Physical difference convolutional neural network multiphase seepage simulation method, device and medium

Abstract

The present application discloses a physical differential convolutional neural network multiphase seepage simulation method, device and medium, which relates to the field of oil and gas exploration technology. By obtaining the geological model of the oil and gas reservoir, the deep learning data domain is determined to construct a data-driven convolutional neural network model; the loss function of the physical constraint differential convolutional neural network is constructed according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the residual of the initial condition; the physical constraint differential convolutional neural network is trained and the loss function weights and network structure parameters are optimized to predict the pressure field and saturation field of the oil and gas reservoir. The above scheme utilizes the similarity between the form of the convolution kernel of the convolutional neural network and the form of the finite difference, and adds the multiphase seepage control equation to the neural network in the form of a differential convolution kernel, so as to realize the joint drive of physical constraints and data in the neural network as a physical constraint term, thereby improving the prediction effect, thereby obtaining a more realistic prediction result of the oil and gas reservoir pressure field and saturation field.

Description

Technical Field

[0001] The present application relates to the field of oil and gas exploration technology, and in particular to a physical differential convolutional neural network multiphase seepage simulation method, device and medium. Background Art

[0002] An oil and gas reservoir is a trap that gathers a certain amount of oil and gas, and is the basic unit of oil and gas accumulation in the earth's crust. An oil and gas reservoir exists in an independent trap, in which oil and gas have a certain distribution law and a unified pressure system. At present, in the exploitation of oil and gas reservoirs, the prediction of physical quantities such as pressure field and saturation field of oil and gas reservoirs is mainly achieved through multiphase seepage simulation of oil and gas reservoirs. Multiphase seepage simulation of oil and gas reservoirs refers to the use of computers to solve the mathematical model of multiphase seepage to obtain the numerical distribution of physical quantities such as pressure field and saturation field that change with time, thereby reflecting their movement laws. Traditional oil and gas seepage simulation is often restricted by the contradiction between computer computing power and accuracy, so machine learning is often used as an alternative model for multiphase seepage prediction of oil and gas reservoirs.

[0003] However, this traditional machine learning method uses purely data-driven multiphase seepage simulation in oil and gas reservoirs. The model itself has no relevant mathematical laws, which results in the parameters in the model having no actual spatial physical meaning and being unexplainable.

[0004] In view of the above problems, how to introduce physical constraints into the machine learning proxy model of oil and gas reservoir seepage model prediction in oil and gas reservoir development in order to obtain physical results of the real pressure and saturation of the oil and gas reservoir that are more in line with reality is an urgent problem to be solved by technicians in this field. Summary of the invention

[0005] The purpose of this application is to provide a physical differential convolutional neural network multiphase seepage simulation method, device and medium, so as to introduce physical constraints into the machine learning proxy model oil and gas reservoir prediction process in oil and gas reservoir development, so as to obtain more realistic pressure and saturation distribution results of oil and gas reservoirs.

[0006] In order to solve the above technical problems, the present application provides a physical difference convolutional neural network multiphase seepage simulation method, including:

[0007] Obtain a geological model of the oil and gas reservoir and determine the spatial size of the deep learning data domain;

[0008] Constructing a training set and a test set of a data-driven convolutional neural network according to the geological model and the deep learning data domain to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are accurate solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance;

[0009] Obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution, and constructing the loss function of the physically constrained differential convolutional neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the residual of the initial condition; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial condition is the initial value of the pressure field and the initial value of the saturation field;

[0010] The physically constrained differential convolutional neural network is trained by using the training set and the test set, and the weights of the loss function and the network structure parameters of the physically constrained differential convolutional neural network are optimized, so as to predict the pressure field and the saturation field of the oil and gas reservoir by using the optimized physically constrained differential convolutional neural network.

[0011] Preferably, the process of generating the exact solution of the pressure field and the exact solution of the saturation field includes:

[0012] Obtaining basic reservoir characteristics, fluid high-pressure physical properties and multiphase seepage mechanism of the oil and gas reservoir;

[0013] The multiphase seepage control equation is discretized using a fully implicit finite difference method according to the geological model, the spatial size of the deep learning data domain, the basic reservoir characteristics, the fluid high-pressure physical properties and the multiphase seepage mechanism to generate an exact solution to the pressure field and an exact solution to the saturation field.

[0014] Preferably, the step of obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution comprises:

[0015] The multiphase seepage control equation and the initial condition are discretized, and the multiphase seepage control equation residual expressed in the form of differential convolution is generated through a cross convolution kernel.

[0016] Preferably, the residual of the multiphase seepage control equation expressed in the form of differential convolution generated by the cross convolution kernel includes:

[0017] Setting the closed boundary condition of the multiphase seepage control equation by an equivalent padding operation;

[0018] The constant pressure boundary condition of the multiphase seepage control equation is set by constant value Padding operation.

[0019] Preferably, the loss function of constructing a physical constraint differential convolutional neural network based on the residual of the multiphase seepage control equation and the initial condition residual expressed in the form of differential convolution includes:

[0020] Obtaining the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the data-driven convolutional neural network data matching;

[0021] The mean square error of the residuals of the multiphase seepage control equation expressed in differential convolution form, the mean square error of the residuals of the initial conditions, and the mean square error of the residuals of the data-driven convolutional neural network data matching are added together, and weight coefficients are assigned to generate the loss function of the physically constrained differential convolutional neural network.

[0022] Preferably, the step of training the physical constraint difference convolutional neural network by using the training set and the test set, and optimizing the weights and network structure parameters of the loss function of the physical constraint difference convolutional neural network comprises:

[0023] The network structure parameters and the weight coefficients of the loss function are optimized by the gradient back propagation algorithm to obtain the weight of the loss function whose error satisfies the first preset condition and the network structure parameters whose error satisfies the second preset condition.

[0024] Preferably, after predicting the pressure field and the saturation field of the oil and gas reservoir by the optimized physical constraint difference convolutional neural network, the method further comprises:

[0025] The prediction results of the pressure field and the saturation field of the oil and gas reservoir are output.

[0026] In order to solve the above technical problems, the present application also provides a physical difference convolutional neural network multiphase seepage simulation device, comprising:

[0027] An acquisition module is used to obtain the geological model of the oil and gas reservoir and determine the spatial size of the deep learning data domain;

[0028] A construction module is used to construct a training set and a test set of a data-driven convolutional neural network according to the geological model and the deep learning data domain, so as to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are accurate solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance;

[0029] A generation module, used to obtain the residual of the multiphase seepage control equation expressed in the form of differential convolution, and construct a loss function of the physical constraint differential convolution neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes a data matching term loss term, a boundary condition loss term and a control equation loss term of the data-driven convolution neural network, and the initial condition is an initial value of the pressure field and an initial value of the saturation field;

[0030] An optimization module is used to train the physical constraint differential convolutional neural network through the training set and the test set, optimize the weights of the loss function and the network structure parameters of the physical constraint differential convolutional neural network, so as to predict the pressure field and the saturation field of the oil and gas reservoir through the optimized physical constraint differential convolutional neural network.

[0031] In order to solve the above technical problems, the present application also provides another physical difference convolutional neural network multiphase seepage simulation device, including:

[0032] Memory for storing computer programs;

[0033] A processor is used to implement the steps of the above-mentioned physical difference convolutional neural network multiphase seepage simulation method when executing the computer program.

[0034] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned physical differential convolutional neural network multiphase seepage simulation method are implemented.

[0035] The physical differential convolutional neural network multiphase seepage simulation method provided in the present application obtains the geological model of the oil and gas reservoir and determines the spatial size of the deep learning data domain; constructs a training set and a test set of the data-driven convolutional neural network according to the geological model and the deep learning data domain to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are the precise solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance; obtains the residual of the multiphase seepage control equation expressed in the form of differential convolution, and constructs the loss function of the physically constrained differential convolutional neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the residual of the initial condition; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial condition is the initial value of the pressure field and the initial value of the saturation field; trains the physically constrained differential convolutional neural network through the training set and the test set, optimizes the weights of the loss function and the network structure parameters of the physically constrained differential convolutional neural network, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physically constrained differential convolutional neural network. It can be seen that when constructing a convolutional neural network to predict oil and gas reservoir pressure and saturation, the above scheme uses the similarity between the form of the convolutional neural network convolution kernel and the form of the finite difference to add the multiphase seepage control equation to the neural network in the form of a differential convolution kernel, so that the multiphase seepage control equation can be used as a physical constraint to realize the joint drive of physical constraints and data in the convolutional neural network, which greatly improves the prediction effect of the neural network for multiphase seepage simulation, thereby obtaining more realistic prediction results of the oil and gas reservoir pressure field and saturation field.

[0036] In addition, an embodiment of the present application also provides a physical differential convolutional neural network multiphase seepage simulation device and medium, with the same effect as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A flow chart of a physical difference convolutional neural network multiphase seepage simulation method provided in an embodiment of the present application;

[0039] Figure 2 It is a schematic diagram of a cross convolution kernel provided in an embodiment of the present application;

[0040] Figure 3 A flow chart of differential physical convolution provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the principle of the physical difference convolutional neural network multiphase seepage simulation method provided in the embodiment of the present application;

[0042] Figure 5 A schematic diagram of a physical difference convolutional neural network multiphase seepage simulation device provided in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of another physical differential convolutional neural network multiphase seepage simulation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] The core of this application is to provide a physical differential convolutional neural network multiphase seepage simulation method, device and medium to introduce physical constraints into the process of predicting oil and gas reservoir pressure and saturation of machine learning agent model in oil and gas reservoir development, so as to obtain the actual pressure and saturation results of oil and gas reservoirs that are more in line with reality.

[0046] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] Multiphase seepage simulation of oil and gas reservoirs refers to the use of computers to solve the mathematical model of multiphase seepage to obtain the numerical distribution of physical quantities such as pressure and saturation that change over time, thereby reflecting its movement law. Traditional multiphase seepage simulation of oil and gas reservoirs mainly uses Fourier neural operators and recurrent neural networks to convert the solution domain of the model from the time domain to the frequency domain, which greatly reduces the complexity of the model solution while obtaining the results. However, this kind of multiphase seepage simulation of oil and gas reservoirs driven by pure data has no relevant mathematical laws in the model itself, resulting in the parameters in the model having no physical meaning in the actual space and being unexplainable. Therefore, an embodiment of the present application provides a physical differential convolutional neural network multiphase seepage simulation method, which can introduce physical constraints in the prediction process of oil and gas reservoir pressure and saturation of the machine learning proxy model in oil and gas reservoir development, so as to obtain more realistic results of the actual pressure and saturation of the oil and gas reservoir. Figure 1 A flow chart of a physical difference convolutional neural network multiphase seepage simulation method provided in an embodiment of the present application. Figure 1 As shown, the method includes:

[0048] S10: Obtain a geological model of the oil and gas reservoir and determine the spatial size of the deep learning data domain.

[0049] In the specific implementation, the geological model of the oil and gas reservoir is first obtained. It can be understood that the geological model of the oil and gas reservoir is a mathematical model that describes the type, geometry, scale, internal structure, reservoir parameters and fluid distribution of the oil and gas reservoir. It is the basis for the comprehensive evaluation of the oil and gas reservoir, and is also an important basis for the numerical simulation of the oil and gas reservoir and the basis for the optimization of the development plan. The geological model of the oil and gas reservoir is a three-dimensional visual model established through data analysis based on structural geology, sedimentology, petroleum geology, reservoir geology and oil and gas reservoir geochemistry as theoretical foundations, with seismic information database, geological information database, logging information database, test information database and dynamic information database as pillars. In this embodiment, the geological model of the oil and gas reservoir can be directly obtained, or the geological model of the oil and gas reservoir can be established according to the above-mentioned model establishment process. In this embodiment, there is no limitation, depending on the specific implementation situation.

[0050] At the same time, it is necessary to determine the spatial size of the deep learning data domain, including the length, width, and height of the space. In the specific implementation, it can be determined randomly or uniformly, which is not limited in this embodiment and depends on the specific implementation situation.

[0051] S11: Construct training sets and test sets of data-driven convolutional neural networks based on geological models and deep learning data domains to build data-driven convolutional neural network models.

[0052] The label values ​​of the training set and the test set are the exact solutions of the pressure field and the saturation field of the pre-generated oil and gas reservoirs.

[0053] Furthermore, in order to predict the pressure field and saturation field of oil and gas reservoirs, it is necessary to construct a convolutional neural network. Convolutional Neural Networks (CNN) is a type of feedforward neural network that contains convolution calculations and has a deep structure. It is one of the representative algorithms of deep learning. Convolutional neural networks have representation learning capabilities and can perform shift-invariant classification on input information according to their hierarchical structure. Therefore, they are also called "Shift-Invariant Artificial Neural Networks (SIANN)."

[0054] Specifically, the training set and test set of the data-driven convolutional neural network are constructed according to the geological model and deep learning data domain obtained in the previous step, thereby obtaining a pure data-driven convolutional neural network model. The convolutional neural network model includes an input layer, an encoding layer, a linear layer, a decoding layer, and an output layer. The input layer includes but is not limited to specific parameters such as time and permeability.

[0055] It should be noted that the label values ​​of the training set and the test set of the data-driven convolutional neural network in this embodiment are the precise solutions of the pressure field and the precise solutions of the saturation field of the oil and gas reservoir generated in advance. In this embodiment, there is no restriction on the method of obtaining the precise solutions of the pressure field and the saturation field of the oil and gas reservoir, which depends on the specific implementation situation. As a preferred embodiment, the generation process of the precise solution of the pressure field and the precise solution of the saturation field includes:

[0056] Obtain basic reservoir characteristics, fluid high-pressure properties and multiphase flow mechanisms of oil and gas reservoirs;

[0057] According to the geological model, the spatial size of the deep learning data domain, the basic reservoir characteristics, the high-pressure physical properties of the fluid and the multiphase seepage mechanism, the multiphase seepage control equations are discretized using a fully implicit finite difference method to generate exact solutions of the pressure field and the saturation field.

[0058] Specifically, the basic reservoir characteristics, fluid high-pressure properties and multiphase seepage mechanism of the oil and gas reservoir are first obtained. It can be understood that the basic reservoir characteristics include data such as porosity, permeability, and fault distribution; the fluid high-pressure properties include data such as volume coefficient, compressibility, density, and viscosity; and the multiphase seepage mechanism includes data such as multiphase relative permeability curves, capillary force, and imbibition. In specific implementation, the basic reservoir characteristics, fluid high-pressure properties, and multiphase seepage mechanism can be obtained based on the specific conditions of the oil and gas reservoir.

[0059] Furthermore, the basic reservoir characteristics, fluid high-pressure physical properties and multiphase seepage mechanism obtained are functionalized, and according to the geological model, the spatial size of the deep learning data domain, the basic reservoir characteristics, fluid high-pressure physical properties and multiphase seepage mechanism, the multiphase seepage control equation is discretized by fully implicit finite difference, and the large matrix is ​​solved to obtain the exact solution of the pressure field and the exact solution of the saturation field. This data is used as the label value of the training set and test set of the convolutional neural network.

[0060] It should be noted that the multiphase flow control equation is the most basic mathematical law describing the flow of fluid in oil and gas reservoirs. The multiphase flow control equation is as follows:

[0061]

[0062]

[0063] Where k is the intrinsic permeability of the porous medium, k ro is the relative permeability of the oil body, k rw is the liquid relative permeability, μ o is the viscosity coefficient of the oil, μ w is the liquid viscosity coefficient, S o is the oil phase saturation, S w is the water phase saturation, q o is the reservoir oil phase fraction flow rate, q w is the reservoir water phase flow rate, and p is the pressure.

[0064] S12: Obtain the residual of the multiphase seepage control equation expressed in the form of differential convolution, and construct the loss function of the physical constraint differential convolution neural network based on the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual.

[0065] Among them, the loss function includes the data matching loss term, boundary condition loss term and control equation loss term of the data-driven convolutional neural network, and the initial conditions are the initial values ​​of the pressure field and the initial values ​​of the saturation field.

[0066] Furthermore, the residual of the multiphase seepage control equation expressed in the form of differential convolution is obtained, and it is used as a physical constraint together with the initial condition residual to construct the loss function of the physical constraint differential convolution neural network. Among them, the loss function includes the data matching term loss term, boundary condition loss term and control equation loss term of the above-mentioned data-driven convolution neural network, and the initial condition is the natural state of the oil and gas reservoir without initial development and human intervention, which specifically refers to the initial pressure field and saturation initial values. As a preferred embodiment, obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution includes: discretizing the multiphase seepage control equation and the initial condition, and generating the residual of the multiphase seepage control equation expressed in the form of differential convolution through a cross convolution kernel.

[0067] Figure 2 is a schematic diagram of the cross convolution kernel provided in the embodiment of the present application. Figure 2 As shown in the figure, the black part in the center represents the cross convolution kernel. The cross convolution kernel moves rightward and downward along the grid in turn to calculate the pressure and saturation data of each grid. c, a, e, b and d represent the parameters of the cross convolution, which change with the movement of the cross convolution. The calculation domain is the area that needs to be calculated.

[0068] In the specific implementation, multiple seepage control equations and initial conditions are discretized, and a fully implicit discretization method is adopted. In data processing, the absolute permeability is processed by harmonic mean; the volume coefficient, relative permeability, and viscosity are determined in the form of upstream weights. The convolution operation is further used to replace the finite difference operation to express the differential expression, so as to obtain the residuals of the multiple seepage control equations expressed in the form of differential convolution. Specifically, the residuals of the multiphase seepage control equations expressed in the form of differential convolution are generated by the cross convolution kernel. In the process of generating the residuals of the multiphase seepage control equations in the form of differential convolution, Padding operation is required.

[0069] Padding operation is a data filling operation in convolution operation, which can be used to set boundary conditions in partial differential equations. Figure 2 As shown in the figure, the outermost grid pointed by the arrow is the boundary condition-grid, which represents the implementation result of the padding operation of the boundary condition. That is to say, in the original state, there is only the penultimate grid where the arrow starts, and the outermost grid is filled in by the padding operation to implement the boundary condition. Specifically, the closed boundary condition of the multiphase seepage control equation is set by the equal value Padding operation, and the constant pressure boundary condition of the multiphase seepage control equation is set by the constant value Padding operation, thereby realizing the setting of the boundary condition of the residual of the multiphase seepage control equation expressed in the form of differential convolution.

[0070] After obtaining the residual of the multiphase seepage control equation and the initial condition residual expressed in the form of differential convolution, in order to introduce them into the data-driven convolutional neural network as physical constraints, the loss function of the physical constraint differential convolutional neural network is constructed according to the multiphase seepage control equation and the initial condition residual expressed in the form of differential convolution, which specifically includes:

[0071] Obtain the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the data-driven convolutional neural network data matching;

[0072] The mean square error of the residuals of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residuals of the initial conditions, and the mean square error of the residuals of the data-driven convolutional neural network data matching are added together, and weight coefficients are assigned to generate the loss function of the physically constrained differential convolutional neural network.

[0073] Figure 3 This is a flow chart of differential physical convolution provided in the embodiment of the present application. Figure 3 As shown, we first receive the pressure / saturation distribution from the output of the convolutional neural network and use Figure 2 The cross convolution kernel shown is subjected to boundary processing and calculation to obtain a series of pressure / saturation distribution fragments. The conductivity coefficient calculation method is then used to convert the pressure / saturation distribution fragment into a conductivity coefficient convolution kernel to obtain a conductivity coefficient convolution kernel. The conductivity coefficient convolution kernel and the pressure / saturation distribution fragment are then used to calculate the outer product to obtain the residual of the control equation, and the mean square error of the residual is further taken and added to the pure data-driven loss function to obtain a differential convolution neural network. Specifically, the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the convolution neural network data matching are obtained. Among them, the residual of the multiphase seepage control equation and the residual of the initial condition are the differences corresponding to the subtraction of the left and right sides of the multiphase seepage control equation. The residual of the data-driven convolution neural network data matching is determined based on the data set of the multiphase seepage control equation and the convolution neural network, specifically the difference between the true value and the predicted value. The mean square error is the mean of the sum of squares of the residuals. The mean square error of the residuals of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residuals of the initial conditions, and the mean square error of the residuals of the convolutional neural network data matching are further added, and certain weight coefficients are assigned to generate the loss function of the convolutional neural network.

[0074] S13: Train the physical constraint differential convolutional neural network through the training set and the test set, optimize the weights of the loss function and the network structure parameters of the physical constraint differential convolutional neural network, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physical constraint differential convolutional neural network.

[0075] Finally, in order to obtain more accurate prediction results, it is also necessary to optimize the loss function weights and network structure parameters of the convolutional neural network by physical constraint differences. In this embodiment, there is no restriction on the optimization method of the loss function weights and network structure parameters, which depends on the specific implementation situation. Specifically, as a preferred embodiment, optimizing the loss function weights and network structure parameters of the physical constraint difference convolutional neural network includes: optimizing the network structure parameters and various weight coefficients of the loss function by the gradient back propagation algorithm to obtain the weights of the loss function whose error satisfies the first preset condition and the network structure parameters whose error satisfies the second preset condition.

[0076] The gradient back propagation algorithm is a combination of the back propagation algorithm and the gradient descent method. The gradient descent method is a first-order optimization algorithm. If the gradient descent method is used to find the local minimum of a function, it is necessary to iteratively search for points of a specified step distance in the opposite direction of the gradient (or approximate gradient) corresponding to the current point on the function. The back propagation algorithm is a learning algorithm for multi-layer neural networks, which is based on the gradient descent method. The input-output relationship of the back propagation network is essentially a mapping relationship: the function performed by a back propagation neural network with n inputs and m outputs is a continuous mapping from the n-dimensional Euclidean space to the finite field in the m-dimensional Euclidean space, and this mapping is highly nonlinear. Its information processing ability comes from the multiple compounding of simple nonlinear functions, so it has a strong function reproduction ability. This is the basis for the application of the back propagation algorithm.

[0077] In this embodiment, the network structure parameters and the weight coefficients of the loss function are optimized by the gradient back propagation algorithm to obtain the weight of the loss function whose error satisfies the first preset condition and the network structure parameters whose error satisfies the second preset condition. In this embodiment, there is no restriction on the first preset condition and the second preset condition, which depends on the specific implementation situation. Preferably, the network structure parameters and the weight coefficients of the loss function can be optimized by the gradient back propagation algorithm until the loss function approaches zero and the network structure parameters are the smallest among all the errors obtained.

[0078] Finally, after obtaining the trained and optimized physical constraint differential convolutional neural network, the pressure field and saturation field predictions of the subsequent time step are obtained through the optimized physical constraint differential convolutional neural network. In addition, as a preferred embodiment, after predicting the pressure field and saturation field of the oil and gas reservoir through the optimized physical constraint differential convolutional neural network, the prediction results of the pressure field and saturation field of the oil and gas reservoir can also be output to enable the user to better understand the final prediction results, which is convenient for the subsequent calculation of the flow process and production change data of the oil and gas reservoir flow.

[0079] The following use Figure 4 Provide an overall description of this application. Figure 4 The schematic diagram of the principle of the physical difference convolutional neural network multiphase seepage simulation method provided in the embodiment of the present application. Figure 4 As shown in the figure, the time-space coordinates and other variables (such as permeability, etc.) are input into the pure data-driven convolutional neural network. Among them, the pure data-driven convolutional neural network is divided into a pressure grid and a saturation grid. After obtaining the pressure output and saturation output respectively, the output pressure and saturation are converted into a differential partial differential equation (Partial Differential Equation, PDE) as a loss function item using the above-mentioned cross convolution kernel and differential physical convolution. The loss function also includes a purely data-driven data matching item, as well as the initial pressure and initial saturation of the oil and gas reservoir. Compared with the traditional pure data-driven method, which only has a data matching item as a loss function, the differential physical convolutional network adds soft constraints of differential PDE and initial conditions and hard constraints of boundary conditions. This loss function is used to perform physical information back propagation to optimize the network structure parameters to obtain the optimal model parameters.

[0080] In this embodiment, a geological model of an oil and gas reservoir is obtained, and the spatial size of a deep learning data domain is determined; a training set and a test set of a data-driven convolutional neural network are constructed according to the geological model and the deep learning data domain to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are the precise solutions of the pressure field and the saturation field of the oil and gas reservoir that are pre-generated; the residual of the multiphase seepage control equation expressed in the form of differential convolution is obtained, and the loss function of the physically constrained differential convolutional neural network is constructed according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes a data matching term loss term, a boundary condition loss term, and a control equation loss term of the data-driven convolutional neural network, and the initial conditions are the initial values ​​of the pressure field and the initial values ​​of the saturation field; the physically constrained differential convolutional neural network is trained by the training set and the test set, and the weights of the loss function and the network structure parameters of the physically constrained differential convolutional neural network are optimized, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physically constrained differential convolutional neural network. It can be seen that when constructing a convolutional neural network to predict oil and gas reservoir pressure and saturation, the above scheme uses the similarity between the form of the convolutional neural network convolution kernel and the form of the finite difference to add the multiphase seepage control equation to the neural network in the form of a differential convolution kernel, so that the multiphase seepage control equation can be used as a physical constraint to realize the joint drive of physical constraints and data in the convolutional neural network, which greatly improves the prediction effect of the neural network for multiphase seepage simulation, thereby obtaining more realistic prediction results of the oil and gas reservoir pressure field and saturation field.

[0081] In order to enable those skilled in the art to better understand the technical solution of the present application, the following takes two-dimensional oil-water two-phase seepage simulation as an example to specifically illustrate the multiphase seepage simulation prediction effect of the differential physical convolution kernel neural network proposed in the present application.

[0082] First, a quarter five-point well network with a reservoir size of 510m×510m×10m is established. The grid length of the numerical benchmark model is 10 meters, and there are closed boundaries on all sides. The water wells are fixed with a fixed water injection volume, and the oil wells are fixed with a fixed bottom hole pressure. Assume that the fluid is isothermal seepage, capillary pressure and gravity are not considered, and the formation pressure is always higher than the saturation pressure, and oil and water are immiscible. The original bottom pressure is 25Mpa, and the depth of the oil and gas reservoir is 2500m. Assume that the formation permeability is generated by a Gaussian distribution, isotropic, and has a porosity of 0.2. The viscosity coefficient of the oil phase is 3mPa·s, the viscosity coefficient of the water phase is 0.3mPa·s, the initial water saturation is 0.2, and the bound water saturation is also 0.2. The oil wells are fixed with a bottom hole pressure of 15Mpa for production, and the water wells are fixed with a fixed water injection volume for water injection. The simulation time step is 1month, the total number of time steps is 100, and the neural network training set is the first 60 steps. Then the oil-water two-phase model of the oil and gas reservoir area is:

[0083]

[0084]

[0085] p(t=0)=250;

[0086] S w (t=0)=0.2;

[0087]

[0088]

[0089] Where k is the intrinsic permeability of the porous medium, k ro is the relative permeability of the oil body, k rw is the liquid relative permeability, μ o is the viscosity coefficient of the oil, μ w is the liquid viscosity coefficient, S o is the oil phase saturation, S w is the water phase saturation, q o is the reservoir oil phase fraction flow rate, q w is the reservoir water phase fraction flow rate. p(t=0)=250 is the initial pressure condition, S w (t=0)=0.2 is the initial saturation condition, is the boundary condition in the x direction, is the boundary condition in the y direction.

[0090]

[0091]

[0092] Table 1

[0093] Table 1 provides a network model structure parameter table for an embodiment of the present application. In the process of constructing the prediction model, the network structure model of the structure shown in Table 1 is adopted, wherein the model adopts the tanh activation function, batch_size=5, and the optimizer is the Adam optimizer. The training set is the numerical results of pressure and saturation for the first 60 months. Because it is a two-dimensional two-phase, the pressure and saturation are coupled at this time, and the pressure term coefficient is not a constant. The equation has strong nonlinearity, which increases the difficulty of solving the model. By adopting dual neural network joint training, the two networks train pressure and saturation respectively, and the pressure output value is convolved with the basic difference operator, further multiplied by the pressure coefficient operator, and the convolution padding operation is used to replace the boundary problem of the model. Considering the source and sink terms, the pressure value of the central grid of the previous time step is obtained, and the difference between the label value and the true value is calculated using the mean square error. It is repeatedly iterated to approximate the true pressure value and saturation value, so as to represent the evolution mode of the partial differential equation learned by the model.

[0094] In the above embodiments, the physical difference convolutional neural network multiphase seepage simulation method is described in detail, and the present application also provides embodiments corresponding to the physical difference convolutional neural network multiphase seepage simulation device. It should be noted that the present application describes the embodiments of the device part from two perspectives, one is based on the perspective of functional modules, and the other is based on the perspective of hardware structure.

[0095] Figure 5 A schematic diagram of a physical difference convolutional neural network multiphase seepage simulation device provided in an embodiment of the present application. Figure 5 As shown, the physical difference convolutional neural network multiphase seepage simulation device includes:

[0096] The acquisition module 10 is used to obtain the geological model of the oil and gas reservoir and determine the spatial size of the deep learning data domain.

[0097] Construction module 11 is used to construct a training set and a test set of a data-driven convolutional neural network based on the geological model and the deep learning data domain to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are the precise solutions of the pressure field and the saturation field of the pre-generated oil and gas reservoir.

[0098] The generation module 12 is used to obtain the residual of the multiphase seepage control equation expressed in the form of differential convolution, and construct the loss function of the physically constrained differential convolutional neural network based on the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial conditions are the initial value of the pressure field and the initial value of the saturation field.

[0099] The optimization module 13 is used to train the physical constraint differential convolutional neural network through a training set and a test set, optimize the weights of the loss function and the network structure parameters of the physical constraint differential convolutional neural network, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physical constraint differential convolutional neural network.

[0100] In this embodiment, the physical difference convolutional neural network multiphase seepage simulation device includes an acquisition module, a construction module, a generation module and an optimization module. The physical difference convolutional neural network multiphase seepage simulation device can implement all the steps of the above-mentioned physical difference convolutional neural network multiphase seepage simulation method when running. By obtaining the geological model of the oil and gas reservoir and determining the spatial size of the deep learning data domain; constructing the training set and test set of the data-driven convolutional neural network according to the geological model and the deep learning data domain to construct the data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are the precise solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance; obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution, and constructing the loss function of the physically constrained differential convolutional neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial conditions are the initial values ​​of the pressure field and the initial values ​​of the saturation field; training the physically constrained differential convolutional neural network through the training set and the test set, optimizing the weights of the loss function and the network structure parameters of the physically constrained differential convolutional neural network, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physically constrained differential convolutional neural network. It can be seen that when constructing a convolutional neural network to predict oil and gas reservoir pressure and saturation, the above scheme uses the similarity between the form of the convolutional neural network convolution kernel and the form of the finite difference to add the multiphase seepage control equation to the neural network in the form of a differential convolution kernel, so that the multiphase seepage control equation can be used as a physical constraint to realize the joint drive of physical constraints and data in the convolutional neural network, which greatly improves the prediction effect of the neural network for multiphase seepage simulation, thereby obtaining more realistic prediction results of the oil and gas reservoir pressure field and saturation field.

[0101] Figure 6 A schematic diagram of another physical difference convolutional neural network multiphase seepage simulation device provided in an embodiment of the present application. Figure 6 As shown, the physical difference convolutional neural network multiphase seepage simulation device includes:

[0102] The memory 20 is used to store computer programs.

[0103] The processor 21 is used to implement the steps of the method of multiphase seepage simulation using physical difference convolutional neural network as mentioned in the above embodiment when executing the computer program.

[0104] The physical differential convolutional neural network multiphase seepage simulation device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer or a desktop computer.

[0105] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of a digital signal processor (DSP), a field programmable gate array (FPGA), and a programmable logic array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 21 may be integrated with a graphics processing unit (GPU), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0106] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used to store the following computer program 201, wherein, after the computer program is loaded and executed by the processor 21, it can implement the relevant steps of the physical differential convolutional neural network multiphase seepage simulation method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 202 may include Windows, Unix, Linux, etc. Data 203 may include but is not limited to data involved in the physical differential convolutional neural network multiphase seepage simulation method.

[0107] In some embodiments, the physical differential convolutional neural network multiphase seepage simulation device may also include a display screen 22, an input and output interface 23, a communication interface 24, a power supply 25 and a communication bus 26.

[0108] Those skilled in the art will understand that Figure 3 The structure shown in the figure does not constitute a limitation of the physical difference convolutional neural network multiphase permeation simulation device, and may include more or fewer components than those shown in the figure.

[0109] In this embodiment, the physical difference convolutional neural network multiphase seepage simulation device includes a memory and a processor. The memory is used to store a computer program. The processor is used to implement the steps of the physical difference convolutional neural network multiphase seepage simulation method mentioned in the above embodiment when executing the computer program. By obtaining the geological model of the oil and gas reservoir and determining the spatial size of the deep learning data domain; constructing the training set and test set of the data-driven convolutional neural network according to the geological model and the deep learning data domain to construct the data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are the precise solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance; obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution, and constructing the loss function of the physically constrained differential convolutional neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial conditions are the initial values ​​of the pressure field and the initial values ​​of the saturation field; training the physically constrained differential convolutional neural network through the training set and the test set, optimizing the weights of the loss function and the network structure parameters of the physically constrained differential convolutional neural network, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physically constrained differential convolutional neural network. It can be seen that when constructing a convolutional neural network to predict oil and gas reservoir pressure and saturation, the above scheme uses the similarity between the form of the convolutional neural network convolution kernel and the form of the finite difference to add the multiphase seepage control equation to the neural network in the form of a differential convolution kernel, so that the multiphase seepage control equation can be used as a physical constraint to realize the joint drive of physical constraints and data in the convolutional neural network, which greatly improves the prediction effect of the neural network for multiphase seepage simulation, thereby obtaining more realistic prediction results of the oil and gas reservoir pressure field and saturation field.

[0110] Finally, the present application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps recorded in the above method embodiment are implemented.

[0111] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0112] In this embodiment, a computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, the steps recorded in the above method embodiment are implemented. By obtaining the geological model of the oil and gas reservoir and determining the spatial size of the deep learning data domain; constructing the training set and test set of the data-driven convolutional neural network according to the geological model and the deep learning data domain to construct the data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are the precise solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance; obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution, and constructing the loss function of the physically constrained differential convolutional neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial conditions are the initial values ​​of the pressure field and the initial values ​​of the saturation field; training the physically constrained differential convolutional neural network through the training set and the test set, optimizing the weights of the loss function and the network structure parameters of the physically constrained differential convolutional neural network, so as to predict the pressure field and saturation field of the oil and gas reservoir through the optimized physically constrained differential convolutional neural network. It can be seen that when constructing a convolutional neural network to predict oil and gas reservoir pressure and saturation, the above scheme uses the similarity between the form of the convolutional neural network convolution kernel and the form of the finite difference to add the multiphase seepage control equation to the neural network in the form of a differential convolution kernel, so that the multiphase seepage control equation can be used as a physical constraint to realize the joint drive of physical constraints and data in the convolutional neural network, which greatly improves the prediction effect of the neural network for multiphase seepage simulation, thereby obtaining more realistic prediction results of the oil and gas reservoir pressure field and saturation field.

[0113] The physical differential convolutional neural network multiphase seepage simulation method, device and medium provided by the present application are introduced in detail above. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can refer to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

[0114] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. A physical difference convolutional neural network multiphase seepage simulation method, characterized in that: include: Obtain a geological model of the oil and gas reservoir and determine the spatial size of the deep learning data domain; Constructing a training set and a test set of a data-driven convolutional neural network according to the geological model and the deep learning data domain to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are accurate solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance; Obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution, and constructing the loss function of the physically constrained differential convolutional neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the residual of the initial condition; wherein the loss function includes the data matching term loss term, the boundary condition loss term and the control equation loss term of the data-driven convolutional neural network, and the initial condition is the initial value of the pressure field and the initial value of the saturation field; The physically constrained differential convolutional neural network is trained by using the training set and the test set, and the weight of the loss function and the network structure parameters of the physically constrained differential convolutional neural network are optimized, so as to predict the pressure field and the saturation field of the oil and gas reservoir by using the optimized physically constrained differential convolutional neural network; The generation process of the exact solution of the pressure field and the exact solution of the saturation field includes: Obtaining basic reservoir characteristics, fluid high-pressure physical properties and multiphase seepage mechanism of the oil and gas reservoir; Performing fully implicit finite difference discretization on the multiphase seepage control equation according to the geological model, the spatial size of the deep learning data domain, the basic reservoir characteristics, the fluid high-pressure physical properties and the multiphase seepage mechanism to generate an exact solution to the pressure field and an exact solution to the saturation field; The method of obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution includes: Discretizing the multiphase seepage control equation and the initial condition, and generating the multiphase seepage control equation residual expressed in the form of differential convolution through a cross convolution kernel; The residual of the multiphase seepage control equation expressed in the form of differential convolution generated by the cross convolution kernel includes: Setting the closed boundary condition of the multiphase seepage control equation by an equivalent padding operation; Setting a constant pressure boundary condition of the multiphase seepage control equation by constant value Padding operation; Use cross convolution kernel for boundary processing and calculation to obtain pressure / saturation distribution fragments; Using the conductivity calculation method, the pressure / saturation distribution fragment is converted into a conductivity convolution kernel; The residual of the control equation is obtained by using the outer product of the conductivity convolution kernel and the pressure / saturation distribution fragment; The loss function of constructing a physical constraint differential convolution neural network based on the residual of the multiphase seepage control equation and the initial condition residual expressed in the form of differential convolution includes: Obtaining the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the data-driven convolutional neural network data matching; Adding the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the data-driven convolutional neural network data matching, and assigning weight coefficients to generate the loss function of the physical constraint differential convolutional neural network; The step of training the physical constraint difference convolutional neural network by using the training set and the test set and optimizing the weights and network structure parameters of the loss function of the physical constraint difference convolutional neural network includes: Optimizing the network structure parameters and various weight coefficients of the loss function by a gradient back propagation algorithm to obtain the weight of the loss function whose error satisfies a first preset condition and the network structure parameters whose error satisfies a second preset condition; The first preset condition is that the weight satisfies the loss function and approaches 0, and the second preset condition is that the error of the network structure parameter is the smallest among all the errors obtained; After predicting the pressure field and the saturation field of the oil and gas reservoir by the optimized physical constraint difference convolutional neural network, the method further includes: The prediction results of the pressure field and the saturation field of the oil and gas reservoir are output.

2. A physical differential convolutional neural network multiphase seepage simulation device, characterized in that: include: An acquisition module is used to obtain the geological model of the oil and gas reservoir and determine the spatial size of the deep learning data domain; A construction module is used to construct a training set and a test set of a data-driven convolutional neural network according to the geological model and the deep learning data domain, so as to construct a data-driven convolutional neural network model; wherein the label values ​​of the training set and the test set are accurate solutions of the pressure field and the saturation field of the oil and gas reservoir generated in advance; A generation module, used to obtain the residual of the multiphase seepage control equation expressed in the form of differential convolution, and construct a loss function of the physical constraint differential convolution neural network according to the residual of the multiphase seepage control equation expressed in the form of differential convolution and the initial condition residual; wherein the loss function includes a data matching term loss term, a boundary condition loss term and a control equation loss term of the data-driven convolution neural network, and the initial condition is an initial value of the pressure field and an initial value of the saturation field; An optimization module, used for training the physical constraint differential convolutional neural network through the training set and the test set, optimizing the weight of the loss function and the network structure parameters of the physical constraint differential convolutional neural network, so as to predict the pressure field and the saturation field of the oil and gas reservoir through the optimized physical constraint differential convolutional neural network; The generation process of the exact solution of the pressure field and the exact solution of the saturation field includes: Obtaining basic reservoir characteristics, fluid high-pressure physical properties and multiphase seepage mechanism of the oil and gas reservoir; Performing fully implicit finite difference discretization on the multiphase seepage control equation according to the geological model, the spatial size of the deep learning data domain, the basic reservoir characteristics, the fluid high-pressure physical properties and the multiphase seepage mechanism to generate an exact solution to the pressure field and an exact solution to the saturation field; The method of obtaining the residual of the multiphase seepage control equation expressed in the form of differential convolution includes: Discretizing the multiphase seepage control equation and the initial condition, and generating the multiphase seepage control equation residual expressed in the form of differential convolution through a cross convolution kernel; The residual of the multiphase seepage control equation expressed in the form of differential convolution generated by the cross convolution kernel includes: Setting the closed boundary condition of the multiphase seepage control equation by an equivalent padding operation; Setting a constant pressure boundary condition of the multiphase seepage control equation by constant value Padding operation; Use cross convolution kernel for boundary processing and calculation to obtain pressure / saturation distribution fragments; Using the conductivity calculation method, the pressure / saturation distribution fragment is converted into a conductivity convolution kernel; The residual of the control equation is obtained by using the outer product of the conductivity convolution kernel and the pressure / saturation distribution fragment; The loss function of constructing a physical constraint differential convolution neural network based on the residual of the multiphase seepage control equation and the initial condition residual expressed in the form of differential convolution includes: Obtaining the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the data-driven convolutional neural network data matching; Adding the mean square error of the residual of the multiphase seepage control equation expressed in the form of differential convolution, the mean square error of the residual of the initial condition, and the mean square error of the residual of the data-driven convolutional neural network data matching, and assigning weight coefficients to generate the loss function of the physical constraint differential convolutional neural network; The step of training the physical constraint difference convolutional neural network by using the training set and the test set and optimizing the weights and network structure parameters of the loss function of the physical constraint difference convolutional neural network includes: Optimizing the network structure parameters and various weight coefficients of the loss function by a gradient back propagation algorithm to obtain the weight of the loss function whose error satisfies a first preset condition and the network structure parameters whose error satisfies a second preset condition; The first preset condition is that the weight satisfies the loss function and approaches 0, and the second preset condition is that the error of the network structure parameter is the smallest among all the errors obtained; After predicting the pressure field and the saturation field of the oil and gas reservoir by the optimized physical constraint difference convolutional neural network, the method further includes: The prediction results of the pressure field and the saturation field of the oil and gas reservoir are output.

3. A physical difference convolutional neural network multiphase seepage simulation device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the physical difference convolutional neural network multiphase seepage simulation method as claimed in claim 1 when executing the computer program.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the physical difference convolutional neural network multiphase seepage simulation method as claimed in claim 1 are implemented.

Images