Methods for predicting the drag coefficient of a car body side profile

Abstract

This invention relates to a method for predicting the drag coefficient of a car body side profile, comprising: S1, acquiring side profile data parameters of the car body and determining the Bézier curve of the side profile; S2, extracting the drag coefficient feature values ​​of the Bézier curve of the side profile and using feature parameters obtained by LASSO regression; S3, training the feature parameters using an LSTM deep learning model; and S4, constructing a car body side profile prediction model as a LASSO-LSTM drag coefficient prediction model to achieve drag coefficient prediction. This invention solves the problem of unclear car body profile features and the inability to extract features when data volume is limited. It improves the model's understanding of the car body profile, reduces the error value of the predicted drag coefficient, and uses LASSO regression parameters as penalty parameters to correct features, avoiding overfitting and achieving fast and automated prediction and calculation of the car body side drag profile.

Description

Technical Field

[0001] This invention belongs to the field of automotive drag prediction technology, and specifically relates to a method for predicting the drag coefficient of the side profile of a car body. Background Technology

[0002] The design of a car's exterior often requires collaboration between CAE engineers and body designers. CAE engineers cannot assess the feasibility of the numerous outline models drawn by body designers in a short period of time. Therefore, improving the efficiency of vehicle exterior design, especially for passenger cars, is crucial; and the key to improving the efficiency of vehicle exterior design lies in improving the analytical efficiency of mechanical designers.

[0003] Deep learning has been widely applied in image recognition, pattern recognition, autonomous driving, and fault detection. In recent years, deep learning has also begun to be combined with CAD and CAE: traditional methods for predicting the drag coefficient of a car profile mainly involve placing the car profile into industrial software for prediction. While deep learning prediction methods exist abroad, these methods still suffer from problems such as large prediction errors and difficulty in feature extraction.

[0004] To address the aforementioned problems, this invention combines the emerging artificial intelligence method of Deep Long Short-Term Memory (LSTM) with Lasso Regression (LASSO) and employs a unique feature extraction method. This results in a system for predicting the drag coefficient of a car's side profile. The method decomposes the car's profile, obtaining corresponding feature values. These feature values ​​are then combined with the penalty parameters of Lasso Regression and input into a deep learning model to predict the drag coefficient of the corresponding car profile. This novel feature extraction and deep learning approach predicts the drag coefficient of the car profile. Furthermore, by integrating deep learning theory with the prediction of the car's drag coefficient, it achieves superior drag coefficient prediction results. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for predicting the drag coefficient of a car body side profile. This method solves the problem that the car body profile features are not obvious and features cannot be extracted when the amount of data is small. It improves the model's understanding of the car body profile, reduces the error in predicting the drag coefficient, and uses LASSO regression parameters as penalty parameters to correct features, avoiding overfitting. This enables fast and automated prediction and calculation of the car body side drag profile.

[0006] To achieve the above objectives, the present invention discloses the following technical solution:

[0007] S1: Obtain the side profile data parameters of the car body and determine the Bezier curve of the side profile of the car body;

[0008] The side profile data parameters of the car body are obtained from the car parameter database. The car profile consists of N data parameters, one of which is the drag coefficient Cd, and N-1 data parameters are the (N-1) / 2 coordinate points that make up the car profile, including the x-coordinate. i and the vertical coordinate y i , i = 1, 2, ..., (N-1) / 2; the Bézier curves of the side profile of the car body include a quadratic Bézier curve B2(t) and a cubic Bézier curve B3(t), and the specific calculation formula is as follows:

[0009]

[0010] Where B3(t) is the cubic Bézier curve formula; B2(t) is the quadratic Bézier curve formula; P 30 P is the first coefficient of the cubic Bézier curve; 31 P is the second coefficient of the cubic Bézier curve; 32 P is the third coefficient of the cubic Bézier curve; 33 P is the fourth coefficient of the cubic Bézier curve. 20 P is the first coefficient of the quadratic Bézier curve; 21 P is the second coefficient of the quadratic Bézier curve; 23 The third coefficient of the quadratic Bézier curve; t is the input to the Bézier curve;

[0011] S2: Extract the drag coefficient characteristic value of the Bezier curve of the side profile of the car body, and use the regression method to obtain the characteristic parameters;

[0012] S21: Extract the drag coefficient feature value from the Bézier curve car profile obtained in step S1; extract the first position feature of each Bézier curve that makes up the side profile of the car body. Second position feature d and third position feature sinθ: Extract the first shape feature Δy1, the second shape feature Δy2, the third shape feature P′0, and the fourth shape feature P′3 of each Bézier curve that makes up the side profile of the car body; construct the drag coefficient feature value X. zu ,for:

[0013]

[0014] Among them, X zu The characteristic value of the drag coefficient; x zj Let j be the characteristic value of the drag coefficient; d is the first positional feature; sinθ is the second positional feature; Δy1 is the first shape feature; Δy2 is the second shape feature; P′0 is the third shape feature; P′3 is the fourth shape feature.

[0015] S22: Divide the drag coefficient feature values ​​extracted in step S21 and their corresponding drag coefficients Cd determined in step 1 into training set and test set;

[0016] S23: Perform LASSO regression on the training set from step S22 to obtain the LASSO regression prediction set X. 输入 The formula for calculating LASSO regression is:

[0017]

[0018] Where ta≥0 is the harmonic parameter; x ij Y is the first input parameter; i This is the second input parameter; The parameter set obtained from LASSO regression; αi 为 First prediction parameter; β j This is the second prediction parameter;

[0019] S3: Train feature parameters using an LSTM deep learning model;

[0020] Obtain the feature parameters from step S2 and input them into the LSTM deep learning model for training; the loss function used for each LSTM layer is MSLE, i.e., mean squared logarithmic error, which is:

[0021]

[0022] in, This is a predicted value; The true value is represented by L; the MSLE loss function is represented by n. 总 This represents the total number of feature parameters.

[0023] S4: Construct a car body side profile prediction model as a LASSO-LSTM drag coefficient prediction model to achieve drag coefficient prediction;

[0024] The LASSO regression prediction set obtained in step S2 and the LSTM deep learning model constructed in step S3 form the LASSO-LSTM drag coefficient prediction model. To ensure the accuracy of the prediction results, m trained LASSO-LSTM drag coefficient prediction models are used for prediction. The average of the m predicted values ​​is then used as the prediction result of the car's drag coefficient.

[0025] Preferably, the first position feature in step S21 The method for obtaining it is as follows:

[0026] Calculate the starting point (x) of each Bézier curve. n-1 ,y n-1 ) and endpoint (x) n ,y n The slope of ) compared to k n The starting point (x) of the previous Bézier curve n-1 ,y n-1 ) and endpoint (x) n ,y n ) connecting line k (n-1) The first position feature The method for obtaining it is as follows:

[0027]

[0028]

[0029] in, The first positional feature; k n k is the slope of the line connecting the start and end points of the current Bézier curve. (n-1) y is the slope of the line connecting the start and end points of the previous Bézier curve; n y is the ordinate of the endpoint of the Bézier curve; (n-1) x represents the ordinate of the starting point of the Bézier curve; n x is the x-coordinate of the endpoint of the Bézier curve; (n-1) is the x-coordinate of the starting point of the Bézier curve.

[0030] Preferably, the method for obtaining the second position feature d in step S21 is as follows:

[0031] Calculate the starting point (x) of each Bézier curve. n-1 ,y n-1 ) and endpoint (x) n ,y n The distance d between the lines connecting the two sides is:

[0032]

[0033] Where d is the second positional feature.

[0034] Preferably, the method for obtaining the third position feature sinθ in step S21 is as follows:

[0035] Calculate the starting point (x) of each Bézier curve n-1 ,y n-1 ) and endpoint (x) n ,y nThe sine of the acute angle θ formed by the line connecting the two lines and the horizontal line, sinθ, is:

[0036]

[0037] Wherein, sinθ is the third position feature;

[0038] For the starting point (x) of a portion of the Bézier curve n-1 ,y n-1 ) and endpoint (x) n ,y n The acute angle θ formed by the line connecting the two points and the horizontal line may have a case where the third position feature sinθ does not exist; in this case, the value of the third position feature sinθ is set to 1.

[0039] Preferably, the method for obtaining the first shape feature Δy1 and the second shape feature Δy2 in step S21 is as follows:

[0040] The constituent points of the third-order Bézier curve are: starting point P0(x0,y0), ending point P3(x3,y3), first control point P1(x1,y1), and second control point P2(x2,y2). The first shape feature Δy1 and the second shape feature Δy2 corresponding to the same abscissa of the control points P1(x1,y1) and P2(x2,y2) of the third-order Bézier curve as the line connecting the starting point P0(x0,y0) and the ending point P3(x3,y3) are calculated as follows:

[0041]

[0042] Where k0 is the slope of the line connecting the start and end points of the Bézier curve; x0 is the x-coordinate of the start point of the Bézier curve; x3 is the x-coordinate of the end point of the Bézier curve; y0 is the y-coordinate of the start point of the Bézier curve; y3 is the y-coordinate of the end point of the Bézier curve; y1 is the y-coordinate of the first control point; y2 is the y-coordinate of the second control point; x1 is the x-coordinate of the first control point; x2 is the x-coordinate of the second control point.

[0043] The second-order Bézier curve has only one control point P1(x1,y1) and no intermediate control point P2(x2,y2). Therefore, let the first shape feature Δy1 be equal to the second shape feature Δy2.

[0044] Preferably, the method for obtaining the third shape feature P′0 and the fourth shape feature P′3 in step S21 is as follows:

[0045] Calculate the third shape feature P′0 and the fourth shape feature P′3 of the Bézier curve corresponding to the starting point P0(x0,y0) and the ending point P3(x3,y3) of each Bézier curve;

[0046] The third shape feature P′0 and the fourth shape feature P′3 of a cubic Bézier curve are:

[0047]

[0048] Where P0 is the starting point of the Bézier curve; P1 is the first control point; P3 is the ending point of the Bézier curve; and P2 is the second control point.

[0049] The third shape feature P′0 and the fourth shape feature P′3 of the quadratic Bézier curve are:

[0050]

[0051] Preferably, the LASSO regression prediction set X in step S23 输入 The method for obtaining it is as follows:

[0052] The second prediction parameter β j With the corresponding drag coefficient characteristic value x zj Multiply to construct the LASSO regression prediction set X 输入 ,for:

[0053] X 输入 =x zj β j ;

[0054] Among them, X 输入 This is the LASSO regression prediction set.

[0055] Preferably, the structure of the LSTM deep learning model in step S3 is as follows:

[0056] The structure of the LSTM deep learning model consists of a forget gate, an output gate, and an input gate in each layer; the forget gate enables the transfer between each layer, thereby enhancing the training effect of the model; the last LSTM structure is connected to the fully connected layer.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] This invention uses car body side profile data as input and drag coefficient as output to optimize the design of car body side profile. It solves the problem of indistinct car profile features, making feature extraction difficult with limited data. This improves the model's understanding of the car profile, reduces the error value in predicting the car body side profile, and uses LASSO regression parameters as penalty parameters to correct features, avoiding overfitting. This enables rapid and automated prediction and calculation of car body side drag profile. Compared to traditional methods, this invention allows for real-time observation of drag coefficient changes during the design process, improving design efficiency. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method for predicting the drag coefficient of the side profile of a passenger car according to the present invention.

[0060] Figure 2 This is a side profile view of the car body of the present invention;

[0061] Figure 3 This is a side profile drawing of the car body according to the present invention;

[0062] Figure 4 This is a diagram illustrating the method for extracting positional features from Bézier curves according to the present invention.

[0063] Figure 5 This is a diagram illustrating the method for extracting shape features from Bézier curves according to the present invention.

[0064] Figure 6 This is a flowchart of the dataset input network model and preprocessing process for the present invention.

[0065] Figure 7 The flowchart of the prediction process from inputting the prediction object into the model is shown below.

[0066] Figure 8 This is a side view of a car body, an example used in this invention.

[0067] Figure 9 This is a side profile view of a car body, used as an example in this invention.

[0068] Figure 10 This is another example of a car body side view used in this invention;

[0069] Figure 11 This is another example of a car body side profile view used in this invention. Detailed Implementation

[0070] Exemplary embodiments, features, and aspects of the present invention will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0071] This invention uses the exterior structures of two different sedans for analysis. Example 1 shows the side profile of one sedan body, and Example 2 shows the side profile of another sedan body. The method for predicting the drag coefficient of the sedan body side profile provided by this invention is as follows: Figure 1 As shown, it includes:

[0072] Step S1: Obtain the side profile data parameters of the car body and determine the Bezier curve of the side profile of the car body;

[0073] Figure 2 This image shows the side profile of a car, illustrating the distribution of the nine Bézier curves that make up the car's silhouette. All Bézier curves were created using 2D modeling; the original data was not an image. Figure 3 This is a side profile diagram of the car body of the present invention, showing the starting point P0 and the ending point P3 of the overall profile, with intermediate control points P1 and P2.

[0074] The side profile data parameters of the car body were obtained from the car parameter database. A total of 53 data parameters were found for a car profile, including one drag coefficient Cd and 52 coordinate points that make up the car profile, including the x-coordinate. i and the vertical coordinate y i , i = 1, 2, ..., 26; the Bézier curves of the side profile of the car body include the quadratic Bézier curve B2(t) and the cubic Bézier curve B3(t), and the specific calculation formula is as follows:

[0075]

[0076] Where B3(t) is the cubic Bézier curve formula; B2(t) is the quadratic Bézier curve formula; P 30 P is the first coefficient of the cubic Bézier curve; 31 P is the second coefficient of the cubic Bézier curve; 32 P is the third coefficient of the cubic Bézier curve; 33 P is the fourth coefficient of the cubic Bézier curve. 20 P is the first coefficient of the quadratic Bézier curve; 21 P is the second coefficient of the quadratic Bézier curve; 23 t represents the third coefficient of the quadratic Bézier curve; t is the input to the Bézier curve.

[0077] Step S2: Extract the drag coefficient feature value of the Bezier curve of the car body side profile and use the feature parameters obtained by LASSO regression.

[0078] Step S21: Extract the drag coefficient feature value from the Bézier curve car profile obtained in Step S1; extract features from the 26 coordinate points mentioned above, and calculate the feature value of each Bézier curve based on the coordinates of the 26 points. A total of 9 Bézier curves are calculated, which are 7 cubic Bézier curves and 2 quadratic Bézier curves.

[0079] like Figure 4This is a diagram illustrating the method for extracting positional features from Bézier curves according to the present invention; it extracts the first positional feature of each Bézier curve that makes up the side profile of a car body. The second positional feature d and the third positional feature sinθ.

[0080] like Figure 5 This diagram illustrates the method for extracting shape features from Bézier curves according to the present invention. It extracts the first shape feature Δy1, the second shape feature Δy2, the third shape feature P′0, and the fourth shape feature P′3 for each Bézier curve constituting the side profile of a car body. After feature extraction, the Bézier curve (B3(t) or B2(t)) is converted into nine variables, namely: position features (…). d and sinθ) are 3 features, and shape features (Δy1, Δy2, P′0, and P′3) are 6 features. For shape features P′0 and P′3, each is 2 features.

[0081] First positional features The method for obtaining this is as follows: calculate the starting point (x) of each Bézier curve. n-1 ,y n-1 ) and endpoint (x) n ,y n The slope of ) compared to k n The starting point (x) of the previous Bézier curve n-1 ,y n-1 ) and endpoint (x) n ,y n ) connecting line k (n-1) First positional feature The method for obtaining it is as follows:

[0082]

[0083] in, The first positional feature; k n The line connecting the start and end points of the current Bézier curve; k (n-1) y is the line connecting the start and end points of the previous Bézier curve; n y is the ordinate of the endpoint of the Bézier curve; (n-1) x represents the ordinate of the starting point of the Bézier curve; n x is the x-coordinate of the endpoint of the Bézier curve; (n-1) is the x-coordinate of the starting point of the Bézier curve.

[0084] The method for obtaining the second positional feature d is as follows: calculate the starting point (x) of each Bézier curve. n-1 ,y n-1 ) and endpoint (x) n ,y n The distance d between the lines connecting the two sides is:

[0085]

[0086] Where d is the second positional feature.

[0087] The method for obtaining the third positional feature sinθ is as follows: calculate the starting point (x) of each Bézier curve. n-1 ,y n-1 ) and endpoint (x) n ,y n The sine of the acute angle θ formed by the line connecting the two lines and the horizontal line, sinθ, is:

[0088]

[0089] Here, sinθ is the third position feature.

[0090] For the starting point (x) of a portion of the Bézier curve n-1 ,y n-1 ) and endpoint (x) n ,y n The acute angle θ formed by the line connecting the two points and the horizontal line may have a case where the third position feature sinθ does not exist; in this case, the value of the third position feature sinθ is set to 1.

[0091] The method for obtaining the first shape feature Δy1 and the second shape feature Δy2 is as follows: The constituent points of the third-order Bézier curve are: starting point P0(x0,y0), ending point P3(x3,y3), first control point P1(x1,y1), and second control point P2(x2,y2); the first shape feature Δy1 and the second shape feature Δy2 corresponding to the same abscissa of the control points P1(x1,y1) and P2(x2,y2) of the third-order Bézier curve as the line connecting the starting point P0(x0,y0) and the ending point P3(x3,y3) are:

[0092]

[0093] Where k0 is the line connecting the start and end points of the Bézier curve; x0 is the x-coordinate of the start point of the Bézier curve; x3 is the x-coordinate of the end point of the Bézier curve; y0 is the y-coordinate of the start point of the Bézier curve; y3 is the y-coordinate of the end point of the Bézier curve; y1 is the y-coordinate of the first control point; y2 is the y-coordinate of the second control point; x1 is the x-coordinate of the first control point; and x2 is the x-coordinate of the second control point.

[0094] The second-order Bézier curve has only one control point P1(x1,y1) and no intermediate control point P2(x2,y2). Therefore, let the first shape feature Δy1 be equal to the second shape feature Δy2.

[0095] The method for obtaining the third shape feature P′0 and the fourth shape feature P′3 is as follows: calculate the third shape feature P′0 and the fourth shape feature P′3 of the Bézier curve corresponding to the starting point P0(x0,y0) and the ending point P3(x3,y3) of each Bézier curve.

[0096] The third shape feature P′0 and the fourth shape feature P′3 of a cubic Bézier curve are:

[0097]

[0098] Where P0 is the starting point of the Bézier curve; P1 is the first control point; P3 is the ending point of the Bézier curve; and P2 is the second control point.

[0099] The third shape feature P′0 and the fourth shape feature P′3 of the quadratic Bézier curve are:

[0100]

[0101] Constructing the drag coefficient characteristic value X zu ,for:

[0102]

[0103] Among them, X zu The characteristic value of the drag coefficient; x zj Let j be the characteristic value of the drag coefficient; d is the first position feature; d is the second position feature; sinθ is the third position feature; Δy1 is the first shape feature; Δy2 is the second shape feature; P′0 is the third shape feature; P′3 is the fourth shape feature.

[0104] Step S22: The drag coefficient feature values ​​extracted in step S21 and their corresponding drag coefficients Cd determined in step 1 are divided into training and testing sets. The drag coefficient feature values ​​obtained from 1000 vehicle body data are a matrix of [1000, 82], with each row corresponding to a feature value of one vehicle body. For this invention example, the 1000 basic data are divided into 960 training sets and 40 testing sets. After the division, the training set is a [960, 82] matrix, and the testing set is a [40, 81] matrix. In particular, for using more data than the experimental data of this invention, adjustments can be made based on the test results. Figure 6 The flowchart shows the input network model and preprocessing process for the dataset of this invention, and the final flowchart shows the input of the total data into the model.

[0105] Step S23: Perform LASSO regression on the training set from Step S22, and input the entire set as a matrix [960, 82]. The last column of this matrix represents the drag coefficient value for each vehicle body. Input the values ​​into LASSO regression line by line. Obtain the LASSO regression prediction set X. 输入 The formula for calculating LASSO regression is:

[0106]

[0107] Where ta≥0 is the harmonic parameter; x ij Y is the first input parameter; i This is the second input parameter; The parameter set obtained from LASSO regression; α i β is the first prediction parameter. j This is the second prediction parameter.

[0108] LASSO regression prediction set X 输入 The method for obtaining the second prediction parameter β is as follows: j With the corresponding drag coefficient characteristic value x zj Multiply to construct the LASSO regression prediction set X 输入 ,for:

[0109] X 输入 =x zj β j ;

[0110] Among them, X 输入 This is the LASSO regression prediction set. β j The parameters obtained are a matrix of size [1, 81], from which the final X is obtained. 输入 The value is [960, 81]. Table 1 shows the X value for a certain sedan model. 输入 Feature value excerpt table; as shown in Table 2, another sedan part X 输入 Eigenvalue selection table.

[0111] Table 1. X section of a certain car model 输入 Eigenvalue Excerpt

[0112] 0.052415228 0.055105538 0 0.004822765 -0.000561562 0.001253243 0 -0.001543843 0

[0113] Table 2. Another sedan model, part X 输入 Eigenvalue Excerpt

[0114]

[0115]

[0116] Step S3: Train the feature parameters using the LSTM deep learning model; obtain the feature parameters X from step S2.输入 The input is used for training in an LSTM deep learning model; the loss function used for each LSTM layer is MSLE, i.e., mean squared logarithmic error, which is:

[0117]

[0118] in, This is a predicted value; The true value is represented by L; the MSLE loss function is represented by n. 总 This represents the total number of feature parameters.

[0119] The LSTM deep learning model structure consists of a forget gate, an output gate, and an input gate in each layer. The forget gate enables data transfer between layers, thus enhancing the training effect. The final LSTM layer is connected to a fully connected layer. The LSTM deep learning model network has 5 LSTM layers and 1 fully connected layer, with data input in a [9,9,300] 3D matrix. The 81 feature values ​​of a car are divided into a [9,9] matrix according to the number of curves forming the car's contour. The entire training set of 960 data points is then divided into groups of 300 according to predefined parameters and input sequentially. This process trains 10 LSTM deep learning models.

[0120] Step S4: Construct a car body side profile prediction model as a LASSO-LSTM drag coefficient prediction model to achieve drag coefficient prediction.

[0121] like Figure 7 The flowchart of the prediction process for the prediction object of this invention is as follows: The LASSO regression prediction set obtained in step S2 and the LSTM deep learning model constructed in step S3 form the LASSO-LSTM drag coefficient prediction model; To ensure the accuracy of the prediction results, 10 trained LASSO-LSTM drag coefficient prediction models are used for prediction, and the average of the 10 predicted values ​​is processed as the prediction result of the car drag coefficient.

[0122] This invention also conducted a preliminary analysis of the side profile of the example sedan to verify the method's validity: the predicted example 1 is the side profile of a certain sedan body, such as... Figure 8 Example 1 is a side view of a car body, used as an example of the present invention. Example 2 shows the side profile of another car body, as shown below. Figure 10 This is a side view of a sedan body, used as an example in this invention. The corresponding side profile of the sedan body is drawn from an image. After drawing, the outline is scaled according to the normal car length ratio, and its length is scaled to the actual length of the sedan according to the coordinate system, as shown below. Figure 9 This is an example of the side profile of a car body used in this invention; for example... Figure 11The image shown is a side profile of a sedan body, used as an example in this invention. The final side profile result of the sedan body calculated using this method, along with the percentage error, is shown in Table 3 below:

[0123] Table 3. Results and Percentage of Error for Sedan Body Side Profile

[0124] True drag coefficient 0.28 0.29 Predicted drag coefficient 0.2867 0.2743 error -0.0067 0.0157 Error percentage 2.36％ 5.41％

[0125] The data in the table above leads to the conclusion that, since the actual drag coefficient is the solid drag coefficient of the vehicle, and the car's outline does not consider its three-dimensional features, errors exist. Secondly, compared to existing known drag coefficients, the predicted drag coefficient has higher accuracy, which also introduces errors. For these reasons, this invention aims to improve prediction performance and is best suited for the initial design outline stage of vehicle body design. At this stage, compared to traditional CAE methods, this invention improves prediction speed while maintaining a certain level of prediction accuracy.

[0126] The beneficial effects of this invention are as follows: This invention provides a method for predicting the drag coefficient of a car body side profile. Using car body side profile data as input and the drag coefficient as output, it optimizes the design of the car body side profile. It solves the problem of unclear car profile features and the inability to extract features when data is limited, improving the model's understanding of the car profile and reducing the error value in predicting the car body side profile. Using LASSO regression parameters as penalty parameters to correct features avoids overfitting, achieving rapid and automated prediction and calculation of the car body side drag profile. Compared to traditional methods, this invention allows for real-time observation of drag coefficient changes during the design process, improving design efficiency.

[0127] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting the drag coefficient of a car body side profile, characterized in that, It includes: S1: Obtain the side profile data parameters of the car body and determine the Bezier curve of the side profile of the car body; The side profile data parameters of the car body are obtained from the car parameter database. The car profile is composed of... It consists of several data parameters, one of which is the drag coefficient. , The data parameters constitute the outline of the car. A coordinate point, including the x-coordinate. and ordinate , The Bézier curve of the car body side profile includes a quadratic Bézier curve. and cubic Bézier curve The specific calculation formula is as follows: ； in, The formula for a cubic Bézier curve; The formula for a quadratic Bézier curve; The first coefficient of the cubic Bézier curve; The second coefficient of the cubic Bézier curve; The third coefficient of the cubic Bézier curve; The fourth coefficient of the cubic Bézier curve; The first coefficient of the quadratic Bézier curve; The second coefficient of the quadratic Bézier curve; The third coefficient of the quadratic Bézier curve; Input for Bézier curves; S2: Extract the drag coefficient characteristic value of the Bezier curve of the side profile of the car body, and use the regression method to obtain the characteristic parameters; S21: Extract the drag coefficient feature value from the Bézier curve car profile obtained in step S1; extract the first position feature of each Bézier curve that makes up the side profile of the car body. Second positional features and third position features Extract the first shape features of each Bézier curve that makes up the side profile of the car body. Second shape features Third shape feature and fourth shape features Constructing the characteristic value of the drag coefficient ,for: ； in, This is the characteristic value of the drag coefficient; Let j be the characteristic value of the drag coefficient; This is the first positional feature; This is a second positional feature; This is a third positional feature; This is the first shape feature; This is a second shape feature; It is a third shape feature; This is the fourth shape feature; S22: Combine the drag coefficient characteristic values ​​extracted in step S21 with the corresponding drag coefficients determined in step S1. Divided into training set and test set; S23: Perform LASSO regression on the training set from step S22 to obtain the LASSO regression prediction set. The formula for calculating LASSO regression is: ； in, For harmonic parameters; This is the first input parameter; This is the second input parameter; This is the set of parameters obtained from LASSO regression; The first prediction parameter; This is the second prediction parameter; S3: Train feature parameters using an LSTM deep learning model; Obtain the feature parameters from step S2 and input them into the LSTM deep learning model for training; the loss function used for each LSTM layer is MSLE, i.e., mean squared logarithmic error, which is: ； in, This is a predicted value; The actual value; The MSLE loss function; This represents the total number of feature parameters. S4: Construct a car body side profile prediction model as a LASSO-LSTM drag coefficient prediction model to achieve drag coefficient prediction; The LASSO regression prediction set obtained in step S2 and the LSTM deep learning model constructed in step S3 form the LASSO-LSTM drag coefficient prediction model. To ensure the accuracy of the prediction results, m trained LASSO-LSTM drag coefficient prediction models are used for prediction. The average of the m predicted values ​​is then used as the prediction result of the car's drag coefficient.

2. The method for predicting the drag coefficient of a car body side profile according to claim 1, characterized in that: The first position feature in step S21 The method for obtaining it is as follows: ； in, This is the first positional feature; The slope of the line connecting the start and end points of the current Bézier curve; The slope of the line connecting the start and end points of the previous Bézier curve; y is the ordinate of the endpoint of the Bézier curve; y is the ordinate of the starting point of the Bézier curve; is the x-coordinate of the endpoint of the Bézier curve; is the x-coordinate of the starting point of the Bézier curve.

3. The method for predicting the drag coefficient of a car body side profile according to claim 2, characterized in that: The second position feature in step S21 The method for obtaining it is as follows: Calculate the starting point of each Bézier curve. and end point The distance of the line connecting them ,for: ； in, This is the second positional feature.

4. The method for predicting the drag coefficient of a car body side profile according to claim 3, characterized in that: The third position feature in step S21 The method for obtaining it is as follows: Calculate the starting point of each Bézier curve and end point The acute angle between the line connecting the two lines and the horizontal line. sine value ,for: ； in, This is a third positional feature; For the starting point of some Bézier curves and end point The acute angle between the line connecting the two lines and the horizontal line. There is a third position feature The case where it does not exist; in this case, the third position feature will be used. The value is set to 1.

5. The method for predicting the drag coefficient of a car body side profile according to claim 1, characterized in that: The first shape feature in step S21 Second shape features The method for obtaining it is as follows: The constituent points of a third-order Bézier curve are: the starting point. End point First control point Second control point First shape feature Second shape features for: ； in, The slope of the line connecting the start and end points of the Bézier curve; The x-coordinate of the starting point of the Bézier curve; The x-coordinate of the endpoint of the Bézier curve; The ordinate of the starting point of the Bézier curve; The ordinate of the endpoint of the Bézier curve; The ordinate of the first control point; The ordinate of the second control point; The x-coordinate of the first control point; The x-coordinate of the second control point; A second-order Bézier curve has only one control point. There are no intermediate control points. Therefore, let the first shape feature Equal to the second shape feature .

6. The method for predicting the drag coefficient of a car body side profile according to claim 3, characterized in that: The third shape feature in step S21 and fourth shape features The method for obtaining it is as follows: Calculate the starting point of each Bézier curve and end point The third shape feature of the corresponding Bézier curve and fourth shape features ; The third shape characteristic of cubic Bézier curves and fourth shape features ,for: ； in, This is the starting point of the Bézier curve; This is the first control point; This is the endpoint of the Bézier curve; This is the second control point; The third shape characteristic of a quadratic Bézier curve and fourth shape features for: 。 7. The method for predicting the drag coefficient of a car body side profile according to claim 1, characterized in that: LASSO regression prediction set in step S23 The method for obtaining it is as follows: The second prediction parameter With the corresponding drag coefficient characteristic value Multiply to construct the LASSO regression prediction set. ,for: ； in, This is the LASSO regression prediction set.

8. The method for predicting the drag coefficient of a car body side profile according to claim 1, characterized in that: The specific structure of the LSTM deep learning model in step S3 is as follows: The structure of the LSTM deep learning model includes a forget gate, an output gate, and an input gate in each layer; the forget gate enables the transmission between each layer, and the last LSTM structure is connected to the fully connected layer.

Images