Hypersonic target wake electron density flow field reconstruction method, system and device based on physical information neural network and medium

Through the flow field reconstruction method based on physical information neural network, the loss function is defined using CFD simulation and N-S equations, the problem of reconstructing electron density data in hypersonic flow field is solved, and high-precision and reliability flow field data are achieved Refactoring, suitable for multiple application fields.

CN118917234AActive Publication Date: 2024-11-08XIDIAN UNIV

Patent Information

Application Number
CN202410945327.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-08
Estimated Expiration
2044-07-15

AI Technical Summary

Technical Problem

The prior art is difficult to effectively reconstruct complete electron density data in hypersonic flow fields, especially in experimental scenarios where the flow field has a great impact, and convolutional neural networks have existed in complex and uncertain scenarios. The problem of fitting and lack of interpretability.

Method used

Using a method based on physical information neural network, the electron density numerical and two-dimensional distribution is obtained through CFD simulation calculation, a flow field reconstruction model is constructed, and the loss function is defined by the N-S equation for training to generate a complete two-dimensional electron density distribution of the target trail flow field .

Benefits of technology

It realizes high accuracy and reliability of flow field data reconstruction, can adapt to the application background of different Mach numbers, and provides reliable data support in fields such as hypersonic aircraft experiments and plasma diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917234A_ABST
    Figure CN118917234A_ABST
Patent Text Reader

Abstract

The invention discloses a hypersonic target wake electron density flow field reconstruction method, system, equipment and medium based on a physical information neural network, and the method comprises the steps: respectively obtaining an electron density numerical value and electron density two-dimensional distribution through CFD simulation calculation, respectively storing the electron density numerical value and the electron density two-dimensional distribution, and sorting the electron density numerical value and the electron density two-dimensional distribution to obtain a training set; training the training set by using the flow field reconstruction model under the supervision of the loss function so as to obtain a flow field reconstruction model with a trained weight, and inputting axis electron density vector data with unknown two-dimensional distribution into the flow field reconstruction model with the trained weight so as to obtain two-dimensional electron density distribution of a target wake flow field; the system, the device and the medium are used for realizing the hypersonic target wake electron density flow field reconstruction method. The method has the advantages of high practicability, high reliability and high accuracy, and can be widely applied to the fields of hypersonic flight vehicle experiment and design, plasma diagnosis, material scientific research, quantum calculation and simulation and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of hypersonic flow field reconstruction, and in particular to a method, system, device and medium for reconstructing the electron density flow field of a hypersonic target wake based on a physical information neural network. Background Art

[0002] The navigation process of a hypersonic target will produce ions, electrons and other components, forming a plasma sheath at the head and a wake plasma at the downstream. In experimental research, it is often necessary to measure the electron density in the wake. In experimental scenarios such as wind tunnel shock tubes, researchers will use probes and other equipment to measure the electron density at a certain point in the flow field, but because the probe will affect the surrounding flow field during the experiment, only a small number of measurement points will be designed on the wake axis in actual operation. However, scientific research often requires electron density data for the entire flow field, and experimental measurement schemes cannot meet the requirements.

[0003] The graph convolutional neural network (GCN) model encodes a batch of data into a set of nodes and uses these nodes as the input of the model. The model structure usually consists of multiple layers of graph convolution layers and activation layers. As a generalization and application of convolutional neural networks on graph structured data, GCN can achieve end-to-end learning of node feature information and structural information, and has shown good performance on multiple public datasets. Although GCN has excellent performance in many fields, it still faces some challenges in some application scenarios with unclear contours, complexity and uncertainty. Especially in the application scenario of the wake, since the electron density of the flow field in the head shock wave will gradually stabilize as it moves away from the aircraft, this characteristic may bring certain difficulties to the modeling ability of GCN.

[0004] Similar technologies include image super-resolution technology, which aims to use multi-layer nonlinear transformations to extract high-level abstract features of data and learn the potential distribution patterns of data, thereby acquiring the ability to make reasonable judgments or predictions about new data. Similarly, the nonlinear transformation in this method is also achieved through multiple convolutional layers. Image super-resolution builds a network model with a combined convolutional neural network and multiple residual blocks, and designs a loss function based on prior knowledge, ultimately achieving image reconstruction and prediction capabilities. Image super-resolution technology can improve the enhancement effects of flow field, remote sensing and other images, and can obtain high-precision and high-efficiency remote sensing image target recognition results. However, for the prediction function, it is limited to the enhancement of local fuzzy features and cannot meet the application background of reconstructing vector data into matrix data.

[0005] In addition, for the above two solutions, their structures are based on convolutional layers, which results in a large amount of data requirements for GNN and image super-resolution. However, the material and time costs required for flow field experiments are high, which cannot meet the large amount of data requirements. In addition, the network is prone to overfitting when faced with such a small amount of data. Although convolutional neural networks can achieve transformations such as translation and rotation, they are poor at handling more complex spatial transformations such as scale changes and nonlinear distortions. Although data enhancement can be used to alleviate the problem, it cannot completely solve the problem. For physical property problems, convolutional neural networks are "black box" models that lack an explainable working mechanism and decision-making process.

[0006] Zhu Rui (Zhu Rui. Research on image recognition algorithm based on convolutional neural network [D]. Beijing University of Posts and Telecommunications, 2018.) proposed a study on image recognition algorithm based on convolutional neural network. Based on the compression and acceleration technology of image recognition model of convolutional neural network, this study proposed an image recognition algorithm based on artificial neural network with low complexity, high recognition accuracy, high efficiency, and fast real-time. However, image recognition algorithm is prone to overfitting when faced with less data. In the application scenario of wake, the electron density of the flow field in the head shock wave will gradually stabilize as it moves away from the aircraft. This scenario will bring difficulties to graph convolutional neural network. The output results cannot meet the physical mechanism and lack an explainable working mechanism and decision-making process.

[0007] Liu Chao (Liu Chao. Research and Implementation of Image Super-resolution Reconstruction Algorithm Based on Deep Learning [D]. University of Electronic Science and Technology of China, 2023. DOI: 10.27005 / d.cnki.gdzku.2022.001896.) proposed a research and implementation of an image super-resolution reconstruction algorithm based on deep learning. Specifically, the generative adversarial network model based on multi-scale feature extraction was improved. The model uses a densely connected multi-scale feature extraction module in the generator to extract image features, and uses a progressive two-stage reconstruction process instead of the original single-stage reconstruction process. It can accurately extract original features and high-level features, alleviate problems such as high computational complexity and gradient disappearance, and improve the network's ability to express features. At the same time, it effectively improves the quality of the reconstructed image. However, for physical property problems, convolutional neural networks are "black box" models that lack an interpretable working mechanism and decision-making process. Summary of the invention

[0008] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method, system, equipment and medium for reconstructing the electron density flow field of a hypersonic target wake based on a physical information neural network. The electron density value and the two-dimensional distribution of the electron density are obtained by CFD simulation calculation, and are stored respectively, and a training set is obtained by sorting them out. The training set is trained with a flow field reconstruction model under the supervision of a loss function to obtain a flow field reconstruction model with trained weights. The axis electron density vector data of the unknown two-dimensional distribution is input into the flow field reconstruction model with trained weights to obtain the two-dimensional electron density distribution of the target wake flow field. The method has the advantages of strong practicability, good reliability and high accuracy, and can be widely used in hypersonic aircraft experiments and design, plasma diagnosis, materials science research, quantum computing and simulation and other fields.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is:

[0010] A method for reconstructing the electron density flow field of a hypersonic target wake based on a physical information neural network comprises the following steps:

[0011] Step 1, obtaining the electron density value on the central axis of the target wake flow field with different Mach numbers, storing the electron density value as vector data, and obtaining the axis electron density vector; obtaining the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located, and storing the two-dimensional distribution of the electron density as matrix data; arranging the vector data and the matrix data into a training set;

[0012] Step 2: construct a flow field reconstruction model, wherein the flow field reconstruction model includes a first fully connected layer, an activation layer, a deconvolution layer, a flattening layer, a second fully connected layer, a third fully connected layer, and a reorganization layer; define the continuity equation in the NS equation as the loss function of the flow field reconstruction model;

[0013] Step 3, using the flow field reconstruction model constructed in step 2 to train the training set in step 1 in the hypersonic target wake flow field at different Mach numbers, and using the loss function of step 2 to supervise the training until the loss function converges, thereby obtaining a flow field reconstruction model with trained weights;

[0014] Step 4: Use the axis electron density vector data of the unknown two-dimensional distribution as the input vector, and use the flow field reconstruction model with the trained weights in step 3 to predict the input vector to obtain the two-dimensional electron density distribution of the target wake flow field.

[0015] In the step 1, the electron density value on the central axis and the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located are obtained by CFD simulation calculation.

[0016] The flow field reconstruction model of step 2 is used to generate two-dimensional flow field data. The input layer receives vector data, the first fully connected layer increases the data volume, and adjusts the data value through the activation layer; the two deconvolution layers perform feature reconstruction and increase the data dimension, and the high-dimensional data output by the deconvolution layer is flattened into vector data using the flattening layer. The data volume is standardized through the second fully connected layer and the third fully connected layer, and finally the data is reshaped through the reorganization layer, and finally the data is output to the desired size.

[0017] The step 2 defines the continuity equation in the NS equation as the loss function of the flow field reconstruction model, specifically including:

[0018] The continuity equation of NS electrons in a steady fluid is used for spatial gradient feature extraction and discretized using the central difference method. The discretized continuity equation expression is used as a conservation term, and the mean square error between the true value of the conservation term and the predicted value constitutes a loss function. The loss function is combined with the flow field reconstruction model so that the loss function is reversely propagated in the model to optimize the model weights.

[0019] The axis electron density vector data of the unknown two-dimensional distribution in step 4 is the electron number density measured by the probe on the central axis of the blunt cone in the ground test.

[0020] A hypersonic target wake electron density flow field reconstruction system based on physical information neural network, comprising:

[0021] Data acquisition module: The electron density values ​​on the central axis of the wake flow field of the target with different Mach numbers and the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located are obtained by CFD simulation calculation, and the electron density values ​​are stored as vector data to obtain the axis electron density vector; the two-dimensional distribution of the electron density is stored as matrix data, and finally the vector data and the matrix data are organized into a training set;

[0022] Deep learning module: The deep learning model consists of two main parts: one is to construct a flow field reconstruction model and define the loss function of the flow field reconstruction model. The flow field reconstruction model is composed of the first fully connected layer, activation layer, deconvolution layer, flattening layer, second fully connected layer, third fully connected layer and reorganization layer. The loss function is defined by the continuity equation in the NS equation. The second is the training and supervision of the flow field reconstruction model. The flow field reconstruction model is trained under the supervision of the loss function.

[0023] Prediction module: Use the unknown two-dimensional distributed axis electron density vector data as the input vector of the flow field reconstruction model. The trained flow field reconstruction model reconstructs the input vector to obtain the two-dimensional electron density distribution of the target wake flow field where the vector data is located.

[0024] A device for reconstructing the electron density flow field of a super-target wake based on a physical information neural network, comprising:

[0025] Memory: used to store computer programs to implement the above-mentioned method for reconstructing the electron density flow field of the wake of a super-superior target based on a physical information neural network;

[0026] Processor: used to implement the above-mentioned method for reconstructing the electron density flow field of the wake of a super-target based on a physical information neural network when executing the computer program.

[0027] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] Step 1 of the present invention utilizes the electron density on the central axis of the hypersonic target wake flow field to reconstruct the flow field, which is consistent with the actual experimental setting and has strong practicality. The flow field data can be reconstructed through the central axis measurement data. Therefore, the reconstructed flow field data is highly correlated with the experimental data and has high accuracy.

[0030] Step 2 of the present invention uses the NS equation to define the loss function of the flow field reconstruction model. Compared with the existing graph convolutional neural network and image super-resolution technology, it has the advantages of an explainable physical mechanism and is easier to converge with a small amount of training data.

[0031] Step 3 of the present invention uses a training set composed of super-target wake flow field data of different Mach numbers to train the flow field reconstruction model, which has the advantage of being adaptable to different application backgrounds.

[0032] Step 4 of the present invention can obtain a complete two-dimensional electron density distribution of the target wake flow field, which can provide reliable data support for the calculation of target electromagnetic scattering, stimulated electromagnetic radiation, and infrared and ultraviolet radiation characteristics;

[0033] In summary, through the above method, the present invention can obtain a complete two-dimensional electron density distribution of the target wake flow field, solving the problem that the probe will affect the surrounding flow field in the experiment, resulting in a small number of measurement points designed on the wake axis; the present invention has the advantages of strong practicality, good reliability and high accuracy, and can be widely used in hypersonic aircraft experiments and design, plasma diagnosis, materials science research, quantum computing and simulation and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The figure is a flow chart of the method of the present invention.

[0035] Figure 2 This is a structural diagram of the physical information coupled flow field reconstruction model of the present invention.

[0036] Figure 3 The hypersonic aircraft geometric model is used by the present invention to obtain the wake electron density simulation.

[0037] Figure 4 This is the convergence process of the loss function training of the flow field reconstruction model of the present invention.

[0038] Figure 5 It is the predicted result and true value of the two-dimensional tail electron number density of the present invention.

[0039] Figure 6 For the present invention Figure 5 Schematic diagram of electron density data extraction.

[0040] Figure 7 This is a comparison chart of the effects of a convolutional neural network (CNN network) and the flow field reconstruction model provided by the present invention.

[0041] Figure 8 It is a comparison diagram of the correlation between the prediction results of the present invention and the true value. DETAILED DESCRIPTION

[0042] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0043] See also Figure 1 , a hypersonic target wake electron density flow field reconstruction method based on physical information neural network, comprising the following steps:

[0044] Step 1, obtaining the electron density value on the central axis of the target wake flow field with different Mach numbers, storing the electron density value as vector data, and obtaining the axis electron density vector; obtaining the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located, and storing the two-dimensional distribution of the electron density as matrix data; arranging the vector data and the matrix data into a training set;

[0045] Step 2: construct a flow field reconstruction model, wherein the flow field reconstruction model includes a first fully connected layer, an activation layer, a deconvolution layer, a flattening layer, a second fully connected layer, a third fully connected layer, and a reorganization layer; define the continuity equation in the NS equation as the loss function of the flow field reconstruction model;

[0046] Step 3, using the flow field reconstruction model constructed in step 2 to train the training set in step 1 in the hypersonic target wake flow field at different Mach numbers, and using the loss function of step 2 to supervise the training until the loss function converges, thereby obtaining a flow field reconstruction model with trained weights;

[0047] Step 4: Use the axis electron density vector data of the unknown two-dimensional distribution as the input vector, and use the flow field reconstruction model with the trained weights in step 3 to predict the input vector to obtain the two-dimensional electron density distribution of the target wake flow field.

[0048] In step 1, the electron density values ​​on the central axis of the target wake flow field at different Mach numbers and the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located are both obtained by CFD simulation calculation.

[0049] The hypersonic vehicle geometry model established by this model is used to perform CFD simulation of the target flow field. In the simulation, the k-ω model is used to describe the turbulent kinetic energy in the fluid, and the Gupta11 component chemical reaction model is used to describe the content change and reaction degree of each component in the fluid. The initial Mach number is set to 11Ma, 12Ma, ..., 18Ma for simulation. The simulation is terminated after the electron density residual converges, and the simulation results are obtained. The electron density value on the central axis is extracted as a vector data of 1×n, and the overall flow field is saved as a matrix data of n×n.

[0050] The flow field reconstruction model of step 2 is used to generate two-dimensional flow field data. The input layer receives vector data, the first fully connected layer increases the data volume, and adjusts the data value through the activation layer; the two deconvolution layers perform feature reconstruction and increase the data dimension, and the high-dimensional data output by the deconvolution layer is flattened into vector data using the flattening layer. The data volume is standardized through the second fully connected layer and the third fully connected layer, and finally the data is reshaped through the reorganization layer, and finally the data is output to the desired size.

[0051] The step 2 defines the continuity equation in the NS equation as the loss function of the flow field reconstruction model, specifically including:

[0052] The continuity equation of NS electrons in a steady fluid is used for spatial gradient feature extraction and discretized using the central difference method. The discretized continuity equation expression is used as a conservation term, and the mean square error between the true value of the conservation term and the predicted value constitutes a loss function. The loss function is combined with the flow field reconstruction model so that the loss function is reversely propagated in the model to optimize the model weights.

[0053] The axis electron density vector data of the unknown two-dimensional distribution in step 4 is the electron number density measured by the probe on the central axis of the blunt cone in the ground test.

[0054] A hypersonic target wake electron density flow field reconstruction system based on physical information neural network, comprising:

[0055] Data acquisition module: The electron density values ​​on the central axis of the wake flow field of the target with different Mach numbers and the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located are obtained by CFD simulation calculation, and the electron density values ​​are stored as vector data to obtain the axis electron density vector; the two-dimensional distribution of the electron density is stored as matrix data, and finally the vector data and the matrix data are organized into a training set;

[0056] Deep learning module: The deep learning model consists of two main parts: one is to construct a flow field reconstruction model and define the loss function of the flow field reconstruction model. The flow field reconstruction model is composed of the first fully connected layer, activation layer, deconvolution layer, flattening layer, second fully connected layer, third fully connected layer and reorganization layer. The loss function is defined by the continuity equation in the NS equation. The second is the training and supervision of the flow field reconstruction model. The flow field reconstruction model is trained under the supervision of the loss function.

[0057] Prediction module: Use the unknown two-dimensional distributed axis electron density vector data as the input vector of the flow field reconstruction model. The trained flow field reconstruction model reconstructs the input vector to obtain the two-dimensional electron density distribution of the target wake flow field where the vector data is located.

[0058] A device for reconstructing the electron density flow field of a super-target wake based on a physical information neural network, comprising:

[0059] Memory: used to store computer programs to implement the above-mentioned method for reconstructing the electron density flow field of the wake of a super-superior target based on a physical information neural network;

[0060] Processor: used to implement the above-mentioned method for reconstructing the electron density flow field of the wake of a super-target based on a physical information neural network when executing the computer program.

[0061] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network.

[0062] The application effects of the present invention are described in detail below in conjunction with the embodiments.

[0063] See also Figure 1 and Figure 2 , a hypersonic target wake electron density flow field reconstruction method based on physical information neural network, comprising the following steps:

[0064] Step 1: Use CFD simulation method to simulate the electron density of target wake at different Mach numbers. Figure 3This is a geometric diagram of the blunt cone model, with a bottom diameter of 1m and a semi-cone angle of 8°. After obtaining the target flow field, the electron density value on the central axis of the target wake flow field is obtained and stored as vector data. The two-dimensional distribution of the electron density of the flow field where the central axis is located is obtained and stored as a matrix; the data set covers the flow field electron density data from 11Ma to 18Ma, and the two-dimensional flow field and the corresponding central axis data are organized into a training set.

[0065] Step 2: The flow field reconstruction model is used to generate two-dimensional flow field data. The input layer receives vector data. The first fully connected layer will increase the data volume and adjust the data value through the activation layer. The two deconvolution layers perform feature reconstruction and increase the data dimension. The flattening layer is used to flatten the high-dimensional data output by the deconvolution layer into vector data. The data volume is standardized through the second and third fully connected layers. Finally, the data is reshaped through the reorganization layer and finally output to the desired size. The flow field reconstruction model is supervised and trained using the constructed loss function. The loss function feeds back the model output in the reverse propagation to facilitate the improvement of the model during the training process. The input data passes through the trained flow field reconstruction model, and the vector data can be reconstructed into two-dimensional flow field data through the expansion and reshaping functions of the model.

[0066] In addition, the loss function based on physical information is constructed in the near-wake flow field of the hypersonic target. The NS equation is usually used to describe the spatial distribution of various parameters in the flow field. The loss function is constructed through the continuity equation describing the physical quantity in the NS equation group:

[0067]

[0068] Among them, ρ m is the density in the flow field, t represents time, v m Represents a physical quantity of component m in the fluid.

[0069] In a steady fluid, removing the time gradient, based on the two-dimensional case, the above gradient equation is expanded to obtain a conservation expression:

[0070]

[0071] The above expression is conserved in the flow field region, so it can be used as a loss function to supervise the training of the flow field reconstruction model. When m represents an electron, v x is the x-direction component of the physical quantity v, v y is the component of the physical quantity v in the y direction. The above formula is expressed using the central difference with four surrounding points:

[0072] The expression in the wake domain is:

[0073]

[0074] The trail boundary expression is:

[0075]

[0076] Where h represents the difference step size, i and j represent the difference coordinates. The continuity equation in the NS equation is defined as the loss function of the flow field reconstruction model, so:

[0077]

[0078] The above loss function can capture the spatial characteristics of drastic changes in physical quantities, such as the parameter gradient at the shock wave contour, the post-stagnation point, etc. In addition, the traditional loss calculation method can ensure the numerical accuracy of the output. Therefore, the loss function is a combination of the above spatial gradient term and the traditional mean square error loss term:

[0079] The network will calculate the residual module of the entire smooth area and add the spatial physical quantity difference loss. The specific calculation method is:

[0080]

[0081] Among them, i represents the data sample number, and N represents the total number of data samples. x is the x-direction component of the physical quantity v, v y is the component of the physical quantity v in the y direction. The formula for the mean square error between the predicted value and the true value.

[0082] The loss value after evaluation is inserted into the flow field reconstruction model to supervise the flow field reconstruction results. For the non-uniform spatial distribution of wake plasma density, the physical information extraction flow field reconstruction model has physical theoretical advantages in training and predicting plasma density.

[0083] Step 3: Use the training set consisting of the 11-18 Mach hypersonic target wake flow field data in step 1 to train the flow field reconstruction model constructed in step 2, and use the loss function of step 2 to supervise the training until the loss function converges. The convergence is shown in Figure 4 , at this time, the flow field reconstruction model with trained weights is obtained;

[0084] Step 4, using the axis electron density vector of the unknown two-dimensional distribution in step 1 as the input vector, and reconstructing the input vector using the flow field reconstruction model trained in step 3 to obtain the reconstructed two-dimensional electron density distribution of the target wake flow field.

[0085] Use the trained flow field reconstruction model to obtain the electron density distribution of the target wake flow field, refer to Figure 5, the predicted flow fields at 12Ma, 14Ma, 16Ma and 18Ma are specifically predicted, and the true value of the flow field electron density distribution obtained by CFD simulation is given on the right. The electron density generated by the hypersonic target in the flow field at 4 different Mach numbers is different. As the Mach number increases, the compression and heating effects of the airflow are stronger, causing more air molecules to be ionized and the electron density to increase significantly. At lower Mach numbers, ionization is mainly concentrated near the shock layer. As the Mach number increases, the ionization area expands to a larger range, and the electron density inside the shock layer increases significantly.

[0086] from Figure 6 From the prediction results, the flow field reconstruction model proposed in the present invention has a better prediction effect on the flow field electron density on the straight line at y=4m, which has undergone a sudden change in electron density caused by the head shock wave. In addition, the straight line at y=0m passes through the post-stagnation point of the hypersonic target flow field, and the electron density spatial gradient in this part is relatively large, so the advantage of this model is more obvious here.

[0087] Figure 7 The Pearson correlation coefficient between the convolutional neural network (CNN network) and the present model for the reconstruction of the two-dimensional flow field at 11Ma, 13Ma, 15Ma and 17Ma is shown, namely Among them, cov represents covariance. Compared with the Pearson correlation coefficient, this model is still better than the convolutional neural network, which shows that this model can take into account both small-scale features and overall structure. Figure 7 The Pearson correlation shows that the present invention has better performance than the convolutional neural network at different Mach numbers, especially at locations with higher electron density, the flow field reconstruction model is closer to the true value.

[0088] In summary, compared with the existing technology:

[0089] In step 1 of this embodiment, the flow field is reconstructed using the electron density on the central axis of the hypersonic target wake flow field, which is consistent with the actual experimental setting and has strong practicality. The flow field data can be reconstructed using the central axis measurement data. Therefore, the reconstructed flow field data is highly correlated with the experimental data and has high accuracy.

[0090] In step 2, the NS equation is used to define the loss function of the flow field reconstruction model. Compared with the existing graph convolutional neural network and image super-resolution technology, it has the advantages of explainable physical mechanism and easier convergence with a small amount of training data.

[0091] In step 3, the flow field reconstruction model is trained using a training set consisting of hypersonic target wake flow field data of different Mach numbers, which has the advantage of being adaptable to different application backgrounds.

[0092] Step 4 can obtain the complete two-dimensional electron density distribution of the target wake flow field, which can provide reliable data support for the calculation of target electromagnetic scattering, stimulated electromagnetic radiation, and infrared and ultraviolet radiation characteristics.

[0093] The above method can obtain the complete two-dimensional electron density distribution of the target wake flow field, which solves the problem that the probe will affect the surrounding flow field in the experiment, resulting in a small number of measurement points designed on the wake axis. This method has the advantages of strong practicality, good reliability and high accuracy, and is widely used in many fields such as hypersonic aircraft experiments and design, plasma diagnosis, materials science research, quantum computing and simulation, etc.

[0094] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for reconstructing the electron density flow field of a hypersonic target wake based on a physical information neural network, characterized in that: The following steps are involved: Step 1, obtaining the electron density value on the central axis of the target wake flow field with different Mach numbers, storing the electron density value as vector data, and obtaining the axis electron density vector; obtaining the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located, and storing the two-dimensional distribution of the electron density as matrix data; arranging the vector data and the matrix data into a training set; Step 2: construct a flow field reconstruction model, wherein the flow field reconstruction model includes a first fully connected layer, an activation layer, a deconvolution layer, a flattening layer, a second fully connected layer, a third fully connected layer, and a reorganization layer; define the continuity equation in the NS equation as the loss function of the flow field reconstruction model; Step 3, using the flow field reconstruction model constructed in step 2 to train the training set in step 1 in the hypersonic target wake flow field at different Mach numbers, and using the loss function of step 2 to supervise the training until the loss function converges, thereby obtaining a flow field reconstruction model with trained weights; Step 4: Use the axis electron density vector data of the unknown two-dimensional distribution as the input vector, and use the flow field reconstruction model with the trained weights in step 3 to predict the input vector to obtain the two-dimensional electron density distribution of the target wake flow field.

2. The method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network according to claim 1 is characterized in that: In the step 1, the electron density value on the central axis and the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located are obtained by CFD simulation calculation.

3. The method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network according to claim 1 is characterized in that: The flow field reconstruction model of step 2 is used to generate two-dimensional flow field data. The input layer receives vector data, the first fully connected layer increases the data volume, and adjusts the data value through the activation layer; the two deconvolution layers perform feature reconstruction and increase the data dimension, and the high-dimensional data output by the deconvolution layer is flattened into vector data using the flattening layer. The data volume is standardized through the second fully connected layer and the third fully connected layer, and finally the data is reshaped through the reorganization layer, and finally the data is output to the desired size.

4. The method for reconstructing the electron density flow field of a hypersonic target wake based on a physical information neural network according to claim 1, characterized in that: The step 2 defines the continuity equation in the NS equation as the loss function of the flow field reconstruction model, specifically including: The continuity equation of NS electrons in a steady fluid is used for spatial gradient feature extraction and discretized using the central difference method. The discretized continuity equation expression is used as a conservation term, and the mean square error between the true value of the conservation term and the predicted value constitutes a loss function. The loss function is combined with the flow field reconstruction model so that the loss function is reversely propagated in the model to optimize the model weights.

5. The method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network according to claim 1 is characterized in that: The axis electron density vector data of the unknown two-dimensional distribution in step 4 is the electron number density measured by the probe on the central axis of the blunt cone in the ground test.

6. A hypersonic target wake electron density flow field reconstruction system based on physical information neural network, characterized by: include: Data acquisition module: The electron density values ​​on the central axis of the wake flow field of the target with different Mach numbers and the two-dimensional distribution of the electron density corresponding to the flow field where the central axis is located are obtained by CFD simulation calculation, and the electron density values ​​are stored as vector data to obtain the axis electron density vector; the two-dimensional distribution of the electron density is stored as matrix data, and finally the vector data and the matrix data are organized into a training set; Deep learning module: The deep learning model consists of two main parts: one is to construct a flow field reconstruction model and define the loss function of the flow field reconstruction model. The flow field reconstruction model is composed of the first fully connected layer, activation layer, deconvolution layer, flattening layer, second fully connected layer, third fully connected layer and reorganization layer. The loss function is defined by the continuity equation in the NS equation. The second is the training and supervision of the flow field reconstruction model. The flow field reconstruction model is trained under the supervision of the loss function. Prediction module: Use the unknown two-dimensional distributed axis electron density vector data as the input vector of the flow field reconstruction model. The trained flow field reconstruction model reconstructs the input vector to obtain the two-dimensional electron density distribution of the target wake flow field where the vector data is located.

7. A hypersonic target wake electron density flow field reconstruction device based on physical information neural network, characterized by: include: Memory: used to store a computer program to implement a method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network as described in any one of claims 1 to 5; Processor: used to implement a method for reconstructing the electron density flow field of a super-target wake based on a physical information neural network as described in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for reconstructing the electron density flow field of the wake of a super-superior target based on a physical information neural network are implemented.

Citation Information

Patent Citations

  • High-time-resolution flow field reconstruction method based on bidirectional gating circulation unit

    CN115114859A

  • Vortex-induced vibration calculation method and device for rotating cylinder

    CN116702649A

  • Method, system and equipment for super-resolution reconstruction by importing hypersonic flow field into convolutional neural network

    CN116843544A

  • Method and system for reconstructing airflow distribution field of toxic gas in transformer substation, and electronic equipment

    CN117787143A

  • Supersonic air inlet test flow field reconstruction method and device based on PINNs and readable medium

    CN117973036A

Cited By

  • Graph neural network ship flow field forecasting method based on high-order statistical aggregation

    CN120372815A

  • A ship flow field prediction method based on high-order statistical aggregation graph neural network

    CN120372815B

  • PINN-based sand mill dynamic separator flow field reconstruction method and system

    CN120850857A

  • Aircraft state prediction method and device based on quantum computing, equipment and medium

    CN121189519A

  • Flow field adaptive calculation method and system based on acoustic Doppler flow meter

    CN121456407A