A deep learning-based method for predicting wall pressure fluctuations in turbulent boundary layers

Abstract

The present invention discloses a method for predicting turbulent boundary layer wall pressure pulsation based on deep learning, which belongs to the field of noise prediction. The method trains a neural network based on unsteady and steady flow simulation data of wall grid points to obtain a turbulent boundary layer wall pressure pulsation prediction model, and obtains a large number of samples in one simulation calculation, which can significantly reduce the cost of sample acquisition, save a lot of time, and thus ensure the prediction accuracy and timeliness of the deep learning model; the nonlinear relationship between turbulent boundary layer parameters and wall pressure pulsation spectrum established by the deep learning method does not require clear mathematical formulas, and at the same time reduces the difficulty of mathematical requirements; compared with traditional prediction methods, the method provided by the present invention has stronger applicability, wider application fields, and stronger generalization ability; in addition, simulation verification shows that the method proposed by the present invention has the advantages of accurate results, small errors, and efficient calculation.

Description

Technical Field

[0001] The present invention belongs to the field of noise prediction, and more specifically, relates to a method for predicting turbulent boundary layer wall pressure pulsation based on deep learning. Background Art

[0002] The noise generated by aircraft, underwater vehicles and rotating machinery is mainly caused by the excitation and radiation of the pressure pulsation on the solid boundary wall and the disturbance in the surrounding turbulent boundary layer. Excessive wall pressure pulsation not only produces large noise, but also causes structural fatigue and component failure. Considering that the wall pressure pulsation of the turbulent boundary layer is the main source of noise, it is very necessary to develop a fast and accurate method for predicting the wall pressure pulsation of the turbulent boundary layer. At present, the methods for predicting wall pressure pulsation mainly include experimental measurement methods, numerical simulation methods and semi-empirical methods. The advantages and disadvantages of these three methods are as follows:

[0003] (1) Experimental measurement: This method mainly measures the transient wall surface pulsating pressure field signal by placing a large number of sensors on the wall. Its advantages are that although the experimental results are relatively accurate, the experimental conditions are harsh, the cycle is long, and the cost is high, which makes it difficult to be widely applied in different practical projects.

[0004] (2) Numerical simulation: This method mainly uses high-precision numerical simulation methods to simulate the unsteady flow in the flow field, and then directly obtains the wall pressure pulsation based on this. Its advantages are: compared with experimental testing methods, numerical simulation methods are less expensive, but obtaining very accurate unsteady constants usually requires a lot of computing resources. In addition, the time period is relatively long.

[0005] (3) Semi-empirical models: Several semi-empirical models are established to describe the nonlinear relationship between turbulent boundary layer parameters and the wall pressure pulsation spectrum. The wall pressure pulsation spectrum can be obtained by obtaining the turbulent boundary layer parameters through experiments or numerical simulations. Semi-empirical models can quickly obtain the wall pressure pulsation spectrum through boundary layer parameters, but each semi-empirical model is based on very limited experimental tests or numerical simulations, resulting in a very limited application area. Summary of the Invention

[0006] In response to the above-mentioned defects or improvement needs of the prior art, the present invention provides a deep learning-based method for predicting the wall pressure pulsation of the turbulent boundary layer, which can quickly, efficiently and accurately predict the wall pressure pulsation spectrum of the turbulent boundary layer of a solid structure.

[0007] To achieve the above objectives, according to a first aspect of the present invention, a method for constructing a turbulent boundary layer wall pressure pulsation prediction model based on deep learning is provided, comprising:

[0008] S1, establishing a geometric simulation model of the external flow field of the sample solid structure, gridding the flow domain therein, and obtaining the flow field simulation data of each grid point;

[0009] S2, calculating the boundary layer thickness, boundary layer displacement thickness, and boundary layer momentum thickness of each grid point according to the tangential flow velocity in the normal direction of each grid point in the flow field simulation data, and using these together with the boundary layer edge velocity, wall friction velocity, and pressure gradient in the flow field simulation data as turbulent boundary layer parameters; converting the pressure pulsation time domain signal of each grid point in the flow field simulation data into a frequency domain signal;

[0010] S3, using the turbulent boundary layer parameters of each grid point as samples and the pressure pulsation frequency domain signals of each grid point as labels, to train the deep learning model, and use the trained model as a turbulent boundary layer wall pressure pulsation prediction model.

[0011] According to a second aspect of the present invention, a method for predicting turbulent boundary layer wall pressure pulsation based on deep learning is provided, comprising:

[0012] The turbulent boundary layer parameters of the solid structure to be predicted are input into the turbulent boundary layer wall pressure pulsation prediction model constructed by the construction method described in the first aspect to obtain the pressure pulsation frequency domain signal of the solid structure to be predicted.

[0013] According to a third aspect of the present invention, there is provided an electronic device comprising: a computer-readable storage medium and a processor;

[0014] The computer-readable storage medium is used to store executable instructions;

[0015] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the construction method described in the first aspect or the prediction method described in the second aspect.

[0016] 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, and the computer instructions are used to enable a processor to execute the construction method as described in the first aspect or the prediction method as described in the second aspect.

[0017] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0018] 1. Traditional machine learning methods require a large number of unsteady numerical simulation results as sample data for prediction input. Compared with traditional machine learning methods, the method of the present invention is based on the unsteady and steady flow simulation data of wall grid points to train neural networks, which can obtain a large number of samples in one simulation calculation, significantly reducing the cost of sample acquisition and saving a lot of time, thereby ensuring the prediction accuracy and timeliness of the deep learning model.

[0019] 2. The method provided by the present invention establishes a nonlinear relationship between turbulent boundary layer parameters and wall pressure pulsation spectrum through a deep learning method. It does not require clear mathematical formulas and reduces the difficulty of mathematical requirements. The user only needs to know the physical quantities of input and output without understanding the intermediate process.

[0020] 3. Compared with the traditional semi-empirical model, the deep learning-based turbulent wall pressure pulsation prediction method proposed in the present invention is more applicable (it only needs to establish the relationship between the boundary layer parameters and the frequency domain signal, so it is applicable to objects with different geometric structures, such as forward pressure / zero pressure / adverse pressure gradient structures), has a wider application field (can be applied to all areas of land, sea and air), and has stronger generalization ability (using limited data for training, more data can be predicted).

[0021] 4. Simulation verification shows that the deep learning-based turbulent boundary layer wall pressure pulsation prediction method proposed in this invention has the advantages of accurate results, small errors, and efficient calculation compared with traditional prediction methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of the outer flow field grid of an airfoil provided in an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of the grid at the boundary layer of an airfoil provided in an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of the wall pressure pulsation spectrum at a certain point on the airfoil surface provided by an embodiment of the present invention;

[0025] Figure 4 Schematic diagram of the loss function of the training set and validation set provided by the embodiment of the present invention;

[0026] Figure 5 A schematic diagram showing a comparison of the wall pressure pulsation spectrum at a certain point on the airfoil surface provided by an embodiment of the present invention;

[0027] Figure 6 Schematic diagram of the streamline direction 1MSE of the airfoil surface provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining 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 may be combined with each other as long as they do not conflict with each other.

[0029] Existing wall pressure pulsation prediction methods all have shortcomings. Considering that machine learning methods have also been gradually applied to the field of fluid mechanics, they are mainly used to predict steady-state numerical simulation results (i.e., the predicted results do not change with time) and periodic unsteady numerical simulation results (i.e., the predicted results change periodically with time). There are not many applications for predicting random disturbance components in physical quantities. Based on this, in order to address the shortcomings of existing methods, the present invention proposes a method for constructing a turbulent boundary layer wall pressure pulsation prediction model based on deep learning, including:

[0030] S1. Establish an external flow field geometric model of the sample solid structure in the geometric modeling software, grid the flow domain in the above model, and obtain the flow field simulation data of each grid point (hereinafter referred to as flow field data) through fluid mechanics simulation software.

[0031] It can be understood that the solid structure can be any solid structure such as an aircraft, an underwater vehicle, a rotating mechanical blade, etc. When it is in the flow field, it will generate turbulent boundary layer wall pressure due to the interaction with the fluid; the fluid is air fluid or fluid of any other medium.

[0032] Specifically, the geometric structure and external flow field model of the sample solid structure are created in geometric modeling software (such as Catia, Spaceclaim, etc.); then, the flow domain in the model is meshed using meshing software such as ICEM and Pointwise to obtain a high-quality mesh; finally, the mesh data is imported into fluid mechanics simulation software (such as AnsysFluent, STAR-CCM, etc.), and the corresponding turbulence model, flow medium and other related parameters and iterative time step are defined according to the usage scenario of the solid structure to obtain the flow field data of each grid point under various working conditions.

[0033] S2, calculating the boundary layer thickness, boundary layer displacement thickness, and boundary layer momentum thickness of each grid point according to the tangential flow velocity in the normal direction of each grid point in the flow field data, and using them together with the boundary layer edge velocity, wall friction velocity, and pressure gradient in the flow field simulation data as turbulent boundary layer parameters; converting the pressure pulsation time domain signal of each grid point in the flow field data into a frequency domain signal.

[0034] Specifically, according to the normal direction tangential flow velocity U of each grid point in the flow field data y , and the definition of the boundary layer thickness δ: from the wall to the height perpendicular to the wall where the tangential flow velocity on the wall reaches 0.99 times the free flow velocity, obtain the boundary layer thickness δ of each grid point, and compare it with the boundary layer displacement thickness δ according to δ. * , the relationship between the momentum thickness of the boundary layer θ, and the calculation of δ at each grid point * and θ.

[0035] Boundary layer displacement thickness δ * The calculation formulas for and momentum thickness θ are:

[0036]

[0037] Among them, U e is the velocity at the edge of the boundary layer, which is the free stream velocity multiplied by 99%, U y is the tangential flow velocity in the normal direction starting from the wall.

[0038] The boundary layer edge velocity, wall friction velocity and pressure gradient are all directly read from the flow field simulation data.

[0039] The pressure pulsation time domain signal of each grid point in the flow field data is read, the time-averaged value and the root mean square value of the pressure are calculated, and then the time domain signal of the pressure pulsation is converted into a frequency domain signal using Fourier transform (FFT).

[0040] To improve computational efficiency, the relevant calculations in step S2 can be performed using data processing software such as MATLAB and PYTHON.

[0041] S3, using the turbulent boundary layer parameters of each grid point as samples and the pressure pulsation frequency domain signals of each grid point as labels, to train the deep learning model to obtain a trained turbulent boundary layer wall pressure pulsation prediction model.

[0042] Specifically, the turbulent boundary layer parameters of each grid point are used as input and the pressure pulsation frequency domain signal of the corresponding grid point is used as output to train the deep learning model, and the trained deep learning model is used as the turbulent boundary layer wall pressure pulsation prediction model, thus completing the construction of the turbulent boundary layer wall pressure pulsation prediction model.

[0043] It is understandable that the above-mentioned deep learning model can adopt MLP, RNN, LSTM, etc., and the embodiment of the present invention does not make any unique limitation to this.

[0044] Furthermore, in order to improve the convergence speed of the model, preferably, before step S3, the method further includes:

[0045] The turbulent boundary layer parameters and pressure pulsation frequency domain signals are dimensionally non-quantized.

[0046] Conventional dimensionless processing methods may be used to perform dimensionless processing on the turbulent boundary layer parameters and the pressure pulsation frequency domain signal, and the embodiments of the present invention do not impose a unique limitation on this.

[0047] Considering the characteristics of the pressure pulsation frequency domain signal, in order to transform the complex physical problem into a simple mathematical problem, preferably, the following formula is used to perform dimensionless processing on the pressure pulsation frequency domain signal:

[0048]

[0049] Where Φ(ω) represents the wall pressure pulsation spectrum, Φ(ω) * represents the dimensionless wall pressure pulsation spectrum, U ∞ represents the velocity of the free flow, q∞ represents the dynamic pressure of the free flow, and L represents the characteristic length of the solid structure. For different solid structures, the characteristic length is selected in different ways. If the solid structure is a body of revolution, its characteristic length is the diameter or radius. If the solid structure is an airfoil, its characteristic length is the chord length. If the solid structure is an aircraft or an underwater vehicle, its characteristic length is usually the streamwise length, which is similar to the definition of the Reynolds number characteristic length.

[0050] Furthermore, considering the large difference in the magnitude of the output data, in order to speed up the gradient descent to find the optimal solution and improve the accuracy of the model, the output data is logarithmized, and then the input and output data are normalized (MIN-MAX) to map the data to the range of [0,1]. That is, preferably, before step S3, the following is also included:

[0051] The pressure pulsation frequency domain signal after dimensionless processing is logarithmized, and the pressure pulsation frequency domain signal after logarithmization processing and the turbulent boundary layer parameters after dimensionless processing are subjected to deviation normalization processing.

[0052] An embodiment of the present invention provides a method for predicting turbulent boundary layer wall pressure pulsation based on deep learning, comprising:

[0053] The turbulent boundary layer parameters of the solid structure to be predicted are input into the turbulent boundary layer wall pressure pulsation prediction model constructed by the construction method described in any of the above embodiments to obtain the pressure pulsation frequency domain signal of the solid structure to be predicted.

[0054] Specifically, an airfoil and external flow field geometric model of the solid structure to be predicted is established in the geometric modeling software, the airfoil is gridded, the flow field data of each grid point is obtained through the fluid mechanics simulation software, the turbulent boundary layer parameters of each grid point are calculated according to the tangential flow velocity in the normal direction of each grid point in the flow field data, and the pressure pulsation time domain signal of each grid point in the flow field data is converted into a frequency domain signal, which is input into the turbulent boundary layer wall pressure pulsation prediction model to obtain the pressure pulsation frequency domain signal of the solid structure to be predicted.

[0055] It can be understood that the application fields of the above-mentioned sample solid structures and the solid structures to be predicted should be the same. For example, if it is necessary to predict the wall pressure pulsation of the turbulent boundary layer of an aircraft, the training data set of the corresponding prediction model should also be obtained based on the sample aircraft; if it is necessary to predict the wall pressure pulsation of the turbulent boundary layer of an underwater vehicle, the training data set of the corresponding prediction model should also be obtained based on the sample underwater vehicle; if it is necessary to predict the wall pressure pulsation of the turbulent boundary layer of a rotating machinery blade, the training data set of the corresponding prediction model should also be obtained based on the sample rotating machinery blade.

[0056] The method provided by the present invention is further described below by taking a solid structure as a rotating machine blade as an example.

[0057] (1) Create a geometric model and perform flow field calculations.

[0058] Use geometric modeling software to create a two-dimensional airfoil and external flow field geometry model of the blade. In this example, the airfoil chord length c is 1m, the external flow field inlet is 10c from the airfoil leading edge, the outlet is 15c from the airfoil trailing edge, the upper and lower lengths are 20c, and the airfoil is symmetrical. Use ICEM meshing software to mesh the two-dimensional airfoil and obtain a high-quality mesh, such as Figure 1 and Figure 2 shown.

[0059] The grid data is exported as an .msh file, and the obtained grid data .msh file is imported into the Ansys Fluent numerical simulation software, and the relevant models and parameters are defined according to the actual application scenario of the blade: In this example, the turbulence model is defined as SST k-ω (not limited to this model, considering the computational cost, this model is preferred); the flow medium is defined as air (it can also be other fluids, defined according to the actual application scenario of the blade); the airfoil wall is defined as a no-slip boundary condition; the inlet condition is defined as a velocity inlet, and the speed is set to 17m / s, 34m / s and 51m / s, that is, the Mach number is 0.05, 0.1, and 0.15, the incoming flow angle of attack is 0° to 8°, the static pressure is set to 1 standard atmospheric pressure, that is, 101325Pa, and the outlet condition is 0° to 1°. The pressure outlet was defined, the static pressure was set to 101325 Pa, and second-order equations were used for both the pressure and momentum solvers. The iterative time step was defined as 2e-4s. After the steady-state numerical calculations (e.g., near-wall tangential velocity and wall shear stress) converged, the unsteady numerical calculations (e.g., wall static pressure) began. After 8000 time steps of unsteady numerical calculations, flow field data (including friction velocity, pressure gradient, wall shear stress, etc.) were extracted based on the steady-state and unsteady numerical calculation results. A total of 3 × 9 = 27 operating conditions were calculated, with 260 mesh nodes on the airfoil surface, resulting in a total of 7020 sample data.

[0060] (2) The flow field data are post-processed to obtain the turbulent boundary layer parameters and pressure pulsation frequency domain signals of each grid point.

[0061] First, the tangential velocity U in the normal direction of the wall grid node in (1) is y Output in .out format, compile the program in MATLAB according to the definition of boundary layer thickness δ, read and calculate the boundary layer thickness δ and boundary layer displacement thickness δ at each wall grid node * and momentum thickness θ, where the boundary layer thickness δ is defined as the height perpendicular to the wall from the wall to the point where the tangential velocity at the wall reaches 0.99 times the free stream velocity. Then, according to the boundary layer displacement thickness δ * The calculation formula of the momentum thickness θ is used to compile the program to calculate the boundary layer displacement thickness and momentum thickness at each wall grid node, and the boundary layer edge velocity, wall friction velocity and pressure gradient are directly read from the flow field simulation data. The specific turbulent boundary layer parameters are shown in Table 1, where u τ is the wall friction velocity, is the pressure gradient.

[0062] Table 1 Turbulent boundary layer parameters

[0063]

[0064] At the same time, the pressure pulsation time domain signal at the wall grid node is read in MATLAB, the program is compiled to calculate the time average value and root mean square value of the pressure, and then the Fourier transform (FFT) is used to convert the pressure pulsation time domain signal into a frequency domain signal. Figure 3 At this point, the input and output data of the deep learning model are generated. The input data is the turbulent boundary layer parameters, and the output data is the frequency domain signal of the pressure pulsation.

[0065] (3) Use input and output data to train the deep learning agent model.

[0066] First, the input and output data are dimensionless. Since the output data has a large difference in magnitude, the output data is logarithmized. Then, both the input and output data are normalized (MIN-MAX) to map the data to the range of [0, 1]. The purpose is to speed up the gradient descent to find the optimal solution while improving the accuracy.

[0067] Then, the input and output data sets are randomly sorted and divided into 8:2, that is, 80% of the data is selected as the training set and 20% of the data is selected as the validation set; the deep learning model is trained using the above data set to obtain the trained model and training weights, which includes: creating a Tensorflow environment framework on the server and loading the corresponding algorithm function library; setting hyperparameters, in this case, the number of neural network layers n L Set to 6 layers, the number of neurons in each layer is n N Set to 256 and decrease in two-fold order, the initial learning rate L r The optimization is set to 0.0005, the optimizer is set to Adam, the batch size of the one-time input model is set to 64, and the number of epochs is set to 10,000 steps. The mean square error (MSE) between the target and the neural network output is selected as the loss function to evaluate the prediction performance of the deep learning model. The L2 regularization term is further added to prevent overfitting problems. The data is read and input into the model to start training. The loss functions of the training set and the validation set during the training process are as follows: Figure 4 As shown; save the trained model and training weights.

[0068] Finally, the generalization ability of the deep learning proxy model was evaluated using test data. Data such as boundary layer thickness, friction velocity, and pressure gradient at the wall mesh nodes, outside the training and validation set conditions, were input into the deep learning model. The output data was compared with existing CFD (Computational Fluid Dynamics) numerical simulation data. The error between the wall pressure pulsation spectrum obtained by the deep learning proxy model and the CFD numerical simulation was expressed using lMSE, which is defined as:

[0069]

[0070] Among them, ω1 and ω2 represent the minimum and maximum frequencies in the spectrum, Φ(ω)1 * and Φ(ω)2 * The wall pressure pulsation spectra output by the CFD numerical simulation and the deep learning proxy model are dimensionless. The error between the wall pressure pulsation spectra at multiple mesh nodes on the airfoil surface is calculated to evaluate the generalization ability of the deep learning proxy model.

[0071] The boundary layer parameters with Mach number of 0.2 and angle of attack of 0° in the test set are input into the deep learning model, and the output data is compared with the CFD numerical simulation data. The comparison result of a certain node is as follows: Figure 5 As shown in Figure 2, the deep learning proxy model provided by the present invention can well fit the wall pressure pulsation spectrum. lMSE is used to represent the error between the deep learning proxy model and the wall pressure pulsation spectrum obtained by CFD numerical simulation. The error between the wall pressure pulsation spectrum at multiple grid nodes on the airfoil surface is calculated respectively, where the lMSE comparison at Mach number 0.2 and 0° angle of attack is as follows: Figure 6 As shown in the figure, the lMSE of all nodes is below 2.5dB. The location with larger errors on the left is the laminar flow region at the leading edge of the airfoil, and the location with larger errors on the right is the vortex shedding region at the trailing edge of the airfoil.

[0072] Comparing the prediction time of the deep learning agent model with the computational time required for traditional CFD unsteady simulation, it is found that the deep learning agent model proposed in this invention has a very obvious advantage, and the prediction time is much shorter than the CFD unsteady simulation time, as shown in Table 2.

[0073] Table 2 Comparison of prediction time

[0074]

[0075] It can be seen that compared with existing methods, the deep learning-based turbulent boundary layer wall pressure pulsation prediction method provided by the present invention has the advantages of being fast and accurate.

[0076] An embodiment of the present invention provides an electronic device, comprising: a computer-readable storage medium and a processor;

[0077] The computer-readable storage medium is used to store executable instructions;

[0078] The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the construction method or prediction method described in any one of the above embodiments.

[0079] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the construction method or prediction method described in any of the above embodiments.

[0080] It will be easily understood by those skilled in the art that the above description is merely 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 scope of protection of the present invention.

Claims

1. A method for constructing a turbulent boundary layer wall pressure pulsation prediction model based on deep learning, characterized in that: include: S1, establishing a geometric simulation model of the external flow field of the sample solid structure, gridding the flow domain therein, and obtaining the flow field simulation data of each grid point; S2, calculating the boundary layer thickness, boundary layer displacement thickness, and boundary layer momentum thickness of each grid point according to the tangential flow velocity in the normal direction of each grid point in the flow field simulation data, and using these together with the boundary layer edge velocity, wall friction velocity, and pressure gradient in the flow field simulation data as turbulent boundary layer parameters; converting the pressure pulsation time domain signal of each grid point in the flow field simulation data into a frequency domain signal; S3, using the turbulent boundary layer parameters of each grid point as samples and the pressure pulsation frequency domain signals of each grid point as labels, to train the deep learning model, and use the trained model as a turbulent boundary layer wall pressure pulsation prediction model.

2. The method according to claim 1, wherein Before step S3, the method further includes: The turbulent boundary layer parameters and pressure pulsation frequency domain signals are dimensionally non-quantized.

3. The method according to claim 2, wherein The following formula is used to perform dimensionless processing on the pressure pulsation frequency domain signal: Where Φ(ω) represents the wall pressure pulsation spectrum, Φ(ω) * represents the dimensionless wall pressure pulsation spectrum, U ∞ represents the free stream velocity, q ∞ represents the dynamic pressure of the free flow, and L represents the characteristic length of the solid structure.

4. The method according to claim 2, wherein Before step S3, the method further includes: The pressure pulsation frequency domain signal after dimensionless processing is logarithmized, and the pressure pulsation frequency domain signal after logarithmization processing and the turbulent boundary layer parameters after dimensionless processing are subjected to deviation normalization processing.

5. A method for predicting turbulent boundary layer wall pressure fluctuations based on deep learning, characterized in that: include: The turbulent boundary layer parameters of the solid structure to be predicted are input into the turbulent boundary layer wall pressure pulsation prediction model constructed by the construction method according to any one of claims 1 to 4 to obtain the pressure pulsation frequency domain signal of the solid structure to be predicted.

6. An electronic device, characterized in that: include: Computer-readable storage medium and processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the construction method according to any one of claims 1 to 4 or the prediction method according to claim 5.

7. 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 construction method according to any one of claims 1 to 4 or the prediction method according to claim 5.

Images