A multi-axle distributed drive vehicle steering assist trajectory tracking method based on ampc

Abstract

The application provides a kind of AMPC-based multi-axle distributed drive vehicle steering auxiliary trajectory tracking method, and a prediction model is established based on linear two-degree-of-freedom single-track model, which can achieve high prediction accuracy when tire force is in linear region;The driver model uses a linear quadratic regulator, which can simulate the conditions of skilled drivers tracking the expected trajectory as much as possible;In the method, the multi-cost objective function is constructed by considering the intervention auxiliary quantity, the reference trajectory tracking error, the front wheel angle and its change rate and other factors, and the driving weight decay coefficient and the driving weight coefficient are introduced to measure and represent the confidence degree of the driver's intention in the prediction window, and the optimal front wheel angle output value obtained by finally solving can make the vehicle quickly and smoothly reach the target trajectory, and reduce the jitter and ensure the smoothness in the process.

Description

Technical Field

[0001] This invention belongs to the field of intelligent electric vehicle assisted driving technology, specifically involving a multi-axis distributed drive vehicle steering assistance trajectory tracking method based on AMPC (Adaptive Model Predictive Control). Background Technology

[0002] Currently, driver assistance technologies applicable to intelligent electric vehicles leverage the environmental perception capabilities of various types of sensors and utilize high-performance onboard computing platforms to process and integrate this perception information, thereby assisting drivers in decision-making and improving vehicle safety to a certain extent. However, most existing technologies in this field are primarily designed for passenger vehicles and are not well-suited for certain specialized commercial vehicles, especially multi-axle commercial transport vehicles. Due to their large weight, size, and high cab, these vehicles generally offer poor driver visibility, and their inferior vehicle agility and braking performance make driving significantly more difficult than in passenger cars. Drivers also struggle to control the vehicle's surroundings and posture. Fatigue and distraction during long-distance transport missions pose serious safety hazards. Driver assistance systems for such vehicles must also consider operational tasks, but existing task-oriented driver assistance systems often require frequent human-machine interaction, leading to low driver acceptance and trust. Furthermore, in human-machine collaborative driver assistance trajectory tracking, the system still lacks effective solutions for incorporating driver intent and permissions. Therefore, there is an urgent need in this field to provide more comprehensive driver assistance and steering trajectory tracking strategies tailored to the characteristics of multi-axle distributed drive vehicles. Summary of the Invention

[0003] In view of the above-mentioned technical problems existing in this field, the present invention provides a multi-axis distributed drive vehicle steering assistance trajectory tracking method based on AMPC, specifically including the following steps:

[0004] S1. In the upper-level control, for an n-axis distributed drive vehicle, the vehicle's horizontal and vertical coordinates (xy) and longitudinal speed (v) are used. x Heading angle Lateral velocity v y and yaw rate ω r As input variables, a linear two-degree-of-freedom single-track vehicle model is established to predict the future dynamic state of the vehicle, and outputs the predicted optimal steering wheel angle control value, the lateral position error e between the predicted trajectory and the reference trajectory. d and heading angle error The reference trajectory is external input information for the proposed method and is a known quantity required for auxiliary tracking control.

[0005] S2. In the upper-level control, the vehicle's horizontal and vertical coordinates (xy) and longitudinal velocity (v) are used as the reference points. x Heading angle Lateral velocity v y and yaw rate ω r As input variables, a linear quadratic driver adjustment model is established. The cost function is constructed by minimizing the weighted sum of cumulative lateral error, lateral error and steering wheel angle to calculate the driver's intended front wheel angle.

[0006] S3. In the lower-level control, the difference between the optimal steering wheel angle control amount predicted by the upper-level step S1 and the front wheel angle determined by the driver's intention front wheel angle obtained in step S2 is defined as the intervention assistance amount, and the lateral position error e between the predicted trajectory and the reference trajectory is used as the intervention assistance amount. d and heading angle error Minimizing all of these is taken as the optimization objective, while minimizing the steering wheel angle and its rate of change is also taken as the optimization objective. A multi-objective cost function is constructed by combining the above optimization objectives.

[0007] S4. Solve for the optimal front wheel steering angle output value at the next moment based on the multi-objective cost function.

[0008] Furthermore, the specific construction process of the linear two-degree-of-freedom monorail vehicle model described in step S1 is as follows:

[0009] A dynamic analysis is performed on a five-axle distributed drive vehicle, considering the control variable u(k) = [δ]. r (k)],δ r Let v(k) be the front wheel steering angle, and v(k) be the disturbance quantity, where ρ is the road curvature. Let the vehicle dynamic state variables be: Among them, e d and These are the lateral position error and the heading angle error, respectively; the following continuous state-space equations are established:

[0010]

[0011] Z = C c x+D u，c u

[0012] In the formula, C i Let L be the lateral stiffness along the i-th axis. i Let I be the distance from the i-th axis to the vehicle's center of mass, I be the vehicle's moment of inertia along the yaw direction, and Cc be the output matrix of the continuous state equation.

[0013] In the formula,

[0014]

[0015]

[0016] D u，c =O

[0017] Discretization yields the discrete state-space equations:

[0018] x(k+1)=A d x(k)+B u，d u(k)+B v，d v(k)

[0019] Z(k)=C d x(k)

[0020] In the formula, k represents time, and A d B is the state matrix of the discrete state-space equations. u，d B is the control matrix of the discrete state-space equations. v，d Cd is the measurable perturbation matrix of the discrete state-space equations, and the subscript d indicates the discrete processing of the corresponding parameters. Cd is the output matrix of the discrete state equations of the same form as Cc.

[0021] The iterative form of the dynamic state of the system at future p time steps is:

[0022] X p =[x(k),x(k+1),…,x(k+p)] T

[0023] U p =[u(k),u(k+1),…,u(k+p)] T

[0024] V p =[v(k), v(k+1),..., v(k+p)] T

[0025] Among them, U p The optimal sequence for solving adaptive model predictive control is given by p, where p is the prediction time window and the superscript T denotes the transpose of the vector.

[0026] Within the prediction time window, the model prediction output is represented as:

[0027] z(k+p|k)=C d x(k+p|k)

[0028] Z p =[y(k),y(k+1),...,y(k+p)] T ;

[0029] In the formula, Z pLet y be the sequence of model prediction outputs at time k within the prediction time window p, and y be a single model prediction output within the prediction time window.

[0030] The optimal steering wheel angle control value, the lateral position error e between the predicted trajectory and the reference trajectory d and heading angle error Defined as:

[0031]

[0032] In the formula, P ego and P i These represent the vehicle's current position and discrete points on the reference trajectory, respectively, with β representing the centroid sideslip angle. The angle between the tangent at the road mapping point where the vehicle is located and the angle in the global coordinate system is given by the superscript ·, which indicates the derivative with respect to the corresponding parameter.

[0033] Furthermore, the specific process of establishing the linear quadratic driver adjustment model in step S2 includes:

[0034] Assuming the road curvature is small in actual roads:

[0035]

[0036] In the formula, k r For measurable interference terms, u x The vehicle's longitudinal speed and the front wheel steering angle δ f To control the quantity;

[0037] Establish the following state-space equations:

[0038]

[0039] In the formula,

[0040]

[0041]

[0042] The meaning of d and the measurable interference term k r same;

[0043] Construct the following weighted sum cost function for cumulative lateral error, lateral error, and steering wheel angle.

[0044]

[0045] In the formula, Q and R are the weighted matrices of the state variables and control variables, respectively, both of which are positive semi-definite matrices, specifically expressed as:

[0046]

[0047] Each element in the matrix represents the weight of a specific parameter;

[0048] In the process of finding the driver's intended front wheel steering angle that minimizes the cost function, a curvature feedforward element u is introduced. f The control quantity can then be expressed as:

[0049] u(k)=-Kx(k)+u f

[0050] Using the algebraic Riccati equation Another e d = 0 to solve for the optimal control quantity u(k) to obtain the corresponding driver's intended front wheel steering angle; where K is the state feedback gain, specifically...

[0051] Furthermore, in step S3, the state variable predicted in step S1 is defined as follows:

[0052]

[0053] The lateral position error e between the predicted trajectory and the reference trajectory respectively d and heading angle error Define the road tracking error cost function:

[0054] J z，k (k)=[z(k)-z ref (k)] 2

[0055] In the formula, the subscript ref represents the reference trajectory;

[0056] Define the cost function for intervention auxiliary quantities:

[0057] J Δu，k (k)=[u Out (k)-u d (k)] 2

[0058] In the formula, u out (k) represents the control quantity predicted by the model in step S1, u d (k) represents the driver's intention control quantity obtained in step S2;

[0059] And define the cost function J for steering wheel angle and its rate of change. u，k (k) and J du，k (k);

[0060] The following multi-objective cost function is constructed by combining the various cost functions:

[0061] J k =Jzk +J uk +J duk +J Δuk +J εk

[0062] Among them, J εk =ρ ε ·ε k 2 ε is the cost function for slack variables, used to penalize violations of soft constraints in the optimization problem by state variables, control variables, etc. k ρ is a slack variable. ε Let be the weighting coefficient, and satisfy ρ ε >0.

[0063] Road tracking error cost function J zk The specific form can be expressed as:

[0064]

[0065] Among them, s z The normalized coefficient diagonal matrix is ​​specifically represented as follows:

[0066]

[0067] Each element in the matrix represents the normalization coefficient of the corresponding element in the state vector;

[0068] Z(k+h|k) and Z ref (k+h|k) represent the predicted state and the reference state at time k+h, respectively. Q(k+h|k) is the positive semidefinite diagonal cost weight matrix used to penalize tracking errors, which will be adjusted and updated in real time during the rolling optimization process.

[0069] The specific forms of the intervention assistance cost function and the cost function defining the steering wheel angle and its rate of change are as follows:

[0070]

[0071]

[0072]

[0073] Where R(K), E(k), and F(k) are the weight matrices at time t = k, and u(k + h|k) represents the predicted optimal control quantity at time k. d (k+h|k) represents the driver's input intention control quantity at time k;

[0074] The above normalization coefficients are determined by the maximum and minimum values ​​of the corresponding parameters, respectively:

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082] A driving right decay coefficient Γ and a driving weight coefficient Θ are introduced to measure the confidence level of the driver's intention input and control the driving weight between the human and machine within the prediction time window, respectively. A cost function J is constructed as follows: k Solve for the coefficients:

[0083] J k = (1-Γ(t))·Q Track (k)·J Track，k +ΘΓ(t)J Δuk +n ε ·max{Γ(t), (1-Γ(t))}·J εk

[0084] coefficient

[0085] Where, n Γ，min This represents the short-term prediction portion within the prediction time window t = 1, 2, ..., p, i.e., the first few steps of the prediction step size, n. Γ，max This represents the long-term prediction portion within the prediction time window, i.e., the last few steps of the prediction step size. ΔT is the prediction sampling time interval, and J... track，，k The partial cost function reflecting the vehicle state is denoted as: J track，k =[J zk J uk J duk ] T ;

[0086] Q(k), R(k), E(k), and F(k) are respectively represented as:

[0087] Q(k)=Q z，Track (k)·(1-Γ(t))

[0088] R(k)=Q u，Track (k)·(1-Γ(t))

[0089] E(k)=Qdu，Track (k)·(1-Γ(t))

[0090] F(k)=ΘΓ(t)

[0091] Among them, Q z，Track Q u，Track Q du，Track The road tracking error cost function J is respectively zk Weighting coefficients, intervention auxiliary quantity cost function J uk Weighting coefficients, steering wheel angle cost function J duk Weighting coefficients;

[0092] The optimal front wheel steering angle output value is finally obtained by solving the problem.

[0093] The multi-axis distributed drive vehicle steering assistance trajectory tracking method based on AMPC provided by the present invention establishes a prediction model based on a linear two-degree-of-freedom single-track model, which enables the vehicle to achieve high prediction accuracy when the tire force is in the linear region. The driver model adopts a linear quadratic regulator, which can simulate the conditions of a skilled driver tracking the expected trajectory as much as possible. The method considers multiple factors such as intervention assistance amount, reference trajectory tracking error, front wheel angle and its rate of change to construct a multi-cost objective function, and introduces a driving power attenuation coefficient and a driving weight coefficient to measure and characterize the confidence level of the driver's intention within the prediction window. The optimal front wheel angle output value obtained by the final solution enables the vehicle to quickly and smoothly reach the target trajectory, while reducing vibration and ensuring smoothness during the process. Attached Figure Description

[0094] Figure 1 This is a schematic diagram illustrating the principle framework of the method provided by the present invention.

[0095] Figure 2 This is a block diagram illustrating the planar structure and dynamics of a five-axle distributed drive vehicle.

[0096] Figure 3 This is a schematic diagram illustrating the principle of the method provided by the present invention for controlling the steering angle of driving intention and the reference trajectory error.

[0097] Figure 4 The dynamic characteristic diagram of the driving rights attenuation coefficient introduced by the method provided in this invention. Detailed Implementation

[0098] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0099] The present invention provides a multi-axis distributed drive vehicle steering assistance trajectory tracking method based on AMPC, such as... Figure 1 As shown, the specific steps include:

[0100] S1. In the upper-level control, for an n-axis distributed drive vehicle, the vehicle's horizontal and vertical coordinates (xy) and longitudinal speed (v) are used. x Heading angle Lateral velocity v y and yaw rate ω r As input variables, a linear two-degree-of-freedom single-track vehicle model is established to predict the future dynamic state of the vehicle, and outputs the predicted optimal steering wheel angle control value, the lateral position error e between the predicted trajectory and the reference trajectory. d and heading angle error The reference trajectory is external input information for the proposed method and is a known quantity required for auxiliary tracking control.

[0101] S2. In the upper-level control, the vehicle's horizontal and vertical coordinates (xy) and longitudinal velocity (v) are used as the reference points. x Heading angle Lateral velocity v y and yaw rate ω r As input variables, a linear quadratic driver adjustment model is established. The cost function is constructed by minimizing the weighted sum of cumulative lateral error, lateral error and steering wheel angle to calculate the driver's intended front wheel angle.

[0102] S3. In the lower-level control, the difference between the optimal steering wheel angle control amount predicted by the upper-level step S1 and the front wheel angle determined by the driver's intention front wheel angle obtained in step S2 is defined as the intervention assistance amount, and the lateral position error e between the predicted trajectory and the reference trajectory is used as the intervention assistance amount. d and heading angle error Minimizing all of these is taken as the optimization objective, while minimizing the steering wheel angle and its rate of change is also taken as the optimization objective. A multi-objective cost function is constructed by combining the above optimization objectives.

[0103] S4. Solve for the optimal front wheel steering angle output value at the next moment based on the multi-objective cost function.

[0104] In a preferred embodiment of the present invention, the specific construction process of the linear two-degree-of-freedom monorail vehicle model in step S1 is as follows:

[0105] For five-axle distributed drive vehicles, such as Figure 2 The dynamic analysis shown considers the control variable u(k) = [δ] r (k)],δ rLet v(k) be the front wheel steering angle, and v(k) be the disturbance quantity, where ρ is the road curvature. Let the vehicle dynamic state variables be: Among them, e d and These are the lateral position error and the heading angle error, respectively; the following continuous state-space equations are established:

[0106]

[0107] z = C c x+D u，c u

[0108] In the formula,

[0109]

[0110]

[0111] D u，c =O

[0112] In the formula, C i Let L be the lateral stiffness along the i-th axis. i Let I be the distance from the i-th axis to the vehicle's center of mass, I be the vehicle's moment of inertia along the yaw direction, and Cc be the output matrix of the continuous state equation.

[0113] Discretization yields the discrete state-space equations:

[0114] x(k+1)=A d x(k)+B u，d u(k)+B v，d v(k)

[0115] z(k)=C d x(k)

[0116] In the formula, k represents time, and A d B is the state matrix of the discrete state-space equations. u，d B is the control matrix of the discrete state-space equations. v，d Cd is the measurable perturbation matrix of the discrete state-space equations, and the subscript d indicates the discrete processing of the corresponding parameters. Cd is the output matrix of the discrete state equations of the same form as Cc.

[0117] The iterative form of the dynamic state of the system at future p time steps is:

[0118] X p =[x(k),x(k+1),...,x(k+p)] T

[0119] U p=[u(k),u(k+1),...,u(k+p)] T

[0120] V p =[v(k), v(k+1),..., v(k+p)] T

[0121] Among them, U p The optimal sequence for solving adaptive model predictive control is given by p, where p is the prediction time window and the superscript T denotes the transpose of the vector.

[0122] Within the prediction time window, the model prediction output is represented as:

[0123] z(k+p|k)=C d x(k+p|k)

[0124] Z p =[y(k),y(k+1),...,y(k+p)] T ;

[0125] In the formula, Z p Let y be the sequence of model prediction outputs at time k within the prediction time window p, and y be a single model prediction output within the prediction time window.

[0126] The optimal steering wheel angle control value, the lateral position error e between the predicted trajectory and the reference trajectory d and heading angle error Defined as:

[0127]

[0128] In the formula, P ego and P i These represent the vehicle's current position and discrete points on the reference trajectory, respectively, with β representing the centroid sideslip angle. This is the angle between the tangent line at the road mapping point where the vehicle is located and the global coordinate system. The superscript · indicates the derivative with respect to the corresponding parameter.

[0129] In a preferred embodiment of the present invention, the specific process of establishing the linear quadratic adjustment driver model in step S2 includes:

[0130] Assuming the road curvature is small in actual roads:

[0131]

[0132] In the formula, k r For measurable interference terms, u x The vehicle's longitudinal speed and the front wheel steering angle δ f To control the quantity;

[0133] Establish the following state-space equations:

[0134]

[0135] In the formula,

[0136]

[0137]

[0138] The meaning of d and the measurable interference term k r same;

[0139] Construct the following weighted sum cost function for cumulative lateral error, lateral error, and steering wheel angle.

[0140]

[0141] In the formula, Q and R are the weighted matrices of the state variables and control variables, respectively, both of which are positive semi-definite matrices, specifically expressed as:

[0142]

[0143] Each element in the matrix represents the weight of a specific parameter;

[0144] In the process of finding the driver's intended front wheel steering angle that minimizes the cost function, a curvature feedforward element u is introduced. f The control quantity can then be expressed as:

[0145] u(k)=-Kx(k)+u f

[0146] Using the algebraic Riccati equation Another e d = 0 to solve for the optimal control quantity u(k) to obtain the corresponding driver's intended front wheel steering angle; where K is the state feedback gain, specifically...

[0147] In a preferred embodiment of the present invention, such as Figure 3 As shown, in step S3, the state variable predicted in step S1 is defined as follows:

[0148]

[0149] The lateral position error e between the predicted trajectory and the reference trajectory respectively d and heading angle error Define the road tracking error cost function:

[0150] J z，k (k)=[z(k)-z ref(k)] 2

[0151] In the formula, the subscript ref represents the reference trajectory;

[0152] Define the cost function for intervention auxiliary quantities:

[0153] J Δu，k (k)=[u out (k)-u d (k)] 2

[0154] In the formula, u out (k) represents the control quantity predicted by the model in step S1, u d (k) represents the driver's intention control quantity obtained in step S2;

[0155] And define the cost function J for steering wheel angle and its rate of change. u，k (k) and J du，k (k);

[0156] The following multi-objective cost function is constructed by combining the various cost functions:

[0157] J k =J zk +J uk +J duk +J Δuk +J εk

[0158] Among them, J εk =ρ ε ·ε k 2 ε is the cost function for slack variables, used to penalize violations of soft constraints in the optimization problem by state variables, control variables, etc. k ρ is a slack variable. ε Let be the weighting coefficient, and satisfy ρ ε >0.

[0159] Road tracking error cost function J zk The specific form can be expressed as:

[0160]

[0161] Among them, s z The normalized coefficient diagonal matrix is ​​specifically represented as follows:

[0162]

[0163] Each element in the matrix represents the normalization coefficient of the corresponding element in the state vector;

[0164] Z(k+h|k) and Zref (k+h|k) represent the predicted state and the reference state at time k+h, respectively. Q(k+h|k) is the positive semidefinite diagonal cost weight matrix used to penalize tracking errors, which will be adjusted and updated in real time during the rolling optimization process.

[0165] The specific forms of the intervention assistance cost function and the cost function defining the steering wheel angle and its rate of change are as follows:

[0166]

[0167]

[0168]

[0169] Where R(k), E(k), and F(k) are the weight matrices at time t = k, and u(k + h|k) represents the predicted optimal control quantity at time k. d (k+h|k) represents the driver's input intention control quantity at time k;

[0170] The above normalization coefficients are determined by the maximum and minimum values ​​of the corresponding parameters, respectively:

[0171]

[0172]

[0173]

[0174]

[0175]

[0176]

[0177]

[0178] A driving right decay coefficient Γ and a driving weight coefficient Θ are introduced to measure the confidence level of the driver's intention input and control the driving weight between the human and machine within the prediction time window, respectively. A cost function J is constructed as follows: k Solve for the coefficients:

[0179] J k = (1-Γ(t))·Q Track (k)·J Track，k +ΘΓ(t)J Δuk +n ε ·max{Γ(t), (1-Γ(t))}·J εk

[0180] coefficient

[0181] Where, n Γ，min This represents the short-term prediction portion within the prediction time window t = 1, 2, ..., p, i.e., the first few steps of the prediction step size, n. Γ，max This represents the long-term prediction portion within the prediction time window, i.e., the last few steps of the prediction step size. ΔT is the prediction sampling time interval, and J... track，k The partial cost function reflecting the vehicle state is denoted as: J track，k =[J zk J uk J duk ] T ;

[0182] Q(k), R(k), E(k), and F(k) are respectively represented as:

[0183] Q(k)=Q z，Track (k)·(1-Γ(t))

[0184] R(k)=Q u，Track (k)·(1-Γ(t))

[0185] E(k)=Q du，Track (k)·(1-Γ(t))

[0186] F(k)=ΘΓ(t)

[0187] Among them, Q z，Track Q u，Track Q du，Track The road tracking error cost function J is respectively zk Weighting coefficients, intervention auxiliary quantity cost function J uk Weighting coefficients, steering wheel angle cost function J duk Weighting coefficients;

[0188] The optimal front wheel steering angle output value is finally obtained through calculation. The dynamic change of the driving attenuation coefficient is as follows: Figure 4 As shown, the e-exponential decay effect occurs over time, which means that short-term predictions tend to focus more on meeting the driver's input, while long-term predictions rely more on the reference target trajectory.

[0189] It should be understood that the sequence number of each step in the embodiments of the present invention does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0190] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for tracking the steering assist trajectory of a multi-axis distributed drive vehicle based on AMPC, characterized in that: Specifically, the following steps are included: S1. In the upper-level control, for n Axis-distributed drive vehicle, with the vehicle's lateral and longitudinal coordinates xy Longitudinal velocity Heading angle Lateral velocity and yaw rate As input variables, a linear two-degree-of-freedom single-track vehicle model is established to predict the future dynamic state of the vehicle, and outputs the predicted optimal steering wheel angle control value, the lateral position error between the predicted trajectory and the externally input reference trajectory, and so on. and heading angle error ; S2. In the upper-level control, the vehicle's horizontal and vertical coordinates are used as the basis. xy Longitudinal velocity Heading angle Lateral velocity and yaw rate As input variables, a linear quadratic driver adjustment model is established. The cost function is constructed by minimizing the weighted sum of cumulative lateral error, lateral error and steering wheel angle to calculate the driver's intended front wheel angle. S3. In the lower-level control, the difference between the optimal steering wheel angle control amount predicted by the upper-level step S1 and the front wheel angle determined by the driver's intention front wheel angle obtained in step S2 is defined as the intervention assistance amount, and the intervention assistance amount, the lateral position error between the predicted trajectory and the reference trajectory are used. and heading angle error Minimizing all of these is taken as the optimization objective, while minimizing the steering wheel angle and its rate of change is also taken as the optimization objective. A multi-objective cost function is constructed by combining the above optimization objectives. S4. Solve for the optimal front wheel steering angle output value at the next moment based on the multi-objective cost function; The specific construction process of the linear two-degree-of-freedom monorail vehicle model described in step S1 is as follows: A dynamic analysis of a five-axle distributed drive vehicle is performed, taking into account control variables. , δ r For the front wheel steering angle, the disturbance amount , ρ Let the vehicle dynamic state variables be defined as follows, given the road curvature: ,in, and These are the lateral position error and the heading angle error, respectively; the following continuous state-space equations are established: In the formula, In the formula, C i For the first i Lateral stiffness of the shaft, L i For the first i The distance between the axis and the vehicle's center of gravity. I Let be the moment of inertia of the vehicle along the yaw direction. Cc The output matrix of the continuous state equation; Discretization yields the discrete state-space equations: x ( k+ 1)= A d x ( k )+ B u,d u ( k )+ B v,d v ( k ) Z ( k ) =C d x ( k ) In the formula, k Indicates time, It is the state matrix of the discrete state-space equation. It is the control matrix of the discrete state-space equations. It is the measurable perturbation matrix of the discrete state-space equations, with subscripts... d This indicates the discrete processing of the corresponding parameters. Cd for Cc The output matrix of the discrete state equations of the same form; Then the system in the future p The iterative form of the dynamic state at each time step is: in, The optimal sequence for solving adaptive model predictive control. p For the prediction time window, superscript T Represents the transpose of a vector; Within the prediction time window, the model prediction output is represented as: ； In the formula, Z p for k Always within the prediction time window p The model predicts the output sequence within the sequence. y For the single model prediction output within this prediction time window; The optimal steering wheel angle control value, the lateral position error between the predicted trajectory and the reference trajectory and heading angle error Defined as: In the formula, and These represent the vehicle's current position and various discrete points on the reference trajectory, respectively. The sideslip angle is the angle of the centroid. The angle between the tangent at the road mapping point where the vehicle is located and the angle in the global coordinate system is given by the superscript ·, which indicates the derivative with respect to the corresponding parameter.

2. The method as described in claim 1, characterized in that: The specific process of establishing the linear quadratic driver adjustment model in step S2 includes: Assuming the road curvature is small in actual roads: In the formula, , For measurable interference terms, u x The vehicle's longitudinal speed and the front wheel steering angle are given. To control the quantity; Establish the following state-space equations: In the formula, d Meaning and measurable interference terms same; Construct the following weighted sum cost function for cumulative lateral error, lateral error, and steering wheel angle. In the formula, Q and R These are the weighted matrices for the state variables and the control variables, respectively, both of which are positive semi-definite matrices, specifically represented as follows: Each element in the matrix represents the weight of a specific parameter; In the process of finding the driver's intended front wheel steering angle that minimizes the cost function, a curvature feedforward stage is introduced. The control quantity can then be expressed as: Using the algebraic Riccati equation ,Other To solve for the optimal control quantity The corresponding driver's intended front wheel steering angle is obtained; where K For state feedback gain, specifically .

3. The method as described in claim 2, characterized in that: In step S3, the state variables predicted in step S1 are defined as follows: Lateral position errors for the predicted trajectory and the reference trajectory respectively and heading angle error Define the road tracking error cost function: In the formula, the subscript ref Represents a reference trajectory; Define the cost function for intervention auxiliary quantities: In the formula, The control quantity predicted by the model in step S1, The driver's intention control quantity obtained in step S2; And define the cost function for steering wheel angle and its rate of change. and ; The following multi-objective cost function is constructed by combining the various cost functions: in, This is the cost function for slack variables, used to penalize violations of soft constraints in the optimization problem by state variables, control variables, etc. As slack variables, Let be the weight coefficient, and satisfy... ; Road tracking error cost function The specific form can be expressed as: in, The normalized coefficient diagonal matrix is ​​specifically represented as follows: Each element in the matrix represents the normalization coefficient of the corresponding element in the state vector; and They are respectively The predicted state and reference state at any given time. The positive semidefinite diagonal cost weight matrix used to penalize tracking errors will be adjusted and updated in real time during the rolling optimization process. The specific forms of the intervention assistance cost function and the cost function defining the steering wheel angle and its rate of change are as follows: in, for The weight matrix at time step, express Predict the optimal control input at all times. for The driver inputs the intended control quantity at all times; The above normalization coefficients are determined by the maximum and minimum values ​​of the corresponding parameters, respectively: Introducing a driving rights attenuation coefficient and driving weight coefficient These are used to measure the confidence level of the driver's intention input and control the human-machine driving weights within the prediction time window, respectively, and to construct the following cost function. Solve for the coefficients: coefficient in, Indicates the prediction time window The short-term forecast portion, i.e., the first few steps of the forecast step size, This represents the long-term forecast portion within the forecast time window, specifically the last few steps of the forecast step size, Δ. T To predict the sampling time interval, J track,k The partial cost function reflecting the vehicle's state is denoted as: J track,k =[ J zk , J uk , J duk ] T ; They are represented as follows: in, Q z,Track , Q u,Track , Q du,Track These are the road tracking error cost functions. J zk Weighting coefficients, intervention auxiliary cost function J uk Weighting coefficients, steering wheel angle cost function J duk Weighting coefficients; The optimal front wheel steering angle output value is finally obtained by solving the problem.

Images