Method, device, electronic device and storage medium for determining wind resistance coefficient of vehicle

Abstract

The present application provides a method, apparatus, electronic device, and storage medium for determining the wind resistance coefficient of a vehicle, including: performing standardization processing on the three-dimensional shape heterogeneous point cloud data of the target vehicle to determine a plurality of target three-dimensional point cloud data; performing spatial order encoding processing, attention calculation processing, and sparse convolution processing on the plurality of target three-dimensional point cloud data to determine sparse three-dimensional point cloud data; performing embedded feature processing on the sparse three-dimensional point cloud data to determine the embedded features of the sparse three-dimensional point cloud data; performing downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing on the embedded features to output the global features of each stage processor; inputting the multiple global features into a fully connected layer for multi-scale feature regression prediction to output the wind resistance coefficient of the target vehicle. It realizes the ability to quickly and accurately complete the prediction of the wind resistance coefficient in various actual industrial environments, improving the prediction speed and accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle dynamics, and in particular to a method, device, electronic device and storage medium for determining the drag coefficient of a vehicle. Background Art

[0002] Aerodynamic performance (especially the drag coefficient) has a significant impact on the overall vehicle energy consumption, fuel economy, and cruising range, etc., and is the goal of design optimization in many industrial fields such as aerospace, automotive manufacturing, and high-speed trains. With the continuous improvement of the requirements for industrial design accuracy and iteration efficiency, traditional aerodynamic analysis technologies are facing huge challenges: Although numerical simulation methods represented by Computational Fluid Dynamics (CFD) have clear physical mechanisms, the computational cost increases rapidly with the complexity of the problem, making it difficult to meet the urgent needs of the industrial community for rapid and even real-time design optimization. One simulation or experiment often takes several hours or even days, and the repeated iteration of complex three-dimensional models is a huge investment in time and cost. In recent years, with the rapid development of deep learning technology, several neural network-based surrogate methods have emerged to replace CFD simulations. However, most existing methods still rely on low-dimensional shape parameters or 2D views and are difficult to fully capture the complex flow information in the real 3D shape. In addition, in the actual industrial environment, the diverse vehicle sizes and inconsistent grid densities will double the difficulty of directly learning the "geometry → drag coefficient" mapping. Therefore, how to improve the accuracy of determining the drag coefficient of a vehicle has become a technical problem that cannot be underestimated. Summary of the Invention

[0003] In view of this, the purpose of the present application is to provide a method, device, electronic device and storage medium for determining the drag coefficient of a vehicle. Through the standardized processing of three-dimensional point cloud heterogeneous data and the vehicle drag coefficient prediction model, it is possible to quickly and accurately complete the prediction of the drag coefficient in a variety of actual industrial environments, improving the prediction speed and accuracy.

[0004] An embodiment of the present application provides a method for determining the drag coefficient of a vehicle, characterized in that the drag coefficient determination method includes:

[0005] Resample the three-dimensional shape heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle;

[0006] Input multiple pieces of the target three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data;

[0007] Input the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedded feature processing to determine the embedded features of the sparse three-dimensional point cloud data;

[0008] Input the embedded features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and output the global features of each stage processor;

[0009] Input multiple global features into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and output the drag coefficient of the target vehicle; wherein, the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network.

[0010] In a possible implementation manner, the resampling processing of the three-dimensional heterogeneous point cloud data of the target vehicle, and the coordinate normalization processing and voxel pooling processing of the resampled three-dimensional heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle include:

[0011] Perform resampling processing on the three-dimensional heterogeneous point cloud data based on grid sampling and uniform sampling to determine the resampled three-dimensional heterogeneous point cloud data;

[0012] Establish a reference coordinate system based on the center points of the front and rear axles of the target vehicle;

[0013] Perform translation transformation on the resampled three-dimensional heterogeneous point cloud data based on the reference coordinate system, and perform scale scaling on the resampled three-dimensional heterogeneous point cloud data based on the vehicle length of the target vehicle as a reference dimension to determine the three-dimensional heterogeneous point cloud data after coordinate normalization processing;

[0014] Perform voxel pooling processing on the three-dimensional heterogeneous point cloud data after coordinate normalization processing to determine multiple target three-dimensional point cloud data.

[0015] In a possible implementation manner, the input of multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model for spatial order encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data includes:

[0016] Perform spatial order encoding processing on each target three-dimensional point cloud data based on a space-filling curve to determine the target three-dimensional point cloud data after encoding processing;

[0017] Perform attention calculation processing on the encoded target three-dimensional point cloud data to determine the target three-dimensional point cloud data after attention calculation processing;

[0018] Perform sparse convolution processing on the target three-dimensional point cloud data after attention calculation processing to determine the sparse three-dimensional point cloud data.

[0019] In a possible implementation manner, inputting the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle aerodynamic coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and outputting the global features of each stage processor, including:

[0020] Input the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling processing, position encoding processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and output the global features of the first-stage encoder;

[0021] Input the global features of the first-stage encoder into the next-stage encoder for processing, and output the global features of the next-stage encoder.

[0022] In a possible implementation manner, inputting the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling processing, position encoding processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and outputting the global features of the first-stage encoder, including:

[0023] Perform downsampling processing on the embedding features, and perform adjacent point cloud feature aggregation processing on the downsampled embedding features based on serial encoding to determine the features of the new three-dimensional point cloud data;

[0024] Perform normalization processing on the features of the new three-dimensional point cloud data and the downsampled embedding features to determine the first features of the three-dimensional point cloud data;

[0025] Perform attention processing on the first features of the three-dimensional point cloud data to determine the second features of the three-dimensional point cloud data, and perform normalization processing on the second features and the features of the three-dimensional point cloud data to determine the third features of the three-dimensional point cloud data;

[0026] Perform perceptron processing on the third features to determine the fourth features of the three-dimensional point cloud data;

[0027] Perform global pooling processing on the fourth features and the second features, and output the global features.

[0028] In a possible implementation, the vehicle aerodynamic drag coefficient prediction model is determined through the following steps:

[0029] Input the sample three-dimensional point cloud data of multiple sample vehicles into a multi-scale Transformer network, process the sample three-dimensional point cloud data of each sample vehicle, and predict the predicted vehicle aerodynamic drag coefficient of each sample vehicle;

[0030] Calculate the mean square error and the mean absolute error for the predicted vehicle aerodynamic drag coefficient and the actual vehicle aerodynamic drag coefficient of each sample vehicle, and determine the mean square error value and the mean absolute error value of the multi-scale Transformer network;

[0031] If any of the mean square error value and the mean absolute error value is greater than the corresponding preset threshold, optimize the network parameters of the multi-scale Transformer network through the backpropagation algorithm, and continue to perform iterative training on the optimized multi-scale Transformer network;

[0032] If both the mean square error value and the mean absolute error value are less than or equal to the corresponding preset threshold, use the multi-scale Transformer network as the vehicle aerodynamic drag coefficient prediction model.

[0033] In a possible implementation, the three-dimensional shape heterogeneous point cloud data is obtained through the following steps:

[0034] Perform computational fluid dynamics numerical simulation processing on the target vehicle after grid division to determine the three-dimensional flow field data of the target vehicle;

[0035] Normalize the shape data of the target vehicle based on the lowest grid points in the three-dimensional flow field data to determine the normalized shape data;

[0036] Perform fixed-dimension conversion processing, outlier removal processing, and mean normalization processing on the normalized shape data to determine the three-dimensional shape heterogeneous point cloud data of the target vehicle.

[0037] The embodiment of the present application also provides a device for determining the aerodynamic drag coefficient of a vehicle. The device for determining the aerodynamic drag coefficient includes:

[0038] A standardization processing module, configured to perform resampling processing on the three-dimensional shape heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle;

[0039] A data preprocessing module for inputting multiple pieces of the target 3D point cloud data into the data preprocessing module of a vehicle drag coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse 3D point cloud data;

[0040] A feature embedding module for inputting the sparse 3D point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedded feature processing to determine the embedded features of the sparse 3D point cloud data;

[0041] A multi-stage encoder module for inputting the embedded features of the sparse 3D point cloud data into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing to output the global features of each stage processor;

[0042] A regression prediction module for inputting multiple pieces of the global features into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction to output the drag coefficient of the target vehicle; wherein, the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network.

[0043] An embodiment of the present application further provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the method for determining the drag coefficient of a vehicle as described above are executed.

[0044] An embodiment of the present application further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is run by a processor, the steps of the method for determining the drag coefficient of a vehicle as described above are executed.

[0045] The method, device, electronic device and storage medium for determining the drag coefficient of a vehicle provided by the embodiments of the present application, the method for determining the drag coefficient includes: resampling the three-dimensional shape heterogeneous point cloud data of the target vehicle, and performing coordinate normalization processing and voxel pooling processing on the resampled three-dimensional shape heterogeneous point cloud data to determine a plurality of target three-dimensional point cloud data of the target vehicle; inputting the plurality of target three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model for spatial sequence encoding processing, attention calculation processing and sparse convolution processing to determine sparse three-dimensional point cloud data; inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data; inputting the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing and global pooling processing, and outputting the global features of each stage processor; inputting the plurality of global features into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and outputting the drag coefficient of the target vehicle; wherein, the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network. Through the standardization processing of three-dimensional point cloud heterogeneous data and the vehicle drag coefficient prediction model, it is possible to quickly and accurately complete the prediction of the drag coefficient in a variety of actual industrial environments, improving the prediction speed and accuracy.

[0046] To make the above objects, features and advantages of the present application more obvious and understandable, the following specific embodiments are given, and in conjunction with the accompanying drawings, the detailed description is as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0048] Figure 1 It is a flowchart of a method for determining the drag coefficient of a vehicle provided by an embodiment of the present application;

[0049] Figure 2 It is a schematic diagram of a method for determining the drag coefficient of a vehicle provided by an embodiment of the present application;

[0050] Figure 3 It is one of the structural schematic diagrams of a device for determining the drag coefficient of a vehicle provided by an embodiment of the present application;

[0051] Figure 4 This is the second schematic structural diagram of an air resistance coefficient determination device for a vehicle provided by an embodiment of the present application;

[0052] Figure 5 This is the schematic structural diagram of an electronic device provided by an embodiment of the present application. Specific embodiments

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some, rather than all, of the embodiments of the present application. Components of the embodiments of the present application generally described and illustrated in the figures herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application that is claimed, but is merely representative of selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0054] First, the applicable application scenarios of the present application are introduced. The present application can be applied to the technical field of vehicle dynamics.

[0055] Through research, it is found that aerodynamic performance (especially the drag coefficient) has a significant impact on the overall vehicle energy consumption, fuel economy, and cruising range, etc., and is the goal of design optimization in many industrial fields such as aerospace, automotive manufacturing, and high-speed trains. With the continuous improvement of the requirements for industrial design accuracy and iteration efficiency, traditional aerodynamic analysis technologies are facing huge challenges: Numerical simulation methods represented by Computational Fluid Dynamics (CFD) have clear physical mechanisms, but the computational cost increases rapidly with the complexity of the problem, making it difficult to meet the urgent needs of the industrial community for rapid and even real-time design optimization. One simulation or experiment often takes several hours or even days, and the repeated iteration of complex three-dimensional models requires a huge investment of time and cost. In recent years, deep learning technologies have developed rapidly, and several neural network-based surrogate methods have emerged to replace CFD simulations. However, most existing methods still rely on low-dimensional shape parameters or 2D views and are difficult to fully capture the complex flow information in the real 3D shape. In addition, in the actual industrial environment, the diverse vehicle sizes and inconsistent grid densities will all double the difficulty of directly learning the "geometry → drag coefficient" mapping. Therefore, how to improve the accuracy of determining the air resistance coefficient of a vehicle has become a technical problem that cannot be underestimated.

[0056] Based on this, the embodiments of the present application provide a method for determining the aerodynamic drag coefficient of a vehicle. Through the standardized processing of three-dimensional point cloud heterogeneous data and the vehicle aerodynamic drag coefficient prediction model, it is possible to quickly and accurately complete the prediction of the aerodynamic drag coefficient in a variety of actual industrial environments, improving the prediction speed and accuracy.

[0057] Please refer to Figure 1 , Figure 1 which is a flowchart of a method for determining the aerodynamic drag coefficient of a vehicle provided by the embodiments of the present application. As Figure 1 shown in

[0058] S101: Resample the three-dimensional shape heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle.

[0059] In this step, resample the three-dimensional shape heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle.

[0060] Here, by performing resampling processing, coordinate normalization processing, and voxel pooling processing on the target three-dimensional shape heterogeneous point cloud data, the standardized processing of the target three-dimensional shape heterogeneous point cloud data is completed.

[0061] In a possible implementation manner, the three-dimensional shape heterogeneous point cloud data is obtained through the following steps:

[0062] A: Perform computational fluid dynamics numerical simulation processing on the target vehicle after mesh division to determine the three-dimensional flow field data of the target vehicle.

[0063] Here, various automotive 3D shape structures are obtained, and then the automotive shape is discretized into meshes at a specified resolution. According to the set target oncoming flow and boundary conditions, perform CFD numerical simulation on the automotive shape of the target vehicle to generate the three-dimensional flow field data of the target shape of the target vehicle.

[0064] Specifically, set the regional oncoming flow and boundary conditions: for the target calculation region, set the boundary oncoming flow conditions. Further, set the inflow direction and wind speed. Set the bottom surface of the computational domain as a no-slip boundary, and the rest as free boundary conditions. According to the initial and boundary conditions, use the RANS model to perform numerical simulation on the target computational domain. After the flow becomes stable, select the three-dimensional flow field result at the final fixed time step as the three-dimensional flow field data corresponding to the target region.

[0065] B: Normalize the external shape data of the target vehicle based on the lowest grid point in the three-dimensional flow field data to determine the normalized external shape data.

[0066] Here, subtract the minimum value of the external shape data of the target vehicle to unify the relative height, and determine the normalized external shape data.

[0067] C: Perform fixed-dimension conversion processing, outlier removal processing, and mean normalization processing on the normalized external shape data to determine the three-dimensional external shape heterogeneous point cloud data of the target vehicle.

[0068] Here, convert the normalized external shape data into a unified format through linear interpolation to ensure that it can adapt to the input requirements of the model. Eliminate the cases where the CFD calculation does not converge through outlier identification to ensure data quality.

[0069] Among them, the formula for mean normalization processing is:

[0070]

[0071] Among them, is the three-dimensional external shape heterogeneous point cloud data, is the original three-dimensional external shape heterogeneous point cloud data, is the statistical mean of the external shape data of the target vehicle, is the statistical standard deviation of the external shape data of the target vehicle.

[0072] In a possible implementation manner, resample the three-dimensional external shape heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional external shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle, including:

[0073] (1): Resample the three-dimensional external shape heterogeneous point cloud data based on grid sampling and uniform sampling to determine the resampled three-dimensional external shape heterogeneous point cloud data.

[0074] Here, due to the difference in point cloud density of different vehicle models, this application combines grid sampling and uniform sampling. Specifically, first sample with the original grid points as the basic points to ensure the full representation of local geometric details; then uniformly sample the vehicle body to control the overall number of points, avoid the excessive influence of high-density areas on global features, ensure that local fine features such as high-curvature areas are fully expressed, and at the same time avoid the overrepresentation of dense areas, eliminate the density difference of multi-source point cloud data, and achieve unified feature distribution. After processing, the data storage overhead is reduced, the key geometric details are retained, and the neural network can focus more on the modeling of the correlation between geometry and flow field, significantly improving the feature generalization ability of the model.

[0075] B: Establish a reference coordinate system based on the center points of the front and rear axles of the target vehicle; perform a translation transformation on the resampled three-dimensional heterogeneous point cloud data based on the reference coordinate system, and perform a scaling on the resampled three-dimensional heterogeneous point cloud data with the vehicle length of the target vehicle as a reference dimension to determine the three-dimensional heterogeneous point cloud data after coordinate normalization processing.

[0076] Here, the purpose of coordinate normalization processing is to eliminate the feature distribution bias caused by the scale differences of multi-source vehicle type data and improve the focusing ability of the neural network on the key geometric topology differences. This application realizes geometric standardization through rigid body registration and affine scaling: First, establish a reference coordinate system based on the center points of the front and rear axles, and align the wheelbase through translation transformation to eliminate the representational ambiguity caused by spatial displacement; then select the wheelbase or vehicle length as a reference dimension for scaling, and map each vehicle type to a dimensionless space. This processing mechanism effectively suppresses the interference of absolute size on the feature learning process by constraining the scale invariance of geometric representations, making the point clouds of different vehicle types show comparability in key geometric dimensions such as curvature distribution and local concave and convex features, and significantly reducing the optimization difficulty of model generalization.

[0077] C: Perform voxel pooling processing on the three-dimensional heterogeneous point cloud data after coordinate normalization processing to determine multiple pieces of the target three-dimensional point cloud data.

[0078] Here, performing voxel pooling processing on the three-dimensional heterogeneous point cloud data after coordinate normalization processing reduces the amount of data to be processed, so that the features of the point cloud in each voxel can reflect the geometric information of the local area.

[0079] S102: Input multiple pieces of the target three-dimensional point cloud data into the data preprocessing module of the vehicle aerodynamic drag coefficient prediction model to perform spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data.

[0080] In this step, input multiple pieces of the target three-dimensional point cloud data into the data preprocessing module of the vehicle aerodynamic drag coefficient prediction model to perform spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data.

[0081] In a possible implementation manner, the inputting multiple pieces of the target three-dimensional point cloud data into the data preprocessing module of the vehicle aerodynamic drag coefficient prediction model to perform spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data includes:

[0082] a: Perform spatial order encoding processing on each of the target three-dimensional point cloud data based on a space-filling curve to determine the encoded target three-dimensional point cloud data.

[0083] Here, use a space-filling curve (such as a Z-order curve, Hilbert curve, etc.) to perform spatial order encoding processing on each target three-dimensional point cloud data according to the spatial order of the shape of the target vehicle, so as to complete the sorting of the target three-dimensional point cloud data and obtain the encoded target three-dimensional point cloud data.

[0084] b: Perform attention calculation processing on the encoded target three-dimensional point cloud data to determine the target three-dimensional point cloud data after attention calculation processing.

[0085] Here, perform attention calculation processing on the encoded target three-dimensional point cloud data to better capture spatial neighborhood information.

[0086] c: Perform sparse convolution processing on the target three-dimensional point cloud data after attention calculation processing to determine the sparse three-dimensional point cloud data.

[0087] Here, use sparse convolution to convert the target three-dimensional point cloud data after attention calculation processing into a sparse convolution tensor. In a three-dimensional space where the point cloud is relatively scattered, sparse convolution can significantly reduce invalid convolution operations and improve memory utilization and calculation efficiency.

[0088] S103: Input the sparse three-dimensional point cloud data into the feature embedding module of the vehicle aerodynamic drag coefficient prediction model for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data.

[0089] In this step, input the sparse three-dimensional point cloud data into the feature embedding module for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data.

[0090] Among them, the embedding feature processing is to perform convolution transformation and normalization processing on the features (such as coordinates, normals, intensities, additional parameters, etc.) of the sparse three-dimensional point cloud data to obtain a unified embedding feature representation. Specifically, it includes: using three-dimensional convolution to perform convolution operations on the sparse three-dimensional point cloud data to capture local geometric patterns, then further normalizing and enhancing the non-linear expression; finally, obtaining embedding features of a specified dimension for subsequent use by the encoder.

[0091] S104: Input the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle aerodynamic drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and output the global features of each stage processor.

[0092] In this step, the embedded features of the sparse three-dimensional point cloud data are input into a multi-stage encoder for downsampling, aggregation, normalization, attention, perceptron, and global pooling operations, and the global features of each stage processor are output.

[0093] Among them, each stage encoder consists of downsampling and a Block stack.

[0094] In a possible implementation, the operation of inputting the embedded features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle aerodynamic drag coefficient prediction model for downsampling, aggregation, normalization, attention, perceptron, and global pooling operations, and outputting the global features of each stage processor, includes:

[0095] (1): Input the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling, position encoding, normalization, attention, perceptron, and global pooling operations, and output the global features of the first-stage encoder.

[0096] Here, first input the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling, position encoding, normalization, attention, perceptron, and global pooling operations, and output the global features of the first-stage encoder.

[0097] In a possible implementation, the operation of inputting the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling, position encoding, normalization, attention, perceptron, and global pooling operations, and outputting the global features of the first-stage encoder, includes:

[0098] i: Perform downsampling on the embedded features, and perform neighboring point cloud feature aggregation on the downsampled embedded features based on sequence encoding to determine the features of the new three-dimensional point cloud data.

[0099] Here, the purpose of downsampling is to perform hierarchical abstraction at the point cloud level, "aggregate" the fine-grained point cloud to a coarser spatial scale, reduce the number of points, and highlight important features. Search for spatially neighboring points through sequence encoding, and merge them according to "clusters" for feature aggregation, and output the features of the new three-dimensional point cloud data.

[0100] ii: Perform normalization on the features of the new three-dimensional point cloud data and the downsampled embedded features to determine the first features of the three-dimensional point cloud data.

[0101] Here, the purpose of performing normalization on the features of the new three-dimensional point cloud data and the downsampled embedded features is to retain the expressive ability of the features.

[0102] iii: Perform attention processing on the first feature of the three-dimensional point cloud data to determine the second feature of the three-dimensional point cloud data, and perform normalization processing on the second feature and the features of the three-dimensional point cloud data to determine the third feature of the three-dimensional point cloud data.

[0103] Here, the purpose of the attention processing is to divide the point cloud into several slices, and perform multi-head attention on the points within the same slice. This attention can use FlashAttention to accelerate the attention operation of large-scale sequences.

[0104] iv: Perform perceptron processing on the third feature to determine the fourth feature of the three-dimensional point cloud data; perform global pooling processing on the fourth feature and the second feature, and output the global feature.

[0105] Here, a multi-layer perceptron (MLP) is used to expand the channel dimension, enhance the non-linearity, and then project back to the original channel dimension. After each stage, the model will perform global pooling on the features to aggregate all the features in the current Batch into a global vector.

[0106] In this application, after multiple stage encoders are connected in series, richer local-global hybrid features can be gradually refined. Let the initial embedded feature be , and the feature after being processed by the -th stage encoder is:

[0107]

[0108] where is all the features in the k -th stage encoder ( ), is all the features in the k -1-th stage encoder, and S is the total number of stage encoders.

[0109] Among them, at each scale stage, scale-related global features are extracted through global pooling operations :

[0110]

[0111] where is the global feature output by the k -th stage encoder, and is the global pooling processing.

[0112] (2): Input the global features of the first-stage encoder into the next-stage encoder for processing, and output the global features of the next-stage encoder.

[0113] Here, input the global features of the first-stage encoder into the next-stage encoder for processing, and output the global features of the next-stage encoder. Repeat the above steps until each stage encoder has processed the global features and then stop processing.

[0114] S105: Input the multiple global features into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and output the drag coefficient of the target vehicle; wherein, the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network.

[0115] In this step, the problem of insufficient modeling of a single-scale network is avoided. In the fully connected layer, multi-scale feature regression prediction is performed on multiple global features to output the drag coefficient of the target vehicle.

[0116] Among them, constructing multi-scale features for multiple global features is expressed as:

[0117]

[0118] Here, linear processing, batch normalization processing, activation function processing, and multi-layer perceptron processing are performed on the multi-scale features in the fully connected layer to obtain the drag coefficient of the target vehicle.

[0119] In a possible implementation manner, the vehicle drag coefficient prediction model is determined through the following steps:

[0120] I: Input the sample three-dimensional point cloud data of multiple sample vehicles into the multi-scale Transformer network, process the sample three-dimensional point cloud data of each sample vehicle, and predict the predicted vehicle drag coefficient of each sample vehicle.

[0121] Here, input the sample three-dimensional point cloud data of multiple sample vehicles into the multi-scale Transformer network, process the sample three-dimensional point cloud data of each sample vehicle, and predict the predicted vehicle drag coefficient of each sample vehicle.

[0122] Among them, the processing process of predicting the vehicle drag coefficient is consistent with the processing process of the above vehicle drag coefficient, and this part will not be elaborated here.

[0123] Here, a new multi-scale Transformer architecture is designed based on the Transformer network, and a global pooling module is embedded between the deep encoders to improve the model feature learning ability.

[0124] II: Calculate the mean square error and the mean absolute error for the predicted vehicle drag coefficient and the actual vehicle drag coefficient of each of the said sample vehicles, and determine the mean square error value and the mean absolute error value of the said multi-scale Transformer network.

[0125] Here, the mean square error value is determined by the following formula and the mean absolute error value :

[0126]

[0127]

[0128] where, is the actual vehicle drag coefficient, is the predicted vehicle drag coefficient, m is the number of sample vehicles.

[0129] III: If any one of the said mean square error value and the said mean absolute error value is greater than the corresponding preset threshold, optimize the network parameters of the said multi-scale Transformer network by the backpropagation algorithm, and continue to perform iterative training on the optimized said multi-scale Transformer network.

[0130] Here, in addition to optimizing the network parameters of the said multi-scale Transformer network by the backpropagation algorithm, techniques such as hyperparameter tuning, network parameter adjustment and regularization can also be used to improve the prediction accuracy of the model under different vehicle shapes.

[0131] IV: If both the said mean square error value and the said mean absolute error value are less than or equal to the corresponding preset thresholds, then regard the said multi-scale Transformer network as the said vehicle drag coefficient prediction model.

[0132] In this application, by proposing a new network structure and algorithm, the prediction speed and accuracy are improved, while the adaptability and generality of the model are enhanced, the dependence on a large amount of historical data and high-performance hardware is reduced, and an efficient, flexible and easy-to-deploy solution for predicting the drag coefficient is provided.

[0133] Further, please refer to Figure 2 , Figure 2 which is a schematic diagram of a method for determining the drag coefficient of a vehicle provided by an embodiment of this application. As Figure 2As shown in the figure, resampling processing, coordinate normalization processing, and voxel pooling processing are performed on the three-dimensional heterogeneous point cloud data of the vehicle to obtain target three-dimensional point cloud data. The target three-dimensional point cloud data is input into the data preprocessing module to obtain sparse three-dimensional point cloud data, and the sparse three-dimensional point cloud data is input into the feature embedding module to obtain embedded features. The embedded features are input into the multi-stage encoder for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing to obtain multiple global features. The multiple global features are input into the fully connected layer to output the wind resistance coefficient of the target vehicle.

[0134] A method for determining the wind resistance coefficient of a vehicle provided by an embodiment of the present application. The method for determining the wind resistance coefficient includes: performing resampling processing on the three-dimensional heterogeneous point cloud data of the target vehicle, and performing coordinate normalization processing and voxel pooling processing on the resampled three-dimensional heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle; inputting the multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle wind resistance coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data; inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle wind resistance coefficient prediction model for embedded feature processing to determine the embedded features of the sparse three-dimensional point cloud data; inputting the embedded features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle wind resistance coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing to output the global features of each stage processor; inputting the multiple global features into the fully connected layer of the vehicle wind resistance coefficient prediction model for multi-scale feature regression prediction to output the wind resistance coefficient of the target vehicle; wherein, the vehicle wind resistance coefficient prediction model is obtained by iteratively training a multi-scale Transformer network. Through the standardization processing of three-dimensional point cloud heterogeneous data and the vehicle wind resistance coefficient prediction model, it is possible to quickly and accurately complete the prediction of the wind resistance coefficient in various actual industrial environments, improving the prediction speed and accuracy.

[0135] Please refer to Figure 3 、 Figure 4 , Figure 3 which is one of the structural schematic diagrams of a device for determining the wind resistance coefficient of a vehicle provided by an embodiment of the present application; Figure 4 which is the second structural schematic diagram of a device for determining the wind resistance coefficient of a vehicle provided by an embodiment of the present application. As Figure 3 shown in the figure, the device 300 for determining the wind resistance coefficient of the vehicle includes:

[0136] The normalization processing module 310 is used to resample the three-dimensional heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle;

[0137] The data preprocessing module 320 is used to input the multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data;

[0138] The feature embedding module 330 is used to input the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data;

[0139] The multi-stage encoder module 340 is used to input the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and output the global features of each stage processor;

[0140] The regression prediction module 350 is used to input the multiple global features into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction and output the drag coefficient of the target vehicle; wherein, the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network.

[0141] Further, when the normalization processing module 310 is used to resample the three-dimensional heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle, the normalization processing module 310 specifically is used for:

[0142] Based on grid sampling and uniform sampling, resample the three-dimensional heterogeneous point cloud data to determine the resampled three-dimensional heterogeneous point cloud data;

[0143] Establish a reference coordinate system based on the center points of the front and rear axles of the target vehicle;

[0144] Based on the reference coordinate system, perform translation transformation on the resampled three-dimensional heterogeneous point cloud data, and based on the vehicle length of the target vehicle as a reference dimension, perform scale scaling on the resampled three-dimensional heterogeneous point cloud data to determine the three-dimensional heterogeneous point cloud data after coordinate normalization processing;

[0145] Perform voxel pooling on the three-dimensional shape heterogeneous point cloud data after coordinate normalization to determine multiple pieces of the target three-dimensional point cloud data.

[0146] Further, when the data preprocessing module 320 is used to input multiple pieces of the target three-dimensional point cloud data into the data preprocessing module of the vehicle aerodynamic drag coefficient prediction model for spatial order encoding processing, attention calculation processing, and sparse convolution processing to determine the sparse three-dimensional point cloud data, the data preprocessing module 320 is specifically used for:

[0147] Perform spatial order encoding processing on each piece of the target three-dimensional point cloud data based on a space-filling curve to determine the target three-dimensional point cloud data after encoding processing;

[0148] Perform attention calculation processing on the target three-dimensional point cloud data after encoding processing to determine the target three-dimensional point cloud data after attention calculation processing;

[0149] Perform sparse convolution processing on the target three-dimensional point cloud data after attention calculation processing to determine the sparse three-dimensional point cloud data.

[0150] Further, when the multi-stage encoder module 340 is used to input the embedded features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle aerodynamic drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing to output the global features of each stage processor, the multi-stage encoder module 340 is specifically used for:

[0151] Input the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling processing, position encoding processing, normalization processing, attention processing, perceptron processing, and global pooling processing to output the global features of the first-stage encoder;

[0152] Input the global features of the first-stage encoder into the next-stage encoder for processing to output the global features of the next-stage encoder.

[0153] Further, when the multi-stage encoder module 340 is used to input the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling processing, position encoding processing, normalization processing, attention processing, perceptron processing, and global pooling processing to output the global features of the first-stage encoder, the multi-stage encoder module 340 is specifically used for:

[0154] Perform downsampling processing on the embedded features, and perform adjacent point cloud feature aggregation processing on the downsampled embedded features based on serialized encoding to determine the features of the new three-dimensional point cloud data;

[0155] Normalize the features of the new three-dimensional point cloud data and the embedded features after the downsampling to determine the first feature of the three-dimensional point cloud data;

[0156] Perform attention processing on the first feature of the three-dimensional point cloud data to determine the second feature of the three-dimensional point cloud data, and normalize the second feature and the features of the three-dimensional point cloud data to determine the third feature of the three-dimensional point cloud data;

[0157] Perform perceptron processing on the third feature to determine the fourth feature of the three-dimensional point cloud data;

[0158] Perform global pooling processing on the fourth feature and the second feature to output the global feature.

[0159] Further, as Figure 4 shown, the vehicle drag coefficient determination device 300 further includes a model training module 360, and the model training module 360 is used for:

[0160] Input the sample three-dimensional point cloud data of multiple sample vehicles into the multi-scale Transformer network, process the sample three-dimensional point cloud data of each sample vehicle, and predict the predicted vehicle drag coefficient of each sample vehicle;

[0161] Calculate the mean square error and the mean absolute error of the predicted vehicle drag coefficient and the actual vehicle drag coefficient of each sample vehicle to determine the mean square error value and the mean absolute error value of the multi-scale Transformer network;

[0162] If any one of the mean square error value and the mean absolute error value is greater than the corresponding preset threshold, optimize the network parameters of the multi-scale Transformer network through the backpropagation algorithm, and continue to perform iterative training on the optimized multi-scale Transformer network;

[0163] If both the mean square error value and the mean absolute error value are less than or equal to the corresponding preset threshold, use the multi-scale Transformer network as the vehicle drag coefficient prediction model.

[0164] Further, as Figure 4 shown, the vehicle drag coefficient determination device 300 further includes a data acquisition module 370, and the data acquisition module 370 is used for:

[0165] Perform computational fluid dynamics numerical simulation processing on the target vehicle after grid division to determine the three-dimensional flow field data of the target vehicle;

[0166] Normalize the external shape data of the target vehicle based on the lowest grid point in the three-dimensional flow field data to determine the normalized external shape data;

[0167] Perform fixed-dimension conversion processing, outlier removal processing, and mean normalization processing on the normalized external shape data to determine the three-dimensional external shape heterogeneous point cloud data of the target vehicle.

[0168] A wind resistance coefficient determination device for a vehicle provided by an embodiment of the present application. The wind resistance coefficient determination device includes: a standardization processing module, configured to resample the three-dimensional external shape heterogeneous point cloud data of the target vehicle, and perform coordinate normalization processing and voxel pooling processing on the resampled three-dimensional external shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle; a data preprocessing module, configured to input the multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle wind resistance coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data; a feature embedding module, configured to input the sparse three-dimensional point cloud data into the feature embedding module of the vehicle wind resistance coefficient prediction model for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data; a multi-stage encoder module, configured to input the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle wind resistance coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and output the global features of each stage processor; a regression prediction module, configured to input the multiple global features into the fully connected layer of the vehicle wind resistance coefficient prediction model for multi-scale feature regression prediction, and output the wind resistance coefficient of the target vehicle; wherein, the vehicle wind resistance coefficient prediction model is obtained by iteratively training a multi-scale Transformer network. Through the standardization processing of three-dimensional point cloud heterogeneous data and the vehicle wind resistance coefficient prediction model, it is possible to quickly and accurately complete the prediction of the wind resistance coefficient in a variety of actual industrial environments, improving the prediction speed and accuracy.

[0169] Please refer to Figure 5 , Figure 5 which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 5 shown in

[0170] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 runs, the processor 510 communicates with the memory 520 through the bus 530. When the machine-readable instructions are executed by the processor 510, they can execute as described above Figure 1and Figure 2 The steps of the method for determining the drag coefficient of a vehicle in the method embodiments shown. For the specific implementation manners, reference may be made to the method embodiments and will not be elaborated herein.

[0171] The embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it can execute the steps of the method for determining the drag coefficient of a vehicle in the method embodiments shown above Figure 1 and Figure 2 The steps of the method for determining the drag coefficient of a vehicle in the method embodiments shown. For the specific implementation manners, reference may be made to the method embodiments and will not be elaborated herein.

[0172] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0173] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division manners in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some communication interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0174] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0175] In addition, in each embodiment of the present application, the functional units can be integrated into one processing unit, or each unit exists physically alone, or two or more units can be integrated into one unit.

[0176] When the above-described functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0177] Finally, it should be noted that the above-described embodiments are only specific implementation manners of the present application, used to illustrate the technical solutions of the present application, rather than limiting them. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed in the present application can still modify the technical solutions described in the foregoing embodiments, or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for determining the drag coefficient of a vehicle, characterized in that, The method for determining the drag coefficient includes: Resampling the three-dimensional heterogeneous point cloud data of the target vehicle, and performing coordinate normalization processing and voxel pooling processing on the resampled three-dimensional heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle; Inputting the multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data; Inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedded feature processing to determine the embedded features of the sparse three-dimensional point cloud data; wherein, the embedded feature processing is to perform convolution transformation and normalization processing on the features of the sparse three-dimensional point cloud data to obtain a unified representation of the embedded features; Inputting the embedded features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and outputting the global features of each stage processor; Inputting the multiple global features into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction to output the drag coefficient of the target vehicle; wherein, the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network; Determining the multiple target three-dimensional point cloud data of the target vehicle through the following steps: Performing resampling processing on the three-dimensional heterogeneous point cloud data based on grid sampling and uniform sampling to determine the resampled three-dimensional heterogeneous point cloud data; Establishing a reference coordinate system based on the center points of the front and rear axles of the target vehicle; Performing translation transformation on the resampled three-dimensional heterogeneous point cloud data based on the reference coordinate system, and performing scale scaling on the resampled three-dimensional heterogeneous point cloud data using the vehicle length of the target vehicle as a reference dimension to determine the three-dimensional heterogeneous point cloud data after coordinate normalization processing; Performing voxel pooling processing on the three-dimensional heterogeneous point cloud data after coordinate normalization processing to determine multiple target three-dimensional point cloud data.

2. The method for determining the drag coefficient according to claim 1, characterized in that The step of inputting the multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data includes: Performing spatial sequence encoding processing on each target three-dimensional point cloud data based on a space-filling curve to determine the target three-dimensional point cloud data after encoding processing; Performing attention calculation processing on the target three-dimensional point cloud data after encoding processing to determine the target three-dimensional point cloud data after attention calculation processing; Performing sparse convolution processing on the target three-dimensional point cloud data after attention calculation processing to determine the sparse three-dimensional point cloud data.

3. The method for determining the drag coefficient according to claim 1, characterized in that, Input the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle aerodynamic drag coefficient prediction model for downsampling, aggregation, normalization, attention, perceptron, and global pooling processes, and output the global features of each stage processor, including: Input the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling, position encoding, normalization, attention, perceptron, and global pooling processes, and output the global features of the first-stage encoder; Input the global features of the first-stage encoder into the next-stage encoder for processing, and output the global features of the next-stage encoder.

4. The method for determining the drag coefficient according to claim 3, wherein The step of inputting the features of the sparse three-dimensional point cloud data into the first-stage encoder for downsampling, position encoding, normalization, attention, perceptron, and global pooling processes, and outputting the global features of the first-stage encoder includes: Perform downsampling on the embedding features, and perform adjacent point cloud feature aggregation on the downsampled embedding features based on sequential encoding to determine the features of the new three-dimensional point cloud data; Perform normalization on the features of the new three-dimensional point cloud data and the downsampled embedding features to determine the first features of the three-dimensional point cloud data; Perform attention processing on the first features of the three-dimensional point cloud data to determine the second features of the three-dimensional point cloud data, and perform normalization on the second features and the features of the three-dimensional point cloud data to determine the third features of the three-dimensional point cloud data; Perform perceptron processing on the third features to determine the fourth features of the three-dimensional point cloud data; Perform global pooling on the fourth features and the second features, and output the global features.

5. The method for determining the drag coefficient according to claim 1, wherein Determine the vehicle aerodynamic drag coefficient prediction model through the following steps: Input the sample three-dimensional point cloud data of multiple sample vehicles into the multi-scale Transformer network, process the sample three-dimensional point cloud data of each sample vehicle, and predict the predicted vehicle aerodynamic drag coefficient of each sample vehicle; Calculate the mean square error and mean absolute error between the predicted vehicle aerodynamic drag coefficient and the actual vehicle aerodynamic drag coefficient of each sample vehicle, and determine the mean square error value and mean absolute error value of the multi-scale Transformer network; If any of the mean square error value and the mean absolute error value is greater than the corresponding preset threshold, optimize the network parameters of the multi-scale Transformer network through the backpropagation algorithm, and continue to perform iterative training on the optimized multi-scale Transformer network; If both the mean square error value and the mean absolute error value are less than or equal to the corresponding preset thresholds, use the multi-scale Transformer network as the vehicle aerodynamic drag coefficient prediction model.

6. The method for determining the drag coefficient according to claim 1, characterized in that, Obtain the three-dimensional shape heterogeneous point cloud data through the following steps: Perform a hydrodynamic numerical simulation on the target vehicle after grid division to determine the three-dimensional flow field data of the target vehicle; Based on the lowest grid points in the three-dimensional flow field data, perform normalization processing on the shape data of the target vehicle to determine the normalized shape data; Perform fixed-dimension conversion processing, outlier removal processing, and mean normalization processing on the normalized shape data to determine the three-dimensional shape heterogeneous point cloud data of the target vehicle.

7. An air resistance coefficient determination device for a vehicle, characterized in that, The wind resistance coefficient determination device includes: A normalization processing module for resampling the three-dimensional shape heterogeneous point cloud data of the target vehicle, and performing coordinate normalization processing and voxel pooling processing on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle; A data preprocessing module for inputting multiple target three-dimensional point cloud data into the data preprocessing module of the vehicle wind resistance coefficient prediction model for spatial sequence encoding processing, attention calculation processing, and sparse convolution processing to determine sparse three-dimensional point cloud data; A feature embedding module for inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle wind resistance coefficient prediction model for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data; wherein, the embedding feature processing is to perform convolution transformation and normalization processing on the features of the sparse three-dimensional point cloud data to obtain a unified representation of the embedding features; A multi-stage encoder module for inputting the embedding features of the sparse three-dimensional point cloud data into the multi-stage encoder of the vehicle wind resistance coefficient prediction model for downsampling processing, aggregation processing, normalization processing, attention processing, perceptron processing, and global pooling processing, and outputting the global features of each stage processor; A regression prediction module for inputting multiple global features into the fully connected layer of the vehicle wind resistance coefficient prediction model for multi-scale feature regression prediction, and outputting the wind resistance coefficient of the target vehicle; wherein, the vehicle wind resistance coefficient prediction model is obtained by iteratively training a multi-scale Transformer network; The data preprocessing module is further configured to: perform resampling processing on the three-dimensional shape heterogeneous point cloud data based on grid sampling and uniform sampling to determine the resampled three-dimensional shape heterogeneous point cloud data; Establish a reference coordinate system based on the center points of the front and rear axles of the target vehicle; Perform a translation transformation on the resampled three-dimensional shape heterogeneous point cloud data based on the reference coordinate system, and perform a scale scaling on the resampled three-dimensional shape heterogeneous point cloud data based on the vehicle length of the target vehicle as a reference dimension to determine the three-dimensional shape heterogeneous point cloud data after coordinate normalization processing; Perform voxel pooling processing on the three-dimensional shape heterogeneous point cloud data after coordinate normalization processing to determine multiple target three-dimensional point cloud data.

8. An electronic device, characterized in that, Including: A processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device operates, the processor communicates with the memory through the bus, and when the machine-readable instructions are executed by the processor, the steps of the method for determining the drag coefficient of a vehicle according to any one of claims 1 to 6 are performed.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method for determining the drag coefficient of a vehicle according to any one of claims 1 to 6 are performed.

Images