Rocket transonic flight wall surface pressure pulsation rapid prediction method based on convolutional neural network

Abstract

The invention discloses a rocket transonic flight wall surface pressure pulsation rapid prediction method based on a convolutional neural network, belongs to the field of rocket transonic numerical simulation calculation, and constructs a rocket transonic flight wall surface pressure pulsation prediction model based on the convolutional neural network. The function of predicting unsteady wall surface pressure pulsation under the same working condition through rocket steady numerical simulation data is achieved. The construction process of the prediction model comprises the following steps: performing steady numerical simulation on each structured grid point of the flow field outside the rocket, and exporting steady flow field data with stable numerical values; performing unsteady simulation on each wall surface grid point and exporting wall surface pressure data of each time step after numerical value stabilization; processing the data to obtain a training data set; and training the neural network model according to the training data set. The method provided by the invention can quickly and accurately predict the wall surface pressure pulsation power-density spectrum during the transonic flight of the rocket.

Description

Technical Field

[0001] The present invention belongs to the field of rocket transonic numerical simulation calculation, and more specifically, relates to a method for quickly predicting rocket transonic flight wall pressure pulsation based on a convolutional neural network. Background Art

[0002] Large launch vehicles and their onboard instruments and equipment must experience a variety of dynamic environments during the rocket ignition, launch, and air operation phases. Among them, the aerodynamic noise generated by the launch vehicle during transonic flight will cause severe mechanical effects, posing a huge challenge to the safe flight of the rocket. Therefore, conducting research on rocket noise environment prediction, developing rapid prediction technology for rocket aerodynamic noise, timely evaluating the health status of rocket system equipment, and forming the optimal design plan for the rocket are of great significance for ensuring the safe flight of the rocket in harsh noise environments.

[0003] When a rocket is flying at transonic speed, there are complex unsteady flow phenomena in the flow field around it, such as boundary layer disturbance evolution, flow separation and shock waves. The flow characteristics are relatively complex and difficult to describe directly through formulas. At the same time, pressure pulsation is unsteady data, and it takes a lot of computing resources to obtain it. If other unsteady data are still used for prediction, it will inevitably be counterproductive. Summary of the invention

[0004] In response to the above defects or improvement needs of the prior art, the present invention provides a method for quickly predicting the wall pressure pulsation of a rocket during transonic flight based on a convolutional neural network, which can quickly and accurately predict the wall pressure pulsation power-density spectrum of a rocket during transonic flight.

[0005] To achieve the above object, according to a first aspect of the present invention, a method for establishing a rocket transonic flight wall pressure pulsation prediction model is provided, comprising:

[0006] S1, gridding the rocket external flow field to obtain a three-dimensional structured grid of the rocket external flow field, and using the first and second turbulence models to obtain the steady flow field simulation data of each grid point and the pressure pulsation time domain data of each wall grid point in a one-to-one correspondence;

[0007] S2, taking the rocket body rotation axis as a reference, select multiple sections from the rocket external flow field, obtain the steady flow field simulation data of each grid point in each section, the pressure pulsation time domain data of each wall grid point in each section, and perform FFT processing on the pressure pulsation time domain data of each wall grid point in each section to obtain the pressure pulsation frequency domain data; wherein the number of grid points in each section is N x *N y , N x is the number of wall grid points along the flow direction, N y is the number of grid points along the wall normal;

[0008] S3, taking the average value of the steady flow field simulation data of each grid point of the multiple cross sections as a sample and the average value of the pressure pulsation frequency domain data of each wall grid point of the multiple cross sections as a label, training the convolutional neural network, and using the trained convolutional neural network as a wall pressure pulsation prediction model for rocket transonic flight.

[0009] According to a second aspect of the present invention, a method for rapid prediction of rocket transonic flight wall pressure pulsation based on a convolutional neural network is provided, comprising:

[0010] The average value of the steady flow field simulation data of each grid point of multiple cross-sections of the rocket to be predicted is input into the rocket transonic flight wall pressure pulsation prediction model obtained by using the method described in the first aspect to obtain the pressure pulsation frequency domain data of each wall grid point of the rocket to be predicted.

[0011] According to a third aspect of the present invention, there is provided an electronic device, comprising: a computer-readable storage medium and a processor;

[0012] The computer-readable storage medium is used to store executable instructions;

[0013] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the method as described in the first aspect or the method as described in the second aspect.

[0014] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a processor to execute the method as described in the first aspect or the method as described in the second aspect.

[0015] According to a fifth aspect of the present invention, there is provided a computer program product, comprising a computer program or instructions, which, when executed by a processor, implements the method as described in the first aspect or the method as described in the second aspect.

[0016] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0017] 1. The method provided by the present invention uses a trained convolutional neural network and fast-obtained steady data to predict the wall pressure pulsation power-density spectrum during the transonic flight of a rocket, thereby being able to quickly and accurately predict the wall pressure pulsation power-density spectrum during the transonic flight of a rocket, especially the low-frequency part that occupies a large amount of energy, and both the pulsation peak frequency and amplitude can be predicted accurately.

[0018] 2. As a further preference, the method provided by the present invention divides the pressure pulsation time domain data of each wall grid point into K groups according to a preset time interval, performs FFT processing on each group of pressure pulsation time domain data to obtain corresponding pressure pulsation frequency domain data, and takes the average value of the K groups of pressure pulsation frequency domain data as the pressure pulsation frequency domain data of each wall grid point; this can avoid the pressure pulsation frequency domain data being too sharp, which is not conducive to the training of the convolutional neural network, thereby improving the prediction accuracy of the convolutional neural network.

[0019] 3. As a further preference, the method provided by the present invention adopts the first and second turbulence models to obtain the steady flow field simulation data of each grid point under different incoming flow Mach numbers and the pressure pulsation time domain data of each wall grid point in a one-to-one correspondence, which can increase the speed domain width of the neural network to predict the flow conditions, thereby increasing the scope of application of the convolutional neural network.

[0020] 4. As a further preference, the convolutional neural network adopted in the method provided by the present invention is a neural network architecture combining a convolutional layer, a self-attention layer and a fully connected layer, which can process and predict complex data. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of grid points and wall grid points provided in an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of the convolutional neural network structure provided by an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of NASA's standard rocket model;

[0024] Figure 4 A schematic diagram of a three-dimensional rocket transonic flight structured grid provided by an embodiment of the present invention, wherein: Figure 4 (a) and (b) are schematic diagrams of the overall mesh, the Z = 0 plane mesh, and the wall y + Change diagram along the x direction;

[0025] Figure 5 A graph showing the change of the rocket wall pressure coefficient in the x direction during 0.8 Ma flight provided by an embodiment of the present invention;

[0026] Figure 6 A diagram showing the change of the pressure pulsation coefficient along the x-direction during flight at 0.8 Ma provided by an embodiment of the present invention;

[0027] Figure 7 A schematic diagram of the azimuth angle of the arrow body provided by an embodiment of the present invention;

[0028] Figure 8 A schematic diagram of an interpolation cross section provided by an embodiment of the present invention;

[0029] Fig. 9 A comparison diagram of the wall pressure pulsation power density spectrum prediction and actual value provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] The embodiment of the present invention provides a method for establishing a rocket transonic flight wall pressure pulsation prediction model, comprising:

[0032] S1, the rocket external flow field is gridded to obtain the three-dimensional structured grid of the rocket external flow field, and the first and second turbulence models are used to obtain the steady flow field simulation data of each grid point and the pressure pulsation time domain data of each wall grid point one by one.

[0033] Specifically, by determining the inlet flow characteristics (including the free flow velocity U, the flow density ρ, and the flow dynamic viscosity coefficient μ), the height h of the first layer of grid close to the wall is calculated. Given the free flow velocity U, the rocket characteristic length L, the flow density ρ, and the flow dynamic viscosity coefficient μ, the specific calculation formula is: Reynolds number Wall resistance coefficient C f =[2log 10 (Re)-0.65] -2.3 , wall shear stress Friction coefficient near the wall First layer grid height At this time, let y + = 1. Draw the three-dimensional structured grid of the rocket's external flow field according to the first-layer grid height h, and set the pressure far-field boundary conditions.

[0034] Fluid mechanics simulation software is used to define corresponding turbulence models to obtain steady simulation data and unsteady simulation data. The first and second turbulence models may use the same turbulence model or different turbulence models, and the embodiment of the present invention does not limit this to uniqueness, for example:

[0035] The steady data are calculated using the SST-kω turbulence model, density basis, and second-order upwind discrete format. The drag coefficient of the rocket body is monitored during the calculation process. Once the drag coefficient is stable, the steady calculation result can be considered stable. The data file with stable numerical calculation is exported by node to obtain the steady data of each grid point.

[0036] The Les turbulence model is used to calculate the unsteady data (i.e., wall pressure data), and the time step is 1×10 -5 s. The drag coefficient of the rocket body is monitored during the calculation process. Once the drag coefficient is stable, the unsteady calculation result can be considered stable. After each time step is settled, the static pressure of the rocket body wall is exported according to the node, and a total of 5000 time step wall pressure data are exported.

[0037] S2, taking the rocket body rotation axis as a reference, select multiple sections from the rocket external flow field, obtain the steady flow field simulation data of each grid point in each section, the pressure pulsation time domain data of each wall grid point in each section, and perform FFT processing on the pressure pulsation time domain data of each wall grid point in each section to obtain the pressure pulsation frequency domain data; wherein the number of grid points in each section is N x *N y , N x is the number of wall grid points along the flow direction, N y is the number of grid points along the wall normal;

[0038] S3, taking the average value of the steady flow field simulation data of each grid point of the multiple cross sections as a sample and the average value of the pressure pulsation frequency domain data of each wall grid point of the multiple cross sections as a label, training the convolutional neural network, and using the trained convolutional neural network as a wall pressure pulsation prediction model for rocket transonic flight.

[0039] Specifically, in step S2, the steady data and the unsteady data are processed respectively.

[0040] Process the steady data, take the rocket body rotation axis (i.e., the central axis) as the reference, divide the rocket external flow field into multiple sections at equal angles, and extract the flow section data near the rocket body flow separation zone. For example, 8 sections are divided at equal angles to extract the flow section data near the rocket body flow separation zone, and each section is 45° apart. The flow field data near the wall on a certain section can be regarded as N x *N y * n matrix, where N x is the number of nodes along the axis, N y is the number of nodes along the normal direction, n is the original physical quantity of the flow field, such as pressure P, density ρ, velocity V x 、V y 、V z Etc. Figure 1 As shown in the figure, the data are recombined, sorted and the relevant partial derivatives are calculated, and the corresponding flow field data are derived one by one according to the order of the wall nodes. Considering that the selected physical quantities should conform to the NS equation, preferably, the derived flow field data contains 8 physical quantities. The specific flow field physical quantities (i.e., steady-state data) are shown in Table 1, where x is the direction along the stream, y is the normal direction of the wall, and ρ ∞is the incoming flow density, U ∞ is the incoming flow velocity, D is the characteristic diameter. The steady data under one working condition can be obtained by averaging the 8 sections. The data format exported after the steady data is dimensionless is N y *8 matrix, each exported data file corresponds to a wall node, a total of N x data files.

[0041] Table 1 Steady data table

[0042]

[0043] To process unsteady data, the wall pressure data (i.e., the pressure data of each wall grid point) is extracted with the same cross section in the same processing method as the steady data. The pressure information of the preset number of time steps is converted into frequency domain information using the fast Fourier transform and dimensionless calculation is performed. The dimensionless calculation formula is: in is the result after fast Fourier transform.

[0044] At the same time, in order to avoid the result being too sharp, which is not conducive to the subsequent neural network training, preferably, in step S2, the pressure pulsation time domain data of each wall grid point in each section is subjected to FFT processing to obtain the pressure pulsation frequency domain data, including:

[0045] The pressure pulsation time domain data of each wall grid point are divided into K groups according to the preset time interval, and each group of pressure pulsation time domain data is processed by FFT to obtain the corresponding pressure pulsation frequency domain data. The average value of the K groups of pressure pulsation frequency domain data is used as the pressure pulsation frequency domain data of each wall grid point.

[0046] Specifically, taking the preset number of time steps as 5000 time steps and K=7 as an example, the 5000 data are divided into 7 groups, each group contains 2000 time-continuous data, and each group is separated by 500 data, namely 1-2000, 501-2500...3001-5000. The 7 groups of data are fast Fourier transformed and averaged to obtain the final unsteady data. The derived unsteady data is a column vector composed of 1000 dimensionless pulsation amplitude data at a certain point, and the resolution of the data is 25HZ. Then, the dimensionless pulsation data on the 8 sections are averaged to obtain the unsteady data under one working condition. Corresponding to the steady data, the unsteady results are derived in total N x data files.

[0047] In order to increase the speed domain width of the neural network in predicting flow conditions, preferably, in step S1, the first and second turbulence models are respectively used to obtain the steady flow field simulation data of each grid point under different incoming flow Mach numbers and the pressure pulsation time domain data of each wall grid point in a one-to-one correspondence; in step S3, the average value of the steady flow field simulation data of each grid point of the multiple cross-sections under each incoming flow Mach number is used as a sample, and the average value of the pressure pulsation frequency domain data of each grid point of the multiple cross-sections under the corresponding incoming flow Mach number is used as a label to train the convolutional neural network.

[0048] In step S3, a neural network model is constructed, and the neural network model is trained according to the training data. Any convolutional neural network can be used, and the embodiment of the present invention does not limit this to uniqueness.

[0049] As a further preference, an embodiment of the present invention proposes a neural network architecture combining a convolutional layer, a self-attention layer and a fully connected layer, that is, the convolutional neural network includes an input layer, a convolutional layer, a self-attention layer, a fully connected layer and an output layer connected in sequence, for processing and predicting complex data. The following is a detailed description of the neural network architecture, including the input and output data structures of each layer and their implementation details, such as Figure 2 Shown is a schematic diagram of the structure of a neural network.

[0050] The input layer receives raw data. The shape of the input data is (N x *9,N y ,8), where N x is the number of wall nodes; N y is the number of normal space points; N x *9 represents the number of samples. The reason for multiplying by 9 is to increase the speed domain width of the neural network to predict the flow conditions. The flow data under multiple different Mach numbers are put together for neural network training. The 9 Mach numbers are 0.8, 0.85, 0.9...1.2Ma; 8 is the number of physical quantities. Therefore, each sample is of the shape (N y ,8) two-dimensional matrix data. The input layer inputs the original data into the subsequent convolution layer for feature extraction.

[0051] After the input layer is the convolution layer, which includes two convolution layers and two maximum pooling layers. The number of convolution kernels in the first convolution layer is 32, the convolution kernel size is 3, and the activation function is ReLU; the pooling window size of the first maximum pooling layer is 2; the number of convolution kernels in the second convolution layer is 64, and the convolution kernel size and activation function are the same as those in the first layer; the pooling window size of the second maximum pooling layer is the same as that in the first layer. A self-attention layer is added after the convolution layer to capture long-distance dependencies in the data and enhance the feature representation capability through the multi-head attention mechanism. The embedding dimension of this self-attention layer is 64, and the number of heads is 4.

[0052] After the self-attention layer, there is a flattening layer and three fully connected layers. The flattening layer is used to transform a tensor of shape (N,n,m) into a vector of shape (N,n*m). The number of neurons in the first fully connected layer is 128, and the activation function is ReLU; the number of neurons in the second fully connected layer is 512, and the activation function is the same as the first fully connected layer; the number of neurons in the third fully connected layer is 256, and the activation function is the same as the first fully connected layer.

[0053] The final output layer is used for direct output of the regression task, and the number of its neurons must match the data structure of the label data. From the above, we can see that the label data is a column vector containing 1000 data, so the number of neurons in the last layer is 1000.

[0054] It can be understood that the sample data at 9 Mach numbers correspond one by one to the label data at 9 Mach numbers.

[0055] The trained neural network can be used to predict the wall pressure pulsation of other rockets at transonic speeds that have not participated in the neural network training.

[0056] The embodiment of the present invention provides a method for quickly predicting rocket transonic flight wall pressure pulsation based on a convolutional neural network, comprising:

[0057] The average value of the steady flow field simulation data of each grid point of multiple cross-sections of the rocket to be predicted is input into the rocket transonic flight wall pressure pulsation prediction model obtained by using the rocket transonic flight wall pressure pulsation prediction model establishment method as described in the above embodiment, to obtain the pressure pulsation frequency domain data of each wall grid point of the rocket to be predicted.

[0058] The method provided by the present invention is further illustrated by taking a specific example as follows.

[0059] This example shows the specific steps of using this method to predict the wall pressure pulsation of the NASA standard rocket model in the transonic stage. Its geometric configuration is as follows Figure 3 shown.

[0060] Plot the 3D structured grid. The result is as follows: Figure 4 As shown in (a) and (b), the grid size is 3 million grids. At the same time, the calculation verifies that the y of the first layer of grid is perpendicular to the wall. + They are all less than 1, which meets the calculation requirements of the turbulence model.

[0061] To further illustrate and verify the accuracy of this numerical simulation method, the wall pressure coefficient, i.e., the root mean square value of the wall pressure pulsation, at an atmospheric height of h = 5 km and Ma = 0.8 is calculated on the numerical simulation software using the drawn grid, and compared with the experimental data, as shown in Figure 2. Figure 5 ,6 The experimental data comes from: DMSchuster…, "Investigation of unsteady Pressure Sensitive paint (uPSP) and a dynamic loads balance to predict launch vehicle buffet environments," Report No. NASA / TM-2028-219364 (NASA Langley Research Center, 2028).

[0062] like Figure 8 As shown, the area with the largest wall pressure pulsation when the rocket is flying at transonic speed is in the diameter contraction section of the rocket body behind the fairing. This part is the flow separation / reattachment area. In order to accurately predict the rocket wall pressure pulsation and reduce the workload to a certain extent, the data preprocessing will be concentrated in this area.

[0063] Use Fluent to export the calculation results of the constant numerical simulation, and export the CAS and DAT files to the TECPLOT post-processing software. In order to accurately export the flow field information on the sections with different azimuth angles, it is necessary to draw the grid of the corresponding section in the ICEM software, such as Figure 7 As shown, it is the azimuth based on the symmetry axis of the arrow body. The azimuth angles of the selected sections are 0°, 45°, 90°...315°, with a total of 8 sections.

[0064] Use the interpolation algorithm to interpolate the data of the entire flow field into 8 sections, such as Figure 7 The figure shows a schematic diagram of the interpolation section. The data on the section are exported, reordered, and calculated to obtain the eight physical quantities shown in Table 1. The data of the eight sections are averaged and exported in the order of the wall nodes to obtain all samples under one working condition. All the steady data of the nine working conditions required for training (different working conditions, different incoming flow Mach numbers) are calculated and exported in the same way to obtain all the data.

[0065] The processing of unsteady data is similar to that of steady data. Eight interpolation sections are used to obtain static pressure data on the wall of the rocket body at eight azimuth angles. In order to obtain the pressure pulsation power-density spectrum at each azimuth angle, the pressure information of 5000 time steps is converted into frequency domain information using fast Fourier transform and dimensionless calculation is performed. At the same time, in order to avoid the result being too sharp and not conducive to the subsequent neural network training, the 5000 data are divided into 7 groups, each containing 2000 time-continuous data, and each group is separated by 500 data, namely 1-2000, 501-2500...3001-5000. The 7 groups of data are fast Fourier transformed and averaged to obtain the final unsteady data. The derived unsteady data is a column vector composed of 1000 dimensionless pulsation amplitude data at a certain point, and the resolution of the data is 25HZ. The frequency spectrum of the eight azimuth angles is averaged to obtain all label data.

[0066] Use tensorflow to build the neural network code. The basic structure of the neural network has been explained and will not be repeated here. Figure 7 As shown in the figure, three samples that are not involved in the training are randomly selected, and the predicted values ​​and the actual values ​​are compared when the trained neural network is used for prediction. As can be seen from the figure, the wall pressure pulsation power-density spectrum of the rocket during transonic flight can be well predicted by using this method, especially the low-frequency part that occupies a large amount of energy, and both the pulsation peak frequency and amplitude can be predicted accurately.

[0067] An embodiment of the present invention provides an electronic device, including: a computer-readable storage medium and a processor;

[0068] The computer-readable storage medium is used to store executable instructions;

[0069] The processor is used to read the executable instructions stored in the computer-readable storage medium, and execute the method for establishing a rocket transonic flight wall pressure pulsation prediction model as described in any of the above embodiments, or the method for quickly predicting rocket transonic flight wall pressure pulsation based on a convolutional neural network as described in any of the above embodiments.

[0070] An embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to execute a method for establishing a rocket transonic flight wall pressure pulsation prediction model as described in any of the above embodiments, or a method for quickly predicting a rocket transonic flight wall pressure pulsation based on a convolutional neural network as described in any of the above embodiments.

[0071] An embodiment of the present invention provides a computer program product, including a computer program or instructions. When the computer program or instructions are executed by a processor, the method for establishing a rocket transonic flight wall pressure pulsation prediction model as described in any of the above embodiments, or the method for quickly predicting rocket transonic flight wall pressure pulsation based on a convolutional neural network as described in any of the above embodiments.

[0072] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for establishing a prediction model for wall pressure pulsation during rocket transonic flight, characterized in that: include: S1, gridding the rocket external flow field to obtain a three-dimensional structured grid of the rocket external flow field, and using the first and second turbulence models to obtain the steady flow field simulation data of each grid point and the pressure pulsation time domain data of each wall grid point in a one-to-one correspondence; S2, taking the rocket body rotation axis as a reference, select multiple sections from the rocket external flow field, obtain the steady flow field simulation data of each grid point in each section, the pressure pulsation time domain data of each wall grid point in each section, and perform FFT processing on the pressure pulsation time domain data of each wall grid point in each section to obtain the pressure pulsation frequency domain data; wherein the number of grid points in each section is N x *N y , N x is the number of wall grid points along the flow direction, N y is the number of grid points along the wall normal; S3, taking the average value of the steady flow field simulation data of each grid point of the multiple cross sections as a sample and the average value of the pressure pulsation frequency domain data of each wall grid point of the multiple cross sections as a label, training the convolutional neural network, and using the trained convolutional neural network as a wall pressure pulsation prediction model for rocket transonic flight.

2. The method according to claim 1, characterized in that The steady flow field simulation data includes ρU, ρV, P. Among them, U is the velocity of the free flow, ρ is the flow density, P is the pressure, x is the flow direction, and y is the wall normal direction.

3. The method according to claim 1 or 2, characterized in that In step S1, the first and second turbulence models are respectively used to obtain the steady flow field simulation data of each grid point under different incoming flow Mach numbers and the pressure pulsation time domain data of each wall grid point in a one-to-one correspondence; in step S3, the average value of the steady flow field simulation data of each grid point of the multiple cross sections under each incoming flow Mach number is used as a sample, and the average value of the pressure pulsation frequency domain data of each grid point of the multiple cross sections under the corresponding incoming flow Mach number is used as a label to train the convolutional neural network.

4. The method according to claim 1, characterized in that Before step S2, the method further includes performing dimensionless processing on the steady flow field simulation data of each grid point and the pressure pulsation time domain data of each wall grid point.

5. The method according to claim 1, characterized in that In step S2, the pressure pulsation time domain data of each wall grid point in each cross section is subjected to FFT processing to obtain the pressure pulsation frequency domain data, including: The pressure pulsation time domain data of each wall grid point are divided into K groups according to preset intervals, and each group of pressure pulsation time domain data is processed by FFT to obtain the corresponding pressure pulsation frequency domain data. The average value of the K groups of pressure pulsation frequency domain data is used as the pressure pulsation frequency domain data of each wall grid point.

6. The method according to claim 1, characterized in that The convolutional neural network includes an output layer, a convolution layer, a self-attention layer, a fully connected layer and an output layer which are connected in sequence.

7. A method for rapid prediction of rocket transonic flight wall pressure pulsation based on convolutional neural network, characterized in that: include: The average value of the steady flow field simulation data of each grid point of multiple cross-sections of the rocket to be predicted is input into the rocket transonic flight wall pressure pulsation prediction model obtained by the method described in any one of claims 1 to 6, and the pressure pulsation frequency domain data of each wall grid point of the rocket to be predicted is obtained.

8. An electronic device, characterized in that: include: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium, and execute the method according to any one of claims 1 to 6, or the method according to claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the method according to any one of claims 1 to 6, or the method according to claim 7.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 6 or the method according to claim 7 is implemented.

Images