Vehicle kinematic parameter identification model training method, identification method and vehicle

Abstract

The application relates to the technical field of vehicles, in particular to a training method and an identification method of a vehicle kinematics parameter identification model and a vehicle, which comprises the following steps: acquiring a running parameter time sequence of a vehicle; determining a hidden state parameter estimated value time sequence of the vehicle based on the running parameter time sequence and inherent kinematics parameters of the vehicle; determining a parameter identification value of a to-be-identified kinematics parameter of the vehicle based on the running parameter time sequence, the hidden state parameter estimated value time sequence and an initial vehicle kinematics parameter identification model; determining a state parameter prediction value time sequence of the vehicle based on the hidden state parameter estimated value time sequence, the parameter identification value of the to-be-identified kinematics parameter, a dynamics equivalent model of the vehicle and a tire equivalent model; and optimizing the initial vehicle kinematics parameter identification model based on a first state prediction deviation of a state parameter true value time sequence and the state parameter prediction value time sequence, so as to obtain the vehicle kinematics parameter identification model.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, specifically to a training method, a recognition method, and a vehicle for a vehicle kinematic parameter recognition model. Background Technology

[0002] Vehicle kinematic parameters, such as the distance from the center of gravity to the front axle, yaw angle, moment of inertia, drag coefficient, rolling resistance coefficient, tire stiffness, and road adhesion coefficient, are fundamental inputs for vehicle dynamics control, state estimation, and autonomous driving decisions. Related technologies primarily acquire these parameters through: theoretical calculations based on vehicle factory design parameters, offline calibration on dedicated test benches, and data fitting based on real-vehicle road tests.

[0003] In related technologies, state parameters and observation parameters of the vehicle during actual operation are collected, and the collected data are input into a pre-built regression model or neural network model, directly outputting the estimated values ​​of the kinematic parameters to be identified. During the training phase, the model uses the deviation between the estimated parameter values ​​and the manually calibrated true parameter values ​​as the loss function, and optimizes the model parameters through backpropagation. However, the above-mentioned related technical solutions only use the deviation between the identified parameter values ​​and the calibrated true values ​​as the optimization target during the training process. Although the parameter identification values ​​output by the model are numerically close to the calibrated true values, after substituting the identified parameters into the vehicle dynamics model, the vehicle state response predicted by the model may have a large deviation from the actual state response, making it difficult to directly use the identified parameter values ​​in vehicle dynamics control. Summary of the Invention

[0004] This application provides a training method, an identification method, and a vehicle for a vehicle kinematic parameter identification model, in order to at least solve the problem in related technologies where the optimization target for training a vehicle kinematic parameter identification model is only the deviation between the identified parameter values ​​and the calibration true values, resulting in the identified parameters being difficult to use directly in vehicle dynamics control.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, this application provides a training method for a vehicle kinematic parameter identification model, comprising: acquiring a time series sequence of vehicle operating parameters; wherein the time series sequence of operating parameters includes a time-aligned time series sequence of true values ​​of state parameters and a time series sequence of true values ​​of observed parameters; determining a time series sequence of estimated hidden state parameters of the vehicle based on the time series sequence of operating parameters and the inherent kinematic parameters of the vehicle; determining the parameter identification values ​​of the kinematic parameters to be identified of the vehicle based on the time series sequence of operating parameters, the time series sequence of estimated hidden state parameters, and an initial vehicle kinematic parameter identification model; determining a time series sequence of predicted state parameters of the vehicle based on the time series sequence of estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and a tire equivalent model; the tire equivalent model is coupled with the vehicle's dynamic equivalent model; and optimizing the initial vehicle kinematic parameter identification model based on a first state prediction deviation between the time series sequence of true state parameters and the time series sequence of predicted state parameters to obtain the vehicle kinematic parameter identification model.

[0006] Based on the aforementioned technical means, by acquiring time-aligned time series sequences of true state parameter values ​​and observation parameter values, and determining time series sequences of estimated hidden state parameter values ​​based on these time series sequences of operating parameters and the vehicle's inherent kinematic parameters, the model can obtain key intermediate state parameters that are difficult to measure directly during vehicle operation, providing richer physical information for subsequent parameter identification. Furthermore, the time series sequences of operating parameters and estimated hidden state parameters are processed using an initial vehicle kinematic parameter identification model to obtain the parameter identification values ​​of the kinematic parameters to be identified. On this basis, the parameter identification values ​​are substituted into a tire equivalent model coupled with a dynamic equivalent model, and combined with the hidden state parameters... By using a numerical estimation time series and a dynamic equivalent model, the predicted time series of vehicle state parameters is determined, enabling the parameter identification values ​​to predict the corresponding vehicle state response after forward derivation from the vehicle dynamics model. Finally, using the first state prediction deviation between the predicted time series of state parameters and the actual time series of state parameters as the optimization objective, a vehicle kinematic parameter identification model is trained. This ensures that the parameter identification values ​​output by the trained model are not only numerically accurate, but also that the vehicle state response predicted by the model remains physically consistent with the actual state response after the parameter identification values ​​are substituted into the vehicle dynamics model, thereby improving the direct usability of the parameter identification values ​​in vehicle dynamics control.

[0007] In one possible implementation, the state parameters in the time series of true state parameter values ​​include: longitudinal velocity of the center of mass, lateral velocity of the center of mass, yaw rate, and wheel angular velocity of the target wheel; the target wheel is one of the wheels on the vehicle. The observed parameters in the time series of true observed parameter values ​​include: steering wheel angle, driving force and braking force provided by the power system. The hidden state parameters in the time series of estimated hidden state parameters include: wheel angle of the target wheel, wheel center velocity, slip ratio, sideslip angle, wheel-end driving force, wheel-end braking force, and vertical load. The kinematic parameters to be identified include: parameters varying according to parameter values. The kinematic parameters to be identified for the first, second, and third targets are defined by the velocity classification. The first target kinematic parameters include: the first distance from the vehicle's center of gravity to the front axle, the yaw angle, moment of inertia, air resistance coefficient, the vehicle's frontal area, rolling resistance coefficient, and braking system response delay. The second target kinematic parameters include: the actual longitudinal stiffness and actual lateral stiffness of the target axle; the target axle is either the front axle or the rear axle. The third target kinematic parameters include: the road surface equivalent adhesion coefficient.

[0008] Based on the aforementioned technical means, by dividing the kinematic parameters to be identified into first target kinematic parameters, second target kinematic parameters, and third target kinematic parameters according to the rate of change of parameter values, the model can identify parameters with different rates of change. For parameters such as center of mass position, yaw moment of inertia, and air drag coefficient that change slowly or remain basically unchanged, as well as parameters such as tire stiffness that change slowly with wear, and parameters such as road surface adhesion coefficient that change rapidly with working conditions, different identification strategies are used to process them separately, taking into account both the timeliness and stability of parameter identification.

[0009] In one possible implementation, the parameter identification values ​​of the vehicle's kinematic parameters to be identified are determined based on the time series sequence of operating parameters, the time series sequence of latent state parameter estimates, and the initial vehicle kinematic parameter identification model. This includes: inputting the time series sequence of operating parameters and the time series sequence of latent state parameter estimates into the first parameter identification layer of the initial vehicle kinematic parameter identification model to obtain the initial identification value of the first target kinematic parameter to be identified output by the first parameter identification layer; and inputting the time series sequence of operating parameters and the time series sequence of latent state parameter estimates into the second parameter identification layer of the initial vehicle kinematic parameter identification model to obtain the initial identification value of the second target kinematic parameter to be identified output by the first parameter identification layer. The time series of running parameters and the time series of hidden state parameter estimates are input into the third parameter recognition layer of the initial vehicle kinematics parameter recognition model to obtain the initial recognition value of the third target kinematic parameter to be identified output by the first parameter recognition layer; the initial recognition values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified are input into the parameter correction layer of the initial vehicle kinematics parameter recognition model to obtain the parameter recognition values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified output by the parameter correction layer.

[0010] Based on the above technical means, by sending the three types of kinematic parameters to be identified to the first parameter recognition layer, the second parameter recognition layer, and the third parameter recognition layer for independent identification, and obtaining their respective initial identification values, they are then uniformly sent to the parameter correction layer for correction. This allows parameters with different rates of change to have their features extracted by their respective appropriate recognition layers. At the same time, the parameter correction layer coordinates the initial identification values ​​output independently by the three layers, avoiding the problem of loss of correlation between parameters caused by hierarchical identification.

[0011] In one possible implementation, a parameter correction layer is used to correct the initial identification value of any kinematic parameter to be identified among the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified, by using the maximum allowable upper limit and the minimum allowable lower limit of the parameter of any kinematic parameter to be identified, so as to obtain the parameter identification value of any kinematic parameter to be identified.

[0012] Based on the above technical means, the initial identification value is corrected by the parameter correction layer using the maximum allowable upper limit and minimum allowable lower limit of the kinematic parameters to be identified, so that the corrected parameter identification value is always within the physically feasible range, avoiding the model output of parameter values ​​that violate physical constraints.

[0013] In one possible implementation, based on the time series of latent state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model, the predicted time series of vehicle state parameters is determined. This includes: determining the longitudinal force and lateral force of the target wheel in its own coordinate system based on the time series of latent state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, and the tire equivalent model; wherein, the tire equivalent model is used to characterize: the correspondence between the longitudinal force of the target wheel in its own wheel coordinate system, the actual longitudinal stiffness of the target wheel, the slip ratio of the target wheel, the vertical load of the target wheel, and the road surface equivalent adhesion coefficient; and the lateral force and actual lateral stiffness of the target wheel in its own wheel coordinate system. The correspondence between the angle of slip, the side slip angle of the target wheel, the vertical load of the target wheel, and the equivalent adhesion coefficient of the road surface is defined. The actual longitudinal stiffness and actual lateral stiffness of the target wheel are determined based on the actual longitudinal stiffness and actual lateral stiffness of the axle on which the target wheel is located, as well as the ratio of the vertical loads of the two wheels on the axle on which the target wheel is located. The time series of predicted values ​​of the vehicle's state parameters is determined based on the longitudinal and lateral forces of the target wheel in its own coordinate system, the time series of estimated values ​​of hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, and the dynamic equivalent model. Among them, the dynamic equivalent model is used to characterize the correspondence between the longitudinal and lateral forces of the target wheel in its own coordinate system, the parameter values ​​of hidden state parameters, the parameter values ​​of the kinematic parameters to be identified, and the state parameters.

[0014] Based on the above technical means, by coupling the tire equivalent model with the dynamic equivalent model, the parameter identification values ​​are first converted into the longitudinal force and lateral force of the target wheel in its own coordinate system based on the tire equivalent model. Then, the tire force and the hidden state parameters are combined through the dynamic equivalent model to deduce the vehicle-level state parameter prediction value. This forms a complete physical transmission link of "parameter-tire force-vehicle state" from the parameter identification value to the state prediction value, ensuring the physical interpretability of the state prediction.

[0015] In one possible implementation, the dynamic equivalent model is specifically used to characterize: the vehicle mass, longitudinal acceleration of the center of mass, lateral velocity, yaw rate, the resultant longitudinal force of each wheel in the vehicle's coordinate system, air resistance, and rolling resistance; and the relationship between the vehicle mass, longitudinal velocity, lateral acceleration of the center of mass, yaw rate, and the resultant lateral force of each wheel in the vehicle's coordinate system; and the moment of inertia of the target wheel, the wheel angular acceleration of the target wheel, and the wheel-end drive of the target wheel. The relationships between dynamic torque, wheel-end braking torque, effective rolling radius of the wheel, and longitudinal force of the target wheel in its own wheel coordinate system; and the relationships between the longitudinal coordinate of the target wheel in the vehicle coordinate system, the lateral coordinate of the target wheel in the vehicle coordinate system, the lateral force of the target wheel in the vehicle coordinate system, and the longitudinal force, yaw angle moment of inertia, and yaw angle acceleration of the target wheel in the vehicle coordinate system; and the relationships between the lateral velocity, lateral velocity, and yaw angle velocity of the target wheel at its wheel center in the vehicle coordinate system. The correspondence between longitudinal coordinates in the body coordinate system; and the correspondence between the longitudinal velocity, longitudinal velocity, yaw rate, and lateral coordinates of the target wheel in the vehicle body coordinate system; and the correspondence between the lateral force, longitudinal force, lateral force, and longitudinal force of the target wheel in its own wheel coordinate system, and the rotation matrix between the wheel coordinate system and the vehicle body coordinate system; wherein, wheel angular acceleration is the derivative of wheel angular velocity; yaw rate is the derivative of yaw rate; longitudinal acceleration is the derivative of longitudinal velocity; lateral acceleration is the derivative of longitudinal velocity; air resistance is determined based on air resistance coefficient and frontal area; rolling resistance is determined based on rolling resistance coefficient; longitudinal and lateral coordinates are determined based on first distance; the rotation matrix is ​​determined by the rotation angle of the target wheel; the vehicle body coordinate system is constructed with the center of mass as the origin and the directions of the vehicle's longitudinal and lateral forces as coordinate axes.

[0016] Based on the aforementioned technical means, the dynamic equivalent model covers the vehicle's longitudinal translational motion, lateral translational motion, wheel rotational motion, yaw rotational motion, and the mechanical transformation relationship between the wheel coordinate system and the vehicle body coordinate system, forming a multi-degree-of-freedom coupled vehicle dynamics description. This enables the predicted state parameters to comprehensively reflect the vehicle's dynamic response in the four dimensions of longitudinal, lateral, yaw, and wheel rotation, improving the completeness and accuracy of state prediction.

[0017] In one possible implementation, based on the first state prediction deviation between the time series of true state parameter values ​​and the time series of predicted state parameter values, an initial vehicle kinematic parameter identification model is optimized to obtain a vehicle kinematic parameter identification model. This includes: determining a loss function based on at least one of the first state prediction deviation and the target deviation; and optimizing the model parameters of the initial vehicle kinematic parameter identification model based on the loss function to obtain the vehicle kinematic parameter identification model. The target deviation includes: a first identification deviation between the identified parameter value and the true value of the kinematic parameter to be identified; a parameter correction deviation between the identified parameter value and the initial identification value of the kinematic parameter to be identified; and the target wheel's position in its own coordinate system. The second identification deviation between the torque and the actual target torque of the target wheel in its own coordinate system; the target torque is one of the longitudinal force and the lateral force; the second state prediction deviation between the time series of theoretical values ​​of the vehicle's state parameters and the time series of actual values ​​of the state parameters; wherein, the time series of theoretical values ​​of the state parameters is determined based on the time series of actual values ​​of the hidden state parameters, the actual values ​​of the kinematic parameters to be identified, the dynamic equivalent model, and the tire equivalent model; the parameter smoothing deviation between the parameter identification value of the kinematic parameters to be identified and the parameter identification value of the kinematic parameters to be identified at the previous moment; and the cumulative prediction deviation of the parameter identification value of the kinematic parameters to be identified for predicting the vehicle state at multiple future moments.

[0018] Based on the above technical means, in addition to introducing the first state prediction bias into the loss function, at least one of the following is also incorporated: the first identification bias, parameter correction bias, tire force identification bias, second state prediction bias, parameter smoothing bias, and cumulative prediction bias between the parameter identification value and the true value. This makes the model training process not only focus on the accuracy of the parameter identification value itself, but also constrain the performance of the parameter identification value in multiple dimensions such as physical rationality, temporal smoothness, future prediction ability, and torque level accuracy, thereby obtaining a vehicle kinematic parameter identification model that is optimized at multiple physical levels.

[0019] Secondly, this application provides a method for identifying vehicle kinematic parameters, comprising: acquiring a time series sequence of the vehicle's current operating parameters; wherein the time series sequence of the current operating parameters includes a time-aligned time series sequence of the true values ​​of the current state parameters and a time series sequence of the true values ​​of the current observed parameters; determining a time series sequence of the estimated values ​​of the vehicle's current hidden state parameters based on the time series sequence of the current operating parameters and the vehicle's inherent kinematic parameters; and determining the current parameter identification value of the vehicle's kinematic parameters to be identified based on the time series sequence of the current operating parameters, the time series sequence of the estimated values ​​of the current hidden state parameters, and a vehicle kinematic parameter identification model.

[0020] Based on the aforementioned technical means, by acquiring time-aligned time series sequences of the true values ​​of the current state parameters and the true values ​​of the current observed parameters, and determining the time series sequence of the estimated values ​​of the current hidden state parameters based on the time series sequence of the current operating parameters and the inherent kinematic parameters, key intermediate state parameters that are difficult to measure directly can be estimated online in real time. Furthermore, using the vehicle kinematic parameter identification model obtained through the aforementioned training method, the time series sequence of the current operating parameters and the time series sequence of the estimated values ​​of the current hidden state parameters are processed to obtain the current parameter identification values ​​of the kinematic parameters to be identified. This ensures that the identified parameters are not only numerically accurate but also physically consistent with the actual dynamic response of the vehicle, thus enabling direct application to vehicle dynamics control and achieving seamless integration of online real-time identification of vehicle kinematic parameters and dynamics control.

[0021] Thirdly, this application provides a training device for a vehicle kinematic parameter recognition model, the device comprising: an acquisition module, a first determination module, a second determination module, a third determination module, and an optimization module.

[0022] The acquisition module is used to acquire the time series sequence of the vehicle's operating parameters; wherein, the time series sequence of operating parameters includes the time-aligned time series sequence of the true values ​​of the state parameters and the time series sequence sequence of the true values ​​of the observation parameters; The first determining module is used to determine the time series of hidden state parameter estimates of the vehicle based on the time series of operating parameters and the inherent kinematic parameters of the vehicle. The second determining module is used to determine the parameter identification values ​​of the vehicle's kinematic parameters to be identified based on the time series sequence of operating parameters, the time series sequence of hidden state parameter estimates, and the initial vehicle kinematic parameter identification model. The third determination module is used to determine the time series of predicted values ​​of vehicle state parameters based on the time series of estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model; the tire equivalent model is coupled with the vehicle's dynamic equivalent model. The optimization module is used to optimize the initial vehicle kinematic parameter recognition model based on the first state prediction deviation between the time series sequence of the true state parameter values ​​and the time series sequence of the predicted state parameter values, thereby obtaining the vehicle kinematic parameter recognition model.

[0023] The state parameters in the above-mentioned true value time series of state parameters include: longitudinal velocity of the center of mass, lateral velocity of the center of mass, yaw rate, and wheel angular velocity of the target wheel; the target wheel is one of the wheels on the vehicle. The observed parameters in the observed parameter true value time series include: steering wheel angle, driving force and braking force provided by the power system. The hidden state parameters in the hidden state parameter estimation time series include: wheel angle of the target wheel, wheel center velocity, slip ratio, sideslip angle, wheel-end driving force, wheel-end braking force, and vertical load. The kinematic parameters to be identified include: classified according to the rate of change of parameter values. The first target kinematic parameters to be identified are: a first target kinematic parameters to be identified, a second target kinematic parameters to be identified, and a third target kinematic parameters to be identified. The first target kinematic parameters to be identified include: the first distance from the vehicle's center of mass to the front axle, the yaw angle moment of inertia, the air drag coefficient, the vehicle's frontal area, the rolling resistance coefficient, and the vehicle's braking system response delay. The second target kinematic parameters to be identified include: the actual longitudinal stiffness and actual lateral stiffness of the target axle of the vehicle; the target axle is one of the front axle and the rear axle of the vehicle. The third target kinematic parameters to be identified include: the road surface equivalent adhesion coefficient.

[0024] The second determining module mentioned above is also used to input the time sequence of the running parameters and the time sequence of the estimated hidden state parameters into the first parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the first target kinematic parameter to be identified output by the first parameter recognition layer; The time series sequence of running parameters and the time series sequence of hidden state parameter estimates are input into the second parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the second target kinematic parameter to be identified output by the first parameter recognition layer. The time series sequence of running parameters and the time series sequence of hidden state parameter estimates are input into the third parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the third target kinematic parameter to be identified output by the first parameter recognition layer. The initial identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified are input into the parameter correction layer in the initial vehicle kinematic parameter identification model to obtain the parameter identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified output by the parameter correction layer.

[0025] The aforementioned parameter correction layer is used to correct the initial identification value of any kinematic parameter to be identified among the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified, by using the maximum allowable upper limit and the minimum allowable lower limit of the parameter of any kinematic parameter to be identified, so as to obtain the parameter identification value of any kinematic parameter to be identified.

[0026] The second determining module is further used to determine the longitudinal and lateral forces of the target wheel in its own coordinate system based on the time series of latent state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, and the tire equivalent model. The tire equivalent model is used to characterize the correspondence between: the longitudinal force of the target wheel in its own wheel coordinate system, the actual longitudinal stiffness of the target wheel, the slip ratio of the target wheel, the vertical load of the target wheel, and the equivalent road surface adhesion coefficient; and the correspondence between: the lateral force of the target wheel in its own wheel coordinate system, the actual lateral stiffness of the target wheel, the sideslip angle of the target wheel, the vertical load of the target wheel, and the equivalent road surface adhesion coefficient. The relationship is as follows: The actual longitudinal stiffness and actual lateral stiffness of the target wheel are determined based on the actual longitudinal stiffness and actual lateral stiffness of the axle on which the target wheel is located, as well as the vertical load ratio of the two wheels on the axle on which the target wheel is located; Based on the longitudinal and lateral forces of the target wheel in its own coordinate system, the time series of the estimated values ​​of the latent state parameters, the parameter identification values ​​of the kinematic parameters to be identified, and the dynamic equivalent model, the time series of the predicted values ​​of the vehicle's state parameters are determined; Among them, the dynamic equivalent model is used to characterize the correspondence between the longitudinal and lateral forces of the target wheel in its own coordinate system, the parameter values ​​of the latent state parameters, the parameter values ​​of the kinematic parameters to be identified, and the state parameters.

[0027] The aforementioned equivalent dynamic model is specifically used to characterize: the vehicle's mass, longitudinal acceleration of the center of mass, lateral velocity, yaw rate, the resultant longitudinal force of each wheel in the vehicle's coordinate system, air resistance, and rolling resistance; the vehicle's mass, longitudinal velocity, lateral acceleration of the center of mass, yaw rate, and the resultant lateral force of each wheel in the vehicle's coordinate system; the target wheel's moment of inertia, wheel angular acceleration, wheel-end driving torque, wheel-end braking torque, effective rolling radius, and longitudinal force in its own wheel coordinate system; the target wheel's longitudinal coordinate in the vehicle's coordinate system, lateral coordinate in the vehicle's coordinate system, lateral force in the vehicle's coordinate system, longitudinal force, yaw rate, moment of inertia, and yaw rate in the vehicle's coordinate system; and the target wheel's wheel-center lateral velocity, lateral velocity, yaw rate, and the resultant lateral force of each wheel in the vehicle's coordinate system. The correspondence between the longitudinal coordinates; and the correspondence between the longitudinal velocity, longitudinal velocity, yaw rate, and lateral coordinates of the target wheel in the vehicle coordinate system; and the correspondence between the lateral force, longitudinal force, lateral force, and longitudinal force of the target wheel in its own wheel coordinate system, and the rotation matrix between the wheel coordinate system and the vehicle coordinate system; wherein, wheel angular acceleration is the derivative of the wheel angular velocity of the target wheel; yaw rate is the derivative of the yaw rate; longitudinal acceleration is the derivative of the longitudinal velocity; lateral acceleration is the derivative of the longitudinal velocity; air resistance is determined based on the air resistance coefficient and the frontal area; rolling resistance is determined based on the rolling resistance coefficient; longitudinal and lateral coordinates are determined based on the first distance; the rotation matrix is ​​determined by the rotation angle of the target wheel; the vehicle coordinate system is constructed with the center of mass as the origin and the directions of the longitudinal and lateral forces of the vehicle as the coordinate axes.

[0028] The aforementioned third determining module is further configured to determine a loss function based on at least one of the first state prediction deviation and the target deviation; and to optimize the model parameters of the initial vehicle kinematic parameter identification model based on the loss function to obtain the vehicle kinematic parameter identification model; wherein, the target deviation includes: a first identification deviation between the parameter identification value and the true value of the kinematic parameter to be identified; a parameter correction deviation between the parameter identification value and the initial identification value of the kinematic parameter to be identified; and a second identification deviation between the target torque of the target wheel in its own coordinate system and the true target torque of the target wheel in its own coordinate system; the target Torque is one of longitudinal force and lateral force; the second state prediction deviation between the time series of theoretical values ​​of vehicle state parameters and the time series of actual values ​​of state parameters; wherein, the time series of theoretical values ​​of state parameters is determined based on the time series of actual values ​​of hidden state parameters, the actual values ​​of kinematic parameters to be identified, the dynamic equivalent model, and the tire equivalent model; the parameter smoothing deviation between the parameter identification value of the kinematic parameter to be identified and the parameter identification value of the kinematic parameter to be identified at the previous moment; and the cumulative prediction deviation of the parameter identification value of the kinematic parameter to be identified for predicting the vehicle state at multiple future moments.

[0029] Fourthly, this application provides a vehicle kinematics parameter identification device, comprising: an acquisition module, a first determination module, and a second determination module.

[0030] The acquisition module is used to acquire the time series sequence of the vehicle's current operating parameters; wherein, the time series sequence of the current operating parameters includes the time series sequence of the true values ​​of the current state parameters aligned with time and the time series sequence of the true values ​​of the current observation parameters. The first determining module is used to determine the time series of estimated values ​​of the vehicle's current hidden state parameters based on the time series of current operating parameters and the vehicle's inherent kinematic parameters. The second determining module is used to determine the current parameter identification value of the vehicle's kinematic parameters to be identified based on the current operating parameter time series, the current hidden state parameter estimation time series, and the vehicle kinematic parameter identification model; wherein, the vehicle kinematic parameter identification model is trained using the training method of the vehicle kinematic parameter identification model as described in the first aspect.

[0031] Fifthly, this application provides a vehicle that includes a training device for the vehicle kinematic parameter recognition model described in the third aspect above.

[0032] In a sixth aspect, this application provides an electronic device, comprising: a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the methods of the first aspect and the second aspect described above and any possible implementation thereof.

[0033] In a seventh aspect, this application provides a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the methods described in the first and second aspects and any possible implementation thereof.

[0034] Eighthly, this application provides a computer program product including computer instructions that, when executed on an electronic device, cause the electronic device to perform the method described in the first aspect and any possible implementation thereof.

[0035] It should be noted that the technical effects of any of the implementation methods in aspects three through eight can be found in the technical effects of the corresponding implementation methods in aspects one or two, and will not be repeated here.

[0036] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0037] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application, and do not constitute an undue limitation of this application.

[0038] Figure 1 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application.

[0039] Figure 2 This is a flowchart illustrating a training method for a vehicle kinematic parameter recognition model provided in an embodiment of this application.

[0040] Figure 3 This is a flowchart illustrating another method for training a vehicle kinematic parameter recognition model provided in this application embodiment.

[0041] Figure 4 This is a flowchart illustrating another method for training a vehicle kinematic parameter recognition model provided in this application embodiment.

[0042] Figure 5 This is a flowchart illustrating another method for training a vehicle kinematic parameter recognition model provided in this application embodiment.

[0043] Figure 6 This is a flowchart illustrating a method for identifying vehicle kinematic parameters according to an embodiment of this application.

[0044] Figure 7 This is a block diagram of a training device for a vehicle kinematic parameter recognition model, as shown in an embodiment of this application.

[0045] Figure 8 This is a block diagram of a vehicle kinematics parameter recognition device shown in an embodiment of this application.

[0046] Figure 9 This is a block diagram illustrating an electronic device according to an embodiment of this application. Detailed Implementation

[0047] To enable those skilled in the art to better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0048] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0049] In the embodiments of this application, the words "exemplary," "for example," or "for instance" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a specific manner.

[0050] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0052] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0053] The vehicle kinematic parameter recognition model training device provided in this application embodiment is used to execute the vehicle kinematic parameter recognition model training method described in the first aspect or any possible implementation above, so as to perform vehicle dynamics control based on the parameter recognition values ​​output by the trained vehicle kinematic parameter recognition model, thereby enabling the vehicle's kinematic parameters to be accurately identified in real time, improving the accuracy of vehicle dynamics control and its adaptability under different working conditions. The vehicle can also be referred to as a vehicle, mobile carrier, electric vehicle (EV), hybrid electric vehicle (HEV), plug-in hybrid electric vehicle (PHEV), fuel cell vehicle (FCV), autonomous vehicle, intelligent and connected vehicle (ICV), driverless vehicle, etc.

[0054] In this application embodiment, the vehicle can be a sedan, bus, sport utility vehicle (SUV), truck, special vehicle, driverless taxi, intelligent connected bus, autonomous logistics vehicle, electric truck, etc. In addition, the method is also applicable to other mobile vehicles equipped with vehicle communication terminals, such as tricycles, two-wheeled vehicles, trains and other transportation devices that carry people or goods, or other types of vehicles powered by power batteries, etc. This application does not impose specific limitations on these.

[0055] Figure 1 This is a structural schematic diagram of a vehicle provided in an embodiment of this application. For example... Figure 1 As shown, the vehicle 100 includes a training device 101, a data acquisition device 102, and an execution device 103 for a vehicle kinematic parameter recognition model.

[0056] The training device 101 for the vehicle kinematic parameter recognition model is a processing device for executing the training method of the vehicle kinematic parameter recognition model and the vehicle kinematic parameter recognition method.

[0057] The training device 101 for the vehicle kinematics parameter recognition model has independent instruction space, power domain, and clock domain, and runs independently of the application processor's operating system or software stack. The training device 101 for the vehicle kinematics parameter recognition model can be a standalone microcontroller or a dedicated processing unit integrated into the vehicle domain controller.

[0058] The aforementioned acquisition device 102 is used to acquire the time series sequence of vehicle operating parameters, wherein the time series sequence of operating parameters includes a time-aligned time series sequence of real values ​​of state parameters and a time series sequence sequence of real values ​​of observation parameters.

[0059] The data acquisition device 102 may include at least one of the following mounted on the vehicle: an inertial measurement unit, a wheel speed sensor, a steering wheel angle sensor, a drive motor controller, and a braking system controller, for acquiring operating parameters such as longitudinal velocity of the center of gravity, lateral velocity of the center of gravity, yaw rate, wheel angular velocity of the target wheel, steering wheel angle, driving force, and braking force.

[0060] The actuator 103 uses the current parameter identification value of the kinematic parameter to be identified output by the training device 101 of the vehicle kinematic parameter identification model to perform vehicle dynamics control. The actuator 103 may include at least one of the vehicle's drive system, braking system, steering system, and suspension system.

[0061] The training device 101 for the vehicle kinematic parameter recognition model is used to determine the time series of estimated values ​​of the current hidden state parameters of the vehicle based on the time series of current operating parameters acquired by the acquisition device 102 and the inherent kinematic parameters of the vehicle during vehicle operation, and to determine the current parameter recognition value of the kinematic parameter to be identified based on the time series of current operating parameters, the time series of estimated values ​​of current hidden state parameters and the pre-trained vehicle kinematic parameter recognition model.

[0062] Specifically, during the training phase, the training device 101 for the vehicle kinematic parameter recognition model performs hidden state parameter estimation, parameter identification value determination, state parameter prediction value determination, and model optimization based on the first state prediction deviation in sequence, based on the historical operating parameter time sequence obtained by the acquisition device 102, to obtain the vehicle kinematic parameter recognition model.

[0063] During the inference phase, the training device 101 of the vehicle kinematic parameter recognition model outputs the current parameter identification values ​​of the kinematic parameters to be identified in real time, based on the current operating parameter time sequence obtained by the acquisition device 102 and the already trained vehicle kinematic parameter recognition model. These values ​​include the first distance from the center of mass to the front axle, yaw angle moment of inertia, air resistance coefficient, rolling resistance coefficient, actual longitudinal stiffness and actual lateral stiffness of the target axle, and the road surface equivalent adhesion coefficient.

[0064] In some embodiments, the training device 101 for the vehicle kinematics parameter recognition model further constructs a loss function based on at least one target deviation other than the first state prediction deviation during the training phase. The target deviation includes at least one of the following: a first recognition deviation between the parameter recognition value and the true value, a parameter correction deviation, a tire force recognition deviation, a second state prediction deviation, a parameter smoothing deviation, and a cumulative prediction deviation, in order to optimize the model parameters under multi-dimensional constraints.

[0065] This application does not limit the specific physical form of the training device 101 for the vehicle kinematic parameter identification model, as long as it can realize the above-mentioned functions of acquiring the time sequence of operating parameters, estimating hidden state parameters, determining parameter identification values, determining state parameter prediction values, and optimizing the model based on state prediction deviation.

[0066] The acquisition device 102 is used to acquire the real-time time series of state parameters and the real-time time series of observation parameters during vehicle operation.

[0067] Specifically, the acquisition device 102 acquires the longitudinal velocity and lateral velocity of the center of mass through the inertial measurement unit, the yaw rate through the yaw rate sensor, the wheel angular velocity of the target wheel through the wheel speed sensor, the steering wheel angle through the steering wheel angle sensor, the driving force through the drive motor controller, and the braking force through the braking system controller. The data acquired by the above sensors are then time-synchronized and aligned to obtain the time-aligned time sequence of the true values ​​of the state parameters and the time sequence of the true values ​​of the observation parameters.

[0068] As one possible approach, the acquisition device 102 also includes a data preprocessing unit for filtering, noise reduction, and outlier removal of the acquired raw sensor data, and for aligning sensor data with different sampling frequencies to a unified time axis through interpolation or downsampling.

[0069] The actuator 103 receives the current parameter identification value of the kinematic parameter to be identified from the training device 101 of the vehicle kinematic parameter identification model, and adjusts the vehicle's dynamic control strategy based on the parameter identification value. For example, the actuator 103 adjusts the control parameters of the anti-lock braking system based on the current parameter identification value of the road surface equivalent adhesion coefficient, or adjusts the control threshold of the electronic stability program based on the current parameter identification values ​​of the actual longitudinal stiffness and actual lateral stiffness of the target axle.

[0070] In this embodiment, the training device 101, the acquisition device 102, and the execution device 103 of the vehicle kinematic parameter recognition model work together. The collecting device 102 collects vehicle operating parameters in real time and provides them to the training device 101 for the vehicle kinematic parameter recognition model; the training device 101 for the vehicle kinematic parameter recognition model identifies the current value of the kinematic parameter to be identified online based on the received operating parameters and outputs the identification result to the execution device 103; the execution device 103 executes the corresponding dynamic control action according to the identification result, so that the vehicle dynamic control parameters can be adaptively adjusted according to the changes in the actual physical state of the vehicle, thereby improving the control accuracy and driving safety of the vehicle under different working conditions.

[0071] like Figure 2 As shown, this embodiment provides a training method for a vehicle kinematic parameter recognition model, including the following steps: S201. Obtain the timing sequence of the vehicle's operating parameters.

[0072] The time series of operating parameters includes time-aligned time series of state parameter real values ​​and time series of observation parameter real values.

[0073] The true value time series of state parameters refers to the sequence of parameters that reflect the motion state of the vehicle, collected by sensors or output by a state estimator during the actual operation of the vehicle, arranged in chronological order. These parameters include longitudinal velocity of the center of gravity, lateral velocity of the center of gravity, yaw rate, and wheel angular velocity of the target wheel.

[0074] The time series of observed parameters refers to the sequence of parameters that are directly measured by sensors or output by controllers during the actual operation of the vehicle, and that reflect the driver's operating intentions and the state of the vehicle's actuators, arranged in chronological order. These parameters include steering wheel angle, driving force and braking force provided by the power system, etc.

[0075] The aforementioned time alignment refers to the correspondence between each sampling moment in the time series of the true values ​​of the state parameters and each sampling moment in the time series of the true values ​​of the observation parameters on the time axis, with the state parameters and observation parameters at the same moment forming a one-to-one data pair.

[0076] The aforementioned inherent kinematic parameters refer to parameters that are determined during the vehicle's design and manufacturing phase and do not change or change very little during a single recognition session. These parameters include vehicle mass, wheelbase, front track, rear track, center of gravity height, effective rolling radius of the wheels, wheel moment of inertia, steering system transmission ratio, transmission ratio, transmission system mechanical efficiency, and front axle lateral load transfer distribution coefficient.

[0077] In one possible implementation, sensor data of the vehicle under various driving conditions is acquired, including longitudinal and lateral acceleration collected by the inertial measurement unit, yaw rate collected by the yaw rate sensor, four-wheel wheel speed collected by the wheel speed sensor, steering wheel angle collected by the steering wheel angle sensor, drive torque output by the drive motor controller, and braking pressure output by the braking system controller. The above raw sensor data is then processed for time synchronization. Sensor data with different sampling frequencies are aligned to a unified time axis through interpolation or downsampling. The data is then filtered, denoised, and outlier removed to obtain a time-aligned time series of state parameter true values ​​and a time series of observation parameter true values.

[0078] In some embodiments, the training data comes from the simulation data generated by the high-fidelity vehicle dynamics simulation platform. During the simulation, the parameters to be identified, sensor noise, sensor zero bias, road slope, air relative speed and wheel-end torque ratio error are randomized in the domain to cover a sufficiently rich combination of parameters and a range of working conditions. The training set, validation set and test set are divided according to the complete running trajectory and parameter combination before the sliding window is split, so as to avoid different windows of the same trajectory crossing the boundary of the dataset.

[0079] S202. Based on the time series of operating parameters and the inherent kinematic parameters of the vehicle, determine the time series of the estimated hidden state parameters of the vehicle.

[0080] Among them, the time series of hidden state parameter estimates refers to the sequence of intermediate state parameters that are difficult to measure directly by sensors during vehicle operation and need to be calculated from operating parameters and inherent kinematic parameters through kinematic relationships or quasi-static models, arranged in chronological order. These parameters include the wheel angle, wheel center speed, slip ratio, sideslip angle, wheel end driving force, wheel end braking force, and vertical load of the target wheel.

[0081] The target wheel is one of the wheels on the vehicle, which can be any one of the left front wheel, right front wheel, left rear wheel, or right rear wheel.

[0082] After obtaining the timing sequence of operating parameters, the actual front wheel steering angle is calculated based on the steering system transmission ratio and steering wheel angle from the inherent kinematic parameters. For vehicles without rear-wheel steering, the rear wheel steering angle is taken as zero.

[0083] Specifically, the steering wheel angle is subtracted from the steering wheel zero-position offset and then divided by the steering system transmission ratio to obtain the reference front wheel angle. The reference front wheel angle is then allocated to the actual steering angles of the left and right front wheels according to the Ackermann steering relationship or steering system calibration table.

[0084] Based on the vehicle mass, wheelbase, center of gravity height, front track and rear track in the inherent kinematic parameters, and the longitudinal acceleration and lateral acceleration in the time series of the true values ​​of the state parameters, the vertical load of each wheel is calculated by a quasi-static longitudinal and lateral load transfer model.

[0085] Specifically, the longitudinal load transfer between the front and rear axles is first calculated based on the longitudinal acceleration and vehicle mass to obtain the total vertical load of the front axle and the total vertical load of the rear axle. Then, based on the lateral acceleration, center of gravity height, wheelbase, and front axle lateral load transfer distribution coefficient, the lateral load transfer between the left and right wheels is calculated, and the total load of the front and rear axles is distributed to each wheel. After calculating the original vertical load of each wheel, a positive correction is applied to the original vertical load, and then it is normalized again according to the load ratio of the left and right wheels on the same axle so that the sum of the vertical loads of the left and right wheels on the same axle is equal to the total load of that axle, so as to maintain load conservation.

[0086] S203. Based on the time series sequence of operating parameters, the time series sequence of hidden state parameter estimates, and the initial vehicle kinematic parameter identification model, determine the parameter identification values ​​of the vehicle's kinematic parameters to be identified.

[0087] The initial vehicle kinematic parameter identification model refers to a neural network model that has not yet been trained and is used to identify the kinematic parameters to be identified from the operating parameters and latent state parameters.

[0088] The kinematic parameters to be identified refer to parameters that change during vehicle operation due to wear, aging, load changes, or changes in road conditions. These include the first target kinematic parameters to be identified, the second target kinematic parameters to be identified, and the third target kinematic parameters to be identified, classified according to the rate of change of parameter values.

[0089] The first target's kinematic parameters change the slowest, including the first distance from the vehicle's center of gravity to the front axle, yaw angle, moment of inertia, drag coefficient, frontal area, rolling resistance coefficient, and braking system response delay; the second target's kinematic parameters change at a moderate rate, including the actual longitudinal stiffness and actual lateral stiffness of the vehicle's front and rear axles; the third target's kinematic parameters change the fastest, including the road surface equivalent adhesion coefficient.

[0090] In one possible implementation, the time series of running parameters and the time series of hidden state parameter estimates are input into the initial vehicle kinematics parameter identification model. The initial vehicle kinematics parameter identification model extracts time series features and maps parameters from the input data, and outputs the parameter identification values ​​of the kinematic parameters to be identified.

[0091] The aforementioned initial vehicle kinematics parameter identification model can employ one or more combinations of gated recurrent unit networks, temporal convolutional networks, or lightweight temporal networks, and set multiple parameter output heads to correspond to different levels of the kinematic parameters to be identified.

[0092] S204. Based on the time series of hidden state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model, determine the time series of predicted state parameters of the vehicle.

[0093] The tire equivalent model is coupled with the vehicle dynamics equivalent model.

[0094] The dynamic equivalent model is a mathematical model used to describe the relationship between the motion and forces of a vehicle in seven degrees of freedom: longitudinal, lateral, yaw, and rotation of the four wheels. Its inputs include the tire force, wheel-end driving torque, and braking torque of each wheel, and its output is the vehicle state parameters at the next moment.

[0095] The tire equivalent model is a mathematical model used to describe the relationship between the longitudinal and lateral forces generated by the wheel in its own coordinate system and the wheel slip ratio, sideslip angle, vertical load and road adhesion coefficient. The tire equivalent model is coupled with the dynamic equivalent model, that is, the tire force output by the tire equivalent model is directly used as the input of the dynamic equivalent model.

[0096] In one possible implementation, the slip ratio, sideslip angle, and vertical load of each wheel in the time series of the hidden state parameter estimates, as well as the road equivalent adhesion coefficient and the actual longitudinal stiffness and actual lateral stiffness of the target axle in the kinematic parameters to be identified, are input into the tire equivalent model to obtain the longitudinal force and lateral force of each wheel in its own coordinate system.

[0097] The actual longitudinal stiffness and actual lateral stiffness of the target axle are distributed to each wheel according to the vertical load ratio of the left and right wheels of the axle.

[0098] After transforming the longitudinal and lateral forces of each wheel in its own coordinate system to the vehicle coordinate system through the wheel rotation matrix, along with the wheel-end driving torque and braking torque of each wheel in the time series of the latent state parameter estimates, as well as the first distance, yaw angle moment of inertia, air resistance coefficient, frontal area and rolling resistance coefficient among the kinematic parameters to be identified, they are input into the dynamic equivalent model.

[0099] The dynamic equivalent model recursively derives the predicted values ​​of the longitudinal velocity, lateral velocity, yaw rate, and wheel angular velocity of the target wheel at the next moment based on the three translational motion equations of the vehicle in the longitudinal, lateral, and yaw directions, as well as the rotational motion equations of each wheel, forming a time series of predicted state parameter values.

[0100] S205. Based on the first state prediction deviation between the time series of true state parameter values ​​and the time series of predicted state parameter values, optimize the initial vehicle kinematic parameter identification model to obtain the vehicle kinematic parameter identification model.

[0101] The first state prediction bias is a quantitative measure of the deviation between the predicted state parameter value at each time step in the state parameter prediction time series and the actual state parameter value at the corresponding time step in the state parameter actual value time series.

[0102] Specifically, the time series of predicted state parameter values ​​is compared with the time series of actual state parameter values ​​on a time-by-time basis. The deviations between the predicted and actual values ​​of the longitudinal velocity of the center of mass, the lateral velocity of the center of mass, the yaw rate, and the wheel angular velocity of the target wheel are calculated. The first state prediction deviation is obtained by weighted summation.

[0103] In one possible implementation, the first state prediction bias is used as the loss function, the gradient is calculated through the backpropagation algorithm and the model parameters of the initial vehicle kinematics parameter recognition model are updated, and the vehicle kinematics parameter recognition model is obtained after multiple rounds of iterative training.

[0104] Since the loss function is based on the deviation between the predicted and actual values ​​of the state parameters, and the predicted values ​​of the state parameters are derived from the parameter identification values ​​through the dynamic equivalent model and the tire equivalent model, the optimization process forces the parameter identification values ​​output by the model to maintain physical consistency between the predicted vehicle state response and the actual state response after being substituted into the vehicle dynamics model, thereby improving the direct usability of the parameter identification values ​​in vehicle dynamics control.

[0105] In one embodiment, the state parameters in the time sequence of true state parameter values ​​include: longitudinal velocity of the center of mass, lateral velocity of the center of mass, yaw rate, and wheel angular velocity of the target wheel.

[0106] The target wheel mentioned above is one of the wheels on the vehicle.

[0107] The observed parameters in the time series of the true values ​​of the observed parameters include: steering wheel angle, driving force and braking force provided by the power system.

[0108] The hidden state parameters in the time series of the estimated hidden state parameters include: the wheel angle, wheel center speed, slip ratio, sideslip angle, wheel end driving force, wheel end braking force, and vertical load of the target wheel.

[0109] The kinematic parameters to be identified include: a first target kinematic parameter to be identified, a second target kinematic parameter to be identified, and a third target kinematic parameter to be identified, classified according to the rate of change of parameter values.

[0110] The first target kinematic parameters to be identified include: the first distance from the vehicle's center of mass to the front axle, the yaw angle, moment of inertia, drag coefficient, frontal area, rolling resistance coefficient, and braking system response delay.

[0111] The second target kinematic parameters to be identified include: the actual longitudinal stiffness and actual lateral stiffness of the target axle of the vehicle.

[0112] The target axle is one of the front axle and the rear axle of the vehicle; the third target is the kinematic parameters to be identified, including the road surface equivalent adhesion coefficient.

[0113] Specifically, state parameters refer to physical quantities that can directly or indirectly reflect the motion state of a vehicle.

[0114] The longitudinal velocity of the center of gravity refers to the velocity of the vehicle's center of gravity along the longitudinal axis of the vehicle's coordinate system.

[0115] Lateral velocity of the center of mass refers to the velocity of the vehicle's center of mass along the lateral axis of the vehicle's coordinate system.

[0116] Yaw rate is the angular velocity of a vehicle's rotation around an axis perpendicular to the ground, with counterclockwise rotation being positive.

[0117] The angular velocity of the target wheel refers to the rotational angular velocity of the target wheel about its own axis of rotation. The target wheel is any one of the four wheels on the vehicle that is selected as the object of observation.

[0118] Observational parameters refer to physical quantities that can be directly measured by sensors or obtained directly from controllers, reflecting the driver's operating intentions and the response status of vehicle actuators.

[0119] Steering wheel angle refers to the angle at which the driver turns the steering wheel, used to determine the steering angle of the front wheels.

[0120] The driving force provided by the power system refers to the equivalent driving force corresponding to the driving torque transmitted from the drive motor or engine to the wheels through the transmission system.

[0121] Braking force refers to the equivalent braking force corresponding to the braking torque applied to the wheels by the braking system.

[0122] Implicit state parameters refer to intermediate state physical quantities that are difficult to measure directly by sensors during vehicle operation and need to be deduced from state parameters and observation parameters through kinematic relationships or quasi-static models.

[0123] The wheel angle of the target wheel is the angle between the steering plane of the target wheel and the longitudinal symmetry plane of the vehicle. The front wheel angle is obtained by converting the steering wheel angle into the steering system transmission ratio and then distributing it. The rear wheel angle is zero when the vehicle does not have rear wheel steering function.

[0124] Wheel center velocity refers to the longitudinal and lateral velocities of the target wheel center in the wheel's own coordinate system. The longitudinal velocity is along the rolling direction of the wheel, while the lateral velocity is perpendicular to the rolling direction of the wheel.

[0125] Slip ratio is the ratio of the difference between the rolling linear velocity and the longitudinal velocity of the wheel center to the longitudinal velocity of the wheel center. It is used to characterize the degree of slippage of the wheel in the longitudinal direction.

[0126] Side slip angle is the angle between the direction of the target wheel's center velocity and the direction of the wheel's rolling motion, used to characterize the degree of lateral deviation of the wheel.

[0127] Wheel-end driving force refers to the actual driving force transmitted to the wheel end of the target wheel through the transmission system.

[0128] Wheel-end braking force refers to the actual braking force exerted by the braking system on the wheel end of the target wheel.

[0129] Vertical load refers to the vertical reaction force exerted by the ground on the target wheel, reflecting the contact state between the target wheel and the ground.

[0130] The rate of change of parameter values ​​refers to the rate at which the kinematic parameters to be identified change due to wear, aging, load changes, or changes in road conditions during the actual operation of the vehicle.

[0131] The first target's kinematic parameters change the slowest, and can be considered to remain basically unchanged or change slowly during a single journey.

[0132] The first distance from the center of gravity to the front axle refers to the distance from the vehicle's center of gravity along the vehicle's longitudinal axis to the center line of the front axle. This parameter changes slowly as the positions of the occupants and cargo change.

[0133] Yaw angle moment of inertia refers to the moment of inertia of a vehicle about an axis perpendicular to the ground. This parameter changes slowly with the distribution of the loaded mass.

[0134] The drag coefficient refers to the aerodynamic drag coefficient of a vehicle's shape, and together with the frontal area, it determines the magnitude of air resistance. The frontal area is the projected area of ​​the vehicle's front in the direction of travel. The rolling resistance coefficient is the ratio of tire rolling resistance to the vertical load on the wheel, and it changes slowly with tire wear and tire pressure. Braking system response delay refers to the time delay between the issuance of a braking command and the actual establishment of braking torque, and it changes slowly with brake pad wear and brake fluid condition.

[0135] The second target is to identify kinematic parameters whose rate of change is moderate, and which may change to some extent during a single trip due to changes in tire temperature, tire pressure, or road conditions.

[0136] The actual longitudinal stiffness of the target axle refers to the equivalent stiffness of a single wheel on the target axle in the longitudinal direction, reflecting the deformation characteristics of the tire under longitudinal force.

[0137] The actual lateral stiffness of the target axle refers to the equivalent stiffness of a single wheel on the target axle in the lateral direction, reflecting the deformation characteristics of the tire under lateral force.

[0138] The target axle is one of the front and rear axles. The actual longitudinal stiffness and actual lateral stiffness of the front and rear axles are identified independently to accommodate the actual situation where the wear and model of the front and rear axles may be different.

[0139] The kinematic parameters of the third target to be identified change the fastest, and may change significantly in a short period of time as the vehicle travels on different road sections. The equivalent coefficient of adhesion of the road surface refers to the ratio of the maximum friction force that the current road surface can provide to the vertical load of the wheel, reflecting the adhesion performance of the road surface. The equivalent coefficients of adhesion of dry asphalt road surface, wet and slippery road surface and icy and snowy road surface are significantly different.

[0140] By dividing the kinematic parameters to be identified into three levels according to the rate of change of parameter values, the parameter identification model can identify parameters with different time window lengths and update frequencies for different rates of change, thus balancing the identification stability of slowly changing parameters with the tracking response speed of rapidly changing parameters.

[0141] In some embodiments, the kinematic parameters of the first target to be identified are updated in time windows of minutes or longer, the kinematic parameters of the second target to be identified are updated in time windows of seconds, and the kinematic parameters of the third target to be identified are updated in time windows of sub-seconds. During training, the training device constructs training samples for each level of parameters based on the sampling data of different time windows, so that the output heads of each level of parameters can extract features from their respective adapted time scales.

[0142] As another example, in the design of the parameter recognition network, the parameter output head corresponding to the kinematic parameter to be identified for the first target adopts a large receptive field, enabling the network to aggregate information over a longer time range to stably identify slowly changing parameters; the parameter output head corresponding to the kinematic parameter to be identified for the second target adopts a medium receptive field; and the parameter output head corresponding to the kinematic parameter to be identified for the third target adopts a small receptive field, enabling the network to quickly respond to transient changes in the road adhesion coefficient.

[0143] like Figure 3 As shown, this embodiment provides another method for training a vehicle kinematic parameter recognition model, including the following steps: S301. Obtain the timing sequence of the vehicle's operating parameters.

[0144] For details, please refer to S201 above; further details will not be provided here.

[0145] S302. Based on the time series of operating parameters and the inherent kinematic parameters of the vehicle, determine the time series of the estimated hidden state parameters of the vehicle.

[0146] For details, please refer to S202 above; further details will not be provided here.

[0147] S303. Input the time series sequence of the running parameters and the time series sequence of the estimated hidden state parameters into the first parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the first target kinematic parameter to be identified output by the first parameter recognition layer.

[0148] The first parameter identification layer is a network layer in the initial vehicle kinematic parameter identification model specifically used to identify the first target kinematic parameters to be identified. Its identification objects are the first distance from the vehicle's center of mass to the front axle, the yaw angle, the moment of inertia, the air resistance coefficient, the vehicle's frontal area, the rolling resistance coefficient, and the vehicle's braking system response delay, which have the slowest rate of change.

[0149] In one possible implementation, the time sequence of operating parameters and the time sequence of hidden state parameter estimates are input into the first parameter identification layer. The first parameter identification layer extracts features related to the aforementioned slowly varying parameters, such as the correspondence between wheel-end driving force and longitudinal acceleration under linear acceleration conditions, and the correspondence between yaw rate and lateral acceleration under steady-state turning conditions, and outputs the initial identification values ​​of the kinematic parameters to be identified for each first target.

[0150] Since the kinematic parameters of the first target to be identified remain basically unchanged or only change slowly during a single driving process, the first parameter identification layer uses a longer time window to sample the input sequence, and improves the identification stability of slowly changing parameters by aggregating observation information from multiple working conditions over a longer time range.

[0151] As an example, the first parameter recognition layer takes the relationship between the longitudinal velocity of the center of gravity and the driving force at the wheel end under high-speed cruising conditions as its main feature, extracts the combined influence of the air drag coefficient and the frontal area, and obtains the initial recognition value of the air drag coefficient.

[0152] As another example, the first parameter identification layer uses the time difference sequence between the establishment time of braking torque and the deceleration response time under braking conditions as a feature, and statistically analyzes its distribution characteristics to obtain the initial identification value of the braking system response delay.

[0153] S304. Input the time series sequence of the running parameters and the time series sequence of the estimated hidden state parameters into the second parameter recognition layer in the initial vehicle kinematics parameter recognition model to obtain the initial recognition value of the second target kinematic parameters to be identified output by the second parameter recognition layer.

[0154] The second parameter identification layer is a network layer in the initial vehicle kinematics parameter identification model specifically used to identify the kinematic parameters of the second target. Its identification objects are the actual longitudinal stiffness and actual lateral stiffness of the target axis with the changing speed in the middle.

[0155] In one possible implementation, the time sequence of operating parameters and the time sequence of hidden state parameter estimates are input into the second parameter identification layer. The second parameter identification layer extracts features related to tire mechanical characteristics, such as the correspondence between slip ratio and wheel-end driving force, and the correspondence between sideslip angle and yaw rate, and outputs the initial identification values ​​of the actual longitudinal stiffness and actual lateral stiffness of the target shaft.

[0156] Since tire stiffness changes slowly with tire temperature and tire pressure on a scale of tens of seconds to several minutes, the second parameter recognition layer uses a medium-length time window to sample the input sequence, taking into account both the observability of stiffness changes and response speed.

[0157] As an example, the second parameter identification layer uses the slip ratio sequence and the wheel-end driving force sequence as feature inputs to fit the slope of the mapping relationship between slip ratio and longitudinal force, thereby obtaining the initial identification value of the actual longitudinal stiffness. As another example, the second parameter identification layer uses the sideslip angle sequence and the yaw rate sequence as feature inputs to obtain the initial identification value of the actual lateral stiffness through time-series regression.

[0158] S305. Input the time series sequence of the running parameters and the time series sequence of the estimated hidden state parameters into the third parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the kinematic parameters of the third target to be identified output by the third parameter recognition layer.

[0159] The third parameter recognition layer is a network layer in the initial vehicle kinematics parameter recognition model specifically used to identify the kinematic parameters of the third target to be identified. Its identification object is the road surface equivalent adhesion coefficient, which changes the fastest.

[0160] In one possible implementation, the time series of operating parameters and the time series of estimated hidden state parameters are input into the third parameter identification layer. The third parameter identification layer extracts features that are strongly correlated with the road surface adhesion state, such as the rapid increase in slip ratio accompanied by wheel end force saturation and the increase in sideslip angle accompanied by the cessation of lateral force growth. The initial identification value of the road surface equivalent adhesion coefficient is then output.

[0161] Compared to the first and second parameter recognition layers, the road surface equivalent adhesion coefficient may change significantly within seconds when a vehicle enters an icy or waterlogged section of road. Therefore, the third parameter recognition layer uses a short time window to sample the input sequence and gives higher attention to the abrupt changes in the input sequence, so that the initial recognition value of the road surface equivalent adhesion coefficient can quickly follow the changes in road conditions.

[0162] As an example, the third parameter recognition layer takes the combined characteristics of the target wheel's slip ratio change rate and wheel-end driving force as input. When the slip ratio increases rapidly but the wheel-end driving force no longer increases, the initial recognition value of the road surface equivalent adhesion coefficient is corrected towards low adhesion.

[0163] As another example, the third parameter identification layer takes slip ratio, sideslip angle and vertical load sequence from multiple recent moments as input and directly outputs the initial identification value of the road surface equivalent adhesion coefficient through nonlinear mapping.

[0164] S306. Input the initial identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified into the parameter correction layer in the initial vehicle kinematic parameter identification model to obtain the parameter identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified output by the parameter correction layer.

[0165] The parameter correction layer is a network layer in the initial vehicle kinematics parameter recognition model used to uniformly correct the initial recognition values ​​output by the three parameter recognition layers.

[0166] Since the first parameter recognition layer, the second parameter recognition layer, and the third parameter recognition layer output initial recognition values ​​independently, each initial recognition value may deviate from the physically reasonable range due to noise in the input sequence or insufficient coverage of the working condition. The parameter correction layer corrects the three initial recognition values ​​respectively and outputs the corrected results as the parameter recognition values ​​of each kinematic parameter to be identified.

[0167] The parameter identification values ​​output by the parameter correction layer are used as inputs to determine the time sequence of predicted values ​​of the vehicle's state parameters, ensuring that the parameter identification values ​​participating in the forward derivation of the dynamic equivalent model and the tire equivalent model are always within the physically feasible range.

[0168] S307. Based on the time series of hidden state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model, determine the time series of predicted values ​​of the vehicle's state parameters.

[0169] For details, please refer to S204 above; further details will not be provided here.

[0170] S308. Based on the first state prediction deviation between the time series of true state parameter values ​​and the time series of predicted state parameter values, optimize the initial vehicle kinematic parameter identification model to obtain the vehicle kinematic parameter identification model.

[0171] For details, please refer to S205 above; further details will not be provided here.

[0172] In one embodiment, the parameter correction layer is used to correct the initial identification value of any kinematic parameter to be identified among the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified, by using the maximum allowable upper limit and the minimum allowable lower limit of the parameter of any kinematic parameter to be identified, so as to obtain the parameter identification value of any kinematic parameter to be identified.

[0173] Specifically, the maximum allowable upper limit and minimum allowable lower limit of the parameters are the boundaries of the physically feasible intervals preset for each kinematic parameter to be identified based on the vehicle's physical priors. For example, the maximum allowable upper limit and minimum allowable lower limit of the first distance are set according to the vehicle's wheelbase range, and the maximum allowable upper limit and minimum allowable lower limit of the road surface equivalent adhesion coefficient are set according to the dry and wet road surface adhesion performance range.

[0174] For any kinematic parameter to be identified, the parameter correction layer compares the initial identification value of the kinematic parameter to its corresponding physical feasible range. When the initial identification value is within the range, it remains unchanged. When the initial identification value exceeds the range, it is corrected to the vicinity of the range boundary. This ensures that the output parameter identification value is always within the physical feasible range, and avoids abnormal identification values ​​from entering the subsequent dynamic equivalent model and tire equivalent model, which would cause distortion in state prediction.

[0175] As an example, the parameter correction layer uses a linear saturation method to correct the initial identification value. When the initial identification value is less than the minimum allowable lower limit, the parameter identification value is taken as the minimum allowable lower limit. When the initial identification value is greater than the maximum allowable upper limit, the parameter identification value is taken as the maximum allowable upper limit. When the initial identification value is in between, it remains unchanged.

[0176] As another example, the parameter correction layer uses a bounded mapping function to correct the initial identification values. The initial identification values ​​are input into a differentiable function whose range is consistent with the physically feasible interval, and the function output is the parameter identification value. This method maintains differentiability throughout the correction process, allowing the gradients during the training phase to be smoothly backpropagated to each parameter identification layer via the parameter correction layer, avoiding gradient vanishing at interval boundaries, which is beneficial to the convergence of model training.

[0177] For example, the maximum permissible upper limit value of the road surface equivalent adhesion coefficient in the kinematic parameters to be identified for the third target is 1.2 and the minimum permissible lower limit value is 0.1, and the maximum permissible upper limit value of the air resistance coefficient in the kinematic parameters to be identified for the first target is 1.0 and the minimum permissible lower limit value is 0.2, so as to cover the parameter value range under common vehicle and road surface conditions.

[0178] like Figure 4 As shown, this embodiment provides another method for training a vehicle kinematic parameter recognition model, including the following steps: S401. Obtain the timing sequence of the vehicle's operating parameters.

[0179] For details, please refer to S201 above; further details will not be provided here.

[0180] S402. Based on the time series of operating parameters and the inherent kinematic parameters of the vehicle, determine the time series of the estimated hidden state parameters of the vehicle.

[0181] For details, please refer to S202 above; further details will not be provided here.

[0182] S403. Based on the time series sequence of operating parameters, the time series sequence of hidden state parameter estimates, and the initial vehicle kinematic parameter identification model, determine the parameter identification values ​​of the vehicle's kinematic parameters to be identified.

[0183] For details, please refer to S203 above; further details will not be provided here.

[0184] S404. Based on the time series of hidden state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, and the tire equivalent model, determine the longitudinal force and lateral force of the target wheel in its own coordinate system.

[0185] The tire equivalent model is a mathematical model used to characterize the relationship between the tire force generated by the tire and the tire deformation state and road conditions. Specifically, it represents two sets of correspondences: one set is the correspondence between the longitudinal force of the target wheel in its own wheel coordinate system, the actual longitudinal stiffness of the target wheel, the slip ratio of the target wheel, the vertical load of the target wheel, and the equivalent adhesion coefficient of the road surface; the other set is the correspondence between the lateral force of the target wheel in its own wheel coordinate system, the actual lateral stiffness of the target wheel, the sideslip angle of the target wheel, the vertical load of the target wheel, and the equivalent adhesion coefficient of the road surface.

[0186] The aforementioned self-wheel coordinate system is a coordinate system constructed with the center of the target wheel as the origin, the longitudinal axis along the rolling direction of the wheel, and the lateral axis perpendicular to the rolling direction of the wheel.

[0187] The actual longitudinal stiffness and actual lateral stiffness of the target wheel are determined based on the actual longitudinal stiffness and actual lateral stiffness of the axle on which the target wheel is located, as well as the ratio of the vertical loads on the two wheels on the axle on which the target wheel is located.

[0188] Specifically, the actual longitudinal stiffness of the target axle is distributed to the target wheel according to the proportion of the vertical load of the target wheel to the total vertical load of the axle, thus obtaining the actual longitudinal stiffness of the target wheel. The actual lateral stiffness is obtained in the same way, so that the stiffness of each wheel under different load conditions matches its respective load state.

[0189] In one possible implementation, the slip ratio and actual longitudinal stiffness of the target wheel are input into the tire equivalent model to obtain the original longitudinal force of the target wheel, which is proportional to both the actual longitudinal stiffness and slip ratio; the sideslip angle and actual lateral stiffness of the target wheel are input into the tire equivalent model to obtain the original lateral force of the target wheel.

[0190] Based on this, the tire equivalent model uses the product of the road surface equivalent adhesion coefficient and the vertical load of the target wheel as the maximum friction force available to the target wheel. When the resultant force of the original longitudinal force and the original lateral force exceeds the maximum friction force, the original longitudinal force and the original lateral force are proportionally corrected so that the corrected resultant force falls on the boundary defined by the maximum friction force, thus obtaining the longitudinal force and lateral force of the target wheel in its own coordinate system.

[0191] As an example, the tire equivalent model determines the longitudinal force and lateral force independently. When the slip ratio or sideslip angle is small, the tire force and deformation maintain a linear relationship. When the resultant force is close to the maximum friction force, the tire force enters the saturation zone through amplitude limiting correction, reflecting the physical characteristics of the tire's adhesion reaching its limit under large slip or large sideslip conditions.

[0192] As another example, the tire equivalent model uses a smooth saturation function to continuously transition between the linear segment and the saturation segment, so that the longitudinal force and lateral force remain continuously differentiable throughout the slip ratio and sideslip angle range, which is beneficial for the stable backpropagation of gradients during the training phase.

[0193] S405. Based on the longitudinal and lateral forces of the target wheel in its own coordinate system, the time series of estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, and the dynamic equivalent model, determine the time series of predicted state parameters of the vehicle.

[0194] Among them, the dynamic equivalent model is used to characterize the longitudinal and lateral forces of the target wheel in its own coordinate system, the parameter values ​​of the hidden state parameters, the parameter values ​​of the kinematic parameters to be identified, and the correspondence between the state parameters.

[0195] Specifically, in this step, based on the longitudinal and lateral forces of the target wheel in its own coordinate system, combined with the time series of the estimated hidden state parameters and the parameter identification values ​​of the kinematic parameters to be identified, the state parameters of the vehicle are recursively deduced time by time according to the correspondence represented by the dynamic equivalent model, so as to obtain the time series of the predicted state parameters.

[0196] In the recursive process at a single moment, the longitudinal and lateral forces of each target wheel in its own wheel coordinate system are first transformed by a rotation matrix between the wheel coordinate system and the vehicle coordinate system to obtain the longitudinal and lateral forces of each target wheel in the vehicle coordinate system. Then, the parameter identification values ​​of each kinematic parameter to be identified are substituted into the corresponding relationship in which they participate. Among them, the parameter identification value of the first distance from the center of mass to the front axle is used to determine the longitudinal and lateral coordinates of each target wheel in the vehicle coordinate system, the parameter identification value of the yaw angle moment of inertia is used for yaw moment balance, the parameter identification values ​​of the air resistance coefficient and the frontal area are used to determine the air resistance, and the parameter identification value of the rolling resistance coefficient is used to determine the rolling resistance. Then, based on the correspondence of the longitudinal force balance, lateral force balance and yaw moment balance represented by the dynamic equivalent model, the longitudinal acceleration, lateral acceleration and yaw angle acceleration of the center of mass are determined, and based on the correspondence of the wheel rotation balance, the wheel angular acceleration of the target wheel is determined.

[0197] After obtaining each acceleration, the longitudinal acceleration, lateral acceleration, yaw acceleration and wheel angular acceleration of the center of mass are integrated at sampling intervals based on the state parameters of the previous moment to obtain the longitudinal velocity, lateral velocity, yaw velocity and wheel angular velocity of the target wheel at the next moment, which are used as the predicted values ​​of the state parameters at the next moment.

[0198] Starting from the true state parameter value at the beginning of the time series of running parameters, the above transformation, solution and integration process is executed step by step along the time axis until the end of the time series is reached, so as to obtain the predicted state parameter value at each time and form the time series of predicted state parameter value.

[0199] As an example, the explicit Euler method is used to perform the above integration, where the product of acceleration and sampling interval is used as the increment of the state parameter. With a sampling frequency of 100Hz, the sampling interval is 0.01 seconds, resulting in low computational cost per step, suitable for large-scale recursion during batch training. As another example, the training device uses the fourth-order Runge-Kutta method to perform the above integration, calculating and weighting the acceleration multiple times within a single sampling interval, maintaining high integration accuracy even under conditions with large sampling intervals or drastic dynamic changes.

[0200] Since all the correspondences in the above recursive link are analytically differentiable expressions, the gradient of the deviation of the state parameter prediction value at any time in the time series can be propagated back along the recursive link to the parameter identification value of each kinematic parameter to be identified and the model parameters of the initial vehicle kinematic parameter identification model. Furthermore, by continuously recursively proposing multiple times from the state parameter prediction value at a certain time, the vehicle state prediction for multiple future times can be obtained, which provides a basis for subsequent optimization based on the cumulative prediction deviation.

[0201] S406. Based on the first state prediction deviation between the time series of true state parameter values ​​and the time series of predicted state parameter values, optimize the initial vehicle kinematic parameter identification model to obtain the vehicle kinematic parameter identification model.

[0202] For details, please refer to S205 above; further details will not be provided here.

[0203] In one embodiment, the above-described dynamic equivalent model is specifically used to characterize: The relationship between vehicle mass, longitudinal acceleration of center of gravity, lateral velocity, yaw rate, resultant longitudinal force of each wheel in the vehicle coordinate system, air resistance, and rolling resistance. In addition, the corresponding relationships between vehicle mass, longitudinal velocity, lateral acceleration of the center of mass, yaw rate, and the resultant lateral forces of each wheel of the vehicle in the vehicle coordinate system. In addition, the corresponding relationships between the moment of inertia of the target wheel, the wheel angular acceleration of the target wheel, the wheel-end driving torque of the target wheel, the wheel-end braking torque of the wheel, the effective rolling radius of the wheel, and the longitudinal force of the target wheel in its own wheel coordinate system. In addition, the corresponding relationships between the longitudinal coordinates of the target wheel in the vehicle coordinate system, the lateral coordinates of the target wheel in the vehicle coordinate system, the lateral force of the target wheel in the vehicle coordinate system, the longitudinal force of the target wheel in the vehicle coordinate system, the yaw angle moment of inertia and the yaw angle acceleration. In addition, the correspondence between the lateral velocity, lateral velocity, and yaw rate of the target wheel in the vehicle coordinate system and the longitudinal coordinate of the target wheel in the vehicle coordinate system; And the correspondence between the longitudinal velocity, longitudinal velocity, yaw rate of the target wheel in the vehicle coordinate system and the lateral coordinate of the target wheel in the vehicle coordinate system; In addition, the correspondence between the lateral force of the target wheel in the vehicle coordinate system, the longitudinal force of the target wheel in the vehicle coordinate system, the lateral force of the target wheel in its own wheel coordinate system, the longitudinal force of the target wheel in its own wheel coordinate system, and the rotation matrix between the wheel coordinate system and the vehicle coordinate system. Among them, wheel angular acceleration is the derivative of the target wheel's wheel angular velocity; yaw acceleration is the derivative of the yaw velocity; longitudinal acceleration is the derivative of the longitudinal velocity; lateral acceleration is the derivative of the longitudinal velocity; air resistance is determined based on the air resistance coefficient and the frontal area; rolling resistance is determined based on the rolling resistance coefficient; longitudinal and lateral coordinates are determined based on the first distance; the rotation matrix is ​​determined by the target wheel's rotation angle; the vehicle coordinate system is constructed with the center of mass as the origin and the directions of the vehicle's longitudinal and lateral forces as coordinate axes.

[0204] Specifically, the dynamic equivalent model is a vehicle dynamics model constructed in the form of a set of equations, which is used to correlate the wheel-end forces and wheel-end torques of each wheel with the motion state of the vehicle's center of mass.

[0205] In this application, the vehicle coordinate system is a coordinate system constructed with the vehicle's center of mass as the origin and the vehicle's longitudinal and lateral axes as coordinate directions.

[0206] The longitudinal and lateral coordinates of the target wheel in the vehicle coordinate system are determined based on a first distance: when the target wheel is located on the front axle, the longitudinal coordinate of the target wheel is the first distance; when the target wheel is located on the rear axle, the longitudinal coordinate of the target wheel is the difference between the wheelbase and the first distance, where the wheelbase is the distance between the front and rear axles. The lateral coordinates of the target wheel are determined based on the track width of the axle on which the target wheel is located. For example, the lateral coordinate of a wheel located on the left half of the axle is half the track width.

[0207] For ease of description, remember For longitudinal velocity, r is the lateral velocity, and r is the yaw rate. Let i be the wheel angular velocity of the i-th wheel. and These represent the longitudinal force and lateral force of the i-th wheel in the vehicle coordinate system. and These represent the longitudinal force and lateral force of the i-th wheel in its own wheel coordinate system, respectively.

[0208] As an example, the vehicle's mass, longitudinal acceleration of the center of gravity, lateral velocity, yaw rate, resultant longitudinal force of each wheel in the vehicle's coordinate system, air resistance, and rolling resistance satisfy the following relationships: ; In the formula, m represents the vehicle mass. This represents the longitudinal acceleration of the center of mass. Indicates air resistance, This represents rolling resistance. The left side of the equation represents the longitudinal inertial force of the vehicle due to yaw motion, and the right side represents the resultant longitudinal force of all wheels after deducting air resistance and rolling resistance.

[0209] The vehicle's mass, longitudinal velocity, lateral acceleration of its center of mass, yaw rate, and the resultant lateral forces of each wheel in the vehicle's coordinate system satisfy the following relationship: ; In the formula, This represents the lateral acceleration of the center of mass.

[0210] In the moment balance in the yaw direction, the longitudinal coordinates, lateral coordinates, lateral forces, longitudinal forces, yaw angle moments of inertia, and yaw angle accelerations of each wheel in the vehicle coordinate system satisfy the following relationships: ; In the formula, Indicates the moment of inertia of the yaw angle. Indicates yaw acceleration. and These represent the longitudinal and lateral coordinates of the i-th wheel in the vehicle coordinate system, respectively, meaning that the lateral and longitudinal forces of each wheel generate yaw moments around the center of mass.

[0211] For each target wheel, the moment of inertia, wheel angular acceleration, wheel-end driving torque, wheel-end braking torque, effective rolling radius, and longitudinal force of the target wheel in its own wheel coordinate system satisfy the following relationships: ; In the formula, This represents the moment of inertia of the target wheel. This represents the wheel angular acceleration of the target wheel. and These represent the wheel-end driving torque and wheel-end braking torque of the target wheel, respectively. This indicates the effective rolling radius of the wheel, which is the combined force of the wheel end torque and the longitudinal ground force after being converted through the effective rolling radius to drive the wheel to rotate.

[0212] The lateral velocity, yaw rate, and longitudinal coordinate of the target wheel at its wheel center in the vehicle coordinate system satisfy the following relationships: The longitudinal velocity, yaw rate, and lateral coordinate of the target wheel at its wheel center in the vehicle coordinate system satisfy the following relationships: ; In the formula, and These represent the lateral velocity and longitudinal velocity of the i-th wheel center in the vehicle coordinate system, respectively, which are the superposition of the center of mass velocity and the yaw motion-induced velocity at the target wheel mounting position.

[0213] The longitudinal and lateral forces of the target wheel in the vehicle coordinate system are transformed with those in its own wheel coordinate system through a rotation matrix, satisfying the following relationship: ; In the formula, This represents the turning angle of the i-th wheel. The rotation matrix is ​​determined by this turning angle and is used to project the tire force in its own wheel coordinate system to the vehicle coordinate system.

[0214] Air resistance is determined based on the drag coefficient and frontal area, and satisfies the following relationship: ; In the formula, Indicates air density, Indicates the air drag coefficient. Indicates the windward area. This indicates the relative velocity of the air.

[0215] Rolling resistance is determined based on the rolling resistance coefficient and satisfies the following relationship: ; In the formula, Represents gravitational acceleration. Indicates the rolling resistance coefficient. This indicates a smoothing scale used to avoid low-speed numerical jitter.

[0216] As can be seen, the air drag coefficient, frontal area, and rolling drag coefficient are the first target kinematic parameters to be identified. Substituting these three parameters into the above relationship with their parameter identification values ​​directly affects the accuracy of the predicted state parameters.

[0217] As an example, when the target wheel is a rear axle wheel, the rotation angle of the target wheel is zero. At this time, the rotation matrix is ​​an identity matrix, and the longitudinal and lateral forces of the target wheel in its own wheel coordinate system are equal to the longitudinal and lateral forces in the vehicle coordinate system, respectively, so no coordinate rotation is required.

[0218] As another example, the air density is taken as 1.2 kg / m³, and the relative air velocity is approximated by the longitudinal velocity when wind speed information is unavailable; the smoothness scale is taken as 1 m / s, so that the rolling resistance of the vehicle smoothly approaches zero when it is traveling at low speed.

[0219] The above seven sets of correspondences are all analytically differentiable expressions, which enables the gradient determined based on the prediction deviation of the first state in step S505 to be back-propagated layer by layer to each kinematic parameter to be identified through the dynamic equivalent model, ensuring that the initial vehicle kinematic parameter identification model can be trained end-to-end.

[0220] like Figure 5As shown, this embodiment provides another method for training a vehicle kinematic parameter recognition model, including the following steps: S501. Obtain the timing sequence of the vehicle's operating parameters.

[0221] For details, please refer to S201 above; further details will not be provided here.

[0222] S502. Based on the time series of operating parameters and the inherent kinematic parameters of the vehicle, determine the time series of the estimated hidden state parameters of the vehicle.

[0223] For details, please refer to S202 above; further details will not be provided here.

[0224] S503. Based on the time series sequence of operating parameters, the time series sequence of hidden state parameter estimates, and the initial vehicle kinematic parameter identification model, determine the parameter identification values ​​of the vehicle's kinematic parameters to be identified.

[0225] For details, please refer to S203 above; further details will not be provided here.

[0226] S504. Based on the time series of hidden state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model, determine the time series of predicted state parameters of the vehicle.

[0227] For details, please refer to S204 above; further details will not be provided here.

[0228] S505. Determine the loss function based on at least one of the first state prediction bias and the target bias.

[0229] The loss function is determined based on at least one of the first state prediction bias and the target bias, wherein the first state prediction bias reflects the overall prediction bias between the time series of predicted state parameter values ​​and the time series of true state parameter values, and the target bias applies auxiliary constraints to the parameter identification values ​​from the constraint dimensions of parameter supervision, physical consistency, temporal smoothing and multi-step prediction.

[0230] Optionally, the above "at least one" covers two cases: determining the loss function based solely on the first state prediction bias; or determining the loss function based on a weighted combination of at least one of the first state prediction bias and the target bias.

[0231] As an example, the loss function satisfies the following relationship: ; In the formula, Lx represents the state prediction loss determined based on the first state prediction deviation, Lk represents the loss term determined based on the k-th deviation in the target deviation, and λx and λk represent the weight coefficients of each term. Any loss term can be turned off by configuring its weight coefficient to zero, thereby achieving various value scenarios for at least one of the above.

[0232] In some embodiments, the state prediction loss satisfies the following relationship: ; In the formula, N represents the length of the time series, and nx represents the number of state parameters. t represents the predicted value of the state parameter at time t, xt represents the actual value of the state parameter at time t, and Wx represents the dimensional balance weight matrix. Since the longitudinal velocity of the center of mass, the lateral velocity of the center of mass, the yaw rate, and the wheel angular velocity of the target wheel have different dimensions and numerical ranges, Wx is used to normalize the deviations of each state parameter to a comparable order of magnitude.

[0233] For the first identification bias in the target bias, in scenarios where the true values ​​of the kinematic parameters to be identified can be obtained (e.g., bench calibration or field testing scenarios), the training device compares the identified parameter values ​​with the true parameter values ​​time-by-time to obtain a first identification bias loss term similar to the state prediction loss mentioned above, so that the identified parameter values ​​directly approximate the true parameter values ​​within the data segment with truth-supervised data. For the parameter correction bias, the training device compares the identified parameter values ​​with the initial identified values, penalizing large corrections in the parameter correction layer, preventing boundary corrections from obscuring the hierarchical identification results of each parameter identification layer, and maintaining consistency between the output of the parameter correction layer and the output of each parameter identification layer.

[0234] Regarding the second identification bias, the true target force is determined based on the wheel-end driving torque, wheel-end braking torque, and effective rolling radius of the target wheel. The training device compares the longitudinal or lateral force of the target wheel in its own coordinate system output by the tire equivalent model with the corresponding true target force, thereby directly supervising the tire force calculation process.

[0235] For the second state prediction bias, the training device substitutes the time series of the true values ​​of the hidden state parameters and the true values ​​of the kinematic parameters to be identified into the equivalent dynamic model and the equivalent tire model, and recursively obtains the time series of the theoretical values ​​of the state parameters, comparing them with the time series of the true values ​​of the state parameters step by step. This bias measures the residual level of the model structure itself. Incorporating it into the loss function can prevent the parameter identification values ​​from overcompensating for model structure errors during parameter optimization, thereby improving the physical reliability of the parameter identification values.

[0236] As an example, the weighting coefficients are configured as follows: the weights of the state prediction loss, the first identification bias loss, and the second state prediction bias loss are all 1; the weights of the second identification bias loss and the cumulative prediction bias loss are 0.5; the total weight of the parameter smoothing bias loss is 0.1; and the smoothing weights of the kinematic parameters to be identified at each level are 1, 0.3, and 0.05 respectively for the first, second, and third target kinematic parameters to be identified.

[0237] The kinematic parameters of the first target to be identified are physically close to constant, so a strong smoothing constraint is applied; the kinematic parameters of the third target to be identified change rapidly with road conditions, so a weaker smoothing constraint is applied, allowing it to quickly track changes in road conditions.

[0238] As another example, on public road data without parameter truth labels, the weights of the first identification bias loss and the second state prediction bias loss are set to zero, and the loss function degenerates into a weighted combination of state prediction loss, parameter smoothing bias loss and cumulative prediction bias loss.

[0239] S506. Based on the loss function, optimize the model parameters of the initial vehicle kinematics parameter recognition model to obtain the vehicle kinematics parameter recognition model.

[0240] The target deviation includes: The first identification deviation between the identified value and the true value of the kinematic parameter to be identified; The parameter correction deviation between the identified values ​​and the initial identified values ​​of the kinematic parameters to be identified; The second identification deviation between the target torque of the target wheel in its own coordinate system and the actual target torque of the target wheel in its own coordinate system; the target torque is one of the longitudinal force and the lateral force; The second state prediction deviation between the time series of theoretical values ​​of vehicle state parameters and the time series of actual values ​​of state parameters; wherein, the time series of theoretical values ​​of state parameters is determined based on the time series of actual values ​​of hidden state parameters, the actual values ​​of the kinematic parameters to be identified, the dynamic equivalent model, and the tire equivalent model; The parameter smoothing deviation between the parameter identification value of the kinematic parameter to be identified and the parameter identification value of the kinematic parameter to be identified at the previous time step; The cumulative prediction bias of the identified kinematic parameters for predicting vehicle state at multiple future moments.

[0241] Specifically, the model parameters of the initial vehicle kinematics parameter identification model include the weights and biases of each network layer in the first, second, and third parameter identification layers; the maximum allowable upper limit and minimum allowable lower limit of each kinematic parameter to be identified in the parameter correction layer are pre-configured hyperparameters, which remain fixed in this step and do not participate in optimization.

[0242] The training device is based on the loss function and calculates the gradient of the loss function with respect to each model parameter through the backpropagation algorithm.

[0243] Since the time-series sequence of state parameter predictions is obtained by recursively deriving parameter identification values ​​from the tire equivalent model and the dynamic equivalent model, and the correspondences between the two sets of models are analytically differentiable expressions, the gradient can be propagated back layer by layer along the link of "first state prediction bias - dynamic equivalent model - tire equivalent model - parameter identification value - output head of each parameter identification layer - internal weights of each parameter identification layer". The training device updates the model parameters in the opposite direction of the gradient, and the update of the model parameters reduces the loss function by changing the parameter identification values.

[0244] As an example, the training device employs the Adaptive Moment Estimation (Adam) optimizer to perform parameter updates, with a learning rate of 1e-3, a batch size of 64, and a weight decay of 1e-5. Gradient norms are also pruned to avoid aberrant gradient amplification caused by isolated outliers. The training device iteratively performs forward computation, loss function determination, and model parameter updates. Training terminates when the change in the loss function is less than a preset threshold or the maximum number of iterations is reached. The initial vehicle kinematics parameter recognition model at this point is then determined as the vehicle kinematics parameter recognition model.

[0245] This embodiment provides a method for identifying vehicle kinematic parameters. This method utilizes the vehicle kinematic parameter identification model trained in the above embodiment to perform online identification of kinematic parameters to be identified during vehicle operation. It can be executed by a controller on the vehicle (e.g., the electronic control unit of a vehicle stability control system). Figure 6 As shown, the method includes the following steps.

[0246] S601. Obtain the current operating parameter time sequence of the vehicle.

[0247] The current operating parameter time series includes a time-aligned time series of the current state parameter true values ​​and a time series of the current observation parameter true values.

[0248] The state parameters in the current state parameter true value time series include the longitudinal velocity of the center of mass, the lateral velocity of the center of mass, the yaw rate, and the wheel angular velocity of the target wheel, which are measured by the integrated navigation device or vehicle speed sensor, inertial measurement unit, and wheel speed sensor, respectively. The observation parameters in the current observation parameter true value time series include the steering wheel angle, the driving force and braking force provided by the power system, which are obtained by the steering wheel angle sensor, the output signal of the power system controller, and the pressure signal of the braking system, respectively.

[0249] The controller timestamps the data collected by each sensor according to a unified clock and interpolates and aligns signals with different sampling frequencies to obtain a time-aligned sequence of current operating parameters.

[0250] As an example, the controller continuously collects the above data at a fixed sampling frequency and extracts data from the most recent period according to the same time window configuration as the training phase (e.g., 500 sampling points for a long window, 100 sampling points for a medium window, and 30 sampling points for a short window). As the vehicle moves, the controller continuously updates the data by the sliding step size, so that the input window of each parameter recognition layer remains consistent with the training phase.

[0251] For example, the sliding step size can be, for instance, 10 sampling points.

[0252] S602. Based on the current operating parameter time series and the vehicle's inherent kinematic parameters, determine the current hidden state parameter estimate time series of the vehicle.

[0253] Among them, the inherent kinematic parameters include vehicle mass, wheelbase, center of gravity height, track width, effective rolling radius of the wheels, and steering system transmission ratio, which are parameters known from the vehicle's manufacture.

[0254] The specific implementation method for determining the time series of estimated values ​​of the current hidden state parameters based on the time series of current operating parameters and inherent kinematic parameters is the same as the implementation method for determining the time series of estimated values ​​of hidden state parameters in the above training method embodiment. For example, it includes the conversion of the wheel angle of the target wheel, the determination of the wheel center speed, the calculation of slip ratio and sideslip angle, and the estimation of vertical load, etc., which will not be elaborated here.

[0255] S603. Based on the current operating parameter time series, the current hidden state parameter estimation time series, and the vehicle kinematic parameter identification model, determine the current parameter identification value of the vehicle's kinematic parameters to be identified.

[0256] The vehicle kinematic parameter identification model is trained using the training method described in any of the above embodiments. The controller inputs the current operating parameter time series and the current hidden state parameter estimate time series into the vehicle kinematic parameter identification model. After one forward computation through the first parameter identification layer, the second parameter identification layer, the third parameter identification layer, and the parameter correction layer, it outputs the current parameter identification values ​​for the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified. Unlike the training phase, this step does not require recursively deriving state parameter prediction values ​​based on the dynamic equivalent model and the tire equivalent model, nor does it require calculating the loss function and gradient; it only performs forward computation, significantly reducing the computational load.

[0257] As an example, the controller performs the above forward calculation once every sliding step, and can output the latest current parameter identification value within a control cycle of hundreds of milliseconds, which meets the real-time requirements of vehicle dynamics control.

[0258] As another example, when a vehicle moves from a dry road surface to a wet road surface, the current parameter identification value of the third target kinematic parameter (i.e., the road equivalent adhesion coefficient) decreases rapidly as the short time window slides, and the vehicle stability control system adjusts the braking force distribution and driving force threshold accordingly. When the vehicle load changes (e.g., changes in the number of occupants or cargo load), the current parameter identification value of the first target kinematic parameter is updated slowly as the long time window slides, reflecting the actual change in the center of gravity position. Thus, the identified current parameter identification value can serve as a real-time input for downstream functions such as vehicle dynamics control and state estimation, improving the vehicle's adaptive capability under different operating conditions.

[0259] like Figure 7 As shown in the figure, this application provides a training device for a vehicle kinematic parameter recognition model, including: an acquisition module 701, a first determination module 702, a second determination module 703, a third determination module 704, and an optimization module 705.

[0260] The acquisition module 701 is used to acquire the time series sequence of the vehicle's operating parameters; wherein, the time series sequence of operating parameters includes the time series sequence of the real values ​​of the time-aligned state parameters and the time series sequence of the real values ​​of the observation parameters; The first determining module 702 is used to determine the time series of the hidden state parameter estimates of the vehicle based on the time series of operating parameters and the inherent kinematic parameters of the vehicle. The second determining module 703 is used to determine the parameter identification value of the vehicle's kinematic parameters to be identified based on the time series sequence of operating parameters, the time series sequence of hidden state parameter estimates, and the initial vehicle kinematic parameter identification model. The third determining module 704 is used to determine the time series of predicted values ​​of the vehicle's state parameters based on the time series of the estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model; the tire equivalent model is coupled with the vehicle's dynamic equivalent model. The optimization module 705 is used to optimize the initial vehicle kinematic parameter identification model based on the first state prediction deviation between the time series sequence of the true state parameter values ​​and the time series sequence of the predicted state parameter values, thereby obtaining the vehicle kinematic parameter identification model.

[0261] The state parameters in the above-mentioned true value time series of state parameters include: longitudinal velocity of the center of mass, lateral velocity of the center of mass, yaw rate, and wheel angular velocity of the target wheel; the target wheel is one of the wheels on the vehicle. The observed parameters in the observed parameter true value time series include: steering wheel angle, driving force and braking force provided by the power system. The hidden state parameters in the hidden state parameter estimation time series include: wheel angle of the target wheel, wheel center velocity, slip ratio, sideslip angle, wheel-end driving force, wheel-end braking force, and vertical load. The kinematic parameters to be identified include: classified according to the rate of change of parameter values. The first target kinematic parameters to be identified are: a first target kinematic parameters to be identified, a second target kinematic parameters to be identified, and a third target kinematic parameters to be identified. The first target kinematic parameters to be identified include: the first distance from the vehicle's center of mass to the front axle, the yaw angle moment of inertia, the air drag coefficient, the vehicle's frontal area, the rolling resistance coefficient, and the vehicle's braking system response delay. The second target kinematic parameters to be identified include: the actual longitudinal stiffness and actual lateral stiffness of the target axle of the vehicle; the target axle is one of the front axle and the rear axle of the vehicle. The third target kinematic parameters to be identified include: the road surface equivalent adhesion coefficient.

[0262] The second determining module 703 described above is also used to input the time sequence of the running parameters and the time sequence of the estimated hidden state parameters into the first parameter recognition layer in the initial vehicle kinematics parameter recognition model to obtain the initial recognition value of the first target kinematic parameter to be identified output by the first parameter recognition layer; The time series sequence of running parameters and the time series sequence of hidden state parameter estimates are input into the second parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the second target kinematic parameter to be identified output by the first parameter recognition layer. The time series sequence of running parameters and the time series sequence of hidden state parameter estimates are input into the third parameter recognition layer in the initial vehicle kinematic parameter recognition model to obtain the initial recognition value of the third target kinematic parameter to be identified output by the first parameter recognition layer. The initial identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified are input into the parameter correction layer in the initial vehicle kinematic parameter identification model to obtain the parameter identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified output by the parameter correction layer.

[0263] The aforementioned parameter correction layer is used to correct the initial identification value of any kinematic parameter to be identified among the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified, by using the maximum allowable upper limit and the minimum allowable lower limit of the parameter of any kinematic parameter to be identified, so as to obtain the parameter identification value of any kinematic parameter to be identified.

[0264] The second determining module 703 is further configured to determine the longitudinal and lateral forces of the target wheel in its own coordinate system based on the time series of latent state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, and the tire equivalent model. The tire equivalent model characterizes the relationships between: the longitudinal force of the target wheel in its own wheel coordinate system, the actual longitudinal stiffness of the target wheel, the slip ratio of the target wheel, the vertical load of the target wheel, and the equivalent road surface adhesion coefficient; and the relationships between: the lateral force of the target wheel in its own wheel coordinate system, the actual lateral stiffness of the target wheel, the sideslip angle of the target wheel, the vertical load of the target wheel, and the equivalent road surface adhesion coefficient. The actual longitudinal stiffness and actual lateral stiffness of the target wheel are determined based on the actual longitudinal stiffness and actual lateral stiffness of the axle on which the target wheel is located, as well as the vertical load ratio of the two wheels on the axle. Based on the longitudinal and lateral forces of the target wheel in its own coordinate system, the time series of estimated values ​​of hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, and the dynamic equivalent model, the time series of predicted values ​​of the vehicle's state parameters are determined. Among them, the dynamic equivalent model is used to characterize the correspondence between the longitudinal and lateral forces of the target wheel in its own coordinate system, the parameter values ​​of hidden state parameters, the parameter values ​​of the kinematic parameters to be identified, and the state parameters.

[0265] The aforementioned equivalent dynamic model is specifically used to characterize: the vehicle's mass, longitudinal acceleration of the center of mass, lateral velocity, yaw rate, the resultant longitudinal force of each wheel in the vehicle's coordinate system, air resistance, and rolling resistance; the vehicle's mass, longitudinal velocity, lateral acceleration of the center of mass, yaw rate, and the resultant lateral force of each wheel in the vehicle's coordinate system; the target wheel's moment of inertia, wheel angular acceleration, wheel-end driving torque, wheel-end braking torque, effective rolling radius, and longitudinal force in its own wheel coordinate system; the target wheel's longitudinal coordinate in the vehicle's coordinate system, lateral coordinate in the vehicle's coordinate system, lateral force in the vehicle's coordinate system, longitudinal force, yaw rate, moment of inertia, and yaw rate in the vehicle's coordinate system; and the target wheel's wheel-center lateral velocity, lateral velocity, yaw rate, and the resultant lateral force of each wheel in the vehicle's coordinate system. The correspondence between the longitudinal coordinates; and the correspondence between the longitudinal velocity, longitudinal velocity, yaw rate, and lateral coordinates of the target wheel in the vehicle coordinate system; and the correspondence between the lateral force, longitudinal force, lateral force, and longitudinal force of the target wheel in its own wheel coordinate system, and the rotation matrix between the wheel coordinate system and the vehicle coordinate system; wherein, wheel angular acceleration is the derivative of the wheel angular velocity of the target wheel; yaw rate is the derivative of the yaw rate; longitudinal acceleration is the derivative of the longitudinal velocity; lateral acceleration is the derivative of the longitudinal velocity; air resistance is determined based on the air resistance coefficient and the frontal area; rolling resistance is determined based on the rolling resistance coefficient; longitudinal and lateral coordinates are determined based on the first distance; the rotation matrix is ​​determined by the rotation angle of the target wheel; the vehicle coordinate system is constructed with the center of mass as the origin and the directions of the longitudinal and lateral forces of the vehicle as the coordinate axes.

[0266] The third determining module 704 is further configured to determine a loss function based on at least one of the first state prediction deviation and the target deviation; and optimize the model parameters of the initial vehicle kinematic parameter identification model based on the loss function to obtain the vehicle kinematic parameter identification model; wherein the target deviation includes: a first identification deviation between the parameter identification value and the true value of the kinematic parameter to be identified; a parameter correction deviation between the parameter identification value and the initial identification value of the kinematic parameter to be identified; and a second identification deviation between the target torque of the target wheel in its own coordinate system and the true target torque of the target wheel in its own coordinate system. The target torque is one of the longitudinal force and the lateral force; the second state prediction deviation between the time series of theoretical values ​​of the vehicle's state parameters and the time series of actual values ​​of the state parameters; wherein, the time series of theoretical values ​​of the state parameters is determined based on the time series of actual values ​​of the hidden state parameters, the actual values ​​of the kinematic parameters to be identified, the dynamic equivalent model, and the tire equivalent model; the parameter smoothing deviation between the parameter identification value of the kinematic parameters to be identified and the parameter identification value of the kinematic parameters to be identified at the previous moment; and the cumulative prediction deviation of the parameter identification value of the kinematic parameters to be identified for predicting the vehicle's state at multiple future moments.

[0267] This application provides a vehicle kinematic parameter identification device. This device is used to perform the steps in the above-described vehicle kinematic parameter identification method embodiments. It can be a controller independently installed on the vehicle, or it can be integrated into the electronic control unit or vehicle domain controller of the vehicle stability control system. Figure 8 As shown, the vehicle kinematic parameter identification device includes an acquisition module 801, a first determination module 802, and a second determination module 803.

[0268] The acquisition module 801 is used to acquire the time series sequence of the vehicle's current operating parameters. This time series sequence includes a time-aligned time series sequence of the actual values ​​of the current state parameters and a time series sequence sequence of the actual values ​​of the current observed parameters.

[0269] The acquisition module 801 communicates with the vehicle speed sensor, inertial measurement unit, wheel speed sensor, steering wheel angle sensor, power system controller, and braking system pressure sensor on the vehicle. It timestamps the data collected by each sensor according to a unified clock and interpolates and aligns signals with different sampling frequencies. It continuously captures data within the most recent period according to the same time window configuration as the training phase to obtain the current operating parameter time sequence.

[0270] The first determining module 802 is used to determine the time series of estimated values ​​of the vehicle's current hidden state parameters based on the time series of the current operating parameters and the vehicle's inherent kinematic parameters.

[0271] Among them, the inherent kinematic parameters include vehicle mass, wheelbase, center of gravity height, track width, effective rolling radius of the wheels, and steering system transmission ratio, which are parameters known from the vehicle's manufacture.

[0272] The first determining module 802 performs wheel angle conversion, wheel center speed determination, slip ratio and sideslip angle calculation, and vertical load estimation for the target wheel based on the current operating parameter time sequence. This results in a time sequence of estimated values ​​for the current hidden state parameters, including the target wheel's wheel angle, wheel center speed, slip ratio, sideslip angle, wheel end driving force, wheel end braking force, and vertical load. The specific implementation method can be referred to the above embodiment of the vehicle kinematic parameter identification method, which will not be repeated here.

[0273] The second determining module 803 is used to determine the current parameter identification value of the vehicle's kinematic parameters to be identified based on the current operating parameter time series, the current hidden state parameter estimation time series, and the vehicle kinematic parameter identification model.

[0274] The vehicle kinematics parameter identification model is trained using the training method of the vehicle kinematics parameter identification model in any of the above embodiments.

[0275] The second determining module 803 inputs the current operating parameter time series and the current hidden state parameter estimate time series into the vehicle kinematic parameter recognition model. After one forward calculation by the first parameter recognition layer, the second parameter recognition layer, the third parameter recognition layer and the parameter correction layer, it outputs the current parameter recognition values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified and the third target kinematic parameter to be identified, for real-time use by downstream functions such as vehicle dynamics control.

[0276] In some embodiments, the above-mentioned functional modules can be implemented in software, for example, by a processor executing program instructions stored in memory to complete the respective functions of the acquisition module, the first determination module, and the second determination module; they can also be implemented in hardware, for example, by using an application-specific integrated circuit or a field-programmable gate array; or they can be implemented by a combination of software and hardware. The above module division is only an exemplary functional division. In actual implementation, the modules can be split or merged as needed. For example, the vertical load estimation function in the first determination module can be divided into a separate module, as long as the corresponding function can be achieved. This application does not impose any restrictions on this.

[0277] Figure 9 This is a block diagram illustrating an electronic device according to an embodiment of this application. Figure 9 As shown, the electronic device includes, but is not limited to, a processor 901 and a memory 902.

[0278] The aforementioned memory 902 is used to store the executable instructions of the aforementioned processor 901. It is understood that the aforementioned processor 901 is configured to execute instructions to implement the network maintenance method described in the above embodiments.

[0279] It should be noted that those skilled in the art will understand that Figure 9 The electronic device structure shown does not constitute a limitation on the electronic device; the electronic device may include, but is not limited to, other electronic devices. Figure 9 This may indicate more or fewer components, or combinations of certain components, or different component arrangements.

[0280] Processor 901 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in memory 902, and by calling data stored in memory 902, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Processor 901 may include one or more processing units. Processor 901 may integrate an application processor and a modem processor. The application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into processor 901.

[0281] The memory 902 can be used to store software programs and various data. The memory 902 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required by at least one functional module (such as deterministic components, integrated components, etc.), etc. Furthermore, the memory 902 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0282] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 902 including instructions, which can be executed by a processor 901 of an electronic device to implement the methods in the above embodiments.

[0283] Optionally, the computer-readable storage medium may be a non-transitory computer-readable storage medium, such as a read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device. In an exemplary embodiment, this application also provides a computer program product including one or more instructions, which can be executed by a processor 901 of an electronic device to perform the methods in the above embodiments.

[0284] It should be noted that when one or more instructions in the computer-readable storage medium or computer program product are executed by the processor of an electronic device, they implement the various processes of the above method embodiments and achieve the same technical effect as the above method. To avoid repetition, they will not be described again here.

[0285] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0286] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0287] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0288] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0289] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, essentially, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0290] This application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods described in the above method embodiments.

[0291] This application also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method in the method flow shown in the above method embodiments.

[0292] The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, a register, a hard disk, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an application-specific integrated circuit (ASIC). In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0293] Since the network maintenance device, computer-readable storage medium, and computer program product in the embodiments of this application can be applied to the above method, the technical effects that can be obtained can also be referred to the above method embodiments. The embodiments of this application will not be repeated here.

[0294] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A training method for a vehicle kinematic parameter recognition model, characterized in that, The training method for the vehicle kinematic parameter model includes: Obtain the time-series sequence of vehicle operating parameters; wherein, the time-series sequence of operating parameters includes a time-aligned time-series sequence of actual values ​​of state parameters and a time-series sequence of actual values ​​of observation parameters; Based on the time series of the operating parameters and the inherent kinematic parameters of the vehicle, determine the time series of the hidden state parameter estimates of the vehicle; Based on the time series of the operating parameters, the time series of the estimated hidden state parameters, and the initial vehicle kinematics parameter identification model, the parameter identification values ​​of the kinematic parameters to be identified for the vehicle are determined. Based on the time series of the estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, the dynamic equivalent model of the vehicle, and the tire equivalent model, the time series of the predicted state parameters of the vehicle is determined; the tire equivalent model is coupled with the vehicle dynamic equivalent model. Based on the first state prediction deviation between the time series of the true state parameter values ​​and the time series of the predicted state parameter values, the initial vehicle kinematic parameter identification model is optimized to obtain the vehicle kinematic parameter identification model.

2. The training method for the vehicle kinematic parameter recognition model according to claim 1, characterized in that, The state parameters in the time sequence of the true values ​​of the state parameters include: longitudinal velocity of the center of mass, lateral velocity of the center of mass, yaw rate, and wheel angular velocity of the target wheel; the target wheel is one of the wheels on the vehicle. The observed parameters in the time series of the true values ​​of the observed parameters include: steering wheel angle, driving force and braking force provided by the power system; The hidden state parameters in the time series of the estimated hidden state parameters include: the wheel angle, wheel center speed, slip ratio, sideslip angle, wheel end driving force, wheel end braking force, and vertical load of the target wheel. The kinematic parameters to be identified include: a first target kinematic parameter to be identified, a second target kinematic parameter to be identified, and a third target kinematic parameter to be identified, classified according to the rate of change of parameter values. The kinematic parameters to be identified for the first target include: the first distance from the vehicle's center of mass to the front axle of the vehicle, the yaw angle moment of inertia, the air resistance coefficient, the vehicle's frontal area, the rolling resistance coefficient, and the vehicle's braking system response delay. The second target kinematic parameters to be identified include: the actual longitudinal stiffness and actual lateral stiffness of the target axle of the vehicle; the target axle is one of the front axle and the rear axle of the vehicle; The third target kinematic parameter to be identified includes: the road surface equivalent adhesion coefficient.

3. The training method for the vehicle kinematic parameter recognition model according to claim 2, characterized in that, The step of determining the parameter identification values ​​of the vehicle's kinematic parameters to be identified based on the time series sequence of the operating parameters, the time series sequence of the estimated hidden state parameters, and the initial vehicle kinematic parameter identification model includes: The time series sequence of the operating parameters and the time series sequence of the estimated hidden state parameters are input into the first parameter recognition layer in the initial vehicle kinematics parameter recognition model to obtain the initial recognition value of the first target kinematic parameter to be identified output by the first parameter recognition layer. The time series sequence of the operating parameters and the time series sequence of the estimated hidden state parameters are input into the second parameter recognition layer in the initial vehicle kinematics parameter recognition model to obtain the initial recognition value of the second target kinematic parameters to be identified output by the second parameter recognition layer. The time series sequence of the operating parameters and the time series sequence of the estimated hidden state parameters are input into the third parameter recognition layer in the initial vehicle kinematics parameter recognition model to obtain the initial recognition value of the third target kinematics parameter to be identified output by the third parameter recognition layer. The initial identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified are input into the parameter correction layer in the initial vehicle kinematic parameter identification model to obtain the parameter identification values ​​of the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified output by the parameter correction layer. The parameter correction layer is used to correct the initial identification value of any kinematic parameter to be identified among the first target kinematic parameter to be identified, the second target kinematic parameter to be identified, and the third target kinematic parameter to be identified, by using the maximum allowable upper limit and the minimum allowable lower limit of the parameter of the kinematic parameter to be identified, so as to obtain the parameter identification value of the kinematic parameter to be identified.

4. The training method for the vehicle kinematic parameter recognition model according to claim 3, characterized in that, The step of determining the time series of predicted state parameters of the vehicle based on the time series of the estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, the dynamic equivalent model of the vehicle, and the tire equivalent model includes: Based on the time series of the latent state parameter estimates, the parameter identification values ​​of the kinematic parameters to be identified, and the tire equivalent model, the longitudinal force and lateral force of the target wheel in its own coordinate system are determined. The tire equivalent model characterizes the following relationships: the longitudinal force of the target wheel in its own wheel coordinate system, the actual longitudinal stiffness of the target wheel, the slip ratio of the target wheel, the vertical load of the target wheel, and the equivalent road surface adhesion coefficient; and the lateral force of the target wheel in its own wheel coordinate system, the actual lateral stiffness of the target wheel, the sideslip angle of the target wheel, the vertical load of the target wheel, and the equivalent road surface adhesion coefficient. The actual longitudinal stiffness and actual lateral stiffness of the target wheel are determined based on the actual longitudinal stiffness and actual lateral stiffness of the axle on which the target wheel is located, and the ratio of the vertical loads of the two wheels on the axle on which the target wheel is located. Based on the longitudinal and lateral forces of the target wheel in its own coordinate system, the time series of the estimated values ​​of the latent state parameters, the parameter identification values ​​of the kinematic parameters to be identified, and the dynamic equivalent model, the time series of the predicted values ​​of the vehicle's state parameters is determined; wherein, the dynamic equivalent model is used to characterize the correspondence between the longitudinal and lateral forces of the target wheel in its own coordinate system, the parameter values ​​of the latent state parameters, the parameter values ​​of the kinematic parameters to be identified, and the state parameters.

5. The training method for the vehicle kinematic parameter recognition model according to claim 4, characterized in that, The aforementioned dynamic equivalent model is specifically used to characterize: The relationship between the vehicle mass, the longitudinal acceleration of the center of mass, the lateral velocity, the yaw rate, the resultant longitudinal force of each wheel of the vehicle in the vehicle body coordinate system, air resistance, and rolling resistance. And the correspondence between the vehicle mass, the longitudinal velocity, the lateral acceleration of the center of mass, the yaw rate, and the resultant lateral forces of each wheel of the vehicle in the vehicle coordinate system; And the correspondence between the moment of inertia of the target wheel, the wheel angular acceleration of the target wheel, the wheel-end driving torque of the target wheel, the wheel-end braking torque of the wheel, the effective rolling radius of the wheel, and the longitudinal force of the target wheel in its own wheel coordinate system; And the correspondence between the longitudinal coordinate of the target wheel in the vehicle coordinate system, the lateral coordinate of the target wheel in the vehicle coordinate system, the lateral force of the target wheel in the vehicle coordinate system, the longitudinal force of the target wheel in the vehicle coordinate system, the yaw angle moment of inertia, and the yaw angle acceleration; And the correspondence between the lateral velocity of the wheel center, the lateral velocity, the yaw rate and the longitudinal coordinate of the target wheel in the vehicle coordinate system; And the correspondence between the longitudinal velocity of the wheel center, the longitudinal velocity, the yaw rate and the lateral coordinates of the target wheel in the vehicle coordinate system; In addition, the correspondence between the lateral force of the target wheel in the vehicle coordinate system, the longitudinal force of the target wheel in the vehicle coordinate system, the lateral force of the target wheel in its own wheel coordinate system, the longitudinal force of the target wheel in its own wheel coordinate system, and the rotation matrix between the wheel coordinate system of the target wheel and the vehicle coordinate system. Wherein, the wheel angular acceleration is the derivative of the wheel angular velocity of the target wheel; the yaw acceleration is the derivative of the yaw velocity; the longitudinal acceleration is the derivative of the longitudinal velocity; the lateral acceleration is the derivative of the longitudinal velocity; the air resistance is determined based on the air resistance coefficient and the frontal area; the rolling resistance is determined based on the rolling resistance coefficient; the longitudinal coordinate and the lateral coordinate are determined based on the first distance; the rotation matrix is ​​determined by the rotation angle of the target wheel; the vehicle coordinate system is constructed with the center of mass as the origin and the directions of the longitudinal and lateral forces of the vehicle as coordinate axes.

6. The training method for the vehicle kinematic parameter recognition model according to claim 4, characterized in that, The process of optimizing the initial vehicle kinematic parameter identification model based on the first state prediction deviation between the time series sequence of the true state parameter values ​​and the time series sequence of the predicted state parameter values, to obtain the vehicle kinematic parameter identification model, includes: The loss function is determined based on at least one of the first state prediction deviation and the target deviation; Based on the loss function, the model parameters of the initial vehicle kinematics parameter recognition model are optimized to obtain the vehicle kinematics parameter recognition model; The target deviation includes: The first identification deviation between the identified value and the true value of the kinematic parameter to be identified; The parameter correction deviation between the identified value and the initial identified value of the kinematic parameter to be identified; The second identification deviation between the target torque of the target wheel in its own coordinate system and the actual target torque of the target wheel in its own coordinate system; the target torque is one of longitudinal force and lateral force; The second state prediction deviation between the time series of theoretical state parameter values ​​and the time series of actual state parameter values ​​of the vehicle; wherein the time series of theoretical state parameter values ​​is determined based on the time series of actual hidden state parameter values, the actual values ​​of the kinematic parameters to be identified, the dynamic equivalent model, and the tire equivalent model; The parameter smoothing deviation between the parameter identification value of the kinematic parameter to be identified and the parameter identification value of the kinematic parameter to be identified at the previous moment; The cumulative prediction deviation of the identified kinematic parameters for predicting vehicle state at multiple future moments.

7. A method for identifying vehicle kinematic parameters, characterized in that, The vehicle kinematic parameter identification method includes: Obtain the time-series sequence of the vehicle's current operating parameters; wherein, the time-series sequence of the current operating parameters includes a time-aligned time-series sequence of the true values ​​of the current state parameters and a time-series sequence of the true values ​​of the current observation parameters; Based on the current operating parameter time series and the vehicle's inherent kinematic parameters, determine the current hidden state parameter estimate time series of the vehicle; Based on the current operating parameter time series, the current hidden state parameter estimation time series, and the vehicle kinematic parameter identification model, the current parameter identification value of the vehicle's kinematic parameter to be identified is determined. The vehicle kinematics parameter identification model is trained using the method described in any one of claims 1-6.

8. A training device for a vehicle kinematic parameter recognition model, characterized in that, The training device for the vehicle kinematic parameter model includes: The acquisition module is used to acquire the time-series sequence of vehicle operating parameters; wherein, the time-series sequence of operating parameters includes a time-aligned time-series sequence of real values ​​of state parameters and a time-series sequence of real values ​​of observation parameters; The first determining module is used to determine the time series of the hidden state parameter estimates of the vehicle based on the time series of the operating parameters and the inherent kinematic parameters of the vehicle. The second determining module is used to determine the parameter identification value of the vehicle's kinematic parameters to be identified based on the time series sequence of the operating parameters, the time series sequence of the estimated hidden state parameters, and the initial vehicle kinematic parameter identification model. The third determining module is used to determine the time series of predicted values ​​of the vehicle's state parameters based on the time series of the estimated hidden state parameters, the parameter identification values ​​of the kinematic parameters to be identified, the vehicle's dynamic equivalent model, and the tire equivalent model; the tire equivalent model is coupled with the vehicle's dynamic equivalent model. An optimization module is used to optimize the initial vehicle kinematic parameter recognition model based on the first state prediction deviation between the time series sequence of the true state parameter values ​​and the time series sequence of the predicted state parameter values, thereby obtaining the vehicle kinematic parameter recognition model.

9. A vehicle kinematics parameter identification device, characterized in that, The vehicle kinematics parameter identification device includes: The acquisition module is used to acquire the time sequence of the vehicle's current operating parameters; wherein, the time sequence of the current operating parameters includes a time-aligned time sequence of the true values ​​of the current state parameters and a time sequence of the true values ​​of the current observation parameters; The first determining module is used to determine the time series of estimated values ​​of the current hidden state parameters of the vehicle based on the time series of the current operating parameters and the inherent kinematic parameters of the vehicle. The second determining module is used to determine the current parameter identification value of the vehicle's kinematic parameters to be identified based on the current operating parameter time series, the current hidden state parameter estimation time series, and the vehicle kinematic parameter identification model; wherein the vehicle kinematic parameter identification model is trained using the method described in any one of claims 1-6.

10. A vehicle, characterized in that, The vehicle includes the training device for the vehicle kinematic parameter recognition model as described in claim 8 and the vehicle kinematic parameter recognition device as described in claim 9.

Images