A method for aerodynamic shape design of automobile wheels based on machine learning optimization

Abstract

The application relates to a kind of automobile wheel aerodynamic shape design methods based on machine learning optimization, comprising the following steps: S1: obtaining the sample data of automobile wheel aerodynamic performance parameter and design parameter variable;S2: screening multiple design parameter variables as optimization parameters;S3: the sample data containing optimization parameters is randomly split to obtain training set and test set, and data enhancement is carried out;S4: the hyperparameters of gradient boosting tree model are optimized by improved particle swarm algorithm, the training set is trained, and the first aerodynamic performance prediction model is constructed;S5: the test set is predicted by the first aerodynamic performance prediction model to obtain the determination coefficient of test set, the determination coefficient is compared with the preset prediction accuracy, and the first aerodynamic performance prediction model is determined as the final aerodynamic performance prediction model;S6: the first aerodynamic performance prediction model is iteratively optimized, and the optimal value combination of optimization parameter is determined as the wheel structure optimization scheme.The application can improve design efficiency and design quality.

Description

Technical Field

[0001] This specification relates to the field of automotive wheel aerodynamic shape design technology, specifically to a method for automotive wheel aerodynamic shape design based on machine learning optimization. Background Technology

[0002] The geometry of a car tire is a complex three-dimensional curved surface. The parameters of the shape include tread width, section height, aspect ratio, tread pattern shape and depth, etc. The influence of each parameter on the performance is complex and contradictory. Performance parameters include rolling resistance, grip, wear characteristics, noise, and handling stability and wet skid resistance at high speeds.

[0003] The geometry of automobile tires is a spatial structure with complex curvature variations. Currently, various parametric design methods for automobile tire shapes exist, which can be used to quickly generate multiple tire shapes with different parameter combinations. However, limited by the number of design variables, these methods typically only describe tire shapes with relatively simple geometries, failing to fully express the detailed characteristics of real tires under complex operating conditions. Most current research on tire aerodynamic or aero-structure interaction optimization design is still based on these simplified parametric models. In addition, some studies have attempted to optimize tire shapes based on mesh deformation methods. This method drives surface deformation through control points, but a complete description of the tire shape often requires hundreds of control points, resulting in high computational cost and low convergence efficiency. Therefore, the application of existing automobile tire shape optimization methods in engineering practice is significantly limited, making it difficult to fundamentally improve design efficiency and quality. Summary of the Invention

[0004] This specification provides a machine learning-based optimization method for automotive wheel aerodynamic shape design to address the issues of insufficient design efficiency and quality in existing technologies.

[0005] This specification adopts the following technical solution: a method for designing the aerodynamic shape of automobile wheels based on machine learning optimization, comprising the following steps: S1: Obtain sample data of aerodynamic performance parameters and design parameter variables of automobile wheels; S2: Based on the sample data, the maximum mutual information coefficient combined with hierarchical clustering method is used to select multiple design parameter variables as optimization parameters; S3: Randomly split the sample data containing the optimization parameters into a training set and a test set according to a preset ratio, and perform data augmentation on the data in the training set; S4: The hyperparameters of the gradient boosting tree model are optimized using an improved particle swarm optimization algorithm, and the training set is used to train the model to construct the first aerodynamic performance prediction model. S5: The determination coefficient of the test set is obtained by predicting the test set using the first aerodynamic performance prediction model, and the determination coefficient is compared with the preset prediction accuracy: If the determination coefficient is greater than or equal to the preset prediction accuracy, then the first aerodynamic performance prediction model is determined as the final aerodynamic performance prediction model. If the determination coefficient is less than the preset prediction accuracy, the sample data is adjusted and steps S2 to S4 are repeated until the determination coefficient is greater than or equal to the preset prediction accuracy. S6: Set the numerical range and minimum change of the optimization parameters as optimization input, and use the improved particle swarm optimization algorithm to iteratively optimize the final aerodynamic performance prediction model to obtain the optimal combination of the optimization parameters as the wheel structure optimization scheme.

[0006] Based on the above technical means, by using an improved particle swarm optimization algorithm to optimize the gradient boosting tree model, a high-precision aerodynamic performance prediction model is constructed. This model can predict the aerodynamic performance under different combinations of wheel structure parameters without requiring a large amount of repetitive computation of fluid dynamics simulations, significantly shortening the design cycle and improving design efficiency. At the same time, through data augmentation and parameter screening, the model's generalization ability and prediction accuracy are improved, thereby obtaining a better wheel aerodynamic shape design scheme.

[0007] This solution can be flexibly adapted to different wheel types, driving environments and aerodynamic performance requirements by adjusting sample data and optimizing parameters and model hyperparameters, and has good engineering applicability and scalability.

[0008] Furthermore, the aerodynamic performance parameters of the automobile wheel in S1 specifically include the aerodynamic drag coefficient and the pressure pulsation coefficient.

[0009] Furthermore, S2 specifically includes the following steps: S21: The improved maximum mutual information coefficient method is used to calculate the correlation strength between each of the design parameters and the aerodynamic drag coefficient; S22: Using the correlation strength as a distance metric, hierarchical clustering is performed on the design parameters, and parameters with similarity higher than a set threshold are grouped into the same cluster; S23: Retain the design parameter with the highest association strength in each cluster as the optimization parameter.

[0010] Based on the above technical means, by performing hierarchical clustering on the design parameters, parameters with similarity higher than a set threshold are grouped into the same cluster. This allows the granularity of design parameter selection to be flexibly adjusted according to specific data characteristics and accuracy requirements, enabling it to adapt to design datasets of different vehicle models and wheel types, thus improving the versatility and robustness of the method.

[0011] Furthermore, S6 specifically includes the following steps: S61: Set the numerical range and minimum change of each optimization parameter as the search space boundary and step size constraint of the improved particle swarm algorithm, and initialize the particle swarm. In this context, the position of each particle in the particle swarm represents a combination of parameters from a set of optimization parameters; S62: Calculate the predicted aerodynamic performance value corresponding to the parameter combination represented by each particle through the final aerodynamic performance prediction model, and update the individual historical best position of each particle and the global best position of the particle swarm. S63: Determine whether the iteration termination condition has been met based on the preset iteration conditions. If the iteration termination condition has been met, output the parameter combination corresponding to the current global optimal position as the wheel structure optimization scheme. If the termination condition is not met, repeat step S62 until the iteration termination condition is met.

[0012] Based on the above technical means, the performance of each parameter combination is evaluated by using a trained final aerodynamic performance prediction model, which replaces the extremely time-consuming computational fluid dynamics simulation that needs to be performed in each iteration in the traditional method, thereby improving optimization efficiency and reducing computational resource consumption.

[0013] Furthermore, S63 also includes: if the iteration termination condition is not met, randomly perturb the positions of all or some of the particles, and repeat steps S62-S63 until the iteration termination condition is met.

[0014] Based on the above technical means, the positions of all or some particles are randomly perturbed, which greatly increases the diversity of the population, avoids the algorithm from getting stuck in local optima, and increases the probability of finding the global optimum or a better combination of parameters.

[0015] Furthermore, S62 also includes: according to preset conditions, after each iteration, a certain proportion of particles are retained and directly enter the next generation of particle swarm.

[0016] Based on the above technical means, according to the preset conditions, a certain proportion of particles are retained after each iteration, so that the historical optimal solution generated in each iteration can be continued without being lost. This can prevent the optimization process from "regressing" or "oscillating" due to the randomness of the algorithm, and ensure the monotonic non-deterioration trend of the overall performance of the particle swarm.

[0017] Furthermore, step S4 specifically includes: S41: Select the hyperparameters to be optimized and their value ranges for the gradient boosting tree model, map the hyperparameters to be optimized to the position vectors of particles in the improved particle swarm optimization algorithm, that is, the position vector of each particle represents a set of values ​​of the hyperparameters to be optimized; and initialize the particle swarm. S42: Construct a gradient boosting tree model based on the hyperparameter values ​​corresponding to the position vector of each particle, train the gradient boosting tree model using the training set, and calculate the determination coefficient of the gradient boosting tree model using the test set as the fitness value of each particle. S43: The improved particle swarm optimization algorithm updates the historical best position of each particle and the global best position of the particle swarm based on the fitness value of each particle, and also updates the velocity and position of the particles. S44: Take the hyperparameter combination corresponding to the global optimal position of the population at this time as the optimal hyperparameter combination, and retrain the training set based on the optimal hyperparameter combination using the gradient boosting tree model to construct the first aerodynamic performance prediction model.

[0018] Based on the above technical means, hyperparameters are mapped to particle positions in a particle swarm, and an improved particle swarm algorithm is used to perform intelligent search within the defined value space. With the test set determination coefficient as the fitness guide, a set of hyperparameter combinations that maximizes the model's prediction accuracy can be accurately determined, ensuring that the constructed first aerodynamic performance prediction model has high performance and robustness.

[0019] Furthermore, the adjustment of the sample data in S5 specifically includes: obtaining aerodynamic performance parameter data of automobile wheels, and removing the aerodynamic performance parameter data of the wheels corresponding to the largest outlier based on the distribution of the aerodynamic performance parameter data, to obtain the adjusted sample data.

[0020] Based on the above technical means, the training set data is cleaned by removing the wheel aerodynamic performance parameter data corresponding to the largest outlier, thereby reducing the interference of abnormal data on model training and improving the stability and generalization ability of the prediction model.

[0021] Furthermore, S5 also includes: if the determination coefficient is less than the preset prediction accuracy, adjusting the clustering threshold in the clustering process in step S2 and / or adjusting the depth of the gradient boosting tree model in step S4.

[0022] Furthermore, the data augmentation in S3 specifically includes: adding noise and rotating / scaling the data in the training set.

[0023] Based on the above technical means, adding noise can simulate measurement errors, manufacturing tolerances or environmental disturbances in actual engineering, and enhance the stability of the model in practical applications; scaling can expand the range of morphological changes of the samples, so that the model can adapt to wheel structures of different sizes, angles or postures, and avoid overfitting.

[0024] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: 1. By using an improved particle swarm optimization algorithm to optimize the gradient boosting tree model, a high-precision aerodynamic performance prediction model is constructed. This model can predict the aerodynamic performance under different combinations of wheel structure parameters without requiring extensive repetitive computation of fluid dynamics simulations, significantly shortening the design cycle. At the same time, through data augmentation and parameter screening, the model's generalization ability and prediction accuracy are improved, thereby obtaining a better wheel aerodynamic shape design scheme.

[0025] 2. This manual allows for flexible adaptation to different wheel types, driving environments, and aerodynamic performance requirements by adjusting sample data and optimizing parameters and model hyperparameters, demonstrating good engineering applicability and scalability. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall structure and flow of this embodiment; Figure 2 This is a flowchart illustrating the process of using maximum mutual information coefficient combined with hierarchical clustering in this embodiment; Figure 3 This is a schematic diagram of the process for constructing the first aerodynamic performance prediction model in this embodiment; Figure 4 This is a schematic diagram of the iterative optimization process for the final aerodynamic performance prediction model in this embodiment; Figure 5 This is a diagram showing the position of the design parameter variables at the wheels of the automobile in this embodiment; Figure 6 This is a partial schematic diagram of the wheel shape sample in this embodiment; Figure 7 This is a schematic diagram of the prediction results for the training set and the test set in this embodiment; Figure 8 This is a schematic diagram of the optimized wheel structure in this embodiment.

[0027] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings. The same or similar reference numerals correspond to the same or similar components. The terms describing positional relationships in the drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0029] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0030] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0031] In the embodiments of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" and "second" may explicitly or implicitly include one or more of that feature.

[0032] In the embodiments of this application, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium.

[0033] In embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0034] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0035] Example 1 like Figure 1 As shown, this embodiment provides a method for designing the aerodynamic shape of automobile wheels based on machine learning optimization, including the following steps: S1: Obtain sample data of aerodynamic performance parameters and design parameter variables of automobile wheels.

[0036] In this preferred embodiment, the aerodynamic performance parameters of the automobile wheel include the aerodynamic drag coefficient. and pressure pulsation coefficient Furthermore, the aerodynamic drag coefficient is the core optimization objective, while the pressure pulsation coefficient is the secondary constraint objective.

[0037] like Figure 5 As shown, in this preferred embodiment, the design parameter variables include 10 key parameters affecting the wheel body, biomimetic surface, and environment, specifically: 3 body parameters, namely wheel height H2, horizontal position W1, and wheel diameter D1; 5 surface parameters, namely pattern quantity A1, pattern width C1, pattern diagonal length C2, pit depth d, and rib height h; and 2 environmental parameters, namely air density B1 and wind speed V. In this preferred embodiment, the slope length C2 and width C1 satisfy C2 / C1∈[0.15,0.25], wind speed∈[40km / h,120km / h], and air density∈[1.18,1.4]. .

[0038] In this preferred embodiment, such as Figure 6 As shown, 5760 wheel shape samples were constructed (using an L16 orthogonal experimental design to cover the parameter range); computational fluid dynamics (CFD) calculations were performed using a k-ω turbulence model to obtain the aerodynamic drag coefficient for each sample. and pressure pulsation coefficient The convergence criterion for the iterative residual is set as follows: .

[0039] S2: Based on the sample data, the maximum mutual information coefficient combined with hierarchical clustering method is used to select multiple design parameter variables as optimization parameters.

[0040] like Figure 2 As shown, in this embodiment, S2 specifically includes the following steps: S21: The correlation strength between each design parameter and the aerodynamic drag coefficient is calculated using the improved Maximum Information Coefficient (MIC) method. In this preferred embodiment, an adaptive binning method is used to discretize the 10-dimensional parameters, setting the maximum hysteresis order k=5 and the significance level α=0.01 (after Bonferroni correction), to calculate each design parameter and the aerodynamic drag coefficient. The MIC values ​​are as follows: pit depth d=0.82, rib height h=0.73, spoke inclination angle θ=0.68, wheel diameter D1=0.65, wind speed V=0.62, number of patterns A1=0.59, rim curvature=0.58, etc. In this preferred embodiment, the number of bins in the adaptive binning method is dynamically adjusted according to the parameter distribution, specifically ranging from 5 to 12.

[0041] S22: Using the correlation strength as a distance metric, hierarchical clustering is performed on the design parameters, and parameters with similarity higher than a set threshold are grouped into the same cluster; In this preferred embodiment, the distance between each design parameter and the aerodynamic drag coefficient MIC value is used as a distance metric (distance = 1 - MIC value). Clustering is performed using the sum of squared deviations method (Ward's method), and parameters with a similarity higher than 0.85 are grouped into the same cluster. Each cluster retains the parameter with the highest MIC value.

[0042] In this embodiment, "rim curvature" and "wheel diameter D1" are clustered together.

[0043] S23: Retain the design parameter with the highest association strength in each cluster as the optimization parameter.

[0044] In this embodiment, seven important variables were ultimately selected as optimization parameters: wheel diameter D1, distance from wheel center to ground H2, distance from wheel center to wheel cavity wall W1, air density B1, number of patterns A1, pattern width C1, and pattern slant length C2.

[0045] S3: Randomly split the sample data containing the optimization parameters into a training set and a test set according to a preset ratio, and perform data augmentation on the data in the training set; In this preferred embodiment, the 5760 samples are randomly split into a training set (4608 samples) and a test set (1152 samples) at a ratio of 4:1.

[0046] In this embodiment, data enhancement includes: ① to The minority class samples with a value < -0.35 (accounting for 18%) were subjected to Synthetic Minority Oversampling (SMOTE oversampling), which increased the minority class sample ratio to 42% after a 3-fold amplification; ② Gaussian noise (σ=0.5% of the range, such as D1=650±3.25mm) was added to all parameters; ③ The biomimetic surface parameters were randomly rotated (±5°) and scaled (±3%), and the final training set sample size was expanded to 6496.

[0047] S4: The hyperparameters of the gradient boosting tree model are optimized using an improved particle swarm optimization algorithm, and the training set is used to train the model to construct the first aerodynamic performance prediction model. like Figure 3 As shown, in this embodiment, S4 specifically includes the following steps: S41: Select the hyperparameters to be optimized and their value ranges for the gradient boosting tree model, map the hyperparameters to be optimized to the position vectors of particles in the improved particle swarm optimization algorithm, that is, the position vector of each particle represents a set of values ​​of the hyperparameters to be optimized; and initialize the particle swarm. In this embodiment, the basic learner used in the construction of the optimized gradient boosting tree model is a classification and regression tree (CART), and the loss function is the mean squared error (MSE). The model parameters in the optimized gradient boosting tree model are updated and optimized through gradient descent.

[0048] In this preferred embodiment, five hyperparameters to be optimized and their value ranges are determined: decision tree depth, learning rate, number of estimators, minimum number of sample splits, and minimum number of sample leaf nodes.

[0049] In this preferred embodiment, the decision tree depth ranges from 3 to 15, the learning rate ranges from 0.01 to 0.3, the estimator / estimator number ranges from 50 to 500, the minimum number of sample splits ranges from 2 to 20, and the minimum number of sample leaf nodes ranges from 1 to 10.

[0050] In this preferred embodiment, the particle swarm size is set to 30, and the maximum number of iterations is iter. max =50, inertia weight ω max =0.9、 ω min =0.4, initial learning factor values ​​c1=2.5, c2=0.5.

[0051] S42: Construct a gradient boosting tree model based on the hyperparameter values ​​corresponding to the position vector of each particle, train the gradient boosting tree model using the training set, and calculate the determination coefficient of the gradient boosting tree model using the test set as the fitness value of each particle. S43: The improved particle swarm optimization algorithm updates the historical best position of each particle and the global best position of the particle swarm based on the fitness value of each particle, and also updates the velocity and position of the particles. In this preferred embodiment, the dynamic adjustment strategy is specifically as follows: ω = ω max -( ω max - ω min )×iter / iter max c1 = 2.5 - 2 × iter / iter max c2 = 0.5 + 2 × iter / iter max ,in, ω This represents the weight of the current particle swarm, iter represents the current iteration number, and the learning factors c1 and c2 are updated.

[0052] In this preferred embodiment, the top 10% of elite particles are retained during the iteration process. Elite particles refer to particles whose fitness ranking is in the top 10% of all particles.

[0053] S44: Take the hyperparameter combination corresponding to the global optimal position of the population at this time as the optimal hyperparameter combination, and retrain the training set based on the optimal hyperparameter combination using the gradient boosting tree model to construct the first aerodynamic performance prediction model.

[0054] S5: The determination coefficient of the test set is obtained by predicting the test set using the first aerodynamic performance prediction model, and the determination coefficient is compared with the preset prediction accuracy: If the determination coefficient is greater than or equal to the preset prediction accuracy, then the first aerodynamic performance prediction model is determined as the final aerodynamic performance prediction model. If the determination coefficient is less than the preset prediction accuracy, the sample data is adjusted and steps S2 to S4 are repeated until the determination coefficient is greater than or equal to the preset prediction accuracy. In this embodiment, the preset prediction accuracy is 0.9.

[0055] In this embodiment, the decision tree depth is 8, the learning rate is 0.12, the number of estimators is 280, the minimum number of sample splits is 12, and the minimum number of sample leaf nodes is 5.

[0056] In this embodiment, a GBDT model is constructed based on this combination. After 5-fold cross-validation, the test set R² = 0.963, which meets the accuracy requirement of R² ≥ 0.9. The prediction results of the training set and the test set are compared as follows: Figure 7 As shown, the average absolute error between the predicted and actual values ​​is 0.012.

[0057] S6: Set the numerical range and minimum change of the optimization parameters as optimization inputs, and use the improved particle swarm optimization algorithm to iteratively optimize the final aerodynamic performance prediction model to obtain the optimal combination of the optimization parameters as the wheel structure optimization scheme.

[0058] like Figure 4 As shown, in this embodiment, step S6 specifically includes the following steps: S61: Set the numerical range and minimum change of each optimization parameter as the search space boundary and step size constraint of the improved particle swarm algorithm, and initialize the particle swarm. In this preferred embodiment, the numerical range and minimum change of each optimization parameter are shown in Table 1:

[0059] In this embodiment, the particle swarm is initialized (size 40), and the velocity range is set to ±10% of the parameter value range.

[0060] In this context, the position of each particle in the particle swarm represents a combination of parameters from a set of optimization parameters; S62: Calculate the predicted aerodynamic performance value corresponding to the parameter combination represented by each particle through the final aerodynamic performance prediction model, and update the individual historical best position of each particle and the global best position of the particle swarm. S63: Determine whether the iteration termination condition has been met based on the preset iteration conditions. If the iteration termination condition has been met, output the parameter combination corresponding to the current global optimal position as the wheel structure optimization scheme. If the termination condition is not met, repeat step S62 until the iteration termination condition is met.

[0061] In this preferred embodiment, the iteration termination condition is that the number of iterations is 50.

[0062] In this embodiment, in step S6, the fitness function is specifically defined as follows: Minimize as the objective, satisfying ≤0.08; if If the value is greater than 0.08, a penalty is imposed on the fitness value, where the penalty coefficient is 1.5; The iterative process specifically includes: each iteration calculating the values ​​of each particle using the GBDT model. and Update the optimal (pbest) and global optimal (gbest) values ​​for each particle; randomly mutate the particle positions (mutation probability 0.05, mutation range is ±2 times the minimum change) to avoid local optima.

[0063] In this embodiment, the globally optimal parameter values ​​are: D1=680mm, H2=320mm, W1=85mm, B1=1.25kg / m³, A1=30, C1=25mm, C2=5.5mm, corresponding to the prediction. =-0.19, =0.072.

[0064] The wheel model constructed using these parameters was verified by CFD simulation (with parameters consistent with S1). The simulated values ​​were Cd = -0.1921 and Cp = 0.073, with errors of 0.11% and 1.39% respectively compared to the predicted values, both less than 2%, demonstrating the effectiveness of the method. The optimized wheel structure is shown below. Figure 8 As shown.

[0065] In summary, this embodiment employs an improved particle swarm optimization algorithm. By combining dynamic inertia weights, adaptive learning factors, and an elite retention strategy, it avoids the limitations of the traditional dragonfly algorithm in the integer search space and can effectively escape local optima, ensuring a higher global optimum for the optimization results. It also boasts broader engineering applicability: the model training process does not rely on complex neural network structure design, the hyperparameter optimization process is simpler, and the constraint settings during the optimization iteration process are more flexible. It can adapt to the wheel structure optimization needs of different vehicle models and environments, exhibiting stronger scalability.

[0066] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. A method for designing the aerodynamic shape of automobile wheels based on machine learning optimization, characterized in that, Includes the following steps: S1: Obtain sample data of aerodynamic performance parameters and design parameter variables of automobile wheels; S2: Based on the sample data, the maximum mutual information coefficient combined with hierarchical clustering method is used to select multiple design parameter variables as optimization parameters; S3: Randomly split the sample data containing the optimization parameters according to a preset ratio to obtain a training set and a test set, and perform data augmentation on the data in the training set; S4: The hyperparameters of the gradient boosting tree model are optimized using an improved particle swarm optimization algorithm, and the training set is used to train the model to construct the first aerodynamic performance prediction model. S5: The determination coefficient of the test set is obtained by predicting the test set using the first aerodynamic performance prediction model, and the determination coefficient is compared with the preset prediction accuracy: If the determination coefficient is greater than or equal to the preset prediction accuracy, then the first aerodynamic performance prediction model is determined as the final aerodynamic performance prediction model. If the determination coefficient is less than the preset prediction accuracy, the sample data is adjusted and steps S2 to S4 are repeated until the determination coefficient is greater than or equal to the preset prediction accuracy. S6: Set the numerical range and minimum change of the optimization parameters as optimization input, and use the improved particle swarm optimization algorithm to iteratively optimize the final aerodynamic performance prediction model to obtain the optimal combination of the optimization parameters as the wheel structure optimization scheme.

2. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 1, characterized in that, The aerodynamic performance parameters of the automobile wheel in S1 specifically include the aerodynamic drag coefficient and the pressure pulsation coefficient.

3. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 2, characterized in that, S2 specifically includes the following steps: S21: The improved maximum mutual information coefficient method is used to calculate the correlation strength between each of the design parameters and the aerodynamic drag coefficient; S22: Using the correlation strength as a distance metric, hierarchical clustering is performed on the design parameters, and parameters with similarity higher than a set threshold are grouped into the same cluster; S23: Retain the design parameter with the highest association strength in each cluster as the optimization parameter.

4. A method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 2 or 3, characterized in that, S6 specifically includes the following steps: S61: Set the numerical range and minimum change of each optimization parameter as the search space boundary and step size constraint of the improved particle swarm algorithm, and initialize the particle swarm. In this context, the position of each particle in the particle swarm represents a combination of parameters from a set of optimization parameters; S62: Calculate the predicted aerodynamic performance value corresponding to the parameter combination represented by each particle through the final aerodynamic performance prediction model, and update the individual historical best position of each particle and the global best position of the particle swarm. S63: Determine whether the iteration termination condition has been met based on the preset iteration conditions. If the iteration termination condition has been met, output the parameter combination corresponding to the current global optimal position as the wheel structure optimization scheme. If the termination condition is not met, repeat step S62 until the iteration termination condition is met.

5. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 4, characterized in that, S63 further includes: if the iteration termination condition is not met, randomly perturb the positions of all or some of the particles, and repeat steps S62-S63 until the iteration termination condition is met.

6. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 4, characterized in that, S62 also includes: according to preset conditions, after each iteration, a certain proportion of particles are retained and directly enter the next generation of particle swarm.

7. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 1, characterized in that, Step S4 specifically includes: S41: Select the hyperparameters to be optimized and their value ranges for the gradient boosting tree model, map the hyperparameters to be optimized to the position vectors of particles in the improved particle swarm optimization algorithm, that is, the position vector of each particle represents a set of values ​​of the hyperparameters to be optimized; and initialize the particle swarm. S42: Construct a gradient boosting tree model based on the hyperparameter values ​​corresponding to the position vector of each particle, train the gradient boosting tree model using the training set, and calculate the determination coefficient of the gradient boosting tree model using the test set as the fitness value of each particle. S43: The improved particle swarm optimization algorithm updates the historical best position of each particle and the global best position of the particle swarm based on the fitness value of each particle, and also updates the velocity and position of the particles. S44: Take the hyperparameter combination corresponding to the global optimal position of the population at this time as the optimal hyperparameter combination, and retrain the training set based on the optimal hyperparameter combination using the gradient boosting tree model to construct the first aerodynamic performance prediction model.

8. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 1, characterized in that, The adjustment of the sample data in S5 specifically includes: obtaining aerodynamic performance parameter data of automobile wheels, and removing the aerodynamic performance parameter data of the wheels corresponding to the largest outlier based on the distribution of the aerodynamic performance parameter data, to obtain the adjusted sample data.

9. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 1, characterized in that, S5 further includes: if the determination coefficient is less than the preset prediction accuracy, adjusting the clustering threshold in the clustering process in step S2 and / or adjusting the depth of the gradient boosting tree model in step S4.

10. The method for designing the aerodynamic shape of automobile wheels based on machine learning optimization according to claim 1, characterized in that, The data augmentation in S3 specifically includes adding noise and rotating / scaling the data in the training set.

Images