Construction method and device of automobile wind resistance prediction model, electronic equipment and medium

Abstract

The invention relates to the field of wind resistance prediction, in particular to a construction method and device of an automobile wind resistance prediction model, electronic equipment and a medium. The construction method of the automobile wind resistance prediction model comprises the steps of obtaining sample data of automobile wind resistance, wherein the sample data comprises automobile appearance point cloud data and an automobile wind resistance coefficient; in response to a to-be-trained model, obtaining prediction data of the sample data; the model to be trained comprises a PointNet + + model; model parameters of the to-be-trained model are updated to reduce loss of the prediction data and the sample data, a wind resistance prediction model is obtained, and the wind resistance prediction model is used for responding to the automobile appearance point cloud data and predicting the automobile wind resistance coefficient. The method is high in model prediction precision and prediction efficiency.

Description

Technical Field

[0001] The present application relates to the field of wind resistance prediction, and more particularly, to a method, device, electronic device and medium for constructing an automobile wind resistance prediction model. Background Art

[0002] The current trend of rapid iteration of vehicle models in the automotive industry has put forward higher efficiency requirements for the optimization of the overall vehicle aerodynamic drag. Quickly providing high-precision CFD (computational fluid dynamics) numerical simulation results is an important way to meet this demand. Low-order moment turbulence models based on physical models and traditional numerical calculation methods have been widely used in actual engineering CFD calculations, but their simulation speed is not sufficient to effectively support the requirements of current rapid concept design of automobile shapes and full-space aerodynamic optimization.

[0003] To solve the above problems, the research on rapid prediction of the automobile wind resistance coefficient has begun to attract the attention of some scholars. At present, the prediction of the wind resistance coefficient mainly adopts the parameterization method, that is, by selecting external styling parameters that have a greater impact on the automobile wind resistance coefficient, and using genetic algorithms or response surface models to establish a mapping relationship between the styling parameters and the wind resistance coefficient. The parameterization method adopts the method of simplifying the three-dimensional complex model into low-dimensional features in the processing of the input parameters of the prediction model. However, there are a large number of complex curved surfaces in the automobile external styling, and one-dimensional or two-dimensional features are difficult to fully express it, resulting in generally low prediction accuracy of the parameterization method. In addition, the form of the input parameters also limits the development of the prediction model.

[0004] In view of this, the present application is specifically proposed. Summary of the Invention

[0005] The purpose of the present application is to provide a method, device, electronic device and medium for constructing an automobile wind resistance prediction model, so as to solve the problems of low computational fluid dynamics calculation efficiency and difficulty in accurately characterizing the automobile external styling existing in the prior art.

[0006] To achieve the above purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides a method for constructing an automobile wind resistance prediction model, including: Obtaining sample data of the automobile wind resistance, where the sample data includes automobile external shape point cloud data and the automobile wind resistance coefficient; In response to the model to be trained, obtaining prediction data of the sample data; the model to be trained includes a PointNet++ model; Updating the model parameters of the model to be trained to reduce the loss between the prediction data and the sample data, and obtaining a wind resistance prediction model, where the wind resistance prediction model is used to respond to the automobile external shape point cloud data and predict the automobile wind resistance coefficient.

[0007] As a further preferred technical solution, after obtaining the sample data of the vehicle aerodynamic drag, it further includes: Using the farthest point sampling algorithm to simplify the vehicle exterior point cloud data to obtain a set of sampled points.

[0008] As a further preferred technical solution, the step of using the farthest point sampling algorithm to simplify the vehicle exterior point cloud data to obtain a set of sampled points includes: Selecting an initial point from the vehicle exterior point cloud data; Determining a second point according to the Euclidean distance between the initial point and each remaining point; the remaining points are the other points in the vehicle exterior point cloud data except the initial point; the second point is the point with the largest Euclidean distance from the initial point; Determining an initial set of sampled points according to the initial point and the second point; Performing normalization processing on the initial set of sampled points to obtain a set of sampled points.

[0009] As a further preferred technical solution, the PointNet++ model includes at least two set abstraction layers, a global feature extraction layer, and a fully connected layer; The set abstraction layer samples and groups the set of sampled points to obtain multiple local neighborhoods, and then extracts local features from the multiple local neighborhoods; The global feature extraction layer uses max pooling operation to summarize the local features to obtain global features; The fully connected layer receives the global features and outputs prediction data.

[0010] As a further preferred technical solution, the step of sampling and grouping the set of sampled points to obtain multiple local neighborhoods, and then extracting local features from the multiple local neighborhoods includes: Using the farthest point sampling algorithm to sample the set of sampled points to obtain a simplified set of sampled points; Using a query ball or classification algorithm to group the simplified set of sampled points to obtain multiple local neighborhoods; Using a PointNet network to extract local features from the multiple local neighborhoods.

[0011] As a further preferred technical solution, the global feature extraction layer uses a PointNet network for max pooling operation.

[0012] As a further preferred technical solution, the loss function in the process of updating the model parameters of the model to be trained is: , Where,MSE is the loss function, y i is the true value of the i th automotive aerodynamic drag sample, is the i th predicted value of the automotive aerodynamic drag sample, m is the number of samples of the automotive aerodynamic drag samples.

[0013] In a second aspect, the present application provides an apparatus for constructing an automotive aerodynamic drag prediction model, including: A sample data acquisition module, configured to acquire sample data of automotive aerodynamic drag, where the sample data includes automotive contour point cloud data and an automotive aerodynamic drag coefficient; A sample data prediction module, configured to obtain prediction data of the sample data in response to a model to be trained; the model to be trained includes a PointNet++ model; An aerodynamic drag prediction model determination module, configured to update model parameters of the model to be trained to reduce the loss between the prediction data and the sample data, and obtain an aerodynamic drag prediction model, where the aerodynamic drag prediction model is configured to predict an automotive aerodynamic drag coefficient in response to the automotive contour point cloud data.

[0014] In a third aspect, the present application provides an electronic device, including: At least one processor, and a memory communicatively connected to at least one of the processors; wherein, the memory stores instructions executable by at least one of the processors, and the instructions are executed by at least one of the processors to enable at least one of the processors to execute the above method.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the above method.

[0016] Compared with the prior art, the beneficial effects of the present application are: The method for constructing an automobile aerodynamic drag prediction model provided by this application includes: obtaining sample data of automobile aerodynamic drag, where the sample data includes automobile external shape point cloud data and automobile aerodynamic drag coefficient; in response to the model to be trained, obtaining prediction data of the sample data; the model to be trained includes a PointNet++ model; updating the model parameters of the to-be-trained model to reduce the loss between the prediction data and the sample data, and obtaining an aerodynamic drag prediction model, where the aerodynamic drag prediction model is used to respond to the automobile external shape point cloud data and predict the automobile aerodynamic drag coefficient. This method uses sample data including automobile external shape point cloud data and automobile aerodynamic drag coefficient to train a specific model to be trained, and then updates the model parameters to obtain an aerodynamic drag prediction model. This aerodynamic drag prediction model can predict the automobile aerodynamic drag coefficient based on the automobile external shape point cloud data. The automobile external shape point cloud data can accurately represent the external shape of the automobile. This model has high prediction efficiency. Compared with the traditional CFD simulation method, the prediction time can be reduced by three orders of magnitude. Moreover, the data used is the automobile external shape point cloud data, and the PointNet++ model used can effectively extract multi-scale features of the point cloud data and can predict the aerodynamic drag coefficient of complex geometric shapes with high prediction accuracy.

[0017] Further, this application uses the farthest point sampling algorithm to simplify the automobile external shape point cloud data, further reducing the data volume and improving the prediction efficiency. Description of the Drawings

[0018] In order to more clearly illustrate the specific embodiments of this application or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of the method for constructing an automobile aerodynamic drag prediction model provided by this application; Figure 2 It is a structural schematic diagram of the device for constructing an automobile aerodynamic drag prediction model provided by this application; Figure 3 It is a structural schematic diagram of the electronic device provided by this application. Detailed Embodiments

[0020] The following makes an explanation of the exemplary embodiments of this application in conjunction with the drawings, including various details of the embodiments of this application to facilitate understanding. It should be considered that they are only exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described here without departing from the scope and spirit of this application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted below.

[0021] As mentioned in the background art, the prior art has problems such as low computational efficiency in computational fluid dynamics and difficulty in accurately characterizing the external shape of an automobile by parametric methods. In response to this, based on the sample data of the aerodynamic drag of an automobile including the point cloud data of the automobile shape and the aerodynamic drag coefficient of the automobile, this application uses the PointNet++ model to be trained to predict the sample data, and then updates the model parameters to obtain an aerodynamic drag prediction model. The following further describes this application in detail with reference to embodiments.

[0022] Embodiment 1 Figure 1 FIG. is a flowchart of a method for constructing an aerodynamic drag prediction model of an automobile provided in this embodiment. This method can be executed by a device for constructing an aerodynamic drag prediction model of an automobile. The device can be composed of software and / or hardware and is generally integrated in an electronic device. The electronic device can be an electronic computer or other mobile terminals (such as a smart phone, a tablet computer, etc.). For the convenience of understanding, each step in the construction method of this embodiment takes an electronic computer as the execution subject.

[0023] As Figure 1 shown, this embodiment provides a method for constructing an aerodynamic drag prediction model of an automobile, including the following steps: S110. Obtain sample data of the aerodynamic drag of an automobile, where the sample data includes point cloud data of the automobile shape and the aerodynamic drag coefficient of the automobile.

[0024] Among them, the "point cloud data of the automobile shape" refers to the point cloud data that can reflect the shape characteristics of the automobile. This point cloud data can be obtained after data conversion from a publicly used three-dimensional object database (such as ShapeNet) or a custom dataset (such as Dataset) (that is, converting the original data into point cloud data. For example, if the original data is the mesh data of the automobile shape generated by a CAD model, the mesh data needs to be converted into point cloud data). It can be understood that only the vehicle-related data in the above databases or datasets is used. The "aerodynamic drag coefficient of the automobile" refers to the aerodynamic drag coefficient calculated by using an aerodynamic drag coefficient simulation software. The simulation software includes CFD simulation tools (such as OpenFOAM). The aerodynamic drag coefficient calculated above is used as the true value of the sample to optimize the parameters of the prediction model.

[0025] The point cloud data of the shape of an automobile has about 300,000 points, with a large amount and high density, which is not conducive to data storage, transmission, and calculation. Therefore, in this embodiment, when using this point cloud data for prediction, it can be simplified first.

[0026] In an alternative embodiment, after obtaining the sample data of the vehicle aerodynamic drag, it further includes: simplifying the vehicle contour point cloud data by using the farthest point sampling algorithm to obtain a set of sampled points. In this embodiment, the farthest point sampling algorithm is used to simplify the vehicle contour point cloud data, which can still maintain the sharp characteristics of the point cloud after simplification.

[0027] In an alternative embodiment, the step of simplifying the vehicle contour point cloud data by using the farthest point sampling algorithm to obtain a set of sampled points includes: Selecting an initial point from the vehicle contour point cloud data; Determining a second point according to the Euclidean distance between the initial point and each remaining point; the remaining points are the other points in the vehicle contour point cloud data except the initial point; the second point is the point with the largest Euclidean distance from the initial point; Determining an initial set of sampled points according to the initial point and the second point; Performing normalization processing on the initial set of sampled points to obtain a set of sampled points.

[0028] Suppose n is the number of original points in the point cloud data and m is the number of points after simplification. When using the farthest point sampling algorithm for data simplification, first, an initial point needs to be selected. This initial point can be randomly chosen and denoted as p0. Then, calculate the Euclidean distance between each remaining point and this initial point. The point farthest from p0 is the second point, denoted as p1, and the expression is as follows: p1 = argmax(d(p i , p0)), i = 1, 2,..., n - 1. In the formula, argmax represents the subscript for taking the maximum value, and d(p i , p0) represents the Euclidean distance between the data point p i and p0, and i is the subscript of the point. For each of the remaining points, calculate the distances to p0 and p1 respectively, and select the shortest distance as the overall distance of this point to p0 and p1. After calculating these distances, select the point with the largest distance and add it to the set (as the 3rd point in the set, and the other 2 points are the initial point and the second point respectively). For each of the remaining points, select in a similar way as determining the above 3rd point until m points are selected. Slightly different is that when selecting subsequent points, since there are already several points in the set (such as 3 points), when calculating the distances, it is necessary to calculate the distances between each remaining point and the several existing points in the set (such as 3 points) respectively. The selection expression is as follows: p m = argmax(min(d(p i , p j))), i ≠ j, i, j = 1, 2, ..., m - 1. After the above processing, the initial sampling point set is obtained, and then normalization processing is performed to make the scale mean 0 and the variance 1 in the XYZ three directions. After processing the coordinates of the points in the initial sampling point set through the mean and variance, the coordinates of the finally point-normalized are obtained to eliminate the dimension. This algorithm essentially translates the point cloud through the mean and scales the point cloud through the variance, placing the center of the point cloud data at the origin of the coordinate system.

[0029] S120. In response to the model to be trained, obtain the predicted data of the sample data; the model to be trained includes a PointNet++ model.

[0030] In an alternative embodiment, the PointNet++ model includes at least two set abstraction layers, a global feature extraction layer, and a fully connected layer; The set abstraction layer samples and groups the sampling point set to obtain multiple local neighborhoods, and then extracts local features from the multiple local neighborhoods; The global feature extraction layer uses max pooling operation to summarize the local features to obtain global features; The fully connected layer receives the global features and outputs the predicted data.

[0031] The PointNet++ model in this embodiment includes at least two set abstraction layers, a global feature extraction layer, and a fully connected layer. Each layer has its own function. After sampling, grouping, local feature extraction, and global feature summarization, the predicted data is output. This model can integrate local and global information layer by layer and has strong processing ability for local geometric structures. Among them, the set abstraction layer can perform hierarchical abstraction on the point cloud data to capture local geometric information at different scales. Sampling the sampling point set in the set abstraction layer can further reduce the number of points. "Local neighborhood" refers to the neighborhood corresponding to a single point.

[0032] In an alternative embodiment, the sampling and grouping the sampling point set to obtain multiple local neighborhoods, and then extracting local features from the multiple local neighborhoods includes: Use the farthest point sampling algorithm to sample the sampling point set to obtain a simplified sampling point set; Use the query ball or classification algorithm to group the simplified sampling point set to obtain multiple local neighborhoods; Use the PointNet network to extract local features from the multiple local neighborhoods.

[0033] In this embodiment, the farthest point sampling algorithm is used to sample the set of sampling points, and the data simplification effect is good. The query ball (Ball Query) or classification algorithm (such as KNN) is used for grouping to find the points within a certain range around each point, forming a new point cloud data (i.e., multiple local neighborhoods). Finally, the PointNet network is used to extract local features from multiple said local neighborhoods. The PointNet network includes a multi-layer perceptron and a pooling operation. After feature extraction through the multi-layer perceptron, it is aggregated into a feature vector through the pooling operation, thereby obtaining local features, and mapping the point features of a point cloud into a global point feature vector to represent the features of the entire point cloud.

[0034] In an alternative embodiment, the global feature extraction layer performs a max pooling operation using the PointNet network. In this embodiment, the PointNet network is also used when extracting global features. Specifically, a max pooling operation is performed based on the PointNet network to accurately extract global information.

[0035] S130. Update the model parameters of the to-be-trained model to reduce the loss between the predicted data and the sample data, and obtain a wind resistance prediction model, where the wind resistance prediction model is used to respond to the automotive body point cloud data and predict the automotive wind resistance coefficient.

[0036] In an alternative embodiment, the loss function in the process of updating the model parameters of the to-be-trained model is: , where, MSE is the loss function, y i is the true value of the i th automotive wind resistance sample, is the predicted value of the i th automotive wind resistance sample, m is the number of automotive wind resistance samples.

[0037] The mean square error can directly calculate the average of the squares of the differences between the predicted data and the sample data, reflecting the deviation degree between the predicted value and the true value. Using the mean square error as the loss function can simplify the optimization process and make the calculation more efficient.

[0038] Optionally, when performing model optimization, the learning rate is 0.001, the batch size is 32, and the number of training epochs is 500.

[0039] The method for constructing the above vehicle aerodynamic drag prediction model includes: obtaining sample data of vehicle aerodynamic drag, where the sample data includes vehicle shape point cloud data and vehicle aerodynamic drag coefficient; obtaining prediction data of the sample data in response to the model to be trained, where the model to be trained includes a PointNet++ model; updating the model parameters of the model to be trained to reduce the loss between the prediction data and the sample data, and obtaining an aerodynamic drag prediction model, where the aerodynamic drag prediction model is used to predict the vehicle aerodynamic drag coefficient in response to the vehicle shape point cloud data. This method uses sample data including vehicle shape point cloud data and vehicle aerodynamic drag coefficient to train a specific model to be trained, and then updates the model parameters to obtain an aerodynamic drag prediction model. The aerodynamic drag prediction model can predict the vehicle aerodynamic drag coefficient based on the vehicle shape point cloud data. The vehicle shape point cloud data can accurately represent the vehicle exterior shape. The model has high prediction efficiency. Compared with the traditional CFD simulation method, the prediction time can be reduced by three orders of magnitude. Moreover, the data used is vehicle shape point cloud data, and the PointNet++ model used can effectively extract multi-scale features of the point cloud data and can predict the aerodynamic drag coefficient of complex geometric shapes with high prediction accuracy.

[0040] Further, the present application uses the farthest point sampling algorithm to simplify the vehicle shape point cloud data, further reducing the data volume and improving the prediction efficiency.

[0041] To comprehensively evaluate the performance of the model, the present application uses the following evaluation metrics: (1) Mean Squared Error (MSE): Measures the error between the aerodynamic drag coefficient predicted by the model and the true aerodynamic drag coefficient.

[0042] (2) Mean Absolute Error (MAE): Calculates the average of the absolute differences between the predicted values and the true values.

[0043] (3) R² coefficient of determination: Reflects the prediction effect of the model. The closer the value is to 1, the more accurate the prediction of the model.

[0044] (4) Computation time: Compares the prediction time of the traditional CFD simulation and the model obtained in this embodiment to evaluate the computational efficiency of the model.

[0045] The following table shows the comparison of the aerodynamic drag coefficient prediction results of the model obtained in Example 1 on different datasets with the true values of CFD simulation:

[0046] It can be seen that the model obtained in Example 1 shows high accuracy on multiple datasets, and the R² coefficient of determination is close to 1, indicating that the model can well capture the relationship between the three-dimensional geometry and the aerodynamic drag coefficient.

[0047] The following table shows the comparison of the wind resistance prediction time based on CFD simulation and the model obtained in Example 1 (unit: seconds):

[0048] As can be seen from the table, the traditional CFD simulation time usually takes about 1 hour for the calculation of each object, while the model obtained based on Example 1 only takes about 10 seconds, and the calculation time is reduced by three orders of magnitude. This significant time advantage makes the model more suitable for scenarios of rapid iterative design.

[0049] Example 2 As Figure 2 shown, this embodiment provides a device for constructing an automobile wind resistance prediction model, including: A sample data acquisition module 201, configured to acquire sample data of automobile wind resistance, where the sample data includes automobile shape point cloud data and automobile wind resistance coefficient; A sample data prediction module 202, configured to obtain prediction data of the sample data in response to a model to be trained; the model to be trained includes a PointNet++ model; A wind resistance prediction model determination module 203, configured to update the model parameters of the model to be trained to reduce the loss between the prediction data and the sample data, and obtain a wind resistance prediction model, where the wind resistance prediction model is configured to predict the automobile wind resistance coefficient in response to the automobile shape point cloud data.

[0050] This device is used to execute the above method, and thus has at least corresponding functional modules and beneficial effects as the above method.

[0051] Example 3 As Figure 3 shown, this embodiment provides an electronic device, including: At least one processor; and A memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute the above method. At least one processor in this electronic device can execute the above method, and thus has at least the same advantages as the above method.

[0052] Optionally, the electronic device further includes interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Each component is interconnected using different buses and can be mounted on a common motherboard or otherwise mounted as required. The processor can process instructions executed within the electronic device, including instructions for storing graphical information in the memory or on the memory to display a GUI (Graphical User Interface) on an external input / output device (such as a display device coupled to the interface). In other embodiments, if necessary, multiple processors can be used in conjunction with multiple memories, and / or multiple buses can be used in conjunction with multiple memories. Similarly, multiple electronic devices can be connected (e.g., as a server array, a set of blade servers, or a multiprocessor system), with each device providing part of the necessary operations. Figure 3 Take a processor 301 as an example.

[0053] The memory 302, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the method for constructing an automotive aerodynamic drag prediction model in the embodiments of the present application (e.g., the sample data acquisition module, the sample data prediction module, and the aerodynamic drag prediction model determination module in the device for constructing an automotive aerodynamic drag prediction model). The processor 301 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 302, thereby implementing the above-mentioned method for constructing an automotive aerodynamic drag prediction model.

[0054] The memory 302 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory 302 can include high-speed random access memory and can also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some instances, the memory 302 can further include a memory remotely located relative to the processor 301, and these remote memories can be connected to the device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0055] The electronic device may further include: an input device 303 and an output device 304. The processor 301, the memory 302, the input device 303, and the output device 304 can be connected through a bus or other means. Figure 3 Take the connection through a bus as an example.

[0056] The input device 303 can receive input digital or character information. The output device 304 can include a display device, an auxiliary lighting device (e.g., an LED), a haptic feedback device (e.g., a vibration motor), etc. The display device can include, but is not limited to, a liquid crystal display (LCD), a light-emitting diode (LED) display, and a plasma display. In some embodiments, the display device can be a touch screen.

[0057] Example 4 This embodiment provides a computer-readable storage medium, on which computer instructions are stored. The computer instructions are used to cause a computer to execute the above method. The computer instructions on the computer-readable storage medium are used to cause a computer to execute the above method, and thus have at least the same advantages as the above method.

[0058] The medium in this application can adopt any combination of one or more computer-readable media. The medium can be a computer-readable signal medium or a computer-readable storage medium. The medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the medium (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0059] The computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0060] The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF (Radio Frequency), etc., or any suitable combination of the above.

[0061] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., connected through the Internet using an Internet service provider).

[0062] It should be understood that the various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in this application can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this application can be achieved, and no limitation is imposed herein.

[0063] The above specific embodiments do not constitute a limitation on the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for constructing a vehicle wind resistance prediction model, characterized in that: include: Acquire sample data of automobile wind resistance, wherein the sample data includes automobile appearance point cloud data and automobile wind resistance coefficient; In response to the model to be trained, obtaining prediction data of the sample data; the model to be trained includes a PointNet++ model; The model parameters of the model to be trained are updated to reduce the loss of the predicted data and the sample data, and obtain a drag prediction model, wherein the drag prediction model is used to respond to the vehicle shape point cloud data and predict the vehicle drag coefficient.

2. The method for constructing a vehicle wind resistance prediction model according to claim 1, characterized in that: After obtaining the sample data of the car's wind resistance, it also includes: The farthest point sampling algorithm is used to simplify the automobile shape point cloud data to obtain a sampling point set.

3. The method for constructing a vehicle wind resistance prediction model according to claim 2, characterized in that: The farthest point sampling algorithm is used to simplify the automobile shape point cloud data to obtain a sampling point set, including: Selecting an initial point from the automobile shape point cloud data; Determine a second point according to the Euclidean distance between the initial point and each remaining point; the remaining points are other points in the automobile shape point cloud data except the initial point; the second point is the point with the largest Euclidean distance from the initial point; Determine an initial sampling point set according to the initial point and the second point; The initial sampling point set is normalized to obtain a sampling point set.

4. The method for constructing a vehicle wind resistance prediction model according to claim 3, characterized in that: The PointNet++ model includes at least two set abstraction layers, a global feature extraction layer and a fully connected layer; The set abstraction layer samples and groups the sampling point set to obtain a plurality of local neighborhoods, and then extracts local features from the plurality of local neighborhoods; The global feature extraction layer uses a maximum pooling operation to aggregate the local features to obtain global features; The fully connected layer receives the global features and outputs prediction data.

5. The method for constructing a vehicle wind resistance prediction model according to claim 4, characterized in that: The step of sampling and grouping the sampling point set to obtain a plurality of local neighborhoods, and then extracting local features from the plurality of local neighborhoods includes: The farthest point sampling algorithm is used to sample the sampling point set to obtain a simplified sampling point set; Using a query ball or a classification algorithm to group the simplified sampling point set to obtain a plurality of local neighborhoods; A PointNet network is used to extract local features from the plurality of local neighborhoods.

6. The method for constructing a vehicle wind resistance prediction model according to claim 4, characterized in that: The global feature extraction layer uses the PointNet network to perform maximum pooling operations.

7. The method for constructing a vehicle wind resistance prediction model according to claim 1, characterized in that: The loss function in the process of updating the model parameters of the model to be trained is: ， in, MSE is the loss function, y i For the i The true value of the car wind resistance sample, For the i The predicted value of the wind resistance sample of each car, m is the sample number of the automobile wind resistance sample.

8. A device for constructing a vehicle wind resistance prediction model, characterized in that: include: A sample data acquisition module is used to acquire sample data of automobile wind resistance, wherein the sample data includes automobile appearance point cloud data and automobile wind resistance coefficient; A sample data prediction module, used for obtaining prediction data of the sample data in response to a model to be trained; the model to be trained includes a PointNet++ model; The wind resistance prediction model determination module is used to update the model parameters of the model to be trained to reduce the loss of the prediction data and the sample data, and obtain the wind resistance prediction model, wherein the wind resistance prediction model is used to respond to the vehicle shape point cloud data and predict the vehicle drag coefficient.

9. An electronic device, characterized in that: include: at least one processor, and a memory communicatively coupled to at least one of the processors; The memory stores instructions executable by at least one of the processors, and the instructions are executed by at least one of the processors so that the at least one processor can execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images