CN114792069A

CN114792069A - Modeling method of vehicle dynamics model

Info

Publication number: CN114792069A
Application number: CN202210450494.8A
Authority: CN
Inventors: 柴嘉峰; 梁元波; 洪志福
Original assignee: Hozon New Energy Automobile Co Ltd
Current assignee: Hozon New Energy Automobile Co Ltd
Priority date: 2022-04-26
Filing date: 2022-04-26
Publication date: 2022-07-26

Abstract

The present invention provides a modeling method for a vehicle dynamics model, including establishing a neural network dynamics model, using first drive test data to train the neural network dynamics model; analyzing the first drive test data to at least determining a special usage scenario not corresponding to the first drive test data; acquiring second drive test data corresponding to the special usage scenario, and establishing a first physical model corresponding to the special usage scenario based on the second drive test data; Acquire theoretical drive test data based on the first physical model; and use the theoretical drive test data to train the neural network dynamics model. The present invention combines the neural network and the physical model, and can obtain a credible and stable data model according to the physical laws while taking into account the good consistency with the real vehicle data and the special usage scenarios with limited data amount.

Description

Modeling method of vehicle dynamics model

Technical Field

The invention relates to the field of automobiles, in particular to a modeling method of a vehicle dynamics model.

Background

Most vehicle dynamics modeling in the current market is a physical model-based modeling method, a large number of professional tests are needed to obtain accurate dynamics models, and accurate modeling is needed for all subsystems. To obtain a high fidelity kinetic model, it is often necessary to rely on carsim, carmaker, etc., which are expensive commercial software, and on test data from expensive test fields. In recent years, with the rapid development of the automatic driving industry, the data-driven iterative mode increasingly highlights the advantages of the automatic driving industry, the research and development of the automatic driving often need to run millions of kilometers in simulation, large-scale concurrent simulation is a necessary trend, large-scale dynamic models are necessarily required to run simultaneously, the expensive price of commercial software makes many companies to be forbidden, and a dynamic model which gives consideration to both price and simulation precision is urgently needed.

With the rapid development and landing of deep learning in these years, many companies use a neural network method for dynamic modeling based on a large amount of drive test data, train a dynamic model by using a large amount of data and strong calculation power, and greatly reduce the cost. However, there is a risk that the established dynamic model lacks interpretability and the scene that data cannot cover is not reliable, and vehicle aging, wear such as steering wheel zero offset, clearance, tire wear, environmental changes such as road surface wet and slippery, etc. cannot be well simulated.

Therefore, there is a need for a modeling method of a vehicle dynamics model that solves the problems of the prior art.

Disclosure of Invention

The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

In order to better solve the above problems, the present invention provides a modeling method of a vehicle dynamics model, comprising: establishing a neural network dynamic model, and training the neural network dynamic model by using first path measurement data; analyzing the first path measurement data to at least determine a special use scene which does not correspond to the first path measurement data; acquiring second road measurement data corresponding to the special use scene, and establishing a first physical model corresponding to the special use scene based on the second road measurement data; theoretical drive test data are obtained based on the first physical model; training the neural network dynamics model by using the theoretical drive test data; wherein at least the trained neural network dynamics model is used as the vehicle dynamics model.

In an embodiment of the modeling method, optionally, analyzing the first path of test data is further used to determine a special working condition that does not correspond to the first path of test data; wherein the modeling method further comprises: acquiring third path measurement data corresponding to the special working condition, and establishing a second physical model corresponding to the special working condition based on the third path measurement data; wherein the second physical model and the trained neural network dynamics model are used as the vehicle dynamics model.

In an embodiment of the modeling method, optionally, the modeling method further includes: verifying the second physical model to judge whether the second physical model meets vehicle dynamics in a normal working condition except the special working condition; wherein the vehicle dynamics model is a verified second physical model.

In an embodiment of the modeling method, optionally, in response to a condition corresponding to a normal operating condition, the vehicle dynamics model is based on the trained neural network dynamics model; and responding to the special working condition, wherein the vehicle dynamic model is based on the second physical model.

In an embodiment of the modeling method, optionally, the input of the neural network dynamic model, and/or the input of the first physical model, and/or the input of the second physical model at least include vehicle state parameters and road state parameters.

In an embodiment of the modeling method, optionally, the vehicle state parameter includes at least a steering wheel zero offset parameter and/or a tire aging parameter.

In an embodiment of the modeling method, optionally, the obtaining theoretical drive test data based on the first physical model further includes: and applying at least one excitation signal of frequency sweep, step and triangular wave to the first physical model to acquire the theoretical drive test data.

In an embodiment of the modeling method, optionally, the modeling method further includes: verifying the first physical model to judge whether the first physical model meets vehicle dynamics in a common use scene except the special use scene; wherein the theoretical drive test data is obtained based on the validated first physical model.

In an embodiment of the modeling method, optionally, the modeling method further includes: verifying the trained neural network dynamic model to judge whether the trained neural network dynamic model conforms to the amplitude-frequency characteristic of the vehicle; wherein at least the neural network dynamics model after the training passing the verification is used as the vehicle dynamics model.

Another aspect of the present invention also provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the steps of the method for modeling a vehicle dynamics model as described in any one of the embodiments of the present invention.

The method combines the neural network with physical (mathematical) modeling, takes the training of the neural network as a basis, utilizes the physical (mathematical) modeling to integrate the objective rule of a vehicle physical model into the neural network training data as theoretical guidance, and ensures that the neural network training model can obtain a credible and stable data model according to the physical rule under the extreme condition of limited training samples while considering good consistency with real vehicle data.

Drawings

The above features and advantages of the present disclosure will be better understood upon reading the detailed description of embodiments of the disclosure in conjunction with the following drawings. In the drawings, components are not necessarily drawn to scale, and components having similar relative characteristics or features may have the same or similar reference numerals.

FIG. 1 illustrates a flow diagram of an embodiment of a modeling method provided by an aspect of the present invention.

FIG. 2 illustrates a flow diagram and a diagram of a neural network architecture of another embodiment of a modeling method provided by an aspect of the present invention.

Detailed Description

The invention is described in detail below with reference to the figures and specific embodiments. It is noted that the aspects described below in connection with the figures and the specific embodiments are only illustrative and should not be construed as imposing any limitation on the scope of the present invention.

The following description is presented to enable any person skilled in the art to make and use the invention, and is incorporated in the context of a particular application. Various modifications, as well as various uses in different applications will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to a wide range of embodiments. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

In the following detailed description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. It will be apparent, however, to one skilled in the art that the invention may be practiced without necessarily being limited to these specific details. In other instances, well-known structures and devices are shown in block diagram form, rather than in detail, in order to avoid obscuring the present invention.

The reader's attention is directed to all papers and documents which are filed concurrently with this specification and which are open to public inspection with this specification, and the contents of all such papers and documents are incorporated herein by reference. All the features disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise. Thus, unless expressly stated otherwise, each feature disclosed is one example only of a generic series of equivalent or similar features.

Note that where used, the designations left, right, front, back, top, bottom, positive, negative, clockwise, and counterclockwise are used merely for convenience and do not imply any particular fixed orientation. In fact, they are used to reflect the relative position and/or orientation between various parts of the object. Furthermore, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

It is noted that, where used, further, preferably, still further and more preferably is a brief introduction to the exposition of the alternative embodiment on the basis of the preceding embodiment, the contents of the further, preferably, still further or more preferably back band being combined with the preceding embodiment as a complete constituent of the alternative embodiment. Several further, preferred, still further or more preferred arrangements of the belt after the same embodiment may be combined in any combination to form a further embodiment.

As described above, in order to solve various problems in the prior art, the present invention provides a modeling method of a vehicle dynamics model, and referring to fig. 1, the modeling method provided by the present invention comprises:

step S100: establishing a neural network dynamic model, and training the neural network dynamic model by using first path measurement data;

step S200: analyzing the first path measurement data to at least determine a special use scene which does not correspond to the first path measurement data;

step S300: acquiring second road measurement data corresponding to the special use scene, and establishing a first physical model corresponding to the special use scene based on the second road measurement data;

step S400: acquiring theoretical drive test data based on the first physical model; and

step S500: training the neural network dynamics model using the theoretical drive test data.

Wherein at least the trained neural network dynamic model is used as the vehicle dynamic model.

Please further refer to fig. 2 to understand the specific structure of the neural network dynamics model. First in step S100, a neural network structure is established that includes an input, a neural network substructure, an intermediate layer signal, a neural network substructure, and an output signal.

The neural network sub-structures described above correspond to various systems of the vehicle, such as EPS, brakes, drives, body modules, and the like. It should be noted that those skilled in the art can adjust the structure of the neural network according to actual needs.

In a preferred embodiment, the inputs to the neural network comprise at least a vehicle state parameter and a road state parameter. Wherein the vehicle state parameters comprise at least a steering wheel zero offset parameter and/or a tire aging parameter. The road condition parameters include at least the wet condition, the flatness, and the like of the road surface.

When the neural network dynamics model is designed, the vehicle state parameters and/or the road state parameters are used as input, so that the dynamics model can be better close to the real vehicle to run, and a credible, stable and widely applicable data model can be obtained.

Furthermore, the first road test data refers to data which is sampled and obtained based on a conventional method and is derived from a real road test, and noise characteristics can be well simulated. However, due to the limitation of sampling, the sampled data generally conforms to normal distribution, and the first path of test data cannot represent all use scenes and all working conditions. That is, even in the normal operation, there is a problem that the data amount of the sampled drive test data corresponding to a partial physical scene is small. Similarly, the first path measurement data has no way of covering all working conditions, especially some unstable and dangerous working conditions of the vehicle.

Therefore, after the neural network dynamics model is trained by using the first road test data, the obtained vehicle dynamics model can better correspond to most of conventional use scenes under normal working conditions, but for some special use scenes and special working conditions, the reliability of the neural network dynamics model obtained by training the first road test data is greatly reduced.

Therefore, in the present invention, step S200 is executed to analyze the first route measurement data to determine at least special usage scenarios, such as emergency obstacle avoidance, wet and slippery road surface, which do not correspond to the first route measurement data, and which have a very low probability of appearing in the first route measurement data compared to normal high-speed driving, and which cannot be characterized by the first route measurement data. Since the first path measurement data has no way of characterizing the special usage scenarios, step S300 needs to be executed to obtain second path measurement data corresponding to the special usage scenarios, and establish a first physical model corresponding to the special usage scenarios based on the second path measurement data.

In the step S200, since the scenes covered by the data of the conventional drive test data (the first drive test data) are limited, the scenes covered by the conventional drive test data (the first drive test data) can be obtained through analysis, and then, according to the scene set to be covered, which scenes are rarely covered by the conventional drive test can be divided from the scene set. Meanwhile, it can be understood that the fact that the first route test data is the corresponding special usage scenario does not mean that the first route test data does not exist in the conventional route test data (the first route test data), and the conventional route test data may also have a corresponding small amount of data.

The second path measurement data is also derived from the real path measurement, but the second path measurement data makes up the partial data which is not covered by the first path measurement data in a targeted manner. And after second path measurement data are obtained, establishing a first physical model corresponding to the special use scene based on the second path measurement data. Different from the training of a neural network, a large amount of data is needed, and the establishment of the physical model can be based on a small amount of experimental test data, so that the actual measurement cost of a special use scene can be reduced.

In a preferred embodiment, the inputs of the first physical model comprise at least vehicle state parameters and road state parameters, wherein the vehicle state parameters comprise at least steering wheel zero deflection parameters and/or tire aging parameters. The road condition parameters include at least the wet condition, the flatness, and the like of the road surface. By taking the vehicle state parameters and/or the road state parameters as input, the dynamic model can be better close to the real vehicle to operate, so that a credible, stable and widely-applicable data model can be obtained.

It should be understood that, in order to ensure the smooth proceeding of the subsequent steps, before step S400, the first physical model needs to be verified to determine whether the first physical model meets the amplitude-frequency characteristic of the vehicle and can satisfy vehicle dynamics in all scenarios, that is, although the first physical model is established based on the second measurement data corresponding to the special usage scenario, the first physical model needs to maintain the vehicle dynamics in all scenarios.

In response to being able to pass the verification, step S400 is performed to acquire theoretical drive test data based on the verified first physical model. In this step, the theoretical drive test data may be acquired by applying at least one excitation signal of frequency sweep, step, or triangle wave to the first physical model.

Although the first physical model is established only by a small amount of second test data, the amount of the second test data is not enough to train the neural network dynamic model. In step S400, a large amount of theoretical drive test data can be obtained by exciting the verified first physical model, and the part of theoretical drive test data with better reliability obtained based on the second drive test data — the first physical model can be used as a data source corresponding to a special use scenario, so as to train a neural network dynamic model, thereby reducing the cost for obtaining drive test data corresponding to the special use scenario while ensuring the reliability of the data.

And for a neural network dynamic model obtained after training based on the first path measurement data and the theoretical path measurement data, frequency sweeping, step input and the like are also required to be verified, and whether the amplitude-frequency characteristic of the vehicle is met or not is verified. So as to ensure the reliability and stability of the neural network dynamic model.

Thus, a specific implementation of an embodiment of the modeling method of a vehicle dynamics model provided by the present invention has been described. In the embodiment, the first physical model is established by acquiring the second road test data, the theoretical road test data is acquired through the first physical model, and the neural network dynamics model can be comprehensively trained based on the first road test data and the theoretical road test data, so that the neural network dynamics model can cover almost all use scenes under normal working conditions. The defect of insufficient samples is made up, so that the trained model conforms to the physical rule of the vehicle, and meanwhile, the fitting precision of real vehicle data is considered.

In another aspect of the invention, another embodiment of a method for modeling a vehicle dynamics model is also provided. In this embodiment, analyzing the first path measurement data is further used to determine a special operating condition (e.g., unstable vehicle, dangerous operating condition, etc.) to which the first path measurement data does not correspond.

In this embodiment, the modeling method further includes obtaining third measurement data corresponding to the special operating condition, establishing a second physical model corresponding to the special operating condition based on the third measurement data, and using the second physical model and the trained neural network dynamics model as the vehicle dynamics model.

The third path measurement data is also derived from the real path measurement, but the partial data which is not covered by the first path measurement data is compensated in a targeted manner. And after third path measurement data are obtained, establishing a second physical model corresponding to the special working condition based on the third path measurement data. Different from the training of a neural network, a large amount of data is needed, and the establishment of the physical model can be based on a small amount of experimental test data, so that the actual measurement cost of special working conditions can be reduced.

In a preferred embodiment, the inputs of the second physical model comprise at least vehicle state parameters and road state parameters, wherein the vehicle state parameters comprise at least steering wheel zero deflection parameters and/or tire aging parameters. The road condition parameters include at least the wet condition, the flatness, and the like of the road surface. By taking the vehicle state parameters and/or the road state parameters as input, the dynamic model can be better close to the real vehicle to run, so that a credible, stable and widely applicable data model can be obtained.

It should be understood that, in order to ensure the stability and reliability of the model, the second physical model needs to be verified to determine whether the second physical model meets the amplitude-frequency characteristics of the vehicle and can satisfy vehicle dynamics in all the operating conditions, that is, although the second physical model is established based on the third measured data corresponding to a special operating condition, the second physical model needs to maintain the vehicle dynamics in all the operating conditions.

After obtaining the second physical model, in an embodiment, at least one excitation signal of frequency sweep, step, and triangular wave may also be applied to the verified second physical model to obtain theoretical drive test data corresponding to a specific working condition. And training a neural network dynamics model by using theoretical drive test data corresponding to the special working conditions.

However, due to the nonlinearity of the special working condition, the reliability of the result obtained by the neural network dynamic model is not as good as that of the second physical model for the special working condition. Therefore, as a preferred embodiment, the second physical model and the trained neural network dynamic model are used as the vehicle dynamic model.

Further, in response to the response corresponding to the normal operating condition, the vehicle dynamics model is based on the trained neural network dynamics model, that is, for the normal operating condition, the weight of the result obtained by the trained neural network dynamics model is larger. In response to the response corresponding to the special operating condition, the vehicle dynamics model is based on the second physical model, that is, the results obtained by the second physical model are weighted more heavily for the special operating condition.

In this embodiment, the final vehicle dynamics model can better conform to the physical laws of the vehicle based on the second physical model, and the fitting accuracy of the real vehicle data can be considered based on the neural network dynamics model trained by the first physical model.

The method utilizes the fitting capability of the neural network on the nonlinear characteristic to fit a large amount of drive test data, and the data is from real drive test, so the noise characteristic can be well simulated, but the coverage of the drive test data only basically conforms to normal distribution relative to all working conditions, namely the data of most working conditions in the stable working state of the vehicle is covered. In some special use scenes and special working conditions, for example, data of unstable vehicles and even dangerous working conditions are few, so that the trained model is over-fitted to the normal working conditions, but is under-fitted to the unstable vehicles and even the dangerous working conditions, the trained model is very unlikely to conform to the physical characteristics of the vehicles, for example, the dynamic model trained by the neural network is swept, the law of the amplitude-frequency characteristic of the approximate frequency is not in accordance with the characteristics of the real vehicles, and the amplitude-frequency characteristic represents the key dynamic characteristics and laws. In order to better compensate the defect, a small amount of experimental tests of unstable and dangerous working conditions are needed to establish a physical (mathematical) model to compensate the defect of sample shortage. And then, the physical characteristics of the physical (mathematical) model are utilized to further guide and verify the neural network, so that the trained model conforms to the physical rule of the vehicle, and meanwhile, the fitting precision of the real vehicle data is considered. A specific implementation of a specific embodiment of the present invention is shown in fig. 2. In summary, the method comprises the following steps:

1. setting a neural network dynamics model, in particular the individual subsystems (neural network substructures) of the vehicle;

2. training a neural network by using the existing drive test data (normal drive test data);

3. according to the using scene of the vehicle, carrying out real-vehicle test on the working conditions with few samples and unstable vehicle and dangerous working conditions, and acquiring data to ensure that the scene covered by the samples is enough (such as a physical scene which is split and controlled to cover according to product definition in figure 2; analyzing the covering probability of different scenes, and combing out the scene lacking in normal drive test and the special working condition scene with unstable vehicle; designing a test method and acquiring data by a test field);

4. carrying out physical (mathematical) modeling on unstable and dangerous working conditions of the vehicle, and simultaneously verifying whether the stability and the physical characteristics of the physical model meet the vehicle dynamics (establishing a nonlinear model for optimizing the vehicle model and analyzing whether the dynamic characteristics meet the physical characteristics) under all the working conditions;

5. considering the influence of vehicle parameter change such as steering wheel zero offset and tire aging, expanding the physical model, and adding a vehicle parameter interface as input to the neural network (the input signal in the neural network structure comprises an aging parameter interface, a road environment interface and the like);

6. the physical (mathematical) model and the road test data are fused together, the data model (generating a theoretical data set adaptive to the neural network training and fusing the road test data) is used for guiding the neural network training, the reliability of the road test data under normal working conditions is high, and the reliability of the mathematical model under unstable and dangerous working conditions is high;

7. verifying the dynamics of neural network training, performing frequency sweep, step input and the like on the trained neural network dynamics model, verifying whether the neural network dynamics model meets the amplitude-frequency characteristics of the vehicle, finishing if the neural network dynamics model meets the amplitude-frequency characteristics, adjusting 6 if the neural network dynamics model does not meet the amplitude-frequency characteristics, and further optimizing the verification;

8. with the increase of test data, the mathematical model and the neural network can be further iteratively optimized to further approach the real vehicle characteristics.

The method has the advantages that the advantages of the neural network and the physical model are combined skillfully, the physical model can represent the physical characteristic rule to supplement the neural network data sample and guide the neural network training, the training data can be well fitted, and the amplitude-frequency characteristic rule of the vehicle can be met.

The dynamic model of the large-scale concurrent simulation is supported, the cost is low, the iteration efficiency is high, the model fidelity is high, the defect that the model interpretability of deep learning training is poor is overcome, the model has strong transplantable characteristics, the model is trained stably, and after a vehicle model is replaced, the whole modeling process can be converged more quickly and more stably due to the fact that the existing neural network conforms to the physical law, and the dynamic model of the vehicle can be obtained more quickly.

Another aspect of the present invention further provides a computer-readable storage medium, which stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the modeling method for a vehicle dynamics model as described in any one of the above embodiments, which are not repeated herein.

The various illustrative logical modules, and circuits described in connection with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal.

In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a web site, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk (disk) and disc (disc), as used herein, includes Compact Disc (CD), laser disc, optical disc, Digital Versatile Disc (DVD), floppy disk and blu-ray disc where disks (disks) usually reproduce data magnetically, while discs (discs) reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

The previous description is provided to enable any person skilled in the art to practice the various aspects described herein. It is to be understood that the scope of the invention is to be defined by the appended claims and not by the specific constructions and components of the embodiments illustrated above. Those skilled in the art can make various changes and modifications to the embodiments within the spirit and scope of the present invention, and such changes and modifications also fall within the scope of the present invention.

Claims

1. a modeling method of vehicle dynamics model, is characterized in that, comprises:

establishing a neural network dynamics model, and using the first drive test data to train the neural network dynamics model;

analyzing the first drive test data to at least determine a special usage scenario not corresponding to the first drive test data;

acquiring second drive test data corresponding to the special usage scenario, and establishing a first physical model corresponding to the special usage scenario based on the second drive test data;

Obtaining theoretical drive test data based on the first physical model; and

using the theoretical drive test data to train the neural network dynamics model; wherein

At least the trained neural network dynamics model is used as the vehicle dynamics model.

2 . The modeling method according to claim 1 , wherein analyzing the first drive test data is also used to determine special operating conditions not corresponding to the first drive test data; wherein

The modeling method also includes:

acquiring third drive test data corresponding to the special working condition, and establishing a second physical model corresponding to the special working condition based on the third drive test data; wherein

The second physical model and the trained neural network dynamics model are used as the vehicle dynamics model.

3. The modeling method according to claim 2, wherein the modeling method further comprises:

Verifying the second physical model to determine whether the second physical model satisfies vehicle dynamics in normal operating conditions other than the special operating conditions; wherein

As the vehicle dynamics model is a verified second physical model.

4. The modeling method according to claim 2, wherein, in response to a normal operating condition, the vehicle dynamics model is based on the trained neural network dynamics model;

In response to corresponding to the special operating condition, the vehicle dynamics model is dominated by the second physical model.

5. The modeling method according to any one of claims 1-4, wherein the neural network dynamics model, and/or the first physical model, and/or the second The input of the physical model includes at least vehicle state parameters and road state parameters.

6 . The modeling method according to claim 5 , wherein the vehicle state parameters at least include steering wheel bias parameters and/or tire aging parameters. 7 .

7. The modeling method according to claim 1, wherein the obtaining theoretical drive test data based on the first physical model further comprises:

At least one excitation signal of sweep frequency, step and triangular wave is applied to the first physical model to obtain the theoretical drive test data.

8. The modeling method according to claim 1, wherein the modeling method further comprises:

Verifying the first physical model to determine whether the first physical model satisfies vehicle dynamics in a normal usage scenario other than the special usage scenario; wherein

The theoretical drive test data is acquired based on the verified first physical model.

9. The modeling method of claim 1, wherein the modeling method further comprises:

Verifying the trained neural network dynamics model to determine whether the trained neural network dynamics model conforms to the amplitude-frequency characteristics of the vehicle; wherein

At least the verified trained neural network dynamics model is used as the vehicle dynamics model.

10. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the modeling method of the vehicle dynamics model according to any one of claims 1-9 is implemented A step of.