-15
Filing date: 2020-11-03
Publication date: 2022-01-20
Also published as: CN111950192A; CN111950192B

Abstract

A convolutional neural network-based modeling method and device for a pore network model. The method comprises: obtaining a nuclear magnetic image of a rock core obtained by means of nuclear magnetic resonance imaging; obtaining a correlation length according to the nuclear magnetic image, and calculating a three-dimensional tensor convolution kernel according to the correlation length; obtaining a T2 spectrum of the rock core obtained by means of nuclear magnetic resonance, and obtaining pore-throat radius frequency distribution according to the T2 spectrum; forming an initial three-dimensional tensor data volume according to the pore-throat radius frequency distribution; generating, according to the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume and by using a convolutional neural network forward propagation algorithm, a three-dimensional tensor data volume conforming to the pore-throat radius frequency distribution of the rock core; constructing a disordered spatial structure of a pore network model according to the initial three-dimensional tensor data volume; and establishing the pore network model according to the three-dimensional tensor data volume conforming to the pore-throat radius frequency distribution of the rock core and the disordered space structure.

Description

Modeling method and device of pore network model based on convolutional neural network

technical field

The embodiments of the present disclosure relate to, but are not limited to, the field of petroleum logging, and in particular, relate to a method and device for modeling a pore network model based on a convolutional neural network.

Background technique

Oil and gas is a widely used and extremely important energy source in the world. With the continuous development of oil and gas resources in many countries, many conventional oil reservoirs have been developed on a large scale and have entered the middle and late stages of production. Considering the utilization of oil and gas resources in the future, it is necessary to deepen the research on unconventional oil and gas reservoirs with complex and diverse conditions. Therefore, by establishing a core-scale model, the development scale of the pore space inside the porous medium, the influence of the spatial distribution on the fluid seepage, the distribution law of the fluid in it, the interaction mechanism, etc. It is very important to carry out research on essential issues.

SUMMARY OF THE INVENTION

The following is an overview of the topics detailed in this article. This summary is not intended to limit the scope of protection of the claims.

An embodiment of the present disclosure provides a method for modeling a pore network model based on a convolutional neural network, including:

Obtain the nuclear magnetic image of the core obtained by nuclear magnetic resonance imaging;

The correlation length is obtained according to the nuclear magnetic image, and a three-dimensional tensor convolution kernel is calculated according to the correlation length; wherein, the correlation length represents the average radius of the preselected bright clusters in the nuclear magnetic image;

obtaining the T2 spectrum of the core obtained by nuclear magnetic resonance, and obtaining the frequency distribution of the pore throat radius according to the T2 spectrum; forming an initial three-dimensional tensor data volume according to the frequency distribution of the pore throat radius;

According to the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume, a convolutional neural network forward propagation algorithm is used to generate a three-dimensional tensor data volume that conforms to the frequency distribution of the core pore throat radius;

Construct the disordered spatial structure of the pore network model according to the initial 3D tensor data volume;

A pore network model is established according to the three-dimensional tensor data volume conforming to the frequency distribution of the core pore throat radius and the disordered spatial structure.

In an exemplary embodiment, the above method further includes:

Obtain core images obtained by micro-CT scanning;

Obtain the number of fractures and the direction of fractures according to the core image;

A pore network model with fracture distribution is established according to the pore network model, the number of fractures and the fracture direction.

In an exemplary embodiment, the calculating a three-dimensional tensor convolution kernel according to the correlation length includes:

Calculate the three-dimensional tensor convolution kernel E(h) according to the following formula:

E(h)=exp(-2h/L _c )

_{Wherein, h represents the distance from the spherical surface with (L x} , _Ly , L _z ) as the center and the radius less than or equal to Lc to the center in the three-dimensional coordinate system,

L _c represents the correlation length.

In an exemplary embodiment, the above method also has the following characteristics:

The T2 spectrum of the core obtained by the nuclear magnetic resonance logging is obtained, and the pore throat radius distribution is obtained according to the T2 spectrum, including:

Acquire the T2 spectrum of the core obtained by nuclear magnetic resonance; and convert the amplitude value of the T2 spectrum into the frequency distribution of the pore throat radius through a preset quantitative relationship.

In an exemplary embodiment,

The predetermined quantitative relationship r _m = cT _2m;

Wherein, r _m is the m-th pore throat radius, T _2m is the m-th amplitude value of the T2 spectrum, c is a preset conversion coefficient, and m is a positive integer.

In an exemplary embodiment,

The forming an initial three-dimensional tensor data volume according to the frequency distribution of the pore throat radius includes:

According to the frequency distribution of the pore throat radius, a three-dimensional stable random field is established through a random function to form an initial three-dimensional tensor data volume;

Wherein, the random function is the following log-normal distribution random function:

The mathematical expectation μ and the standard deviation σ are obtained by fitting the frequency distribution of the pore throat radius, and x represents the pore throat radius.

In an exemplary embodiment, according to the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume, a convolutional neural network forward propagation algorithm is used to generate a three-dimensional tensor that conforms to the frequency distribution of the core pore throat radius. Data body, including:

Perform tensor point multiplication with the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume in turn, and stack the results of the point multiplication in the order in the initial three-dimensional tensor data volume to generate a pore-throat radius that conforms to the core. A 3D tensor data volume of frequency distributions.

In an exemplary embodiment, the construction of the disordered spatial structure of the pore network model according to the initial three-dimensional tensor data volume includes:

Determine the number of nodes of the disordered space structure according to the data quantity of the initial three-dimensional tensor data volume;

Build a three-dimensional cube network including three directions of X, Y, and Z according to the number of nodes and the preset interval distance L between each node;

calculating the coordinates of each node in the three-dimensional cube network;

Determine whether there is tube bundle connection between each adjacent node in each direction, and assign the tube bundle radius;

Move each node coordinate by preset rules;

The disordered space structure is generated according to the three-dimensional cube network, the result of determining whether there is a tube bundle connected, the assigned tube bundle radius and the moved node coordinates.

In an exemplary embodiment,

Obtain core images obtained by micro-CT scanning;

Obtain the average roar length and coordination number according to the core image; wherein, the configuration number refers to the number of nodes with connected tube bundles between a node and an adjacent node;

The preset interval distance L between each node is the average roar length;

The determining whether there is tube bundle connection between each adjacent node in each direction includes:

Determine whether there is a tube bundle connection between each adjacent node in the X direction according to the preset first probability function;

Determine whether there is a tube bundle connection between each adjacent node in the Y direction according to the preset second probability function;

or,

Whether there is tube bundle communication between each adjacent node in each direction is determined according to the coordination number.

In an exemplary embodiment, the moving each node coordinate according to a preset rule includes:

Move each node coordinate (x, y, z) according to the following formula:

(x,y,z)=[(i-1)L±rand()%(0.5L),(j-1)L±rand()%(0.5L),(k-1)L±rand( )% (0.5L)]

Among them, i is the node serial number in the x direction, j is the node serial number in the y direction, k is the node serial number in the z direction, i, j and k are integers greater than 0, rand()%(0.5L) means randomly generated 0.5L Any integer in the range.

In an exemplary embodiment, establishing a pore network model according to the three-dimensional tensor data volume conforming to the frequency distribution of the core pore throat radius and the disordered spatial structure, including:

A pore network model is established by sequentially assigning the three-dimensional tensor data volume conforming to the frequency distribution of the pore throat radius of the core to the nodes of the disordered spatial structure.

The embodiments of the present disclosure also provide a modeling device for a pore network model based on a convolutional neural network, including: a memory and a processor;

the memory configured to store a program for modeling the pore network model of the core;

The processor is configured to read and execute the program for modeling the pore network model of the core, and execute the following modeling method:

In an exemplary embodiment, the processor, configured to read and execute the program for modeling the pore network model of the core, also executes the following modeling method:

Obtain core images obtained by micro-CT scanning;

E(h)=exp(-2h/L _c )

Among them, h represents the distance from the spherical surface of the three-dimensional coordinate system with (L _x , _Ly , L _z ) as the center of the sphere, and the radius is less than or equal to Lc to the spherical center,

L _c represents the correlation length.

In an exemplary embodiment, obtaining the T2 spectrum of the core obtained by nuclear magnetic resonance logging, and obtaining the pore throat radius distribution according to the T2 spectrum, includes:

An exemplary embodiment, the predetermined quantitative relationship r _m = cT _2m;

In an exemplary embodiment, the forming an initial three-dimensional tensor data volume according to the frequency distribution of the pore throat radius includes:

calculating the coordinates of each node in the three-dimensional cube network;

Move each node coordinate by preset rules;

In an exemplary embodiment, the processor, configured to read and execute the program for modeling the pore network model of the core, also executes the following modeling method: core image;

The preset interval distance L between each node is the average roar length;

or,

Move each node coordinate (x, y, z) according to the following formula:

(x,y,z)=[(i-1)L±rand()%(0.5L),(j-1)L±rand()%(0.5L),(k-1)L±rand( )% (0.5L)]

Among them, i is the node number in the x direction, j is the node number in the y direction, k is the node number in the z direction, i, j, and k are integers greater than 0, and rand()%(0.5L) means randomly generating 0.5L Any integer in the range.

Other aspects will become apparent upon reading and understanding of the drawings and detailed description.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flowchart of a method for modeling a pore network model based on a convolutional neural network according to an embodiment of the present disclosure.

FIG. 2 is an example of mercury intrusion pore throat distribution and relaxation distribution according to an embodiment of the present disclosure.

FIG. 3 is an example of a three-dimensional convolution kernel according to an embodiment of the present disclosure.

FIG. 4 is a schematic diagram of an example of generating a three-dimensional tensor data volume conforming to the frequency distribution of pore throat radius according to an embodiment of the present disclosure.

FIG. 5 is an example of a pore image reconstructed by a CT scan according to an embodiment of the present disclosure.

FIG. 6 is an example of a fracture image reconstructed by a CT scan according to an embodiment of the present disclosure.

FIG. 7 is an example of an ordered spatial structure according to an embodiment of the present disclosure.

FIG. 8 is an example of a disordered space structure according to an embodiment of the present disclosure.

FIG. 9 is an example of an image of a micropore-dissolution pore dual medium model of an embodiment of the present disclosure.

FIG. 10 is an example of an image of a triple medium model of micropore-dissolution-pore-fracture according to an embodiment of the disclosure.

FIG. 11 is a schematic diagram of an apparatus for modeling a pore network model based on a convolutional neural network according to an embodiment of the present disclosure.

detail

Hereinafter, the embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. It should be noted that the embodiments of the present disclosure and the features of the embodiments may be arbitrarily combined with each other under the condition of no conflict.

FIG. 1 is a flowchart of a method for modeling a pore network model based on a convolutional neural network according to an embodiment of the present disclosure. As shown in FIG. 1 , the modeling method in this embodiment includes steps S11-S16:

S11. Acquire the nuclear magnetic image of the core obtained by the nuclear magnetic resonance imaging.

S12, obtaining a correlation length according to the nuclear magnetic image, and calculating a three-dimensional tensor convolution kernel according to the correlation length;

S13. Obtain the T2 spectrum of the core obtained by nuclear magnetic resonance, and obtain the frequency distribution of the pore throat radius according to the T2 spectrum; form an initial three-dimensional tensor data volume according to the frequency distribution of the pore throat radius;

S14. According to the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume, a convolutional neural network forward propagation algorithm is used to generate a three-dimensional tensor data volume that conforms to the frequency distribution of the core pore throat radius;

S15, constructing the disordered spatial structure of the pore network model according to the initial three-dimensional tensor data volume;

S16. Establish a pore network model according to the three-dimensional tensor data volume conforming to the frequency distribution of the pore throat radius of the core and the disordered spatial structure.

In the above steps, S13 may be executed before or after S11 and S12, and may be executed in parallel with S11 and S12; the sequence of S14 and S15 is not limited, and may also be executed in parallel.

In an exemplary embodiment, it may further include:

Obtain core images obtained by micro-CT scanning;

Core images can be obtained without limitation by micro-CT scanning, so as to determine the number and direction of fractures. The number and direction of cracks can also be determined in other ways. For example, the number of cracks and the direction of cracks can be artificially set, or they can be determined by a random function.

The three-dimensional tensor convolution kernel E(h) can be calculated as follows:

E(h)=exp(-2h/L _c )

L _c represents the correlation length.

The three-dimensional tensor convolution kernel can be calculated using the above formula or a modified formula of this formula.

The preset quantitative relationship may be a preset corresponding relationship between the amplitude value of the T2 spectrum and the frequency distribution of the pore throat radius, or a preset calculation formula for calculating the pore throat radius through the amplitude value, or the like.

An exemplary embodiment includes:

The predetermined quantitative relationship r _m = cT _2m;

The random function is not limited to the above formula, and a logarithmic uniform random function can be used.

calculating the coordinates of each node in the three-dimensional cube network;

Move each node coordinate by preset rules;

In an exemplary embodiment, the method further includes,

Obtain core images obtained by micro-CT scanning;

The preset interval distance L between each node is the average roar length;

or,

Move each node coordinate (x, y, z) according to the following formula:

(x,y,z)=[(i-1)L±rand()%(0.5L),(j-1)L±rand()%(0.5L),(k-1)L±rand( )% (0.5L)]

In an exemplary embodiment, the nuclear magnetic resonance T2 spectrum and nuclear magnetic resonance imaging of the core can be obtained in the following manner.

Measure the target core. For the target core, the carbonate core collected from the formation is washed with oil and salt, and then fully dried at a temperature of 80 °C until the weight does not change. Using a vacuum pressure saturator, KCl2 brine was used as the medium, and the carbonate core was saturated for 48 hours and then the NMR measurement experiment was carried out to obtain the NMR image of the core.

Then, combining the nuclear magnetic T2 spectrum (obtained by the nuclear magnetic resonance experiment) with the core pore size distribution (obtained by the conventional rock mercury intrusion experiment), a conversion coefficient c (the value of which has regional experience) can be obtained. The pore size distribution is obtained by multiplying the coordinates by c, as shown in Figure 2.

Magnetic resonance imaging can be used to obtain cross-sectional, coronal, and sagittal images of rock samples. For dual-medium rocks, the image signal can be used to represent the distribution of single-phase fluid in the core space. The brighter the image pixel is, the larger the pore throat radius is, and the place where the bright color is concentrated in the image is displayed as a dissolved pore. Conversely, the darker the image, the more unrecognizable areas of the resolution, and the smaller the pore throat radius. Therefore, the distribution characteristics and correlation lengths of dissolved pores or dissolved pore development zones in the core can be observed from two-dimensional images.

In an exemplary embodiment, the convolution kernel can be generated as follows:

Figure 3 is an example of a convolution kernel. By acquiring the features of NMR images, the empirical relationship proposed by geostatistics in the past is:

E(h)=exp(-2h/L _c )

Lc—correlation length, which is obtained from the analysis of nuclear magnetic resonance images; h represents _{the distance from the spherical surface with (L x} , _Ly , L _z ) as the center in the three-dimensional coordinate system, and the radius is less than or equal to Lc to the center of the sphere,

For example, the three-dimensional convolution kernel can be implemented by the following procedure.

In an exemplary embodiment, a three-dimensional tensor data volume may be formed as follows:

According to the distribution characteristics of the pore throat radius extracted from the nuclear magnetic resonance T2 spectrum, a three-dimensional stable random field is established through a random function (logarithmic uniform or logarithmic normal, etc.) to form a three-dimensional tensor data volume. For example, the random function can be the following log-normally distributed random function:

Using the Tensorflow platform, three-dimensional convolution is performed with a three-dimensional tensor data volume as input. As shown in Figure 4, the convolution kernel is used to slide over each element of the input 3D stable random field tensor data volume, the results are formed into a new matrix, and finally the result matrices of all channels are stacked in the original order to form a third-order tensor, and use it as the output to get the 3D tensor data volume of pore throat radius, which contains the spatial information of the pore throat radius distribution of the (micro)pore-dissolution pore dual medium.

In an exemplary embodiment, a CT image of a carbonate rock with a triple medium of micro-pore-dissolved-pore-fracture can be generated as follows.

As shown in Figures 5 and 6, the samples were scanned based on computer high-resolution tomography imaging technology (ie MicroCT) and the digital core 3D reconstruction was carried out using the equivalent sphere method and the maximum sphere method to construct the hole network, respectively. Layers were analyzed for structural characteristics. Among them, the length of the roar can be calculated by the following formula:

L=DR ₁ -R ₂

In the formula, R1 and R2 are the radii of the two pores connected by the roar, respectively, in μm; D is the actual coordinate distance between the center points of the two pores, in μm.

The micro-CT experimental processing results are as follows: the resolution is 8 μm, the size is 710 × 710 × 710 μm, the color concentration is dissolved pores, the rest are pores, the analytical porosity is 1.12%, and the volume percentage is 71.3%, the average pore radius is 18.97 μm, the average roar radius is 17.8 μm, the average roar length is 131.8 μm, and the coordination number is 1.13. The red balls in Fig. 5 are pores; the white rods are the throats. The volume of a single connected pore in Fig. 6 is displayed by size and color scale. main. At the same time, the distribution of dissolved pores and fractures in carbonate rocks can be clearly observed, and based on this, the number and volume of dissolved pores, the number, direction, number-to-volume ratio of fractures, and the distribution of pore throat radius are calculated.

In an exemplary embodiment, the disordered space structure can be constructed as follows:

The modeling method of the disordered network model (as shown in Figure 8) is introduced with a square grid network model. The ordered network model (as shown in Figure 7) is characterized by a regular shape of the model and fixed positions of nodes. According to this characteristic, a modeling method of disordered network model is proposed.

(1) Specify the number of nodes in the model (the number of nodes corresponds to the number of data in the three-dimensional tensor data volume), and construct a three-dimensional simple cubic grid of X×Y×Z. Each node represents a pore, and the nodes are connected by a throat. In the network thus established, each node representing a pore is connected by six throats; similarly, each throat is also connected by six pores. The spacing distance between each node in each direction (ie, the x direction, the y direction, and the z direction) is set to L (the average roar length of 131.8 μm can be determined as the node spacing according to the results of the micro-CT experiment), and the number of nodes is set to d, the side length of the model is (d-1)×L.

(2) Calculate the coordinates of each node in the network model. The calculation formula is: (x, y, z)=[(i-1)l, (j-1)l, (k-1)l], where i, j, k are the x direction, y direction and The node number in the z direction, the values are 1, 2, 3... respectively.

(3) A probability function with probability p is set in the program, and a (pseudo) random number generator is used to determine whether there is a tube bundle connection between each adjacent node in the x direction. Use a function to generate random numbers, thereby generating random probabilities. In the C/C++ programming language, the rand() function can be used to generate random numbers, thereby generating random probabilities. The specific C/C++ code is:

if(rand()%100＜p×100)

In the formula, rand()%100—the computer randomly generates any integer in the range of 0-99.

When the penetration probability p is 50%, 50% of the integers randomly generated by the rand() function (probability) are less than 50, and another 50% of the integers (probability) are greater than 50. Therefore, this expression can realize the pipe bundle connection with probability p=50%, that is, when a number less than 50 is generated, the expression is true, and the assignment task of the pipe bundle radius is performed (the assignment pipe bundle radius is r); For numbers less than 50, the expression is false and no action is performed.

(4) Establish a connection probability function through a (pseudo) random number generator to determine whether there is a tube bundle connection between each adjacent node in the y and z directions. The method is the same as the tube bundle allocation process in the x-direction.

Instead of (3) and (4), the tube bundle connectivity of the nodes can also be determined by the coordination number obtained from the micro-CT experimental results.

(5) Move each node coordinate, and generate a disordered space structure according to the above-mentioned three-dimensional cube network, the result of determining whether there is tube bundle connection, the assigned tube bundle radius and the moved node coordinates.

Each node coordinate (x, y, z) can be moved according to the following formula:

(x,y,z)=[(i-1)L±rand()%(0.5L),(j-1)L±rand()%(0.5L),(k-1)L±rand( )% (0.5L)]

In an exemplary embodiment, a pore network model of a (micro)pore-dissolution pore dual media characteristic can be generated as follows.

By sequentially assigning the data in the 3D tensor data volume to the pore throat radius grid points of the disordered structure network model, a pore network model with (micro)pore-dissolution pore dual medium characteristics can be constructed (as shown in Figure 9). , model scale: length×width×height=3cm×1cm×1cm).

In an exemplary embodiment, the pore network model of the (micro)pore-dissolution-pore-fracture triple media feature can be generated as follows:

Insert a plane equation into the node of the pore network model with the characteristics of (micro)pore-dissolved pore dual medium that has been constructed above to generate two-dimensional plane fractures, and construct a (micro)pore-dissolution pore-fracture triple medium. Let n be the plane normal vector:

M(x, y, z), N(x0, y0, z0) are any two points on the plane, then there are:

The point-form equation of the plane is thus:

A(xx ₀ )+B(yy ₀ )+C(zz ₀ )=0

Spatial circular cracks can be generated by the following methods:

Take the random point on the normal vector as the center of the circle o(a,b,c), and a is a random number from 0 to A, b is a random number from 0 to B, and c is a random number from 0 to C. Calculate the spherical equation first:

(xa) ² +(yb) ² +(zc) ² =r ²

where the radius r is the radius of the sphere; then the equation of the plane where the circle is located satisfies

The circular equation (circular crack) can be obtained from the two. The number of plane equations determines the number of fractures. According to the results of CT experiments, the number of fractures i in the core with a diameter of 2.5cm and a length of 3cm for multiple samples is counted, and the fracture is roughly the direction; The point coordinates) control the crack angle and distribution position, and control the number of cracks generated by the number of cycles.

Use GPU parallel computing to call the workstation equipped with two TITAN graphics cards, insert the surface equation into the micropore-dissolved pore model, and generate the fractured medium as shown in Figure 11 (n is the normal vector, M and N are any two points on the plane) . In the CT experimental sample with a diameter of 2.5cm, a length of 4cm, and a resolution of 8μm, the number of statistical cracks (including micro-cracks) is 4, the distribution is random, and the direction of the cracks is random. The normal vector and the center point position are randomly generated and inserted. The random surface equation is used to obtain the final carbonate micropore-dissolution pore-fracture model (as shown in Figure 10).

The carbonate core model established in this example combines experimental methods and computer algorithms, and uses convolutional neural network to process the eigenvalues and correlation lengths of nuclear magnetic imaging images to generate models. Reservoir physical parameters, to a certain extent, solve the problem that the geological model established by the existing modeling technology lacks a certain physical meaning. Combined with the CT experimental analysis, the micro-pore-dissolved-pore-fracture network model is generated in the form of adding a random surface equation to the model, which solves the difficulty of modeling the triple medium of carbonate rocks.

FIG. 11 is a schematic diagram of an apparatus for modeling a pore network model based on a convolutional neural network according to an embodiment of the disclosure. The modeling device includes: a memory and a processor;

In an exemplary embodiment, the processor is configured to read and execute the program for modeling the pore network model of the core, and also execute the following modeling method:

Obtain core images obtained by micro-CT scanning;

E(h)=exp(-2h/L _c )

L _c represents the correlation length.

An exemplary embodiment, the predetermined relationship may be quantified r _m = cT _2m;

In an exemplary embodiment, the disordered spatial structure of the pore network model is constructed according to the initial three-dimensional tensor data volume, including:

calculating the coordinates of each node in the three-dimensional cube network;

Move each node coordinate by preset rules;

In an exemplary embodiment, the processor is configured to read and execute the program for modeling the pore network model of the core, and also execute the following modeling method: core image;

The preset interval distance L between each node is the average roar length;

or,

Move each node coordinate (x, y, z) according to the following formula:

(x,y,z)=[(i-1)L±rand()%(0.5L),(j-1)L±rand()%(0.5L),(k-1)L±rand( )% (0.5L)]

Those skilled in the art can understand that all or part of the steps in the above method can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disk. Optionally, all or part of the steps in the above embodiments may also be implemented using one or more integrated circuits. Correspondingly, the modules/units in the foregoing embodiments may be implemented in the form of hardware, and may also be implemented in the form of software functional modules. The present disclosure is not limited to any particular form of combination of hardware and software.

The present disclosure may also have other various embodiments. Without departing from the spirit and essence of the present disclosure, those skilled in the art can make various corresponding changes and modifications according to the present disclosure, but these corresponding changes and All modifications should fall within the protection scope of the appended claims of the present disclosure.

Claims

A modeling method for a pore network model based on a convolutional neural network, comprising:

Obtain the nuclear magnetic image of the core obtained by nuclear magnetic resonance imaging;

The correlation length is obtained according to the nuclear magnetic image, and a three-dimensional tensor convolution kernel is calculated according to the correlation length; wherein, the correlation length represents the average radius of the preselected bright clusters in the nuclear magnetic image;

obtaining the T2 spectrum of the core obtained by nuclear magnetic resonance, and obtaining the frequency distribution of the pore throat radius according to the T2 spectrum; forming an initial three-dimensional tensor data volume according to the frequency distribution of the pore throat radius;

According to the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume, a convolutional neural network forward propagation algorithm is used to generate a three-dimensional tensor data volume that conforms to the frequency distribution of the core pore throat radius;

Construct the disordered spatial structure of the pore network model according to the initial 3D tensor data volume;

A pore network model is established according to the three-dimensional tensor data volume conforming to the frequency distribution of the core pore throat radius and the disordered spatial structure.
The method of claim 1, further comprising:

Obtain core images obtained by micro-CT scanning;

Obtain the number of fractures and the direction of fractures according to the core image;

A pore network model with fracture distribution is established according to the pore network model, the number of fractures and the fracture direction.
The method of claim 1, wherein,

The calculation of the three-dimensional tensor convolution kernel according to the correlation length includes:

Calculate the three-dimensional tensor convolution kernel E(h) according to the following formula:

E(h)=exp(-2h/L c )

Wherein, h represents the distance from the spherical surface with (L x , Ly , L z ) as the center and the radius less than or equal to Lc to the center in the three-dimensional coordinate system,
L c represents the correlation length.
The method of claim 1, wherein,

The T2 spectrum of the core obtained by the nuclear magnetic resonance logging is obtained, and the pore throat radius distribution is obtained according to the T2 spectrum, including:

Acquire the T2 spectrum of the core obtained by nuclear magnetic resonance; convert the amplitude value of the T2 spectrum into the frequency distribution of the pore throat radius through a preset quantitative relationship.
The method of claim 4, wherein,

The predetermined quantitative relationship r m = cT 2m;

Wherein, r m is the m-th pore throat radius, T 2m is the m-th amplitude value of the T2 spectrum, c is a preset conversion coefficient, and m is a positive integer.
The method of claim 4, wherein,

The forming an initial three-dimensional tensor data volume according to the frequency distribution of the pore throat radius includes:

According to the frequency distribution of the pore throat radius, a three-dimensional stable random field is established through a random function to form an initial three-dimensional tensor data volume;

Wherein, the random function is the following log-normal distribution random function:

The mathematical expectation μ and the standard deviation σ are obtained by fitting the frequency distribution of the pore throat radius, and x represents the pore throat radius.
The method of claim 1, wherein,

The three-dimensional tensor data volume conforming to the frequency distribution of the pore throat radius of the core is generated by using the convolutional neural network forward propagation algorithm according to the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume, including:

Perform tensor point multiplication with the three-dimensional tensor convolution kernel and the initial three-dimensional tensor data volume in turn, and stack the results of the point multiplication in the order in the initial three-dimensional tensor data volume to generate a pore-throat radius that conforms to the core. A 3D tensor data volume of frequency distributions.
The method of claim 1, wherein,

The construction of the disordered spatial structure of the pore network model according to the initial three-dimensional tensor data volume includes:

Determine the number of nodes of the disordered space structure according to the data quantity of the initial three-dimensional tensor data volume;

Build a three-dimensional cube network including three directions of X, Y, and Z according to the number of nodes and the preset interval distance L between each node;

calculating the coordinates of each node in the three-dimensional cube network;

Determine whether there is tube bundle connection between each adjacent node in each direction, and assign the tube bundle radius;

Move each node coordinate by preset rules;

The disordered space structure is generated according to the three-dimensional cube network, the result of determining whether there is a tube bundle connected, the assigned tube bundle radius and the moved node coordinates.
The method of claim 8, further comprising:

Obtain core images obtained by micro-CT scanning;

Obtain the average roar length and coordination number according to the core image; wherein, the configuration number refers to the number of nodes with connected tube bundles between a node and an adjacent node;

The preset interval distance L between each node is the average roar length;

The determining whether there is tube bundle connection between each adjacent node in each direction includes:

Determine whether there is a tube bundle connection between each adjacent node in the X direction according to the preset first probability function;

Determine whether there is a tube bundle connection between each adjacent node in the Y direction according to the preset second probability function;

or,

Whether there is tube bundle communication between each adjacent node in each direction is determined according to the coordination number.
The method of claim 8 or 9, wherein,

The moving each node coordinate according to the preset rule includes:

Move each node coordinate (x, y, z) according to the following formula:

(x,y,z)=[(i-1)L±rand()%(0.5L),(j-1)L±rand()%(0.5L),(k-1)L±rand( )% (0.5L)]

Among them, i is the node number in the x direction, j is the node number in the y direction, k is the node number in the z direction, i, j, and k are integers greater than 0, and rand()%(0.5L) means randomly generating 0.5L Any integer in the range.
The method of claim 1, wherein,

A pore network model is established according to the three-dimensional tensor data volume conforming to the frequency distribution of the pore throat radius of the core and the disordered spatial structure, including:

A pore network model is established by sequentially assigning the three-dimensional tensor data volume conforming to the frequency distribution of the pore throat radius of the core to the nodes of the disordered spatial structure.
A modeling device for a pore network model based on a convolutional neural network, comprising: a memory and a processor;

the memory configured to store a program for modeling the pore network model of the core;

The processor, configured to read and execute the program for modeling the pore network model of the core, executes the modeling method according to any one of claims 1-11.