Vehicle wind resistance coefficient determination method and device, electronic equipment and storage medium

Abstract

The invention provides a vehicle wind resistance coefficient determination method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the standardization processing of three-dimensional shape heterogeneous point cloud data of a target vehicle, and determining a plurality of target three-dimensional point cloud data; performing spatial sequential coding processing, attention calculation processing and sparse convolution processing on the multiple pieces 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 embedded features of the sparse three-dimensional point cloud data; performing down-sampling processing, aggregation processing, normalization processing, attention processing, perceptron processing and global pooling processing on the embedded features, and outputting the global features of the processor in each stage; and inputting the plurality of global features into a full connection layer to carry out multi-scale feature regression prediction, and outputting a wind resistance coefficient of the target vehicle. According to the method, the wind resistance coefficient can be quickly and accurately predicted in various actual industrial environments, and the prediction speed and precision are improved.

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 a vehicle's wind resistance coefficient. Background Art

[0002] Aerodynamic performance (especially the drag coefficient) has a significant impact on vehicle energy consumption, fuel economy, and cruising range, and is the goal of design optimization in many industrial fields such as aerospace, automobile manufacturing, and high-speed trains. With the continuous improvement of industrial design accuracy and iteration efficiency requirements, traditional aerodynamic analysis technology is facing huge challenges: although the numerical simulation method represented by computational fluid dynamics (CFD) has a clear physical mechanism, the computational cost increases rapidly with the complexity of the problem, and it is difficult to meet the urgent needs of the industry for fast or even real-time design optimization. A simulation or experiment often takes hours or even days, and repeated iterations of complex three-dimensional models are huge in terms of time and cost. In recent years, deep learning technology has developed rapidly, and several neural network-based approximate model (surrogate) methods have emerged to replace CFD simulation, but most of the existing methods still rely on low-dimensional shape parameters or 2D views, which makes it difficult to fully capture the complex flow information in the real 3D shape. In addition, in actual industrial environments, the diverse sizes of vehicle models and inconsistent grid density will double the difficulty of directly learning the "geometry→drag coefficient" mapping. Therefore, how to improve the accuracy of determining the vehicle's drag coefficient has become a technical issue that cannot be underestimated. Summary of the invention

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

[0004] The embodiment of the present application provides a method for determining a drag coefficient of a vehicle, wherein the method for determining a drag coefficient comprises: Resampling the three-dimensional shape heterogeneous point cloud data of the target vehicle, and performing coordinate normalization and voxel pooling on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle; Inputting the plurality of target three-dimensional point cloud data into a 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 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 to perform embedding feature processing, and determining the embedding 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 drag coefficient prediction model for downsampling, aggregation, normalization, attention, perception and global pooling processing, and outputting the global features of each stage processor; The multiple global features are input into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and the drag coefficient of the target vehicle is output; wherein the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network.

[0005] In a possible implementation manner, the three-dimensional shape heterogeneous point cloud data of the target vehicle is resampled, and the coordinate normalization and voxel pooling are performed on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle, including: Resampling 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; Establishing a reference coordinate system based on the center points of the front and rear axles of the target vehicle; The resampled three-dimensional shape heterogeneous point cloud data is translated based on the reference coordinate system, and the resampled three-dimensional shape heterogeneous point cloud data is scaled based on the length of the target vehicle as a reference dimension to determine the three-dimensional shape heterogeneous point cloud data after coordinate normalization processing; The three-dimensional shape heterogeneous point cloud data after coordinate normalization is subjected to voxel pooling processing to determine a plurality of target three-dimensional point cloud data.

[0006] In a possible implementation manner, the inputting of the plurality of target three-dimensional point cloud data into a 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 three-dimensional point cloud data includes: Performing spatial sequential 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 the encoding processing to determine the target three-dimensional point cloud data after the attention calculation processing; Sparse convolution processing is performed on the target three-dimensional point cloud data after the attention calculation processing to determine the sparse three-dimensional point cloud data.

[0007] In a possible implementation manner, the embedded features of the sparse three-dimensional point cloud data are input 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 the global features of the processor of each stage are output, including: 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; The global features of the first stage encoder are input to the next stage encoder for processing, and the global features of the next stage encoder are output.

[0008] In a possible implementation manner, the 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 includes: Downsampling the embedded features, and performing neighboring point cloud feature aggregation processing on the downsampled embedded features based on serialized coding to determine features of new three-dimensional point cloud data; Normalizing the features of the new three-dimensional point cloud data and the embedded features after the downsampling to determine a first feature of the three-dimensional point cloud data; Performing attention processing on the first feature of the three-dimensional point cloud data to determine a second feature of the three-dimensional point cloud data, and performing normalization processing on the second feature and the feature of the three-dimensional point cloud data to determine a third feature of the three-dimensional point cloud data; Performing perceptron processing on the third feature to determine a 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.

[0009] In a possible implementation manner, the vehicle drag coefficient prediction model is determined by the following steps: Inputting sample 3D point cloud data of multiple sample vehicles into a multi-scale Transformer network, processing the sample 3D point cloud data of each sample vehicle, and predicting a predicted vehicle drag coefficient of each sample vehicle; Performing mean square error calculation and mean absolute error calculation on the predicted vehicle drag coefficient and the actual vehicle drag coefficient of each of the sample vehicles to determine the mean square error value and the 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, the network parameters of the multi-scale Transformer network are optimized by a back propagation algorithm, and the optimized multi-scale Transformer network is continued to be iteratively trained; If the mean square error value and the mean absolute error value are both less than or equal to the corresponding preset thresholds, the multi-scale Transformer network is used as the vehicle drag coefficient prediction model.

[0010] In a possible implementation manner, the three-dimensional shape heterogeneous point cloud data is obtained by the following steps: Performing fluid dynamics numerical simulation processing on the target vehicle after meshing to determine three-dimensional flow field data of the target vehicle; Normalizing the shape data of the target vehicle based on the lowest grid point in the three-dimensional flow field data to determine the normalized shape data; The normalized shape data is subjected to fixed dimension conversion, outlier removal and mean normalization to determine the three-dimensional shape heterogeneous point cloud data of the target vehicle.

[0011] The embodiment of the present application further provides a device for determining a drag coefficient of a vehicle, the device for determining a drag coefficient comprising: A standardization processing module is used to 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; A data preprocessing module, used for 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, so as to determine sparse three-dimensional point cloud data; A feature embedding module, used for inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedding feature processing, and determining the embedding features of the sparse three-dimensional point cloud data; A multi-stage encoder module, used for 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; A regression prediction module 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.

[0012] An embodiment of the present application also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate 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 the vehicle as described above are performed.

[0013] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for determining the wind resistance coefficient of a vehicle as described above are executed.

[0014] The embodiments of the present application provide a method, device, electronic device and storage medium for determining the drag coefficient of a vehicle. 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 and voxel pooling on the resampled three-dimensional shape 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 a data preprocessing module of a vehicle drag coefficient prediction model for spatial sequence encoding, attention calculation and sparse convolution to determine sparse three-dimensional point cloud data; inputting the sparse three-dimensional point cloud data into the data preprocessing module of the vehicle drag coefficient prediction model to perform spatial sequence encoding, attention calculation and sparse convolution to determine sparse three-dimensional point cloud data; The feature embedding module performs embedded feature processing to determine the embedded features of the sparse three-dimensional point cloud data; the embedded features of the sparse three-dimensional point cloud data are input into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling, aggregation, normalization, attention, perceptron and global pooling, and the global features of each stage processor are output; multiple global features are input into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and the drag coefficient of the target vehicle is output; wherein the vehicle drag coefficient prediction model is obtained by iteratively training the multi-scale Transformer network. Through the standardized processing of heterogeneous three-dimensional point cloud 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, thereby improving the prediction speed and accuracy.

[0015] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A flow chart of a method for determining a vehicle's wind resistance coefficient provided in an embodiment of the present application; Figure 2 A schematic diagram of a method for determining a vehicle's wind resistance coefficient provided in an embodiment of the present application; Figure 3 One of the structural schematic diagrams of a device for determining a drag coefficient of a vehicle provided in an embodiment of the present application; Figure 4 A second structural schematic diagram of a device for determining a drag coefficient of a vehicle provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various 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 application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0019] First, the application scenarios to which the present application is applicable are introduced. The present application can be applied in the field of vehicle dynamics technology.

[0020] Research has found that aerodynamic performance (especially the drag coefficient) has a significant impact on vehicle energy consumption, fuel economy, and cruising range, and is the goal of design optimization in many industrial fields such as aerospace, automobile manufacturing, and high-speed trains. With the continuous improvement of industrial design accuracy and iteration efficiency requirements, traditional aerodynamic analysis technology is facing huge challenges: although the numerical simulation method represented by computational fluid dynamics (CFD) has a clear physical mechanism, the computational cost increases rapidly with the complexity of the problem, and it is difficult to meet the urgent needs of the industry for fast or even real-time design optimization. A simulation or experiment often takes hours or even days, and repeated iterations of complex three-dimensional models are huge in terms of time and cost. In recent years, deep learning technology has developed rapidly, and several neural network-based approximate model (surrogate) methods have emerged to replace CFD simulation, but most of the existing methods still rely on low-dimensional shape parameters or 2D views, which makes it difficult to fully capture the complex flow information in the real 3D shape. In addition, in actual industrial environments, the diverse sizes of vehicle models and inconsistent grid density will double the difficulty of directly learning the "geometry→drag coefficient" mapping. Therefore, how to improve the accuracy of determining the vehicle's drag coefficient has become a technical issue that cannot be underestimated.

[0021] Based on this, an embodiment of the present application provides a method for determining a vehicle's drag coefficient. Through standardized processing of three-dimensional point cloud heterogeneous data and a vehicle drag coefficient prediction model, it is possible to quickly and accurately predict the drag coefficient in a variety of actual industrial environments, thereby improving prediction speed and accuracy.

[0022] See also Figure 1 , Figure 1 This is a flow chart of a method for determining the drag coefficient of a vehicle provided in an embodiment of the present application. Figure 1 As shown in , the method for determining the drag coefficient provided in the embodiment of the present application includes: S101: resampling the three-dimensional shape heterogeneous point cloud data of the target vehicle, and performing coordinate normalization and voxel pooling 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.

[0023] In this step, the three-dimensional shape heterogeneous point cloud data of the target vehicle is resampled, and the resampled three-dimensional shape heterogeneous point cloud data is subjected to coordinate normalization processing and voxel pooling processing to determine multiple target three-dimensional point cloud data of the target vehicle.

[0024] Here, the target three-dimensional shape heterogeneous point cloud data is standardized by resampling, normalizing, and voxel pooling the target three-dimensional shape heterogeneous point cloud data.

[0025] In a possible implementation manner, the three-dimensional shape heterogeneous point cloud data is obtained by the following steps: A: Perform fluid dynamics numerical simulation processing on the target vehicle after grid division to determine the three-dimensional flow field data of the target vehicle.

[0026] Here, various 3D vehicle shape structures are obtained, and then the vehicle shape is discretized with a specified resolution. According to the set target incoming flow and boundary conditions, the CFD numerical simulation of the target vehicle's vehicle shape is performed to generate the 3D flow field data of the target vehicle's target shape.

[0027] Specifically, set the regional inflow and boundary conditions: for the target calculation area, set the boundary inflow conditions. Furthermore, set the inflow direction and wind speed. Set the bottom surface of the calculation 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 calculation domain. After the flow stabilizes, select the final three-dimensional flow field result of the fixed time step as the three-dimensional flow field data corresponding to the target area.

[0028] B: normalizing the shape data of the target vehicle based on the lowest grid point in the three-dimensional flow field data to determine the normalized shape data.

[0029] Here, the shape data of the target vehicle is subtracted from its minimum value to unify the relative height, so as to determine the normalized shape data.

[0030] C: Performing 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.

[0031] Here, the normalized shape data is converted into a unified format by a linear interpolation method to ensure that it can adapt to the input requirements of the model. The CFD calculation examples that have not converged are eliminated by outlier identification to ensure data quality.

[0032] Among them, the formula for mean normalization is:

[0033] in, It is the three-dimensional shape heterogeneous point cloud data. is the original 3D shape heterogeneous point cloud data, is the statistical mean of the target vehicle’s appearance data, is the statistical standard deviation of the target vehicle appearance data.

[0034] In a possible implementation manner, the three-dimensional shape heterogeneous point cloud data of the target vehicle is resampled, and the coordinate normalization and voxel pooling are performed on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle, including: (1): Based on grid sampling and uniform sampling, the three-dimensional shape heterogeneous point cloud data is resampled to determine the resampled three-dimensional shape heterogeneous point cloud data.

[0035] Here, due to the differences in point cloud density of different car models, this application combines grid sampling and uniform sampling. Specifically, the original grid points are first used as the basic points for sampling to ensure the full representation of local geometric details; then the car body is uniformly sampled to control the overall number of points, avoid excessive influence of high-density areas on global features, and ensure that local fine features such as high curvature areas are fully expressed, while avoiding excessive representation of dense areas, eliminating density differences in multi-source point cloud data, and achieving unified feature distribution. After processing, data storage overhead is reduced, key geometric details are retained, and the neural network can focus more on modeling the relationship between geometry and flow field, significantly improving the feature generalization ability of the model.

[0036] B: Establishing a reference coordinate system based on the center points of the front and rear axles of the target vehicle; performing a translation transformation on the resampled three-dimensional shape heterogeneous point cloud data based on the reference coordinate system, and scaling the resampled three-dimensional shape heterogeneous point cloud data based on the length of the target vehicle as a reference dimension, and determining the three-dimensional shape heterogeneous point cloud data after coordinate normalization.

[0037] Here, the purpose of coordinate normalization is to eliminate the feature distribution bias caused by the scale differences of multi-source vehicle model data, and to enhance the neural network's ability to focus on key differences in geometric topology. This application achieves geometric standardization through rigid body registration and affine scaling: first, a reference coordinate system is established based on the center points of the front and rear axles, and the wheelbase is aligned through translation transformation to eliminate the representation ambiguity caused by spatial displacement; then the wheelbase or vehicle length is selected as the reference dimension for proportional scaling, and each vehicle model is uniformly mapped 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 representation, making the point clouds of different models comparable in key geometric dimensions such as curvature distribution and local concave-convex features, significantly reducing the optimization difficulty of model generalization.

[0038] C: Perform voxel pooling processing on the three-dimensional shape heterogeneous point cloud data after coordinate normalization processing to determine a plurality of target three-dimensional point cloud data.

[0039] Here, voxel pooling is performed on the three-dimensional shape heterogeneous point cloud data after coordinate normalization, which reduces the amount of data to be processed and enables the features of the point cloud in each voxel to reflect the geometric information of the local area.

[0040] S102: Input the plurality of target three-dimensional point cloud data into a 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 three-dimensional point cloud data.

[0041] In this step, multiple target three-dimensional point cloud data are input 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. In a possible implementation manner, the inputting of the plurality of target three-dimensional point cloud data into a 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 three-dimensional point cloud data includes: a: performing spatial sequential 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.

[0042] Here, a space filling curve (such as a Z-order curve, a Hilbert curve, etc.) is used 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.

[0043] b: Performing attention calculation processing on the target three-dimensional point cloud data after the encoding processing to determine the target three-dimensional point cloud data after the attention calculation processing.

[0044] Here, attention calculation is performed on the encoded target three-dimensional point cloud data to better capture the spatial neighborhood information.

[0045] 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.

[0046] Here, sparse convolution is used to convert the target three-dimensional point cloud data after attention calculation into a sparse convolution tensor. In the three-dimensional space where the point cloud is more dispersed, sparse convolution can significantly reduce invalid convolution operations and improve memory utilization and computing efficiency.

[0047] S103: Inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model to perform embedding feature processing, and determining the embedding features of the sparse three-dimensional point cloud data.

[0048] In this step, the sparse three-dimensional point cloud data is input into a feature embedding module for embedding feature processing to determine the embedding features of the sparse three-dimensional point cloud data.

[0049] Among them, the embedded feature processing is to perform convolution transformation and normalization processing on the features of sparse 3D point cloud data (such as coordinates, normals, intensity, additional parameters, etc.) to obtain a unified embedded feature representation. Specifically, it includes: using 3D convolution to perform convolution operations on sparse 3D point cloud data to capture local geometric patterns, and then further normalize and enhance nonlinear expressions; finally, the embedded features of the specified dimension are obtained for subsequent encoders.

[0050] S104: 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, aggregation, normalization, attention, perception and global pooling processing, and output the global features of each stage processor.

[0051] In this step, the embedded features of the sparse three-dimensional point cloud data are input into the multi-stage encoder for downsampling, aggregation, normalization, attention, perception, and global pooling, and the global features of each stage processor are output.

[0052] Among them, each stage encoder consists of downsampling and Block stacking.

[0053] In a possible implementation manner, the embedded features of the sparse three-dimensional point cloud data are input 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 the global features of the processor of each stage are output, including: (I): 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, and output the global features of the first-stage encoder.

[0054] Here, the features of the sparse three-dimensional point cloud data are first input into the first-stage encoder for downsampling, position encoding, normalization, attention, perception, and global pooling, and the global features of the first-stage encoder are output.

[0055] In a possible implementation manner, the 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 includes: i: down-sample the embedded features, and perform neighboring point cloud feature aggregation on the down-sampled embedded features based on serialized coding to determine features of new three-dimensional point cloud data.

[0056] Here, the downsampling process aims to make hierarchical abstractions at the point cloud level, "aggregate" fine-grained point clouds to a coarser spatial scale, reduce the number of points and highlight important features. Through serialized encoding, spatially adjacent points are found and merged according to "clusters" for feature aggregation, and the features of new 3D point cloud data are output.

[0057] ii: normalizing the features of the new three-dimensional point cloud data and the embedded features after the downsampling to determine the first features of the three-dimensional point cloud data.

[0058] Here, the purpose of normalizing the features of the new 3D point cloud data and the embedded features after downsampling is to retain the expressive power of the features.

[0059] iii: performing 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 performing 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.

[0060] Here, the purpose of attention processing is to divide the point cloud into several slices and perform multi-head attention on the points in the same slice. This attention can use FlashAttention to accelerate the attention operation of large-scale sequences.

[0061] iv: performing perceptron processing on the third feature to determine a fourth feature of the three-dimensional point cloud data; performing global pooling processing on the fourth feature and the second feature to output the global feature.

[0062] Here, a multi-layer perceptron (MLP) is used to expand the channel dimension, enhance nonlinearity, and then project back to the original channel dimension. After each stage, this model will perform a global pooling on the features and aggregate all features in the current Batch into a global vector.

[0063] In this application, multiple stage encoders are connected in series to gradually extract richer local-global hybrid features. Suppose the initial embedding feature is , after the The features after the stage encoder processing are:

[0064] in, For the k Stage encoder ( ), For the k - all features in 1 stage encoder, S is the total number of stage encoders.

[0065] At each scale stage, the global pooling operation is used to extract scale-related global features. :

[0066] in, For the k The global features output by the encoder in each stage, This is global pooling processing.

[0067] (ii): Input the global features of the first stage encoder to the next stage encoder for processing, and output the global features of the next stage encoder.

[0068] Here, the global features of the first stage encoder are input to the next stage encoder for processing, and the global features of the next stage encoder are output. The above steps are repeated until each stage encoder processes the global features and stops processing.

[0069] 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.

[0070] In this step, the problem of insufficient single-scale network modeling is avoided. Multi-scale feature regression prediction is performed on multiple global features in the fully connected layer to output the drag coefficient of the target vehicle.

[0071] Among them, multi-scale features are constructed for multiple global features The expression is:

[0072] Here, the multi-scale features are processed linearly, batch normalized, activated, and processed by a multi-layer perceptron in the fully connected layer to obtain the drag coefficient of the target vehicle.

[0073] In a possible implementation manner, the vehicle drag coefficient prediction model is determined by the following steps: 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.

[0074] Here, the sample three-dimensional point cloud data of multiple sample vehicles are input into the multi-scale Transformer network, the sample three-dimensional point cloud data of each sample vehicle is processed, and the predicted vehicle drag coefficient of each sample vehicle is predicted.

[0075] Among them, the processing process of predicting the vehicle drag coefficient is consistent with the processing process of the above-mentioned vehicle drag coefficient, and this part is no longer decorated.

[0076] Here, a new multi-scale Transformer architecture is designed based on the Transformer network, and a global pooling module is embedded between deep encoders to improve the model's feature learning ability.

[0077] II: performing mean square error calculation and mean absolute error calculation on the predicted vehicle drag coefficient and the actual vehicle drag coefficient of each of the sample vehicles, and determining the mean square error value and mean absolute error value of the multi-scale Transformer network.

[0078] Here, the mean square error value is determined by the following formula and mean absolute error :

[0079]

[0080] in, is the actual vehicle drag coefficient, To predict the vehicle drag coefficient, m is the number of sample vehicles.

[0081] III: If any of the mean square error value and the mean absolute error value is greater than the corresponding preset threshold, the network parameters of the multi-scale Transformer network are optimized by a back propagation algorithm, and the optimized multi-scale Transformer network is continued to be iteratively trained.

[0082] Here, in addition to optimizing the network parameters of the multi-scale Transformer network through the back propagation algorithm, the prediction accuracy of the model under different car shapes can also be improved through hyperparameter tuning, network parameter adjustment, and regularization techniques.

[0083] IV: If the mean square error value and the mean absolute error value are both less than or equal to the corresponding preset thresholds, the multi-scale Transformer network is used as the vehicle drag coefficient prediction model.

[0084] In this application, a new network structure and algorithm are proposed to improve the prediction speed and accuracy, enhance the adaptability and versatility of the model, reduce the dependence on large amounts of historical data and high-performance hardware, and provide an efficient, flexible and easy-to-deploy solution for drag coefficient prediction.

[0085] For further information, see Figure 2 , Figure 2 A schematic diagram of a method for determining a vehicle's wind resistance coefficient provided in an embodiment of the present application. Figure 2 As shown in , the three-dimensional shape heterogeneous point cloud data of the vehicle is resampled, normalized, and voxel pooled to obtain the target three-dimensional point cloud data, and 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, aggregation, normalization, attention, perception, and global pooling to obtain multiple global features. Multiple global features are input into the fully connected layer to output the drag coefficient of the target vehicle.

[0086] A method for determining a vehicle's drag coefficient is provided in an embodiment of the present application. The method for determining a vehicle's drag coefficient includes: resampling three-dimensional shape heterogeneous point cloud data of a target vehicle, and performing coordinate normalization and voxel pooling on the resampled three-dimensional shape 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 a data preprocessing module of a vehicle drag coefficient prediction model to perform spatial sequence encoding, attention calculation and sparse convolution processing to determine sparse three-dimensional point cloud data; inputting the sparse three-dimensional point cloud data into a feature embedding module of the vehicle drag coefficient prediction model The embedded features of the sparse three-dimensional point cloud data are processed to determine the embedded features of the sparse three-dimensional point cloud data; the embedded features of the sparse three-dimensional point cloud data are input into the multi-stage encoder of the vehicle drag coefficient prediction model for downsampling, aggregation, normalization, attention, perceptron and global pooling, and the global features of each stage processor are output; multiple global features are input into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and the drag coefficient of the target vehicle is output; wherein the vehicle drag coefficient prediction model is obtained by iteratively training the multi-scale Transformer network. Through the standardized processing of heterogeneous three-dimensional point cloud data and the vehicle drag coefficient prediction model, the drag coefficient can be predicted quickly and accurately in a variety of actual industrial environments, improving the prediction speed and accuracy.

[0087] See also Figure 3 , Figure 4 , Figure 3 One of the structural schematic diagrams of a device for determining a drag coefficient of a vehicle provided in an embodiment of the present application; Figure 4 This is a second structural diagram of a vehicle drag coefficient determination device provided in an embodiment of the present application. Figure 3 As shown in FIG. 3 , the vehicle drag coefficient determination device 300 includes: The standardization processing module 310 is used to 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; A data preprocessing module 320 is used to input the plurality of target three-dimensional point cloud data into a 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 three-dimensional point cloud data; A feature embedding module 330, used for inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedding feature processing, and determining the embedding features of the sparse three-dimensional point cloud data; A multi-stage encoder module 340, used to 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; 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.

[0088] Furthermore, when the standardization processing module 310 is used to 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, the standardization processing module 310 is specifically used to: Resampling 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; Establishing a reference coordinate system based on the center points of the front and rear axles of the target vehicle; The resampled three-dimensional shape heterogeneous point cloud data is translated based on the reference coordinate system, and the resampled three-dimensional shape heterogeneous point cloud data is scaled based on the length of the target vehicle as a reference dimension to determine the three-dimensional shape heterogeneous point cloud data after coordinate normalization processing; The three-dimensional shape heterogeneous point cloud data after coordinate normalization is subjected to voxel pooling processing to determine a plurality of target three-dimensional point cloud data.

[0089] Further, the data preprocessing module 320 performs spatial sequence encoding processing, attention calculation processing and sparse convolution processing in the data preprocessing module for inputting the plurality of target three-dimensional point cloud data into the vehicle drag coefficient prediction model, and when determining the sparse three-dimensional point cloud data, the data preprocessing module 320 is specifically used for: Performing spatial sequential 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 the encoding processing to determine the target three-dimensional point cloud data after the attention calculation processing; Sparse convolution processing is performed on the target three-dimensional point cloud data after the attention calculation processing to determine the sparse three-dimensional point cloud data.

[0090] Further, 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 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. The multi-stage encoder module 340 is specifically used to: 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; The global features of the first stage encoder are input to the next stage encoder for processing, and the global features of the next stage encoder are output.

[0091] 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, and output the global features of the first-stage encoder, the multi-stage encoder module 340 is specifically used to: Downsampling the embedded features, and performing neighboring point cloud feature aggregation processing on the downsampled embedded features based on serialized coding to determine features of new three-dimensional point cloud data; Normalizing the features of the new three-dimensional point cloud data and the embedded features after the downsampling to determine a first feature of the three-dimensional point cloud data; Performing attention processing on the first feature of the three-dimensional point cloud data to determine a second feature of the three-dimensional point cloud data, and performing normalization processing on the second feature and the feature of the three-dimensional point cloud data to determine a third feature of the three-dimensional point cloud data; Performing perceptron processing on the third feature to determine a 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.

[0092] Further, such as Figure 4 As shown, the vehicle drag coefficient determination device 300 further includes a model training module 360, and the model training module 360 ​​is used to: Inputting sample 3D point cloud data of multiple sample vehicles into a multi-scale Transformer network, processing the sample 3D point cloud data of each sample vehicle, and predicting a predicted vehicle drag coefficient of each sample vehicle; Performing mean square error calculation and mean absolute error calculation on the predicted vehicle drag coefficient and the actual vehicle drag coefficient of each of the sample vehicles to determine the mean square error value and the 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, the network parameters of the multi-scale Transformer network are optimized by a back propagation algorithm, and the optimized multi-scale Transformer network is continued to be iteratively trained; If the mean square error value and the mean absolute error value are both less than or equal to the corresponding preset thresholds, the multi-scale Transformer network is used as the vehicle drag coefficient prediction model.

[0093] Further, such as Figure 4 As shown, the vehicle drag coefficient determination device 300 further includes a data acquisition module 370, and the data acquisition module 370 is used to: Performing fluid dynamics numerical simulation processing on the target vehicle after meshing to determine three-dimensional flow field data of the target vehicle; Normalizing the shape data of the target vehicle based on the lowest grid point in the three-dimensional flow field data to determine the normalized shape data; The normalized shape data is subjected to fixed dimension conversion, outlier removal and mean normalization to determine the three-dimensional shape heterogeneous point cloud data of the target vehicle. The embodiment of the present application provides a device for determining the wind resistance coefficient of a vehicle, and the device for determining the wind resistance coefficient includes: a standardization processing module, which is used to 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; a data preprocessing module, which is used 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, which is used to input the sparse three-dimensional point cloud data into the feature embedding module of the vehicle wind resistance coefficient prediction model The feature embedding module is used to perform embedded feature processing to determine the embedded features of the sparse three-dimensional point cloud data; the multi-stage encoder module is used to 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, aggregation, normalization, attention, perceptron and global pooling, and output the global features of each stage processor; the regression prediction module is used to 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 the multi-scale Transformer network. Through the standardized processing of heterogeneous three-dimensional point cloud 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, thereby improving the prediction speed and accuracy.

[0094] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown in , the electronic device 500 includes a processor 510 , a memory 520 and a bus 530 .

[0095] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 communicates with the memory 520 via the bus 530. When the machine-readable instructions are executed by the processor 510, the above-mentioned Figure 1 as well as Figure 2 The steps of the method for determining the drag coefficient of a vehicle in the method embodiment shown in the figure, and the specific implementation manner can be referred to the method embodiment, which will not be repeated here.

[0096] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 as well as Figure 2 The steps of the method for determining the drag coefficient of a vehicle in the method embodiment shown in the figure, and the specific implementation manner can be referred to the method embodiment, which will not be repeated here.

[0097] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0098] In the several embodiments provided in 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 schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For 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 mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0099] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0100] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0101] If the 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 that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0102] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; 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 be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for determining a vehicle's drag coefficient, characterized in that: The method for determining the drag coefficient comprises: Resampling the three-dimensional shape heterogeneous point cloud data of the target vehicle, and performing coordinate normalization and voxel pooling on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle; Inputting the plurality of target three-dimensional point cloud data into a 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 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 to perform embedding feature processing, and determining the embedding 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 drag coefficient prediction model for downsampling, aggregation, normalization, attention, perception and global pooling processing, and outputting the global features of each stage processor; The multiple global features are input into the fully connected layer of the vehicle drag coefficient prediction model for multi-scale feature regression prediction, and the drag coefficient of the target vehicle is output; wherein the vehicle drag coefficient prediction model is obtained by iteratively training a multi-scale Transformer network.

2. The method for determining the drag coefficient according to claim 1, characterized in that: The three-dimensional shape heterogeneous point cloud data of the target vehicle is resampled, and the coordinate normalization and voxel pooling are performed on the resampled three-dimensional shape heterogeneous point cloud data to determine multiple target three-dimensional point cloud data of the target vehicle, including: Resampling 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; Establishing a reference coordinate system based on the center points of the front and rear axles of the target vehicle; The resampled three-dimensional shape heterogeneous point cloud data is translated based on the reference coordinate system, and the resampled three-dimensional shape heterogeneous point cloud data is scaled based on the length of the target vehicle as a reference dimension to determine the three-dimensional shape heterogeneous point cloud data after coordinate normalization processing; The three-dimensional shape heterogeneous point cloud data after coordinate normalization is subjected to voxel pooling processing to determine a plurality of target three-dimensional point cloud data.

3. The method for determining the drag coefficient according to claim 1, characterized in that: The step of inputting the plurality of target three-dimensional point cloud data into a 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 three-dimensional point cloud data includes: Performing spatial sequential 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 the encoding processing to determine the target three-dimensional point cloud data after the attention calculation processing; Sparse convolution processing is performed on the target three-dimensional point cloud data after the attention calculation processing to determine the sparse three-dimensional point cloud data.

4. The method for determining the drag coefficient according to claim 1, characterized in that: The embedded features of the sparse three-dimensional point cloud data are input 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 the global features of the processor of each stage are output, including: 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; The global features of the first stage encoder are input to the next stage encoder for processing, and the global features of the next stage encoder are output.

5. The method for determining the drag coefficient according to claim 4, characterized in that: The step of 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 includes: Downsampling the embedded features, and performing neighboring point cloud feature aggregation processing on the downsampled embedded features based on serialized coding to determine features of new three-dimensional point cloud data; Normalizing the features of the new three-dimensional point cloud data and the embedded features after the downsampling to determine a first feature of the three-dimensional point cloud data; Performing attention processing on the first feature of the three-dimensional point cloud data to determine a second feature of the three-dimensional point cloud data, and performing normalization processing on the second feature and the feature of the three-dimensional point cloud data to determine a third feature of the three-dimensional point cloud data; Performing perceptron processing on the third feature to determine a 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.

6. The method for determining the drag coefficient according to claim 1, characterized in that: The vehicle drag coefficient prediction model is determined by the following steps: Inputting sample 3D point cloud data of multiple sample vehicles into a multi-scale Transformer network, processing the sample 3D point cloud data of each sample vehicle, and predicting a predicted vehicle drag coefficient of each sample vehicle; Performing mean square error calculation and mean absolute error calculation on the predicted vehicle drag coefficient and the actual vehicle drag coefficient of each of the sample vehicles to determine the mean square error value and the 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, the network parameters of the multi-scale Transformer network are optimized by a back propagation algorithm, and the optimized multi-scale Transformer network is continued to be iteratively trained; If the mean square error value and the mean absolute error value are both less than or equal to the corresponding preset thresholds, the multi-scale Transformer network is used as the vehicle drag coefficient prediction model.

7. The method for determining the drag coefficient according to claim 1, characterized in that: The three-dimensional shape heterogeneous point cloud data is obtained by the following steps: Performing fluid dynamics numerical simulation processing on the target vehicle after meshing to determine three-dimensional flow field data of the target vehicle; Normalizing the shape data of the target vehicle based on the lowest grid point in the three-dimensional flow field data to determine the normalized shape data; The normalized shape data is subjected to fixed dimension conversion, outlier removal and mean normalization to determine the three-dimensional shape heterogeneous point cloud data of the target vehicle.

8. A device for determining a drag coefficient of a vehicle, characterized in that: The drag coefficient determination device comprises: A standardization processing module is used to 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; A data preprocessing module, used for 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, so as to determine sparse three-dimensional point cloud data; A feature embedding module, used for inputting the sparse three-dimensional point cloud data into the feature embedding module of the vehicle drag coefficient prediction model for embedding feature processing, and determining the embedding features of the sparse three-dimensional point cloud data; A multi-stage encoder module, used for 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; A regression prediction module 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.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the method for determining the wind resistance coefficient of a vehicle as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining the wind resistance coefficient of a vehicle as claimed in any one of claims 1 to 7 are executed.

Images