A method and apparatus for estimating forces on a vehicle tire

Abstract

The application provides a vehicle tire force estimation method and device, comprising: acquiring state response data and control quantity data of a vehicle in a continuous time window, and performing normalization processing to obtain input feature data; inputting the input feature data into a pre-constructed full connection neural network model, and outputting lateral force estimation values of each wheel of the vehicle through nonlinear mapping processing of the full connection neural network model; inputting the lateral force estimation values as intermediate physical variables into a vehicle dynamics mechanism model containing the influence of a pitch angle and a roll angle, and calculating vehicle state prediction values at a next moment based on Newton-Euler theorem; constructing a loss function based on the deviation between the vehicle state prediction values and actually measured vehicle state response data at the next moment, and iteratively optimizing parameters of the full connection neural network model by using the loss function until a final tire force estimation result is output after a convergence condition is met.

Description

Technical Field

[0001] This invention belongs to the field of vehicle dynamics modeling technology, specifically relating to a method and apparatus for estimating vehicle tire forces. Background Technology

[0002] Tires are the only medium for force transmission between a vehicle and the road surface, and estimating tire forces is fundamental to predicting vehicle conditions. The difficulty in tire force estimation lies in its time-varying nature, strong coupling, and the inherently high nonlinearity of the tire itself, making it challenging to establish accurate, universal models. Key states such as the road adhesion coefficient and tire deformation cannot be directly measured and can only be indirectly observed through vehicle sensors, which are also susceptible to noise interference. Under extreme conditions such as emergency braking or low-adhesion surfaces, tire characteristics enter the nonlinear saturation region, and traditional methods easily fail. Therefore, accurate tire force estimation remains a significant challenge.

[0003] Traditional tire force estimation methods typically utilize tire models or vehicle dynamics models. The accuracy of tire models depends on input and model parameters, which vary due to tire aging, wear, and complex driving conditions. When certain key parameters are time-varying and difficult to calibrate precisely, tire model-based methods struggle to guarantee accurate tire force estimation. Vehicle dynamics model-based methods avoid the influence of tire parameters, but the accuracy of the estimation results also decreases when the tire enters a highly nonlinear region. Using highly complex physical models can better describe the vehicle's dynamic behavior in such scenarios. However, as the model's degrees of freedom increase, more precise hyperparameters (such as center of gravity position, tire lateral angular stiffness, and road adhesion coefficient) are required. Furthermore, most methods can only estimate the lateral tire forces of the front and rear axles; subsequent allocation of individual tire lateral forces introduces further errors.

[0004] For nonlinear operating conditions, data-driven methods offer satisfactory performance. However, the "black box" nature of neural networks leads to a lack of sufficient interpretability and generalization. Furthermore, data-driven tire force estimation methods are highly dependent on data; model performance is affected by data quality and quantity. In practical applications, it is necessary to collect a sufficient amount of high-quality data to improve the performance of machine learning models. However, the actual tire force, which serves as the true label for network training, is difficult to obtain directly. These issues limit the scope and conditions for the application of data-driven models.

[0005] In summary, improving the accuracy of vehicle tire force estimation is a problem that needs to be solved. Summary of the Invention

[0006] The present invention aims to at least partially solve one of the technical problems in the related art.

[0007] Therefore, the first objective of this invention is to provide a method for estimating vehicle tire force.

[0008] The second objective of this invention is to provide a vehicle tire force estimation device.

[0009] The third objective of this invention is to provide a computer device.

[0010] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0011] To achieve the above objectives, a first aspect of the present invention provides a method for estimating vehicle tire force, comprising:

[0012] S1, acquire the vehicle's state response data and control quantity data within a continuous time window, and normalize the state response data and control quantity data to obtain input feature data; S2, The input feature data is input into a pre-built fully connected neural network model, and the lateral force estimate of each wheel of the vehicle is output through the nonlinear mapping processing of the fully connected neural network model; S3, the estimated lateral force is input as an intermediate physical variable into the vehicle dynamics mechanism model that includes the effects of pitch and roll angles, the vehicle acceleration is calculated based on the Newton-Euler theorem, and the vehicle state at the current moment is updated by integration to obtain the predicted vehicle state value at the next moment. S4. Based on the deviation between the predicted vehicle state value and the actual measured vehicle state response data at the next moment, a loss function is constructed, and the parameters of the fully connected neural network model are iteratively optimized using the loss function until the convergence condition is met, and the final tire force estimation result is output.

[0013] In one embodiment of the present invention, acquiring the vehicle's state response data and control quantity data within a continuous time window includes: The system collects lateral vehicle speed, longitudinal vehicle speed, yaw rate, and steering wheel angle at the current and historical moments. Based on a preset time window length, it unfolds and splices the vehicle state response and control quantities at multiple consecutive moments into a time series to construct an input feature matrix containing time-series information. The input feature matrix is ​​then normalized to obtain input feature data that eliminates orders of magnitude differences.

[0014] In one embodiment of the present invention, the output of the estimated lateral force values ​​for each wheel of the vehicle includes: The input feature data is fed into the input layer of a fully connected neural network. Affine transformation and nonlinear activation operations are performed layer by layer in several hidden layers. The output of the last hidden layer is linearly mapped by the output layer to produce an output vector containing the estimated lateral forces of the left front wheel, right front wheel, left rear wheel, and right rear wheel.

[0015] In one embodiment of the present invention, the step of calculating vehicle acceleration based on the Newton-Euler theorem and updating the vehicle state at the current moment by integration to obtain the predicted vehicle state value at the next moment includes: The estimated four-wheel lateral forces output by the fully connected neural network and the longitudinal forces of the wheels are used together as mechanical inputs and substituted into the vehicle planar dynamics equation based on the Newton-Euler theorem. In the calculation process, the vehicle pitch rate and roll rate are introduced as coupling variables. Combined with the vehicle mass, moment of inertia and center of mass position parameters, the vehicle's longitudinal acceleration, lateral acceleration and yaw acceleration are calculated. The vehicle state at the current moment is updated by integrating the calculated acceleration to obtain the predicted vehicle state at the next moment, which includes longitudinal velocity, lateral velocity, and yaw rate.

[0016] In one embodiment of the present invention, the longitudinal force of the wheel is calculated based on the wheel driving torque, braking torque, wheel moment of inertia, wheel angular velocity and effective rolling radius of the tire; the calculated longitudinal forces of each wheel and the lateral forces of each wheel output by the neural network are substituted into the differential equations of the vehicle's longitudinal velocity, lateral velocity and yaw rate, and the solution process explicitly includes the vertical velocity coupling term caused by the pitch rate and the lateral velocity coupling term caused by the roll rate.

[0017] In one embodiment of the present invention, the construction of the loss function and iterative optimization includes: Extract the predicted values ​​of longitudinal velocity, lateral velocity, and yaw rate from the vehicle state prediction values, and obtain the actual values ​​corresponding to the next moment obtained from the actual measurement. Calculate the squared deviation between each predicted value and the true value, multiply each squared deviation by the corresponding weight coefficient, and then sum them to construct a loss function with vehicle state response error as the core. Based on the loss function, the optimizer uses the backpropagation algorithm to calculate gradients to update network parameters.

[0018] To achieve the above objectives, a second aspect of the present invention provides a vehicle tire force estimation device, comprising: The data acquisition and processing module is used to acquire the vehicle's state response data and control quantity data within a continuous time window, and to normalize the state response data and control quantity data to obtain input feature data. The neural network estimation module is used to input the input feature data into a pre-built fully connected neural network model, and output the estimated lateral force values ​​of each wheel of the vehicle through the nonlinear mapping processing of the fully connected neural network model. The mechanism model prediction module is used to input the estimated lateral force value as an intermediate physical variable into the vehicle dynamics mechanism model that includes the effects of pitch angle and roll angle, calculate the vehicle acceleration based on the Newton-Euler theorem, and update the vehicle state at the current moment by integration to obtain the predicted vehicle state value at the next moment. The parameter optimization module is used to construct a loss function based on the deviation between the predicted vehicle state value and the actual measured vehicle state response data at the next moment, and to use the loss function to iteratively optimize the parameters of the fully connected neural network model until the convergence condition is met, and then output the final tire force estimation result.

[0019] The present invention provides a vehicle tire force estimation method and apparatus that can achieve high-precision estimation without the need for real tire force labels. It effectively overcomes the problems of traditional methods being highly dependent on parameter calibration and the poor generalization of data-driven methods, and improves the estimation accuracy and robustness under complex working conditions.

[0020] To achieve the above objectives, a third aspect of this application provides a computer device comprising a processor and a memory; wherein the processor runs a program corresponding to the executable program code stored in the memory to implement a vehicle tire force estimation method as described in the first aspect embodiment.

[0021] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements a vehicle tire force estimation method as described in the first aspect embodiment.

[0022] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a vehicle tire force estimation method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a three-degree-of-freedom planar dynamics model of a vehicle according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a tire dynamics model according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a mechanism-data fusion vehicle dynamics model according to an embodiment of the present invention; Figure 5 This is a comparison chart of vehicle state response estimation results according to an embodiment of the present invention; Figure 6 This is a comparison chart of the four-wheel lateral force estimation results according to an embodiment of the present invention; Figure 7 This is a structural diagram of a vehicle tire force estimation device according to an embodiment of the present invention; Figure 8 It is a computer device according to an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] A method and apparatus for estimating vehicle tire force according to an embodiment of the present invention will now be described with reference to the accompanying drawings.

[0027] Example 1 Figure 1 This is a flowchart of a vehicle tire force estimation method according to an embodiment of the present invention, such as... Figure 1 As shown, it includes: S1, acquire the vehicle's state response data and control quantity data within a continuous time window, and normalize the state response data and control quantity data to obtain input feature data; S2, The input feature data is input into a pre-built fully connected neural network model, and the lateral force estimate of each wheel of the vehicle is output through the nonlinear mapping processing of the fully connected neural network model; S3, the estimated lateral force is input as an intermediate physical variable into the vehicle dynamics mechanism model that includes the effects of pitch and roll angles, the vehicle acceleration is calculated based on the Newton-Euler theorem, and the vehicle state at the current moment is updated by integration to obtain the predicted vehicle state value at the next moment. S4. Based on the deviation between the predicted vehicle state value and the actual measured vehicle state response data at the next moment, a loss function is constructed, and the parameters of the fully connected neural network model are iteratively optimized using the loss function until the convergence condition is met, and the final tire force estimation result is output.

[0028] The vehicle tire force estimation method of this invention can achieve high-precision estimation without the need for real tire force labels, effectively overcoming the problems of strong dependence on parameter calibration and poor generalization of data-driven methods in traditional methods, and improving the estimation accuracy and robustness under complex working conditions.

[0029] Example 2 The following description, in conjunction with the accompanying drawings, details a tire lateral force estimation method based on a mechanism-data fusion vehicle dynamics model according to an embodiment of the present invention.

[0030] 1. Vehicle planar dynamics model.

[0031] like Figure 2 As shown, for a four-wheel independently driven vehicle, its planar dynamics equations are constructed based on the Newton-Euler theorem, as follows:

[0032] in , , , These are the vehicle's longitudinal velocity, lateral velocity, vertical velocity, and yaw rate. , , These represent the vehicle's longitudinal acceleration, lateral acceleration, and yaw rate, respectively. Represents the longitudinal force of the wheel. Represents the lateral force of the wheel, where the subscript They represent Figure 2 The front and rear axles of the vehicle They represent Figure 2 The right and left wheels, For the front wheel steering angle, , , , , , , , These represent the vehicle's mass, roll moment of inertia, pitch moment of inertia, yaw moment of inertia, distance from the center of mass to the front (rear) axle, width between the coaxial wheels and the wheel centerline, pitch angular velocity, and roll angular velocity, respectively. This model comprehensively considers the vehicle's longitudinal, lateral, and yaw motions to describe the vehicle's dynamic response characteristics under complex operating conditions, providing mechanistic constraints for subsequent lateral force estimation and state prediction.

[0033] like Figure 3 As shown, the longitudinal force of the four wheels can be expressed as:

[0034] in, For wheel drive torque, This refers to the braking torque of the wheels. For the moment of inertia of the wheel, The angular velocity of the wheel. The effective rolling radius of the tire is given by the subscript. They represent Figure 2 The front and rear axles of the vehicle They represent Figure 2 The right and left wheels.

[0035] To simplify the expression, the vehicle state response and control variables can be defined as follows:

[0036] The overall dynamic model of the vehicle can be expressed as the following nonlinear function:

[0037] Among them, subscript Indicates the current moment. Indicates the next moment.

[0038] 2. Data acquisition and processing.

[0039] Using vehicle simulation software or actual vehicles, set relevant vehicle parameters, operating road and adhesion coefficient, sampling cutoff time, and sampling time to collect four state variables (including lateral vehicle speed). Longitudinal speed yaw rate Steering wheel angle The collected dataset was divided into training and validation sets. To avoid performance degradation due to order-of-magnitude differences between different datasets, a normalization technique was used to process the dataset, keeping all data within the range of -1 to 1, as shown below:

[0040] in Represents the original data. Represents normalized data. as well as These represent the maximum and minimum values ​​of the original data, respectively. This represents the mean of the original data.

[0041] 3. Construct a neural network model.

[0042] In this invention, a fully connected neural network structure is used to establish a nonlinear mapping relationship between measurable vehicle state variables and lateral forces of each wheel. Specifically, a multilayer perceptron structure is constructed, comprising an input layer, multiple hidden layers, and an output layer. The input vector consists of vehicle state responses and control variables. Furthermore, the input features in this invention are presented in a time window expansion format, incorporating time series information to enhance dynamic modeling capabilities.

[0043]

[0044] in, Indicates the current moment. Indicates the length of the time window.

[0045] A fully connected neural network consists of an input layer, several hidden layers, and an output layer. The network has a depth of J layers. Each layer performs affine transformation and nonlinear activation operations. Its computation process can be uniformly represented as:

[0046] This represents the feature vector of the j-th layer. and They are the first The weights and biases of the hidden layer, y j-1 This represents the output of the (j-1)th layer of the network, which serves as the input of the jth layer. This represents the activation function. The hidden layer uses the ReLU activation function to enhance the network's ability to represent sparse features and improve training stability. The output layer uses a linear activation function, and its output is an estimate of the tire's lateral force.

[0047] The output vector is defined as the estimated lateral force values ​​for the four wheels:

[0048] The four components correspond to the lateral forces of the four wheels of the vehicle at the current moment. Regarding the training mechanism, because this invention couples the neural network with a three-degree-of-freedom vehicle dynamics model, a mechanism-data fusion vehicle dynamics model is formed, such as... Figure 4As shown. Unlike traditional supervised learning, the output of this invention is not directly compared with the actual tire force, but is input as an intermediate physical variable into the vehicle dynamics mechanism model, thereby constructing an implicit learning framework based on physical constraints. Specifically, the four-wheel lateral forces output by the network are input into the vehicle dynamics model, and the predicted vehicle state value is calculated through the dynamic equations and compared with the actual measured value to construct a loss function. The loss function can be expressed as:

[0049] in For each state variable, the weighting coefficient is... The state quantity at the next moment is calculated from the mechanistic model.

[0050] During parameter optimization, the gradient of the loss function with respect to the network parameters is calculated through backpropagation, and the learning rate is set. The parameters are then updated using the Adam optimizer.

[0051] After multiple rounds of iterative training, the optimal network parameters were found.

[0052] Based on the above solution, the present invention also has the following embodiments: In this embodiment, a joint simulation test was conducted using the vehicle simulation software CarSim and the mathematical software Python. The training set was set to continuous lane changes and continuous curves, with different speed gradients ranging from 60km / h to 95km / h. The sampling interval was 0.025s, and a total of 22,377 training set data points were collected. The test set was set to variable speed curves, with a sampling interval of 0.025s, and a total of 6,004 data points were collected to evaluate the model's generalization ability under unseen conditions.

[0053] (1) Analysis of vehicle state response prediction results.

[0054] The vehicle state response prediction results for the test set are shown in Table 1 and Figure 5 As shown, three output state variables—longitudinal velocity, lateral velocity, and yaw rate—are selected as evaluation objects, and root mean square error (RMSE), mean absolute error (MAE), and maximum absolute error (MAX) are used as evaluation indicators. Figure 5 (a) shows the longitudinal velocity estimation results. Figure 5 (b) shows the lateral velocity estimation results. Figure 5 (c) shows the estimated yaw rate.

[0055] The model exhibits extremely low prediction errors across the three key state variables, with RMSE values ​​all below 0.005 (0.005 for longitudinal velocity, 0.001 for lateral velocity, and 0.002 for yaw rate), and maximum errors not exceeding 0.008. This result fully validates the effectiveness and generalization ability of the proposed method in vehicle state prediction.

[0056] Table 1. Prediction results of vehicle state response in the test set.

[0057] (2) Analysis of the estimation results of the four-wheel lateral forces of the vehicle.

[0058] As shown in Table 2 and Figure 6 As shown, where Figure 6 (a) shows the estimated lateral force of the right front wheel. Figure 6 (b) shows the estimated lateral force of the left front wheel. Figure 6 (c) shows the estimated lateral force of the right rear wheel. Figure 6 (d) shows the estimation results of the lateral force of the left rear wheel. Judging from the RMSE and MAE values, the estimation errors of the lateral forces of all four wheels remain at a low level, indicating high overall estimation accuracy. The error indicators of the rear wheels are significantly lower than those of the front wheels. This may be because the front wheels, as steering wheels, are more affected by steering input and complex suspension movements, resulting in more nonlinear force characteristics and making estimation relatively more difficult.

[0059] This estimation method can effectively capture the changing trend of lateral forces on all four wheels of a vehicle, especially showing excellent performance in estimating the lateral forces on the rear wheels. It can fully meet the needs of upper-level control logic such as vehicle stability control or path tracking for lateral force perception.

[0060] Table 2. Test Set Tire Lateral Force Prediction Results

[0061] Example 3 To achieve the above embodiments, such as Figure 7 As shown, this embodiment also provides a vehicle tire force estimation device 10, which includes a data acquisition and processing module 100, a neural network estimation module 200, a mechanism model prediction module 300, and a parameter optimization module 400.

[0062] The data acquisition and processing module 100 is used to acquire the vehicle's state response data and control quantity data within a continuous time window, and to normalize the state response data and control quantity data to obtain input feature data. The neural network estimation module 200 is used to input the input feature data into a pre-built fully connected neural network model and output the estimated lateral force values ​​of each wheel of the vehicle through the nonlinear mapping processing of the fully connected neural network model. The mechanism model prediction module 300 is used to input the estimated lateral force value as an intermediate physical variable into the vehicle dynamics mechanism model that includes the effects of pitch angle and roll angle, calculate the vehicle acceleration based on the Newton-Euler theorem, and update the vehicle state at the current moment by integration to obtain the predicted vehicle state value at the next moment. The parameter optimization module 400 is used to construct a loss function based on the deviation between the predicted vehicle state value and the actual measured vehicle state response data at the next moment, and to use the loss function to iteratively optimize the parameters of the fully connected neural network model until the convergence condition is met, and then output the final tire force estimation result.

[0063] Furthermore, the aforementioned data acquisition and processing module 100 is also used for: The system collects lateral vehicle speed, longitudinal vehicle speed, yaw rate, and steering wheel angle at the current and historical moments. Based on a preset time window length, it unfolds and splices the vehicle state response and control quantities at multiple consecutive moments into a time series to construct an input feature matrix containing time-series information. The input feature matrix is ​​then normalized to obtain input feature data that eliminates orders of magnitude differences.

[0064] Furthermore, the aforementioned neural network estimation module 200 is also used for: The input feature data is fed into the input layer of a fully connected neural network. Affine transformation and nonlinear activation operations are performed layer by layer in several hidden layers. The output of the last hidden layer is linearly mapped by the output layer to produce an output vector containing the estimated lateral forces of the left front wheel, right front wheel, left rear wheel, and right rear wheel.

[0065] Furthermore, the aforementioned mechanism model prediction module 300 is also used for: The estimated lateral forces of the four wheels output by the fully connected neural network and the longitudinal forces of the wheels are used as mechanical inputs and substituted into the vehicle planar dynamics equation based on the Newton-Euler theorem. During the calculation, the vehicle pitch velocity and roll velocity are introduced as coupling variables. Combined with the vehicle mass, moment of inertia and center of mass position parameters, the vehicle's longitudinal acceleration, lateral acceleration and yaw acceleration are calculated. The vehicle state at the current moment is updated by integrating the calculated accelerations to obtain the predicted vehicle state value at the next moment, which includes the longitudinal velocity, lateral velocity and yaw velocity.

[0066] Furthermore, the longitudinal force of the wheel is calculated based on the wheel driving torque, braking torque, wheel rotational inertia, wheel angular velocity, and effective rolling radius of the tire. The calculated longitudinal forces of each wheel and the lateral forces of each wheel output by the neural network are substituted into the differential equations of the vehicle's longitudinal velocity, lateral velocity, and yaw rate. In the solution process, the vertical velocity coupling term caused by the pitch rate and the lateral velocity coupling term caused by the roll rate are explicitly included.

[0067] Furthermore, the parameter optimization module 400 is also used for: Extract the predicted longitudinal velocity, lateral velocity, and yaw rate from the vehicle state prediction values, and obtain the actual values ​​corresponding to the next moment obtained from actual measurements; calculate the squared deviation between each predicted value and the actual value, multiply each squared deviation by the corresponding weight coefficient, and sum them to construct a loss function with the vehicle state response error as the core; based on the loss function, use the optimizer to calculate the gradient through the backpropagation algorithm to update the network parameters.

[0068] The vehicle tire force estimation device of this invention can achieve high-precision estimation without the need for real tire force labels, effectively overcoming the problems of strong dependence on parameter calibration and poor generalization of data-driven methods in traditional methods, and improving the estimation accuracy and robustness under complex working conditions.

[0069] To implement the methods of the above embodiments, the present invention also provides a computer device, such as... Figure 8 As shown, the computer device 600 includes a memory 601 and a processor 602; wherein, the processor 602 reads executable program code stored in the memory 601 to run a program corresponding to the executable program code, so as to implement the various steps of the method described above.

[0070] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0071] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0072] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for estimating vehicle tire force, characterized in that, include: S1, acquire the vehicle's state response data and control quantity data within a continuous time window, and normalize the state response data and control quantity data to obtain input feature data; S2, The input feature data is input into a pre-built fully connected neural network model, and the lateral force estimate of each wheel of the vehicle is output through the nonlinear mapping processing of the fully connected neural network model; S3, the estimated lateral force is input as an intermediate physical variable into the vehicle dynamics mechanism model that includes the effects of pitch and roll angles, the vehicle acceleration is calculated based on the Newton-Euler theorem, and the vehicle state at the current moment is updated by integration to obtain the predicted vehicle state value at the next moment. S4. Based on the deviation between the predicted vehicle state value and the actual measured vehicle state response data at the next moment, a loss function is constructed, and the parameters of the fully connected neural network model are iteratively optimized using the loss function until the convergence condition is met, and the final tire force estimation result is output.

2. The method as described in claim 1, characterized in that, The acquisition of vehicle status response data and control quantity data within a continuous time window includes: Collect lateral vehicle speed, longitudinal vehicle speed, yaw rate, and steering wheel angle at the current and historical moments; Based on a preset time window length, the vehicle state response and control quantity at multiple consecutive moments are spliced ​​together in time series to construct an input feature matrix containing time series information; The input feature matrix is ​​normalized to obtain input feature data with eliminated order-of-magnitude differences.

3. The method as described in claim 1, characterized in that, The output of the estimated lateral force values ​​for each wheel of the vehicle includes: The input feature data is fed into the input layer of a fully connected neural network, and affine transformations and nonlinear activation operations are performed layer by layer in several hidden layers. The output of the last hidden layer is linearly mapped by the output layer, resulting in an output vector containing estimates of the lateral forces of the left front wheel, right front wheel, left rear wheel, and right rear wheel.

4. The method as described in claim 1, characterized in that, The process of calculating vehicle acceleration based on Newton-Euler theorem and updating the vehicle state at the current moment by integration to obtain the predicted vehicle state at the next moment includes: The estimated four-wheel lateral forces output by the fully connected neural network and the longitudinal forces of the wheels are used together as mechanical inputs and substituted into the vehicle planar dynamics equation based on the Newton-Euler theorem. In the calculation process, the vehicle pitch rate and roll rate are introduced as coupling variables. Combined with the vehicle mass, moment of inertia and center of mass position parameters, the vehicle's longitudinal acceleration, lateral acceleration and yaw acceleration are calculated. The vehicle state at the current moment is updated by integrating the calculated acceleration to obtain the predicted vehicle state at the next moment, which includes longitudinal velocity, lateral velocity, and yaw rate.

5. The method as described in claim 4, characterized in that, The longitudinal force of the wheel is calculated based on the wheel driving torque, braking torque, wheel rotational inertia, wheel angular velocity, and effective rolling radius of the tire. The calculated longitudinal forces of each wheel and the lateral forces of each wheel output by the neural network are substituted into the differential equations of the vehicle's longitudinal velocity, lateral velocity, and yaw rate. In the solution process, the vertical velocity coupling term caused by the pitch rate and the lateral velocity coupling term caused by the roll rate are explicitly included.

6. The method as described in claim 1, characterized in that, The iterative optimization of the parameters of the fully connected neural network model using the loss function includes: Extract the predicted values ​​of longitudinal velocity, lateral velocity, and yaw rate from the vehicle state prediction values, and obtain the actual values ​​corresponding to the next moment obtained from the actual measurement. Calculate the squared deviation between each predicted value and the true value, multiply each squared deviation by the corresponding weight coefficient, and then sum them to construct a loss function with vehicle state response error as the core. Based on the loss function, the optimizer uses the backpropagation algorithm to calculate gradients to update network parameters.

7. A vehicle tire force estimation device, characterized in that, include: The data acquisition and processing module is used to acquire the vehicle's state response data and control quantity data within a continuous time window, and to normalize the state response data and control quantity data to obtain input feature data. The neural network estimation module is used to input the input feature data into a pre-built fully connected neural network model, and output the estimated lateral force values ​​of each wheel of the vehicle through the nonlinear mapping processing of the fully connected neural network model. The mechanism model prediction module is used to input the estimated lateral force value as an intermediate physical variable into the vehicle dynamics mechanism model that includes the effects of pitch angle and roll angle, calculate the vehicle acceleration based on the Newton-Euler theorem, and update the vehicle state at the current moment by integration to obtain the predicted vehicle state value at the next moment. The parameter optimization module is used to construct a loss function based on the deviation between the predicted vehicle state value and the actual measured vehicle state response data at the next moment, and to use the loss function to iteratively optimize the parameters of the fully connected neural network model until the convergence condition is met, and then output the final tire force estimation result.

8. A computer device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement a vehicle tire force estimation method as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a vehicle tire force estimation method as described in any one of claims 1-6.

Images