Intelligent construction method of shale oil and pore wall interaction energy database

Abstract

The application discloses a shale oil and pore wall surface interaction energy database intelligent construction method, and belongs to the technical field of database construction, and comprises the following steps: acquiring interaction energy data of representative fluid and representative pore wall surface in shale reservoir nanometer pores through a molecular dynamics calculation method; quantifying fluid and pore wall surface characteristic parameters to form neural network model original data; constructing a GAN data enhancement model based on a generative adversarial network; establishing a fluid and pore wall surface interaction energy prediction model in shale reservoir nanometer pores based on a full connection neural network; based on the prediction model, data completion is performed on the area not covered by the GAN data enhancement model to construct a complete interaction energy database, and a visual software system is established. The application is used for constructing a shale reservoir nanometer pore complex fluid component and multi-type pore wall surface interaction energy database, and greatly reduces the expensive cost of interaction theoretical simulation calculation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of database construction, and particularly relates to a shale oil and pore wall surface interaction energy database intelligent construction method. BACKGROUND

[0002] Shale oil is a multi-component fluid, which is usually divided into light components, medium components, heavy components and non-hydrocarbon components. The shale nanopore wall surface types are inorganic mineral wall surfaces and organic matter wall surfaces, the organic matter is mainly cheese root, and the inorganic minerals are mainly clay minerals such as quartz and kaolinite. Due to the differences in molecular structure and chemical properties of each fluid component, the interaction strength with the pore wall surface is different. Clarifying the action mechanism and action strength of the fluid and the pore wall surface is helpful to develop a reasonable and efficient shale oil and gas development plan. The action strength of the fluid and the pore wall surface is the adsorption of the fluid to the pore wall surface, and the interaction energy is one of the key parameters for evaluating the action strength, which can be estimated or calculated by various methods. The interaction energy between the fluid and the pore wall surface can be obtained by interaction theory simulation. In the field of computational chemistry, the interaction energy is calculated by molecular dynamics (MD). However, for complex fluid components and multiple types of pore wall surfaces, it is difficult to quickly obtain the interaction energy between each type of hydrocarbon fluid and each type of pore wall surface by using the molecular dynamics calculation method. Accurate molecular simulation theoretical calculation will consume a lot of cost and time.

[0003] At present, using machine learning to establish a targeted model can help solve the problem of calculating the interaction energy for a long time and cost. Machine learning is a series of statistical methods, which uses various data mining algorithms to obtain information from historical data and predict unknown data. The technical problem to be solved by the present application is to predict the interaction energy between the fluid and the organic matter and inorganic mineral pore wall surface in the shale reservoir, and then to construct the interaction strength database of the fluid and the organic matter and inorganic mineral wall surface in the shale reservoir nanopore. In the experimental study of shale nanopores, there is still no enough experimental means to directly observe and quantify the interaction energy between the fluid and the organic matter and inorganic mineral pore wall surface. In the theoretical calculation aspect, the MD method is usually used to analyze the interaction ability of the fluid component to the pore wall surface, but this method greatly consumes the calculation time and resources, and is no longer applicable to large-scale complex molecular simulation calculation. SUMMARY

[0004] In order to solve the above problems, the present application provides a shale oil and pore wall surface interaction energy database intelligent construction method, which takes the fluid and organic matter and inorganic mineral wall surface in the shale reservoir nanometer pore as the research object, quantifies the interaction energy between the fluid and the organic matter and the inorganic mineral wall surface in the shale reservoir nanometer pore on the basis of MD theoretical calculation, combines the generative adversarial network (GAN) and the fully connected neural network (FCNN) to establish a high-efficiency and high-precision interaction energy prediction model of the fluid and the organic matter and the inorganic mineral wall surface in the shale reservoir nanometer pore, and constructs a complex fluid component and multi-type pore wall surface interaction energy database in the shale reservoir nanometer pore. Then, combined with the statistical analysis of the constructed database, the interaction ability of the fluid to the wall surface in the shale reservoir nanometer pore is comprehensively evaluated, and the expensive cost of the interaction theoretical simulation calculation is greatly reduced. Finally, the software and the visualization module are established, the fluid characteristics and the wall surface characteristic data are input, the interaction energy data are quickly searched and the fluid and pore wall surface interaction strength image schematic diagram is output, the visualization display of the interaction strength is realized, and the more intuitive interaction strength effect is presented.

[0005] The technical scheme of the present application is as follows:

[0006] A shale oil and pore wall surface interaction energy database intelligent construction method adopts an intelligent algorithm of a generative adversarial network and a fully connected neural network to construct a database, and specifically includes the following steps:

[0007] Step 1: acquiring the interaction energy data of the representative fluid and the representative pore wall surface in the shale reservoir nanometer pore through a molecular dynamics calculation method;

[0008] Step 2: quantifying the fluid and pore wall surface characteristic parameters to form neural network model original data;

[0009] Step 3: constructing a GAN data enhancement model based on a generative adversarial network, generating new samples with high quality similar to the original data, and expanding the fluid and pore wall surface interaction energy data set;

[0010] Step 4: establishing a shale reservoir nanometer pore fluid and pore wall surface interaction energy prediction model based on a fully connected neural network;

[0011] Step 5: based on the prediction model, data completion is performed on the area not covered by the GAN data enhancement model, a complete interaction energy database is constructed, and a visualization software system is established.

[0012] Further, the specific process of step 1 is as follows:

[0013] Step 1.1, establish a multi-component shale oil molecular model and a nanopore molecular model, combine the molecular model of shale oil adsorbed in the nanopore, and perform energy minimization and relaxation; the specific process is as follows:

[0014] Step 1.1.1, establish a multi-component shale oil molecular model based on the Ligpargen website;

[0015] First, use the Ligpargen website to establish methane, n-dodecane, asphaltene molecular model, n-decylamine molecular model respectively, after charge balance optimization, export four corresponding.lmp type files and four.xyz type files of each model; then, format conversion is performed on the four.lmp type files, and four.lt type files are obtained; at the same time, four.xyz type is based on methane, n-dodecane, asphaltene, n-decylamine and rock pore size to construct a random distribution file; finally, the random distribution file and the four.lt type files are fused to generate a multi-component shale oil molecular model;

[0016] Step 1.1.2, use Materials Studio software to establish shale nanopore molecular model; the shale nanopore molecular model includes shale kaolinite nanopore molecular model, shale quartz nanopore molecular model, shale calcite nanopore molecular model and shale type II-C kerogen nanopore molecular model;

[0017] First, establish shale kaolinite nanopore molecular model, call the unit cell in Materials Studio software, use the Bulid module in Materials Studio software to cut the crystal surface of kaolinite, obtain the kaolinite unit cell model, optimize the lattice structure parameters of the kaolinite unit cell model, and generate the kaolinite unit cell model file; import the kaolinite unit cell model file into the Lammps molecular simulator for file format conversion to obtain the kaolinite unit cell data file, set the ClayFF force field parameters of the kaolinite mineral, apply the ClayFF force field parameters to the kaolinite unit cell model, and construct the shale kaolinite nanopore molecular model; then, shale quartz nanopore molecular model, shale calcite nanopore molecular model and shale type II-C kerogen nanopore molecular model are established in the same way;

[0018] Step 1.1.3, combine the shale nanopore molecular model with the multi-component shale oil molecular model to obtain a multi-component shale oil nanopore adsorption initial model, perform energy minimization and relaxation on the multi-component shale oil nanopore adsorption initial model, and obtain a multi-component shale oil nanopore adsorption model;

[0019] Firstly, shale kaolinite nanopore molecular model and multi-component shale oil molecular model are combined to form a multi-component shale oil nanopore adsorption initial model, and the energy minimization process is performed on the multi-component shale oil nanopore adsorption initial model to eliminate the overlapping atomic configuration, so that the energy of the multi-component shale oil nanopore adsorption initial model is minimized, and the energy of the multi-component shale oil nanopore adsorption initial model in the energy minimized state is obtained, and the multi-component shale oil nanopore adsorption initial model after energy minimization processing is obtained; then the multi-component shale oil nanopore adsorption initial model after energy minimization processing is optimized, the relaxation step and relaxation time of the multi-component shale oil nanopore adsorption initial model are set, and the multi-component shale oil nanopore adsorption initial model is relaxed under the NVE canonical ensemble with the preset relaxation step and relaxation time, the system temperature value of the multi-component shale oil nanopore adsorption initial model in the stable state is obtained and is consistent with the real formation environment temperature through the relaxation processing, and the multi-component shale oil nanopore adsorption model is obtained; then, shale quartz nanopore molecular model, shale calcite nanopore molecular model and shale II-C type kerogen nanopore molecular model are combined with multi-component shale oil molecular model in turn, and the multi-component shale oil nanopore adsorption models of quartz, calcite and II-C type kerogen are established in the same way;

[0020] Step 1.2, the equilibrium state molecular trajectory of multi-component shale oil in multi-component shale oil nanopore adsorption in the equilibrium state is obtained by using molecular dynamics method, and the adsorption density distribution of each component shale oil in the nanopore is calculated; the specific process is as follows:

[0021] Firstly, based on the molecular dynamics method, the adsorption configuration of shale oil in the kaolinite nanopore in the equilibrium state is obtained by running 10 nanoseconds with a step of 1 femtosecond; further, the adsorption configuration of shale oil in the kaolinite nanopore in the equilibrium state is maintained under the original simulation conditions for 2 nanoseconds, the density distribution data is obtained, and 400 frame molecular trajectory files are outputted with a step of 500 femtoseconds and saved;

[0022] In the process of simulating and analyzing the density distribution data, data statistics boxes are established along the z direction at an interval of 0.025 nm, and the average value of the mass distribution of the multi-component shale oil is taken at 1 femtosecond according to formula (1) for 2000000 groups; the adsorption density distribution curve of each component of shale oil in the nanopore is obtained;

[0023]

[0024] In the formula: ρ ij is the average density of the jth component of shale oil in the ith data statistics box; Nij is the number of the j component in the i data statistical box in shale oil; m j is the relative molecular mass of the j component in shale oil; x is the size of the system in the x direction; y is the size of the system in the y direction; h i is the height of the i data statistical box; N A is the Avogadro constant; then, the molecular trajectory files of quartz, calcite and type II-C kerogen are generated in the same process;

[0025] Step 1.3, setting the force field parameters of shale oil and the force field parameters of nanometer pore, calculating the interaction energy of each component of multi-component shale oil with nanometer pore, and obtaining the interaction energy data of each component of multi-component shale oil;

[0026] The force field parameters of shale oil and the force field parameters of nanometer pore include interatomic parameters, bond parameters and angle parameters; based on the obtained adsorption configuration in nanometer pore and the molecular trajectory file, first, the Coulomb interaction parameters in the adsorption configuration file in the nanometer pore are deleted; then, the average value of 400 groups of data is obtained by taking 500 femtoseconds as a step to obtain the van der Waals interaction energy of each component of shale oil with pore; based on multi-component shale oil, the geological characteristics of multiple shale oil reservoirs are statistically classified and analyzed; considering the density of shale oil, the environmental temperature and pressure conditions, and the nanometer pore matrix, the above-mentioned molecular dynamics simulation is used to obtain the solid-liquid interaction energy data, which is the required interaction energy data of representative fluid and representative pore wall in nanometer pore of shale oil reservoir, and finally a small sample data set of fluid and pore wall interaction energy is formed.

[0027] Further, the specific process of step 2 is: first, according to the interaction energy data of each component of multi-component shale oil, based on the pore characteristics obtained by SEM scanning experiment of multi-component shale matrix; then, the interaction energy of pore wall to fluid is sorted according to the strength, which is mapped into the interval of 0-1, so as to realize the quantification of pore wall characteristics.

[0028] Further, in step 3, the process of data augmentation of the GAN data augmentation model is as follows: the original input data O includes temperature, pore diameter, wall type, shale oil component, pressure, and interaction energy; the GAN data augmentation model first normalizes the original input data O into x1, x2, x3, x4, x5, and x6, the generator projects the original low-dimensional vector into a high-dimensional vector X1', X2', X3', X4', X5', and X6' consistent with the dimension of the original data, and then the discriminator judges the distribution similarity between the normalized original data O and the generated high-dimensional vector. After continuous training of the generator and the discriminator, the data with similar distribution is finally generated, and the generated data O' is obtained through inverse normalization, realizing the data augmentation of the interaction energy between the fluid and the pore wall in the shale oil reservoir nanometer pore, and forming a new sample with high quality similar to the original data.

[0029] The input layer of the generator is a Dense layer, the size of the input data is the product of the number of input data categories and the batch size, and is connected with a LeakyReLU activation function layer; the hidden layer is 3 layers of Dense layers, the activation function is tanh, the first layer has 20 neurons, the second layer has 10 neurons, and the third layer has 6 neurons, and is finally connected with a LeakyReLU activation function layer; the output layer is a Reshape layer, which normalizes the feature value size to 6 categories, and completes data augmentation.

[0030] The input layer of the discriminator is a Flatten layer, which expands all features; the hidden layer is a Dense layer with 50 neurons, connected with a ReLU activation function layer; the output layer is a Dense layer with 1 neuron, connected with a ReLU activation function layer, to judge the authenticity of the output data.

[0031] Further, in step 4, the prediction model based on the fully connected neural network includes an input layer, a hidden layer, an output layer, and an activation function; the input layer is a Dense layer responsible for receiving original data, the input data has 5 dimensions, including temperature, pore diameter, wall type, shale oil component, and pressure, and the number of neurons is 5; the hidden layer is the intermediate layer in the fully connected neural network, the number of neurons in the hidden layer is freely set, each neuron is connected with all neurons in the previous layer and the next layer, the hidden layer is set as 5 layers of Dense layers, the activation function of the first 4 layers is tanh, and the last layer does not add an activation function, and the number of neurons in each layer is 20; the output layer is the last layer of the fully connected neural network, responsible for outputting the prediction result of the network, the prediction result is the interaction energy, and the output layer is set as a Dense layer with 1 neuron.

[0032] Further, in step 5, the data set expanded by the GAN data enhancement model is divided into two disjoint sets of training set and validation set, the interaction energy prediction model is first trained with the training set, and then the prediction model is used to predict the interaction energy of the validation set, and the trained model is evaluated; the interaction energy is predicted by the trained prediction model, the data of the area not covered by the GAN data enhancement model is completed, and then a complete interaction energy database is constructed;

[0033] The mean square error (MSE) is selected for model evaluation, and the average value thereof is taken as the performance index for evaluating the final model; the calculation method of the MSE is as follows:

[0034]

[0035] wherein h(x i ) is the model prediction value, y i is the true value, i represents the ith sample, and n represents the total number of samples.

[0036] Further, in step 5, a visual software system, shale system, is established based on MATLAB; the shale system includes three functional modules: basic physical property database module, solid-liquid interaction strength database module, and state equation module; wherein the basic physical property database module is used to query the related physical properties of shale oil and reservoirs from different places, and the SEM scanning electron microscope images of the reservoir cores are visualized; the solid-liquid interaction strength database module quantitatively characterizes the interaction energy, average force potential and number density distribution of the solid-liquid interaction under different fluid properties, pore wall properties and reservoir conditions, and the micro model is visualized; the state equation module realizes the calculation of the compressibility factor and fugacity coefficient of single-component pure substances and binary mixtures based on the PR state equation; the shale system preliminarily evaluates the mobility of shale oil, the advantages and disadvantages of reservoir properties and the calculation of fluid state parameters through the above three functional modules.

[0037] Further, in the basic physical property database module, the basic physical properties of shale oil and the basic physical properties of reservoir rocks are automatically displayed by selecting the source of shale oil; the basic physical properties of shale oil include shale oil density, shale oil composition and shale oil carbon number distribution, the shale oil carbon number distribution is a curve graph with the number of carbon atoms as the horizontal coordinate and the mass fraction as the vertical coordinate; the basic physical properties of reservoir rocks include TOC content, organic matter type, pore type, porosity and main mineral content, and the SEM scanning electron microscope images of the reservoir rocks are visualized;

[0038] In the solid-liquid interaction intensity database module, fluid properties, nanopore wall properties, reservoir conditions, PMF display of the fluid and the wall object, PMF distribution, and number density distribution are included; the fluid properties include mass fractions of light, medium and heavy components; the nanopore wall properties include upper and lower different wall types; the reservoir conditions include temperature and pressure of shale oil and aperture size of a slit pore; in the PMF display of the fluid and the wall object, according to the set fluid type, wall type, and adsorption energy of the upper and lower walls, a current PMF perspective view is displayed; the PMF distribution and the number density distribution are both curve graph settings;

[0039] In the state equation module, two groups of single-component pure substance related parameters of component 1 and component 2, temperature, pressure, mole fraction of component 1, mole fraction of component 2, compressibility factor, phase state, and fugacity coefficient are included; each group of single-component pure substance related parameters includes critical temperature, critical pressure, and eccentric factor; the state equation module can calculate both single-component pure substance related parameters and double-component binary mixture related parameters; in the single-component pure substance related parameter calculation, the critical temperature, critical pressure, and eccentric factor parameter values of component 1 are set first, and the critical temperature, critical pressure, and eccentric factor parameter values of component 2 are all set to 0, then the temperature and pressure values of the state to be calculated are set, and finally the compressibility factor, fugacity coefficient, and the phase state of the substance are calculated; the mole fraction of component 1 and the mole fraction of component 2 are also set to 0; in the double-component binary mixture related parameter calculation, the critical temperature, critical pressure, and eccentric factor parameters of component 1 and component 2 are set respectively, and the mole fraction values of component 1 and component 2 are set, then the temperature and pressure values of the state to be calculated are set, and finally the compressibility factor, fugacity coefficient, and the phase state of the substance are calculated.

[0040] The beneficial technical effects brought by the present application are as follows:

[0041] 1. For the problem that a machine learning model needs a large amount of data for training, and the cost of using the MD method to calculate fluid and pore wall interaction energy data is high, resulting in data scarcity and low prediction accuracy of the machine learning model, the present application uses the GAN algorithm to learn the sample distribution rule of the fluid and pore wall interaction energy data in the real shale oil reservoir nanopore based on the small sample data calculated by the MD method, to generate new samples of high quality similar to the original data, realize data enhancement and expansion, and improve the accuracy of the subsequent prediction model.

[0042] 2. Existing experimental methods cannot directly observe and quantify the interaction energy between fluids and pore walls in nanopores. While MD theory calculations, combining fundamental theories such as geochemistry and molecular thermodynamics, can obtain the interaction energy between fluids and pore walls, they are computationally expensive and time-consuming, making it impossible to quickly obtain the interaction energy between various types of hydrocarbon fluids and pore walls in shale reservoir nanopores. Since traditional linear methods require the assumption of a linear relationship between the prediction target and the descriptor, which is unsuitable for the practical application of this invention, this invention uses the FCNN algorithm to construct a prediction model for the interaction energy between fluids and pore walls in shale reservoir nanopores. This significantly reduces computation time and cost, enabling rapid and efficient acquisition of interaction energy data between various types of fluids and pore walls in shale reservoir nanopores.

[0043] 3. Based on the constructed fluid-pore wall interaction energy prediction model, the interaction energy between fluid types and pore wall types not covered in the small sample raw data calculated based on the MD method is quantitatively evaluated, forming a database of fluid-pore wall interaction intensity in shale reservoir nanopores.

[0044] 4. Establish an operating software interface that inputs shale reservoir fluid characteristics and pore wall characteristics data, and outputs quantified interaction energy data to achieve rapid and efficient acquisition of interaction intensity parameters. Combined with molecular simulation image data, add a module illustrating the interaction intensity between the fluid and pore walls to visualize the interaction intensity. Compared to simply building a machine learning prediction model, constructing an interaction intensity database enables more efficient data retrieval and presents a more intuitive understanding of the interaction intensity.

[0045] 5. This invention is not only applicable to the construction of a database of fluid-pore wall interaction energies in nanopores of shale oil reservoirs, but also applicable to the construction of other databases related to molecular adsorption interaction energies. Attached Figure Description

[0046] Figure 1 This is a flowchart of the intelligent database construction method for the interaction energy between shale oil and pore walls according to the present invention.

[0047] Figure 2 This is the quantification model of pore wall features in this invention.

[0048] Figure 3 This is a schematic diagram of the structure of the GAN model in this invention.

[0049] Figure 4 This is a schematic diagram of the interaction energy data enhancement model corresponding to various shale oil components and different wall types under the conditions of 30MPa, 333K, and 5nm in this invention.

[0050] Figure 5 This is a flowchart of the data augmentation process of the GAN data augmentation model in this invention.

[0051] Figure 6 This is a schematic diagram of the structure of the fully connected neural network in this invention. Detailed Implementation

[0052] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0053] like Figure 1 As shown, an intelligent method for constructing a database of the interaction intensity between shale oil and pore walls is proposed. This method employs intelligent algorithms based on generative adversarial networks and fully connected neural networks to construct the database, specifically including the following steps:

[0054] Step 1: Obtain the interaction energy data between representative fluids and representative pore walls in the nanopores of shale oil reservoirs using molecular dynamics calculations. The specific process is as follows:

[0055] Step 1.1: First, establish a multi-component shale oil molecular model and a nanopore molecular model, combine the molecular model of shale oil adsorption in nanopores, and perform energy minimization and relaxation. The specific process is as follows:

[0056] Step 1.1.1: Establish a multi-component shale oil molecular model based on the Ligpargen website. The specific process is as follows:

[0057] First, using the Ligpargen website, molecular models of methane, n-dodecane, asphaltenes, and n-decylamine were established. After charge balance optimization, four .lmp files (methane.lmp, n-dodecane.lmp, asparatene.lmp, and n-decylamine.lmp) and four .xyz files (methane.xyz, n-dodecane.xyz, asparatene.xyz, and n-decylamine.xyz) were exported.

[0058] Then, the four.lmp type files of methane.lmp, n-dodecane.lmp, asphaltene.lmp and n-decylamine.lmp are all imported into the moltemplate package for format conversion, and four.lt type files of methane.lt, n-dodecane.lt, asphaltene.lt and n-decylamine.lt are obtained. The specific implementation command for format conversion is: ltemplify.py-name methane methane.lmp>methane.lt, ltemplify.py-name n-dodecane n-dodecane.lmp>n-dodecane.lt, ltemplify.py-name asphaltene asphaltene.lmp>asphaltene.lt, ltemplify.py-name n-decylamine n-decylamine.lmp>n-decylamine.lt.

[0059] At the same time, the four.xyz type files of methane.xyz, n-dodecane.xyz, asphaltene.xyz and n-decylamine.xyz are all imported into the Pcakmol package, and the randomly distributed oil.xyz file is constructed based on methane, n-dodecane, asphaltene, n-decylamine and rock pore size.

[0060] For example, taking 30 MPa, 333 K and 5 nm pore size as an example, when calculating the van der Waals interaction energy of each component shale oil and kaolinite pore, based on the rock pore size of 990 methane, 809 n-dodecane, 75 asphaltene, 86 n-decylamine and 5.92 nm x 8.81 nm x 9.39 nm in Table 1, the randomly distributed oil.xyz file is constructed.

[0061] Table 1 Multi-component shale oil component proportion data

[0062]

[0063] Finally, the oil.xyz file and the four.lt type files of methane.lt, n-dodecane.lt, asphaltene.lt and n-decylamine.lt obtained in step 1.1.1.2 are imported into the moltemplate package to generate the oil.data file, and the oil.data file is the final established multi-component shale oil molecular model. The specific implementation command for generating oil.data is: moltemplate.sh-xyz oil.xyz system.lt.

[0064] Step 1.1.2, using Materials Studio software, establishes a shale nanopore molecular model. The specific process is as follows:

[0065] The shale nanopore molecular model includes shale kaolinite nanopore molecular model, shale quartz nanopore molecular model, shale calcite nanopore molecular model and shale type II-C kerogen nanopore molecular model.

[0066] First, the shale kaolinite nanopore molecular model is established, the unit cell in the Materials Studio software is called, the (001) crystal surface of the kaolinite unit cell is cut using the Bulid module in the Materials Studio software, the model is energy minimized, the kaolinite unit cell model is obtained, the lattice structure parameters of the kaolinite unit cell model are optimized, the kaolinite unit cell model file is generated and exported, and the kaolinite unit cell model file includes two files of kaolinite.car and kaolinite.mdf. The optimized lattice structure parameters are: the length of the unit cell a = 0.51535 nm, the width of the unit cell b = 0.89419 nm, the height of the unit cell c = 0.73906 nm, angle α = 91.926°, angle β = 105.046°, angle γ = 89.797°.

[0067] The two files of kaolinite.car and kaolinite.mdf are imported into the msi2lmp package of the Lammps molecular simulator for file format conversion, and the subsequent Lammps molecular simulator readable kaolinite.data file of the kaolinite unit cell data file is obtained. The ClayFF force field parameters of the kaolinite mineral are set, the ClayFF force field parameters are applied to the kaolinite unit cell model, and the shale kaolinite nanopore molecular model is constructed. The specific implementation command for file format conversion is:. / msi2lmp.exe kaolinite-class I-frc cvff-i>data.kaolinite.

[0068] Then shale quartz nanometer pore molecular model, shale calcite nanometer pore molecular model and shale type II-C kerogen nanometer pore molecular model are established in the same way.

[0069] Step 1.1.3, shale nanometer pore molecular model is combined with multi-component shale oil molecular model to obtain a multi-component shale oil nanometer pore adsorption initial model, and the multi-component shale oil nanometer pore adsorption initial model is subjected to energy minimization and relaxation treatment to obtain a multi-component shale oil nanometer pore adsorption model.

[0070] First, shale kaolinite nanometer pore molecular model is combined with multi-component shale oil molecular model to form a multi-component shale oil nanometer pore adsorption initial model, and the multi-component shale oil nanometer pore adsorption initial model is subjected to energy minimization treatment to eliminate overlapping atomic configurations, so that the multi-component shale oil nanometer pore adsorption initial model is energy minimized, and the energy of the multi-component shale oil nanometer pore adsorption initial model in the energy minimized state is obtained to obtain the multi-component shale oil nanometer pore adsorption initial model after energy minimization treatment; then the multi-component shale oil nanometer pore adsorption initial model after energy minimization treatment is optimized, the relaxation step and relaxation time of the multi-component shale oil nanometer pore adsorption initial model are set, and the multi-component shale oil nanometer pore adsorption initial model is subjected to relaxation treatment under NVE canonical ensemble with the preset relaxation step and relaxation time, the system temperature value of the multi-component shale oil nanometer pore adsorption initial model in the stable state is obtained and compared with the real formation environment temperature, the system temperature value of the multi-component shale oil nanometer pore adsorption initial model in the stable state is consistent with the real formation environment temperature through the relaxation treatment, and a multi-component shale oil nanometer pore adsorption model is obtained.

[0071] Then, shale quartz nanometer pore molecular model, shale calcite nanometer pore molecular model and shale type II-C kerogen nanometer pore molecular model are combined with multi-component shale oil molecular model in sequence, and multi-component shale oil nanometer pore adsorption models of quartz, calcite and type II-C kerogen are established in the same way.

[0072] Take kaolinite as an example, the specific implementation process is: the kaolinite cell data file kaolinite.data generated in step 1.1.2 and the multi-component shale oil molecular model file oil.data generated in step 1.1.1 are combined into a multi-component shale oil nanopore adsorption initial model data file adsorption.data through the Lammps molecular simulator, and visualized into a multi-component shale oil nanopore adsorption initial model through OVITO software. Set 1.1 nm as the force field cutoff radius, that is, the interaction between two atoms outside the cutoff radius can be ignored, the purpose is to reduce the calculation time. And take the periodic boundary in the xyz direction of the shale oil nanopore adsorption model, that is, when the molecules move out of the model boundary, the same number of molecules will return to the model from the opposite interface, so as to ensure the constant particle number of the simulation system, which can effectively reduce the boundary effect caused by the size limitation of the simulation system. The overlapping atomic configuration in the adsorption.data file is eliminated through energy minimization, and the energy of the system is reduced. Finally, after the system energy is stable, the system with minimized structure energy and the multi-component shale oil nanopore adsorption model file min_adsorption.data under the energy minimization state are obtained. At this time, only the preliminary system configuration stabilization is completed by moving the system atoms, and the initial temperature of the system deviates greatly from the actual value. Therefore, the energy minimization file min_adsorption.data is further run at a step of 1 fs for 0.5 ns under an NVT ensemble at a temperature of 333 K, and finally the energy minimized and relaxed multi-component shale oil nanopore adsorption model file relaxation_adsorption.data is obtained. Finally, the temperature of the shale oil is stabilized at 333 K, which is consistent with the occurrence environment of the multi-component shale oil in the underground real environment, thereby enhancing the accuracy of the calculation data.

[0073] Step 1.2, using molecular dynamics method to obtain the equilibrium state of multi-component shale oil in multi-component shale oil nanopore adsorption equilibrium state molecular trajectory, and calculate the adsorption density distribution of each component shale oil in nanopore. The specific process is:

[0074] Firstly, based on the molecular dynamics method, the data file over_adsorption.data of the adsorption configuration of shale oil in kaolinite nanopore at equilibrium state is obtained by running 10 nanoseconds at a step of 1 femtosecond. Further, the adsorption configuration of shale oil in kaolinite nanopore at equilibrium state is maintained under the original simulation condition for 2 nanoseconds simulation, the density distribution data is obtained, and 400 frames of molecular trajectory file system.lammpstrj is outputted at a step of 500 femtoseconds and saved.

[0075] In the process of simulating the acquisition of the analysis density distribution data, the data statistical boxes of the multi-component shale oil in the space along the z direction are established at the interval of 0.025 nm, and 2000000 groups of the average values of the mass distribution of the multi-component shale oil are taken according to the formula (1) with 1 femtosecond as a step. The adsorption density distribution curve of each component of the shale oil in the nanopore is obtained.

[0076]

[0077] In the formula, ρ ij is the average density of the jth component of the shale oil in the ith data statistical box; N ij is the number of the jth component of the shale oil in the ith data statistical box; m j is the relative molecular mass of the jth component of the shale oil; x is the size of the system in the x direction; y is the size of the system in the y direction; h i is the height of the ith data statistical box; N A is the Avogadro constant. Then, the molecular trajectory files of the quartz, calcite and type II-C kerogen are generated according to the same process.

[0078] The process of calculating the adsorption density distribution of each component of the shale oil in the nanopore can be realized by the corresponding command of the internal Lammps molecular simulator, and the specific command is as follows: fix 1 oil ave / chunk 1000 500 500000 cc1 density / number file density_oil.txt.

[0079] Step 1.3, setting the force field parameters of the shale oil and the force field parameters of the nanopore, calculating the interaction energy of each component of the multi-component shale oil with the nanopore, and obtaining the interaction energy data of each component of the multi-component shale oil.

[0080] The force field parameters of the shale oil and the force field parameters of the nanopore include the interatomic parameters, the bond parameters and the angle parameters, and the specific settings are shown in Tables 2-4.

[0081] Table 2 Interatomic parameters

[0082]

[0083]

[0084] Table 3 Bond parameters

[0085]

[0086] Table 4 Angle parameters

[0087]

[0088] Based on the obtained adsorption configuration in nanopore and molecular trajectory file, first, delete the columbus interaction parameter in the adsorption configuration file in nanopore. Then take the way of reproducing molecular trajectory, take 400 groups of data with 500 femtoseconds as step length to obtain the average value of each group of shale oil and pore van der waals interaction energy. Based on the multi-component shale oil, the geological characteristics of the multiple shale oil reservoirs are statistically classified and analyzed. Considering the shale oil density, the storage environment temperature and pressure conditions, the nanopore matrix four factors are considered to obtain the solid-liquid interaction energy data by the above-mentioned molecular dynamics simulation, which is the required shale oil reservoir nanopore representative fluid (light, medium, heavy shale oil) and the interaction energy data of the representative pore wall (quartz, calcite, kaolinite, II-C kerogen), and finally form the small sample data set of fluid and pore wall interaction energy.

[0089] The process of calculating the interaction energy of each component of the multi-component shale oil and the nanopore can be realized by the corresponding command inside the Lammps molecular simulator, and the specific command is: fix en all ave / time 1 100000 100000c_ENERGY file energy_${TEMP}.txt.

[0090] The shale is rich in a large number of nanoscale pores, and the multi-component shale oil storage characteristics are taken as an example for research by the method of molecular dynamics simulation. The present application takes pressure 30MPa, temperature 333K, pore size 5nm as an example, calculates the pore characteristics based on the multi-component shale matrix SEM scanning experiment, and constructs the slit-shaped inorganic nanopore and the round hole-shaped organic nanopore, wherein the inorganic pore matrix is represented by kaolinite, quartz and calcite, and the organic pore is represented by II-C type kerogen as a monomer. The model is completed by the method of simulated annealing. According to the multi-component characteristics of the multi-component shale oil, the shale oil molecular model containing multiple components is constructed according to the same proportion of each component and the equal total density principle. By using the equilibrium state molecular dynamics simulation method, the solid-liquid interaction characteristics such as shale oil density distribution, interaction energy and state change under multi-group temperature and pressure formation conditions are investigated, and the interaction energy data of the multi-component shale oil in the multi-type nanopore are obtained as shown in Table 5.

[0091] Table 5 Interaction energy data of each component of multi-component shale oil

[0092]

[0093] Step 2, quantifying the fluid and pore wall characteristic parameters to form the neural network model original data. The specific process is as follows:

[0094] The pore wall characteristic quantization model is as follows: Figure 2As shown, first, according to the multi-component shale oil component interaction energy data, based on the pore characteristics obtained by SEM scanning experiment of multi-component shale matrix. Then, the interaction energy of the pore wall surface to the fluid is sorted according to the strength, which is mapped to the interval of 0, 1, to realize the quantification of the pore wall surface characteristics. The quantification process makes the wall surface conditions become numerical parameters, optimizes the input features of the neural network model, improves the calculation efficiency on the premise of considering the influence factors of fluid and wall surface interaction energy.

[0095] Step 3, based on the generative adversarial network, a GAN data augmentation model is constructed to generate new samples with high quality similar to the original data, and expand the fluid and pore wall surface interaction energy data set.

[0096] The input features of the GAN data augmentation model include temperature, pressure, pore size, wall surface condition, shale oil component and interaction energy value, wherein the shale oil component is represented by the density of shale oil, which is divided into light, medium and heavy components.

[0097] The training data of the generative adversarial network is 324 sets of interaction energy data, and each set of features corresponds to an interaction energy data through molecular simulation experiment. The pore wall surface features include quartz wall surface, kaolinite wall surface, calcite wall surface and II-C type kerogen wall surface. The temperature features include 333K, 393K and 453K. The pressure features include 20MPa, 30MPa and 40MPa. The shale oil component features (represented by density) include 0.62g / cm 3 , 0.73g / cm 3 and 0.86g / cm 3 , and the pore size features include 2nm, 5nm and 10nm. According to the pore wall surface feature quantization operation, the quartz, kaolinite, calcite and II-C type kerogen wall surfaces are assigned to the interval of 0-1, wherein the quartz wall surface is assigned to 0.1686, the kaolinite wall surface is assigned to 0, the calcite wall surface is assigned to 0.4029, and the II-C type kerogen wall surface is assigned to 1. As shown in Table 6, part of the interaction energy data under the pore size of 5nm.

[0098] Table 6 Multi-component shale oil component interaction energy data

[0099]

[0100]

[0101] The noise data of the generative adversarial network is six-dimensional random generated data, which satisfies the standard normal distribution and corresponds to the pore wall surface feature, temperature feature, pressure feature, pore diameter feature, shale oil component feature and interaction energy feature. The pore wall surface feature is a limited value, which varies between 0.1686, 0, 0.4029 and 1 corresponding to four wall conditions.

[0102] The structure of the generative adversarial network model is shown in Figure 3 The generative adversarial network model includes two neural networks, namely a generator and a discriminator. The generator learns the data distribution of a set of samples by implicit learning, thereby generating artificial samples conforming to the data distribution. The discriminator is used to determine whether the samples synthesized by the generator are true, and the output of the discriminator is fed back to the generator to improve the quality of the artificial samples. This cycle is repeated alternately. After continuous iteration and updating, the generator can generate artificial samples conforming to the real data distribution, and at this time the discriminator cannot distinguish between real samples and artificial samples, and the probability of making a correct judgment is 0.5, i.e. Nash equilibrium is reached. This learning optimization process is a binary maximum minimum game. When training the model, one side needs to be fixed and the parameters of the other side need to be updated. Through alternating iterative training, the error of the other side is maximized, and finally a generated model that can deceive is obtained.

[0103] The input layer of the generator is a Dense layer, the size of the input data is the number of input data categories (6 categories) multiplied by the batch size (batch size = 10), and a LeakyReLU activation function layer is connected. The hidden layer is a 3-layer Dense layer, the activation function is tanh, the first layer has 20 neurons, the second layer has 10 neurons, and the third layer has 6 neurons, and finally a LeakyReLU activation function layer is connected. The output layer is a Reshape layer, which normalizes the feature value size to 6 categories to complete data augmentation.

[0104] The input layer of the discriminator is a Flatten layer, which expands all features. The hidden layer is a 1-layer Dense layer with 50 neurons, and a ReLU activation function layer is connected. The output layer is a Dense layer with 1 neuron, and a ReLU activation function layer is connected to judge the true or false of the output data.

[0105] The loss function of the generator and the discriminator is defined by the BinaryCrossentropy function built-in the artificial neural network library Keras, so that the true value of the output value tends to 1 and the false value tends to 0. The optimizer of the generator and the discriminator uses Adam, the learning rate is set to 0.0001, and the number of iterations Epoch is set to 100 times.

[0106] The representative small sample data obtained by the MD simulation is taken as original data, a generative adversarial network algorithm is used to construct a shale nanopore fluid and wall surface interaction energy data augmentation model, and data expansion is realized based on the data augmentation model. Figure 4 As shown in the figure, the data augmentation model generates shale oil and pore wall surface interaction energy data within a certain range centered on the training data points. For data points outside the range, the data augmentation model cannot cover them. Therefore, a large amount of high-quality data generated by the data augmentation model is used to train a shale oil and pore wall surface interaction energy prediction model to complete the areas not covered by the data augmentation model, and finally a shale oil and pore wall surface interaction energy database is constructed.

[0107] The GAN data augmentation model is constructed based on the generative adversarial network algorithm, and the process of data expansion by the GAN data augmentation model is as shown in the figure. Figure 5 The original input data O includes temperature T, pore diameter r, wall surface type γ, shale oil component X, pressure P and interaction energy E. The GAN data augmentation model first normalizes the original input data O into x1, x2, x3, x4, x5 and x6. Any Gaussian noise sequence projects the original low-dimensional vector into a high-dimensional vector consistent with the dimension of the original data through the generator, i.e. maps the six-dimensional noise data into high-dimensional vectors X1', X2', X3', X4', X5' and X6'. Then the discriminator judges the distribution similarity between the normalized original data O and the generated high-dimensional vector. After continuous training of the generator and the discriminator, distribution similar data can be finally generated, and the generated data O' is obtained through inverse normalization. The data of the fluid and pore wall surface interaction energy in the shale reservoir nanopore is expanded, and high-quality new samples similar to the original data are formed. Finally, a group of newly generated samples are selected, and MD method is used for molecular simulation to verify the accuracy of the data augmentation model.

[0108] Step 4, based on the fully connected neural network FCNN, a fluid and pore wall surface interaction energy prediction model in shale reservoir nanopore is established.

[0109] The fully connected neural network (FCNN) is a basic artificial neural network structure, also known as multilayer perceptron (MLP). In the fully connected neural network, each neuron is connected to all neurons of the previous and next layers, forming a dense connection structure. The fully connected neural network can learn the complex features of the input data and perform classification, regression and other tasks. The structure is as shown in the figure. Figure 6

[0110] ​The prediction model structure based on the full connection neural network comprises an input layer, a hidden layer, an output layer and an activation function. The input layer is a Dense layer, responsible for receiving original data, the input data is 5 dimensions, specifically including temperature, pore size, wall type, shale oil component and pressure, and the number of neurons is 5; the hidden layer is an intermediate layer in the full connection neural network, the number of neurons in the hidden layer can be freely set, each neuron is connected with all neurons of the previous layer and the next layer, the hidden layer is set as 5 layers of Dense layer, the activation function of the first 4 layers is tanh, the last layer does not add the activation function, and the number of neurons in each layer is 20; the output layer is the last layer of the full connection neural network, responsible for outputting the prediction result of the network, and the prediction result is the interaction energy, the output layer is set as 1 layer of Dense layer, and the number of neurons is 1.

[0111] Step 5, based on the prediction model, data completion is performed on the region not covered by the GAN data enhancement model to construct a complete interaction energy database and establish a visual software system.

[0112] The GAN data enhancement model cannot comprehensively cover the interaction energy data of different wall types, different pore sizes, different temperature and pressure conditions and different shale oil components in the database, therefore, the interaction energy prediction model is trained by using the data set expanded by the GAN data enhancement model, the interaction energy is predicted through the trained prediction model, the data of the region not covered by the GAN data enhancement model is completed, and then a complete interaction energy database is constructed.

[0113] The data set expanded by the GAN data enhancement model is divided into two disjoint sets of a training set and a validation set, the interaction energy prediction model is first trained by using the training set, then the prediction model is used to predict the interaction energy of the validation set, and the trained model is evaluated. The present application selects mean squared error (MSE) for predictability evaluation. The present application calculates the mean squared error MSE between the predicted value and the true value, and takes the average value as the performance index for evaluating the final model. The calculation method of MSE is:

[0114]

[0115] Wherein, h(x i ) is the model prediction value, y i is the true value, i represents the i th sample, and n represents the total number of samples.

[0116] Based on the constructed fluid-pore wall interaction energy prediction model, the interaction energy of the fluid type and the pore wall type not calculated by the MD method and the area not covered by the GAN data enhancement model is complemented, so as to form a complete shale reservoir nanometer pore fluid-pore wall interaction energy database, which facilitates the evaluation of the properties of shale oil by reservoir engineers and helps oilfield engineering design.

[0117] The present application establishes a visual software system based on MATLAB: shale system Shale V1.0, which mainly includes three functional modules: basic physical property database module, solid-liquid interaction strength database module and state equation module. Among them, the basic physical property database module is used to query the related physical properties of shale oil and reservoirs from different places, and the SEM scanning electron microscope image of the reservoir core is visualized. The solid-liquid interaction strength database module quantitatively characterizes the interaction energy, average force potential and number density distribution of solid-liquid interaction for different fluid properties, pore wall properties and reservoir conditions, and visualizes the micro model; the state equation module realizes the calculation of the compressibility factor and fugacity coefficient of single-component pure substance and binary mixture based on PR state equation. The shale system can preliminarily evaluate the mobility of shale oil, the advantages and disadvantages of reservoir properties and the calculation of fluid state parameters through the above three functional modules.

[0118] In the basic physical property database module, the basic physical properties of shale oil and reservoir rock are automatically displayed by selecting the source of shale oil; the basic physical properties of shale oil include shale oil density, shale oil composition and shale oil carbon number distribution, the shale oil carbon number distribution is a curve graph with carbon atom number as the horizontal coordinate and mass fraction as the vertical coordinate; the basic physical properties of reservoir rock include TOC content, organic matter type, pore type, porosity and main mineral content, and the SEM scanning electron microscope image of the reservoir rock is visualized.

[0119] Taking the Songliao Basin Gulong Sag as an example, after selecting the source of shale oil as Songliao Basin Gulong Sag, in the shale oil basic physical property part, the shale oil density is automatically displayed as 0.80-0.82 g / cm 3 , the shale oil composition includes light oil (surface density less than 0.8 g / cm 3 ), high gas-oil ratio (more than 50 m 3 / m 3), and a graph of the carbon number distribution of shale oil. In the reservoir rock basic physical property section, the TOC content is automatically displayed as 0.9-9.0, the organic matter type is I-II, the pore type is intergranular space, intragranular pore, intercrystalline pore, organic matter pore, bedding seam, and microcrack, the porosity is 3.4-16.0, and the main mineral content is 2-52% of clay minerals, 19-43% of quartz, and 8-65% of plagioclase. In addition, the current reservoir rock SEM scanning electron microscope image is displayed, which includes multiple images of other magnifications. The previous and next images can be selected for review, and the physical properties of the reservoir rock can be better observed.

[0120] In the solid-liquid interaction strength database module, fluid properties, nanopore wall properties, reservoir conditions, fluid and wall object average force potential (PMF) display, PMF distribution, and number density distribution are included. Fluid properties include the mass fraction of light, medium, and heavy components. The wall properties of the nanopore include different upper and lower wall types. Reservoir conditions include the temperature and pressure of shale oil, and the pore size of the slit pore. In the fluid and wall object average force potential (PMF) display, the current PMF stereogram is displayed according to the set fluid type, wall type, and adsorption energy of the upper and lower walls. PMF distribution and number density distribution are both curve graph settings.

[0121] For example, in the fluid property section, the light, medium, and heavy components in shale oil are set as 10%, 80%, and 10%, respectively. In the nanopore wall property section, the upper and lower wall types are both set as quartz. In the reservoir condition section, the temperature of shale oil is set as 300K, the pressure is set as 20MPa, and the pore size of the slit pore is set as 10nm. The fluid is set as light, the upper wall, and the adsorption energy of the upper wall is set as 20.2Kcal / mol, and the adsorption energy of the lower wall is set as 19.5Kcal / mol. After setting, the module extracts the interaction energy parameters, average force potential, and number density distribution of the corresponding objects, and visualizes the image of the micro-molecular model.

[0122] In the state equation module, two groups of single-component pure substance related parameters, temperature, pressure, mole fraction of component 1, mole fraction of component 2, compressibility factor, phase state, and fugacity coefficient of component 1 and component 2 are included. Each group of single-component pure substance related parameters includes critical temperature, critical pressure, and eccentric factor.

[0123] The state equation module can calculate both single-component pure substance related parameters and two-component binary mixture related parameters.

[0124] In the calculation of single-component pure substance related parameters, first, the critical temperature, critical pressure and eccentric factor of component 1 are set as parameters, the critical temperature, critical pressure and eccentric factor of component 2 are set as 0, then the temperature and pressure values of the state to be calculated are set, and finally the compressibility factor, fugacity coefficient and phase state of the substance are calculated. The mole fraction of component 1 and the mole fraction of component 2 are also set as 0. For example, the critical temperature, critical pressure and eccentric factor of component 1 are set as 304.2K, 73.82bar and 0.228w respectively, the critical temperature, critical pressure and eccentric factor of component 2 are set as 0, the temperature is set as 300K, the pressure is set as 20bar, the mole fraction of component 1 and the mole fraction of component 2 are also set as 0, and after calculation, the compressibility factor of the current substance is 0.8862, the fugacity coefficient is 0.8955, and the phase state of the substance is gas phase.

[0125] In the calculation of two-component binary mixture related parameters, the critical temperature, critical pressure and eccentric factor parameters of component 1 and component 2 are set respectively, and the mole fraction values of component 1 and component 2 are set, then the temperature and pressure values of the state to be calculated are set, and after calculation, the compressibility factor, fugacity coefficient and phase state of component 1 and component 2 can be obtained. For example, the critical temperature, critical pressure and eccentric factor of component 1 are set as 304.2K, 73.82bar and 0.228w respectively, the critical temperature, critical pressure and eccentric factor of component 1 are set as 647.3K, 221.2bar and 0.344w respectively, the temperature is set as 300K, the pressure is set as 20bar, the mole fraction of component 1 and the mole fraction of component 2 are set as 0.2% and 0.8% respectively, and after calculation, the compressibility factor of component 1 is 0.02319, the fugacity coefficient is 0.7031, and the phase state of the substance is liquid phase, the compressibility factor of component 2 is 0.8483, the fugacity coefficient is 0.8673, and the phase state of the substance is gas phase.

[0126] The system of the present application can quickly retrieve interaction energy data and display interaction strength schematic image after inputting fluid and pore wall characteristics, realizes the visual display of interaction strength, and overcomes the problem that the existing MD molecular simulation technology cannot quantitatively analyze the interaction strength between complex fluid components and multiple types of pore walls in shale oil reservoir nanometer pores with high efficiency.

[0127] Of course, the above description is not a limitation of the present application, and the present application is not limited to the above examples, and changes, modifications, additions or replacements made by those skilled in the art within the scope of the present application should also be within the protection scope of the present application.

Claims

1. A method for intelligently constructing a database of the interaction energy between shale oil and pore walls, characterized in that, The database is constructed using intelligent algorithms employing generative adversarial networks and fully connected neural networks, specifically including the following steps: Step 1: Obtain the interaction energy data between representative fluids and representative pore walls in the nanopores of shale oil reservoirs using molecular dynamics calculations; the specific process is as follows: Step 1.1: Establish a multi-component shale oil molecular model and a nanopore molecular model, combine the molecular model of shale oil adsorption in nanopores, and perform energy minimization and relaxation. Step 1.2: Use molecular dynamics to obtain the equilibrium molecular trajectory of multi-component shale oil adsorption in the nanopores of multi-component shale oil under equilibrium conditions, and calculate the adsorption density distribution of each component of shale oil in the nanopores. Step 1.3: Set the shale oil force field parameters and nanopore force field parameters, calculate the interaction energy between each component of the multi-component shale oil and the nanopores, and obtain the interaction energy data of each component of the multi-component shale oil. Step 2: Quantify the characteristic parameters of the fluid and pore walls to form the raw data for the neural network model; Step 3: Construct a GAN data augmentation model based on generative adversarial networks to generate high-quality new samples similar to the original data, thereby expanding the fluid-pore wall interaction energy dataset; Step 4: Based on a fully connected neural network, establish a predictive model for the interaction energy between fluid and pore wall in nanopores of shale oil reservoirs; Step 5: Based on the prediction model, complete the data for the areas not covered by the GAN data augmentation model, build a complete interaction energy database, and establish a visualization software system. A visualization software system, the Shale System, was established based on MATLAB. The Shale System comprises three functional modules: a basic physical property database module, a solid-liquid interaction intensity database module, and a state equation module. The basic physical property database module is used to query the relevant physical properties of shale oil and reservoirs from different producing areas and visualizes SEM images of reservoir cores. The solid-liquid interaction intensity database module quantitatively characterizes the interaction energy, average force potential, and number density distribution of solid-phase interactions for different fluid properties, pore wall properties, and reservoir conditions, and visualizes the microscopic model. The state equation module, based on the PR state equation, calculates the compressibility coefficient and fugacity coefficient of single-component pure substances and two-component binary mixtures.

2. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 1, characterized in that, The specific process of step 1.1 is as follows: Step 1.1.1: Establish a multi-component shale oil molecular model based on the Ligpargen website; First, using the Ligpargen website, molecular models of methane, n-dodecane, asphaltene, and n-decylamine were built. After completing charge balance optimization, four .lmp files and four .xyz files corresponding to each model were exported. Then, the four .lmp type files are converted to obtain four .lt type files; Simultaneously, random distribution files of four .xyz types were constructed based on methane, n-dodecane, asphaltene, n-decylamine, and rock pore size; finally, the random distribution files and four .lt type files were merged to generate a multi-component shale oil molecular model. Step 1.1.2: Using Materials Studio software, establish shale nanoporous molecular models; the established shale nanoporous molecular models include shale kaolinite nanoporous molecular models, shale quartz nanoporous molecular models, shale calcite nanoporous molecular models, and shale type II-C kerogen nanoporous molecular models; First, a shale kaolinite nanoporous molecular model was established. The unit cell function in Materials Studio was used, and the Build module was employed to cut out the kaolinite crystal faces, resulting in a kaolinite unit cell model. The lattice structure parameters of the kaolinite unit cell model were optimized to generate a kaolinite unit cell model file. This kaolinite unit cell model file was then imported into the Lammps molecular simulator for file format conversion, yielding a kaolinite unit cell data file. The ClayFF force field parameters for the kaolinite mineral were set and applied to the kaolinite unit cell model to construct the shale kaolinite nanoporous molecular model. Then, shale quartz nanoporous molecular models, shale calcite nanoporous molecular models, and shale II-C type kerogen nanoporous molecular models were established in the same manner. Step 1.1.3: Combine the shale nanopore molecular model with the multi-component shale oil molecular model to obtain the initial model of multi-component shale oil nanopore adsorption. Perform energy minimization and relaxation processing on the initial model of multi-component shale oil nanopore adsorption to obtain the multi-component shale oil nanopore adsorption model. First, the shale kaolinite nanopore molecular model is combined with the multi-component shale oil molecular model to form an initial adsorption model for multi-component shale oil nanopores. By minimizing the energy of this initial model to eliminate overlapping atomic configurations, the energy of the multi-component shale oil nanopore adsorption initial model is minimized. The energy of the multi-component shale oil nanopore adsorption initial model under the minimized energy state is then obtained, resulting in the energy-minimized initial model. Next, the energy-minimized initial model is optimized by setting the relaxation step size and relaxation time under the NVE canonical ensemble with preset parameters. The relaxation step size and relaxation time were used to relax the initial model of nanoporous adsorption of multi-component shale oil, and the system temperature value of the initial model under steady state was obtained and compared with the actual formation temperature. The relaxation process made the system temperature value of the initial model under steady state consistent with the actual formation temperature, thus obtaining the multi-component shale oil nanoporous adsorption model. Then, the shale quartz nanoporous molecular model, shale calcite nanoporous molecular model, and shale II-C type kerogen nanoporous molecular model were combined with the multi-component shale oil molecular model in turn to establish multi-component shale oil nanoporous adsorption models of quartz, calcite, and II-C type kerogen in the same way. The specific process of step 1.2 is as follows: First, based on the molecular dynamics method, the adsorption configuration of shale oil in kaolinite nanopores under equilibrium state was obtained by running for 10 nanoseconds with a step size of 1 femtosecond. Then, the adsorption configuration of shale oil in kaolinite nanopores under equilibrium state was further simulated for 2 nanoseconds under the original simulation conditions to obtain density distribution data. 400 frames of molecular trajectory files were output and saved with a step size of 500 femtoseconds. In the process of simulating and obtaining density distribution data, a data statistics box was established along the z direction of the space where the multi-component shale oil is located with an interval of 0.025 nm. The average value of the mass distribution of 2,000,000 multi-component shale oil was obtained with a step size of 1 femtosecond according to formula (1). The adsorption density distribution curve of each component of shale oil in nanopores was obtained. （1）； In the formula: Let be the average density of component j in the i-th data statistics box in shale oil; Let j be the number of components j in the i-th data statistics bin in shale oil; denoted as , where is the relative molecular mass of component j in shale oil; x is the dimension of the system in the x-direction; y is the dimension of the system in the y-direction. Let the height be the height of the i-th data collection box; The constant is Avogadro; then, the same process is used to generate molecular trajectory files for quartz, calcite, and type II-C kerogen; In step 1.3, the shale oil force field parameters and nanopore force field parameters include interatomic parameters, bond parameters, and angular parameters. Based on the obtained adsorption configuration and molecular trajectory files in the nanopores, firstly, the Coulomb interaction parameters in the adsorption configuration files in the nanopores are deleted. Then, by reproducing the molecular trajectory, the average value of 400 sets of data with a step size of 500 femtoseconds is taken to obtain the van der Waals interaction energy between each component of shale oil and pores. Based on multi-component shale oil, the geological characteristics of multiple shale oil reservoirs are statistically classified and analyzed. Considering four factors—shale oil density, temperature and pressure conditions of the storage environment, and nanoporous matrix—the above-mentioned molecular dynamics simulation was conducted to obtain solid-liquid interaction energy data. This solid-liquid interaction energy data is the required interaction energy data between representative fluids and representative pore walls in the nanopores of shale oil reservoirs, ultimately forming a small sample dataset of fluid-pore wall interaction energy.

3. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 1, characterized in that, The specific process of step 2 is as follows: First, based on the interaction energy data of each component of multi-component shale oil, the pore characteristics obtained from the SEM scanning experiment of multi-component shale matrix are used; then, the interaction energy of the pore wall surface to the fluid is sorted according to its strength and mapped to the 0-1 interval to realize the quantification of the pore wall surface characteristics.

4. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 1, characterized in that, In step 3, the GAN data augmentation model performs data augmentation as follows: The original input data O includes temperature, pore size, wall type, shale oil composition, pressure, and interaction energy. The GAN data augmentation model first normalizes the original input data O into x1, x2, x3, x4, x5, and x6. The generator projects the original low-dimensional vector into high-dimensional vectors X1', X2', X3', X4', X5', and X6' with the same dimensions as the original data. Then, the discriminator judges the distribution similarity between the normalized original data O and the generated high-dimensional vector. After continuously training the generator and discriminator, data with similar distribution is finally generated. After inverse normalization, the generated data O' is obtained, thus realizing the augmentation of the interaction energy data between fluid and pore wall in the nanopores of shale oil reservoirs and forming a high-quality new sample similar to the original data. The generator's input layer is a Dense layer, with the input data size equal to the number of input data categories multiplied by the batch size, and connected to a LeakyReLU activation function layer. The hidden layers are three Dense layers with tanh activation functions: the first layer has 20 neurons, the second layer has 10 neurons, and the third layer has 6 neurons, all connected to a LeakyReLU activation function layer. The output layer is a Reshape layer, which normalizes the feature values ​​to six categories, thus augmenting the data. The input layer of the discriminator is a Flatten layer, which flattens out all features; The hidden layer is a single Dense layer with 50 neurons, connected to a ReLU activation function layer; the output layer is a Dense layer with one neuron, connected to a ReLU activation function layer, which determines whether the output data is true or false.

5. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 1, characterized in that, In step 4, the prediction model based on the fully connected neural network includes an input layer, a hidden layer, an output layer, and an activation function. The input layer is a Dense layer, responsible for receiving raw data. The input data has five dimensions, specifically including temperature, pore size, wall type, shale oil composition, and pressure, and has five neurons. The hidden layer is the middle layer in the fully connected neural network. The number of neurons in the hidden layer is arbitrarily set. Each neuron is connected to all neurons in the previous and next layers. The hidden layer is set to five Dense layers. The activation function for the first four layers is tanh, and no activation function is added to the last layer. Each layer has 20 neurons. The output layer is the last layer of the fully connected neural network, responsible for outputting the network's prediction result, which is the interaction energy. The output layer is set to one Dense layer with one neuron.

6. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 1, characterized in that, In step 5, the dataset expanded by the GAN data augmentation model is divided into two disjoint sets: a training set and a validation set. First, the interaction energy prediction model is trained using the training set. Then, the prediction model is used to predict the interaction energy of the validation set to evaluate the trained model. The interaction energy is predicted using the trained prediction model to fill in the data in areas not covered by the GAN data augmentation model, thereby constructing a complete interaction energy database. The mean squared error (MSE) is selected for model evaluation, and its average value is used as the performance metric for the final model. The MSE is calculated as follows: (2)； in, These are the model's predicted values. Let i represent the true value, i represent the i-th sample, and n represent the total number of samples.

7. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 1, characterized in that, In step 5, the shale system uses three functional modules to preliminarily evaluate the mobility of shale oil, the quality of reservoir properties, and calculate fluid state parameters.

8. The intelligent database construction method for the interaction energy between shale oil and pore walls according to claim 7, characterized in that, The basic physical property database module automatically displays the basic physical properties of shale oil and reservoir rocks by selecting the shale oil source. The basic physical properties of shale oil include shale oil density, shale oil composition, and shale oil carbon number distribution. The carbon number distribution of shale oil is a curve with the number of carbon atoms on the horizontal axis and the mass fraction on the vertical axis. The basic physical properties of reservoir rocks include TOC content, organic matter type, pore type, porosity, and the content of major minerals, and the SEM scanning electron microscope images of the reservoir rocks are visualized. The solid-liquid interaction strength database module includes fluid properties, nanopore wall properties, reservoir conditions, PMF display of fluid and wall objects, PMF distribution, and number density distribution. Fluid properties include the mass fractions of light, medium, and heavy components. Nanopore wall properties include different wall types on the upper and lower surfaces. Reservoir conditions include the temperature and pressure of shale oil, as well as the pore size of slotted pores. In the PMF display of fluid and wall objects, a three-dimensional PMF map is displayed based on the set fluid type, wall type, and adsorption energy of the upper and lower walls. The PMF distribution and number density distribution are both set as curves. The state equation module includes two sets of parameters related to single-component pure substances, namely component 1 and component 2: temperature, pressure, mole fraction of component 1, mole fraction of component 2, compressibility factor, phase state, and fugacity coefficient. Each set of parameters for single-component pure substances includes critical temperature, critical pressure, and eccentricity factor. The state equation module can calculate parameters for both single-component pure substances and binary mixtures. In the calculation of parameters for single-component pure substances, the critical temperature, critical pressure, and eccentricity factor of component 1 are first set, while the critical temperature, critical pressure, and eccentricity factor of component 2 are all set to 0. Then, the temperature and pressure values ​​of the state to be calculated are set, and finally, the compressibility factor, fugacity coefficient, and phase state of the substance are calculated. The mole fractions of components 1 and 2 are also set to 0. In the calculation of relevant parameters of a two-component binary mixture, the critical temperature, critical pressure and eccentricity factor parameters of component 1 and component 2 are set respectively, and the mole fraction values ​​of component 1 and component 2 are set respectively. Then, the temperature and pressure values ​​of the state to be calculated are set, and the compressibility factor, fugacity coefficient and phase state of the substance of component 1 and component 2 are obtained through calculation.

Images