Neural network model frequency-band-division modeling method for predicting automobile front side window flow field excitation

Abstract

The invention discloses a neural network model frequency-band modeling method for predicting flow field excitation of a front side window of an automobile. The method comprises the following steps: 1) obtaining frequency domain flow field excitation data of the surface of the front side window of the automobile; 2) dividing a prediction frequency band according to a contribution analysis result; 3) screening out parameters with top contribution in each frequency band as key geometric parameters of the frequency band, and performing segmentation processing on the flow field excitation data according to frequency bands; 4) taking the key geometric parameters in each frequency band as the input of a neural network model, taking the excitation data of the corresponding frequency band as the output of the neural network model, and training the neural network model; and 5) combining prediction results of the neural network model of each frequency band, and establishing a neural network model for rapidly predicting front side window flow field excitation in the whole prediction frequency band. According to the method, the frequency bands are divided based on the contribution analysis result, and accurate prediction of front side window excitation is realized through a mode of training the network model by the frequency bands.

Description

Technical Field

[0001] The invention relates to the fields of aerodynamics and machine learning, and in particular to a frequency-band modeling method of a neural network model for predicting flow field excitation of a front side window of an automobile. Background Art

[0002] In the field of car interior wind noise research, computational fluid dynamics (CFD) is currently the most commonly used research method. However, the CFD method has problems such as long calculation time, high consumption of computing resources, and complex model mesh processing. Often a car has multiple sets of shape optimization schemes, and the CFD simulation of each set of schemes takes several days. This seriously hinders the progress of the work in the stages of car shape optimization design and interior wind noise optimization and noise reduction.

[0003] The unique black box properties of neural networks enable them to accurately establish the connection between the input parameters and output values ​​of nonlinear systems, thereby realizing the ability to predict flow field excitation through key geometric parameters. This seems to be able to serve as an efficient prediction method to replace CFD simulation. However, there are many geometric factors that affect the flow field state, which means that the neural network model may select too many input parameters, which in turn leads to an increase in the demand for the number of training samples. In addition, since the degree of influence of different geometric parameters on flow field excitation varies at different frequencies, considering all input parameters will repeatedly consider parameters with low correlation at each frequency. These low-correlation parameters will interfere with the model's identification of the main parameters, increase the degree of nonlinearity of the system, and affect the model training effect. Summary of the invention

[0004] The object of the present invention is to provide a neural network model frequency band modeling method for predicting the flow field excitation of the front side window of an automobile, comprising the following steps:

[0005] 1) Use computational fluid dynamics simulation software to calculate the time domain flow field data of the front side window surface of the car;

[0006] 2) The time domain flow field data is processed using the wave number frequency spectrum decomposition technology to obtain the frequency domain flow field excitation data of the front side window surface of the vehicle;

[0007] 3) Measure the vehicle geometric parameter data and map it with the frequency domain flow field excitation data of the vehicle front side window surface;

[0008] 4) The contribution analysis method is used to calculate the contribution of the vehicle geometric parameters to each flow field excitation, and the predicted frequency segment is divided according to the contribution size to obtain multiple sub-frequency segments;

[0009] 5) The vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% are taken as key geometric parameters;

[0010] According to the sub-frequency segments, the frequency domain flow field excitation data is segmented to obtain the sub-frequency segment frequency domain flow field excitation data;

[0011] Construct a sub-frequency band sample data set with key geometric parameter data as input and sub-frequency band frequency domain flow field excitation data as output;

[0012] 6) Use machine learning technology to build a neural network model;

[0013] 7) Using the sub-frequency band sample data set to train the neural network model, and obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band;

[0014] 8) Combining the front side window flow field excitation prediction model corresponding to each sub-frequency band, thereby constructing the front side window flow field excitation sub-frequency band prediction model;

[0015] 9) Measure the geometric parameter data of the vehicle to be tested and input them into the front side window flow field excitation frequency band prediction model to obtain the vehicle's front side window flow field excitation prediction value.

[0016] Further, in step 1), the step of calculating the time domain flow field data of the front side window surface of the automobile includes:

[0017] 1.1) A virtual numerical wind tunnel is established in the fluid dynamics simulation software. The virtual numerical wind tunnel is a cuboid, in which two walls serve as the velocity inlet wall and the pressure outlet wall;

[0018] 1.2) Build a closed model of the vehicle to be tested;

[0019] 1.3) Place the vehicle model to be tested in the virtual numerical wind tunnel and close to the velocity inlet wall;

[0020] 1.4) Set a dense grid area in the front side window area of ​​the vehicle model to be tested;

[0021] 1.5) The velocity inlet wall is set to the incoming wind speed, the pressure outlet wall is set to 0Pa, and the remaining four walls and the body wall are set to no-slip walls;

[0022] 1.6) Run the simulation software to obtain the time domain flow field data of the front side window surface of the car.

[0023] Furthermore, the length of the virtual numerical wind tunnel is 15 times the length of the vehicle to be tested, the width is 13 times the width of the vehicle, and the height is 7 times the height of the vehicle;

[0024] The distance between the vehicle model to be tested and the speed inlet wall is 3 times the length of the vehicle body, and the distance between the two sides of the vehicle body and the two side walls is 6 times the width of the vehicle body;

[0025] Further, in step 2), the step of processing the time domain flow field data of the front side window surface of the automobile includes:

[0026] 2.1) Perform Fourier transform on the time domain flow field data of the front side window surface of the car to obtain:

[0027]

[0028] Where ω is the circular frequency, x, y are spatial quantities, t is time, e is a natural constant, and i is an imaginary number; p(x, y, t) is the time domain flow field data; p * (x, y, ω) is the time domain flow field data after Fourier transformation;

[0029] 2.2) Perform a second Fourier transform on the pressure pulsation in the frequency domain in the x and y directions to obtain the pressure data P in the wave number domain * (k x ,k y ,ω), that is:

[0030]

[0031] In the formula, k x is the wave number in the x direction, k y is the wave number in the y direction;

[0032] 2.3) In the two-dimensional wave number domain (k x ,k y ), separating the acoustic energy and the water kinetic energy according to the difference in propagation speeds of the acoustic energy and the water kinetic energy;

[0033] The distribution of acoustic energy is as follows:

[0034]

[0035] In the formula, c is the speed of sound; f is the frequency;

[0036] The distribution of water kinetic energy is shown below:

[0037]

[0038] Where u is the air flow velocity.

[0039] 2.4) The decomposed sound energy is used as the sound pressure excitation of the surface of the front side window of the car, and the decomposed water kinetic energy is used as the hydrodynamic pressure excitation of the surface of the front side window of the car; the sound pressure excitation of the surface of the front side window of the car and the hydrodynamic pressure excitation of the surface of the front side window of the car are used as the frequency domain flow field excitation data of the surface of the front side window of the car.

[0040] Furthermore, the vehicle geometric parameter data is obtained by measuring by geometric measurement means;

[0041] The automobile geometric parameters include but are not limited to: vehicle width, front wind breaking angle, A-pillar width, A-pillar deflection, A-pillar side step height, A-pillar roll angle, A-pillar angle, rearview mirror handle thickness, rearview mirror inner side angle, rearview mirror outer side angle, rearview mirror upper deflection, rearview mirror bottom deflection, rearview mirror distance from side window, rearview mirror height, rearview mirror width and other parameters that can describe the geometric shape of the A-pillar and rearview mirror.

[0042] Further, in step 4), the step of calculating the contribution of the vehicle geometric parameters to each flow field excitation includes:

[0043] 4.1) Normalize the vehicle geometric parameters;

[0044] 4.2) With the normalized geometric parameters as input variables and the vehicle front side window flow field excitation as output, an expression in the form of regression analysis is established, namely:

[0045]

[0046] In the formula, X j represents the jth input variable, j = 1, 2, ..., m, β j represents the regression coefficient corresponding to the jth input variable, β0 is a constant; Y is the output;

[0047] 4.3) Using regression problem optimization algorithm to calculate the regression coefficient corresponding to each input variable;

[0048] 4.4) Calculate the contribution of each input variable to the output, that is:

[0049]

[0050] In the formula, For contribution amount.

[0051] Further, in step 5), the step of taking the vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% as the key geometric parameters is as follows:

[0052] In each sub-frequency band, the average contribution of each geometric parameter is calculated, and the geometric parameters whose average contribution is in the top n% are selected as the key geometric parameters of the sub-frequency band.

[0053] Further, in step 6), the step of establishing a neural network model using machine learning technology includes:

[0054] 6.1) Determine the basic structure of the neural network, including an input layer, one or more hidden layers, and an output layer;

[0055] 6.2) Determine the number of neurons in each layer of the neural network.

[0056] Further, in step 7), the step of training the neural network model using the sub-frequency segment sample data set includes:

[0057] 7.1) Normalize the data of the sample data set and divide the normalized sample data set into a test set and a training set;

[0058] 7.2) Select the training function of the neural network model and the activation function of each hidden layer;

[0059] 7.3) Taking the divided sub-frequency segments as units, divide the sample set corresponding to each sub-frequency segment into a training set and a test set;

[0060] 7.4) Use the training set to train the neural network model;

[0061] 7.5) Use the test set to test the neural network model. If the error is less than the preset value, the front side window flow field excitation prediction model is output. Otherwise, the sample data set is reconstructed.

[0062] 7.6) Repeat steps 7.3)-7.5) to obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band.

[0063] Furthermore, the neural network model includes a back-propagation neural network model.

[0064] The technical effect of the present invention is unquestionable. The present invention proposes a frequency-band modeling method for a neural network model for predicting the flow field excitation of the front side window of an automobile. The method divides the prediction frequency band based on the result of parameter contribution analysis, and screens key geometric parameters and divides the excitation data in units of the divided frequency bands. The frequency-band modeling method reduces the input dimension of the neural network model in each frequency band and improves the prediction accuracy of the model. It has the advantages of high prediction efficiency and accurate results. It has great application value and broad application prospects in the fields of automobile aerodynamic shape optimization design and automobile wind noise prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flowchart of a frequency-band modeling method for a neural network model used to predict the flow field excitation of a front side window of an automobile;

[0066] Figure 2 This is a schematic diagram of the appearance of the virtual numerical wind tunnel;

[0067] Figure 3 A schematic diagram of the location of the front side window mesh encryption area;

[0068] Figure 4 The acoustic pressure excitation and hydrodynamic pressure excitation data diagrams are obtained after the time domain flow field data is decomposed by the WFS method;

[0069] Figure 5 This is a schematic diagram of the measurement position of the vehicle's geometric parameters;

[0070] Figure 6 It is a schematic diagram of the neural network structure;

[0071] Figure 7 It is a curve diagram of the contribution of each geometric parameter to the sound pressure excitation at different frequencies;

[0072] Figure 8 It is the variation curve of the contribution of each geometric parameter to the hydrodynamic pressure excitation at different frequencies;

[0073] Fig. 9 This is a schematic diagram of the BPNN neural network structure used in Example 2;

[0074] Fig.10 Comparison between the predicted sound pressure excitation and the expected sound pressure excitation obtained by this method in test set 1;

[0075] Fig.11 Comparison between the predicted sound pressure excitation and the expected sound pressure excitation obtained by this method in test set 2;

[0076] Fig.12 Comparison between the predicted hydrodynamic pressure excitation and the expected acoustic pressure excitation obtained by this method in test set 1;

[0077] Fig.13 Comparison of the predicted hydrodynamic pressure excitation and the expected acoustic pressure excitation obtained by this method in test set 2. DETAILED DESCRIPTION

[0078] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0079] Embodiment 1:

[0080] See also Figures 1 to 13 A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile comprises the following steps:

[0081] 1) Use computational fluid dynamics simulation software to calculate the time domain flow field data of the front side window surface of the car;

[0082] 2) The time domain flow field data is processed using the wave number frequency spectrum decomposition technology to obtain the frequency domain flow field excitation data of the front side window surface of the vehicle;

[0083] 3) Measure the vehicle geometric parameter data and map it with the frequency domain flow field excitation data of the vehicle front side window surface;

[0084] 4) The contribution analysis method is used to calculate the contribution of the vehicle geometric parameters to each flow field excitation, and the predicted frequency segment is divided according to the contribution size to obtain multiple sub-frequency segments;

[0085] 5) The vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% are taken as key geometric parameters; n is a positive integer.

[0086] According to the sub-frequency segments, the frequency domain flow field excitation data is segmented to obtain the sub-frequency segment frequency domain flow field excitation data;

[0087] Construct a sub-frequency band sample data set with key geometric parameter data as input and sub-frequency band frequency domain flow field excitation data as output;

[0088] 6) Use machine learning technology to build a neural network model;

[0089] 7) Using the sub-frequency band sample data set to train the neural network model, and obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band;

[0090] 8) Combining the front side window flow field excitation prediction model corresponding to each sub-frequency band, thereby constructing the front side window flow field excitation sub-frequency band prediction model;

[0091] 9) Measure the geometric parameter data of the vehicle to be tested and input them into the front side window flow field excitation frequency band prediction model to obtain the vehicle's front side window flow field excitation prediction value.

[0092] In step 1), the step of calculating the time domain flow field data of the front side window surface of the automobile includes:

[0093] 1.1) A virtual numerical wind tunnel is established in the fluid dynamics simulation software. The virtual numerical wind tunnel is a cuboid, in which two walls serve as the velocity inlet wall and the pressure outlet wall;

[0094] 1.2) Build a closed model of the vehicle to be tested;

[0095] 1.3) Place the vehicle model to be tested in the virtual numerical wind tunnel and close to the velocity inlet wall;

[0096] 1.4) Set a dense grid area in the front side window area of ​​the vehicle model to be tested;

[0097] 1.5) The velocity inlet wall is set to the incoming wind speed, the pressure outlet wall is set to 0Pa, and the remaining four walls and the body wall are set to no-slip walls;

[0098] 1.6) Run the simulation software to obtain the time domain flow field data of the front side window surface of the car.

[0099] The length of the virtual numerical wind tunnel is 15 times the length of the vehicle to be tested, the width is 13 times the width of the vehicle, and the height is 7 times the height of the vehicle;

[0100] The distance between the vehicle model to be tested and the speed inlet wall is 3 times the length of the vehicle body, and the distance between the two sides of the vehicle body and the two side walls is 6 times the width of the vehicle body;

[0101] In step 2), the step of processing the time domain flow field data of the front side window surface of the automobile includes:

[0102] 2.1) Perform Fourier transform on the time domain flow field data of the front side window surface of the car to obtain:

[0103]

[0104] Where ω is the circular frequency, x, y are spatial quantities, t is time, e is a natural constant, and i is an imaginary number; p(x, y, t) is the time domain flow field data; p * (x, y, ω) is the time domain flow field data after Fourier transformation;

[0105] 2.2) Perform a second Fourier transform on the pressure pulsation in the frequency domain in the x and y directions to obtain the pressure data P in the wave number domain * (k x ,k y ,ω), that is:

[0106]

[0107] In the formula, k x is the wave number in the x direction, k y is the wave number in the y direction;

[0108] 2.3) In the two-dimensional wave number domain (k x ,k y ), separating the acoustic energy and the water kinetic energy according to the difference in propagation speeds of the acoustic energy and the water kinetic energy;

[0109] The distribution of acoustic energy is as follows:

[0110]

[0111] In the formula, c is the speed of sound; f is the frequency;

[0112] The distribution of water kinetic energy is shown below:

[0113]

[0114] Where u is the air flow velocity.

[0115] 2.4) The decomposed sound energy is used as the sound pressure excitation of the surface of the front side window of the car, and the decomposed water kinetic energy is used as the hydrodynamic pressure excitation of the surface of the front side window of the car; the sound pressure excitation of the surface of the front side window of the car and the hydrodynamic pressure excitation of the surface of the front side window of the car are used as the frequency domain flow field excitation data of the surface of the front side window of the car.

[0116] The vehicle geometric parameter data is obtained by measuring by geometric measurement means;

[0117] The automobile geometric parameters include but are not limited to: vehicle width, front wind breaking angle, A-pillar width, A-pillar deflection, A-pillar side step height, A-pillar roll angle, A-pillar angle, rearview mirror handle thickness, rearview mirror inner side angle, rearview mirror outer side angle, rearview mirror upper deflection, rearview mirror bottom deflection, rearview mirror distance from side window, rearview mirror height, rearview mirror width and other parameters that can describe the geometric shape of the A-pillar and rearview mirror.

[0118] In step 4), the step of calculating the contribution of the vehicle geometric parameters to each flow field excitation includes:

[0119] 4.1) Normalize the vehicle geometric parameters;

[0120] 4.2) With the normalized geometric parameters as input variables and the vehicle front side window flow field excitation as output, an expression in the form of regression analysis is established:

[0121]

[0122] In the formula, X j represents the jth input variable, j = 1, 2, ..., m, β j represents the regression coefficient corresponding to the jth input variable, β0 is a constant. Y is the output;

[0123] 4.3) Using regression problem optimization algorithm to calculate the regression coefficient corresponding to each input variable;

[0124] 4.4) Calculate the contribution of each input variable to the output, that is:

[0125]

[0126] In the formula, For contribution amount.

[0127] In step 5), the step of taking the vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% as the key geometric parameters is as follows:

[0128] In each sub-frequency band, the average contribution of each geometric parameter is calculated, and the geometric parameters whose average contribution is in the top n% are selected as the key geometric parameters of the sub-frequency band.

[0129] In step 6), the steps of establishing a neural network model using machine learning technology include:

[0130] 6.1) Determine the basic structure of the neural network, including an input layer, one or more hidden layers, and an output layer;

[0131] 6.2) Determine the number of neurons in each layer of the neural network.

[0132] In step 7), the step of training the neural network model using the sub-frequency segment sample data set includes:

[0133] 7.1) Normalize the data of the sample data set and divide the normalized sample data set into a test set and a training set;

[0134] 7.2) Select the training function of the neural network model and the activation function of each hidden layer;

[0135] 7.3) Taking the divided sub-frequency segments as units, divide the sample set corresponding to each sub-frequency segment into a training set and a test set;

[0136] 7.4) Use the training set to train the neural network model;

[0137] 7.5) Use the test set to test the neural network model. If the error is less than the preset value, the front side window flow field excitation prediction model is output. Otherwise, the sample data set is reconstructed.

[0138] 7.6) Repeat steps 7.3)-7.5) to obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band.

[0139] The neural network model includes a back-propagation neural network model.

[0140] Embodiment 2:

[0141] A neural network model frequency band modeling method for predicting the flow field excitation of the front side window of an automobile comprises the following steps:

[0142] 1) Use computational fluid dynamics simulation software to calculate the time domain flow field data of the front side window surface of the car;

[0143] 2) The time domain flow field data is processed using the wave number frequency spectrum decomposition technology to obtain the frequency domain flow field excitation data of the front side window surface of the vehicle;

[0144] 3) Measure the vehicle geometric parameter data and map it with the frequency domain flow field excitation data of the vehicle front side window surface;

[0145] 4) The contribution analysis method is used to calculate the contribution of the vehicle geometric parameters to each flow field excitation, and the predicted frequency segment is divided according to the contribution size to obtain multiple sub-frequency segments;

[0146] 5) The vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% are taken as key geometric parameters;

[0147] According to the sub-frequency segments, the frequency domain flow field excitation data is segmented to obtain the sub-frequency segment frequency domain flow field excitation data;

[0148] Construct a sub-frequency band sample data set with key geometric parameter data as input and sub-frequency band frequency domain flow field excitation data as output;

[0149] 6) Use machine learning technology to build a neural network model;

[0150] 7) Using the sub-frequency band sample data set to train the neural network model, and obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band;

[0151] 8) Combining the front side window flow field excitation prediction model corresponding to each sub-frequency band, thereby constructing the front side window flow field excitation sub-frequency band prediction model;

[0152] 9) Measure the geometric parameter data of the vehicle to be tested and input them into the front side window flow field excitation frequency band prediction model to obtain the vehicle's front side window flow field excitation prediction value.

[0153] Embodiment 3:

[0154] A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content is the same as that of Example 2, further, in step 1), the step of calculating the time domain flow field data of the surface of the front side window of the automobile comprises:

[0155] 1.1) A virtual numerical wind tunnel is established in the fluid dynamics simulation software. The virtual numerical wind tunnel is a cuboid, in which two walls serve as the velocity inlet wall and the pressure outlet wall;

[0156] 1.2) Build a closed model of the vehicle to be tested;

[0157] 1.3) Place the vehicle model to be tested in the virtual numerical wind tunnel and close to the velocity inlet wall;

[0158] 1.4) Set a dense grid area in the front side window area of ​​the vehicle model to be tested;

[0159] 1.5) The velocity inlet wall is set to the incoming wind speed, the pressure outlet wall is set to 0Pa, and the remaining four walls and the body wall are set to no-slip walls;

[0160] 1.6) Run the simulation software to obtain the time domain flow field data of the front side window surface of the car.

[0161] Embodiment 4:

[0162] A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content of which is the same as any one of Embodiments 2-3, further, the length of the virtual numerical wind tunnel is 15 times the length of the vehicle body to be tested, the width is 13 times the width of the vehicle body, and the height is 7 times the height of the vehicle body;

[0163] The distance between the vehicle model to be tested and the speed inlet wall is 3 times the length of the vehicle body, and the distance between the two sides of the vehicle body and the two side walls is 6 times the width of the vehicle body;

[0164] Embodiment 5:

[0165] A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content of which is the same as any one of embodiments 2-4, further, in step 2), the step of processing the time domain flow field data of the surface of the front side window of the automobile comprises:

[0166] 2.1) Perform Fourier transform on the time domain flow field data of the front side window surface of the car to obtain:

[0167]

[0168] In the formula, ω is the circular frequency, x, y are spatial quantities, t is time, e is a natural constant, and i is an imaginary number;

[0169] 2.2) Perform a secondary Fourier transform on the pressure pulsation in the frequency domain in the x and y directions to obtain the pressure data in the wave number domain, namely:

[0170]

[0171] In the formula, k x is the wave number in the x direction, k y is the wave number in the y direction;

[0172] 2.3) In the two-dimensional wave number domain (k x ,k y ), separating the acoustic energy and the water kinetic energy according to the difference in propagation speeds of the acoustic energy and the water kinetic energy;

[0173] The distribution of acoustic energy is as follows:

[0174]

[0175] In the formula, c is the speed of sound; f is the frequency;

[0176] The distribution of water kinetic energy is shown below:

[0177]

[0178] Where u is the air flow velocity.

[0179] 2.4) The decomposed sound energy is used as the sound pressure excitation of the surface of the front side window of the car, and the decomposed water kinetic energy is used as the hydrodynamic pressure excitation of the surface of the front side window of the car; the sound pressure excitation of the surface of the front side window of the car and the hydrodynamic pressure excitation of the surface of the front side window of the car are used as the frequency domain flow field excitation data of the surface of the front side window of the car.

[0180] Embodiment 6:

[0181] A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content of which is the same as any one of embodiments 2-5, and further, the automobile geometric parameter data is measured by a geometric measurement method;

[0182] The automobile geometric parameters include but are not limited to: vehicle width, front wind breaking angle, A-pillar width, A-pillar deflection, A-pillar side step height, A-pillar roll angle, A-pillar angle, rearview mirror handle thickness, rearview mirror inner side angle, rearview mirror outer side angle, rearview mirror upper deflection, rearview mirror bottom deflection, rearview mirror distance from side window, rearview mirror height, rearview mirror width and other parameters that can describe the geometric shape of the A-pillar and rearview mirror.

[0183] Embodiment 7:

[0184] A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content is the same as any one of embodiments 2-6, further, in step 4), the step of calculating the contribution of the automobile geometric parameters to each flow field excitation includes:

[0185] 4.1) Normalize the vehicle geometric parameters;

[0186] 4.2) With the normalized geometric parameters as input variables and the vehicle front side window flow field excitation as output, an expression in the form of regression analysis is established:

[0187]

[0188] In the formula, X j represents the jth input variable, j = 1, 2, ..., m, β j It represents the regression coefficient corresponding to the j-th input variable, and β0 is a constant.

[0189] 4.3) Using regression problem optimization algorithm to calculate the regression coefficient corresponding to each input variable;

[0190] 4.4) Calculate the contribution of each input variable to the output, that is:

[0191]

[0192] Embodiment 8:

[0193] A neural network model frequency band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content is the same as any one of embodiments 2-7, further, in step 5), the step of taking the automobile geometric parameters whose contribution in each sub-frequency band is in the top n% as the key geometric parameters is:

[0194] In each sub-frequency band, the average contribution of each geometric parameter is calculated, and the geometric parameters whose average contribution is in the top n% are selected as the key geometric parameters of the sub-frequency band.

[0195] Embodiment 9:

[0196] A neural network model frequency band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content is the same as any one of embodiments 2-8, further, in step 6),

[0197] The steps to build a neural network model using machine learning techniques include:

[0198] 6.1) Determine the basic structure of the neural network, including an input layer, one or more hidden layers, and an output layer;

[0199] 6.2) Determine the number of neurons in each layer of the neural network.

[0200] Embodiment 10:

[0201] A neural network model sub-frequency band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content is the same as any one of embodiments 2-9, further, in step 7), the step of training the neural network model using the sub-frequency band sample data set includes:

[0202] 7.1) Normalize the data of the sample data set and divide the normalized sample data set into a test set and a training set;

[0203] 7.2) Select the training function of the neural network model and the activation function of each hidden layer;

[0204] 7.3) Taking the divided sub-frequency segments as units, divide the sample set corresponding to each sub-frequency segment into a training set and a test set;

[0205] 7.4) Use the training set to train the neural network model;

[0206] 7.5) Use the test set to test the neural network model. If the error is less than the preset value, the front side window flow field excitation prediction model is output. Otherwise, the sample data set is reconstructed.

[0207] 7.6) Repeat steps 7.3)-7.5) to obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band.

[0208] Embodiment 11:

[0209] A neural network model frequency band modeling method for predicting the flow field excitation of the front side window of an automobile, the technical content of which is the same as any one of Examples 2-10, and further, the neural network model includes a back propagation neural network model.

[0210] Embodiment 12:

[0211] See also Figures 1 to 9 A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile comprises the following steps:

[0212] 1) Using computational fluid dynamics (CFD) simulation software to solve and obtain the time domain flow field data of the front side window surface of the car;

[0213] 2) Use the wavenumber frequency spectrum decomposition (WFS) technology to process the time domain flow field data and obtain the frequency domain flow field excitation data of the front side window surface of the car;

[0214] 3) Use geometric measurement methods to measure the geometric parameters of the vehicle;

[0215] 4) The contribution analysis method is used to obtain the contribution of the vehicle geometric parameters to the excitation of each flow field, and the predicted frequency band is divided according to the contribution analysis results;

[0216] 5) Taking frequency band as the unit, the parameters with the highest contribution in each frequency band are selected as the key geometric parameters of the frequency band, and the flow field excitation data is processed in sections according to the frequency band;

[0217] 6) Use machine learning technology to build a neural network model;

[0218] 7) Take the key geometric parameters in each frequency band as the input of the neural network model, and take the excitation data of the corresponding frequency band as the output of the neural network model. Train the neural network model in units of frequency bands.

[0219] 8) combining the prediction results of the neural network models of each frequency band to establish a neural network model that can quickly predict the front side window flow field excitation in the entire prediction frequency band;

[0220] 9) Measure the geometric parameter data of the vehicle to be tested, input the frequency-segment prediction model of the front side window flow field excitation established in step 8), and obtain the predicted value of the front side window flow field excitation of the vehicle to be tested.

[0221] The steps of using CFD simulation software to solve and obtain the time domain flow field data of the front side window surface of the car include:

[0222] 1.1) A virtual numerical wind tunnel is established in the CFD simulation software. The length of the virtual numerical wind tunnel is about 15 times the length of the vehicle to be tested, the width is about 13 times the width of the vehicle, and the height is about 7 times the height of the vehicle. The virtual numerical wind tunnel consists of the velocity inlet wall, the pressure outlet wall, the ground boundary, the top boundary, and the remaining two opposite wall boundaries;

[0223] 1.2) Perform geometric processing on the vehicle model to be tested, retain only the outermost geometric surface and strictly ensure that the interior of the vehicle is a closed space as a whole;

[0224] 1.3) Place the vehicle to be tested in the virtual numerical wind tunnel, close to the velocity inlet wall, with the front of the vehicle to be tested about 3 times the length of the vehicle body away from the inlet wall, and the two sides of the vehicle body to be tested about 6 times the width of the vehicle body away from the two side walls;

[0225] 1.4) Setting a dense grid area to improve the accuracy of the acquired flow field data. The position of the dense grid area is set in the front side window area of ​​the vehicle to be tested. The dense grid area is set to 8 layers of quadrangular prism grids with a boundary layer thickness of 2 mm;

[0226] 1.5) Set the simulation software to simulate, set the incoming wind speed on the velocity inlet wall, set the pressure outlet on the pressure outlet wall, the size is 0Pa, and the remaining walls including the body wall are set to no-slip walls;

[0227] 1.6) Run the simulation software to obtain and save the time domain flow field data of the vehicle to be tested.

[0228] The steps of solving the frequency domain flow field excitation of the front side window surface of the car using the WFS method include:

[0229] 2.1) Through Fourier transform, the time domain flow field data (pressure pulsation p(x,y,t)) obtained by CFD simulation software is converted into the frequency domain:

[0230]

[0231] Where ω is the circular frequency, x and y are spatial quantities, t is time, e is a natural constant, and i is an imaginary number.

[0232] 2.2) By performing Fourier transform in the x-direction and y-direction, the pressure pulsation in the frequency domain can be converted into pressure data in the wavenumber domain:

[0233]

[0234] In the formula, k x is the wave number in the x direction, k y is the wave number in the y direction.

[0235] 2.3) According to the difference in propagation speed between acoustic energy and water kinetic energy, in the two-dimensional wave number domain (k x ,k y ) can separate these two energies.

[0236] The distribution of sound energy is:

[0237]

[0238] The distribution of water kinetic energy is:

[0239]

[0240] Where c is the speed of sound, f is the frequency, and u is the air flow velocity.

[0241] 2.4) The acoustic energy decomposed by the WFS method is the acoustic pressure excitation on the surface of the front side window of the car, and the water kinetic energy decomposed by the WFS method is the water dynamic pressure excitation on the surface of the front side window of the car.

[0242] The automobile geometric parameters measured by geometric measurement methods include but are not limited to: vehicle width, front wind breaking angle, A-pillar width, A-pillar deflection, A-pillar side step height, A-pillar roll angle, A-pillar angle, rearview mirror handle thickness, rearview mirror inner side angle, rearview mirror outer side angle, rearview mirror upper deflection, rearview mirror bottom deflection, rearview mirror distance from side window, rearview mirror height, rearview mirror width and other parameters that can describe the geometric shape of the A-pillar and rearview mirror.

[0243] The contribution amount of the vehicle geometric parameters to each flow field excitation is obtained by using the contribution analysis method, and the predicted frequency band is divided according to the contribution analysis result. The steps of obtaining the contribution amount of the vehicle geometric parameters by using the contribution analysis method include:

[0244] 4.1) Use the maximum-minimum normalization method to process the vehicle geometric parameters to the interval [0,1]:

[0245]

[0246] In the formula, is the normalized value, i represents the i-th sample, i = 1, 2, ..., p, m represents the m-th feature in the same sample, m = 1, 2, ..., n, α im Represents the mth eigenvalue in the i-th sample.

[0247] 4.2) The normalized geometric parameter α j As the input variable, the vehicle front side window flow field excitation Y is used as the output value, and the expression in the form of regression analysis is established:

[0248]

[0249] In the formula, α j represents the jth input variable, j = 1, 2, ..., m, β j It represents the regression coefficient corresponding to the j-th input variable, and β0 is a constant.

[0250] 4.3) The ridge regression method is used to solve the regression coefficients corresponding to each input variable. The loss function of the ridge regression is:

[0251]

[0252] In the formula, α j represents the jth input variable, j = 1, 2, ..., m, β j represents the regression coefficient corresponding to the jth input variable, β0 is a constant, Y i represents the output value of the i-th sample, i = 1, 2, …, n, and λ is the regularization parameter.

[0253] Minimize the loss function to obtain the regression coefficients corresponding to each input variable.

[0254] 4.4) Use the following formula to calculate the contribution of each input variable to the output:

[0255]

[0256] In the formula, α j represents the jth input variable, j = 1, 2, ..., m, β j Represents the regression coefficient corresponding to the jth input variable.

[0257] The predicted frequency segment is divided according to the variation characteristics of the geometric parameter contribution with frequency, and the entire predicted frequency segment is divided into several small frequency segments.

[0258] Taking the divided small frequency segments as units, calculate the average contribution of each geometric parameter in each small frequency segment and arrange them in descending order according to the average contribution. Select the top geometric parameters in each frequency segment as the key geometric parameters of the frequency segment.

[0259] Taking several divided small frequency bands as units, the flow field excitation data (acoustic pressure excitation and hydrodynamic pressure excitation) of the front side window of the automobile is divided and processed according to the frequency band as the excitation output data of the frequency band.

[0260] The steps to build a neural network model using machine learning techniques include:

[0261] 6.1) Determine the basic structure of the neural network, i.e., an input layer, one or more hidden layers, and an output layer;

[0262] 6.2) Determine the number of neurons in each layer of the neural network;

[0263] 6.3) Normalize the input and output sample data, using the same maximum-minimum normalization method as step 4.1) to process the data to the interval [0,1];

[0264] 6.4) Select the training function of the neural network, the training function is the trainlm function, that is, the Levenberg-Marquardt (LM) algorithm. For the following nonlinear least squares problem:

[0265]

[0266] Where H(x) is the residual sum of squares, P i ′ is the network model prediction vector, P i is the expected output vector, n is the number of samples, e i (x) is the residual, and a is the vector consisting of the network weights and thresholds.

[0267] The iterative step size of the LM algorithm is:

[0268] I LM =-(J T (a)J(a)+μI) -1 J T (a)e(a) (10)

[0269] Where: μ is the damping factor, I is the identity matrix, and J(a) is the Jacobian matrix.

[0270] 6.5) Select the activation function of each hidden layer of the neural network. The activation function is the Sigmoid function. Its mathematical definition is:

[0271]

[0272] Where f(b) is the output value of the Sigmoid function, e is a natural constant, and b is the input value of the previous layer of neurons.

[0273] 6.6) The performance of the neural network model is expressed by Mean Squared Error (MSE):

[0274]

[0275] Where n is the number of samples, P i ′ is the network model prediction vector, P i is the desired output vector.

[0276] The key geometric parameters in each frequency band are used as the input of the neural network model, and the excitation data of the corresponding frequency band is used as the output of the neural network model. Taking the frequency band as the unit, the steps of training the neural network model include:

[0277] 7.1) Divide the sample data into a training sample set and a test sample set. The training sample set is used to train the network model. The test sample set is used to verify the model prediction results and shall not be used for network training;

[0278] 7.2) Taking the divided small frequency bands as units, the key geometric parameters of the frequency band are used as input samples, and the acoustic pressure excitation and hydrodynamic pressure excitation data of the front side window of the car in the frequency band are used as output samples to train the network model and obtain the network prediction model of the corresponding frequency band;

[0279] 7.3) Use the test set data to verify the model prediction effect;

[0280] By combining the network prediction models of each frequency band, the prediction model of the full-frequency band of the vehicle front side window flow field excitation is obtained.

[0281] Embodiment 13:

[0282] See also Figures 1 to 13 A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile comprises the following steps:

[0283] 1) Using computational fluid dynamics (CFD) simulation software to solve and obtain the time domain flow field data of the front side window surface of the car;

[0284] 2) Use the wavenumber frequency spectrum decomposition (WFS) technology to process the time domain flow field data and obtain the frequency domain flow field excitation data of the front side window surface of the car;

[0285] 3) Use geometric measurement methods to measure the geometric parameters of the vehicle;

[0286] 4) The contribution analysis method is used to obtain the contribution of the vehicle geometric parameters to the excitation of each flow field, and the predicted frequency band is divided according to the contribution analysis results;

[0287] 5) Taking frequency band as the unit, the parameters with the highest contribution in each frequency band are selected as the key geometric parameters of the frequency band, and the flow field excitation data is processed in sections according to the frequency band;

[0288] 6) Use machine learning technology to build a neural network model;

[0289] 7) Take the key geometric parameters in each frequency band as the input of the neural network model, and take the excitation data of the corresponding frequency band as the output of the neural network model. Train the neural network model in units of frequency bands.

[0290] 8) combining the prediction results of the neural network models of each frequency band to establish a neural network model that can quickly predict the front side window flow field excitation in the entire prediction frequency band;

[0291] 9) Measure the geometric parameter data of the vehicle to be tested, input the frequency-segment prediction model of the front side window flow field excitation established in step 8), and obtain the predicted value of the front side window flow field excitation of the vehicle to be tested.

[0292] 1) Using computational fluid dynamics (CFD) simulation software to solve and obtain the time domain flow field data of the front side window surface of the car;

[0293] The steps of using CFD simulation software to solve and obtain the time domain flow field data of the front side window surface of the car include:

[0294] 1.1) A virtual numerical wind tunnel is established in the CFD simulation software. The virtual numerical wind tunnel is 60m long, 24m wide and 12m high. The virtual numerical wind tunnel consists of a velocity inlet wall, a pressure outlet wall, a ground boundary, a top boundary and the remaining two opposite wall boundaries;

[0295] 1.2) Perform geometric processing on the vehicle model to be tested, retain only the outermost geometric surface and strictly ensure that the interior of the vehicle is a closed space as a whole;

[0296] 1.3) Place the vehicle to be tested in the virtual numerical wind tunnel, close to the velocity inlet wall, with the front of the vehicle to be tested about 3 times the length of the vehicle body away from the inlet wall, and the two sides of the vehicle body to be tested are equidistant from the two side walls;

[0297] 1.4) Setting a dense grid area to improve the accuracy of the acquired flow field data. The location of the dense grid area is set in the front side window area of ​​the vehicle to be tested. The dense grid area is set to 8 layers of quadrangular prism grids with a boundary layer thickness of 2 mm, and the grid thickness growth rate of adjacent layers is 1.5;

[0298] 1.5) Set the simulation software to simulate, set the incoming wind speed on the velocity inlet wall to 120km / h, set the pressure outlet on the pressure outlet wall to 0Pa, and set the remaining walls including the vehicle body wall to no-slip walls;

[0299] 1.6) The SSTk-ω model with two turbulence models is used for numerical simulation, and the pressure-velocity coupling term is solved by the semi-implicit-method for pressure linked equations (SIMPLE). The transient simulation is calculated by detached eddy simulation. Some other settings of the solver are shown in the following table.

[0300]

[0301] 1.7) Run the simulation software to solve and save the front side window flow field data of 46 groups of different appearance conditions for 6 different models.

[0302] 2) Use the wavenumber frequency spectrum decomposition (WFS) technology to process the time domain flow field data and obtain the frequency domain flow field excitation data of the front side window surface of the car;

[0303] The steps of solving the frequency domain flow field excitation of the front side window surface of the car using the WFS method include:

[0304] 2.1) Through Fourier transform, the time domain flow field data (pressure pulsation p(x,y,t)) obtained by CFD simulation software is converted into the frequency domain:

[0305]

[0306] Where ω is the circular frequency, x and y are spatial quantities, t is time, e is a natural constant, and i is an imaginary number.

[0307] 2.2) By performing Fourier transform in the x-direction and y-direction, the pressure pulsation in the frequency domain can be converted into pressure data in the wavenumber domain:

[0308]

[0309] In the formula, k x is the wave number in the x direction, k y is the wave number in the y direction.

[0310] 2.3) According to the difference in propagation speed between acoustic energy and water kinetic energy, in the two-dimensional wave number domain (k x ,k y ) can separate these two energies.

[0311] The distribution of sound energy is:

[0312]

[0313] The distribution of water kinetic energy is:

[0314]

[0315] Where c is the speed of sound and u is the air flow velocity.

[0316] 2.4) The acoustic energy decomposed by the WFS method is the acoustic pressure excitation on the surface of the front side window of the car, and the water kinetic energy decomposed by the WFS method is the water dynamic pressure excitation on the surface of the front side window of the car. Both data are saved as one-third octave band results.

[0317] 3) Use geometric measurement methods to measure the geometric parameters of the vehicle;

[0318] The automobile geometric parameters measured using geometric measurement methods are shown in the following table, totaling 15 geometric parameters.

[0319]

[0320] 4) The contribution analysis method is used to obtain the contribution of the vehicle geometric parameters to the excitation of each flow field, and the predicted frequency band is divided according to the contribution analysis results;

[0321] 4.1) Contribution analysis results are as follows Figure 7-Figure 8 As shown;

[0322] 4.2) According to the contribution analysis results, the predicted frequency band is divided into 4 small frequency bands (one-third octave bands), namely: 100Hz-250Hz, 315Hz-800Hz, 1000Hz-2500Hz, 3150Hz-8000Hz.

[0323] 5) Taking the frequency band as the unit, calculate the average contribution of each geometric parameter in each small frequency band, and arrange them in descending order according to the average contribution. Figure * shows the average contribution of each geometric parameter to the acoustic pressure excitation and hydrodynamic pressure excitation in the first frequency band (100Hz-250Hz). The top 5 geometric parameters in each frequency band are selected as the key geometric parameters of the frequency band. The top 5 key geometric parameters in each frequency band are shown in the following table.

[0324]

[0325] Taking the four divided small frequency bands as units, the flow field excitation data (acoustic pressure excitation and hydrodynamic pressure excitation) of the front side window of the automobile is divided and processed according to the frequency band as the excitation output data of the frequency band.

[0326] 6) Use machine learning technology to establish a back propagation neural network (BPNN) model.

[0327] The steps to build a BPNN model using machine learning techniques include:

[0328] 6.1) Determine the basic structure of the neural network, namely one input layer, three hidden layers and one output layer;

[0329] 6.2) Determine the number of neurons in each layer of the neural network. The number of neurons in the input layer is 5, and the number of neurons in each hidden layer is 20;

[0330] 6.3) Normalize the input and output sample data and use the maximum-minimum normalization method to process the data into the interval [0,1];

[0331] 6.4) Select the training function of the neural network, the training function is the trainlm function. Select the activation function of each hidden layer of the neural network, the activation function is the Sigmoid function. The performance of the neural network model is represented by the mean square error (MSE).

[0332] 7) Take the key geometric parameters in each frequency band as the input of the neural network model, and take the excitation data of the corresponding frequency band as the output of the neural network model. Train the neural network model in units of frequency bands. The steps include:

[0333] 7.1) Divide the 46 sets of sample data into 44 sets of training sample sets and 2 sets of test sample sets. The training sample sets are used to train the network model. The test sample sets are used to verify the model prediction results and shall not be used for network training;

[0334] 7.2) Taking the divided several small frequency bands as units, the key geometric parameters of the frequency band are used as input samples, and the acoustic pressure excitation and hydrodynamic pressure excitation data of the front side window of the car in the frequency band are used as output samples to train the network model and obtain the neural network prediction model in each frequency band;

[0335] 7.3) According to the key geometric parameters corresponding to each frequency band, the geometric parameters of the two test sample sets are input into the prediction model of the frequency band to obtain the excitation prediction results of the frequency band;

[0336] 8) Combine the excitation prediction results of each frequency band to obtain the excitation prediction results of the entire prediction frequency range. Compare the predicted data with the expected data of the test sample set, calculate the average sound pressure level error and relative error, and the relative error is defined as follows:

[0337]

[0338] 9) Compare the errors between the front side window excitation prediction data obtained by the method proposed in the present invention and the expected data, as shown in the following table.

[0339]

[0340] It can be seen that the average sound pressure level error is within 1dB for both acoustic pressure excitation and hydrodynamic pressure excitation, and the relative error is controlled within 1.5%. This further verifies the accuracy of the proposed neural network modeling method for predicting the flow field excitation of the front side window of the car.

Claims

1. A neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile, characterized in that: The following steps are involved: 1) Use computational fluid dynamics simulation software to calculate the time domain flow field data of the front side window surface of the car. 2) The time domain flow field data is processed using the wave number frequency spectrum decomposition technology to obtain the frequency domain flow field excitation data of the front side window surface of the vehicle; 3) Measure the vehicle geometric parameter data and map it with the frequency domain flow field excitation data of the vehicle front side window surface; 4) The contribution analysis method is used to calculate the contribution of the vehicle geometric parameters to each flow field excitation, and the predicted frequency segment is divided according to the contribution size to obtain multiple sub-frequency segments; 5) The vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% are taken as key geometric parameters; According to the sub-frequency segments, the frequency domain flow field excitation data is segmented to obtain the sub-frequency segment frequency domain flow field excitation data; Construct a sub-frequency band sample data set with key geometric parameter data as input and sub-frequency band frequency domain flow field excitation data as output; 6) Use machine learning technology to build a neural network model; 7) Using the sub-frequency band sample data set to train the neural network model, and obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band; 8) Combining the front side window flow field excitation prediction model corresponding to each sub-frequency band, thereby constructing the front side window flow field excitation sub-frequency band prediction model; 9) Measure the geometric parameter data of the vehicle to be tested and input them into the front side window flow field excitation frequency band prediction model to obtain the vehicle's front side window flow field excitation prediction value.

2. According to claim 1, a neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile is characterized in that: In step 1), the step of calculating the time domain flow field data of the front side window surface of the automobile includes: 1.1) A virtual numerical wind tunnel is established in the fluid dynamics simulation software. The virtual numerical wind tunnel is a cuboid, in which two walls serve as the velocity inlet wall and the pressure outlet wall; 1.2) Build a closed model of the vehicle to be tested; 1.3) Place the vehicle model to be tested in the virtual numerical wind tunnel and close to the velocity inlet wall; 1.4) Set a dense grid area in the front side window area of ​​the vehicle model to be tested; 1.5) The velocity inlet wall is set to the incoming wind speed, the pressure outlet wall is set to 0Pa, and the remaining four walls and the body wall are set to no-slip walls; 1.6) Run the simulation software to obtain the time domain flow field data of the front side window surface of the car.

3. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: The length of the virtual numerical wind tunnel is 15 times the length of the vehicle to be tested, the width is 13 times the width of the vehicle, and the height is 7 times the height of the vehicle; The distance between the vehicle model to be tested and the speed inlet wall is 3 times the body length, and the distance between the two sides of the body and the two side walls is 6 times the body width.

4. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: In step 2), the step of processing the time domain flow field data of the front side window surface of the automobile includes: 2.1) Perform Fourier transform on the time domain flow field data of the front side window surface of the car to obtain: p * (x,y,ω)=∫p(x,y,t)e (-iωt) dt (1) Where ω is the circular frequency, x, y are spatial quantities, t is time, e is a natural constant, and i is an imaginary number; p(x, y, t) is the time domain flow field data; p * (x, y, ω) is the time domain flow field data after Fourier transformation; 2.2) Perform a second Fourier transform on the pressure pulsation in the frequency domain in the x and y directions to obtain the pressure data P in the wave number domain * (k x ,k y ,ω), that is: In the formula, k x is the wave number in the x direction, k y is the wave number in the y direction; 2.3) In the two-dimensional wave number domain (k x ,k y ), separating the acoustic energy and the water kinetic energy according to the difference in propagation speeds of the acoustic energy and the water kinetic energy; The distribution of acoustic energy is as follows: In the formula, c is the speed of sound; f is the frequency; The distribution of water kinetic energy is shown below: Where u is the air flow velocity. 2.4) The decomposed sound energy is used as the sound pressure excitation of the surface of the front side window of the car, and the decomposed water kinetic energy is used as the hydrodynamic pressure excitation of the surface of the front side window of the car; the sound pressure excitation of the surface of the front side window of the car and the hydrodynamic pressure excitation of the surface of the front side window of the car are used as the frequency domain flow field excitation data of the surface of the front side window of the car.

5. The neural network model frequency segment modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: The vehicle geometric parameter data is obtained by measuring by geometric measurement means; The automobile geometric parameters include but are not limited to: vehicle width, front wind breaking angle, A-pillar width, A-pillar deflection, A-pillar side step height, A-pillar roll angle, A-pillar angle, rearview mirror handle thickness, rearview mirror inner side angle, rearview mirror outer side angle, rearview mirror upper deflection, rearview mirror bottom deflection, rearview mirror distance from side window, rearview mirror height, rearview mirror width and other parameters that can describe the geometric shape of the A-pillar and rearview mirror.

6. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: In step 4), the step of calculating the contribution of the vehicle geometric parameters to each flow field excitation includes: 4.1) Normalize the vehicle geometric parameters; 4.2) With the normalized geometric parameters as input variables and the vehicle front side window flow field excitation as output, an expression in the form of regression analysis is established, namely: Where, X j represents the jth input variable, j = 1, 2, ..., m, β j represents the regression coefficient corresponding to the jth input variable, β0 is a constant; Y is the output; 4.3) Using regression problem optimization algorithm to calculate the regression coefficient corresponding to each input variable; 4.4) Calculate the contribution of each input variable to the output, that is: In the formula, For contribution amount.

7. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: In step 5), the step of taking the vehicle geometric parameters whose contribution in each sub-frequency band is in the top n% as the key geometric parameters is as follows: In each sub-frequency band, the average contribution of each geometric parameter is calculated, and the geometric parameters whose average contribution is in the top n% are selected as the key geometric parameters of the sub-frequency band.

8. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: In step 6), the steps of establishing a neural network model using machine learning technology include: 6.1) Determine the basic structure of the neural network, including an input layer, one or more hidden layers, and an output layer; 6.2) Determine the number of neurons in each layer of the neural network.

9. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: In step 7), the step of training the neural network model using the sub-frequency segment sample data set includes: 7.1) Normalize the data of the sample data set and divide the normalized sample data set into a test set and a training set; 7.2) Select the training function of the neural network model and the activation function of each hidden layer; 7.3) Taking the divided sub-frequency segments as units, divide the sample set corresponding to each sub-frequency segment into a training set and a test set; 7.4) Use the training set to train the neural network model; 7.5) Use the test set to test the neural network model. If the error is less than the preset value, the front side window flow field excitation prediction model is output. Otherwise, the sample data set is reconstructed. 7.6) Repeat steps 7.3)-7.5) to obtain the front side window flow field excitation prediction model corresponding to each sub-frequency band.

10. The neural network model frequency-band modeling method for predicting the flow field excitation of the front side window of an automobile according to claim 1, characterized in that: The neural network model includes a back-propagation neural network model.

Images