A method and system for dynamic aerodynamic testing of a commercial vehicle
By dividing dynamic operating condition clusters in commercial vehicle dynamic aerodynamic testing and using transfer learning and hash indexing algorithms, the problem of low efficiency in detecting deviations between simulation data and measured data in commercial vehicle aerodynamic testing was solved, achieving efficient aerodynamic testing and accurate data matching.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JAINGXI ISUZU AUTOMOBILE CO LTD
- Filing Date
- 2025-12-26
- Publication Date
- 2026-05-19
AI Technical Summary
Existing CFD software cannot accurately detect the deviation between simulation data and measured data in dynamic aerodynamic testing of commercial vehicles, resulting in longer testing cycles, low efficiency, and potential misleading of design direction, thus increasing R&D costs.
A pre-defined clustering algorithm combined with aerodynamic drag sensitivity analysis is used to divide dynamic operating condition clusters. Initial correction coefficients are generated through transfer learning algorithm, and hash index algorithm is used for fast matching. The target deviation value is calculated to complete the aerodynamic test.
Accurate detection of deviations between simulation data and measured data improves testing efficiency, ensures a high degree of consistency between simulation data and measured data, reduces the risk of misleading design, and lowers R&D costs.
Smart Images
Figure CN121409637B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of commercial vehicle testing technology, and in particular to a dynamic aerodynamic testing method and system for commercial vehicles. Background Technology
[0002] The aerodynamic performance of commercial vehicles directly impacts energy consumption, driving stability, and operational safety. Dynamic aerodynamic testing is a core support for their optimized design, and a standard process of "CFD simulation - wind tunnel calibration - real vehicle verification" has been established. Among these, CFD simulation has become a core preliminary step due to its low cost and flexible scenarios, but the accuracy of its matching with measured data directly determines the value of the optimization solution. The unique aerodynamic phenomena of the tractor-trailer combination structure in commercial vehicles, such as the gap effect and tail vortex system, make their flow field far more complex than that of passenger vehicles. Existing CFD software calibration models are mostly developed based on general structures and are not adapted to the characteristics of commercial vehicles, resulting in significant deviations between simulation and wind tunnel / real vehicle measured data. Drag coefficient deviations reach ±5%~8%, far exceeding the ±2% engineering error threshold.
[0003] A more prominent problem is the difficulty in identifying the sources of deviation. Engineers must rely on experience to systematically check factors such as mesh, model, and boundary conditions, which is inefficient and prone to misjudgment. This not only extends the optimization cycle by more than 30% but may also mislead the design direction and increase R&D costs. Therefore, building a testing system adapted to the characteristics of commercial vehicles and solving the problems of missing calibration models and deviation tracing has become an urgent need for the industry.
[0004] Furthermore, and more critically, the existing testing system lacks an efficient mechanism for identifying the sources of deviation. When discrepancies arise between simulation and measured data, engineers must spend a significant amount of time meticulously checking various factors such as mesh generation quality, geometric model simplification, and the rationality of boundary condition settings. This troubleshooting process is highly dependent on individual engineering experience, making misjudgments or omissions likely. This problem of "large deviations and difficulty in tracing their origins" not only reduces testing efficiency, extending the aerodynamic optimization design cycle by more than 30%, but may also mislead optimization directions due to distorted simulation data. This can cause some simulation-based improvement schemes to fail to achieve the expected results in real-world vehicle applications, and even increase the number of repetitions between wind tunnel testing and real-world testing, significantly increasing R&D costs. Summary of the Invention
[0005] Based on this, the purpose of this invention is to provide a dynamic aerodynamic testing method and system for commercial vehicles, so as to solve the problem that the existing technology cannot accurately detect the deviation between simulation data and actual measurement data during the dynamic aerodynamic testing of commercial vehicles, which leads to a longer testing cycle and a corresponding reduction in testing efficiency.
[0006] The first aspect of the present invention proposes:
[0007] A dynamic aerodynamic testing method for commercial vehicles specifically includes the following steps:
[0008] Based on the driving data of commercial vehicles throughout their entire life cycle, a preset clustering algorithm is used in conjunction with aerodynamic drag sensitivity analysis to divide the vehicle into several dynamic operating condition clusters, and the aerodynamic feature thresholds of each dynamic operating condition cluster are extracted.
[0009] A mapping relationship between simulation data and measured data is constructed. Based on the mapping relationship, a transfer learning algorithm is used to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients.
[0010] The boundary conditions and turbulence model parameters of the preset simulation model are pre-adjusted according to the initial correction coefficient and the aerodynamic characteristic threshold to generate the corresponding initial correction simulation data.
[0011] A hash indexing algorithm is used to quickly match the actual collected aerodynamic parameters with the initial corrected simulation data to calculate the corresponding target deviation value. Based on the target deviation value and the initial corrected simulation data, the corresponding target corrected simulation data is output to complete the aerodynamic test of the commercial vehicle.
[0012] The beneficial effects of this invention are as follows: By collecting driving data from commercial vehicles, a dynamic operating condition cluster for subsequent analysis can be generated. Based on this, to facilitate subsequent judgment, the corresponding aerodynamic feature thresholds are extracted. Based on this, the corresponding initial correction coefficients are generated according to the current dynamic operating condition cluster. At the same time, the preset simulation model can be adjusted to obtain the corresponding simulation data. Based on this, using the current aerodynamic parameters and simulation data as the calculation basis, the required target deviation value is calculated, and the final required target correction simulation data is obtained to complete the corresponding aerodynamic test. This allows for accurate detection of the deviation between the simulation data and the measured data, and corresponding adjustments are made, thus improving testing efficiency.
[0013] Furthermore, the step of dividing the driving data based on the entire life cycle of commercial vehicles into several dynamic operating condition clusters using a preset clustering algorithm combined with aerodynamic drag sensitivity analysis includes:
[0014] Data on driving speed, acceleration, three-dimensional environmental wind field, aerodynamic drag, and vehicle accessory status generated by the commercial vehicle during the test are collected. Corresponding strong correlation combinations are identified through association rule mining algorithms to create corresponding aerodynamic sensitive feature sets.
[0015] The aerodynamic sensitive feature set is standardized to generate a corresponding standard feature set;
[0016] The standard feature set is clustered once using the FCM algorithm to generate the corresponding initial operating condition cluster. An appropriate dynamic drag gradient threshold is set for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster.
[0017] Furthermore, the step of setting an appropriate dynamic drag gradient threshold for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster includes:
[0018] The core parameters of each initial operating condition cluster are extracted, and the sensitivity coefficient of each core parameter relative to the starting resistance is calculated using Sobol global sensitivity analysis.
[0019] Each of the aforementioned sensitivity coefficients is fitted using kernel density estimation to generate a corresponding drag distribution curve. The peak value of the drag distribution curve is set as the reference drag, and a dynamic threshold for the adapted working condition is generated by combining it with an altitude correction factor.
[0020] The initial operating condition cluster is substituted into the dynamic threshold, and dynamic verification is performed simultaneously to generate the corresponding dynamic operating condition cluster.
[0021] Furthermore, the step of using a transfer learning algorithm based on the mapping relationship to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate corresponding initial correction coefficients includes:
[0022] The initial correlation features between the correction parameters of the baseline working condition and the aerodynamic characteristic thresholds of the dynamic working condition cluster are extracted. The feature difference distribution is quantified by the KL algorithm and the target correlation features are selected by combining the mutual information method.
[0023] Using the simulation-test mapping as the loss constraint, a corresponding dual-branch network is constructed;
[0024] The target associated features are transferred into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients accordingly.
[0025] Furthermore, the step of transferring the target correlation features into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients includes:
[0026] Deep information in the target association features is extracted by multi-scale dilated convolution, and a dynamic perturbation factor adapted to the dynamic working condition cluster is introduced to perform corresponding feature perturbation training in order to output the corresponding enhanced association features.
[0027] The enhanced correlation features are input into the dual-branch network, and the feature distribution of the dynamic working condition cluster is detected by the meta-learning strategy in order to perform constraint training and output the corresponding candidate correction coefficient set.
[0028] The candidate correction coefficient set is hierarchically verified to screen out effective coefficients, and Gaussian process regression is used to smooth and correct the effective coefficients to output the initial correction coefficients.
[0029] Furthermore, the step of using a hash index algorithm to quickly match the actually collected aerodynamic parameters with the initial corrected simulation data to calculate the corresponding target deviation value includes:
[0030] The actual aerodynamic parameters are analyzed and processed using a dual-channel attention mechanism to output the corresponding steady-state parameters and transient parameters. The temporal features contained in the steady-state parameters and the extreme value features contained in the transient parameters are extracted to create the corresponding structural feature vector.
[0031] The temporal trend of the structural feature vector is encoded into a dynamic hash fragment using the hash indexing algorithm, and aerodynamic theoretical constraints corresponding to the commercial vehicle are collected.
[0032] The target deviation value is generated based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment.
[0033] Furthermore, the step of generating the target deviation value based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment includes:
[0034] Based on the aerodynamic characteristic threshold of the dynamic working condition cluster, the aerodynamic theoretical constraints are transformed into multi-dimensional constraint functions, and the dynamic hash fragments are substituted into the multi-dimensional constraint functions to filter out the effective feature intervals that meet the constraint conditions.
[0035] The long-term dependence features of the aerodynamic parameters are captured by a preset encoder to generate a corresponding correlation feature matrix;
[0036] The effective feature intervals are fused into the interior of the associated feature matrix to generate a corresponding target deviation matrix, and the target deviation matrix is converted into the target deviation value accordingly.
[0037] The second aspect of the present invention proposes:
[0038] A dynamic aerodynamic testing system for commercial vehicles, wherein the system comprises:
[0039] The segmentation module is used to divide the driving data of commercial vehicles throughout their entire life cycle into several dynamic operating condition clusters by using a preset clustering algorithm combined with aerodynamic drag sensitivity analysis, and extracting the aerodynamic feature thresholds of each dynamic operating condition cluster.
[0040] The construction module is used to construct the mapping relationship between simulation data and measured data. Based on the mapping relationship, the transfer learning algorithm is used to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients.
[0041] The adjustment module is used to pre-adjust the boundary conditions and turbulence model parameters of the preset simulation model according to the initial correction coefficient and the aerodynamic characteristic threshold, so as to generate the corresponding initial correction simulation data.
[0042] The matching module is used to quickly match the actually collected aerodynamic parameters with the initial correction simulation data using a hash index algorithm to calculate the corresponding target deviation value. Based on the target deviation value and the initial correction simulation data, the corresponding target correction simulation data is output to complete the aerodynamic test of the commercial vehicle.
[0043] Furthermore, the partitioning module is specifically used for:
[0044] Data on driving speed, acceleration, three-dimensional environmental wind field, aerodynamic drag, and vehicle accessory status generated by the commercial vehicle during the test are collected. Corresponding strong correlation combinations are identified through association rule mining algorithms to create corresponding aerodynamic sensitive feature sets.
[0045] The aerodynamic sensitive feature set is standardized to generate a corresponding standard feature set;
[0046] The standard feature set is clustered once using the FCM algorithm to generate the corresponding initial operating condition cluster. An appropriate dynamic drag gradient threshold is set for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster.
[0047] Furthermore, the partitioning module is specifically used for:
[0048] The core parameters of each initial operating condition cluster are extracted, and the sensitivity coefficient of each core parameter relative to the starting resistance is calculated using Sobol global sensitivity analysis.
[0049] Each of the aforementioned sensitivity coefficients is fitted using kernel density estimation to generate a corresponding drag distribution curve. The peak value of the drag distribution curve is set as the reference drag, and a dynamic threshold for the adapted working condition is generated by combining it with an altitude correction factor.
[0050] The initial operating condition cluster is substituted into the dynamic threshold, and dynamic verification is performed simultaneously to generate the corresponding dynamic operating condition cluster.
[0051] Furthermore, the adjustment module is specifically used for:
[0052] The initial correlation features between the correction parameters of the baseline working condition and the aerodynamic characteristic thresholds of the dynamic working condition cluster are extracted. The feature difference distribution is quantified by the KL algorithm and the target correlation features are selected by combining the mutual information method.
[0053] Using the simulation-test mapping as the loss constraint, a corresponding dual-branch network is constructed;
[0054] The target associated features are transferred into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients accordingly.
[0055] Furthermore, the adjustment module is specifically used for:
[0056] Deep information in the target association features is extracted by multi-scale dilated convolution, and a dynamic perturbation factor adapted to the dynamic working condition cluster is introduced to perform corresponding feature perturbation training in order to output the corresponding enhanced association features.
[0057] The enhanced correlation features are input into the dual-branch network, and the feature distribution of the dynamic working condition cluster is detected by the meta-learning strategy in order to perform constraint training and output the corresponding candidate correction coefficient set.
[0058] The candidate correction coefficient set is hierarchically verified to screen out effective coefficients, and Gaussian process regression is used to smooth and correct the effective coefficients to output the initial correction coefficients.
[0059] Furthermore, the matching module is specifically used for:
[0060] The actual aerodynamic parameters are analyzed and processed using a dual-channel attention mechanism to output the corresponding steady-state parameters and transient parameters. The temporal features contained in the steady-state parameters and the extreme value features contained in the transient parameters are extracted to create the corresponding structural feature vector.
[0061] The temporal trend of the structural feature vector is encoded into a dynamic hash fragment using the hash indexing algorithm, and aerodynamic theoretical constraints corresponding to the commercial vehicle are collected.
[0062] The target deviation value is generated based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment.
[0063] Furthermore, the matching module is specifically used for:
[0064] Based on the aerodynamic characteristic threshold of the dynamic working condition cluster, the aerodynamic theoretical constraints are transformed into multi-dimensional constraint functions, and the dynamic hash fragments are substituted into the multi-dimensional constraint functions to filter out the effective feature intervals that meet the constraint conditions.
[0065] The long-term dependence features of the aerodynamic parameters are captured by a preset encoder to generate a corresponding correlation feature matrix;
[0066] The effective feature intervals are fused into the interior of the associated feature matrix to generate a corresponding target deviation matrix, and the target deviation matrix is converted into the target deviation value accordingly.
[0067] The third aspect of the present invention proposes:
[0068] A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the dynamic aerodynamic testing method for commercial vehicles as described above.
[0069] The fourth aspect of the present invention proposes:
[0070] A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the dynamic aerodynamic testing method for commercial vehicles as described above.
[0071] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0072] Figure 1 A flowchart of the dynamic aerodynamic testing method for commercial vehicles provided in the first embodiment of the present invention;
[0073] Figure 2 This is a structural block diagram of a commercial vehicle dynamic aerodynamic testing system provided in the third embodiment of the present invention.
[0074] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation
[0075] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
[0076] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.
[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0078] Please see Figure 1 The figure shows a dynamic aerodynamic testing method for commercial vehicles provided in the first embodiment of the present invention. The dynamic aerodynamic testing method for commercial vehicles provided in this embodiment can accurately detect the data deviation generated by the commercial vehicle during the testing process, thereby improving the testing efficiency.
[0079] Specifically, this embodiment provides:
[0080] A dynamic aerodynamic testing method for commercial vehicles specifically includes the following steps:
[0081] Step S10: Based on the driving data of the commercial vehicle throughout its entire life cycle, a preset clustering algorithm is used in conjunction with aerodynamic drag sensitivity analysis to divide the vehicle into several dynamic operating condition clusters and extract the aerodynamic feature thresholds of each dynamic operating condition cluster.
[0082] It's important to note that the first step involves dividing the commercial vehicle's aerodynamic drag into dynamic operating condition clusters based on its full lifecycle driving data (covering multiple scenarios including highways, national roads, and mountainous areas). Commercial vehicle aerodynamic drag is influenced by multiple factors such as speed, wind field, and vehicle condition. Directly testing all operating conditions is both time-consuming and redundant. By employing a pre-defined clustering algorithm (such as fuzzy clustering) combined with aerodynamic drag sensitivity analysis, operating conditions with similar aerodynamic characteristics and consistent drag influence patterns can be grouped into one cluster (e.g., the "high speed + tailwind + fairing open" cluster, the "low speed + crosswind + fully loaded cargo box" cluster). Simultaneously, aerodynamic feature thresholds for each cluster are extracted (e.g., the speed range of this cluster is 80-110 km / h, and the crosswind speed is ≤5 m / s). Specifically, clustering allows the testing to focus on key operating conditions, and the feature thresholds provide boundary criteria for subsequent simulations, facilitating subsequent processing.
[0083] Step S20: Construct a mapping relationship between simulation data and measured data. Based on the mapping relationship, use a transfer learning algorithm to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients.
[0084] It's important to note that the second step addresses the issue of insufficient dynamic operating condition data. While baseline operating conditions (such as uniform velocity without crosswind in a standard wind tunnel) have complete data and reliable correction parameters, measured data for dynamic operating conditions (such as random crosswind conditions) is scarce, leading to significant modeling errors when modeled directly. By establishing a mapping relationship between simulation and measured data (clarifying the correspondence between simulation parameters and measured aerodynamic drag), a transfer learning algorithm is used to transfer the correction parameters of the baseline operating conditions (such as turbulence model correction coefficients) to the dynamic operating condition cluster. Specifically, transfer learning can reuse effective information from the baseline operating conditions, reducing the measurement costs of dynamic operating conditions and generating initial correction coefficients (preliminary simulation correction parameters adapted to each dynamic operating condition). This facilitates subsequent processing.
[0085] Step S30: Based on the initial correction coefficient and the aerodynamic characteristic threshold, the boundary conditions and turbulence model parameters of the preset simulation model are pre-adjusted to generate the corresponding initial correction simulation data.
[0086] It should be noted that the third step is to optimize the simulation model: combining the initial correction coefficients (parameter-level correction) and aerodynamic characteristic thresholds (boundary-level constraints), the boundary conditions (such as wind direction and air density) and turbulence model parameters (such as the turbulent kinetic energy coefficient of the k-ε model) of the preset simulation model are adjusted to generate initial corrected simulation data. Specifically, this data has been initially adapted to dynamic working conditions and is closer to reality than the uncorrected simulation, but it still needs to be calibrated with the measured data.
[0087] Step S40: The actual collected aerodynamic parameters are quickly matched with the initial correction simulation data using a hash index algorithm to calculate the corresponding target deviation value. Based on the target deviation value and the initial correction simulation data, the corresponding target correction simulation data is output to complete the aerodynamic test of the commercial vehicle.
[0088] It's important to note that the fourth step, precise matching and testing, employs a hash index algorithm (an efficient data matching algorithm suitable for the real-time needs of dynamic testing) to quickly match the actually collected aerodynamic parameters (such as aerodynamic drag and lift measured during driving) with the initial corrected simulation data, calculating the target deviation value (e.g., the measured drag being 5% greater than the simulation). Based on this deviation value, the initial simulation data is corrected, and the target corrected simulation data is output. Specifically, at this point, the simulation data and the measured data are highly consistent, completing the dynamic aerodynamic testing of commercial vehicles, ensuring both accuracy and efficiency. This facilitates subsequent processing.
[0089] Second Embodiment
[0090] Furthermore, the step of dividing the driving data based on the entire life cycle of commercial vehicles into several dynamic operating condition clusters using a preset clustering algorithm combined with aerodynamic drag sensitivity analysis includes:
[0091] Data on driving speed, acceleration, three-dimensional environmental wind field, aerodynamic drag, and vehicle accessory status generated by the commercial vehicle during the test are collected. Corresponding strong correlation combinations are identified through association rule mining algorithms to create corresponding aerodynamic sensitive feature sets.
[0092] The aerodynamic sensitive feature set is standardized to generate a corresponding standard feature set;
[0093] The standard feature set is clustered once using the FCM algorithm to generate the corresponding initial operating condition cluster. An appropriate dynamic drag gradient threshold is set for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster.
[0094] It should be noted that the first step involves collecting core driving and aerodynamic data, including driving speed (which directly affects aerodynamic drag, increasing quadratically at high speeds), acceleration (which indirectly affects drag due to changes in vehicle attitude during acceleration), three-dimensional environmental wind field (crosswinds and tailwinds are key variables for dynamic drag), aerodynamic drag (a core test indicator), and the status of vehicle accessories (such as the opening and closing of fairings and side skirts, which directly alter the aerodynamic shape of the vehicle). Strongly correlated combinations are identified using association rule mining algorithms (such as the Apriori algorithm). For example, the combination of "speed ≥ 90 km / h + crosswind speed 3-5 m / s + fairing open" is strongly correlated with "a sudden increase in aerodynamic drag of 15%". These combinations constitute a set of aerodynamic sensitive features. Specifically, focusing on sensitive features avoids interference from irrelevant data, making the division of operating condition clusters more closely aligned with the core aerodynamic influencing factors.
[0095] The second step is to standardize the sensitive feature set: different features have large differences in units (such as speed in km / h, wind in m / s, and drag in N). Direct clustering would lead to excessively high weights for features with large values such as "speed". Z-score standardization is used to convert each feature into a standard value with a mean of 0 and a standard deviation of 1, generating a standard feature set. Specifically, this eliminates the influence of units and ensures that the weights of each feature are balanced in clustering.
[0096] The third step employs a two-layer clustering approach to generate operating condition clusters: First, the standard feature set is clustered once using the FCM algorithm (Fuzzy C-means clustering, suitable for scenarios with "fuzzy boundaries," such as the lack of an absolute boundary between "medium speed" and "high speed") to generate initial operating condition clusters (e.g., preliminary classifications like "low speed with no wind" and "high speed with crosswind"). However, these initial clusters do not highlight the differences in the core aerodynamic indicator "drag." Therefore, a dynamic drag gradient threshold is set for each initial cluster through aerodynamic drag sensitivity analysis (e.g., the drag gradient threshold for the "high speed crosswind cluster" is 5 N / km·h, meaning that for every 1 km / h increase in speed, the drag changes by approximately 5 N). Based on this threshold, a second clustering is performed. Specifically, subclusters with consistent drag change patterns are merged, while those with large differences are split, ultimately generating dynamic operating condition clusters. This two-layer clustering approach considers both multiple factors and focuses on the core aerodynamic indicator, ensuring the operating condition clusters are specifically tailored to aerodynamic testing, thus facilitating subsequent processing.
[0097] Furthermore, the step of setting an appropriate dynamic drag gradient threshold for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster includes:
[0098] The core parameters of each initial operating condition cluster are extracted, and the sensitivity coefficient of each core parameter relative to the starting resistance is calculated using Sobol global sensitivity analysis.
[0099] Each of the aforementioned sensitivity coefficients is fitted using kernel density estimation to generate a corresponding drag distribution curve. The peak value of the drag distribution curve is set as the reference drag, and a dynamic threshold for the adapted working condition is generated by combining it with an altitude correction factor.
[0100] The initial operating condition cluster is substituted into the dynamic threshold, and dynamic verification is performed simultaneously to generate the corresponding dynamic operating condition cluster.
[0101] It should be noted that the first step involves extracting the core parameters of the initial operating condition cluster (such as the average speed, maximum crosswind, and typical vehicle body accessory status of the cluster), and using Sobol global sensitivity analysis to calculate the sensitivity coefficient of each parameter to aerodynamic drag. Sobol analysis can quantify the impact of each parameter and parameter interaction on drag (such as the sensitivity coefficient of "speed" 0.7, "crosswind" 0.2, and "vehicle body accessories" 0.1), and identify the core influencing parameters (such as speed). Specifically, this provides a basis for threshold setting of "which parameters need to be considered".
[0102] The second step generates a dynamic drag threshold: Kernel density estimation (KDE) is used to fit the drag data distribution corresponding to each sensitivity coefficient, generating a drag distribution curve. Specifically, the peak value of the curve represents the most common drag value for this operating condition cluster, which is set as the baseline drag. Considering the altitude differences in commercial vehicle operation (altitude affects air density, which in turn affects aerodynamic drag; the higher the altitude, the lower the drag), an altitude correction factor is introduced (e.g., a correction factor of 0.92 for an altitude of 1000m). The baseline drag is multiplied by the correction factor, and combined with the fluctuation range of the core parameters of this cluster (e.g., speed fluctuation ±5km / h), a dynamic threshold adapted to this operating condition is generated (instead of a fixed threshold, such as "baseline drag 800N±50N"). Specifically, the dynamic threshold adapts to different altitudes and parameter fluctuations, conforming to the dynamic scenario of "full life cycle driving" for commercial vehicles.
[0103] The third step is to verify and generate the operating condition cluster: Substitute all samples from the initial operating condition cluster into the dynamic threshold to determine whether the sample drag is within the threshold range; simultaneously, perform dynamic verification. Specifically, collect new driving data within the cluster to verify whether the threshold can cover the drag changes of the new samples. If the deviation exceeds 10%, adjust the threshold. After verification, samples that meet the threshold are retained as the same dynamic operating condition cluster to ensure that the aerodynamic drag change patterns of samples within the cluster are consistent, providing an accurate operating condition basis for subsequent simulation corrections.
[0104] Furthermore, the step of using a transfer learning algorithm based on the mapping relationship to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate corresponding initial correction coefficients includes:
[0105] The initial correlation features between the correction parameters of the baseline working condition and the aerodynamic characteristic thresholds of the dynamic working condition cluster are extracted. The feature difference distribution is quantified by the KL algorithm and the target correlation features are selected by combining the mutual information method.
[0106] Using the simulation-test mapping as the loss constraint, a corresponding dual-branch network is constructed;
[0107] The target associated features are transferred into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients accordingly.
[0108] It should be noted that the first step involves extracting correlation features and quantifying differences: Initial correlation features are extracted between the baseline operating condition correction parameters (such as the turbulence model correction coefficients obtained from wind tunnel testing) and the aerodynamic feature thresholds of the dynamic operating condition cluster (such as the velocity and wind field thresholds of the "high-speed crosswind cluster"). These initial features include the preliminary rule that "the higher the velocity, the larger the correction coefficient." The Kullback-Leibler divergence is used to quantify the distribution of feature differences between the two. Specifically, the smaller the Kullback-Leibler divergence value, the closer the feature distributions of the baseline and dynamic operating conditions are, and the higher the feasibility of migration. If the divergence value is too large, the mutual information method is used to filter target correlation features (only retaining features with small differences and strong correlation, such as "velocity-correction coefficient"), and eliminating redundant features with large differences to ensure the reliability of the migration.
[0109] The second step is to construct a transfer learning network: using the "mapping relationship between simulation data and measured data" as the loss constraint (i.e., the corrected parameters after transfer must meet the requirement that "the simulation results are close to the actual measurements"), a two-branch network is built. Specifically, one branch inputs the corrected parameters and features of the baseline working condition, and the other branch inputs the features of the dynamic working condition cluster. The two networks achieve knowledge transfer by sharing the feature layer, and the loss constraint ensures that the parameters after transfer will not deviate from the actual test requirements.
[0110] The third step involves transferring and outputting the coefficients: The selected target-related features are input into a dual-branch network. The network transfers the correction parameters from the baseline operating condition to the dynamic operating condition cluster branch through a shared layer. The parameters are then adapted and adjusted based on the aerodynamic characteristic thresholds of the dynamic operating conditions. For example, the correction coefficient for the baseline operating condition "no wind" is 1.0, but when transferred to the dynamic operating condition "crosswind 5 m / s," it is adjusted to 1.2 based on the influence of crosswind on turbulence. Finally, the initial correction coefficients for each dynamic operating condition cluster are output. This process avoids testing the correction parameters individually for each dynamic operating condition, significantly reducing testing costs and facilitating subsequent processing.
[0111] Furthermore, the step of transferring the target correlation features into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients includes:
[0112] Deep information in the target association features is extracted by multi-scale dilated convolution, and a dynamic perturbation factor adapted to the dynamic working condition cluster is introduced to perform corresponding feature perturbation training in order to output the corresponding enhanced association features.
[0113] The enhanced correlation features are input into the dual-branch network, and the feature distribution of the dynamic working condition cluster is detected by the meta-learning strategy in order to perform constraint training and output the corresponding candidate correction coefficient set.
[0114] The candidate correction coefficient set is hierarchically verified to screen out effective coefficients, and Gaussian process regression is used to smooth and correct the effective coefficients to output the initial correction coefficients.
[0115] It should be noted that the first step is to strengthen the correlation features: multi-scale dilated convolution is used to extract deep information of the target correlation features. Specifically, dilated convolution can expand the receptive field without increasing the amount of computation, and capture the multi-scale correlation of "speed-wind field-drag" (such as the combined effect of short-term speed fluctuations and long-term wind field changes). Dynamic perturbation factors (simulating random changes in dynamic operating conditions, such as sudden crosswinds and vehicle vibrations) are introduced to perturb the features, making the features more robust and outputting enhanced correlation features. Specifically, these features not only contain the basic correlation rules, but can also adapt to the uncertainty of dynamic operating conditions.
[0116] The second step generates a candidate coefficient set: The enhanced correlation features are input into the dual-branch network, and a meta-learning strategy (MAML, Model Independent Meta-Learning) is used to quickly detect the feature distribution of dynamic working condition clusters. Specifically, meta-learning is good at quickly adapting to new scenarios. Even if there is very little data for a certain dynamic working condition, it can find a suitable feature distribution pattern through meta-learning. Based on this distribution, constraint training is performed (the constraint condition is "the coefficients must make the simulated drag within the feature threshold range"), and multiple candidate correction coefficient sets are output (such as 3 different sets of coefficients for "high-speed crosswind clusters").
[0117] The third step is screening and smoothing correction: The candidate coefficient set undergoes tiered verification. Specifically, the first layer verifies whether the coefficients conform to aerodynamic theory (e.g., drag is proportional to the square of velocity), and the second layer verifies whether the coefficients are suitable for environmental parameters such as altitude and temperature under this operating condition, eliminating invalid coefficients. For valid coefficients, Gaussian process regression is used for smoothing correction. Specifically, the Gaussian process can use a probabilistic model to capture the uncertainty of the coefficients, smoothing out abnormally fluctuating coefficient values (e.g., a coefficient suddenly increasing to 2.0, which clearly violates the rule, is smoothed down to 1.3), ultimately outputting stable and accurate initial correction coefficients for subsequent processing.
[0118] Furthermore, the step of using a hash index algorithm to quickly match the actually collected aerodynamic parameters with the initial corrected simulation data to calculate the corresponding target deviation value includes:
[0119] The actual aerodynamic parameters are analyzed and processed using a dual-channel attention mechanism to output the corresponding steady-state parameters and transient parameters. The temporal features contained in the steady-state parameters and the extreme value features contained in the transient parameters are extracted to create the corresponding structural feature vector.
[0120] The temporal trend of the structural feature vector is encoded into a dynamic hash fragment using the hash indexing algorithm, and aerodynamic theoretical constraints corresponding to the commercial vehicle are collected.
[0121] The target deviation value is generated based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment.
[0122] It should be noted that the first step is to analyze the measured aerodynamic parameters: a dual-channel attention mechanism (one channel focuses on steady-state parameters, and the other channel focuses on transient parameters) is used to analyze the actual collected aerodynamic parameters. Specifically, the steady-state parameters (such as the resistance during constant speed driving) are used to extract time-series features (such as the average and fluctuation of resistance within 10 minutes), and the transient parameters (such as the resistance during rapid acceleration) are used to extract extreme value features (such as the maximum resistance and the rate of change of resistance). The two types of features are then fused into a structural feature vector. Specifically, this vector contains both steady-state laws and transient abrupt changes, comprehensively reflecting the actual aerodynamic state.
[0123] The second step is to construct a fast matching index: the temporal trend of the structural feature vector is encoded into dynamic hash fragments using a hash index algorithm. Specifically, hash encoding can convert high-dimensional features into short codes, significantly improving the matching speed (completed in milliseconds), which meets the "real-time" requirements of commercial vehicle dynamic testing. At the same time, the aerodynamic theoretical constraints of commercial vehicles are collected (such as aerodynamic drag ≤ 0.5×ρ×v²×Cd×A, where ρ is air density, v is speed, Cd is drag coefficient, and A is frontal area). Specifically, the theoretical constraints set "physical boundaries" for deviation calculation to avoid matching results that violate scientific laws.
[0124] The third step is to calculate the target deviation value: Based on aerodynamic theory constraints, initial corrected simulation data that "conforms to physical laws" is selected. Then, through dynamic hashing, the simulation data segment most similar to the measured eigenvector is quickly matched, and the difference between the two is calculated (e.g., if the measured drag is 900N and the matched simulated drag is 850N, the difference is 50N). This difference is then weighted by the eigenvector weights (e.g., the weight of transient extrema is higher than that of the steady-state mean) to obtain the target deviation value. Specifically, this deviation value reflects both data differences and conforms to aerodynamic theory, providing a reliable basis for subsequent simulation corrections and facilitating subsequent processing.
[0125] Furthermore, the step of generating the target deviation value based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment includes:
[0126] Based on the aerodynamic characteristic threshold of the dynamic working condition cluster, the aerodynamic theoretical constraints are transformed into multi-dimensional constraint functions, and the dynamic hash fragments are substituted into the multi-dimensional constraint functions to filter out the effective feature intervals that meet the constraint conditions.
[0127] The long-term dependence features of the aerodynamic parameters are captured by a preset encoder to generate a corresponding correlation feature matrix;
[0128] The effective feature intervals are fused into the interior of the associated feature matrix to generate a corresponding target deviation matrix, and the target deviation matrix is converted into the target deviation value accordingly.
[0129] It should be noted that the first step involves constructing a multi-dimensional constraint function: based on the aerodynamic characteristic thresholds of the dynamic operating condition cluster (such as the speed of 80-110 km / h and the crosswind of 3-5 m / s for the "high-speed crosswind cluster"), the aerodynamic theoretical constraints (such as the drag formula) are converted into a multi-dimensional constraint function. Specifically, for example, the constraint function for this cluster is "when v∈[80,110], wind field∈[3,5], Cd∈[0.5,0.6] and drag∈[800,1000]". The dynamic hash fragment (measured feature encoding) is substituted into this function to filter out the effective feature intervals that satisfy all constraints (such as only retaining the measured and simulated data fragments of "drag 850-950N"). Specifically, abnormal data that violates the operating condition characteristics and theory are eliminated to ensure the reliability of the basic data for deviation calculation.
[0130] The second step is to capture long-term dependent features: Commercial vehicle aerodynamic parameters have long-term dependent features (such as long-term crosswinds leading to stable airflow over the vehicle body and gradual changes in drag). By using a preset encoder (such as an LSTM encoder), the long-term dependent features of the measured aerodynamic parameters (such as the drag change trend within 30 minutes) are captured, and a correlation feature matrix is generated. Specifically, this matrix reflects the temporal correlation of the parameters, avoiding the bias caused by using only instantaneous data to calculate deviations.
[0131] The third step is to generate the deviation value: The effective feature intervals (constrained reliable data) are fused into the correlation feature matrix, ensuring the matrix includes both long-term trends and focuses on effective data. Based on the fused matrix, a target deviation matrix is generated (rows represent time points, columns represent deviation types, such as drag deviation and lift deviation). Through matrix normalization (converting different types of deviations into 0-1 intervals) and weighted summation (with drag deviation having a higher weight than lift deviation), the deviation matrix is converted into a single target deviation value. Specifically, this value comprehensively reflects the overall difference between measured and simulated data, providing a precise quantitative basis for subsequent correction of the initial simulation data. Ultimately, this ensures that the target correction simulation data is highly consistent with the actual aerodynamic state, completing efficient and accurate dynamic aerodynamic testing of commercial vehicles. This facilitates subsequent processing.
[0132] Please see Figure 2 The third embodiment of the present invention provides:
[0133] A dynamic aerodynamic testing system for commercial vehicles, wherein the system comprises:
[0134] The segmentation module is used to divide the driving data of commercial vehicles throughout their entire life cycle into several dynamic operating condition clusters by using a preset clustering algorithm combined with aerodynamic drag sensitivity analysis, and extracting the aerodynamic feature thresholds of each dynamic operating condition cluster.
[0135] The construction module is used to construct the mapping relationship between simulation data and measured data. Based on the mapping relationship, the transfer learning algorithm is used to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients.
[0136] The adjustment module is used to pre-adjust the boundary conditions and turbulence model parameters of the preset simulation model according to the initial correction coefficient and the aerodynamic characteristic threshold, so as to generate the corresponding initial correction simulation data.
[0137] The matching module is used to quickly match the actually collected aerodynamic parameters with the initial correction simulation data using a hash index algorithm to calculate the corresponding target deviation value. Based on the target deviation value and the initial correction simulation data, the corresponding target correction simulation data is output to complete the aerodynamic test of the commercial vehicle.
[0138] Furthermore, the partitioning module is specifically used for:
[0139] Data on driving speed, acceleration, three-dimensional environmental wind field, aerodynamic drag, and vehicle accessory status generated by the commercial vehicle during the test are collected. Corresponding strong correlation combinations are identified through association rule mining algorithms to create corresponding aerodynamic sensitive feature sets.
[0140] The aerodynamic sensitive feature set is standardized to generate a corresponding standard feature set;
[0141] The standard feature set is clustered once using the FCM algorithm to generate the corresponding initial operating condition cluster. An appropriate dynamic drag gradient threshold is set for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster.
[0142] Furthermore, the partitioning module is specifically used for:
[0143] The core parameters of each initial operating condition cluster are extracted, and the sensitivity coefficient of each core parameter relative to the starting resistance is calculated using Sobol global sensitivity analysis.
[0144] Each of the aforementioned sensitivity coefficients is fitted using kernel density estimation to generate a corresponding drag distribution curve. The peak value of the drag distribution curve is set as the reference drag, and a dynamic threshold for the adapted working condition is generated by combining it with an altitude correction factor.
[0145] The initial operating condition cluster is substituted into the dynamic threshold, and dynamic verification is performed simultaneously to generate the corresponding dynamic operating condition cluster.
[0146] Furthermore, the adjustment module is specifically used for:
[0147] The initial correlation features between the correction parameters of the baseline working condition and the aerodynamic characteristic thresholds of the dynamic working condition cluster are extracted. The feature difference distribution is quantified by the KL algorithm and the target correlation features are selected by combining the mutual information method.
[0148] Using the simulation-test mapping as the loss constraint, a corresponding dual-branch network is constructed;
[0149] The target associated features are transferred into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients accordingly.
[0150] Furthermore, the adjustment module is specifically used for:
[0151] Deep information in the target association features is extracted by multi-scale dilated convolution, and a dynamic perturbation factor adapted to the dynamic working condition cluster is introduced to perform corresponding feature perturbation training in order to output the corresponding enhanced association features.
[0152] The enhanced correlation features are input into the dual-branch network, and the feature distribution of the dynamic working condition cluster is detected by the meta-learning strategy in order to perform constraint training and output the corresponding candidate correction coefficient set.
[0153] The candidate correction coefficient set is hierarchically verified to screen out effective coefficients, and Gaussian process regression is used to smooth and correct the effective coefficients to output the initial correction coefficients.
[0154] Furthermore, the matching module is specifically used for:
[0155] The actual aerodynamic parameters are analyzed and processed using a dual-channel attention mechanism to output the corresponding steady-state parameters and transient parameters. The temporal features contained in the steady-state parameters and the extreme value features contained in the transient parameters are extracted to create the corresponding structural feature vector.
[0156] The temporal trend of the structural feature vector is encoded into a dynamic hash fragment using the hash indexing algorithm, and aerodynamic theoretical constraints corresponding to the commercial vehicle are collected.
[0157] The target deviation value is generated based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment.
[0158] Furthermore, the matching module is specifically used for:
[0159] Based on the aerodynamic characteristic threshold of the dynamic working condition cluster, the aerodynamic theoretical constraints are transformed into multi-dimensional constraint functions, and the dynamic hash fragments are substituted into the multi-dimensional constraint functions to filter out the effective feature intervals that meet the constraint conditions.
[0160] The long-term dependence features of the aerodynamic parameters are captured by a preset encoder to generate a corresponding correlation feature matrix;
[0161] The effective feature intervals are fused into the interior of the associated feature matrix to generate a corresponding target deviation matrix, and the target deviation matrix is converted into the target deviation value accordingly.
[0162] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the commercial vehicle dynamic aerodynamic testing method as described above.
[0163] The fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the commercial vehicle dynamic aerodynamic testing method as described above.
[0164] In summary, the commercial vehicle dynamic aerodynamic testing method and system provided in the above embodiments of the present invention can accurately detect the data deviations generated during the testing process of commercial vehicles, thereby significantly improving the testing efficiency of commercial vehicles.
[0165] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.
[0166] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0167] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0168] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0169] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0170] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.
Claims
1. A method for dynamic aerodynamic testing of commercial vehicles, characterized in that, The method includes: Based on the driving data of commercial vehicles throughout their entire life cycle, a preset clustering algorithm is used in conjunction with aerodynamic drag sensitivity analysis to divide the vehicle into several dynamic operating condition clusters, and the aerodynamic feature thresholds of each dynamic operating condition cluster are extracted. A mapping relationship between simulation data and measured data is constructed. Based on the mapping relationship, a transfer learning algorithm is used to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients. The boundary conditions and turbulence model parameters of the preset simulation model are pre-adjusted according to the initial correction coefficient and the aerodynamic characteristic threshold to generate the corresponding initial correction simulation data. A hash index algorithm is used to quickly match the actual collected aerodynamic parameters with the initial correction simulation data to calculate the corresponding target deviation value. Based on the target deviation value and the initial correction simulation data, the corresponding target correction simulation data is output to complete the aerodynamic test of the commercial vehicle. The steps for dividing the driving data based on the entire life cycle of commercial vehicles into several dynamic operating condition clusters using a pre-defined clustering algorithm combined with aerodynamic drag sensitivity analysis include: Data on driving speed, acceleration, three-dimensional environmental wind field, aerodynamic drag, and vehicle accessory status generated by the commercial vehicle during the test are collected. Corresponding strong correlation combinations are identified through association rule mining algorithms to create corresponding aerodynamic sensitive feature sets. The aerodynamic sensitive feature set is standardized to generate a corresponding standard feature set; The standard feature set is clustered once using the FCM algorithm to generate the corresponding initial operating condition cluster. An appropriate dynamic drag gradient threshold is set for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster. The step of using a transfer learning algorithm based on the mapping relationship to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients includes: The initial correlation features between the correction parameters of the baseline working condition and the aerodynamic characteristic thresholds of the dynamic working condition cluster are extracted. The feature difference distribution is quantified by the KL algorithm and the target correlation features are selected by combining the mutual information method. Using the simulation-test mapping as the loss constraint, a corresponding dual-branch network is constructed; The target associated features are transferred into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients accordingly. The step of using a hash index algorithm to quickly match the actually collected aerodynamic parameters with the initial corrected simulation data to calculate the corresponding target deviation value includes: The actual collected aerodynamic parameters are analyzed and processed using a dual-channel attention mechanism to output the corresponding steady-state parameters and transient parameters. The temporal features contained in the steady-state parameters and the extreme value features contained in the transient parameters are extracted to create the corresponding structural feature vector. The temporal trend of the structural feature vector is encoded into a dynamic hash fragment using the hash indexing algorithm, and aerodynamic theoretical constraints corresponding to the commercial vehicle are collected. The target deviation value is generated based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment.
2. The commercial vehicle dynamic aerodynamic testing method according to claim 1, characterized in that, The step of setting an appropriate dynamic drag gradient threshold for the initial operating condition cluster through aerodynamic drag sensitivity analysis to perform secondary clustering and generate the dynamic operating condition cluster includes: The core parameters of each initial operating condition cluster are extracted, and the sensitivity coefficient of each core parameter relative to the starting resistance is calculated using Sobol global sensitivity analysis. Each of the aforementioned sensitivity coefficients is fitted using kernel density estimation to generate a corresponding drag distribution curve. The peak value of the drag distribution curve is set as the reference drag, and a dynamic threshold for the adapted working condition is generated by combining it with an altitude correction factor. The initial operating condition cluster is substituted into the dynamic threshold, and dynamic verification is performed simultaneously to generate the corresponding dynamic operating condition cluster.
3. The dynamic aerodynamic testing method for commercial vehicles according to claim 1, characterized in that, The step of transferring the target associated features into the dynamic operating condition cluster through the dual-branch network to output the initial correction coefficients includes: Deep information in the target association features is extracted by multi-scale dilated convolution, and a dynamic perturbation factor adapted to the dynamic working condition cluster is introduced to perform corresponding feature perturbation training in order to output the corresponding enhanced association features. The enhanced correlation features are input into the dual-branch network, and the feature distribution of the dynamic working condition cluster is detected by the meta-learning strategy in order to perform constraint training and output the corresponding candidate correction coefficient set. The candidate correction coefficient set is hierarchically verified to screen out effective coefficients, and Gaussian process regression is used to smooth and correct the effective coefficients to output the initial correction coefficients.
4. The dynamic aerodynamic testing method for commercial vehicles according to claim 1, characterized in that, The step of generating the target deviation value based on the aerodynamic theoretical constraints and the corresponding dynamic hash fragment includes: Based on the aerodynamic characteristic threshold of the dynamic working condition cluster, the aerodynamic theoretical constraints are transformed into multi-dimensional constraint functions, and the dynamic hash fragments are substituted into the multi-dimensional constraint functions to filter out the effective feature intervals that meet the constraint conditions. The long-term dependence features of the aerodynamic parameters are captured by a preset encoder to generate a corresponding correlation feature matrix; The effective feature intervals are fused into the interior of the associated feature matrix to generate the corresponding target deviation matrix, and the target deviation matrix is converted into the target deviation value accordingly.
5. A dynamic aerodynamic testing system for commercial vehicles, characterized in that, The system for implementing the commercial vehicle dynamic aerodynamic testing method as described in any one of claims 1 to 4 includes: The segmentation module is used to divide the driving data of commercial vehicles throughout their entire life cycle into several dynamic operating condition clusters by using a preset clustering algorithm combined with aerodynamic drag sensitivity analysis, and extracting the aerodynamic feature thresholds of each dynamic operating condition cluster. The construction module is used to construct the mapping relationship between simulation data and measured data. Based on the mapping relationship, the transfer learning algorithm is used to transfer the correction parameters of the baseline working condition to each of the dynamic working condition clusters to generate the corresponding initial correction coefficients. The adjustment module is used to pre-adjust the boundary conditions and turbulence model parameters of the preset simulation model according to the initial correction coefficient and the aerodynamic characteristic threshold, so as to generate the corresponding initial correction simulation data. The matching module is used to quickly match the actually collected aerodynamic parameters with the initial correction simulation data using a hash index algorithm to calculate the corresponding target deviation value. Based on the target deviation value and the initial correction simulation data, the corresponding target correction simulation data is output to complete the aerodynamic test of the commercial vehicle.
6. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the commercial vehicle dynamic aerodynamic testing method as described in any one of claims 1 to 4.
7. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the dynamic aerodynamic testing method for commercial vehicles as described in any one of claims 1 to 4.