Dynamic gradient estimation method based on multi-source information fusion

The dynamic slope estimation method integrates vehicle dynamics and kinematics models with unscented Kalman filters and SVD to address single-direction limitations, achieving accurate and universal slope estimation by decoupling vehicle roll and road slope.

CN119089580APending Publication Date: 2024-12-06NANJING UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411185636.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Current slope estimation methods in vehicles are limited to single-direction estimation (longitudinal or lateral) and lack universality in complex conditions, particularly under varying vehicle dynamics and sensor accuracy.

Method used

A dynamic slope estimation method based on multi-source information fusion, integrating vehicle dynamics and kinematics models with unscented Kalman filters, using singular value decomposition (SVD) to enhance accuracy and stability, and an interactive multi-model filtering algorithm to combine estimates from both models.

Benefits of technology

This method achieves accurate and universal slope estimation in any direction by decoupling vehicle roll and road slope, reducing calculation complexity and improving accuracy through model fusion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089580A_ABST
    Figure CN119089580A_ABST
Patent Text Reader

Abstract

The invention discloses a dynamic gradient estimation method based on multi-source information fusion, which is suitable for various vehicles such as distributed driving electric vehicles and the like, and comprises the following steps: establishing a kinematics model and a dynamics model of a vehicle about a road gradient; a vehicle-road roll dynamic model is established, and superposition of vehicle roll and a road slope angle is decoupled; establishing a slope estimator based on a kinematic model and a dynamic model; and fusing the results of the slope estimators by using an interactive multi-model algorithm, and outputting a road slope estimated value. According to the dynamic slope estimation method provided by the invention, the limitation of a current single slope estimation method is overcome, and more accurate road slope information can be provided in real time when the vehicle moves dynamically, so that the accuracy of road slope estimation is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of interactive multi-model fusion of vehicle dynamics model and kinematics model, in particular to a dynamic slope estimation method based on multi-source information fusion. Background Art

[0002] With the electrification and intelligent technology advancement in the automotive industry, there is an increasing trend towards the widespread adoption of active safety systems in mass-produced vehicles. The integration of multiple control systems has proven to be an effective way to reduce costs. Although these individual control systems are specifically designed for driving on flat roads, the road gradient, as a key external variable, directly affects the estimation of key driving parameters. Therefore, accurate estimation of road gradient becomes a key element of comprehensive control and plays a vital role in improving overall system performance.

[0003] Current slope estimation algorithms mainly focus on the estimation of longitudinal and lateral road slopes. The estimation methods used can be roughly divided into two categories: direct methods, which use on-board sensor data for direct estimation; and indirect methods, which rely on vehicle models for estimation.

[0004] The direct method mainly involves directly estimating the road slope using GPS data and information obtained from the inertial measurement unit. Although the direct method relies heavily on measurement data and is not affected by vehicle parameters or road conditions, it has high external environment requirements for GPS signals. In addition, the slope information in the IMU signal is coupled with the vehicle's own roll information. Therefore, the accuracy of the slope estimation result is easily affected by changes in vehicle driving conditions and sensor accuracy.

[0005] Indirect methods mainly rely on vehicle dynamics for estimation. By building a vehicle model under different working conditions, accurate slope estimation is obtained through a multi-sensor information fusion filtering algorithm. These methods have low sensor dependence and obtain the necessary data through the vehicle dynamics model and the vehicle CAN bus. However, it should be noted that in order to ensure the effectiveness of their estimation, the vehicle must have a certain level of acceleration, and larger errors will occur when the acceleration is low.

[0006] However, it should be emphasized that the above road slope estimation methods have limitations under specific conditions. The current slope estimation mainly focuses on the estimation of a single direction such as longitudinal and lateral. When facing complex working conditions in multiple directions, there is a lack of a more universal slope estimation model. Summary of the invention

[0007] The purpose of the present invention is to provide a dynamic slope estimation method based on multi-source information fusion to solve the problem that the current slope estimation is concentrated in a single direction, longitudinal or lateral, so as to achieve more universal dynamic slope estimation.

[0008] The technical solution to achieve the purpose of the present invention is:

[0009] A dynamic slope estimation method based on multi-source information fusion, comprising:

[0010] Step 1: Establish a vehicle dynamics model and a vehicle kinematics model for a vehicle moving on a slope;

[0011] Step 2: Establishing a vehicle-road roll dynamics model;

[0012] Step 3, constructing an unscented Kalman estimation model based on the vehicle dynamics model and the vehicle-road roll dynamics model, and constructing an unscented Kalman estimation model based on the vehicle kinematics model and the vehicle-road roll dynamics model, and using an unscented Kalman filter algorithm to obtain road slope estimation values ​​of the single unscented Kalman estimation model;

[0013] Step 4: According to the road slope estimation values ​​of the two single unscented Kalman estimation models, based on the interactive multi-model filtering algorithm, the two unscented Kalman filter estimation models are fused to output the final road slope estimation value.

[0014] Compared with the prior art, the present invention has the following significant advantages:

[0015] (1) The present invention discloses a dynamic slope estimation method based on multi-source information fusion, which solves the problem that the current slope estimation is concentrated in a single direction, longitudinal or lateral, and realizes dynamic slope estimation in any direction on the slope.

[0016] (2) The present invention describes a dynamic slope estimation method based on multi-source information fusion, which establishes a vehicle dynamics model, a vehicle kinematic model and a vehicle-road kinematic model on the slope. According to the acceleration information measured by the inertial sensor, the vehicle roll angle and the road slope are decoupled, thereby reducing the amount of calculation and improving the accuracy.

[0017] (3) The present invention describes a dynamic slope estimation method based on multi-source information fusion, which uses an interactive multi-model fusion algorithm to fuse two types of model slope information to obtain a final road slope estimation value as the final road slope output, thereby improving the accuracy of the estimation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the general movement form of a vehicle on a slope.

[0019] Figure 2 For the vehicle model.

[0020] Figure 3 is the vehicle-road kinematic model.

[0021] Figure 4 This is the overall process of the interactive multi-model filtering algorithm adopted by the present invention.

[0022] Figure 5 This is a before-and-after comparison between the algorithm used in the present invention and the Kalman filter algorithm of a single model. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below in conjunction with the accompanying drawings. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0024] like Figure 4 As shown, this example provides a dynamic slope estimation method based on multi-source information fusion, which includes the following five steps:

[0025] Step 1: Establish the dynamic model and kinematic model of the vehicle moving on the slope. Figure 1 , Figure 2 shown.

[0026] The vehicle's acceleration, angular acceleration and other information are obtained through inertial sensors, and the tire force is obtained by the magic tire formula. The X direction is the vertical slope direction, the Y direction is the parallel slope direction, the u direction is the vehicle's forward speed direction, and the v direction is the vehicle's lateral speed direction.

[0027] Based on the corresponding vehicle inherent parameters, the vehicle kinematic model in the u and v directions and the vehicle dynamic model with three degrees of freedom are constructed respectively.

[0028] Vehicle kinematic model:

[0029]

[0030] Where: a u_sen Indicates the acceleration value in the u direction measured by the sensor, a v_sen Indicates the acceleration value in the v direction measured by the sensor, a u Represents the actual acceleration value of the vehicle in the u direction, a v Represents the actual acceleration value of the vehicle in the v direction. ν u represents the vehicle's speed in the u direction, ν v represents the speed of the vehicle in the v direction, represents the derivative of velocity, g represents the velocity due to gravity, represents the vehicle roll angle, φ s Indicates the road slope, φ v represents the angle between the vehicle's forward u direction and the X direction, ψ represents the vehicle's yaw angle, Indicates the vehicle's yaw rate.

[0031] Longitudinal dynamic equation:

[0032]

[0033] Lateral dynamics equation:

[0034]

[0035] Yaw dynamics equation:

[0036]

[0037] Where: Respectively represent the speed ν u ,ν v The derivative of represents the vehicle roll angle, φ s Indicates the road slope, φ v It represents the angle between the vehicle's forward u direction and the X direction. represents the vehicle yaw rate, represents the vehicle's yaw angular acceleration. δ represents the front wheel turning angle, m represents the vehicle mass, g represents the gravitational acceleration speed, and F xp represents the longitudinal tire force on the pth tire of the vehicle (p = 1, 2, 3, 4), F yp represents the directional tire force on the pth tire of the vehicle (p = 1, 2, 3, 4), l r Indicates the distance between the vehicle's center of mass and the rear axle, l f Indicates the distance between the vehicle's center of mass and the front axle, 2d r Indicates the rear track of the vehicle, I z Represents the vehicle's yaw moment of inertia.

[0038] Step 2: Establish a vehicle-road kinematic model, consider the coupling relationship between the slope and the vehicle roll angle in the inertial sensor signal, and analyze the vehicle-road model in detail. Since the change in road slope during vehicle motion is relatively small, it is reasonable to assume that the road slope remains unchanged. Figure 3 As shown, the vehicle-road kinematic model is expressed as follows:

[0039]

[0040] Where: m s represents the sprung mass, g represents the gravitational acceleration, I eq represents the moment of inertia of the vehicle's sprung mass, h represents the height from the vehicle's center of mass to the roll center, represents the vehicle roll angle, φ s Indicates the road slope, φ v It represents the angle between the vehicle's forward u direction and the X direction. represents the derivative of velocity, represents the vehicle yaw rate, represents the vehicle yaw angular acceleration, is the body roll angular acceleration, is the body roll stiffness, is the body roll damping coefficient, a v_sen Indicates the acceleration value of the sensor in the v direction, is the sensor roll angle acceleration.

[0041] Step 3: Based on the vehicle dynamics model, vehicle kinematics model and road-vehicle kinematics model, unscented Kalman filter algorithms are constructed to estimate the road slope.

[0042] Based on the vehicle dynamics model, the unscented Kalman estimation model is constructed according to formulas (2)-(5):

[0043]

[0044] in: is acting on x k-1 The state transition model on is acting on x k-1 The observation model on is the state equation, is the observation equation, the state matrix Input Matrix Observation Matrix The subscript k represents the number of iterations. is the process noise; To measure noise.

[0045] Based on the vehicle kinematic model, the unscented Kalman estimation model is constructed according to formulas (1) and (5):

[0046]

[0047] in: is acting on x k-1 The state transition model on is acting on x k-1 The observation model on is the observation equation, is the state equation, the state matrix Input Matrix Observation Matrix represents the number of iteration steps, is the process noise; To measure noise.

[0048] To alleviate the problem of reduced filtering accuracy or potential non-convergence encountered by the Unscented Kalman Filter (UKF) in scenarios characterized by strong nonlinearity or ambiguity in the state estimate, the SVD-UKF filtering algorithm is used. The algorithm improves the selection of the state covariance matrix in the UKF through the application of singular value decomposition (SVD). This method ensures the non-negativity and stability of the covariance matrix during iteration.

[0049] In the unscented Kalman filter, the covariance matrix P k-1 Perform singular value decomposition:

[0050]

[0051] Where P k-1 Represents the estimated value of the state at step k-1 The covariance of k-1 , D k-1 is the decomposed matrix.

[0052] In equation (8), the singular value decomposition is derived from the previous state iteration, and the sigma point of the state is calculated and updated as follows:

[0053]

[0054] Among them, i,k-1 represents the i-th sigma point corresponding to the k-1th step, i is the value range (1 to 2n), and n represents the state variable x k The dimension of is the estimated value of the k-1th step state, λ=α 2 (n+κ)-n is the scaling parameter. α, β, k are candidate parameters. In this example, α=0.001, β=2, k=0

[0055] Calculate weights:

[0056]

[0057] Where i is in the range of 1 to 2n. represents the weight of the predicted mean value corresponding to the i-th sigma point, Represents the weight of the covariance corresponding to the i-th sigma point.

[0058] In the k+1 step, calculate the predicted mean and the covariance matrix P(k+1|k)

[0059]

[0060] Among them, X (i)(k+1|k) represents the predicted mean value corresponding to the i-th sigma point, and Q represents the covariance matrix of the process noise.

[0061] Based on the predicted mean and covariance matrix, the UT transform is used again to generate a new set of 2n+1 sigma points. These points are substituted into the observation equation to obtain the predicted observations for k+1 steps. The predicted mean of the system observations at k+1 steps is further calculated and the covariance matrix

[0062]

[0063] in, represents the observed mean, Z (i) (k+1|k) represents the observed value at step i, Represents x k and The covariance matrix of express and The covariance matrix of , where j = 1, 2. R represents the covariance matrix of the measurement noise.

[0064] Calculate the Kalman gain K(k+1) of the k+1 step:

[0065]

[0066] Finally, update the system state covariance matrix P(k+1|k+1)

[0067]

[0068] in, is the system state at step k+1, Z(k+1) is the observation value at step k+1, and P(k+1|k+1) is the covariance matrix at step k+1 after update.

[0069] By constructing the unscented Kalman filter algorithm separately, the road slope estimation value based on a single model is obtained

[0070] Step 4: Estimated road slopes from the two single models in step 3 Based on the interactive multi-model filtering algorithm, the two unscented Kalman filter estimation models (6) (7) are fused to output the final road slope estimation value.

[0071] In view of the limitations of a single model, the present invention adopts an interactive multi-model algorithm to fuse the estimated values ​​of two models with appropriate weights. The steps are as follows:

[0072] Initialize the Markov model transfer matrix G and the probability of each filter. Use the kinematic and dynamic models to calculate the optimal estimate and estimated covariance at step k-1, and evaluate the predicted probability and mixing probability of the two models. And calculate the predicted probability c of the model in step 3 j With the mixing probability μ hj .

[0073]

[0074] Where P is the model transformation matrix, h = 1, 2 is the number of transformation matrix rows, and p hj (p 11 ,p 12 ,p 21 ,p 22 ) represents the model transition probability, which is determined by the Markov matrix. h (k-1) is the update probability of the model at the k-1th step, which is calculated by (19) and represents the number of iterations. j = 1, 2 represents the model sequence. Here, the estimation method based on the kinematic model is designated as sequence 1, and the estimation method based on the dynamic model is designated as sequence 2. μ hj (k-1|k-1) is the parameter used to calculate equation (17).

[0075] By interacting to obtain the mixed state estimate for each of the two models and the mixed covariance estimate P 0j (k-1|k-1).

[0076]

[0077] Where: P h (k-1|k-1) represents the state estimation and covariance matrix of the target at the k-1th step, Represents the posterior state estimate at the k-1th step.

[0078] Apply SVD-UKF filtering to the two models separately and update the predicted states and covariances.

[0079] Update the model probability of each filter using the likelihood function.

[0080] The maximum likelihood function matching the j model at step k is:

[0081]

[0082] in

[0083]

[0084] v j (k), S j(k) is the maximum likelihood function Λ j (k) is the required parameter, Z(k) is the observed value of the model, H(k) is the transformation matrix, is the prior estimate at step k, P j (k|k-1) is the covariance matrix of the kth step, and R(k) is the noise matrix.

[0085] The update probability of model j is expressed as:

[0086] μ j (k) = Λ j (k)c j / η (20)

[0087] Where: η is the normalization constant,

[0088] The estimation results from the two models are weighted and fused using the model probability to determine the final slope estimate φ s .

[0089]

[0090] in, is the fusion estimation result, is the estimated result of the jth model.

[0091] Simulation verification: The dynamic slope estimation method based on multi-source information fusion proposed in the present invention is compared with the unscented Kalman filter algorithm based on a single model. Figure 5 As shown, the simulation test is a double lane change simulation verification on a slope road. The vehicle performs a double lane change experiment on a slope road with a slope of 10% and maintains a speed of 40km / h.

[0092] Between 4.5 and 14.5 seconds, the vehicle dynamics gradually increase and the proposed fusion algorithm shows some fluctuations. However, these fluctuations are significantly reduced compared to the single model. The maximum error of the fusion estimation method after convergence is only 0.5%, which effectively verifies the stability and accuracy of the algorithm proposed in this invention.

Claims

1. A dynamic slope estimation method based on multi-source information fusion, characterized in that: include: Step 1: Establish a vehicle dynamics model and a vehicle kinematics model for a vehicle moving on a slope; Step 2: Establishing a vehicle-road roll dynamics model; Step 3, constructing an unscented Kalman estimation model based on the vehicle dynamics model and the vehicle-road roll dynamics model, and constructing an unscented Kalman estimation model based on the vehicle kinematics model and the vehicle-road roll dynamics model, and using an unscented Kalman filter algorithm to obtain road slope estimation values ​​of the single unscented Kalman estimation model; Step 4: According to the road slope estimation values ​​of the two single unscented Kalman estimation models, based on the interactive multi-model filtering algorithm, the two unscented Kalman filter estimation models are fused to output the final road slope estimation value.

2. The dynamic slope estimation method based on multi-source information fusion according to claim 1 is characterized in that: The dynamic model of the vehicle moving on a slope is established as: Where: a u_sen Indicates the acceleration value in the u direction measured by the sensor, a v_sen Indicates the acceleration value in the v direction measured by the sensor, a u Represents the actual acceleration value of the vehicle in the u direction, a v Represents the actual acceleration value of the vehicle in the v direction, ν u represents the vehicle's speed in the u direction, ν v represents the speed of the vehicle in the v direction, g represents the acceleration due to gravity, represents the vehicle roll angle, φ s Indicates the road slope, φ v represents the angle between the vehicle's forward u direction and the X direction, ψ represents the vehicle's yaw angle; the X direction is the vertical slope direction, the u direction is the vehicle's forward speed direction, and the v direction is the vehicle's lateral speed direction.

3. The dynamic slope estimation method based on multi-source information fusion according to claim 1 is characterized in that: The vehicle kinematic model is: Where: u represents the vehicle's speed in the u direction, ν v represents the speed of the vehicle in the v direction, represents the vehicle roll angle, φ s Indicates the road slope, φ v represents the angle between the vehicle's forward u direction and the X direction, ψ represents the vehicle's yaw angle, δ represents the front wheel turning angle, m represents the vehicle's mass, g represents the gravitational acceleration speed, and F xp represents the longitudinal tire force on the pth tire of the vehicle, F yp represents the directional tire force on the pth tire of the vehicle, l r Indicates the distance between the vehicle's center of mass and the rear axle, l f Indicates the distance between the vehicle's center of mass and the front axle, 2d r Indicates the rear track of the vehicle, I z Represents the vehicle's yaw moment of inertia; the X direction is the vertical slope direction, the u direction is the vehicle's forward speed direction, and the v direction is the vehicle's lateral speed direction.

4. The dynamic slope estimation method based on multi-source information fusion according to claim 1 is characterized in that: The vehicle-road roll dynamics model is established as: Where: m s represents the sprung mass, g represents the gravitational acceleration, I eq represents the moment of inertia of the vehicle's sprung mass, h represents the height from the vehicle's center of mass to the roll center, represents the vehicle roll angle, φ s Indicates the road slope, φ v Indicates the angle between the vehicle's forward u direction and the X direction, v u represents the vehicle's speed in the u direction, v v represents the vehicle's v-direction velocity, ψ represents the vehicle's yaw angle, is the body roll stiffness, is the body roll damping coefficient, a v_sen Indicates the acceleration value of the sensor in the v direction, is the sensor roll angle acceleration, the X direction is the vertical slope direction, the u direction is the vehicle forward speed direction, and the v direction is the vehicle lateral speed direction.

5. The dynamic slope estimation method based on multi-source information fusion according to claim 1 is characterized in that: The unscented Kalman estimation model based on the vehicle dynamics model and the vehicle-road roll dynamics model is constructed as follows: in: is the state transition model, is the observation model, the state matrix Input Matrix Observation Matrix is the process noise; To measure the noise, F xp represents the longitudinal tire force on the pth tire of the vehicle, F yp represents the lateral tire force on the pth tire of the vehicle, v u represents the vehicle's speed in the u direction, v v represents the speed of the vehicle in the v direction, represents the vehicle roll angle, φ s Indicates the road slope, φ v It represents the angle between the vehicle's forward u direction and the X direction, δ represents the front wheel turning angle, the X direction is the vertical slope direction, the u direction is the vehicle's forward speed direction, and the v direction is the vehicle's lateral speed direction.

6. The dynamic slope estimation method based on multi-source information fusion according to claim 1 is characterized in that: The unscented Kalman estimation model based on the vehicle kinematic model and the vehicle-road roll dynamics model is constructed as follows: in: is acting on x k-1 The state transition model on is acting on x k-1 The observation model on the state matrix Input Matrix Observation Matrix is the process noise; To measure the noise, v u represents the vehicle's speed in the u direction, v v represents the speed of the vehicle in the v direction, represents the vehicle roll angle, φ s Indicates the road slope, φ v Indicates the angle between the vehicle's forward u direction and the X direction, a u_sen Indicates the acceleration value in the u direction measured by the sensor, a v_sen It represents the acceleration value in the v direction measured by the sensor, ψ represents the yaw angle of the vehicle, the X direction is the vertical slope direction, the u direction is the vehicle forward speed direction, and the v direction is the vehicle lateral speed direction.

7. The dynamic slope estimation method based on multi-source information fusion according to claim 1 is characterized in that: Step 4 specifically includes: Initialize the Markov model transfer matrix and the probability of each filter, calculate the optimal estimate and estimated covariance at the k-1th step, and evaluate the predicted probability and mixing probability of the two models; The interaction is used to obtain the mixture state estimate and mixture covariance estimate for each of the two models; Apply SVD-UKF filtering to two single unscented Kalman estimation models respectively to update the predicted state and covariance; Update the model probability of each filter using the likelihood function; The estimation results from the two models are weighted and fused using the model probabilities to determine the final slope estimate.