CN-121980793-A - Vehicle centroid cornering angle estimation method and system integrating Gaussian process uncertainty and unscented Kalman filtering

CN121980793ACN 121980793 ACN121980793 ACN 121980793ACN-121980793-A

Abstract

The invention discloses a vehicle centroid slip angle estimation method and system integrating Gaussian process uncertainty and unscented Kalman filtering. The method comprises the steps of firstly establishing a nominal vehicle model based on nonlinear three-degree-of-freedom dynamics, calculating tire force by adopting a magic tire formula with longitudinal and lateral coupling, utilizing multi-working-condition data to train a Gaussian process in machine learning in an off-line mode to learn residual errors and uncertainty between a predicted value and a true value of the nominal vehicle model, fusing output of the nominal vehicle model and a predicted mean value of the Gaussian process to construct a hybrid vehicle dynamics model, designing a self-adaptive unscented Kalman filter, taking the hybrid vehicle dynamics model as a state prediction equation, dynamically injecting residual error variance predicted by the Gaussian process into a priori error covariance matrix to realize self-adaptive adjustment of process noise, estimating longitudinal and transverse speeds through the filter, and calculating centroid side deflection angles in real time. The invention utilizes uncertainty quantization to improve the robustness and self-adaptive capacity of the filter.

Inventors

XIAO FENG
ZHAO ZIBO
JIN LIQIANG
LI YANJIE
YANG JUNYAO
GAO YUNZHAO

Assignees

吉林大学

Dates

Publication Date: 20260505
Application Date: 20260123

Claims (10)

1. A vehicle mass center slip angle estimation method integrating Gaussian process uncertainty and unscented Kalman filtering is characterized by comprising the following steps: s1, establishing a nominal vehicle dynamics model which is a nonlinear three-degree-of-freedom monorail vehicle dynamics equation comprising longitudinal, transverse and yaw degrees of freedom, and calculating tire force by adopting a magic tire formula taking longitudinal and lateral coupling into consideration; S2, constructing and offline training through a machine learning algorithm based on a multi-working condition vehicle dynamics data set, wherein the machine learning algorithm adopts a Gaussian process model, the input of the Gaussian process model is a feature vector containing the vehicle state, and the output feature vector is a posterior prediction mean and posterior prediction variance of a residual error between a state variable of a nominal vehicle dynamics model and a state variable of a real vehicle; s3, combining the nominal vehicle dynamics model with the residual error model to obtain a hybrid vehicle dynamics model for centroid slip angle estimation; S4, constructing a self-adaptive unscented Kalman filter, taking a hybrid vehicle dynamics model as a state prediction equation thereof, and establishing a corresponding measurement equation by combining observation information obtained by a vehicle-mounted sensor; S5, acquiring real-time measurement data of a current vehicle during online running, injecting residual variances predicted by a residual model at the current moment into calculation of an priori state error covariance matrix in a prediction stage of an adaptive unscented Kalman filter, estimating the longitudinal speed, the transverse speed and the yaw rate of the vehicle by using the adaptive unscented Kalman filter to obtain a posterior estimated value, and calculating the real-time centroid slip angle of the vehicle according to the posterior estimated value.
2. The method for estimating a vehicle centroid slip angle by combining gaussian process uncertainty and unscented kalman filter according to claim 1, wherein in step S1, the nominal vehicle dynamics model is: wherein the state variables , For the longitudinal speed of the vehicle, For the transverse velocity of the vehicle, Control quantity for yaw rate of vehicle , Is the front wheel corner; Is that The state variable of the moment of time, Is that The state variable of the moment of time, Is that The amount of time control is controlled by the time, Is a discrete state transfer function.
3. The method for estimating the vehicle centroid cornering angle by combining the Gaussian process uncertainty and the unscented Kalman filter according to claim 1, wherein in the step S1, the tire force is calculated by adopting a magic tire formula considering longitudinal and lateral coupling, specifically: Wherein, the For the longitudinal force of the front axle of the vehicle, For the lateral force of the front axle of the vehicle, For the longitudinal force of the rear axle of the vehicle, For the lateral force of the rear axle of the vehicle, For the slip ratio of the front axle tire, For the slip ratio of the rear axle tire, , , , , , 。
4. The method for estimating the vehicle centroid slip angle by fusing gaussian process uncertainty and unscented kalman filter according to claim 2, wherein in step S2, the trained gaussian process model is used as a residual model, and the expression is: Wherein, the As a model of the longitudinal velocity residual, Is the posterior mean value of the longitudinal velocity residual, As a longitudinal speed residual post-test variance, As a model of the lateral velocity residual error, Is the lateral velocity residual posterior mean value, As a side-to-side velocity residual post-test variance, As a model of the yaw rate residual error, Is the posterior mean value of the yaw rate residual, Is the yaw rate residual error post-test variance.
5. The method for estimating a vehicle centroid slip angle by fusing gaussian process uncertainty and unscented kalman filter according to claim 1, wherein in step S4, the adaptive unscented kalman filter is constructed, and a hybrid vehicle dynamics model is used as a state prediction equation thereof, and the specific expression is: extraction matrix The method comprises the following steps: Wherein, the In order to extract the coefficients of the coefficients, As a threshold value for the lateral acceleration, Is a yaw rate threshold; Is that The state vectors of the dynamics model of the vehicle are mixed at the moment, Is that And when the threshold value is exceeded, compensating the nominal vehicle dynamics model by the residual model.
6. The method for estimating the vehicle centroid slip angle by combining Gaussian process uncertainty and unscented Kalman filtering according to claim 1, wherein in step S5, the residual variance predicted by the residual model at the current moment is injected into the calculation of the prior state error covariance matrix, and the calculation formula is as follows: Process noise covariance matrix The expression is: Wherein, the Is the first Weight coefficients of Sigma points in covariance synthesis; Is the first A priori predictions of the Sigma points, As a prior state average value, As a basis process noise covariance matrix, For the matrix of adjustable scaling parameters, Is a cognitive uncertainty process noise covariance matrix.
7. The method for estimating the centroid slip angle of the vehicle by combining the uncertainty of the Gaussian process and the unscented Kalman filter according to claim 1, wherein in the step S5, the real-time centroid slip angle of the vehicle is calculated according to the posterior estimated value, and the specific calculation formula is as follows: Wherein, the A posterior estimated value of the longitudinal speed at the moment k; is a posterior estimate of the lateral velocity at time k.
8. A vehicle centroid slip angle estimation system that fuses gaussian process uncertainty with unscented kalman filtering, comprising: The nominal dynamics modeling module is used for establishing a nominal vehicle dynamics model, wherein the nominal vehicle dynamics model is a nonlinear three-degree-of-freedom monorail vehicle dynamics equation comprising longitudinal, transverse and yaw degrees of freedom, and a magic tire formula taking longitudinal and lateral coupling into consideration is adopted to calculate tire force; The device comprises a residual error learning and uncertainty quantization module, a post-test prediction mean and a post-test prediction variance of residual errors between state variables of a nominal vehicle dynamics model and real vehicle, wherein the residual error learning and uncertainty quantization module is used for constructing and offline training through a machine learning algorithm based on a multi-working-condition vehicle dynamics data set, the machine learning algorithm adopts a Gaussian process model, the input of the Gaussian process model is a feature vector containing the vehicle state, and the output feature vector is a post-test prediction mean and a post-test prediction variance of the residual errors between the state variables of the nominal vehicle dynamics model and the real vehicle; the hybrid model fusion module is used for combining the nominal vehicle dynamics model and the residual model to obtain a hybrid vehicle dynamics model for centroid slip angle estimation; The self-adaptive filter construction module is used for constructing a self-adaptive unscented Kalman filter, taking a hybrid vehicle dynamics model as a state prediction equation thereof, and combining observation information obtained by a vehicle-mounted sensor to establish a corresponding measurement equation; The mass center slip angle calculation module is used for acquiring real-time measurement data of a current vehicle during online running, injecting residual variances predicted by a residual model at the current moment into calculation of a priori state error covariance matrix in a prediction stage of the adaptive unscented Kalman filter, estimating the longitudinal speed, the transverse speed and the yaw rate of the vehicle by using the adaptive unscented Kalman filter to obtain a posterior estimated value, and calculating the real-time mass center slip angle of the vehicle according to the posterior estimated value.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and operable on the processor, wherein the processor, when executing the computer program, implements a method of vehicle centroid slip angle estimation incorporating gaussian process uncertainty and unscented kalman filtering as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon a computer program for causing a computer to perform a vehicle centroid slip angle estimation method that combines Gaussian process uncertainty and unscented Kalman filtering as set forth in any one of claims 1-7.

Description

Vehicle centroid cornering angle estimation method and system integrating Gaussian process uncertainty and unscented Kalman filtering Technical Field The invention relates to the technical field of electric vehicle state estimation, in particular to application of a Gaussian process modeling technology in machine learning in the field, and specifically relates to a vehicle centroid slip angle estimation method and system integrating Gaussian process uncertainty and unscented Kalman filtering. Background The vehicle mass center slip angle is a key state parameter for representing the transverse dynamics behavior of the vehicle, evaluating the yaw stability of the vehicle and identifying the driving limit, and the accurate estimation of the key state parameter has important significance for the safety and performance optimization of a vehicle stability control system, an advanced driving auxiliary system and an automatic driving system. At present, the acquisition of the centroid slip angle mainly depends on an observation method based on the combination of a vehicle-mounted low-cost sensor (such as an inertia measurement unit) and a vehicle dynamics model, and the method generally utilizes a Kalman filter or an improved algorithm thereof, so that the method has a good estimation effect under the conventional driving working condition. However, existing estimation methods based on physical models still face a series of challenges. Firstly, the vehicle dynamics system has the characteristics of strong nonlinearity, time variation, parameter uncertainty and the like, and particularly under the limit working conditions of tire force saturation, large cornering, low adhesion road surface and the like, the traditional linear or simplified nonlinear model is difficult to accurately describe the real behavior of the system, so that the model prediction error is obviously increased. Secondly, due to complexity of a vehicle running environment and simplification of system modeling, the model has problems of unmodeled dynamics, parameter perturbation and the like, and therefore robustness and reliability of state estimation are affected. In terms of filter estimation, common methods such as extended kalman filtering and unscented kalman filtering generally rely on a preset process noise covariance matrix, and consider it as a constant or a simple time-varying function in the whole estimation process. The assumption has certain rationality when the model is well matched with a real system, but under the working condition that the model mismatch is serious, the fixed noise covariance cannot accurately reflect the uncertainty of model prediction, so that excessive confidence of a filter on a physical model is easily caused, and further, state estimation deviation and even filtering divergence are caused. Disclosure of Invention The invention aims to provide a vehicle centroid slip angle estimation method and system integrating Gaussian process uncertainty and unscented Kalman filtering, which are used for solving the problems that a single physical model is insufficient in precision under a vehicle limit working condition and a noise covariance matrix in the traditional unscented Kalman filtering process cannot be self-adaptive. In order to achieve the purpose, the technical scheme provided by the invention is that the vehicle centroid slip angle estimation method integrating Gaussian process uncertainty and unscented Kalman filtering is characterized by comprising the following steps: s1, establishing a nominal vehicle dynamics model which is a nonlinear three-degree-of-freedom monorail vehicle dynamics equation comprising longitudinal, transverse and yaw degrees of freedom, and calculating tire force by adopting a magic tire formula taking longitudinal and lateral coupling into consideration; S2, constructing and offline training through a machine learning algorithm based on a multi-working condition vehicle dynamics data set, wherein the machine learning algorithm adopts a Gaussian process model, the input of the Gaussian process model is a feature vector containing the vehicle state, and the output feature vector is a posterior prediction mean and posterior prediction variance of a residual error between a state variable of a nominal vehicle dynamics model and a state variable of a real vehicle; s3, combining the nominal vehicle dynamics model with the residual error model to obtain a hybrid vehicle dynamics model for centroid slip angle estimation; S4, constructing a self-adaptive unscented Kalman filter, taking a hybrid vehicle dynamics model as a state prediction equation thereof, and establishing a corresponding measurement equation by combining observation information obtained by a vehicle-mounted sensor; S5, acquiring real-time measurement data of a current vehicle during online running, injecting residual variances predicted by a residual model at the current moment into calculation of an priori state error cov