Tire and whole vehicle matching quantitative evaluation method

Abstract

The application discloses a kind of tire and whole vehicle matching quantitative evaluation method, first carry out objective evaluation test, obtain whole vehicle objective index and tire objective index;Next, subjective evaluation test is carried out, and subjective area score is obtained;Then the objective index is processed with feature dimension reduction, then the objective index after dimension reduction and subjective area score are input into subjective-objective fusion generative adversarial network to carry out data enhancement, and matching enhancement dataset is obtained, finally, the matching enhancement dataset is used to train subjective-objective fusion combination model to obtain the quantitative evaluation result of tire and whole vehicle matching.The application is helpful to provide accurate evaluation and guidance for tire selection, and shorten tire matching option cycle.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automobile subjective and objective fusion and dynamics matching, and particularly relates to a tire and vehicle matching quantitative evaluation method. BACKGROUND

[0002] As the only component for transmitting force and torque between a vehicle and a road surface, the dynamics performance of a tire directly affects the driving performance of the vehicle. The matching of the tire and the vehicle dynamics has always been an extremely important part of the positive development process of an enterprise during the development of the vehicle. Whether the tire and the vehicle system are matched or not has a great influence on the performance of the tire and other systems of the vehicle.

[0003] At present, the matching of the tire and the chassis of the vehicle during the development of the vehicle mainly relies on subjective evaluation. However, in the process of subjective evaluation, the advantages and disadvantages of the vehicle handling stability are evaluated by the feeling of the driver to the dynamic response of the vehicle, and such feeling has no direct correlation with the vehicle parameters, resulting in high evaluation cost and long product design cycle during the product development process. In addition, the driver is affected by some instability factors, so that the repeatability of the evaluation is not high, and sometimes the evaluation results need to be corrected repeatedly for many times.

[0004] Therefore, the subjective evaluation is not the optimal choice for the development of the chassis in terms of time, money or repeatability. Compared with the subjective evaluation, the quantifiable objective evaluation has considerable advantages in terms of evaluation cost, simplicity of test method, acquisition of test data and repeatability of test.

[0005] It is particularly important and urgent to extract the vehicle dynamics parameters from the subjective evaluation, to establish the correlation between the vehicle dynamics parameters and the dynamics parameters of the tire, and to build an evaluation method of the subjective evaluation and the dynamics parameters of the tire. The establishment of the system can pre-position the test nodes of the matching of the tire and the chassis, improve the matching accuracy, and greatly shorten the period of selection of the matching of the tire.

[0006] A vehicle chassis controller parameter optimization method for subjective and objective fusion evaluation is disclosed in Chinese Patent No. CN116611228A. The subjective evaluation index and the objective measured data index are fused, and the optimal control parameters under the subjective and objective fusion evaluation index are realized through an automatic optimization method, while the workload of calibration and debugging is greatly reduced. However, this method is for solving control parameters, and it is difficult to be directly applied to the field of automobile subjective evaluation. SUMMARY

[0007] The present application provides a tire and whole vehicle matching quantitative evaluation method to provide accurate evaluation and guidance for tire selection and promote the development of tire design and optimization.

[0008] The present application is characterized in that the following technical solutions are adopted to achieve the object of the application:

[0009] The present application provides a tire and whole vehicle matching quantitative evaluation method, which comprises the following steps:

[0010] Step 1, collecting objective indexes and subjective scores of tires and whole vehicles;

[0011] Step 1.1: obtaining a whole vehicle objective index data set X = {X1, X2,..., X n ,...,X N} composed of objective indexes of a whole vehicle under N test conditions; wherein X n represents the objective index data set of the whole vehicle under the nth test condition, and represents the mth objective index of the whole vehicle under the nth test condition; M represents the number of objective indexes of the whole vehicle, and N represents the number of test conditions;

[0012] Step 1.2: obtaining a tire objective index data set T = {T1, T2,..., T j ,...,T J} composed of objective indexes of a tire under J tire performance test conditions; wherein T j represents the objective index data set of the tire under the jth tire performance test condition, and represents the kth objective index of the tire under the jth tire performance test condition; K represents the number of objective indexes of the tire; and J represents the number of tire performance test conditions;

[0013] Step 1.3: obtaining subjective scores Y = {Y1, Y2,..., Y u ,...,Y U} of U driving experience areas, wherein Y u represents the subjective score of the uth driving experience area; and U represents the number of driving experience areas;

[0014] Step 2: reducing the dimensions of X and T to obtain a reduced whole vehicle objective index data set X' and a reduced tire objective index data set T';

[0015] Step 3: using a subjective and objective fusion generative adversarial network to perform data enhancement on X', T' and Y to obtain an enhanced whole vehicle matching enhanced data set an enhanced tire performance matching enhanced data set and enhanced driving experience area matching enhancement score

[0016] Step 4: training of the subjective and objective fusion combination model and quantitative evaluation result

[0017] Step 4.1: taking as the training sample, taking as the label, training the subjective and objective fusion combination model to obtain the trained subjective and objective fusion combination model

[0018] Step 4.2: taking X and T as the verification sample, taking Y as the verification true value, inputting the trained subjective and objective fusion combination model for processing to obtain the classification score C={C1, C2,..., C u ,...,C U} and the regression score R={R1, R2,..., R u ,...,R U} of the driving experience area, wherein C u represents the classification score of the u-th driving experience area; R u represents the regression score of the u-th driving experience area

[0019] Step 4.3: calculating the difference diff={diff1, diff2,..., diff u ,...,diff U} between the classification score C and the regression score R by using formula (5)

[0020] diff u =C u -R u (5)

[0021] In formula (5): diff u represents the difference between C u and R u ;

[0022] Step 4.4: setting the correction index as δ, the minimum classification scale value as i, and using formula (6) to correct the rating of C to obtain the quantitative evaluation result C'={C'1, C'2,..., C' u ,...,C' U} of the tire and vehicle matching

[0023]

[0024] In formula (6): C' u is the quantitative evaluation result of C u after rating correction.

[0025] The tire and whole vehicle matching quantitative evaluation method has the characteristics that step 2 comprises the following steps:

[0026] Step 2.1: calculating the whole vehicle objective index correlation coefficient p of the whole vehicle objective index data set X by using formula (1) X :

[0027]

[0028] In formula (1), COV is the covariance, is the standard deviation of X , and is the standard deviation of X n .

[0029] Step 2.2: calculating the tire objective index correlation coefficient p of the tire objective index data set T by using formula (2) T :

[0030]

[0031] In formula (2), COV is the covariance, is the standard deviation of T , and is the standard deviation of T j .

[0032] Step 2.3: setting a threshold value Δ, and performing dimension reduction processing on the objective indexes greater than the threshold value Δ in p X and p T , respectively, to obtain the dimension-reduced whole vehicle objective index data set X' = {X'1, X'2,..., X' n} and the dimension-reduced tire objective index data set T' = {T'1, T'2,..., T' j}, wherein X' n represents the dimension-reduced whole vehicle objective index data set under the nth test working condition, and T' j represents the dimension-reduced tire objective index data set under the jth tire performance test working condition. n N j J n j

[0033] The step 3 comprises the following steps:

[0034] Step 3.1: defining the whole vehicle random noise tire performance random noise and driving experience area random noise wherein, is the random noise of the Gaussian distribution with the same dimension as X' n, is the random noise of the Gaussian distribution with the same dimension as T' j. n j ​​​​​​​Random noise with the same Gaussian distribution of the same dimension. Is with Y u Random noise with a Gaussian distribution and the same range of values;

[0035] Step 3.2: Transfer Q X′ Q T′ and Q Y The random noise z, respectively, is input to the generator G of the subjective-objective fusion adversarial generative network for processing, thereby utilizing the loss function L of the generator G defined in equation (3). G The generator G is trained by backpropagation to obtain the trained generator G. * Output the whole vehicle matching enhancement dataset. Tire performance matching enhancement dataset Enhanced score matching driving experience area

[0036] L G =H(1,D(G(z))) (3)

[0037] In equation (3): G is the generator, D is the discriminator, and H is the cross-entropy;

[0038] Step 3.3: Use the vehicle objective index dataset X, the tire objective index dataset T, and the subjective score Y as ground truth samples r, respectively, and the vehicle matching enhancement dataset. Tire performance matching enhancement dataset Enhanced score matching driving experience area These are respectively used as input data for the discriminator D in the subjective-objective fusion adversarial generative network. and the true sample r and its corresponding input data The data are input together into the discriminator D of the subjective-objective fusion adversarial generative network for processing, thereby utilizing the loss function L of the discriminator D defined by equation (4). D The discriminator D is obtained by performing discrimination. * Output the enhanced vehicle matching dataset. Enhanced tire performance matching dataset Enhanced driving experience area matching and improved score

[0039]

[0040] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the tire and vehicle matching quantitative evaluation method, and the processor is configured to execute the program stored in the memory.

[0041] The application is a computer readable storage medium, and a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to perform the steps of the tire and whole vehicle matching quantitative evaluation method.

[0042] Compared with the prior art, the application has the beneficial effects that:

[0043] 1. The application adopts the idea of constructing feature engineering in machine learning to perform feature dimension reduction on a large number of objective indexes, and removes objective indexes with high correlation coefficients between regions, which helps to improve the efficiency of the subsequent evaluation system.

[0044] 2. The application adopts the subjective and objective fusion generative adversarial network to perform data enhancement on real vehicle data, solving the problem of difficult acquisition of test data in subjective and objective evaluation.

[0045] 3. The application adopts the deep learning method, and the attention mechanism is used to fuse the objective indexes and the subjective score, and a subjective and objective fusion combination model is used to comprehensively consider the prediction results of the regression model and the classification model, thereby improving the overall prediction performance and robustness, which not only improves the precision of tire matching selection, but also greatly shortens the period of tire matching selection. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The application is a tire and whole vehicle matching quantitative evaluation method flowchart.

[0047] Figure 2 The application is a principle diagram of the subjective and objective fusion generative adversarial network in the method.

[0048] Figure 3 The application is a combination principle diagram of the subjective and objective fusion model in the method.

[0049] Figure 4 The application is a center feeling prediction diagram output by the subjective and objective fusion combination model in the method.

[0050] Figure 5 The application is a steering force feedback prediction diagram output by the subjective and objective fusion combination model in the method.

[0051] Figure 6 The application is a steering response prediction diagram output by the subjective and objective fusion combination model in the method.

[0052] Figure 7 The application is a linear feeling prediction diagram output by the subjective and objective fusion combination model in the method.

[0053] Figure 8 The application is a lateral grip force prediction diagram output by the subjective and objective fusion combination model in the method.

[0054] Figure 9The method of this invention outputs a straight-line driving stability prediction map from the subjective-objective fusion combined model.

[0055] Figure 10 The lane change stability prediction map is output by the subjective-objective fusion model in the method of this invention.

[0056] Figure 11 This is the steady-state rotation capability prediction diagram output by the subjective-objective fusion model in the method of this invention. Detailed Implementation

[0057] In this embodiment, as Figure 1 As shown, a quantitative evaluation method for tire-vehicle matching aims to address the problem that existing subjective and objective evaluation methods are disconnected and their relationship is ambiguous. The method includes the following steps:

[0058] Step 1: Collect objective indicators and subjective scores for the tires and the whole vehicle;

[0059] Step 1.1: Conduct objective evaluation tests according to the national standard GB / T 6323-2014 "Test Method for Handling Stability of Automobiles", and obtain a vehicle objective index dataset X = {X1, X2, ..., X...} consisting of the vehicle's objective indexes under N=5 test conditions. n ,...,X N}; where X n Let represent the dataset of objective vehicle performance indicators under the nth test condition, and This represents the m-th objective vehicle index under the n-th test condition; M represents the number of objective vehicle indexes, and N represents the number of test conditions. In practice, the five test conditions are: slow incremental test, sinusoidal hysteresis test, central area handling stability test, step input test, and steady-state turning test. 33 items are collected in the slow incremental test; 6 items are collected in the sinusoidal hysteresis test; 6 items are collected in the central area handling stability test; 2 items are collected in the step input test; and 9 items are collected in the steady-state turning test, for a total of 56 objective vehicle indexes.

[0060] Step 1.2: Conduct tire performance tests to obtain a dataset of objective tire indicators T = {T1, T2, ..., Tj, ..., Tj}, consisting of objective tire indicators under J tire performance test conditions. J}; where T j Let represent the dataset of objective tire performance indicators under the j-th tire performance test condition, and This represents the k-th objective tire index under the j-th tire performance test condition; K represents the number of objective tire indexes; J represents the number of tire performance test conditions; in specific implementation, the 6 test conditions include torsional stiffness test, dynamic stiffness test, vertical stiffness test, lateral stiffness test, longitudinal stiffness test, and tire imprint test; the objective tire indexes are collected in the following ways: 3 items in the torsional stiffness test; 4 items in the dynamic stiffness test; 2 items in the vertical stiffness test; 3 items in the lateral stiffness test; 3 items in the longitudinal stiffness test; and 12 items in the tire imprint test, for a total of 27 objective tire indexes.

[0061] Step 1.3: Conduct a subjective evaluation test according to the group standard CSAE278-2022 "Subjective Evaluation Method for Dry Handling and Comfort of Passenger Car Tire Performance", and obtain subjective scores Y = {Y1, Y2, ..., Y} for 8 driving experience areas. u ,...,Y U}, where Y u The subjective score for the u-th driving experience area is represented by 'u'; U represents the number of driving experience areas; the subjective evaluation includes a total of 8 driving experience areas, namely center zone feeling, steering force feedback, steering response, linearity, lateral grip, straight-line driving stability, lane change stability, and steady-state turning ability, and subjective area scores are obtained.

[0062] Step 2: Dimensionality reduction of objective indicators;

[0063] Step 2.1: Calculate the correlation coefficient p of the vehicle objective index dataset X using equation (1). X :

[0064]

[0065] In equation (1): COV is the covariance. for standard deviation For X n Standard deviation;

[0066] Step 2.2: Calculate the correlation coefficient p of the tire objective indexes of the tire objective index dataset T using equation (2). T :

[0067]

[0068] In formula (2): for standard deviation For T j The standard deviation.

[0069] Step 2.3: Set the threshold Δ = 0.7, and apply it to p respectively.X and p T Objective indicators with values ​​greater than a threshold Δ are subjected to dimensionality reduction processing. Highly correlated objective indicators are removed, and feature space dimensionality reduction is performed to obtain the dimensionality-reduced dataset of vehicle objective indicators X′={X′1,X′2,...,X′...} n ,...,X′ N}, Tire objective index dataset T′={T′1,T′2,...,T′ j ,...,T′ J}, where X′ n Let T′ represent the dataset of objective vehicle indicators after dimensionality reduction for the nth test condition. j Let represent the tire objective index dataset after dimensionality reduction for the j-th tire performance test condition; the objective indices corresponding to each subjective region after dimensionality reduction are as follows:

[0070] Among them, the objective indicators corresponding to the central area perception are:

[0071] Steering system friction torque, steering wheel torque gradient at lateral acceleration of 0.1g, steering hysteresis, steering wheel torque corresponding to lateral acceleration of 2m / s^2, steering wheel torque linearity (0.05-0.3g), steering wheel angle corresponding to lateral acceleration of 1m / s^2, steering wheel angle corresponding to lateral acceleration of 2m / s^2, steering wheel angle linearity (0.05-0.3g), Mz-SA slope under 40% load, and lateral stiffness value under 40% load.

[0072] The objective indicators corresponding to steering force feedback are:

[0073] Steering wheel torque gradient at lateral acceleration of 0g, steering wheel torque gradient at lateral acceleration of 0.1g, steering wheel torque corresponding to lateral acceleration of 8m / s^2, steering wheel torque linearity (0.5-0.8g), steering wheel torque linearity (0.05-0.3g), Mz-SA slope under 40% load, and lateral stiffness value under 40% load.

[0074] The objective indicators corresponding to steering response are:

[0075] Peak lateral acceleration, steering wheel angle corresponding to lateral acceleration of 2 m / s², steering wheel angle corresponding to lateral acceleration of 5 m / s², steering wheel angle corresponding to lateral acceleration of 7 m / s², steering wheel angle corresponding to lateral acceleration of 8 m / s², yaw rate corresponding to lateral acceleration of 8 m / s², yaw rate linearity (0.05-0.3g), understeer gradient at 0.2g, understeer gradient at 0.4g, understeer gradient at 0.6g, understeer gradient at 0.8g, Mz-SA slope under 40% load, lateral stiffness value under 40% load.

[0076] wherein the linear feeling corresponds to the objective indicators of:

[0077] steering wheel angle linearity (0.05-0.3g), yaw rate linearity (0.5-0.8g), yaw rate linearity (0.05-0.3g), understeer gradient linearity, Mz-SA slope at 40% load, lateral stiffness value at 40% load.

[0078] wherein the lateral grip corresponds to the objective indicators of:

[0079] peak yaw rate, steering wheel angle corresponding to lateral acceleration of 8 m / s^2, yaw rate linearity (0.5-0.8g), maximum lateral acceleration, lateral acceleration corresponding to peak yaw rate, peak yaw rate, pattern saturation at 6° camber angle at 40% load, pattern saturation at 6° camber angle at 80% load, pattern saturation at 6° camber angle at 120% load.

[0080] wherein the straight-line stability corresponds to the objective indicators of:

[0081] steering system friction torque, steering wheel torque gradient at 0g lateral acceleration, steering wheel torque gradient at 0.1g lateral acceleration, Mz-SA slope at 40% load, lateral stiffness value at 40% load.

[0082] wherein the lane change stability corresponds to the objective indicators of:

[0083] peak lateral acceleration, peak side slip angle, steering wheel angle corresponding to lateral acceleration of 8 m / s^2, yaw rate corresponding to lateral acceleration of 8 m / s^2, understeer gradient at 0.6g, understeer gradient at 0.8g, maximum lateral acceleration, lateral acceleration corresponding to peak yaw rate, peak yaw rate, Mz-SA slope at 40% load, lateral stiffness value at 40% load.

[0084] wherein the steady-state cornering ability corresponds to the objective indicators of:

[0085] body roll gradient, understeer gradient at 0.6g, maximum lateral acceleration, lateral acceleration corresponding to peak yaw rate, peak yaw rate, understeer gradient linearity, Mz-SA slope at 40% load, lateral stiffness value at 40% load.

[0086] Step 3: data augmentation is performed using a subjective-objective fusion generative adversarial network;

[0087] The subjective-objective fusion generative adversarial network is built in the following way:

[0088] The subjective-objective fusion generative adversarial network is composed of a full connection neural network generator and a discriminator with the same skeleton, each part containing 3 layers of hidden layers, wherein the first layer has 128 nodes, the second layer has 256 nodes, and the third layer has a full connection neural network with 128 nodes, using a RELU activation function, using mean square error to define the loss and adaptive learning rate for back propagation, and the principle diagram of the subjective-objective fusion generative adversarial network in the example is as shown in Figure 2 ;

[0089] Step 3.1: defining the whole vehicle random noise tire performance random noise and driving experience area random noise wherein, is a random noise of Gaussian distribution with the same dimension as X' n , is a random noise of Gaussian distribution with the same dimension as T' j , is a random noise of Gaussian distribution with the same value range as Y u .

[0090] Step 3.2: inputting Q X′ , Q T′ and Q Y as input random noise z respectively and inputting into the generator G of the subjective-objective fusion generative adversarial network for processing, so as to define the loss function L G of the generator G by formula (3) * and outputting the whole vehicle matching enhanced data set tire performance matching enhanced data set and driving experience area matching enhanced score

[0091] L G = H(1, D(G(z))) (3)

[0092] In formula (3): G is the generator, D is the discriminator, and H is the cross entropy.

[0093] Step 3.3: inputting the whole vehicle objective index data set X, the tire objective index data set T and the subjective score Y as the true value sample r respectively, the whole vehicle matching enhanced data set tire performance matching enhanced data set and driving experience area matching enhanced score as input data of the discriminator D in the subjective-objective fusion generative adversarial network and inputting the true value sample r and the corresponding input data The data are input together into the discriminator D of the subjective-objective fusion adversarial generative network for processing, thereby utilizing the loss function L of the discriminator D defined by equation (4). D The discriminator D is obtained by performing discrimination. * Output the enhanced vehicle matching dataset. Enhanced tire performance matching dataset Enhanced driving experience area matching and improved score

[0094]

[0095] Step 4: Training and quantifying the results of the combined subjective and objective model;

[0096] The subjective-objective fusion model is constructed as follows:

[0097] Regression and classification models are constructed using the same self-attention mechanism module and different output mapping modules. The self-attention mechanism module adopts an encoder architecture and achieves the fusion of subjective and objective evaluations through the self-attention mechanism. This module is implemented serially by six attention layers, each consisting of eight attention heads and one mapping layer. The first mapping layer uses a 1024-node one-dimensional convolutional neural network, and the second layer uses a 512-node one-dimensional convolutional neural network. The principle diagram of the subjective-objective fusion model in this example is shown below. Figure 3 As shown, the regression model output mapping module uses subjective region scores as the true label values ​​and is formed by a three-layer fully connected neural network with 512, 256, and 1 nodes respectively. Each layer uses the ReLU activation function and the mean absolute error is used as the loss function. The classification model output mapping module maps subjective region scores to 0-N class values ​​as the true label values ​​and is formed by a three-layer fully connected neural network with 512, 256, and N nodes respectively. The first two layers use the ReLU activation function, the last layer uses the softmax activation function, and the cross-entropy is used as the loss function.

[0098] Step 4.1: Step 4.1: will As training samples, As labels, the subjective-objective fusion model is trained to obtain the trained subjective-objective fusion model; the training error and key features of each driving experience area are shown in Table 1.

[0099] Table 1. Subjective Region Algorithm Errors and Most Important Features

[0100]

[0101] Step 4.2: Using X and T as validation samples and Y as the ground truth, input them into the trained subjective-objective fusion model for processing to obtain the classification score C = {C1, C2, ..., C...} for the driving experience area. u ,...,C U} and the regression score R = {R1, R2, ..., R u ,...,R U}, where C u R represents the category rating for the u-th driving experience area; u This represents the regression score for the u-th driving experience region; the subjective-objective fusion model in this example outputs the central area sensory prediction as follows: Figure 4 As shown; steering force feedback prediction is as follows Figure 5 As shown; steering response prediction is as follows Figure 6 As shown; linearity prediction is as follows Figure 7 As shown; lateral grip prediction is as follows Figure 8 As shown; Straight-line driving stability prediction is as follows Figure 9 As shown; lane change stability prediction is as follows Figure 10 As shown; steady-state rotation capability prediction is as follows: Figure 11 As shown in the figure above, the predicted results after training show good consistency with the test results, and can truly reflect the correlation between objective indicators and subjective scores in the driving experience area. This can be used for subsequent tire and vehicle matching evaluation, accelerating the tire selection cycle and reducing development costs.

[0102] Step 4.3: Assess the classification scores C = {C1, C2, ..., C} u ,...,C U} and regression score R = {R1, R2, ..., R u ,...,R U} Perform rating correction: Calculate the difference between the classification score and the regression score using equation (5);

[0103] diff = C u -R u (5)

[0104] Step 4.4: Set the correction index δ = 0.625, the minimum classification scale value i = 0.25, and use equation (6) to perform rating correction to obtain the quantitative evaluation result of tire-vehicle matching C′ = {C′1, C′2, ..., C′}. u ,...,C′ U}:

[0105]

[0106] In equation (6): C′ u It is Cu The result of the rating correction is the quantitative evaluation result of the u-th driving experience area.

[0107] In this example, the rating correction is based on the following to obtain the quantitative evaluation result C' = {C'1, C'2,..., C'N} of the matching of the tire and the whole vehicle: u U

[0108] 1) diff≤-0.25, C u increases by 0.75, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0109] 2) -0.25<diff≤-0.125, C u increases by 0.5, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0110] 3) -0.125<diff<-0.0625, C u increases by 0.25, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0111] 4) |diff|≤0.0625, C u remains unchanged, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0112] 5) 0.0625<diff≤0.125, C u decreases by 0.25, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0113] 6) 0.125<diff<0.25, C u decreases by 0.5, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0114] 7) diff≥0.25, C u decreases by 0.75, as the quantitative evaluation result C' of the u-th driving experience area; u .

[0115] In this embodiment, an electronic device includes a memory for storing a program supporting a processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0116] ​​In this embodiment, a computer readable storage medium stores a computer program, and the computer program is run by a processor to execute the steps of the above method.

Claims

1. A method for quantitatively evaluating tire-vehicle matching, characterized in that, The method comprises the following steps: Step 1, collecting objective indexes and subjective scores of the tire and the whole vehicle; Step 1.1: Obtain the vehicle objective index dataset X = {X1, X2, ..., X...}, consisting of N vehicle objective indexes under test conditions. n ,...,X N }; where X n Let represent the dataset of objective vehicle performance indicators under the nth test condition, and This represents the m-th objective indicator of the vehicle under the n-th test condition; M represents the number of objective indicators of the vehicle, and N represents the number of test conditions; Step 1.2: Obtain a tire objective index dataset T = {T1, T2,..., Tj} composed of tire objective indexes of J tire performance test conditions; wherein Tj represents a tire objective index dataset of the jth tire performance test condition, and j ,...,T J} ; wherein Tj represents a tire objective index dataset of the jth tire performance test condition, and j represents the kth tire objective index of the jth tire performance test condition; K represents the number of tire objective indexes; and J represents the number of tire performance test conditions.​ Step 1.3: Obtain subjective scores Y = {Y1, Y2, ..., Y} for U driving experience areas. u ,...,Y U }, where Y u This represents the subjective rating for the u-th driving experience area; U represents the number of driving experience areas. Step 2, dimension reduction is performed on X and T to obtain a dimension-reduced whole vehicle objective index dataset X' and a dimension-reduced tire objective index dataset T'; Step 3: Data augmentation is performed on X', T' and Y by using the subjective-objective fusion GAN to obtain the enhanced whole vehicle matching augmented dataset Enhanced tire performance matching augmented dataset and enhanced driving experience region matching augmented score Step 4, training of the subjective and objective fusion combination model and quantitative evaluation result; Step 4.1: the following formula is used to calculate the objective value of the sample: As the training sample, the following formula is used to calculate the objective value of the sample: As the label, the following formula is used to train the subjective and objective fusion combination model to obtain the trained subjective and objective fusion combination model: Step 4.2: input X, T as verification samples and Y as verification true values into the trained subjective and objective fusion combination model to obtain classification scores C = {C1, C2,...,C u ,...,C U} and regression scores R = {R1, R2,...,R u ,...,R U} of the driving experience area, wherein C u represents the classification score of the u-th driving experience area; and R u represents the regression score of the u-th driving experience area. Step 4.3: Calculate the difference diff = {diff1, diff2,..., diff u ,...,diff U} between the classification score C and the regression score R using formula (5) for each class c. Step 4.4: Calculate the final score S = {s1, s2,..., sn} for each class c using formula (6) for each class c. Step 4.5: Sort the final scores S in descending order and select the top n classes diff u = C u -R u (5) In formula (5): diff u represents C u and R u between them; Step 4.4: Set the correction index as δ, the minimum classification interval value as i, and correct the C using formula (6) to obtain the quantitative evaluation result C' = {C'1, C'2,..., C'N} of the tire and vehicle matching. u U}​ In formula (6): C' u is C u The quantitatively evaluated result after the rating correction.

2. The method for matching and quantitatively evaluating a tire and a whole vehicle according to claim 1, characterized in that, The step 2 comprises the following steps: Step 2.1: Calculate the vehicle objective index correlation coefficient p of the vehicle objective index data set X using formula (1) X : In formula (1): COV is the covariance, is the standard deviation of X is the standard deviation of X n the standard deviation of X Step 2.2: Calculating the tire objective index correlation coefficient p of the tire objective index dataset T using formula (2) T : In formula (2): is the standard deviation, is T j the standard deviation; Step 2.3: Set threshold Δ, respectively, to p X and p T Objective indicators greater than the threshold Δ are processed for dimension reduction, and the dimension-reduced whole vehicle objective indicator dataset X' = {X'1, X'2,..., X' n} and the tire objective indicator dataset T' = {T'1, T'2,..., T' m} are obtained, wherein X' n represents the dimension-reduced whole vehicle objective indicator dataset under the nth test working condition, and T' m represents the dimension-reduced tire objective indicator dataset under the jth tire performance test working condition. n N j J n j ​​​​​​ 3. The method for matching and quantitatively evaluating a tire and a whole vehicle according to claim 2, characterized in that, The step 3 comprises the following steps: Step 3.1: Define the random noise of the whole vehicle Tire performance random noise and random noise in the driving experience area in, Is with X′ n Random noise with the same Gaussian distribution of the same dimension. Is with T′ j Random noise with the same Gaussian distribution of the same dimension. Is with Y u Random noise with a Gaussian distribution and the same range of values; Step 3.2: Transfer Q X′ Q T′ and Q Y The random noise z, respectively, is input to the generator G of the subjective-objective fusion adversarial generative network for processing, thereby utilizing the loss function L of the generator G defined in equation (3). G The generator G is trained by backpropagation to obtain the trained generator G. * Output the whole vehicle matching enhancement dataset. Tire performance matching enhancement dataset Enhanced score matching driving experience area L G = H(l, D(G(z))) (3) In formula (3), G is a generator, D is a discriminator, and H is cross-entropy; Step 3.3: The vehicle objective index dataset X, the tire objective index dataset T and the subjective score Y are respectively taken as the true value sample r, the vehicle matching enhancement dataset the tire performance matching enhancement dataset and the driving experience area matching enhancement score as the input data of the discriminator D in the subjective and objective fusion generative adversarial network and the true value sample r and its corresponding input data are input into the discriminator D of the subjective and objective fusion generative adversarial network for processing, so as to obtain the loss function L of the discriminator D defined by formula (4) D to perform discrimination, obtain the trained discriminator D * and output the enhanced vehicle matching enhancement dataset the enhanced tire performance matching enhancement dataset and the enhanced driving experience area matching enhancement score 4. An electronic device comprising a memory and a processor, characterized in that The memory is configured to store a program supporting the processor to execute the tire and whole vehicle matching quantitative evaluation method according to any one of claims 1-3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program, when executed by the processor, performs the steps of the tire and whole vehicle matching quantitative evaluation method according to any one of claims 1-3.

Images