Physical constraint double-discriminator GAN-based multivariable flow field correlation analysis method, apparatus and device, and storage medium

Abstract

The invention discloses a multivariable flow field correlation analysis method, device and equipment based on a physical constraint double-discriminator GAN and a storage medium, and relates to the technical field of data analysis, and the method comprises the steps: constructing a physical constraint double-discriminator GAN model to generate a feature coding variable of a target flow field variable; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator comprises a flow field data space discriminator and a flow field data time discriminator; dimensionality reduction is carried out on the feature coding variables through a dimensionality reduction visual space projection method to obtain a flow field variable distribution diagram, and a target multivariable association diagram is constructed based on the mutual relation between the variables in the flow field variable distribution diagram; and performing variable conversion between the target variable pairs by utilizing a physical characteristic relationship between the physical constraint double-discriminator GAN model and the target variable pairs in the target multivariable association graph. In this way, automatic conversion between flow field variables can be achieved, and relevance between data is comprehensively explored.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a multivariate flow field correlation analysis method, device, equipment and storage medium based on a physical constraint dual discriminator GAN. Background Art

[0002] Flow field data has multivariate properties, and there are complex and diverse interactions between variables. Extracting and expressing implicit similarities and differences between variables is an important means to understand flow field data and acquire flow field knowledge. At present, the association relationship of multivariate flow field data can be divided into five types of expressions: inter-variable, inter-voxel, inter-value, inter-feature, and mixed. With the continuous increase in the scale of flow field data and the continuous increase in the dimension of variables, the inter-voxel, inter-variable, and inter-value association analysis methods usually use correlation coefficients and mutual information to measure the similarity between variable pairs, which not only consumes a lot of hardware resources and computing costs, but also has redundant information that causes visual confusion, affecting the efficiency and accuracy of the analysis. In comparison, inter-feature and mixed association analysis methods reduce the amount of data to be processed by defining and extracting the main features in different variables, ignoring most of the redundant or irrelevant data, while maintaining the accuracy of the information, and can solve the problems of low efficiency and visual confusion in association analysis to a certain extent. However, the definition of flow field variable features in these methods often depends on specific analysis objectives and specific task scenarios, lacking universality, which makes it difficult for users to obtain comprehensive and integrated analysis results. Therefore, how to improve the efficiency of flow field data correlation analysis needs to be solved. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a multivariate flow field association analysis method, device, equipment and storage medium based on physical constraint dual discriminator GAN, which can improve the efficiency of flow field data association analysis. The specific scheme is as follows:

[0004] In a first aspect, the present application discloses a multivariate flow field correlation analysis method based on physical constraint dual discriminator GAN, comprising:

[0005] A physical constraint dual discriminator GAN model is constructed based on a first target loss function, a flow field data generator and a flow field data discriminator, and a feature encoding variable of a target flow field variable is generated based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator;

[0006] The feature encoding variables are mapped into a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map, and a target multivariate association map is constructed based on the relationship between the variables in the flow field variable distribution map;

[0007] A target variable pair is determined from the target multivariate association graph, and a variable conversion between the target variable pair is performed using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair.

[0008] Optionally, the flow field data spatial discriminator is used to judge the spatial flow field variables generated by the flow field data generator to obtain a spatial judgment result; the flow field data temporal discriminator is used to judge the continuous spatial flow field variables generated by the flow field data generator to obtain a temporal judgment result.

[0009] Optionally, the step of generating a feature encoding variable of a target flow field variable based on the physical constraint dual discriminator GAN model includes:

[0010] Inputting the target flow field variable into the physical constraint dual discriminator GAN model for training to obtain a trained GAN model, and retaining the encoder in the generator corresponding to the trained GAN model to obtain a variable feature extraction GAN model;

[0011] The target flow field variable is input into the variable feature extraction GAN model to obtain the feature encoding variable of the target flow field variable.

[0012] Optionally, inputting the target flow field variable into the physical constraint dual discriminator GAN model for training to obtain a trained GAN model includes:

[0013] Inputting the target flow field variables into the physical constraint dual discriminator GAN model, and generating spatial flow field variables through the flow field data generator in the physical constraint dual discriminator GAN model;

[0014] Generate a spatial determination result and a temporal determination result of the spatial flow field variable by the flow field data discriminator;

[0015] The physical constraint dual discriminator GAN model is adversarially trained based on the spatial flow field variables, the spatial judgment results and the temporal judgment results to obtain a trained GAN model.

[0016] Optionally, mapping the feature encoding variables into a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map includes:

[0017] The feature encoding variables are mapped to a two-dimensional scatter plot through a nonlinear manifold learning algorithm to obtain a flow field variable distribution map.

[0018] Optionally, constructing a target multivariate association diagram based on the relationship between the variables in the flow field variable distribution diagram includes:

[0019] The mutual relationship between the variables is determined based on the distance relationship between the variables in the flow field variable distribution diagram, and a target multivariate association diagram is constructed based on the mutual relationship.

[0020] Optionally, determining a target variable pair from the target multivariate association graph, and performing variable conversion between the target variable pair using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair, comprises:

[0021] Determining a target variable pair from the target multivariate association graph, and generating a second target loss function based on a physical feature relationship between the target variable pairs;

[0022] The second objective loss function is fused with the physical constraint dual discriminator GAN model to obtain a fused GAN model, and variable conversion between the target variable pairs is performed based on the fused GAN model.

[0023] In the second aspect, the present application discloses a multivariate flow field correlation analysis device based on physical constraint dual discriminator GAN, comprising:

[0024] A feature extraction module, used to construct a physical constraint dual discriminator GAN model based on a first target loss function, a flow field data generator and a flow field data discriminator, and generate feature encoding variables of a target flow field variable based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator;

[0025] An association diagram creation module is used to map the feature encoding variables into a two-dimensional scatter diagram through a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution diagram, and to construct a target multivariate association diagram based on the relationship between the variables in the flow field variable distribution diagram;

[0026] A variable conversion module is used to determine a target variable pair from the target multivariate association diagram, and perform variable conversion between the target variable pairs using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair.

[0027] In a third aspect, the present application discloses an electronic device, including:

[0028] Memory, used to store computer programs;

[0029] A processor is used to execute the computer program to implement the aforementioned multivariate flow field correlation analysis method.

[0030] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program, wherein the computer program implements the aforementioned multivariate flow field correlation analysis method when executed by a processor.

[0031] It can be seen that a physical constraint dual discriminator GAN model is constructed based on the first target loss function, the flow field data generator and the flow field data discriminator, and the feature encoding variable of the target flow field variable is generated based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator; the feature encoding variable is mapped to a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map, and a target multivariate association map is constructed based on the mutual relationship between the variables in the flow field variable distribution map; the target variable pair is determined from the target multivariate association map, and the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair is used to perform variable conversion between the target variable pair. That is, the learning method of the GAN model is used to mine the effective representation of each flow field variable in combination with the physical characteristics of the flow field data. Afterwards, the effective representation of each variable is mapped to a two-dimensional scatter plot through a visual spatial projection method based on dimensionality reduction. With the help of the scatter plot, the relationship between the variables in the original flow field is inferred, and a multivariate association diagram is constructed to support subsequent association analysis and variable conversion. In this way, by taking into account the physical characteristics of the flow field data to extract the flow field variable characteristics, the potential association relationship between variables can be intuitively and quickly revealed, and variable conversion can be supported, which helps to form a comprehensive and integrated analysis perspective. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0033] Figure 1 A flowchart of a multivariate flow field correlation analysis method based on physical constraint dual discriminator GAN disclosed in this application;

[0034] Figure 2 A schematic diagram of a network structure of a generator disclosed in this application;

[0035] Figure 3 A flowchart of a specific multivariate flow field correlation analysis method based on physical constraint dual discriminator GAN disclosed in this application;

[0036] Figure 4 This is a schematic diagram of the structure of a multivariate flow field correlation analysis device based on physical constraint dual discriminator GAN disclosed in this application;

[0037] Figure 5 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0038] 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.

[0039] At present, the definition of flow field variable features often depends on specific analysis objectives and specific task scenarios, and lacks universality, which makes it difficult for users to obtain comprehensive and integrated analysis results. Therefore, this application will specifically introduce a variable flow field correlation analysis method, which can better assist users in understanding the potential deep-level relationship between multivariate flow field data and discover unknown new phenomena.

[0040] See also Figure 1 As shown, the embodiment of the present application discloses a multivariate flow field correlation analysis method based on physical constraint dual discriminator GAN, including:

[0041] Step S11: construct a physical constraint dual discriminator GAN model based on the first target loss function, a flow field data generator and a flow field data discriminator, and generate feature encoding variables of the target flow field variables based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator.

[0042] In this embodiment, the potential physical knowledge of the flow field is first introduced into the construction of the loss function to guide the learning process of the GAN (Generative Adversarial Networks) model to ensure that the generated flow field conforms to physical consistency. The flow field data discriminator in this embodiment includes a flow field data space discriminator and a flow field data time discriminator. Specifically, the flow field data space discriminator is used to judge the spatial flow field variables generated by the flow field data generator to obtain a spatial judgment result; the flow field data time discriminator is used to judge the continuous spatial flow field variables generated by the flow field data generator to obtain a time judgment result. In other words, the spatial discriminator is used to judge the spatial flow field generated by the generator to effectively capture spatial details, while the other is a time discriminator, which is used to judge the continuous multiple spatial flow fields generated by the generator, aiming to capture potential time patterns. In this way, through the dual discriminator model, the generator is guided together, which not only ensures that the generated flow field has spatiotemporal consistency, but also effectively avoids the occurrence of problems such as mode collapse and gradient disappearance. In the model, the generator is divided into two parts: the encoder and the decoder. The encoder includes three learning modules, each of which contains a convolutional layer and a batch normalization layer. The decoder is similar to the encoder. It only needs to change the convolutional layer in the learning module of the encoder into a deconvolutional layer, such as Figure 2 As shown, each square at the feature represents a one-dimensional vector. After the flow field data passes through each learning module in the encoder, the flow field scale is halved, and after each learning module in the decoder, the flow field scale is doubled, and finally the generated flow field of the same scale is obtained. The discriminator is essentially a binary classifier, whose task is to distinguish true and false flow fields as much as possible and give a discriminant score for the authenticity of the flow field. A score of 1 means that the generated flow field is considered to be a true flow field, and a score of 0 means that the generated flow field comes from the generator. First, the flow field generated by the generator is taken as input, and then passes through three discriminant modules in sequence. Each module contains a convolution layer and a spectral normalization layer to obtain the flow field feature map, and finally passes through the fourth convolution layer to obtain the final judgment result. During the training process, the temporal discriminator and the spatial discriminator have the same structure, do not share parameters, and are trained separately. The loss function of the discriminator and the generator uses the standard adversarial loss. The difference is that the loss function of the generator adds the potential physical constraint term of the flow field. In the flow field data, the gradient of the variable encodes key information related to shear, divergence and continuity, which can punish false oscillations in the flow field and maintain the inherent physical properties. Specifically, physical constraints are imposed on the flow field generated by the generator by using physical features such as gradient information, thereby ensuring that the generated flow field conforms to physical laws. , Time Discriminator and spatial discriminator The adversarial loss function The definition is as follows:

[0043] ;

[0044] ;

[0045] ;

[0046] in, represents the expected value of a variable, represents the generated flow field, represents the real flow field, represents the flow field generated by the spatial discriminator; represents the flow field generated by the temporal discriminator; x is a value in the generated flow field; z is a value in the real flow field. In this embodiment, in order to improve the accuracy, the content loss function is defined for:

[0047] ;

[0048] in Represents the mean absolute error of the element. To ensure that the generated flow field conforms to the underlying physical knowledge, a physical loss function is defined for:

[0049] ;

[0050] in express To ensure that the generated flow field can capture high-frequency details, a frequency loss function based on Fourier transform is defined , as shown below:

[0051] ;

[0052] in, Indicated by frequency band Filtered variables The Fourier transform of Indicates each frequency band The weight of Set to 5, Set to Considering all loss functions, the first objective loss function of the generator is defined as:

[0053] ;

[0054] in, , and It is a hyperparameter with a value range between 0 and 1, which is used to control the relative importance of adversarial loss, gradient loss and frequency loss. In the present invention, these three parameters are set to 0.5, 0.01 and 0.001 respectively.

[0055] In this embodiment, the feature encoding variable of the target flow field variable based on the physical constraint dual discriminator GAN model includes: inputting the target flow field variable into the physical constraint dual discriminator GAN model for training to obtain a trained GAN model. Specifically, the target flow field variable is input into the physical constraint dual discriminator GAN model, and the spatial flow field variable is generated by the flow field data generator in the physical constraint dual discriminator GAN model; the spatial determination result and the time determination result of the spatial flow field variable are generated by the flow field data discriminator; the physical constraint dual discriminator GAN model is adversarially trained based on the spatial flow field variable, the spatial determination result and the time determination result to obtain the trained GAN model. Then, the encoder in the generator corresponding to the trained GAN model is retained to obtain a variable feature extraction GAN model; the target flow field variable is input into the variable feature extraction GAN model to obtain the feature encoding variable of the target flow field variable. In other words, the flow field variable is unsupervisedly trained based on the physical constraint dual discriminator GAN model, and the low-dimensional features of the flow field variable can be obtained. After the model training is completed, the trained GAN model is obtained. Only the encoder in the generator is retained. The original flow field variables are sent to the encoder output to obtain the feature encoding of the flow field variables, which is directly used as the input for constructing the variable association graph. These low-dimensional features are not only independent of specific fields, but also provide universal and unified definitions and descriptions, and also reduce the amount of data required for association analysis.

[0056] Step S12: Mapping the feature encoding variables into a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map, and constructing a target multivariate association map based on the relationship between the variables in the flow field variable distribution map.

[0057] In this embodiment, in order to more accurately reflect the structural characteristics of the flow field variables and fully retain the richness of the flow field variables, the flow field variable features obtained based on the physical constraint dual discriminator GAN model are usually still high-dimensional data. If the high-dimensional data is directly subjected to association analysis, it is often difficult to intuitively display the differences and potential relationships between the variables due to the high dimension. Therefore, the feature encoding variable is mapped to a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map, including: mapping the feature encoding variable to a two-dimensional scatter plot by a nonlinear manifold learning algorithm to obtain a flow field variable distribution map. Therefore, by processing the high-dimensional data through a dimensionality reduction method, the effective representation of each variable is mapped to a two-dimensional scatter plot, and the distribution and potential association of different variables in the flow field are intuitively displayed, which effectively alleviates the problem of visual confusion, avoids data understanding bias and misleading caused by redundant information, and thus improves the accuracy and efficiency of multivariate relationship exploration. Considering that the nonlinear manifold learning algorithm t-SNE (t-distributed stochastic neighbor embedding, t-distributed random field embedding) has efficient dimensionality reduction capabilities and excellent visualization effects. Therefore, in this embodiment, the nonlinear manifold learning algorithm maps the feature encoding variables into a two-dimensional scatter plot to obtain a flow field variable distribution map. The high-dimensional data set is effectively reduced to a two-dimensional scatter plot, and the original multivariate flow field sequence is visualized to display the change trend and distribution characteristics between different variables. Then, the user end can iteratively select different variable pairs in combination with the two-dimensional scatter plot to calculate the distance between the scatter points, infer the relationship between the variables in the original flow field, and thus construct a multivariate association map. Figure 3 The fluid simulation big data of the supersonic aircraft is reduced in dimension to a two-dimensional scatter plot, and then the mutual relationship between the variables is determined based on the distance relationship between the variables in the flow field variable distribution diagram, and a target multivariate association diagram is constructed based on the mutual relationship. For example, the distance relationship between the velocity component W and the velocity component V of the supersonic aircraft is calculated to determine the mutual relationship between the variables, and a target multivariate association diagram is constructed based on the mutual relationship.

[0058] Step S13: Determine a target variable pair from the target multivariate association diagram, and perform variable conversion between the target variable pair using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair.

[0059] The flow field variable conversion task is similar to the image conversion task, but the difficulty and requirements of flow field variable conversion are higher than image conversion. Image conversion only requires seeing the basic content of the new image, while flow field variable conversion requires data exploratory analysis capabilities, which requires that when converting variables, not only the basic information and topological structure of the flow field are restored, but also the relationship between each grid point of the flow field is restored. Therefore, in this embodiment, the target variable pair is determined from the target multivariable association diagram, and the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair is used to perform variable conversion between the target variable pairs, including: determining the target variable pair from the target multivariable association diagram, and generating a second target loss function based on the physical feature relationship between the target variable pair; fusing the second target loss function with the physical constraint dual discriminator GAN model to obtain a fused GAN model, and performing variable conversion between the target variable pairs based on the fused GAN model. For example Figure 3 As shown, the velocity component W and the velocity component V are determined as the target data pair from the target multivariate association diagram, and then the second target loss function is generated according to the physical characteristic relationship of the target data pair. That is, the idea of ​​physically embedded neural network is borrowed to increase the loss function that integrates physical information. First, taking the two-dimensional Navier-Stokes (NS) equation as an example, it can be derived into a differential equation of the following form. Based on this differential equation, the loss function is designed as follows:

[0060] ;

[0061] ;

[0062] in, and It means that the function is defined as the NS equation, and Respectively and The velocity field in the direction, represents the pressure field, and Respectively of and The first-order partial derivative in the direction, and Respectively of and The first-order partial derivative in the direction, and Respectively of and The first-order partial derivative in the direction, and Respectively of and The second-order partial derivative in the direction, and express of and The CFD (Computational Fluid Dynamics) value is the solution of this equation under given initial conditions. and Since the velocity and pressure of each point in the generated flow field must satisfy the NS equation, that is, In order to introduce the NS equation constraint into the intelligent model, taking the conversion between speed variables selected by the variable association diagram as an example, the present invention adds a second objective loss function As shown below:

[0063] ;

[0064] Among them, t i Represents the time dimension information, t i Time, x i ,y i is the spatial position information, x i and i Position. Embed them into the loss function to guide network training, so that the transformed flow field is physically consistent. represents the number of grid points in the sampling block, and represents the velocity value of the corresponding grid point in the real velocity field, and It represents the velocity value of the corresponding grid point in the predicted velocity field. By minimizing the reconstruction error and NS equation of the two velocity components at the same time, the converted flow field variables are closer to the flow field variables before the conversion. Then, the physical constraint dual discriminator GAN model is transferred and learned using the second objective loss function to obtain the fused GAN model, and then the fused GAN model is used to perform variable conversion on the target variable pair. For example, the velocity component W is input into the fused GAN model and converted into the velocity component V. After variable conversion, the organic fusion of different variables can be achieved, and different drawing styles can be used for mixed drawing of variables with large differences, which supports comprehensive exploration of the correlation between data and helps to discover new scientific phenomena.

[0065] It can be seen that a physical constraint dual discriminator GAN model is constructed based on the first target loss function, the flow field data generator and the flow field data discriminator, and the feature encoding variable of the target flow field variable is generated based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator; the feature encoding variable is mapped to a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map, and a target multivariate association map is constructed based on the mutual relationship between the variables in the flow field variable distribution map; the target variable pair is determined from the target multivariate association map, and the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair is used to perform variable conversion between the target variable pair. That is, the learning method of the GAN model is used to mine the effective representation of each flow field variable in combination with the physical characteristics of the flow field data. Afterwards, the effective representation of each variable is mapped to a two-dimensional scatter plot through a visual spatial projection method based on dimensionality reduction. With the help of the scatter plot, the relationship between the variables in the original flow field is inferred, and a multivariate association diagram is constructed to support subsequent association analysis and variable conversion. In this way, by taking into account the physical characteristics of the flow field data to extract the flow field variable characteristics, the potential association relationship between variables can be intuitively and quickly revealed, and variable conversion can be supported, which helps to form a comprehensive and integrated analysis perspective.

[0066] refer to Figure 4 The embodiment of the present application also discloses a multivariate flow field correlation analysis device based on a physical constraint dual discriminator GAN, including:

[0067] The feature extraction module 11 is used to construct a physical constraint dual discriminator GAN model based on a first target loss function, a flow field data generator and a flow field data discriminator, and generate feature encoding variables of a target flow field variable based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator;

[0068] An association diagram creation module 12 is used to map the feature encoding variables into a two-dimensional scatter diagram through a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution diagram, and to construct a target multivariate association diagram based on the relationship between the variables in the flow field variable distribution diagram;

[0069] The variable conversion module 13 is used to determine the target variable pair from the target multivariate association diagram, and perform variable conversion between the target variable pair by using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair.

[0070] It can be seen that in this embodiment, by taking into account the physical characteristics of the flow field data to extract the flow field variable characteristics, the potential correlation between variables can be revealed intuitively and quickly, and variable conversion can be supported, which helps to form a comprehensive and integrated analysis perspective.

[0071] In some specific embodiments, the feature extraction module 11 may specifically include:

[0072] A model generation submodule, used for inputting the target flow field variable into the physical constraint dual discriminator GAN model for training to obtain a trained GAN model, and retaining the encoder in the generator corresponding to the trained GAN model to obtain a variable feature extraction GAN model;

[0073] A feature extraction unit is used to input the target flow field variable into the variable feature extraction GAN model to obtain a feature encoding variable of the target flow field variable.

[0074] In some specific embodiments, the model generation submodule may specifically include:

[0075] A flow field variable generating unit, used for inputting the target flow field variable into the physical constraint dual discriminator GAN model, and generating the spatial flow field variable through the flow field data generator in the physical constraint dual discriminator GAN model;

[0076] A determination result generating unit, used for generating a spatial determination result and a temporal determination result of the spatial flow field variable through the flow field data discriminator;

[0077] A model training unit is used to perform adversarial training on the physical constraint dual discriminator GAN model based on the spatial flow field variable, the spatial judgment result and the temporal judgment result to obtain a trained GAN model.

[0078] In some specific embodiments, the association graph creation module 12 may be specifically used to map the feature encoding variables into a two-dimensional scatter plot through a nonlinear manifold learning algorithm to obtain a flow field variable distribution graph.

[0079] In some specific embodiments, the association diagram creation module 12 can be specifically used to determine the mutual relationship between the variables based on the distance relationship between the variables in the flow field variable distribution diagram, and construct a target multivariate association diagram based on the mutual relationship.

[0080] In some specific embodiments, the variable conversion module 13 may specifically include:

[0081] a variable pair determination unit, configured to determine a target variable pair from the target multivariate association diagram, and generate a second target loss function based on a physical feature relationship between the target variable pair;

[0082] A variable conversion unit is used to fuse the second objective loss function with the physical constraint dual discriminator GAN model to obtain a fused GAN model, and perform variable conversion between the target variable pairs based on the fused GAN model.

[0083] Furthermore, the present application also discloses an electronic device. Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0084] Figure 5 A schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the multivariate flow field association analysis method based on the physical constraint dual discriminator GAN disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0085] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0086] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0087] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the multivariate flow field correlation analysis method based on physical constraint dual discriminator GAN performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks.

[0088] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the multivariate flow field correlation analysis method based on the physical constraint dual discriminator GAN disclosed above is implemented. For the specific steps of the method, reference can be made to the corresponding contents disclosed in the aforementioned embodiments, and no further description will be given here.

[0089] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred 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 be referred to the method part.

[0090] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0091] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0092] Finally, it should be noted that, in this article, 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 "include", "comprise" 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, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0093] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A multivariate flow field correlation analysis method based on physical constraint dual discriminator GAN, characterized in that: include: A physical constraint dual discriminator GAN model is constructed based on a first target loss function, a flow field data generator and a flow field data discriminator, and a feature encoding variable of a target flow field variable is generated based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator; The feature encoding variables are mapped into a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution map, and a target multivariate association map is constructed based on the relationship between the variables in the flow field variable distribution map; A target variable pair is determined from the target multivariate association graph, and a variable conversion between the target variable pair is performed using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair.

2. The multivariate flow field correlation analysis method according to claim 1, characterized in that: The flow field data space discriminator is used to judge the spatial flow field variables generated by the flow field data generator to obtain a spatial judgment result; The flow field data time discriminator is used to judge the continuous space flow field variables generated by the flow field data generator to obtain a time judgment result.

3. The multivariate flow field correlation analysis method according to claim 1, characterized in that: The generating of feature encoding variables of target flow field variables based on the physical constraint dual discriminator GAN model includes: Inputting the target flow field variable into the physical constraint dual discriminator GAN model for training to obtain a trained GAN model, and retaining the encoder in the generator corresponding to the trained GAN model to obtain a variable feature extraction GAN model; The target flow field variable is input into the variable feature extraction GAN model to obtain the feature encoding variable of the target flow field variable.

4. The multivariate flow field correlation analysis method according to claim 3, characterized in that: The step of inputting the target flow field variable into the physical constraint dual discriminator GAN model for training to obtain a trained GAN model includes: Inputting the target flow field variables into the physical constraint dual discriminator GAN model, and generating spatial flow field variables through the flow field data generator in the physical constraint dual discriminator GAN model; Generate a spatial determination result and a temporal determination result of the spatial flow field variable by the flow field data discriminator; The physical constraint dual discriminator GAN model is adversarially trained based on the spatial flow field variables, the spatial judgment results and the temporal judgment results to obtain a trained GAN model.

5. The multivariate flow field correlation analysis method according to claim 1, characterized in that: The method of mapping the feature encoding variables into a two-dimensional scatter plot by a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution diagram includes: The feature encoding variables are mapped to a two-dimensional scatter plot through a nonlinear manifold learning algorithm to obtain a flow field variable distribution map.

6. The multivariate flow field correlation analysis method according to claim 1, characterized in that: The constructing of a target multivariate association diagram based on the relationship between the variables in the flow field variable distribution diagram comprises: The mutual relationship between the variables is determined based on the distance relationship between the variables in the flow field variable distribution diagram, and a target multivariate association diagram is constructed based on the mutual relationship.

7. The multivariate flow field correlation analysis method according to any one of claims 1 to 6, characterized in that: The determining of the target variable pair from the target multivariate association graph and performing variable conversion between the target variable pair using the physical constraint dual discriminator GAN model and the physical feature relationship between the target variable pair include: Determining a target variable pair from the target multivariate association graph, and generating a second target loss function based on a physical feature relationship between the target variable pairs; The second objective loss function is fused with the physical constraint dual discriminator GAN model to obtain a fused GAN model, and variable conversion between the target variable pairs is performed based on the fused GAN model.

8. A multivariate flow field correlation analysis device based on physical constraint dual discriminator GAN, characterized in that: include: A feature extraction module, used to construct a physical constraint dual discriminator GAN model based on a first target loss function, a flow field data generator and a flow field data discriminator, and generate feature encoding variables of a target flow field variable based on the physical constraint dual discriminator GAN model; the first target loss function is a loss function constructed based on the physical characteristics of the target flow field variable; the flow field data discriminator includes a flow field data space discriminator and a flow field data time discriminator; An association diagram creation module is used to map the feature encoding variables into a two-dimensional scatter diagram through a preset dimensionality reduction visualization space projection method to obtain a flow field variable distribution diagram, and to construct a target multivariate association diagram based on the relationship between the variables in the flow field variable distribution diagram; A variable conversion module is used to determine a target variable pair from the target multivariate association diagram, and perform variable conversion between the target variable pairs using the physical feature relationship between the physical constraint dual discriminator GAN model and the target variable pair.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the multivariate flow field correlation analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the multivariate flow field correlation analysis method according to any one of claims 1 to 7.

Images