CN111547059A - A distributed drive electric vehicle inertial parameter estimation method - Google Patents

Abstract

The invention relates to a distributed driving electric vehicle inertia parameter estimation method, which comprises the specific steps of firstly considering the change of vehicle inertia parameters caused by uncertain load parameters, establishing a three-degree-of-freedom whole vehicle dynamics model, selecting a nonlinear tire model of a magic formula, designing a state and parameter estimation system of a double-adaptive unscented Kalman filter, and then determining a double-adaptive unscented Kalman filter observer, thereby realizing the estimation of vehicle inertia parameters such as the longitudinal speed of a vehicle, the vehicle mass center and lateral deviation angle and the like, and the vehicle mass, the yaw rotation inertia, the distance from the mass center to the front axle of the vehicle. The method is based on the vehicle dynamics estimation model considering load parameter change, can effectively inhibit the divergence influence of the vehicle state parameter filter by adopting the self-adaptive unscented Kalman filtering method, corrects and predicts the vehicle inertia parameter in real time by utilizing the estimation value of the vehicle state in the double self-adaptive unscented Kalman filter, and has the advantages of high estimation precision and strong reliability.

Description

A distributed drive electric vehicle inertial parameter estimation method

technical field

The invention belongs to the field of active safety control of distributed driving electric vehicles, and particularly relates to a method for estimating inertial parameters of distributed driving electric vehicles.

Background technique

Due to the great potential of energy saving and emission reduction, electric vehicles have become a hot spot in the research and development of energy-saving and environmentally friendly vehicle technology, and the development of electric vehicles has also become the development direction of the future automobile industry. However, the increase in the number of electric vehicles has brought many traffic problems, especially the active safety of vehicles. In recent years, with the development of electronic information technology and intelligent control technology, there have been many automotive chassis electronic control systems that can effectively improve active safety in the modern automobile market. ), Traction Control System (TCS), and Electronic Stability System (ESP) based on vehicle lateral control, Electric Power Steering (EPS), and Direct Yaw Moment Control (DYC) based on vehicle yaw direction control. In order to realize the effective and reliable control of these active safety dynamic systems, it is necessary to obtain some key state parameter information, such as vehicle lateral speed, vehicle center of mass slip angle, etc., accurately and in real time.

However, the measurement of these state parameters requires the installation of expensive on-board sensors, and the reliability of sensor measurement signals has not been fully resolved. These key states of vehicle operation are difficult to measure directly using standard on-board sensors, and can only be observed or estimated. At the same time, from the perspective of practical vehicle engineering application, how to use the existing vehicle sensor measurement information to accurately estimate the vehicle state parameter information that is difficult to directly measure online is an urgent problem to be solved in vehicle engineering applications.

In the current research on vehicle state estimation, most of them focus on the state estimation during vehicle operation, while the research on vehicle inertia parameter estimation is relatively rare. In fact, the increase of the unsprung mass of the distributed drive electric vehicle leads to the redistribution of the whole vehicle mass, especially the uncertainty of the load parameters (passenger or cargo loading) leads to the vehicle inertia parameters including the vehicle mass, the yaw moment of inertia, the position of the vehicle center of mass Changes directly affect the handling characteristics, control performance and stability of the vehicle chassis system, such as lateral stability, and it is particularly important to observe the inertial parameter information of distributed drive electric vehicles in real time.

In addition, the traditional vehicle nonlinear Kalman filter state estimation assumes that the statistical characteristics of noise are known and zero-mean white noise, but there are external and environmental disturbances in the actual vehicle engineering application process, and the statistical characteristics of noise are often unknown. The statistical characteristics of noise under the condition can cause the performance of vehicle state estimation to degrade, and even cause the estimation to diverge. In the process of vehicle state parameter estimation, how to avoid estimation failure caused by the acoustic statistical characteristics of uncertain noise is also an important issue that should be considered.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for estimating inertial parameters of a distributed drive electric vehicle. Based on the established nonlinear dynamic model of the vehicle considering load changes, the dual adaptive unscented Kalman filter algorithm is used to estimate the longitudinal speed of the vehicle, the Based on the state quantities such as the center of mass slip angle, the vehicle inertial parameters such as the mass of the vehicle, the yaw moment of inertia, and the distance from the center of mass to the front axle of the vehicle are estimated in real time, which has the advantages of high accuracy and strong reliability.

In order to achieve the above object, the present invention provides the following scheme:

A method for estimating inertial parameters of a distributed drive electric vehicle adopts a dual-adaptive unscented Kalman filter algorithm, which specifically includes the following steps:

(1) Construct a three-degree-of-freedom vehicle dynamics model that considers uncertain load parameters such as passenger or cargo loading, and establish a relationship with the vehicle longitudinal speed, vehicle yaw rate, vehicle center of mass slip angle, vehicle lateral speed and acceleration, and vehicle mass. , yaw moment of inertia, the distance from the center of mass to the front axle of the vehicle, including vehicle longitudinal, lateral, yaw motion, three degrees of freedom vehicle dynamics model.

In the above step (1), the established dynamic equation of the distributed drive electric vehicle including the longitudinal, lateral and yaw motions of the vehicle is as follows:

in,

in,

in,

当载荷m_p加载后，车辆总质量为m_n＝m_e+m_p The position of the center of mass of the model will change taking into account changes in the load parameters. When the position of the mass center of mass loaded by the vehicle is relative to the original coordinate system, the coordinate vector is

When the load m _p is loaded, the total mass of the vehicle is m _n =m _e +m _p

The yaw moment of inertia at the original center of mass after loading is:

where _Izzo is the yaw moment of inertia when the vehicle is unloaded.

The yaw moment of inertia at the original center of mass after loading:

in

The new centroid position coordinates in the original coordinate system:

且

and

Yaw moment of inertia after loading

At the same time, when the load changes, the relevant geometric parameters of its centroid position also change accordingly:

In the above formula, V _x and V _y are the longitudinal and lateral velocities of the center of mass of the vehicle, respectively; _{r z} is the yaw rate of the center of mass of the vehicle; β is the _{sideslip angle} of the center of mass of the vehicle; Loaded mass, loaded mass, and total mass; F _xij and F _yij are the longitudinal and lateral forces of the ith and jth tires of the vehicle, respectively, where i=f,r; j=l,r. F _w , F _f are the air resistance of the vehicle and the rolling resistance of the ground tires; C _d is the air resistance coefficient; ρ is the air density; A _f is the front windward area of the vehicle; a _x , a _y are the longitudinal and lateral accelerations of the vehicle; μ is the known road adhesion coefficient; δ _fl , δ _fr are the steering angles of the front and left wheels, respectively; I _zz , M _z represent the vehicle yaw moment of inertia and vehicle yaw moment, respectively; l _f , l _r are the center of mass to The horizontal distance between the front and rear axles of the vehicle; b _l and br are the horizontal distance from the center of mass to the center of the left and right wheels respectively; _mp is the load mass of the vehicle; _L and B are the horizontal distance between the front and rear axles of the vehicle and the horizontal distance between the left and right wheels of the vehicle; l _f0 , l _r0 are the horizontal distance from the front and rear axles of the vehicle to the center of mass when the vehicle is not loaded; x _p , y _p are the coordinate values of the load in the original vehicle coordinate system; x _n , y _n are the center of mass coordinates when the vehicle is loaded; b _f0 , b _r0 are the horizontal distances from the left and right wheels to the center of mass when the vehicle is not loaded, respectively.

(2) Build the tire model, select the Pacejka model to build the nonlinear tire, which uses the same set of complex trigonometric function formulas to uniformly express the longitudinal force and lateral force of the tire, which has the advantages of strong adaptability and high precision.

The Pacejka model described in step (2) establishes the nonlinear tire as follows:

In the above formula, the tire model parameters D, B, C, and E are the peak factor, stiffness factor, curve shape factor, and curve curvature factor, respectively; _Sh and S _v are the horizontal drift of the curve and the vertical drift of the curve, respectively. Y is the tire lateral force and longitudinal force; X is the tire side slip angle α or the longitudinal slip rate s. The lateral and longitudinal forces of the tire are calculated as follows:

Among them, for tire lateral force

For tire longitudinal force

In the above calculation of tire lateral force and longitudinal force parameters, a ₁ , a ₂ , b ₁ , b ₂ represent the peak factor calculation coefficients, a ₃ , a ₄ , a ₅ , b ₃ , b ₄ , b ₅ represent the BCD calculation coefficients ,a ₆ ,a ₇ ,a ₈ ,b ₆ ,b ₇ ,b ₈ represent curvature factor calculation coefficients.

(3) According to the constructed three-degree-of-freedom vehicle dynamics model and tire model, the state and parameter estimation system of dual adaptive unscented Kalman filter is designed, and the local observability of vehicle inertial parameters is proved at the same time. The state equation and observation equation of the state estimation system can be expressed in the following form after discretization:

in

∑M _zi (k-1)=[F _yfl (k-1)sin(δ _fl (k-1))-F _xfl (k-1)cos(δ _fl (k-1))]b _l +[ F _xfl (k-1)sin(δ _fl (k-1))+F _yfl (k-1)cos(δ _fl (k-1))]l _f +[F _xfr (k-1)cos(δ _fr (k-1))-F _yfr (k-1)sin(δ _fr (k-1))]b _r +(F _xfr (k-1)sin(δ _fr (k-1))+F _yfr (k-1)cos(δ _fr (k-1))]l _f +(F _xrr (k-1)b _r -F _xrl (k-1)b _l )-(F _yrr (k-1)+ F _yrl (k-1))l _r

In the above state observation system, x(k)=[r _z , V _x , β, a _y , V _y ] ^T , θ(k)=[m _n , I _zz , l _f ] ^T are the vehicle nonlinearities, respectively The state vector and parameter vector of the dynamic observer system, u(k)=[δ _f , w _ij , T _ij ] ^T and z(k)=[r _z , a _x , a _y ] ^T are the vehicle nonlinearities, respectively The input vector and measurement vector of the dynamic observer system, w(k), v(k) are the process noise and measurement noise of the system, respectively, the two are independent of the system, and T _s is the sampling time.

The corresponding parameter estimation system can be further constructed:

in,

In the above parameter estimation system, r(k) and e(k) are the process noise and measurement noise of the system, respectively, and d(k)=[r _z , a _x , a _y ] ^T is the measurement vector.

The local observability is proved by studying the rank of the observability co-distribution matrix of inertial parameters. If the observability co-distribution matrix has full column rank, the inertial parameters are said to be locally observable. The output vector and the derivative of the output vector of the vehicle inertial parameters are defined as:

Its observability co-distribution matrix is:

Among them, the partial derivation of the observability co-distribution matrix is:

满秩，则车辆惯性参数θ(k)＝[m_n,I_zz,l_f]^T局部可观。When the vehicle is running,

Full rank, the vehicle inertia parameter θ(k)=[m _n , I _zz , l _f ] ^T is locally considerable.

(4) The specific operation of the dual-adaptive unscented Kalman filter observer for the inertial parameters of the distributed drive electric vehicle includes the following steps:

P_w,P_v,P_r,P_e；Initialization; the values that need to be initialized are:

P _w , P _v , P _r , P _e ;

和

The time update of the time-varying parameter, we get

and

和

Build the sigma point of the state, complete the time update of the state, and obtain X _i (k|k-1),

and

Construct the sigma points of time-varying parameters to get Θ _j (k-1|k-1);

Calculate the output estimates of the time-varying parameters from the sigma points to obtain D _j (k|k-1) and

Calculate the output estimate of the state according to the sigma points to get _zi (k|k-1) and

和L_x(k)；Calculate the Kalman gain of the state to get

and L _x (k);

和L_θ(k)；Calculate the Kalman gain of the time-varying parameters, and get

and L _θ (k);

和

Complete the measurement and update of the state and time-varying parameters, respectively, to obtain

and

An adaptive update of the covariance of noise in the state and time-varying parameters is done, respectively, resulting in _Pw (k-1), _Pv (k), _Pr (k-1), and _Pe (k).

(5) Write the S-function for the design of the dual-adaptive unscented Kalman filter observer and execute it online. First, build the Simulink-Carsim distributed drive electric vehicle system state estimation co-simulation platform in the Matlab/Simulink environment. Since the Carsim software has not been developed For the power source system of new energy vehicles, the distributed drive system of electric vehicles is built in an external form, and the observer system of electric vehicles is established in Matlab/Simulink, and then CarSim and Simulink are connected through the CarSim-S function connection interface to realize simulation communication , and finally realize the estimation of state and inertia parameters in the process of vehicle driving.

Compared with the prior art, the present invention has the following obvious outstanding substantive features and remarkable technological progress:

1. In the process of establishing the nonlinear dynamic estimation model of the vehicle, the present invention considers that the load parameters of the distributed drive electric vehicle are uncertain, such as the load of passengers or cargo, which causes the vehicle inertia parameters including vehicle mass, yaw moment of inertia, and changes in the position of the center of mass of the vehicle, A dynamic model for the estimation of inertial parameters of a distributed drive electric vehicle is constructed;

2. The present invention estimates the dynamic model tire model based on the inertial parameters of the distributed drive electric vehicle, utilizes the in-wheel motor torque perception information of the distributed drive electric vehicle using the in-wheel motor to directly drive the four wheels, and the design is based on the dual adaptive unscented Kalman The inertial parameter estimation system of the filter, while proving the local observability of the vehicle inertial parameters;

3. In the process of estimating the inertial parameters of the vehicle, the present invention adopts the dual adaptive unscented Kalman filter to estimate the covariance matrix of the process noise and the measurement noise online, avoiding the traditional Kalman filter due to the assumption that the statistical characteristics of the noise in the estimation process are The resulting filter estimation performance is reduced, and even the filter divergence deviates from the true value, which has the advantages of high estimation accuracy and strong reliability.

Description of drawings

FIG. 1 is an overall design frame diagram of a method for estimating inertial parameters of a distributed drive electric vehicle according to the present invention.

FIG. 2 is a schematic diagram of a vehicle dynamics model considering load parameters according to the present invention.

FIG. 3 is a flow chart of the dual-adaptive unscented Kalman filtering algorithm of the present invention.

Detailed ways

The preferred embodiments of the present invention are described in detail in conjunction with the accompanying drawings as follows:

Example 1

In this embodiment, referring to FIG. 1 , a method for estimating inertial parameters of a distributed drive electric vehicle includes the following steps:

S1. Establish a three-degree-of-freedom vehicle nonlinear dynamic model including vehicle longitudinal, lateral, and yaw motions, and consider the system changes of the vehicle dynamics estimation model caused by the uncertainty of the load parameters;

S2. Build a tire model, and select the Pacejka model to build a nonlinear tire;

S3. According to the constructed three-degree-of-freedom vehicle dynamics model and tire model, design an inertial parameter estimation system framework based on dual adaptive unscented Kalman filters, and prove the local observability of vehicle inertial parameters;

S4. Based on the inertial parameter estimation system in the step S3, determine the specific operation method and steps of the dual-adaptive unscented Kalman filter observer, so as to realize the vehicle status such as vehicle longitudinal speed, vehicle center of mass sideslip angle, and vehicle mass, Estimation of vehicle inertial parameters such as yaw moment of inertia, the distance from the center of mass to the front axle of the vehicle.

Embodiment 2

This embodiment is basically the same as the first embodiment, and the special features are as follows:

In this embodiment, the equation of the three-degree-of-freedom vehicle dynamics model in step S1 is:

in,

In the above formula, V _x and V _y are the longitudinal and lateral velocities of the center of mass of the vehicle, respectively; r _z is the yaw rate of the center of mass of the vehicle; β is the sideslip angle of the center of mass of the vehicle; m _n represents the total mass of the vehicle; F _xij , F _yij are the longitudinal and lateral forces of the ith and jth tires of the vehicle, respectively, where i=f, r; j=l, r; F _w , F _f are the air resistance of the vehicle and the rolling resistance of the ground tires, respectively; C _d is the air resistance coefficient ρ is the air density; A _f is the front windward area of the vehicle; a _x and a _y are the longitudinal and lateral accelerations of the vehicle, respectively; μ is the known road adhesion coefficient; δ _fl , δ _fr are the steering angles of the left and right front wheels, respectively ; I _zz , M _z represent the vehicle yaw moment of inertia and vehicle yaw moment, respectively; l _f , l _r are the horizontal distance from the center of mass to the front and rear axles of the vehicle, respectively; b _l , _br are the horizontal distance from the center of mass to the center of the left and right wheels, respectively .

载荷m_p加载后，车辆总质量为m_n＝m_e+m_p，则加载后原质心处的横摆转动惯量为：In the three-degree-of-freedom vehicle dynamics model in the step S1, the position of the center of mass of the model has changed when considering the change of the load parameters; it is assumed that the position of the center of mass of the vehicle loaded with respect to the coordinate vector of the original coordinate system is:

After the load m _p is loaded, the total mass of the vehicle is m _n =m _e +m _p , then the yaw moment of inertia at the original center of mass after loading is:

where _Izzo is the yaw moment of inertia when the vehicle is unloaded;

其中

The yaw moment of inertia at the original center of mass after loading:

in

The new centroid position coordinates in the original coordinate system:

且

and

Yaw moment of inertia after loading

At the same time, when the load is loaded, its related geometric parameters also change accordingly:

In the above formula, m _p is the load mass of the vehicle; I _zzo is the yaw moment of inertia when the vehicle is unloaded; L and B are the horizontal distance between the front and rear axles of the vehicle and the horizontal distance between the left and right wheels of the vehicle; l _f0 , l _r0 are respectively The horizontal distance from the front and rear axles of the vehicle to the center of mass when not loaded; x _p , y _p are the coordinate values of the load in the original vehicle coordinate system; x _n , y _n are the center of mass coordinates when the vehicle is loaded; b _f0 , b _r0 are the vehicle The horizontal distance between the left and right turns to the centroid when not loaded.

In the present embodiment, the Pacejka model in the step S2 uses the same set of compound trigonometric function formulas to uniformly express the longitudinal force, lateral force, etc. of the tire, and its form is:

In the above formula, the tire model parameters D, B, C, and E are the peak factor, stiffness factor, curve shape factor, and curve curvature factor, respectively; _Sh and S _v are the horizontal drift of the curve and the vertical drift of the curve; when X is Tire side slip angle α, Y is the tire lateral force; when X is the tire longitudinal slip rate s, Y is the tire longitudinal force;

The lateral and longitudinal forces of the tire are calculated as follows:

Among them, for tire lateral force

For tire longitudinal force

In the above calculation of tire lateral force and longitudinal force parameters, a ₁ , a ₂ , b ₁ , b ₂ represent the peak factor calculation coefficients, a ₃ , a ₄ , a ₅ , b ₃ , b ₄ , b ₅ represent the BCD calculation coefficients ,a ₆ ,a ₇ ,a ₈ ,b ₆ ,b ₇ ,b ₈ represent curvature factor calculation coefficients.

In this embodiment, the state estimation system of the dual-adaptive unscented Kalman filter in the step S3 is:

in

∑M _zi (k-1)=[F _yfl (k-1)sin(δ _fl (k-1))-F _xfl (k-1)cos(δ _fl (k-1))]b _l +[ F _xfl (k-1)sin(δ _fl (k-1))+F _yfl (k-1)cos(δ _fl (k-1))]l _f +[F _xfr (k-1)cos(δ _fr (k-1))-F _yfr (k-1)sin(δ _fr (k-1))]b _r +(F _xfr (k-1)sin(δ _fr (k-1))+F _yfr (k-1)cos(δ _fr (k-1))]l _f +(F _xrr (k-1)b _r -F _xrl (k-1)b _l )-(F _yrr (k-1)+ F _yrl (k-1))l _r

In the above state observation system, x(k)=[r _z , V _x , β, a _y , V _y ] ^T , θ(k)=[m _n , I _zz , l _f ] ^T are the vehicle nonlinearities, respectively The state vector and parameter vector of the dynamic observer system, u(k)=[δ _f , w _ij , T _ij ] ^T and z(k)=[r _z , a _x , a _y ] ^T are the vehicle nonlinearities, respectively Input vector and measurement vector of the dynamic observer system, w(k), v(k) are the process noise and measurement noise of the system, respectively, the two are independent of the system, and T _s is the sampling time;

The corresponding dual adaptive unscented Kalman filter parameter estimation system can be further constructed:

in,

In the above parameter estimation system, r(k) and e(k) are the process noise and measurement noise of the system, respectively, and d(k)=[r _z , a _x , a _y ] ^T is the measurement vector.

In this embodiment, in the step S3, the local observability is proved by studying the rank of the co-distribution matrix of the observability of the inertia parameter. If the co-distribution matrix of the observability has a full column rank, the inertia parameter is said to be locally observable; The local observability process of the vehicle inertial parameters is as follows:

The output vector of the vehicle inertia parameter and the derivative of the output vector are defined as:

其中部分求导结果为：Then its observability co-distribution matrix is:

Some of the derivation results are:

满秩，则车辆惯性参数θ(k)＝[m_n,I_zz,l_f]^T局部可观。From the above derivation, it can be obtained that when the vehicle is running,

Full rank, the vehicle inertia parameter θ(k)=[m _n , I _zz , l _f ] ^T is locally considerable.

In this embodiment, the specific operation of the dual-adaptive unscented Kalman filter observer in the step S4 includes the following steps:

P_w,P_v,P_r,P_e；(1) Initialization, the values that need to be initialized are:

P _w , P _v , P _r , P _e ;

和

(2) Time update of time-varying parameters, we get

and

和

(3) Construct the sigma point of the state, complete the time update of the state, and obtain X _i (k|k-1),

and

(4) Construct the sigma point of time-varying parameters to obtain Θ _j (k-1|k-1);

(5) Calculate the output estimation of the time-varying parameters according to the sigma point, and obtain D _j (k|k-1) and

(6) Calculate the output estimation of the state according to the sigma point, and obtain _zi (k|k-1) and

和L_x(k)；(7) Calculate the Kalman gain of the state, and get

and L _x (k);

和L_θ(k)；(8) Calculate the Kalman gain of the time-varying parameters, and get

and L _θ (k);

和

(9) Complete the measurement and update of the state and time-varying parameters, respectively, and obtain

and

(10) The adaptive update of the covariance of noise in the state and time-varying parameters is completed respectively, and _Pw (k-1), _Pv (k), _Pr (k-1) and _Pe (k) are obtained.

Embodiment 3

This embodiment is basically the same as the previous embodiment, and the special features are as follows:

In this embodiment, a method for estimating inertial parameters of a distributed driving electric vehicle, as shown in FIG. 1 , includes the vehicle front wheel rotation angle δ _f , the wheel tire angular velocity w _ij , the wheel torque T _ij , the vehicle lateral According to the information such as yaw rate r _z , longitudinal acceleration a _x and lateral acceleration a _y , combined with the established vehicle dynamics model of the relevant degrees of freedom, a double adaptive unscented Kalman filter algorithm is designed to calculate the front wheel rotation angle δ _f of the vehicle, tire The angular velocity w _ij and the tire torque T _ij are the system input quantities, the vehicle yaw angular velocity _{r z} , the longitudinal acceleration a _x and the lateral acceleration a _y are the system observation _quantities . , vehicle center of mass slip angle β, vehicle lateral speed V _y and acceleration a _y and the vehicle mass m _n , yaw moment of inertia I _zz , the center of mass to the vehicle front axle distance _lf estimates.

The specific steps of the estimation method are as follows:

1. Build a vehicle dynamics model with the relevant degrees of freedom:

The dynamic model of the distributed drive electric vehicle considering the load parameters is shown in Figure 2. The origin of the vehicle coordinate system is defined at the center of mass (CG) of the vehicle. The dynamic equation of driving an electric vehicle is as follows:

in,

in,

in,

载荷m_p加载后，车辆总质量为m_n＝m_e+m_p，则加载后原质心处的横摆转动惯量为：The position of the center of mass of the model has changed, taking into account changes in the load parameters. Assume that the position of the mass center of mass loaded by the vehicle is relative to the original coordinate system coordinate vector as

After the load m _p is loaded, the total mass of the vehicle is m _n =m _e +m _p , then the yaw moment of inertia at the original center of mass after loading is:

where _Izzo is the yaw moment of inertia of the vehicle in empty time.

Using the principle of parallel axes, the yaw moment of inertia at the original center of mass after loading can be obtained:

further,

Further, assuming that the height of the center of mass of the whole vehicle does not change, use the lever principle to calculate the new position coordinates of the center of mass in the original coordinate system:

and

In summary:

At the same time, after loading, its related geometric parameters also change accordingly:

2. Build a tire model, select the Pacejka model to build a nonlinear tire, which uses the same set of complex trigonometric function formulas to uniformly express the longitudinal force and lateral force of the tire.

The Pacejka model builds nonlinear tires as follows:

The lateral and longitudinal forces of the tire are calculated as follows:

Among them, for tire lateral force

For tire longitudinal force

In the above calculation of tire lateral force and longitudinal force parameters, a ₁ , a ₂ , b ₁ , b ₂ represent the peak factor calculation coefficients, a ₃ , a ₄ , a ₅ , b ₃ , b ₄ , b ₅ represent the BCD calculation coefficients ,a ₆ ,a ₇ ,a ₈ ,b ₆ ,b ₇ ,b ₈ represent curvature factor calculation coefficients.

The wheel vertical load F _zij is calculated as follows:

The wheel tire slip angle α _ij is calculated as follows:

The wheel tire longitudinal slip s _ij is calculated as follows:

3. Establish a state and parameter estimation system for dual adaptive unscented Kalman filters and prove the local observability of vehicle inertial parameters:

1) According to the above three-degree-of-freedom vehicle dynamics model, the state estimation system can be constructed as:

Specifically:

in,

∑M _zi (k-1)=[F _yfl (k-1)sin(δ _fl (k-1))-F _xfl (k-1)cos(δ _fl (k-1))]b _l +[ F _xfl (k-1)sin(δ _fl (k-1))+F _yfl (k-1)cos(δ _fl (k-1))]l _f +[F _xfr (k-1)cos(δ _fr (k-1))-F _yfr (k-1)sin(δ _fr (k-1))]b _r +(F _xfr (k-1)sin(δ _fr (k-1))+F _yfr (k-1)cos(δ _fr (k-1))]l _f +(F _xrr (k-1)b _r -F _xrl (k-1)b _l )-(F _yrr (k-1)+ F _yrl (k-1))l _r

The above formula is the sampling time in T _s .

The corresponding parameter estimation system can be further constructed:

Specifically:

2) The output vector of the vehicle inertia parameter and the derivative of the output vector are defined as:

Then its observability co-distribution matrix is:

Among them, part of the derivation results are:

满秩，则车辆惯性参数θ(k)＝[m_n,I_zz,l_f]^T局部可观。From the above derivation, it can be obtained that when the vehicle is running,

Full rank, the vehicle inertia parameter θ(k)=[m _n , I _zz , l _f ] ^T is locally considerable.

4. Design a dual-adaptive unscented Kalman filter observer as shown in Figure 3. The specific algorithm steps are as follows:

P_w,P_v,P_r,P_e 1) Initialization; the values that need to be initialized here are:

P _w ,P _v ,P _r ,P _e

2) Vehicle parameter time update

Compute a one-step prediction for a vector of vehicle parameters:

Compute the one-step vehicle parameter prediction error covariance matrix:

3) The sigma point of the build state and the vehicle state time update

According to the state mean and covariance of the system state, the initialized 2L+1 sigma point set is established as:

in,

是矩阵

的第i列。In the above formula, L is the dimension of the state vector, λ _s is the scale parameter,

is the matrix

the i-th column of .

The corresponding rights are:

分别为对应均值和方差的权重，α_s为尺度标量,用来控制每个点到均值的距离且满足0.0001≤α_s≤1,β_s涉及高斯情况下状态的先验分布信息,s为标准参数,通常为0或3-L。In the above formula,

are the weights corresponding to the mean and variance, respectively, α _s is the scale scalar, used to control the distance from each point to the mean and satisfies 0.0001≤α _s ≤1, β _s involves the prior distribution information of the state in the Gaussian case, s is the standard parameter, usually 0 or 3-L.

Calculate the sigma point set and conduction sigma point set of the vehicle state:

X _i (k|k-1)=f(X _i (k-1|k-1),u(k-1))

Compute the one-step prediction of the vehicle state vector:

Compute the one-step vehicle state prediction error covariance matrix:

4) Construct sigma points of time-varying parameters

The 2l+1 sigma point set for constructing parameters according to the vehicle parameter information is:

in,

是矩阵

的第j列。In the above formula, l is the dimension of the estimated parameter vector, λ _θ is the scale parameter,

is the matrix

the jth column of .

The corresponding rights are:

分别为对应均值和方差的权重，α_θ为尺度标量,用来控制每个点到均值的距离且满足0.0001≤α_θ≤1,β_θ涉及高斯情况下状态的先验分布信息,s_θ为标准参数，通常为0或3-l。In the above formula,

are the weights corresponding to the mean and variance, respectively, α _θ is a scale scalar, which is used to control the distance from each point to the mean and satisfies 0.0001≤α _θ ≤1, β _θ relates to the prior distribution information of the state in the Gaussian case, s _θ is Standard parameter, usually 0 or 3-l.

5) Vehicle parameter measurement update

Further calculate the vehicle parameter measurement sigma point set and the new conduction sigma point set:

6) Vehicle state measurement update

Further calculate the vehicle state measurement volume point set and conduction volume point set:

7) Calculate the Kalman gain of the state

Compute the innovation covariance matrix:

Compute the cross-covariance matrix:

Compute the Kalman filter nonlinear state observer gain:

8) Calculate the Kalman gain of time-varying parameters

Compute the parameter innovation covariance matrix:

Compute the parametric cross-covariance matrix:

Compute the Kalman filter nonlinear parameter observer gain:

9) Complete the measurement update of the state and time-varying parameters respectively

Update the state vector at the current moment to obtain the optimal estimate of the nonlinear vehicle state at the current moment:

Also update the error covariance matrix:

Then update the parameter vector at the current moment to obtain the optimal estimated value of the vehicle parameters at the current moment:

Also update the parameter error covariance matrix:

10) Complete the adaptive update of the covariance of the noise in the state and time-varying parameters

Adaptive update of state noise covariance:

in,

In the above formula, n is the number of sampling times.

Adaptive update of parameter noise covariance:

5. In Matlab/Simulink, first build the Simulink-Carsim distributed drive electric vehicle estimation inertial parameter simulation platform, in which the distributed drive system of the electric vehicle is built in an external form, and then set the simulation conditions in the CarSim software, and then pass the CarSim-S The connection interface of the function is used to perform co-simulation communication with Simulink, and finally realize a distributed drive electric vehicle inertial parameter estimation.

Claims (7)

1. a distributed drive electric vehicle inertial parameter estimation method, is characterized in that, comprises the following steps: S1. Establish a three-degree-of-freedom vehicle nonlinear dynamic model including vehicle longitudinal, lateral, and yaw motions, and consider the system changes of the vehicle dynamics estimation model caused by the uncertainty of the load parameters; S2. Build a tire model, and select the Pacejka model to build a nonlinear tire; S3. According to the constructed three-degree-of-freedom vehicle dynamics model and tire model, design an inertial parameter estimation system framework based on dual adaptive unscented Kalman filters, and prove the local observability of vehicle inertial parameters; S4. Based on the inertial parameter estimation system in the step S3, determine the specific operation method and steps of the dual-adaptive unscented Kalman filter observer, so as to realize the vehicle status such as vehicle longitudinal speed, vehicle center of mass sideslip angle, and vehicle mass, Estimation of vehicle inertial parameters such as yaw moment of inertia, the distance from the center of mass to the front axle of the vehicle. 2. The method for estimating inertial parameters of a distributed drive electric vehicle according to claim 1, wherein the equation of the three-degree-of-freedom vehicle dynamics model in the step S1 is:

in,

In the above formula, V _x and V _y are the longitudinal and lateral velocities of the center of mass of the vehicle, respectively; r _z is the yaw rate of the center of mass of the vehicle; β is the sideslip angle of the center of mass of the vehicle; m _n represents the total mass of the vehicle; F _xij , F _yij are the longitudinal and lateral forces of the ith and jth tires of the vehicle, respectively, where i=f, r; j=l, r; F _w , F _f are the air resistance of the vehicle and the rolling resistance of the ground tires, respectively; C _d is the air resistance coefficient ρ is the air density; A _f is the front windward area of the vehicle; a _x and a _y are the longitudinal and lateral accelerations of the vehicle, respectively; μ is the known road adhesion coefficient; δ _fl , δ _fr are the steering angles of the left and right front wheels, respectively ; I _zz , M _z represent the vehicle yaw moment of inertia and vehicle yaw moment, respectively; l _f , l _r are the horizontal distance from the center of mass to the front and rear axles of the vehicle, respectively; b _l , _br are the horizontal distance from the center of mass to the center of the left and right wheels, respectively . 3. The method for estimating inertial parameters of a distributed drive electric vehicle according to claim 1, wherein the three-degree-of-freedom vehicle dynamics model in the step S1 considers the change of the load parameter, and the position of the center of mass of the model has occurred. change; it is assumed that the position of the center of mass of the vehicle loaded relative to the original coordinate system coordinate vector is

After the load m _p is loaded, the total mass of the vehicle is m _n =m _e +m _p , then the yaw moment of inertia at the original center of mass after loading is:

where _Izzo is the yaw moment of inertia when the vehicle is unloaded; The yaw moment of inertia at the original center of mass after loading:

in

The new centroid position coordinates in the original coordinate system:

and

Yaw moment of inertia after loading

At the same time, when the load is loaded, its related geometric parameters also change accordingly:

In the above formula, m _p is the load mass of the vehicle; I _zzo is the yaw moment of inertia when the vehicle is unloaded; L and B are the horizontal distance between the front and rear axles of the vehicle and the horizontal distance between the left and right wheels of the vehicle; l _f0 , l _r0 are respectively The horizontal distance from the front and rear axles of the vehicle to the center of mass when not loaded; x _p , y _p are the coordinate values of the load in the original vehicle coordinate system; x _n , y _n are the center of mass coordinates when the vehicle is loaded; b _f0 , b _r0 are the vehicle The horizontal distance between the left and right turns to the centroid when not loaded. 4. The method for estimating inertial parameters of a distributed drive electric vehicle according to claim 1, wherein the step The Pacejka model in S2 uses the same set of complex trigonometric function formulas to uniformly express the longitudinal force and lateral force of the tire, and its form is:

In the above formula, the tire model parameters D, B, C, and E are the peak factor, stiffness factor, curve shape factor, and curve curvature factor, respectively; _Sh and S _v are the horizontal drift of the curve and the vertical drift of the curve; when X is Tire side slip angle α, Y is the tire lateral force; when X is the tire longitudinal slip rate s, Y is the tire longitudinal force; The lateral and longitudinal forces of the tire are calculated as follows:

Among them, for tire lateral force F _yij ,

For tire longitudinal force F _xij ,

In the above calculation of tire lateral force and longitudinal force parameters, a ₁ , a ₂ , b ₁ , b ₂ represent the peak factor calculation coefficients, a ₃ , a ₄ , a ₅ , b ₃ , b ₄ , b ₅ represent the BCD calculation coefficients ,a ₆ ,a ₇ ,a ₈ ,b ₆ ,b ₇ ,b ₈ represent curvature factor calculation coefficients. 5. The method for estimating inertial parameters of a distributed drive electric vehicle according to claim 1, wherein the state estimation system of the dual adaptive unscented Kalman filter in the step S3 is:

in

∑M _zi (k-1)=[F _yfl (k-1)sin(δ _fl (k-1))-F _xfl (k-1)cos(δ _fl (k-1))]b _l +[ F _xfl (k-1)sin(δ _fl (k-1))+F _yfl (k-1)cos(δ _fl (k-1))]l _f +[F _xfr (k-1)cos(δ _fr (k-1))-F _yfr (k-1)sin(δ _fr (k-1))]b _r +(F _xfr (k-1)sin(δ _fr (k-1))+F _yfr (k-1)cos(δ _fr (k-1))]l _f +(F _xrr (k-1)b _r -F _xrl (k-1)b _l )-(F _yrr (k-1)+ F _yrl (k-1))l _r In the above state observation system, x(k)=[r _z , V _x , β, a _y , V _y ] ^T , θ(k)=[m _n , I _zz , l _f ] ^T are the vehicle nonlinearities, respectively The state vector and parameter vector of the dynamic observer system, u(k)=[δ _f , w _ij , T _ij ] ^T and z(k)=[r _z , a _x , a _y ] ^T are the vehicle nonlinearities, respectively Input vector and measurement vector of the dynamic observer system, w(k), v(k) are the process noise and measurement noise of the system, respectively, the two are independent of the system, and T _s is the sampling time; The corresponding dual adaptive unscented Kalman filter parameter estimation system can be further constructed:

in,

In the above parameter estimation system, r(k) and e(k) are the process noise and measurement noise of the system, respectively, and d(k)=[r _z , a _x , a _y ] ^T is the measurement vector. 6. The method for estimating inertial parameters of a distributed drive electric vehicle according to claim 1, wherein in the step S3, its local observability is proved by studying the rank of the co-distribution matrix of the observability of the inertia parameters. If the distribution matrix has full column rank, the inertial parameters are said to be locally observable; the process of proving the local observability of vehicle inertial parameters is as follows: The output vector of the vehicle inertia parameter and the derivative of the output vector are defined as:

Then its observability co-distribution matrix is:

Some of the derivation results are:

From the above derivation, it can be obtained that when the vehicle is running,

Full rank, the vehicle inertia parameter θ(k)=[m _n , I _zz , l _f ] ^T is locally considerable. 7. The method for estimating inertial parameters of a distributed drive electric vehicle according to claim 1, wherein the specific operation of the dual-adaptive unscented Kalman filter observer in the step S4 comprises the following steps: (1) Initialization, the values that need to be initialized are:

P _w , P _v , P _r , P _e ; (2) Time update of time-varying parameters, we get

and

(3) Construct the sigma point of the state, complete the time update of the state, and obtain X _i (k|k-1),

and

(4) Construct the sigma point of time-varying parameters to obtain Θ _j (k-1|k-1); (5) Calculate the output estimation of the time-varying parameters according to the sigma point, and obtain D _j (k|k-1) and

(6) Calculate the output estimation of the state according to the sigma point, and obtain _zi (k|k-1) and

(7) Calculate the Kalman gain of the state, and get

and L _x (k); (8) Calculate the Kalman gain of the time-varying parameters, and get

and L _θ (k); (9) Complete the measurement and update of the state and time-varying parameters, respectively, and obtain

and

(10) The adaptive update of the covariance of noise in the state and time-varying parameters is completed respectively, and _Pw (k-1), _Pv (k), _Pr (k-1) and _Pe (k) are obtained.

Images