Flow heat transfer prediction method and device based on adversarial neural network, and medium

Abstract

The invention discloses a flow heat transfer prediction method and equipment based on an adversarial neural network and a medium, and relates to the technical field of data prediction.The method comprises the steps that a quality source item and an energy source item are constructed through the geometric structure and physical characteristics of a target prediction device, and simulation is conducted based on the quality source item and the energy source item; by generating the quality source item and the energy source item, the early-stage data error can be reduced in the early-stage parameter acquisition process, and by constructing the order reduction model and performing order reduction processing on the multi-physical field simulation result of the target prediction device, the data calculation complexity is reduced. The flow heat transfer prediction model is constructed through the adversarial neural network model for prediction, the prediction precision of the generated flow field cloud picture and the temperature field cloud picture can be improved, the flow heat transfer characteristics in industrial equipment can be obtained in real time, data support is provided for model development, and an important analysis method and tool are provided for industrial digital twinning.

Description

Technical Field

[0001] The present invention relates to the technical field of data prediction, and in particular to a flow and heat transfer prediction method, device and medium based on an adversarial neural network. Background Technique

[0002] Two-phase flow and heat transfer are widely used in the industrial field. Two-phase flow and heat transfer refer to the flow and heat transfer phenomena occurring in a fluid system where both vapor phase and liquid phase coexist in the channels of certain devices. This process has extensive applications in multiple fields such as chemical engineering, nuclear energy, petroleum, and environmental protection. In many industrial processes, such as air conditioners, refrigerators, boilers, and nuclear power plants, efficient heat transfer technologies are required to improve energy utilization efficiency. By studying the mechanisms of vapor-liquid two-phase flow and heat transfer and combining specific application requirements, the optimization and control of fluid flow and heat transfer can be achieved, which can improve the efficiency and quality of industrial production, and at the same time achieve energy conservation and environmental protection.

[0003] Therefore, in order to provide better control of fluid flow and heat transfer during the research process of energy utilization, it is necessary to predict the two-phase flow and heat transfer of the device. Numerically simulating complex flow and heat transfer phenomena usually requires a large amount of computing resources. However, the existing software is not accurate enough in calculating two-phase flow, with relatively large errors. Due to the pre-existing data errors, the computational complexity is high when constructing a turbulence model and boundary layer transition, resulting in a decrease in prediction accuracy and making it difficult to meet the high-precision requirements of industrial applications. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that the existing prediction methods have high computational complexity and low prediction accuracy, and the existing software is not accurate enough in calculating two-phase flow, with relatively large errors. The purpose is to provide a flow and heat transfer prediction method, device and medium based on an adversarial neural network. By constructing a mass source term and an energy source term based on the geometric structure and physical characteristics of the flow channel, and performing simulations based on the mass source term and the energy source term, a multi-physical field simulation result of the target prediction device is obtained. By customizing the mass source term and the energy source term, the pre-existing data errors can be reduced during the early-stage calculation and data acquisition process. By constructing a reduced-order model, the multi-physical field simulation result of the target prediction device is processed by reducing the order, reducing the data computational complexity. By constructing a flow and heat transfer prediction model through an adversarial neural network model for prediction, the prediction accuracy of the generated flow field cloud map and temperature field cloud map can be improved.

[0005] The present invention is realized through the following technical solutions: The first aspect of the present invention provides a flow and heat transfer prediction method based on an adversarial neural network, including the following specific steps: According to the geometric structure and physical characteristics of the target prediction device, perform mesh division of the target prediction device in a multi-physical field framework; Based on the fluid phase change parameters, heat transfer parameters, and latent heat of phase change parameters for each grid, obtain the mass source term and the energy source term; Perform a simulation based on the custom mass source term and energy source term to obtain the multi-physical field simulation results of the target prediction device; Construct a reduced-order model, and generate a training sample of temperature field cloud map data and a training sample of velocity field cloud map data according to the multi-physical field simulation results of the target prediction device; Construct an adversarial neural network model, and train the adversarial neural network model using the training sample of temperature field cloud map data and the training sample of velocity field cloud map data to obtain a flow and heat transfer prediction model; Input the multi-physical field simulation results of the target prediction device into the flow and heat transfer prediction model to predict and generate a flow field cloud map and a temperature field cloud map.

[0006] Furthermore, the steps for obtaining the mass source term specifically include: Obtain the grid volume to get the characteristic length of the grid cell; Obtain the phase change adjustment coefficient, the latent heat of phase change of the liquid, the gas density, the bubble temperature, the pressure inside the bubble, the molar mass of the gas, and the molar gas constant to obtain the condensation and evaporation heat transfer coefficients at the phase change interface; Obtain the latent heat of phase change of the liquid, the saturation temperature, and the grid cell temperature, and combine the condensation and evaporation heat transfer coefficients at the phase change interface to determine the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid; Obtain the mass transfer rate constraint condition of the vapor-liquid two-phase in the grid cell volume, and combine the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid to obtain the mass source term of each grid.

[0007] Furthermore, obtaining the condensation and evaporation heat transfer coefficients at the phase change interface specifically includes: ; Among them, represents the phase change adjustment coefficient, that is, the non-ideal effect caused by the presence of non-condensable gas molecules at the vapor-liquid interface, h lg is the latent heat of phase change of the liquid, ρ v represents the gas density, T v is the bubble temperature, represents the molar mass of the gas, is the molar gas constant, p v is the pressure inside the bubble.

[0008] Furthermore, determining the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid specifically includes: ; Among them, m lg represents the mass transfer rate of the vapor-liquid two-phase per unit volume in each grid, T sat represents the saturation temperature, T l represents the grid cell temperature, h ev represents the condensation and evaporation heat transfer coefficient of the phase change interface, h lg represents the latent heat of phase change of the liquid, L represents the characteristic length of the grid cell. The characteristic length of the cell is the cube root of the cell volume.

[0009] Furthermore, the mass transfer rate constraint conditions specifically include: During condensation, ; During evaporation, ; Among them, ρ v represents the vapor density, a v represents the gas volume fraction, △ t represents the current calculation time step.

[0010] Furthermore, the steps for obtaining the energy source term specifically include: Obtain the specific heat capacity at constant pressure of the liquid phase, the specific heat capacity at constant pressure of the vapor phase, the number of time steps, the temperature corresponding to the number of time steps, and the reference temperature in each grid, and combine with the mass transfer rate of the vapor-liquid two-phase per unit volume in each grid to obtain the energy source phase.

[0011] Furthermore, the specific calculation steps for obtaining the energy source phase include: ; Among them, represents the energy source phase, m lg represents the mass transfer rate of the vapor-liquid two-phase per unit volume in each grid, c pl represents the specific heat capacity at constant pressure of the liquid phase, c pg represents the specific heat capacity at constant pressure of the vapor phase, T n represents the number of time steps n of the temperature, n represents the number of time steps, T ref represents the reference temperature.

[0012] Further, for constructing the reduced-order model, temperature field cloud map data training samples and velocity field cloud map data training samples are generated according to the multi-physical field simulation results of the target prediction device, which specifically includes: Extract the original data of the temperature field and velocity field from the multi-physical field simulation for preprocessing; Construct a reduced-order model, input the preprocessed data into the reduced-order model to generate a learning matrix; Perform singular value decomposition on the learning matrix to obtain the eigenvalues of the decomposed data; Use the first K eigenvalues of the decomposed data as the reduced-order basis, project the original data into the low-dimensional space to obtain temperature field cloud map data training samples and velocity field cloud map data training samples.

[0013] Further, in the process of obtaining the eigenvalues of the decomposed data, it also includes calculating the average projection error and optimizing the number of eigenvalues of the decomposed data according to the average projection error. K 。

[0014] Further, the calculation steps of the average projection error include: ; where, relativeError represents the average projection error, M represents the learning matrix, K represents the number of eigenvalues, σ represents the deviation, i represents from 1 to K +1.

[0015] Further, constructing the adversarial neural network model specifically includes: Construct a generator; Obtain the random noise data of the temperature field cloud map data training samples and velocity field cloud map data training samples, and input them into the generator; The generator trains on the random noise data of the multi-physical field simulation results of the target prediction device to learn the probability distribution of the physical field cloud map until the training reaches the set threshold to obtain the generated physical field cloud map; Construct a discriminator; Extract the real physical field cloud map based on the temperature field cloud map data training samples and velocity field cloud map data training samples; Discriminate between the generated physical field cloud map and the real physical field cloud map, and optimize the generated physical field cloud map according to the discrimination result until the discrimination result reaches the set threshold to obtain the flow and heat transfer prediction model.

[0016] Further, the construction calculation steps of the adversarial neural network model include: ; Among them, represents the random noise data of the temperature field cloud map data training sample and the velocity field cloud map data training sample, G represents the generator, D represents the discriminator, G ( z ) represents the generated physical field cloud map data, D ( x ) represents the real physical field cloud map, x represents the real physical field cloud map sample, z represents the random noise vector, E represents the expected value, Pdata(x) represents the distribution of the real physical field cloud map.

[0017] The second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, a flow and heat transfer prediction method based on an adversarial neural network is implemented.

[0018] The third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, a flow and heat transfer prediction method based on an adversarial neural network is implemented.

[0019] Compared with the prior art, the present invention has the following advantages and beneficial effects: By constructing the mass source term and the energy source term through the geometric structure and physical characteristics of the target prediction device, and performing simulations based on the mass source term and the energy source term, the multi-physical field simulation results of the target prediction device are obtained. By customizing the mass source term and the energy source term, the pre-data error can be reduced during the pre-data acquisition process. By constructing a reduced-order model, the multi-physical field simulation results of the target prediction device are reduced-order processed to reduce the data calculation complexity. By constructing a flow and heat transfer prediction model through an adversarial neural network model for prediction, the prediction accuracy of the generated flow field cloud map and temperature field cloud map can be improved, the flow and heat transfer characteristics inside the industrial equipment can be obtained in real time, providing data support for model development, and providing important analysis methods and tools for industrial digital twins. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts. In the drawings: Figure 1 is the prediction process in the embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with embodiments and the accompanying drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0022] As a possible implementation manner, as Figure 1 shown, this embodiment provides a flow and heat transfer prediction method based on a generative adversarial neural network, including the following specific steps: According to the geometric structure and physical characteristics of the target prediction device, perform mesh division of the target prediction device in a multi-physics field framework; Based on the fluid phase change parameters, heat transfer parameters, and latent heat of phase change parameters of each grid, obtain the mass source term and the energy source term; Based on the mass source term and the energy source term, perform a simulation to obtain the multi-physics field simulation results of the target prediction device; Construct a reduced-order model, and according to the multi-physics field simulation results of the target prediction device, generate training samples of temperature field contour data and velocity field contour data; Construct a generative adversarial neural network model, and use the training samples of temperature field contour data and velocity field contour data to train the generative adversarial neural network model to obtain a flow and heat transfer prediction model; Input the multi-physics field simulation results of the target prediction device into the flow and heat transfer prediction model to predict and generate a flow field contour and a temperature field contour. In this embodiment, the mass source term and the energy source term are constructed through the geometric structure and physical characteristics of the target prediction device, and a simulation is performed based on the mass source term and the energy source term to obtain the multi-physics field simulation results of the target prediction device. By customizing the mass source term and the energy source term, the pre-data acquisition error can be reduced in the early stage. By constructing a reduced-order model, the multi-physics field simulation results of the target prediction device are processed for reduction, reducing the data calculation complexity. By constructing a flow and heat transfer prediction model through a generative adversarial neural network model for prediction, the prediction accuracy of the generated flow field contour and temperature field contour can be improved.

[0023] In some possible implementation manners, since the source terms of commercial CFD software need to be defined by the user himself, if the source terms are not defined accurately, the boiling two-phase flow heat transfer and phase change calculations are not accurate enough. Therefore, the present invention uses a self-compiled UDF (user-defined function) to correct the source terms of commercial CFD software. Add the customized mass source term and energy source term to the commercial CFD software through the UDF interface to define the mass source term and energy source term in the commercial CFD software. The mass source term and energy source term of the present application are obtained in the following manner: In some possible implementation manners, the specific steps for obtaining the mass source term include: Obtain the grid volume to obtain the characteristic length of the grid cell; Obtain the phase change adjustment coefficient, the latent heat of phase change of the liquid, the gas density, the bubble temperature, the pressure inside the bubble, the molar mass of the gas, and the molar gas constant to obtain the condensation and evaporation heat transfer coefficients at the phase change interface; Obtain the latent heat of phase change of the liquid, the saturation temperature, and the grid cell temperature, and combine the condensation and evaporation heat transfer coefficients at the phase change interface to determine the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid; Obtain the constraint condition of the mass transfer rate of the vapor-liquid two-phase per unit volume of the grid cell, and combine the mass source term to obtain the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid.

[0024] In some possible implementation manners, obtaining the condensation and evaporation heat transfer coefficients at the phase change interface specifically includes: ; Wherein, represents the phase change adjustment coefficient, that is, the non-ideal effect caused by the presence of non-condensable gas molecules on the vapor-liquid interface, h lg is the latent heat of phase change of the liquid, ρ v represents the gas density, T v is the bubble temperature, represents the molar mass of the gas, is the molar gas constant, p v is the pressure inside the bubble.

[0025] In some possible implementation manners, determining the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid specifically includes: ; Wherein, m lg represents the mass transfer rate of the vapor-liquid two-phase per unit volume within each grid, T sat represents the saturation temperature, T l represents the grid cell temperature, h ev represents the condensation and evaporation heat transfer coefficients at the phase change interface, h lg represents the latent heat of phase change of the liquid, L represents the characteristic length of the grid cell. The characteristic length of the cell is the cube root of the cell volume. For liquids with a temperature greater than a certain degree of superheat, a negative sign is taken in front of this formula; for steam with a temperature lower than the saturation temperature T sat a positive sign is taken in front of this formula. That is, in this embodiment, the UDF is set such that only when the superheat of the liquid is greater than the set value (set to 10 in the present inventiono It will only evaporate and turn into steam when the temperature reaches the saturation temperature (C); while when the temperature of the steam is lower than the saturation temperature, condensation will occur and it will turn into a liquid.

[0026] In some possible embodiments, the mass transfer rate constraint conditions specifically include: During condensation, ; During evaporation, ; Among them, ρ v represents the steam density, a v represents the gas-phase volume fraction, △ t represents the current calculation time step.

[0027] In some possible embodiments, the steps for obtaining the energy source term specifically include: Obtain the specific heat capacity at constant pressure of the liquid phase, the specific heat capacity at constant pressure of the vapor phase, the number of time steps, the temperature corresponding to the number of time steps, and the reference temperature in each grid, and combine the mass transfer rate of the vapor-liquid two-phase per unit volume in each grid to obtain the energy source phase.

[0028] In some possible embodiments, the specific calculation steps for obtaining the energy source phase include: ; Among them, represents the energy source phase, m lg represents the mass transfer rate of the vapor-liquid two-phase per unit volume in each grid, c pl represents the specific heat capacity at constant pressure of the liquid phase, c pg represents the specific heat capacity at constant pressure of the vapor phase, T n represents the number of time steps n of the temperature, n represents the number of time steps, T ref represents the reference temperature.

[0029] In some possible embodiments, constructing a reduced-order model, and generating a training sample of the temperature field cloud map data and a training sample of the velocity field cloud map data according to the multi-physical field simulation results of the target prediction device specifically includes: Extract the original data of the temperature field and the velocity field from the multi-physical field simulation for preprocessing; Construct a reduced-order model, input the preprocessed data into the reduced-order model, and generate a learning matrix; Perform singular value decomposition on the learning matrix to obtain the eigenvalues of the decomposed data; Use the first K eigenvalues of the decomposed data as the reduction basis, project the original data into a low-dimensional space, and obtain the training samples of the temperature field cloud map data and the training samples of the velocity field cloud map data.

[0030] In some possible implementation manners, for the reduced-order model, singular value decomposition (SVD) can be used for the reduced-order simulation of the physical field.

[0031] In some possible implementation manners, in the process of obtaining the eigenvalues of the decomposed data, it further includes calculating the average projection error and optimizing the number of eigenvalues of the decomposed data according to the average projection error K .

[0032] In some possible implementation manners, for singular value decomposition (SVD) in the heat conduction problem, mesh division is performed on the entire heat conduction structure or part of the target prediction device to obtain N working conditions (s1, s2,... s N ), each mesh represents a working condition, and the working condition represents the combination of input conditions for each group of calculations.

[0033] Calculate data for the N working conditions, specify the input parameters and output parameters of the entire data set, and divide it into a learning set q and a verification set m. Among them, the input parameters of the learning data and the temperature values of each mesh constitute a learning matrix M; Perform SVD decomposition on the learning matrix M, and then select the first K eigenvalues to obtain an approximate expression of the learning matrix M. Because for the rapid simulation of the heat conduction structure, the number of rows of the matrix of the learning matrix M cannot change, column dimension reduction is required, and the approximate expression of M obtained is: ; Among them, M represents the learning matrix, and its dimension is n × m , where n represents the number of samples, m represents the number of features, X represents the approximate reconstruction data matrix, and its dimension is also n × m , a i represents the i th eigenvalue, represents the proportion of the i th eigenvector in the data variance, U represents that the column vectors of the matrix are the left singular vectors of the data matrix M and constitute the orthogonal basis of the data, K represents the number of eigenvalues and is used to approximate the original data matrix M. K is usually much smaller than m , which can greatly reduce the dimension of the data.

[0034] In some possible embodiments, the calculation steps of the average projection error include: ; wherein, relativeError represents the average projection error, M represents the learning matrix, K represents the number of eigenvalues, σ represents the deviation, i represents from 1 to K +1.

[0035] In some possible embodiments, the average standard projection error: Each learning vector is deleted from the learning set, projected using the pattern calculated by the remaining vectors, and then the resulting projection errors are averaged (cross-validation).

[0036] In some possible embodiments, due to the fact that the eigenvalues of the matrix have the characteristic of rapidly decreasing, the sum of the first 10% or even the first 1% of the eigenvalues accounts for more than 99% of the total sum. Therefore, as K continues to increase, the average projection error will necessarily become smaller and smaller, that is, the degree of dispersion becomes smaller and smaller. When it reaches a certain value, it will drop rapidly, while the average standard projection error will gradually stabilize as K decreases as K continues to increase, that is, the degree of deviation no longer changes. For small K values, the average projection error and the average standard projection error are almost the same.

[0037] In some possible embodiments, it further includes: using dozens of groups of α in the learning set, and the learning set is already known α i … α K (each sample point has a group, a total of q groups), and the validation set is also already known α i … α K (each sample point has a group, a total of m groups). The validation set is used to verify the optimal number of patterns K, and the genetic aggregation interpolation method is used to generate a response surface within the given range (the relationship between the input parameters and α , α is equivalent to the output parameter of this response surface). When using the generated response surface for prediction, each sample point is predicted separately. After obtaining α i（i=1，2,…K） at a certain working condition point, only one column of U is required to obtain the desired learning matrix M, making the dimensionality reduction speed faster. Among them, because for the rapid simulation of the heat conduction structure, the number of rows of the M matrix cannot be changed, column dimensionality reduction is required.

[0038] In some possible implementations, the generative adversarial neural network GAN is composed of two neural network generators G and discriminators D with conflicting objectives, where: Construct an adversarial neural network model, including: Build the generator; Obtain random noise data of temperature field cloud map data training samples and velocity field cloud map data training samples, and input them into the generator; The generator trains random noise data of multi-physics field simulation results of the target prediction device to learn the probability distribution of the physics field cloud map until the training reaches a set threshold, thereby obtaining a generated physics field cloud map; Build the discriminator; Extracting real physical field cloud maps based on temperature field cloud map data training samples and velocity field cloud map data training samples; The generated physical field cloud map and the real physical field cloud map are distinguished, and the generated physical field cloud map is optimized according to the distinction result until the distinction result reaches the set threshold, thereby obtaining the flow heat transfer prediction model.

[0039] In some possible implementations, the generator models the joint probability, represents the distribution of data from a statistical perspective, and describes how the data is generated; the discriminator models the conditional probability, and its main goal is to find the optimal classification surface between different categories. Figure 1 As shown in Figure 1, in GAN, the input of the generator G is random noise z, and it continuously learns the probability distribution of the real data (physical field cloud map) in the training set. The goal is to convert the input random noise z into a physical field cloud map, that is, the generated physical field cloud map is as similar as possible to the physical field cloud map in the training set. The input of the discriminator D is the real physical field cloud map of the image training generated by the generator, and the output is the probability value of the physical field cloud map being true. Its function is to judge whether a physical field cloud map is a real physical field cloud map. The goal is to distinguish the physical field cloud map generated by the generator G from the physical field cloud map in the input training set. Due to the existence of the discriminator D, G can learn to approximate the real data well without a lot of prior knowledge and prior distribution, and finally make the data generated by the model achieve the effect of being indistinguishable from the real.

[0040] The original generative adversarial network GAN's discriminator D and generator G both use multi-layer perceptrons. GAN defines a noise As a priori, it is used to learn the generator G in the training data x The probability distribution on G( z ) represents the input noise z Generated physical field cloud data , D( x ) represents the probability from the real data distribution, so the optimization objective is as follows: ; Among them, represents the random noise data of the temperature field cloud map data training sample and the velocity field cloud map data training sample, G represents the generator, D represents the discriminator, G ( z ) represents the generated physical field cloud map data, D ( x ) represents the real physical field cloud map, x represents the real physical field cloud map sample, z represents the random noise vector, E represents the expected value, Pdata(x) represents the real physical field cloud map distribution.

[0041] The optimization objective function can be interpreted as: when updating the parameters of the discriminator D, for the samples Pdata from the real distribution x , the goal of GAN is that the output of the discriminator D( x ) is close to 1, that is, the larger logD( x ); for the data G( z ) generated by the noise z , the goal of GAN is that the output of the discriminator D(G( z )) is close to 0, that is, the larger log(1 - D ( G (z))), so it is necessary to maximize D. When updating the parameters of the generator G, the goal of the generator is that the generated physical field cloud map is as close as possible to the real physical field cloud map, that is, D(G( z )) is close to 1, and the smaller log(1 - D ( G (z))) is, so it is necessary to minimize G.

[0042] As a possible implementation manner, the target prediction device can be a reactor core, that is, the flow and heat transfer in the reactor core channels are predicted according to the geometric structure and physical characteristics of the reactor core. The target prediction device can also be a heat exchanger, a high-power electronic device, a boiler, etc.

[0043] As a possible implementation manner, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, a flow and heat transfer prediction method based on an adversarial neural network is implemented.

[0044] As a possible implementation manner, this embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, a flow and heat transfer prediction method based on an adversarial neural network is implemented.

[0045] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A flow heat transfer prediction method based on an adversarial neural network, characterized in that It includes the following specific steps: According to the geometric structure and physical properties of the target prediction device, perform mesh generation of the target prediction device in the multi-physics field framework; Based on the fluid phase change parameters, heat transfer parameters, and latent heat of phase change parameters of each grid, obtain the mass source term and the energy source term; Based on the mass source term and the energy source term, perform simulations to obtain the multi-physics field simulation results of the target prediction device; Construct a reduced-order model, and generate a training sample of the temperature field cloud map data and a training sample of the velocity field cloud map data according to the multi-physics field simulation results of the target prediction device; Construct an adversarial neural network model, and use the training sample of the temperature field cloud map data and the training sample of the velocity field cloud map data to train the adversarial neural network model to obtain a flow and heat transfer prediction model; Input the multi-physics field simulation results of the target prediction device into the flow and heat transfer prediction model to predict and generate a flow field cloud map and a temperature field cloud map.

2. The flow and heat transfer prediction method based on a generative adversarial network according to claim 1, characterized in that The steps for obtaining the mass source term specifically include: Obtain the grid volume to obtain the characteristic length of the grid cell; Obtain the phase change adjustment coefficient, the latent heat of phase change of the liquid, the gas density, the bubble temperature, the pressure inside the bubble, the molar mass of the gas, and the molar gas constant to obtain the condensation and evaporation heat transfer coefficients at the phase change interface; Obtain the latent heat of phase change of the liquid, the saturation temperature, and the grid cell temperature, and combine the condensation and evaporation heat transfer coefficients at the phase change interface to determine the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid; Obtain the mass transfer rate constraint condition of the vapor-liquid two-phase in the grid cell volume, and combine the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid to obtain the mass source term of each grid.

3. The flow and heat transfer prediction method based on the adversarial neural network according to claim 2, wherein The specific steps for obtaining the condensation and evaporation heat transfer coefficients at the phase change interface include: ； Among them, represents the phase change adjustment coefficient, that is, the non-ideal effect caused by the presence of non-condensable gas molecules on the vapor-liquid interface, h lg is the latent heat of phase change of the liquid, ρ v represents the gas density, T v is the bubble temperature, represents the molar mass of the gas, is the molar gas constant, p v is the pressure inside the bubble.

4. The flow heat transfer prediction method based on a generative adversarial network according to claim 2, wherein The specific steps for determining the mass transfer rate of the vapor-liquid two-phase per unit volume of each grid include: ； Among them, m lg represents the mass transfer rate of the vapor-liquid two-phase per unit volume within each grid; T sat represents the saturation temperature; T l represents the grid cell temperature; h ev represents the condensation and evaporation heat transfer coefficients of the phase change interface; h lg represents the latent heat of phase change of the liquid; L represents the characteristic length of the grid cell. The characteristic length of the cell is the cube root of the cell volume.

5. The flow and heat transfer prediction method based on the adversarial neural network according to claim 2, characterized in that The mass transfer rate constraint condition specifically includes: During condensation, ； During evaporation, ； Among them, ρ v represents the steam density, a v represents the gas-phase volume fraction, △ t represents the current calculation time step.

6. The flow and heat transfer prediction method based on the adversarial neural network according to claim 1, wherein, The steps for obtaining the energy source term specifically include: Obtain the specific heat capacity at constant pressure of the liquid phase, the specific heat capacity at constant pressure of the vapor phase, the number of time steps, the temperature corresponding to the number of time steps, and the reference temperature in each grid, and combine the mass transfer rate of the vapor-liquid two-phase per unit volume in each grid to obtain the energy source phase.

7. The flow and heat transfer prediction method based on a generative adversarial network according to claim 6, characterized in that The specific calculation steps for obtaining the energy source phase include: ； Among them, represents the energy source phase, m lg represents the mass transfer rate of the vapor-liquid two-phase per unit volume within each grid, c pl represents the specific heat capacity at constant pressure of the liquid phase, c pg represents the specific heat capacity at constant pressure of the vapor phase, T n represents the number of time steps n of the temperature, n represents the number of time steps, T ref represents the reference temperature.

8. The flow and heat transfer prediction method based on an adversarial neural network according to claim 1, wherein The steps for constructing the reduced-order model and generating a training sample of the temperature field cloud map data and a training sample of the velocity field cloud map data according to the multi-physics field simulation results of the target prediction device specifically include: Extract the original data of the temperature field and the velocity field from the multi-physics field simulation for preprocessing; Construct a reduced-order model, input the preprocessed data into the reduced-order model to generate a learning matrix; Perform singular value decomposition on the learning matrix to obtain the eigenvalues of the decomposed data; Using the first K eigenvalues of the decomposed data as the reduced-order basis, project the original data into a low-dimensional space to obtain the training samples of the temperature field contour data and the training samples of the velocity field contour data.

9. The flow and heat transfer prediction method based on an adversarial neural network according to claim 8, characterized in that, In the process of obtaining the eigenvalues of the decomposed data, it also includes calculating the average projection error and optimizing the number of eigenvalues of the decomposed data according to the average projection error K .

10. The flow and heat transfer prediction method based on the adversarial neural network according to claim 9, characterized in that, The calculation steps for the average projection error include: ； Among them, relativeError represents the average projection error, M represents the learning matrix, K represents the number of eigenvalues, σ represents the deviation, i represents from 1 to K +1.

11. The flow and heat transfer prediction method based on the adversarial neural network according to claim 1, wherein The steps for constructing the adversarial neural network model specifically include: Construct a generator; Obtain the random noise data of the training sample of the temperature field cloud map data and the training sample of the velocity field cloud map data, and input it into the generator; The generator trains on the random noise data of the multi-physics field simulation results of the target prediction device, learns the probability distribution of the physical field cloud map, and until the training reaches the set threshold, obtains the generated physical field cloud map; Construct a discriminator; Extract the real physical field cloud map based on the training samples of the temperature field cloud map data and the training samples of the velocity field cloud map data; Discriminate between the generated physical field cloud map and the real physical field cloud map, and optimize the generated physical field cloud map according to the discrimination result until the discrimination result reaches the set threshold to obtain a flow and heat transfer prediction model.

12. The flow heat transfer prediction method based on a generative adversarial network according to claim 11, wherein The construction calculation steps of the adversarial neural network model include: ； Among them, represents the random noise data of the training samples of the temperature field cloud map data and the velocity field cloud map data, G represents the generator, D represents the discriminator, G ( z ) represents the generated physical field cloud map data, D ( x ) represents the real physical field cloud map, x represents the real physical field cloud map sample, z represents the random noise vector, E represents the expected value, Pdata(x) represents the real physical field cloud map distribution.

13. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the flow and heat transfer prediction method based on the adversarial neural network according to any one of claims 1 to 12.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the flow and heat transfer prediction method based on the adversarial neural network according to any one of claims 1 to 12.

Images