Vehicle aerodynamic characteristic evaluation method and device, computer equipment, readable storage medium and program product

Abstract

This application relates to a method, apparatus, computer equipment, readable storage medium, and program product for evaluating vehicle aerodynamic characteristics. The method includes: acquiring a vehicle aerodynamic model, a target outer contour model, and target fluid boundary condition data corresponding to a target vehicle; extracting geometric features from the target outer contour model to obtain the target geometric mesh features of the target vehicle; extracting multi-scale features from the target geometric mesh features using the vehicle aerodynamic model to obtain target multi-scale features; and evaluating the aerodynamic characteristics of the target vehicle based on the target multi-scale features and the target fluid boundary condition data using the vehicle aerodynamic model to obtain the target vehicle aerodynamic characteristics. This method can improve the efficiency of vehicle aerodynamic characteristic evaluation.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for evaluating vehicle aerodynamic characteristics. Background Technology

[0002] With the widespread use of vehicles, the demand for vehicle research and development is increasing, and the technology for evaluating the aerodynamic characteristics of vehicles has emerged accordingly.

[0003] In traditional techniques, geometric and flow field data are collected, and the finite volume method and fluid control equations are combined for discrete solution. After multiple iterations and convergence, the aerodynamic characteristics of the vehicle are obtained by integrating the pressure and friction on the vehicle surface.

[0004] However, when there are a large number of vehicles waiting for aerodynamic characteristic evaluation, applying the above process to each vehicle can easily lead to low efficiency in evaluating their aerodynamic characteristics. Therefore, there is an urgent need for a way to quickly evaluate the aerodynamic characteristics of vehicles. Summary of the Invention

[0005] Based on this, this application addresses the aforementioned technical problems by providing a vehicle aerodynamic characteristic evaluation method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the efficiency of vehicle aerodynamic characteristic evaluation.

[0006] In a first aspect, this application provides a method for evaluating the aerodynamic characteristics of a vehicle, including:

[0007] Acquire the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle;

[0008] Geometric features are extracted from the outer contour model of the target to obtain the target geometric mesh features of the target vehicle.

[0009] Multi-scale feature extraction of the target geometric mesh features is performed using the vehicle aerodynamic model to obtain the target multi-scale features.

[0010] The aerodynamic characteristics of the target vehicle are evaluated using the vehicle aerodynamic model based on the multi-scale characteristics of the target and the target fluid boundary condition data, thereby obtaining the aerodynamic characteristics of the target vehicle.

[0011] Thus, by extracting corresponding geometric features from the vehicle's outer contour model, the extracted target geometric mesh features can characterize the positional and geometric topological features of the vehicle's outer contour. Multi-scale feature extraction of the target geometric features enriches the content of the extracted multi-scale features and involves more feature dimensions. Using a pre-trained vehicle aerodynamic model, the target multi-scale features and target fluid boundary condition data are used as the basis for aerodynamic feature evaluation. This enables the vehicle aerodynamic model to quickly and accurately construct regular flow field mesh data for aerodynamic feature evaluation. Furthermore, by matching the mesh point positions between geometric mesh features and regular flow field mesh data, a certain correspondence exists between irregular geometric mesh features and regular flow field mesh data, improving the computational efficiency of subsequent model training. The resulting vehicle aerodynamic model, which enables aerodynamic feature evaluation, thus improves the efficiency of vehicle aerodynamic feature evaluation.

[0012] In an optional embodiment of the first aspect, the training process of the vehicle aerodynamic model includes:

[0013] Obtain the training outer contour model, training flow field data, and training fluid boundary condition data corresponding to the training vehicle, and construct the flow field regular mesh data of the training vehicle based on the training flow field data.

[0014] Geometric features are extracted from the training outer contour model to obtain the training geometric mesh features of the training vehicle;

[0015] The training geometric mesh features and the flow field volume regular mesh data are matched accordingly to obtain multiple sets of matching data;

[0016] For each set of matching data, the training geometric mesh features and the training fluid boundary conditions in the matching data are determined as the input feature data of a training sample, the flow field regular mesh data in the matching data are determined as the first true label of the training sample, and the true drag coefficient corresponding to the training geometric mesh features is determined as the second true label of the training sample.

[0017] The vehicle aerodynamic model is obtained by iteratively updating multiple training samples determined by the multiple sets of matching data.

[0018] Thus, considering that the volume field distribution density of the training flow field data of the training vehicle is not uniform, the construction of regular flow field mesh data can transform the non-uniform and irregular training flow field data into regular and uniformly distributed flow field mesh data. This allows for the matching of training geometric mesh features with flow field mesh data, ensuring a one-to-one correspondence between the geometric features in the training geometric mesh features and the flow field data in the flow field mesh data, thereby enabling the determination of subsequent training samples. Consequently, the vehicle aerodynamic model obtained through iterative updates using training samples becomes more accurate.

[0019] In an optional embodiment of the first aspect, the step of iteratively updating the vehicle aerodynamic model based on multiple training samples determined from the multiple sets of matching data includes:

[0020] For each training sample, the aerodynamic model of the vehicle to be trained predicts the flow field of the training vehicle based on the input feature data of the training sample, and obtains the predicted flow field data. Also, the aerodynamic model of the vehicle to be trained predicts the wind force of the training vehicle based on the input feature data of the training sample, and obtains the predicted drag coefficient.

[0021] The data constraint conditions corresponding to the predicted volume flow field data are matched with the predicted volume flow field data to obtain the data constraint loss value of the predicted volume flow field data;

[0022] Based on the difference between the first true label in the training sample and the predicted volume flow field data, the volume flow field prediction accuracy of the vehicle aerodynamic model to be trained is evaluated, and the volume flow field prediction loss value of the predicted volume flow field data is obtained.

[0023] The drag prediction accuracy of the aerodynamic model of the vehicle to be trained is evaluated based on the difference between the second true label in the training sample and the predicted drag coefficient, and the drag prediction loss value of the predicted drag coefficient is obtained.

[0024] The data constraint loss value, the volume flow field prediction loss value, and the wind resistance prediction loss value are fused to obtain the total model loss value;

[0025] Based on the total loss value of the model, the aerodynamic model of the vehicle to be trained is updated to obtain the vehicle aerodynamic model.

[0026] Thus, when the aerodynamic model of the vehicle under training predicts the volume flow field, it may fail to consider the actual fluid physics laws, resulting in a discrepancy between the predicted volume flow field data and the actual fluid physics laws. Therefore, by matching the predicted volume flow field data with the corresponding data constraints, the data constraint loss value can quantify the deviation between the predicted volume flow field data and the actual fluid physics laws. The difference between the first true label and the predicted volume flow field data can quantify the volume flow field prediction accuracy of the aerodynamic model of the vehicle under training, and the difference between the second true label and the predicted drag coefficient can quantify the drag prediction accuracy of the aerodynamic model of the vehicle under training. By using the data constraint loss value, the volume flow field prediction loss value, and the drag prediction loss value together as the decision basis for the total loss value of the model, the update of the aerodynamic model of the vehicle under training can ensure that the predicted volume flow field data conforms to the actual fluid physics laws and also ensure the prediction accuracy of the volume flow field and drag, thereby improving the update accuracy of the vehicle aerodynamic model.

[0027] In an optional embodiment of the first aspect, before iteratively updating the vehicle aerodynamic model based on the plurality of training samples determined from the plurality of sets of matching data, the method further includes:

[0028] Obtain the positional relationship between each regular grid point in the flow field regular grid data and the vehicle;

[0029] Filter the target volume flow field data representing the positional relationship inside the vehicle from the flow field regular grid data;

[0030] Based on the flow field data of the target body, the regular grid data of the flow field body is processed.

[0031] Therefore, considering the certain flow field changes inside the vehicle, and the fact that the flow field volume regular grid data may contain the flow field volume data corresponding to the grid points inside the vehicle, but its actual help to vehicle research and development is small, the flow field volume regular grid data is processed based on the target volume flow field data to make the flow field volume regular grid data used for model training more accurate.

[0032] In an optional embodiment of the first aspect, constructing the flow field regular mesh data of the training vehicle based on the training flow field data includes:

[0033] Based on the vehicle size information of the training vehicle, the location range of the body flow field data for the training vehicle is determined;

[0034] Target flow field data that fits the location range of the volume flow field data are selected from the training flow field volume data;

[0035] The target flow field data is normalized to obtain the flow field regular grid data of the training vehicle.

[0036] Therefore, considering that vehicle R&D needs should be focused on aerodynamic evaluation of a certain area around the vehicle, the location range of the volume flow field data obtained by matching the vehicle size information of the training vehicle can achieve accurate screening of the training flow field volume data. Considering that the volume field density distribution of the vehicle's training flow field volume data is not uniform, the target flow field volume data is normalized to make the obtained flow field volume regular grid data presented in a regular grid form. The flow field volume regular grid data is the basis for determining the training samples. Therefore, the training bias caused by the density difference between training samples can be eliminated, thereby improving the model accuracy of the vehicle aerodynamic model subsequently trained with the training samples.

[0037] In an optional embodiment of the first aspect, the training geometric mesh features include geometric features of multiple geometric mesh points, and the flow field volume regular mesh data includes flow field volume data of multiple regular mesh points;

[0038] The step involves matching the training geometric mesh features with the flow field volume regular mesh data to obtain multiple sets of matching data, including:

[0039] For each geometric grid point in the training geometric grid features, the distance between the geometric grid point and each regular grid point in the flow field regular grid is obtained;

[0040] Based on the distance between the geometric grid point and each regular grid point in the flow field regular grid, target regular grid points that match the geometric grid points are selected from the flow field regular grid to obtain matching data.

[0041] Thus, by using the distance between each geometric grid point and each regular grid point in the flow field regular grid as the matching basis, the position matching of irregular geometric grid points and regular flow field regular grids is achieved, so that the target grid features include the geometric grid features of the corresponding geometric grid points and the flow field features of the target regular grid points. The flow field features of the target regular grid points can be used as the true labels of the corresponding geometric grid features, thereby enabling the subsequent training of the vehicle aerodynamic model.

[0042] Secondly, this application also provides a vehicle aerodynamic characteristic evaluation device, comprising:

[0043] The data acquisition module is used to acquire the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle.

[0044] The feature extraction module is used to extract geometric features from the outer contour model of the target to obtain the target geometric mesh features of the target vehicle; and to extract multi-scale features from the target geometric mesh features using the vehicle aerodynamic model to obtain the target multi-scale features.

[0045] The feature evaluation module is used to evaluate the aerodynamic characteristics of the target vehicle based on the multi-scale features of the target and the fluid boundary condition data of the target using the vehicle aerodynamic model, thereby obtaining the aerodynamic characteristics of the target vehicle.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0047] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0048] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the above aspects.

[0049] Regarding the beneficial effects of any of the technical solutions in the second to fifth aspects mentioned above, refer to the beneficial effects of the corresponding technical solutions in the first aspect; repeated examples will not be listed here. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of an optional application environment for a vehicle aerodynamic feature evaluation method in one embodiment;

[0052] Figure 2 This is a schematic diagram of an optional process for evaluating vehicle aerodynamic characteristics in one embodiment;

[0053] Figure 3 This is a schematic diagram of an optional process for training a vehicle aerodynamic model in one embodiment;

[0054] Figure 4This is a schematic diagram of an optional process for constructing the flow field regular grid data of the training vehicle based on the training flow field data in one embodiment.

[0055] Figure 5 This is a schematic diagram of an optional structure of a vehicle aerodynamic feature evaluation device in one embodiment;

[0056] Figure 6 This is a schematic diagram of an optional internal structure of a computer device in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0058] The terms “in,” “comprising,” and “having,” as used herein, and any variations thereof, are intended to cover non-exclusive inclusion. The term “multiple” as used herein refers to two or more. The term “at least one” as used herein refers to one or more. The term “one” as used herein refers to any one of the foregoing.

[0059] The vehicle aerodynamic characteristic evaluation method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, the target vehicle 102 communicates with the server 104 via a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated onto the server 104, or it can be located in the cloud or on another network server. The server 104 acquires the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle 102; it extracts geometric features from the target outer contour model to obtain the target geometric mesh features of the target vehicle 102; it extracts multi-scale features from the target geometric mesh features using the vehicle aerodynamic model to obtain the target multi-scale features; and it evaluates the aerodynamic characteristics of the target vehicle 102 based on the target multi-scale features and target fluid boundary condition data using the vehicle aerodynamic model to obtain the target vehicle aerodynamic features. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0060] In one exemplary embodiment, such as Figure 2 As shown, a method for evaluating the aerodynamic characteristics of a vehicle is provided, which can be applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 204. Wherein:

[0061] Step 201: Obtain the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle.

[0062] In step 201, the target vehicle is the object awaiting research and development testing. The number of target vehicles can be single or multiple, without limitation. The vehicle aerodynamic model can be trained using data corresponding to the target vehicle, or it can be trained using data corresponding to training vehicles of the same type as the target vehicle. Specifically, vehicles belonging to the same vehicle type (where the vehicle type represents at least one of the vehicle's external shape and size) can be considered to belong to the same type, or vehicles belonging to the same vehicle model can be considered to belong to the same type, without limitation.

[0063] The target outer contour model is stored in STL (STereoLithography) format. The target flow field data is obtained through fluid dynamics simulation and stored in VTU (VTK XML Unstructured Grid, an XML-based unstructured grid data format used for 3D mesh storage in scientific computing and engineering simulations). The target flow field data includes at least one of velocity field data, pressure field data, and turbulence data. Velocity field data characterizes the velocity vector, pressure field data characterizes the pressure, and turbulence data characterizes the turbulent eddy viscosity coefficient. The target fluid boundary condition data includes at least one of air density, incoming flow velocity, and vehicle frontal area.

[0064] The vehicle aerodynamic model includes a fully trained feature extraction module, a volume flow field prediction module, and a wind resistance prediction module.

[0065] Step 202: Extract geometric features from the outer contour model of the target to obtain the target geometric mesh features of the target vehicle.

[0066] The target geometric mesh features in step 202 include triangular mesh features and geometric mesh point positions. The triangular mesh features are used to characterize at least one of the triangular patch area and the triangular normal vector.

[0067] As an example, step 202 includes: extracting the triangular mesh features and geometric mesh point positions of the target outer contour model to obtain the target geometric mesh features.

[0068] Step 203: Extract multi-scale features from the target geometric mesh features using the vehicle aerodynamic model to obtain the target multi-scale features.

[0069] As an example, step 203 includes: extracting multi-scale features of the target geometric grid features by using the vehicle aerodynamic model to extract the positions of each geometric grid point and the normal vectors of each triangular face.

[0070] Therefore, considering that the positions of geometric grid points and the normal vectors of triangular faces are data with richer information, multi-scale feature extraction is performed on the positions of geometric grid points and the normal vectors of triangular faces to realize the construction of complex feature engineering.

[0071] Step 203 can be implemented through the feature extraction module in the vehicle aerodynamic model. The extraction module can be a pyramid coding module or other modules that can perform multi-scale feature extraction. No restrictions are placed here.

[0072] Step 204: Based on the multi-scale characteristics of the target and the target fluid boundary condition data, the aerodynamic characteristics of the target vehicle are evaluated using the vehicle aerodynamic model to obtain the aerodynamic characteristics of the target vehicle.

[0073] For example, step 204 includes: generating global aerodynamic features based on the target multi-scale features and target fluid boundary condition data using a vehicle aerodynamic model; and evaluating the aerodynamic features of the target vehicle based on the global aerodynamic features using a vehicle aerodynamic model to obtain the target vehicle aerodynamic features.

[0074] As one embodiment, a global aerodynamic feature is generated based on the target's multi-scale features and target fluid boundary condition data using a vehicle aerodynamic model. This includes: uniformly encoding the target's multi-scale features using the vehicle aerodynamic model to obtain multi-scale encoded features; encoding the target fluid boundary condition data using the vehicle aerodynamic model to obtain fluid boundary condition codes; encoding the normal vectors of each triangular facet in the target's geometric mesh features using the vehicle aerodynamic model to obtain triangular facet normal vector codes; and concatenating the multi-scale encoded features, triangular facet normal vector codes, and fluid boundary condition codes to obtain the global aerodynamic feature.

[0075] The vehicle aerodynamic model also includes an encoding module. The step of generating global aerodynamic features based on the target's multi-scale features and target fluid boundary condition data is implemented through this encoding module. The encoding module includes a unified structure encoding layer, a multi-layer fully connected encoding layer, and an encoding concatenation layer. The step of performing unified structure encoding on the target's multi-scale features to obtain multi-scale encoded features is implemented through the unified structure encoding layer. The step of encoding the target fluid boundary condition data using the vehicle aerodynamic model to obtain fluid boundary condition encoding, and the step of encoding the normal vectors of each triangular facet in the target's geometric mesh features using the vehicle aerodynamic model to obtain triangular facet normal vector encoding, is implemented through the multi-layer fully connected encoding layer. The step of concatenating the multi-scale encoded features, triangular facet normal vector encoding, and fluid boundary condition encoding to obtain global aerodynamic features is implemented through the encoding concatenation layer.

[0076] As one embodiment, the aerodynamic characteristics of the target vehicle include target body flow field data and target drag coefficient; the aerodynamic characteristics of the target vehicle are evaluated based on the global aerodynamic characteristics using a vehicle aerodynamic model to obtain the target vehicle aerodynamic characteristics, including: predicting the body flow field of the target vehicle based on the global aerodynamic characteristics using a vehicle aerodynamic model to obtain target body flow field data; and predicting the wind conditions of the target vehicle based on the global aerodynamic characteristics using a vehicle aerodynamic model to obtain the target drag coefficient.

[0077] In this way, the target body flow field data and target drag coefficient can be directly output through the vehicle aerodynamic model, which improves the efficiency of vehicle aerodynamic characteristic assessment.

[0078] The steps of predicting the volume flow field of the target vehicle based on global aerodynamic characteristics and obtaining the target volume flow field data can be implemented by the volume flow field prediction module. The steps of predicting the wind force of the target vehicle based on global aerodynamic characteristics and obtaining the target drag coefficient can be implemented by the drag prediction module.

[0079] As an example, the target drag coefficient is obtained by predicting the wind conditions of the target vehicle based on global aerodynamic characteristics using a vehicle aerodynamic model, including: extracting volume flow field features based on global aerodynamic characteristics using a vehicle aerodynamic model; and predicting the wind conditions of the vehicle based on global aerodynamic characteristics and volume flow field features using a vehicle aerodynamic model to obtain the predicted drag coefficient.

[0080] Therefore, considering the correlation between volume flow field characteristics and drag coefficient, and that the vehicle aerodynamic model extracts corresponding features for volume flow field prediction based on global aerodynamic features, the volume flow field features extracted by the vehicle aerodynamic model and the global aerodynamic features can be used together as input data for the drag prediction model. This makes the prediction basis for wind conditions richer and improves the prediction accuracy of drag coefficient.

[0081] In the aforementioned vehicle aerodynamic feature evaluation method, geometric features are extracted from the vehicle's outer contour model, allowing the extracted target geometric mesh features to characterize the positional and geometric topological features of the vehicle's outer contour. Multi-scale feature extraction is performed on the target geometric features, resulting in richer feature content and more feature dimensions. A pre-trained vehicle aerodynamic model uses the target multi-scale features and target fluid boundary condition data as the basis for aerodynamic feature evaluation. This enables the vehicle aerodynamic model to quickly and accurately construct regular flow field mesh data for aerodynamic feature evaluation. Furthermore, by matching the mesh point positions between geometric mesh features and regular flow field mesh data, a certain correspondence exists between irregular geometric mesh features and regular flow field mesh data, improving the computational efficiency of subsequent model training. The trained vehicle aerodynamic model enables aerodynamic feature evaluation, thus improving the efficiency of vehicle aerodynamic feature evaluation.

[0082] Understandably, in order to ensure the training efficiency and accuracy of the vehicle aerodynamic model, it is necessary to determine relatively accurate training samples so that the vehicle aerodynamic model can be trained according to the training samples.

[0083] In one exemplary embodiment, such as Figure 3 As shown, the training process for the above-mentioned vehicle aerodynamic model includes steps 301 to 305. Wherein:

[0084] Step 301: Obtain the training outer contour model, training flow field data, and training fluid boundary condition data corresponding to the training vehicle, and construct the flow field regular mesh data of the training vehicle based on the training flow field data.

[0085] Optionally, the contents of the training outer contour model, training flow field data, and training fluid boundary condition data can refer to the above-mentioned vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data, and will not be elaborated here.

[0086] As one embodiment, constructing flow field volume regular mesh data for a training vehicle based on training flow field volume data includes: selecting target flow field volume data from the training flow field volume data for constructing flow field volume regular mesh data, and constructing flow field volume regular mesh data corresponding to the vehicle based on the target flow field volume data.

[0087] Step 302: Extract geometric features from the training outer contour model to obtain the training geometric mesh features of the training vehicle.

[0088] Optionally, the specific implementation steps of step 302 can refer to the specific implementation content of step 202 above, and will not be repeated here.

[0089] Step 303: Match the training geometric mesh features with the flow field volume regular mesh data to obtain multiple sets of matching data.

[0090] In step 303, the training geometric mesh features include the geometric features of multiple geometric mesh points, the flow field volume regular mesh data includes the flow field volume data of multiple regular mesh points, and the target mesh features include multiple sets of one-to-one corresponding geometric features and flow field volume features.

[0091] For example, step 303 includes: for each geometric grid point in the training geometric grid features, obtaining the distance between the geometric grid point and each regular grid point in the flow field regular grid; and filtering target regular grid points that match the geometric grid points from the flow field regular grid based on the distance between the geometric grid points and each regular grid point in the flow field regular grid to obtain matching data.

[0092] As one embodiment, based on the distance between the geometric grid point and each regular grid point in the flow field regular grid, target regular grid points that match the geometric grid points are selected from the flow field regular grid, including: selecting target regular grid points whose corresponding distance is less than a preset distance threshold from the flow field regular grid.

[0093] The preset distance threshold can be an empirical value or it can correspond to the distribution of each regular grid point and each geometric grid point.

[0094] As another embodiment, based on the distance between the geometric grid point and each regular grid point in the flow field regular grid, target regular grid points that match the geometric grid points are selected from the flow field regular grid, including: determining the grid point with the shortest corresponding distance in the flow field regular grid as the target regular grid point.

[0095] Thus, considering that the distribution of each geometric grid point may be irregular, that is, different geometric grid points may have different preset distance thresholds, the shortest distance in the regular grid of the flow field is directly used as the screening condition for the target regular grid point, which improves the screening accuracy of the target regular grid point.

[0096] Step 304: For each set of matching data, the training geometric grid features and training fluid boundary conditions in the matching data are determined as the input feature data of a training sample, the flow field volume regular grid data in the matching data are determined as the first true label of the training sample, and the true drag coefficient corresponding to the training geometric grid features is determined as the second true label of the training sample.

[0097] Optionally, the above method further includes: obtaining the positional relationship between each regular grid point in the flow field regular grid data and the vehicle; filtering the target body flow field data whose positional relationship is characterized inside the vehicle from the flow field regular grid data; and performing data processing on the flow field regular grid data based on the target body flow field data.

[0098] Therefore, considering the certain flow field changes inside the vehicle, and the fact that the flow field volume regular grid data may contain the flow field volume data corresponding to the grid points inside the vehicle, but its actual help to vehicle research and development is small, the flow field volume regular grid data is processed based on the target volume flow field data to make the flow field volume regular grid data used for model training more accurate.

[0099] As one embodiment, obtaining the positional relationship between each regular grid point in the flow field regular grid data and the vehicle includes: for each regular grid point in the flow field regular grid data, obtaining the distance between the regular grid point and the nearest point on the vehicle surface; and determining the positional relationship between the regular grid point and the vehicle based on the distance between the regular grid point and the nearest point on the vehicle surface.

[0100] Furthermore, based on the distance between the regular grid point and the nearest point on the vehicle surface, the positional relationship between the regular grid point and the vehicle is determined, including: the distance between the regular grid point and the nearest point on the vehicle surface includes the SDF (Signed Distance Function) value. If the SDF value between the regular grid point and the nearest point on the vehicle surface is non-negative, the positional relationship between the regular grid point and the vehicle is determined to be represented outside the vehicle; if the SDF value between the regular grid point and the nearest point on the vehicle surface is negative, the positional relationship between the regular grid point and the vehicle is determined to be represented inside the vehicle.

[0101] Furthermore, as an embodiment, data processing is performed on the flow field regular grid data based on the target volume flow field data, including: removing the target volume flow field data from the flow field regular grid data.

[0102] Therefore, considering the certain flow field changes inside the vehicle, and the fact that the flow field volume regular grid data may contain the flow field volume data corresponding to the grid points inside the vehicle, but its actual help to vehicle research and development is small, the target flow field data belonging to the inside of the vehicle is removed when conducting the corresponding vehicle aerodynamic characteristic evaluation. This can reduce the amount of data to be processed to a certain extent and improve the training efficiency of the vehicle aerodynamic model.

[0103] As another embodiment, data processing is performed on the flow field regular grid data based on the target volume flow field data, including: adding mask labels to the target volume flow field data in the flow field regular grid data.

[0104] Therefore, considering the certain flow field changes inside the vehicle, and the fact that the flow field volume regular grid data may contain the flow field volume data corresponding to the grid points inside the vehicle, but its actual help to vehicle research and development is small, the target volume flow field data with masked labels can be ignored when performing the corresponding vehicle aerodynamic feature evaluation. This can reduce the amount of data to be processed to a certain extent and improve the training efficiency of the vehicle aerodynamic model.

[0105] Step 305: Based on multiple training samples determined by multiple sets of matching data, the vehicle aerodynamic model is iteratively updated to obtain the model.

[0106] As an embodiment, step 305 includes: for each training sample, predicting the volume flow field of the training vehicle based on the input feature data of the training sample using the aerodynamic model of the vehicle to be trained, obtaining predicted volume flow field data; and predicting the wind force of the training vehicle based on the input feature data of the training sample using the aerodynamic model of the vehicle to be trained, obtaining a predicted drag coefficient; matching the data constraints corresponding to the predicted volume flow field data with the predicted volume flow field data to obtain a data constraint loss value for the predicted volume flow field data; evaluating the volume flow field prediction accuracy of the aerodynamic model of the vehicle to be trained based on the difference between the first true label in the training sample and the predicted volume flow field data, obtaining a volume flow field prediction loss value for the predicted volume flow field data; evaluating the drag prediction accuracy of the aerodynamic model of the vehicle to be trained based on the difference between the second true label in the training sample and the predicted drag coefficient, obtaining a drag prediction loss value for the predicted drag coefficient; fusing the data constraint loss value, the volume flow field prediction loss value, and the drag prediction loss value to obtain a total model loss value; and updating the aerodynamic model of the vehicle to be trained based on the total model loss value to obtain a vehicle aerodynamic model.

[0107] Thus, when the aerodynamic model of the vehicle under training predicts the volume flow field, it may fail to consider the actual fluid physics laws, resulting in a discrepancy between the predicted volume flow field data and the actual fluid physics laws. Therefore, by using the data constraints corresponding to the predicted volume flow field data, relational matching is performed on the predicted volume flow field data, allowing the data constraint loss value to quantify the deviation between the predicted volume flow field data and the actual fluid physics laws. The difference between the first true label and the predicted volume flow field data can quantify the volume flow field prediction accuracy of the aerodynamic model of the vehicle under training, and the difference between the second true label and the predicted drag coefficient can quantify the drag prediction accuracy of the aerodynamic model of the vehicle under training. By using the data constraint loss value, the volume flow field prediction loss value, and the drag prediction loss value together as the decision basis for the total model loss value, the update of the aerodynamic model of the vehicle under training can ensure that the predicted volume flow field data conforms to the actual fluid physics laws and also ensure the prediction accuracy of the volume flow field and drag, thereby improving the update accuracy of the vehicle aerodynamic model.

[0108] Optionally, the specific implementation of the steps of predicting the flow field of the training vehicle based on the input feature data of the training samples using the aerodynamic model of the vehicle to be trained, and predicting the wind force of the training vehicle based on the input feature data of the training samples, can be referred to the specific implementation of steps 202 to 204 above, and will not be repeated here.

[0109] The predicted volume flow field data includes predicted velocity field data, predicted flow field data, and predicted turbulence data. The data constraints corresponding to the predicted velocity field data include zero constraints, and the data constraints corresponding to the predicted turbulence data include non-negative constraints.

[0110] Furthermore, the data constraint conditions corresponding to the predicted volume flow field data are matched with the predicted volume flow field data to obtain the data constraint loss value of the predicted volume flow field data. This includes: matching the predicted velocity field data with the zero constraint condition to obtain the velocity constraint loss value of the predicted velocity field data; matching the predicted turbulence data with the non-negative constraint condition to obtain the turbulence constraint loss value of the predicted turbulence data; and fusing the flow field data residual value, velocity constraint loss value, and turbulence constraint loss value corresponding to the predicted flow field data to obtain the data constraint loss value of the predicted volume flow field data.

[0111] Thus, for the subdivided predicted velocity field data, predicted flow field data, and predicted turbulence data in the predicted volume flow field data, corresponding constraint matching and residual value evaluation are performed respectively. For the velocity field and turbulence dimensions, constraint matching is performed to quantify the deviation between the velocity field and turbulence field data and the actual fluid physics laws of the predicted volume flow field data. For the flow field dimension, residual value evaluation is performed to quantify the prediction accuracy of the flow field dimension data. Therefore, the flow field data residual value, velocity constraint loss value, and turbulence constraint loss value corresponding to the predicted flow field data are used together as the basis for generating the data constraint loss value, which improves the accuracy of the data constraint loss value generation.

[0112] Among them, the velocity constraint loss value is positively correlated with the difference between the predicted velocity field data and zero.

[0113] Since the velocity field data is zero, which is consistent with the laws of real fluid physics, the velocity constraint loss value is set to be positively correlated with the difference between the predicted velocity field data and zero. This makes the vehicle aerodynamic model update in the direction that the predicted velocity field data is closer to zero, so that the predicted velocity field data corresponding to the vehicle aerodynamic model is more likely to conform to the laws of real fluid physics.

[0114] As one embodiment, matching the predicted velocity field data with a zero constraint condition to obtain the velocity constraint loss value of the predicted velocity field data includes: when the predicted velocity field data is zero, determining a first value as the velocity constraint loss value of the predicted velocity field data; when the predicted velocity field data is not zero, determining a second value as the velocity constraint loss value of the predicted velocity field data, wherein the first value is less than the second value.

[0115] For example, the first value can be set to 0 and the second value to 1, or other settings can be used; there are no restrictions here.

[0116] As another embodiment, matching the predicted velocity field data with a zero constraint condition to obtain the velocity constraint loss value of the predicted velocity field data includes: obtaining the difference between the predicted velocity field data and zero, and determining the velocity constraint loss value of the predicted velocity field data based on the difference between the predicted velocity field data and zero.

[0117] Further, determining the velocity constraint loss value of the predicted velocity field data based on the difference between the predicted velocity field data and zero includes: determining the difference between the predicted velocity field data and zero as the velocity constraint loss value of the predicted velocity field data.

[0118] As an example, the predicted turbulence data is matched with non-negative constraint conditions to obtain the turbulence constraint loss value of the predicted turbulence data, including: when the predicted turbulence data is non-negative, a third value is determined as the turbulence constraint loss value of the predicted turbulence data; when the predicted turbulence data is not non-negative, a fourth value is determined as the turbulence constraint loss value of the predicted turbulence data, wherein the third value is less than the fourth value.

[0119] The third value may or may not be the same as the first value, and the fourth value may or may not be the same as the second value; there are no restrictions here.

[0120] Since turbulence data only conforms to the laws of real fluid physics when it is non-negative, the turbulence constraint loss value is set relatively small when the predicted turbulence data is non-negative. This is to make the vehicle aerodynamic model update in a direction that makes the predicted turbulence data closer to non-negative, so that the predicted turbulence data corresponding to the vehicle aerodynamic model is more likely to conform to the laws of real fluid physics.

[0121] The fourth value is positively correlated with the difference between 0 and the predicted turbulence data. For example, the fourth value can be the difference between 0 and the predicted turbulence data. For example, if the predicted turbulence data is -5, then the fourth value is 0 - (-5) = 5.

[0122] Since turbulence data only conforms to the laws of real fluid physics when it is non-negative, the fourth value is set to be positively correlated with the difference between 0 and the predicted turbulence data. This makes the vehicle aerodynamic model update in a direction that makes the predicted turbulence data closer to non-negative, so that the predicted turbulence data corresponding to the vehicle aerodynamic model is more likely to conform to the laws of real fluid physics.

[0123] As one embodiment, the flow field data residual value, velocity constraint loss value and turbulence constraint loss value corresponding to the predicted flow field data are fused to obtain the data constraint loss value of the predicted volume flow field data, including: determining the flow field data residual value corresponding to the predicted flow field data; and determining the sum of the flow field data residual value, velocity constraint loss value and turbulence constraint loss value corresponding to the predicted flow field data as the data constraint loss value of the predicted volume flow field data.

[0124] The residual values ​​of the flow field data can be calculated using the Navier–Stokes governing equations.

[0125] Among them, the volume flow field prediction loss value of the predicted volume flow field data is positively correlated with the difference between the first true label and the predicted volume flow field data, and the wind resistance prediction loss value of the predicted wind resistance coefficient is positively correlated with the difference between the second true label and the predicted wind resistance coefficient.

[0126] In this way, by setting the predicted loss value to be positively correlated with the difference between the true label and the actual label, the vehicle aerodynamic model is updated in the direction that the predicted data is closer to the actual label, thereby improving the prediction accuracy of the vehicle aerodynamic model.

[0127] As an example, the data constraint loss value, the volume flow field prediction loss value, and the wind resistance prediction loss value are fused to obtain the total model loss value, including: determining the sum of the data constraint loss value, the volume flow field prediction loss value, and the wind resistance prediction loss value as the total model loss value.

[0128] As an example, the aerodynamic model of the vehicle to be trained is updated according to the total loss value of the model to obtain the vehicle aerodynamic model. This includes: updating the aerodynamic model of the vehicle to be trained when the total loss value representation of the model has not converged, and returning to the step of extracting multi-scale features from each geometric grid feature in the target grid feature through the feature extraction module in the aerodynamic model of the vehicle to be trained to obtain multi-scale features, until the total loss value representation of the model converges, then the current aerodynamic model of the vehicle to be trained is determined as the vehicle aerodynamic model.

[0129] The aerodynamic model of the vehicle to be trained can be updated using either gradient ascent or gradient descent, and there is no restriction on this.

[0130] In this embodiment, considering that the volume field distribution density of the training flow field data of the training vehicle is not uniform, the non-uniform and irregular training flow field data can be converted into regular and uniformly distributed flow field regular grid data by constructing flow field regular grid data. This allows the training geometric grid features to be matched with the flow field regular grid data, so that the geometric features in the training geometric grid features can correspond one-to-one with the flow field data in the flow field regular grid data, thereby determining the subsequent training samples. This makes the vehicle aerodynamic model obtained by iteratively updating the training samples more accurate.

[0131] Understandably, given the inconsistent distribution density of flow field data, a method for regularizing the flow field data is needed to ensure the training accuracy and efficiency of the vehicle aerodynamic model.

[0132] In one exemplary embodiment, such as Figure 4 As shown, step 304 includes steps 401 to 403. Wherein:

[0133] Step 401: Based on the vehicle size information of the training vehicle, match the position range of the volume flow field data for the training vehicle.

[0134] In step 401, the vehicle size information is used to characterize at least one of the following: the size of the training vehicle in the length direction (x-direction), the size of the vehicle in the width direction (y-direction), and the size of the training vehicle in the vertical direction (z-direction). The volume flow field data location range includes at least one of the following: length direction range, width direction range, and vertical direction range.

[0135] As an embodiment, step 401 includes: determining the sum of the length direction size in the vehicle size information and the preset length value as the maximum length value corresponding to the length direction range; determining the sum of the width direction size in the vehicle size information and the preset width value as the maximum width value corresponding to the width direction range; and determining the sum of the vertical direction size in the vehicle size information and the preset vertical value as the maximum vertical value corresponding to the vertical direction range.

[0136] The preset length, preset width, and preset vertical values ​​can be empirical values.

[0137] Step 402: Select target flow field volume data that fits the location range of the volume flow field data from the training flow field volume data.

[0138] Step 403: Normalize the target flow field data to obtain the flow field regular grid data corresponding to the training vehicle.

[0139] As an embodiment, step 403 includes: obtaining the prediction requirement accuracy corresponding to the vehicle aerodynamic model; determining the normalization value based on the prediction requirement accuracy; and performing normalization processing on the target flow field data based on the normalization value to obtain the flow field regular grid data corresponding to the training vehicle.

[0140] Among them, the normalized value is positively correlated with the accuracy of the predicted demand.

[0141] As an example, the target flow field data is normalized based on the normalized values ​​to obtain the flow field regular grid data corresponding to the vehicle. This includes: normalized values ​​for length, width, and vertical direction; target flow field data including first flow field data in the length direction, second flow field data in the width direction, and third flow field data in the vertical direction; and flow field regular grid data including first regular grid data in the length direction, second regular flow field data in the width direction, and third regular flow field data in the vertical direction. The ratio between the first flow field data and the length normalized value is determined as the first regular grid data corresponding to the vehicle; the ratio between the second flow field data and the width normalized value is determined as the second regular grid data corresponding to the vehicle; and the ratio between the third flow field data and the vertical normalized value is determined as the third regular grid data corresponding to the vehicle.

[0142] In this embodiment, considering that vehicle R&D needs should be for aerodynamic evaluation of a certain area around the vehicle, the location range of the volume flow field data obtained by matching the vehicle size information can achieve accurate screening of the flow field data. Considering that the distribution of the vehicle's flow field data is uneven, after normalizing the target flow field data, the obtained flow field data is presented in a regular grid form, thereby improving the model accuracy of the vehicle aerodynamic model trained with the regular grid data of the flow field.

[0143] As a detailed embodiment, during model training, the following steps are taken: First, the training outer contour model, training flow field data, and training fluid boundary condition data corresponding to the training vehicle are acquired. Second, the triangular mesh features and geometric mesh point positions of the training outer contour model are extracted to obtain the training geometric mesh features. Third, based on the vehicle size information, the position range of the volume flow field data is matched to the training vehicle. Fourth, target flow field data that fits the position range of the volume flow field data is selected from the training flow field data. Fifth, the target flow field data is normalized to obtain the flow field regular mesh data corresponding to the training vehicle. Sixth, the SDF value between each regular mesh point in the flow field regular mesh data and the nearest point on the vehicle surface is obtained. If the SDF value between the regular mesh point and the nearest point on the vehicle surface is non-negative, the positional relationship between the regular mesh point and the vehicle is determined to be outside the vehicle. In the case of negative values, the positional relationship between the regular grid points and the vehicle is determined to be represented inside the vehicle; target flow field data representing the positional relationship inside the vehicle are filtered from the flow field regular grid data; mask labels are added to the target flow field data in the flow field regular grid data; the distances between the geometric grid points and each regular grid point in the flow field regular grid are obtained; the grid point with the shortest corresponding distance in the flow field regular grid is determined as the target regular grid point; the geometric features of the geometric grid points are matched with the flow field features of the target regular grid points to obtain matching data; the training geometric grid features and training fluid boundary conditions in the matching data are determined as the input feature data of a training sample; the flow field regular grid data in the matching data is determined as the first true label of the training sample; and the true drag coefficient corresponding to the training geometric grid features is determined as the second true label of the training sample.

[0144] Furthermore, multi-scale feature extraction is performed on the positions of each geometric grid point and the normal vectors of each triangular face in the target grid features using the aerodynamic model of the vehicle to be trained, to obtain training multi-scale features; unified structure encoding is applied to the multi-scale features to obtain training multi-scale encoded features; the normal vectors of each triangular face in the training geometric grid features are encoded to obtain training triangular face normal vector encoding; the training fluid boundary condition data is encoded to obtain training fluid boundary condition encoding; the training multi-scale encoded features, training triangular face normal vector encoding, and training fluid boundary condition encoding are concatenated to obtain training global aerodynamic features; the volume flow field of the training vehicle is predicted based on the input feature data of the training samples using the aerodynamic model of the vehicle to be trained, to obtain predicted volume flow field data; and the wind force is predicted based on the input feature data of the training samples using the aerodynamic model of the vehicle to be trained, to obtain predicted drag coefficient; the predicted velocity field data is matched with zero constraint conditions to obtain the velocity constraint loss value of the predicted velocity field data; and the predicted turbulence data is matched with non-negative constraint conditions to obtain predicted turbulence. The data constraint loss value is calculated as follows: Turbulence constraint loss value; The residual value of the flow field data, velocity constraint loss value, and turbulence constraint loss value corresponding to the predicted flow field data are fused to obtain the data constraint loss value of the predicted volume flow field data; The volume flow field prediction accuracy of the vehicle aerodynamic model to be trained is evaluated based on the difference between the first true label in the training samples and the predicted volume flow field data, resulting in the volume flow field prediction loss value; The drag prediction accuracy of the vehicle aerodynamic model to be trained is evaluated based on the difference between the second true label in the training samples and the predicted drag coefficient, resulting in the drag prediction loss value; The sum of the data constraint loss value, the volume flow field prediction loss value, and the drag prediction loss value is determined as the total model loss value; If the total model loss value representation does not converge, the vehicle aerodynamic model to be trained is updated, and the process returns to the step of multi-scale feature extraction of each geometric grid feature in the target grid feature through the feature extraction module in the vehicle aerodynamic model to be trained, until the total model loss value representation converges, at which point the current vehicle aerodynamic model to be trained is determined as the vehicle aerodynamic model.

[0145] Furthermore, during model application, the aerodynamic model, outer contour model, and fluid boundary condition data of the target vehicle are acquired; the triangular mesh features and geometric mesh point positions of the target outer contour model are extracted to obtain the target geometric mesh features; multi-scale feature extraction is performed on the geometric mesh point positions and triangular normal vectors in the target geometric mesh features using the vehicle aerodynamic model to obtain the target multi-scale features; unified structural encoding is performed on the target multi-scale features using the vehicle aerodynamic model to obtain multi-scale encoded features; and encoding processing is performed on the target fluid boundary condition data using the vehicle aerodynamic model to obtain... Fluid boundary condition encoding is performed by encoding the normal vectors of each triangular facet in the target geometric mesh features using a vehicle aerodynamic model, resulting in triangular facet normal vector encoding. The multi-scale encoded features, triangular facet normal vector encoding, and fluid boundary condition encoding are then concatenated to obtain global aerodynamic features. Based on these global aerodynamic features, the vehicle aerodynamic model predicts the volume flow field of the target vehicle, obtaining target volume flow field data. Volume flow field features are extracted from the global aerodynamic features using the vehicle aerodynamic model. Finally, based on the global aerodynamic features and volume flow field features, the vehicle aerodynamic model predicts the wind conditions of the vehicle, obtaining the predicted drag coefficient.

[0146] Thus, geometric features are extracted from the outer contour model of the target vehicle, allowing the extracted geometric mesh features to characterize the positional and topological features of the vehicle's outer contour. Multi-scale feature extraction is performed on the target geometric features, resulting in richer feature content and more feature dimensions. Using a pre-trained vehicle aerodynamic model, the multi-scale features and target fluid boundary condition data are used as the basis for aerodynamic feature evaluation. This enables the vehicle aerodynamic model to quickly and accurately construct regular flow field mesh data for aerodynamic feature evaluation. Furthermore, by matching the mesh point positions between geometric mesh features and regular flow field mesh data, a certain correspondence is established between irregular geometric mesh features and regular flow field mesh data, improving the computational efficiency of subsequent model training. The resulting vehicle aerodynamic model, capable of evaluating vehicle aerodynamic features, thus improves the efficiency of aerodynamic feature evaluation.

[0147] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0148] Based on the same inventive concept, this application also provides a vehicle aerodynamic feature evaluation device for implementing the vehicle aerodynamic feature evaluation method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more vehicle aerodynamic feature evaluation device embodiments provided below can be found in the limitations of the vehicle aerodynamic feature evaluation method described above, and will not be repeated here.

[0149] In one exemplary embodiment, such as Figure 5 As shown, a vehicle aerodynamic feature evaluation device is provided, comprising: a data acquisition module, a feature extraction module, and a feature evaluation module, wherein:

[0150] The data acquisition module is used to acquire the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle.

[0151] The feature extraction module is used to extract geometric features from the outer contour model of the target to obtain the target geometric mesh features of the target vehicle; and to extract multi-scale features from the target geometric mesh features through the vehicle aerodynamic model to obtain the target multi-scale features.

[0152] The feature evaluation module is used to evaluate the aerodynamic characteristics of the target vehicle based on the multi-scale characteristics of the target and the target fluid boundary condition data using the vehicle aerodynamic model, thereby obtaining the aerodynamic characteristics of the target vehicle.

[0153] In one embodiment, the vehicle aerodynamic feature evaluation device further includes: a model training module, used to acquire training outer contour model, training flow field data, and training fluid boundary condition data corresponding to the training vehicle, and to construct flow field regular mesh data of the training vehicle based on the training flow field data; to extract geometric features from the training outer contour model to obtain training geometric mesh features of the training vehicle; to match the training geometric mesh features and the flow field regular mesh data to obtain multiple sets of matching data; for each set of matching data, to determine the training geometric mesh features and training fluid boundary conditions in the matching data as input feature data of a training sample, to determine the flow field regular mesh data in the matching data as the first true label of the training sample, and to determine the true drag coefficient corresponding to the training geometric mesh features as the second true label of the training sample; and to iteratively update the vehicle aerodynamic model based on the multiple training samples determined by the multiple sets of matching data.

[0154] In one embodiment, the model training module is further configured to, for each training sample, predict the volume flow field of the training vehicle based on the input feature data of the training sample using the aerodynamic model of the vehicle to be trained, to obtain predicted volume flow field data; and predict the wind conditions of the training vehicle based on the input feature data of the training sample using the aerodynamic model of the vehicle to be trained, to obtain a predicted drag coefficient; match the data constraints corresponding to the predicted volume flow field data with the predicted volume flow field data to obtain a data constraint loss value for the predicted volume flow field data; evaluate the volume flow field prediction accuracy of the aerodynamic model of the vehicle to be trained based on the difference between the first true label in the training sample and the predicted volume flow field data, to obtain a volume flow field prediction loss value for the predicted volume flow field data; evaluate the drag prediction accuracy of the aerodynamic model of the vehicle to be trained based on the difference between the second true label in the training sample and the predicted drag coefficient, to obtain a drag prediction loss value for the predicted drag coefficient; fuse the data constraint loss value, the volume flow field prediction loss value, and the drag prediction loss value to obtain a total model loss value; and update the aerodynamic model of the vehicle to be trained based on the total model loss value to obtain a vehicle aerodynamic model.

[0155] In one embodiment, the feature extraction module is further configured to match the position range of volume flow field data for the training vehicle based on the vehicle size information of the training vehicle; filter target flow field volume data that conforms to the position range of volume flow field data from the training flow field volume data; and normalize the target flow field volume data to obtain the flow field volume regular grid data of the training vehicle.

[0156] In one embodiment, the training geometric mesh features include geometric features of multiple geometric mesh points, and the flow field volume regular mesh data includes flow field volume data of multiple regular mesh points; the vehicle aerodynamic feature evaluation device further includes: a feature matching module, used to obtain the distance between each geometric mesh point in the training geometric mesh features and each regular mesh point in the flow field volume regular mesh for each geometric mesh point; and to filter target regular mesh points that match the geometric mesh points from the flow field volume regular mesh based on the distances between the geometric mesh points and each regular mesh point in the flow field volume regular mesh to obtain matching data.

[0157] Each module in the aforementioned vehicle aerodynamic characteristic evaluation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0158] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a method for evaluating vehicle aerodynamic characteristics. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0159] Those skilled in the art will understand that Figure 6The structure shown is a block diagram of a partial structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0160] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0161] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0162] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0163] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program mentioned can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0165] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0166] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for evaluating the aerodynamic characteristics of a vehicle, characterized in that, The method includes: Acquire the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle; Geometric features are extracted from the outer contour model of the target to obtain the target geometric mesh features of the target vehicle. Multi-scale feature extraction of the target geometric mesh features is performed using the vehicle aerodynamic model to obtain the target multi-scale features. The aerodynamic characteristics of the target vehicle are evaluated using the vehicle aerodynamic model based on the multi-scale characteristics of the target and the target fluid boundary condition data, thereby obtaining the aerodynamic characteristics of the target vehicle.

2. The method according to claim 1, characterized in that, The training process for the vehicle aerodynamic model includes: Obtain the training outer contour model, training flow field data, and training fluid boundary condition data corresponding to the training vehicle, and construct the flow field regular mesh data of the training vehicle based on the training flow field data. Geometric features are extracted from the training outer contour model to obtain the training geometric mesh features of the training vehicle; The training geometric mesh features and the flow field volume regular mesh data are matched accordingly to obtain multiple sets of matching data; For each set of matching data, the training geometric mesh features and the training fluid boundary conditions in the matching data are determined as the input feature data of a training sample, the flow field regular mesh data in the matching data are determined as the first true label of the training sample, and the true drag coefficient corresponding to the training geometric mesh features is determined as the second true label of the training sample. The vehicle aerodynamic model is obtained by iteratively updating multiple training samples determined by the multiple sets of matching data.

3. The method according to claim 2, characterized in that, The step of iteratively updating the vehicle aerodynamic model based on multiple training samples determined from the multiple sets of matching data includes: For each training sample, the aerodynamic model of the vehicle to be trained predicts the flow field of the training vehicle based on the input feature data of the training sample, and obtains the predicted flow field data. Also, the aerodynamic model of the vehicle to be trained predicts the wind force of the training vehicle based on the input feature data of the training sample, and obtains the predicted drag coefficient. The data constraint conditions corresponding to the predicted volume flow field data are matched with the predicted volume flow field data to obtain the data constraint loss value of the predicted volume flow field data; Based on the difference between the first true label in the training sample and the predicted volume flow field data, the volume flow field prediction accuracy of the vehicle aerodynamic model to be trained is evaluated, and the volume flow field prediction loss value of the predicted volume flow field data is obtained. The drag prediction accuracy of the aerodynamic model of the vehicle to be trained is evaluated based on the difference between the second true label in the training sample and the predicted drag coefficient, and the drag prediction loss value of the predicted drag coefficient is obtained. The data constraint loss value, the volume flow field prediction loss value, and the wind resistance prediction loss value are fused to obtain the total model loss value; Based on the total loss value of the model, the aerodynamic model of the vehicle to be trained is updated to obtain the vehicle aerodynamic model.

4. The method according to claim 2, characterized in that, Before iteratively updating the vehicle aerodynamic model based on multiple training samples determined from the multiple sets of matching data, the method further includes: Obtain the positional relationship between each regular grid point in the flow field regular grid data and the vehicle; Filter the target volume flow field data representing the positional relationship inside the vehicle from the flow field regular grid data; Based on the flow field data of the target body, the regular grid data of the flow field body is processed.

5. The method according to claim 2, characterized in that, The step of constructing the flow field regular mesh data of the training vehicle based on the training flow field data includes: Based on the vehicle size information of the training vehicle, the location range of the body flow field data for the training vehicle is determined; Target flow field data that fits the location range of the volume flow field data are selected from the training flow field volume data; The target flow field data is normalized to obtain the flow field regular grid data of the training vehicle.

6. The method according to any one of claims 2 to 5, characterized in that, The training geometric mesh features include geometric features of multiple geometric mesh points, and the flow field volume regular mesh data includes flow field volume data of multiple regular mesh points; The step involves matching the training geometric mesh features with the flow field volume regular mesh data to obtain multiple sets of matching data, including: For each geometric grid point in the training geometric grid features, the distance between the geometric grid point and each regular grid point in the flow field regular grid is obtained; Based on the distance between the geometric grid point and each regular grid point in the flow field regular grid, target regular grid points that match the geometric grid points are selected from the flow field regular grid to obtain matching data.

7. A vehicle aerodynamic characteristic evaluation device, characterized in that, The device includes: The data acquisition module is used to acquire the vehicle aerodynamic model, target outer contour model, and target fluid boundary condition data corresponding to the target vehicle. The feature extraction module is used to extract geometric features from the outer contour model of the target to obtain the target geometric mesh features of the target vehicle; and to extract multi-scale features from the target geometric mesh features using the vehicle aerodynamic model to obtain the target multi-scale features. The feature evaluation module is used to evaluate the aerodynamic characteristics of the target vehicle based on the target multi-scale features and the target fluid boundary condition data using the vehicle aerodynamic model, thereby obtaining the aerodynamic characteristics of the target vehicle.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images