An adaptive longitudinal vehicle speed estimation method

Abstract

The application relates to a self-adaptive longitudinal vehicle speed estimation method, comprising the following steps: establishing a detailed vehicle dynamics model, including a vehicle longitudinal dynamics model and a wheel speed dynamics model; combining a Sage-Husa self-adaptive unscented Kalman filtering algorithm with the vehicle dynamics model, designing a state equation and an observation equation, taking a vehicle longitudinal speed as a state variable, and constructing an observation vector by using data collected by a wheel speed sensor, an acceleration sensor and the like; introducing divergence calculation for detecting and correcting divergence phenomena in a filtering process; and based on a self-adaptive mechanism of the Sage-Husa algorithm, adjusting a filtering gain in real time and optimizing a vehicle speed estimation process. The improved Sage-Husa self-adaptive filter can dynamically adjust filter parameters according to changes in system states and differences in noises, detect transient disturbances by combining divergence calculation, reduce influences of the transient disturbances on longitudinal vehicle speed estimation, and effectively improve estimation accuracy.

Description

Technical Field

[0001] This invention relates to the field of automotive intelligent safety control technology, specifically to an adaptive longitudinal vehicle speed estimation method. Background Technology

[0002] As a key input parameter for many vehicle control systems (such as ABS, TCS, and ESP), the accuracy of vehicle longitudinal velocity is crucial for improving active safety and driving performance. Traditional speed estimation methods rely on expensive and environmentally sensitive sensors, while low-precision sensors may lead to excessive measurement errors. Furthermore, drastic changes in vehicle operating conditions, such as variations in road surface adhesion coefficient, gradient changes, and instantaneous acceleration or braking, significantly alter vehicle dynamics, further complicating the accurate estimation of longitudinal velocity.

[0003] Currently, Kalman filtering and its derivative algorithms are the mainstream methods in the field of vehicle state estimation. Traditional Kalman filtering (KF) performs excellently in handling linear systems, but vehicle dynamics systems are mostly nonlinear. Extended Kalman filtering (EKF) linearizes nonlinear systems through Taylor expansion, but in strongly nonlinear or high-noise environments, linearization approximation can lead to accumulated estimation errors or even filter divergence. Unscented Kalman filtering (UKF) has gained widespread attention due to its superior estimation performance for nonlinear systems; however, the standard UKF algorithm has some limitations. It relies heavily on prior knowledge of the statistical characteristics of process and measurement noise. In practical applications, the noise covariance matrix is ​​often difficult to obtain accurately and changes dynamically with operating conditions, which can lead to a decrease in filter estimation accuracy or even divergence.

[0004] Existing methods for estimating vehicle longitudinal speed still suffer from insufficient accuracy and poor robustness when faced with complex and variable operating conditions and noise interference. Therefore, a method that can estimate vehicle longitudinal speed in real time and accurately is needed to meet the stringent requirements of modern vehicle control systems for the accuracy and reliability of speed information. Summary of the Invention

[0005] In view of the problems existing in the prior art, the purpose of this invention is to provide an adaptive longitudinal vehicle speed estimation method that can improve the accuracy and anti-interference ability of vehicle longitudinal speed estimation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An adaptive longitudinal vehicle speed estimation method includes the following steps:

[0008] (1) Collect wheel angular velocity signals and acceleration information, filter the wheel speed information and acceleration information and perform time-aligned data synchronization processing to obtain preprocessed data;

[0009] (2) Based on the preprocessed data obtained in step (1), construct a seven-degree-of-freedom vehicle model including longitudinal, lateral, yaw and four wheels of the vehicle body, and output vehicle dynamic parameters;

[0010] (3) Based on the preprocessed data in step (1) and the vehicle dynamics parameters in step (2), the Dugoff tire model is used to simulate and output the longitudinal and lateral forces of the tire;

[0011] (4) Based on the dynamic parameters output in step (2) and the longitudinal and lateral forces of the tires output in step (3), the longitudinal vehicle speed is incorporated into the state variables. The preprocessed data in step (1) is used to construct the observation variables and form the state equation and observation equation.

[0012] (5) Based on the state equation in step (4), the state estimate and covariance matrix are constructed by sampling points, and the sampling points are updated in time by the state transition matrix to predict the state mean and covariance matrix at the next moment.

[0013] (6) Based on the predicted state mean and covariance matrix obtained in step (5), calculate the correlation coefficient of the innovation sequence, and determine whether the filter diverges by setting a critical value. The innovation sequence is obtained by the difference between the actual observed value and the predicted observed value of the observation equation in step (4).

[0014] If the filter is determined to be diverging, the prediction covariance matrix obtained in step (5) is corrected using the covariance of the innovation sequence and the covariance of the predicted observations.

[0015] (7) Based on the covariance matrix corrected in step (6), calculate the mean and covariance of the predicted observations, and then calculate the Kalman gain using the preprocessed data from step (1) to update the state estimate and covariance matrix; at the same time, based on the Sage-Husa adaptive mechanism, update the mean and covariance matrix of the measurement noise using the innovation sequence from step (6).

[0016] (8) Combining the state estimate, covariance matrix and noise characteristic update results obtained in step (7), optimize the state estimate and covariance matrix again, and output the longitudinal vehicle speed estimate.

[0017] Furthermore, the dynamic equations of the seven-degree-of-freedom vehicle model are as follows:

[0018]

[0019]

[0020] Where m is the total mass of the vehicle; β is the sideslip angle; u is the longitudinal velocity; v is the lateral velocity; ω is the yaw rate; a x It is longitudinal acceleration; a y Γ is the lateral acceleration; Γ is the torque about the z-axis; I z δ is the moment of inertia of the car about the z-axis; δ is the wheel steering angle; F xi It is the longitudinal force of the tire; F zi This refers to the lateral force of the tire; i = 1, 2, 3, 4 represent the positions of the wheel at the left front, right front, left rear, and right rear, respectively; δ i The steering angle δ of a specific wheel in response to the wheel's steering angle; a and b represent the distances from the center of mass to the front and rear axles, respectively; t f ,t r This indicates the distance between the front or rear wheels.

[0021] Furthermore, the formulas for simulating the longitudinal and lateral forces of the tire using the Dugoff tire model are as follows:

[0022]

[0023] Where μ is the current road adhesion coefficient; F z ε is the vertical load on each tire; L is the introduced boundary condition; ε is the speed influence factor; C x and C y These refer to the longitudinal stiffness and lateral stiffness of the wheel, respectively.

[0024] Furthermore, the state equations and observation equations are designed as follows:

[0025] State variable x(t) = [v x v y r Γ a x a y ] T

[0026] Observed variable y(t)=[a x a y r] T

[0027] Based on the seven-DOF vehicle model and the Dugoff tire model, the state equation is:

[0028]

[0029] System observation equations:

[0030]

[0031] Where k is the discrete time; Λ is the noise driving matrix; X kY is the state matrix at time k; k is the observation matrix at time k; W(k) is the input white noise; V(k) is the observation noise.

[0032] Further, step (5) includes constructing a series of sampling points for approximating the system state distribution by sampling the state estimates and covariance matrix, with the number of sampling points being 2n+1, where n is the dimension of the state vector:

[0033]

[0034] in This is the current estimate; P k Here is the state covariance matrix; λ = α 2 (n+κ)-n, where α and κ are the parameters for adjusting the release of sampling points, respectively.

[0035] Furthermore, step (5) also includes updating the sampling points over time and calculating the next state of the Sigma point:

[0036] χ i,k|k-1 =f(χ) i,k-1 )

[0037] Where f(·) is the state transition matrix;

[0038] Predicted state mean and covariance matrix:

[0039]

[0040]

[0041] Where, ω i and ω i c These are the state weights and covariance weights of the sampling points, respectively; Q k Let be the system noise covariance matrix. Further, in step (6), the divergence is calculated: a system is described by the following state space:

[0042] X k+1 =φX k +ΛW k

[0043] Y k =HX k +V k

[0044] Where k is the discrete time; φ is the state transition matrix; H is the observation matrix; Λ is the noise driving matrix; X k Y is the state matrix at time k; k W is the observation matrix at time k;k It is input white noise; V k It is observation noise; assuming W k and V k The mean is zero, W k and V k The covariance matrix is ​​Q k and R k ,have:

[0045]

[0046] We can obtain:

[0047]

[0048] The estimated variance of the measured values ​​is:

[0049]

[0050] The covariance matrix can be represented as:

[0051] Cov(ε k+1 ·ε k+1 T ) = HP k+1 H T +R k+1

[0052] Where P k+1 It is the covariance matrix of the prediction error; when the estimation converges, ε k It is white noise that follows a normal distribution, therefore we can obtain the following equation, where ξ xt This is the correlation coefficient, with values ​​between 0 and 1.

[0053]

[0054] The correlation coefficient is used to determine divergence. The critical value for filter divergence is set to ξ0. If the following equation is satisfied, the filter diverges:

[0055]

[0056] Further, step (7) includes determining the divergence of the filter; if the filter diverges, correcting the covariance matrix P of the prediction error. k|k-1 ,

[0057]

[0058] Where ρ is in the range of 0-1.

[0059] Furthermore, step (7) also includes calculating the predicted observations and their mean and covariance matrices:

[0060]

[0061] Where h(·) is the observation function; R k To observe the noise covariance matrix;

[0062] Calculate the Kalman gain matrix and update the state estimate and covariance matrix.

[0063] K k =P xy P yy -1

[0064]

[0065] P k =P k|k-1 -K k P yy K k T .

[0066] Furthermore, step (7) also includes updating the mean and covariance matrix of the measurement noise based on the Sage-Husa adaptive mechanism:

[0067]

[0068] Where, d k =1-γ / 1-γ k γ is the fading factor, with a value range of 0-1; H k d is the Jacobian matrix of the nonlinear measurement matrix. k The function is in The changes are continuous and the computational load is reasonable. It can quickly adjust the Sage-Husa coefficients d, thereby meeting the real-time requirements of filtering calculations.

[0069] In summary, the present invention has the following advantages:

[0070] 1) The estimation method used estimates the measurement noise variance in real time and corrects it dynamically, thus optimizing problems such as vehicle speed estimation due to sensor data fluctuations in a strong noise environment.

[0071] 2) This method can also be used to study vehicle status information such as yaw rate and center of gravity sideslip angle.

[0072] 3) Accurately estimated longitudinal vehicle speed helps control vehicle drive, which is beneficial to improving vehicle dynamic performance and safety performance. Attached Figure Description

[0073] Figure 1 This is a flowchart of the method of the present invention.

[0074] Figure 2 This is a schematic diagram of a seven-degree-of-freedom vehicle model according to the present invention.

[0075] Figure 3 This is a schematic diagram of the simulation estimation process of the present invention.

[0076] Figure 4 This is a schematic diagram of the longitudinal acceleration signal of the present invention.

[0077] Figure 5 This is a schematic diagram comparing the longitudinal velocity estimation results of the present invention. Detailed Implementation

[0078] The inventive concept of this application addresses the problem of insufficient accuracy and poor robustness in longitudinal vehicle speed estimation under complex operating conditions (such as sensor noise interference, changes in road conditions, and transient disturbances). It proposes an adaptive longitudinal vehicle speed estimation method based on an improved Sage-HusaUKF, which achieves high-precision and high-reliability estimation of longitudinal vehicle speed by integrating accurate vehicle dynamics modeling with adaptive filtering optimization. The core idea is as follows:

[0079] 1. Based on accurate modeling, characterize vehicle dynamics: Establish a seven-DOF vehicle model, including three degrees of freedom (longitudinal, lateral, and yaw) and four wheel rotational degrees of freedom. Quantify the relationship between vehicle motion (such as longitudinal velocity, lateral velocity, and yaw rate) and forces (tire longitudinal force, lateral force, etc.) through dynamic equations. Combine this with the Dugoff tire model to simulate tire longitudinal and lateral forces, considering the influence of parameters such as road adhesion coefficient, tire stiffness, and vertical load on tire forces, thus improving the model's adaptability to complex road conditions.

[0080] 2. Improved filtering algorithm to overcome the limitations of traditional methods: Integrating the UKF's advantages in estimating nonlinear systems, avoiding the errors caused by the linearization approximation of the extended Kalman filter, and more accurately handling the nonlinear characteristics of vehicle dynamics systems. Introducing the Sage-Husa adaptive mechanism to dynamically update the mean and covariance matrix of the measurement noise, overcoming the strong dependence of the traditional UKF on prior knowledge of noise statistical characteristics, and adapting to scenarios where noise changes dynamically with operating conditions (such as fluctuations in sensor data).

[0081] 3. Introduce divergence calculation to enhance filtering stability: Detect divergence in the filtering process by calculating the correlation coefficient of the innovation sequence, and set a critical value to determine whether the filter diverges; if divergence occurs, adjust the filtering parameters by correcting the prediction error covariance matrix, etc., to avoid the accumulation of estimation errors and improve the stability of the algorithm under transient disturbances.

[0082] 4. Multi-stage optimization for high-precision estimation: Preprocess sensor data to improve input data quality. Design reasonable state and observation equations, using longitudinal vehicle speed as the core state variable, construct observation vectors using sensor data such as wheel speed and acceleration, and optimize the estimation results through a prediction-update iterative process.

[0083] The present invention will now be described in further detail.

[0084] An adaptive longitudinal vehicle speed estimation method includes the following steps:

[0085] (1) Obtain the preprocessed wheel speed and acceleration information:

[0086] Wheel speed sensors are installed on all four wheels to output the angular velocity signals of the wheels, and the vehicle's own body sensors are used to obtain acceleration information; the collected wheel speed information and acceleration information are preprocessed: (a) filtering: a low-pass filter is used to remove high-frequency noise in the signal; (b) data synchronization: the collected multi-source data is time aligned.

[0087] (2) Establish a seven-degree-of-freedom vehicle model:

[0088] The following assumptions are made for the seven-DOF vehicle model: (a) The vehicle model is assumed to be a rigid body; (b) The effects of pitch and roll on vehicle stability are ignored, and the front and rear track widths are assumed to be the same; (c) The vehicle is assumed to travel on a level road surface, and the effects of slope on vehicle travel are ignored.

[0089] The established seven-DOF vehicle model includes three degrees of freedom for the vehicle body (longitudinal, lateral, and yaw) and four rotational degrees of freedom for the wheels. The dynamic equations obtained from the model are as follows:

[0090]

[0091]

[0092] Where m is the total mass of the vehicle; β is the sideslip angle; u is the longitudinal velocity; v is the lateral velocity; ω is the yaw rate; a x It is longitudinal acceleration; a y Γ is the lateral acceleration; Γ is the torque about the z-axis; I z δ is the moment of inertia of the car about the z-axis; δ is the wheel steering angle; F xi It is the longitudinal force of the tire; F zi This refers to the lateral force of the tire; i = 1, 2, 3, 4 indicate that the wheel is located at the left front, right front, left rear, and right rear, respectively; δ i The steering angle δ of a specific wheel in response to the wheel's steering angle; a and b represent the distances from the center of mass to the front and rear axles, respectively; t f ,t rThis indicates the distance between the front or rear wheels.

[0093] (3) Establish a wheel dynamics model:

[0094] Use the Dugoff tire model to simulate longitudinal and lateral forces on the tire:

[0095]

[0096] Where μ is the current road adhesion coefficient; F z ε is the vertical load on each tire; L is the introduced boundary condition; ε is the speed influence factor; C x and C y These refer to the longitudinal stiffness and lateral stiffness of the wheel, respectively.

[0097] The tire model is closely related to the tire's longitudinal and lateral stiffness, and relevant wheel parameters need to be collected in advance.

[0098] (4) Design the state equations and observation equations:

[0099] Based on the key parameters obtained from the above seven-degree-of-freedom vehicle model and tire model, the state equations and observation equations are designed as follows:

[0100] State variable x(t) = [v x v y rΓa x a y ] T

[0101] Observed variable y(t)=[a x a y r] T

[0102] Based on the seven-DOF vehicle model and the Dugoff tire model, the state equation is:

[0103]

[0104] System observation equations:

[0105]

[0106] Where k is the discrete time; Λ is the noise driving matrix; X k Y is the state matrix at time k; k is the observation matrix at time k; W(k) is the input white noise; V(k) is the observation noise.

[0107] (5) Sampling point construction and time update:

[0108] (a) By constructing sampling points from the state estimates and covariance matrix, the number of sampling points is 2n+1, where n is the dimension of the state vector, thus obtaining a series of sampling points for approximating the system state distribution:

[0109]

[0110] in This is the current estimate; P k Here is the state covariance matrix; λ = α 2 (n+k)-n, where α and κ are the parameters for adjusting the release of sampling points, respectively.

[0111] (b) Update the sampling points in time and calculate the next state of the Sigma point:

[0112] χ i,k|k-1 =f(χ) i,k-1 )

[0113] Where f(·) is the state transition matrix.

[0114] Predicted state mean and covariance matrix:

[0115]

[0116] Where, ω i and ω i c These are the state weights and covariance weights of the sampling points, respectively; Q k Let be the system noise covariance matrix.

[0117] (6) Introducing the divergence calculation:

[0118] A system is described by the following state space:

[0119] X k+1 =φX k +ΛW k

[0120] Y k =HX k +V k

[0121] Where k is the discrete time; φ is the state transition matrix; H is the observation matrix; Λ is the noise driving matrix; X k Y is the state matrix at time k; k W is the observation matrix at time k; k It is input white noise; V k This is observation noise. Assume W k and V k The mean is zero, W k and Vk The covariance matrix is ​​Q k and R k ,have:

[0122]

[0123] We can obtain:

[0124]

[0125] The estimated variance of the measured values ​​is:

[0126]

[0127] The covariance matrix can be represented as:

[0128] Cov(ε k+1 ·ε k+1 T ) = HP k+1 H T +R k+1

[0129] Where P k+1 ε is the covariance matrix of the prediction error. When the estimation converges, ε k It is white noise that follows a normal distribution. Therefore, we can obtain the following equation, where ξ xy It is the correlation coefficient, with a value between 0 and 1.

[0130]

[0131] The correlation coefficient can be used to determine divergence. Let ξ0 be the critical value for filter divergence. If the following equation is satisfied, the filter diverges:

[0132]

[0133] (7) Update of observed variables and noise characteristics:

[0134] (a) Determine the divergence of the filter. If the filter diverges, correct the covariance matrix P of the prediction error. k|k-1 :

[0135]

[0136] Where ρ is in the range of 0-1. A larger ρ will reduce the residual information effect before time k and strengthen the effect of the current residual information.

[0137] (b) Calculate the predicted observations and their mean and covariance matrices:

[0138]

[0139]

[0140] Where h(·) is the observation function; R k To observe the noise covariance matrix.

[0141] Calculate the Kalman gain matrix and update the state estimate and covariance matrix:

[0142] K k =P xy P yy -1

[0143]

[0144] P k =P k|k-1 -K k P yy K k T

[0145] (c) Update the mean and covariance matrix of the measurement noise based on the Sage-Husa adaptive mechanism:

[0146]

[0147] Where, d k =1-γ / 1-γ k γ is the fading factor, with a value range of 0-1; H k is the Jacobian matrix of the nonlinear measurement matrix.

[0148] (8) Update the state and covariance matrix:

[0149] By combining the vehicle dynamics model and observation information, the state estimate and covariance matrix are updated again to obtain the longitudinal velocity estimate. The update process integrates information from multiple aspects such as prediction, observation, and noise characteristic adjustment, as shown in steps 5-7, and achieves accurate estimation of the vehicle's longitudinal speed through iterative optimization.

[0150] (9) Method verification:

[0151] By using software co-simulation and setting a noisy input signal, the results of the algorithm's estimated vehicle speed are compared with the actual vehicle speed. To verify the effectiveness and accuracy of the proposed algorithm in noisy environments, this invention builds a vehicle state estimation simulation model in the Carsim and Simulink co-simulation environment, as follows: Figure 3 As shown, the step size for the Carsim and Simulink co-simulation was set to 0.001s. To meet the experimental conditions, noise was artificially added to some of the required sensor signals in Carsim to simulate a real industrial application environment.

[0152] Table 1 shows some parameters of the reference vehicle model in Carsim:

[0153] Table 1 Vehicle Parameters

[0154]

[0155] Carsim simulation verification and analysis:

[0156] The simulation test of this invention uses a split road surface to verify the effectiveness of the proposed algorithm. The estimation results of the proposed improved Sage-Husa unscented Kalman filter are compared with those of the standard Kalman filter.

[0157] The root mean square error (RMSE) measures the average deviation between the predicted and the true values. The lower the RMSE value, the smaller the overall error. Therefore, this experiment uses RMSE to evaluate the accuracy of the improved algorithm in simulation verification.

[0158] Road conditions: The maximum coefficient of friction for the left wheel is 0.5, and for the right wheel it is 0.8. The vehicle accelerates from a standstill, with the longitudinal acceleration increasing. The input signal is Gaussian white noise. Figure 4 As shown. The estimation results are as follows. Figure 5 As shown in Table 2, the root mean square error of the improved algorithm proposed in this invention and UKF in the simulation results is shown in Table 2.

[0159] Table 2. Root Mean Square Error of Simulation Calculation Results

[0160]

[0161] The root mean square error (RMSE) of longitudinal vehicle speed estimation was reduced by 49.7%. This is because during startup, the longitudinal force difference between the two tires is significant. Traditional UKF algorithms, with a fixed measurement noise covariance, struggle to fully filter sensor-generated noise, leading to divergent results and significant bias, thus reducing estimation accuracy. The improved SH-UKF algorithm, however, optimizes and corrects the measurement noise covariance in noisy environments, reducing the divergent impact of measurement noise on filtering, thereby enhancing the algorithm's robustness and improving the accuracy of driving state observation estimation.

[0162] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. An adaptive longitudinal vehicle speed estimation method, characterized in that: Includes the following steps, (1) Collect wheel angular velocity signals and acceleration information, filter the wheel speed information and acceleration information and perform time-aligned data synchronization processing to obtain preprocessed data; (2) Based on the preprocessed data obtained in step (1), construct a seven-degree-of-freedom vehicle model including longitudinal, lateral, yaw and four wheels of the vehicle body, and output vehicle dynamic parameters; (3) Based on the preprocessed data in step (1) and the vehicle dynamics parameters in step (2), the Dugoff tire model is used to simulate and output the longitudinal and lateral forces of the tire; (4) Based on the dynamic parameters output in step (2) and the longitudinal and lateral forces of the tires output in step (3), the longitudinal vehicle speed is incorporated into the state variables. The preprocessed data in step (1) is used to construct the observation variables and form the state equation and observation equation. (5) Based on the state equation in step (4), the state estimate and covariance matrix are constructed by sampling points, and the sampling points are updated in time by the state transition matrix to predict the state mean and covariance matrix at the next moment. (6) Based on the predicted state mean and covariance matrix obtained in step (5), calculate the correlation coefficient of the innovation sequence, and determine whether the filter diverges by setting a critical value. The innovation sequence is obtained by the difference between the actual observed value and the predicted observed value of the observation equation in step (4). If the filter is determined to be diverging, the prediction covariance matrix obtained in step (5) is corrected using the covariance of the innovation sequence and the covariance of the predicted observations. (7) Based on the covariance matrix corrected in step (6), calculate the mean and covariance of the predicted observations, and then calculate the Kalman gain using the preprocessed data from step (1) to update the state estimate and covariance matrix; at the same time, based on the Sage-Husa adaptive mechanism, update the mean and covariance matrix of the measurement noise using the innovation sequence from step (6). (8) Combining the state estimate, covariance matrix and noise characteristic update results obtained in step (7), optimize the state estimate and covariance matrix again, and output the longitudinal vehicle speed estimate.

2. The method according to claim 1, characterized in that: The dynamic equations of the seven-degree-of-freedom vehicle model are as follows: Where m is the total mass of the vehicle; β is the sideslip angle; u is the longitudinal velocity; v is the lateral velocity; ω is the yaw rate; a x It is longitudinal acceleration; a y Γ is the lateral acceleration; Γ is the torque about the z-axis; I z δ is the moment of inertia of the car about the z-axis; δ is the wheel steering angle; F xi It is the longitudinal force of the tire; F zi This refers to the lateral force of the tire; i = 1, 2, 3, 4 represent the positions of the wheel at the left front, right front, left rear, and right rear, respectively; δ i The steering angle δ of a specific wheel in response to the wheel's steering angle; a and b represent the distances from the center of mass to the front and rear axles, respectively; t f ,t r This indicates the distance between the front or rear wheels.

3. The method according to claim 2, characterized in that: The formulas for simulating longitudinal and lateral forces in a tire using the Dugoff tire model are as follows: Where μ is the current road adhesion coefficient; F z ε is the vertical load on each tire; L is the introduced boundary condition; ε is the speed influence factor; C x and C y These refer to the longitudinal stiffness and lateral stiffness of the wheel, respectively.

4. The method of claim 2, wherein: The state equations and observation equations are designed as follows: State variable x(t) = [v x v y rΓa x a y ] T Observed variable y(t)=[a x a y r] T Based on the seven-DOF vehicle model and the Dugoff tire model, the state equation is: System observation equations: Where k is the discrete time; Λ is the noise driving matrix; X k Y is the state matrix at time k; k is the observation matrix at time k; W(k) is the input white noise; V(k) is the observation noise.

5. The method of claim 2, wherein: Step (5) includes constructing a series of sampling points for approximating the system state distribution by sampling the state estimates and covariance matrix, with the number of sampling points being 2n+1, where n is the dimension of the state vector: in This is the current estimate; P k Here is the state covariance matrix; λ = α 2 (n+κ)-n, where α and κ are the parameters for adjusting the release of sampling points, respectively.

6. The method of claim 5, wherein: Step (5) also includes updating the sampling points over time and calculating the next state of the Sigma point: χ i,k|k-1 = f(χ i,k-1 ) Where f(·) is the state transition matrix; Predicted state mean and covariance matrix: Where, ω i and ω i c These are the state weights and covariance weights of the sampling points, respectively; Q k Let be the system noise covariance matrix.

7. The method of claim 2, wherein: In step (6), the divergence calculation is introduced: a system is described by the following state space: X k+1 =φX k +ΛW k Y k =HX k +V k Where k is the discrete time; φ is the state transition matrix; H is the observation matrix; Λ is the noise driving matrix; X k Y is the state matrix at time k; k W is the observation matrix at time k; k It is input white noise; V k It is observation noise; assuming W k and V k The mean is zero, W k and V k The covariance matrix is ​​Q k and R k ,have: We can obtain: The estimated variance of the measured values ​​is: The covariance matrix can be represented as: Cov(ε k+1 ·ε k+1 T )=HP k+1 H T +R j+1 Where P k+1 It is the covariance matrix of the prediction error; when the estimation converges, ε k It is white noise that follows a normal distribution, therefore we can obtain the following equation, where ξ xt This is the correlation coefficient, with values ​​between 0 and 1. The correlation coefficient is used to determine divergence. The critical value for filter divergence is set to ξ0. If the following equation is satisfied, the filter diverges:

8. The method of claim 7, wherein: Step (7) includes determining the divergence of the filter; if the filter diverges, correcting the covariance matrix P of the prediction error. k|k-1 , Where ρ is in the range of 0-1.

9. The method of claim 8, wherein: Step (7) also includes calculating the predicted observations and their mean and covariance matrices: Where h(·) is the observation function; R k To observe the noise covariance matrix; Calculate the Kalman gain matrix and update the state estimate and covariance matrix. K k =P xy P yy -1 10. The method according to claim 9, characterized in that: Step (7) also includes updating the mean and covariance matrix of the measurement noise based on the Sage-Husa adaptive mechanism: Where, d k =1-γ / 1-γ k γ is the fading factor, with a value range of 0-1; H k is the Jacobian matrix of the nonlinear measurement matrix.

Images