An economic model predictive control method for an autonomous electric vehicle

Abstract

The application discloses an economic model predictive control method for an automatic driving electric vehicle. At each sampling time, whether the vehicle meets a control optimization trigger condition is judged according to a tracking error of the vehicle; if yes, a control trajectory is updated according to current sampling data; a control sequence interval longer than a set time interval is determined, and the updated control trajectory is output to the vehicle in the control sequence interval; if no, the control trajectory is not updated, and the control trajectory is output to the vehicle in the set time interval. The application can guarantee tracking performance in a vehicle control process and reduce the calculation and communication burden of a system.

Description

Technical Field

[0001] This invention relates to the field of autonomous electric vehicle control technology, and specifically to an economical model predictive control method for autonomous electric vehicles. Background Technology

[0002] Autonomous electric vehicles (EVs) have attracted widespread attention in modern autonomous driving, exploration missions, and intelligent cruise control. Due to the nonlinear dynamic characteristics of EVs, appropriate control methods are required to achieve control objectives. First, the control method for EVs needs to be able to calculate the optimal control input in real time based on the vehicle's current traffic environment and ensure that the vehicle maintains a certain level of tracking performance according to the real-time control objectives. Second, during the driving process of EVs, the control method also needs to consider many economic performance factors, such as energy consumption, driving safety, and comfort. Finally, the control method also needs to consider the burden on the vehicle's electronic systems, avoiding excessive consumption of computing resources and improving the stability and robustness of the vehicle's integrated electronic systems. Therefore, there is an urgent need for an EV control method that can improve the economic performance of EVs while reducing the system's computational and communication burden and ensuring control performance during the tracking control process. Summary of the Invention

[0003] In view of this, the present invention provides an economical model predictive control method for autonomous electric vehicles, which can reduce the computational and communication burden of the system while ensuring the tracking performance during the vehicle control process.

[0004] To solve the above-mentioned technical problems, the present invention is implemented as follows:

[0005] An economical model predictive control method for autonomous electric vehicles is provided. At each sampling time, the method determines whether the vehicle meets the control optimization triggering condition based on the vehicle's tracking error. If so, the control trajectory is updated based on the current sampling data. A control sequence interval longer than a set equal time interval is determined, and the updated control trajectory is output to the vehicle at the control sequence interval. If not, the control trajectory is not updated, and the control trajectory is output to the vehicle at a set equal time interval.

[0006] Preferably, the control optimization trigger condition is:

[0007] Design a total cost function based on the vehicle's tracking error; design a rotational total cost function based on the total cost function using dissipation theory; design the control optimization trigger condition based on the rotational total cost function, which ensures that the difference between the rotational total cost functions at adjacent time points decreases.

[0008] Preferably, the control sequence interval is:

[0009] The control sequence interval is obtained based on Lipschitz's theorem and Grangell's inequality. for:

[0010]

[0011] in, To control the upper bound of the sequence interval, m is the sequence interval number, and L... s This is the intermediate Lipschitz constant.

[0012] Preferably, the tracking error is:

[0013] Construct a vehicle kinematic model with the center point of the rear axle as the origin. for:

[0014]

[0015] in, For vehicle status, p h For the vehicle position, θ h For vehicle yaw angle, The control input received by the vehicle, υ h For vehicle speed, ω h Let w be the yaw rate of the vehicle, w be the disturbance experienced by the vehicle, and f(·,·) be the kinematic model function.

[0016] Constructing a vehicle reference trajectory model for:

[0017]

[0018] in, For reference only. For reference position, θ r For reference yaw angle, For reference control input, υ r For the virtual vehicle speed, ω r The yaw rate of the virtual vehicle;

[0019] The vehicle status ξ h With the reference state ξ r The difference between them is the tracking error ξ. e =ξ h -ξ r According to the tracking error ξ e Constructing an error model Among them, u e To control input error, f e (·,·) represents the nonlinear error model function.

[0020] Preferably, the total design cost function is:

[0021] The kth sampling time is t k The next sampling time adjacent to it is t. k+1 =t k +Φ M , Let T be the sampling interval, T be the prediction time domain, and M be the preset value of the control sequence; then, adjacent sampling times t will be... k and t k+1 The m-th control sequence interval σ between i Represented as σ m The control sequence interval number m∈[1,M] is determined according to the sampling time t. k The tracking error ξ e (t k ) and control input error u e (t k Design the total cost function J(ξ). e (t k ),u e (t k )),for:

[0022]

[0023] Wherein, E(ξ) e (t k +l|t k ),u e (t k +l|t k F(ξ) is the stage cost function. e (t k +T|t k ),u e (t k +T|t k )) is the terminal cost function, (t) k +l|t k ) represents the sampling time t k After l steps of prediction information, (t) k +T|t k ) represents the sampling time t k Then, the prediction information is obtained from the prediction time domain T.

[0024] Preferably, the total cost function for the design rotation is:

[0025] Based on the total cost function and dissipation theory, design the rotational total cost function. for:

[0026]

[0027] in, The cost function for the rotation stage. Let be the cost function of the rotating terminal, (t) k +l|t k ) represents the sampling time t k After the prediction information of the next step, (t) k +T|t k ) represents the sampling time t k Then, the prediction information is obtained from the prediction time domain T.

[0028] Preferably, the updated control trajectory is:

[0029] Based on the total cost function, an economic cost function is designed and an economic MPC optimal control problem is constructed; a set of constraints is constructed for the economic MPC optimal control problem to optimize the vehicle's tracking effect and economic performance; based on the optimization constraints of the set of constraints, the economic MPC optimal control problem is solved, and the vehicle's control trajectory is updated.

[0030] Preferably, the set of constraints includes:

[0031] The first constraint reflects the predicted trajectory and prediction error starting from the initial moment;

[0032] The second constraint reflects the control inputs related to the predicted trajectory;

[0033] The third constraint reflects the range of control inputs associated with the predicted trajectory;

[0034] The fourth constraint reflects the range of reference control inputs related to the tracking trajectory;

[0035] The fifth constraint reflects the state equation related to the error trajectory;

[0036] The sixth constraint reflects the feasibility and stability of the system.

[0037] Beneficial effects:

[0038] 1. This invention determines whether to optimize and update the control trajectory and optimize the output sequence interval at the current sampling time by judging whether the control optimization trigger condition is met at the current moment. This avoids redundant optimization calculations when the tracking error is small, and reduces the computational burden of the system while ensuring tracking performance. After optimizing the control trajectory, this invention further outputs the control trajectory with the optimized control sequence interval, reducing the communication burden of the system.

[0039] 2. This invention designs a rotating economic cost function and control optimization triggering conditions by adopting dissipative theory. By constructing and judging the control optimization triggering conditions, not only can the safety and effectiveness of the method be guaranteed under all conditions, but it also achieves the maximum optimization of vehicle communication and computing burden, reduces the burden on the automotive electronic system, and does not have a negative impact on the vehicle's tracking performance.

[0040] 3. The economic model predictive control method for autonomous electric vehicles provided by this invention optimizes the optimal control sequence interval time and obtains an explicit solution in order to reduce the communication burden. By parameterizing the self-triggered EMPC algorithm, the method ensures a balance between control performance and system computational burden, and shows superiority over the existing time-triggered MPC algorithm in terms of computational burden and triggering frequency.

[0041] 4. The economic model predictive control method for autonomous electric vehicles provided by this invention ensures the recursive feasibility of the optimization problem, the convergence of the closed-loop system, and the asymptotic average performance by constructing an economic cost function and an economic MPC optimal control problem. Compared with existing EMPC and other self-triggered control algorithms, the self-triggered EMPC control algorithm designed in this invention has better economic performance.

[0042] 5. By setting a set of constraints, this invention not only solves the optimal control problem of economic MPC, but also allows for targeted settings for factors that need to be optimized, improving the flexibility of the method's application and subsequent upgrades, and has universal application value. Attached Figure Description

[0043] Figure 1 This is a flowchart of the economic model predictive control method for autonomous electric vehicles based on an embodiment of the present invention. Detailed Implementation

[0044] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0045] This invention provides an economical model predictive control algorithm for autonomous electric vehicles, the core idea of ​​which is:

[0046] At each sampling time, the system determines whether the vehicle meets the control optimization trigger condition based on the vehicle's tracking error. If it does, the control trajectory is updated based on the current sampling data. A control sequence interval longer than the set time interval is determined, and the updated control trajectory is output to the vehicle at the control sequence interval. Otherwise, the control trajectory is not updated, and the control trajectory is output to the vehicle at the set time interval.

[0047] As can be seen, this invention determines whether to optimize and update the control trajectory and optimize the output sequence interval at the current sampling time by judging whether the control optimization trigger condition is met at the current moment. This avoids redundant optimization calculations when the tracking error is small, and reduces the computational burden of the system while ensuring tracking performance. After optimizing the control trajectory, this invention further outputs the control trajectory with the optimized control sequence interval, reducing the communication burden of the system.

[0048] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0049] This invention provides an economical model predictive control method for autonomous electric vehicles, such as... Figure 1 As shown. Existing Model Predictive Control (MPC) has advantages such as being adept at handling nonlinear dynamics problems, handling numerous constraints, and being capable of online optimization. The optimal control input obtained using MPC is acquired through repeated online prediction and optimization, and it can be used to solve pre-constructed optimal control problems. However, since autonomous electric vehicles need to consider many economic performance factors during operation, such as energy consumption, driving safety, and comfort, MPC is difficult to directly improve economic performance. Therefore, Economic Model Predictive Control (EMPC) was proposed based on MPC. EMPC is a control algorithm that designs economic factors as economic costs to directly optimize the economic performance of the system. However, during autonomous driving, autonomous electric vehicles also need to calculate the optimal control input in real time based on the current traffic environment, while the calculation and implementation of the control input in existing EMPC are time-driven. Therefore, in order to eliminate the limitations of time-driven MPC, this invention adopts self-triggered MPC, which determines the next trigger time based on the current system state and system dynamics.

[0050] The first embodiment of the present invention specifically includes the following steps:

[0051] Step 1: Construct a kinematic model of the autonomous electric vehicle, incorporating a reference trajectory generated by a virtual vehicle to obtain a tracking error model. This includes the following steps:

[0052] Step 101: Construct a kinematic model of the autonomous electric vehicle with the center point of the rear axle as the origin. for:

[0053]

[0054] Wherein, the nonlinear motion model function f(·,·) satisfies f(·,·): The values ​​in parentheses represent the independent variables of the nonlinear motion model function. The vehicle's status and location are... The vehicle yaw angle is θh Control input u h for With the vehicle's speed υ h =v(t) and vehicle yaw rate in For reference wheel deflection angle, Let w be the vehicle wheelbase, and w ≤ w max For the disturbances experienced by the system, w max To disturb the upper bound, [...] T This represents the transpose of a matrix.

[0055] Considering that the kinematic model is within a controllable range, the control input and state constraints of the system are as follows:

[0056]

[0057] in, and These represent the compact and convex state set and output set, respectively. and They are n-dimensional space and m-dimensional space, respectively.

[0058] Step 102: To obtain the reference trajectory, a virtual vehicle model is introduced as follows:

[0059]

[0060] in, and and It is a compact and convex set. The reference state is determined by... Define, where and θ r Representing position and vehicle yaw angle respectively, the control input is provided by Defined as having the speed υ of a virtual vehicle. r And the yaw rate ω of the virtual vehicle r .

[0061] Step 103, the error ξ between the vehicle state (1) and its reference trajectory (3) e =ξ h -ξ r And construct an error model in and These are the state error and the corresponding control input error, f. e (·,·) represents the nonlinear error model function, and the variables are within the parentheses.

[0062] Step 2: To reduce the communication and computational burden on the vehicle, a dissipative theory is used to design the rotation cost function. A rule ensuring a decreasing difference in the rotation cost function between adjacent time steps is used to design self-triggering conditions, i.e., control optimization triggering conditions. The optimal control sequence interval is then solved using these self-triggering conditions. Specifically, this includes the following steps:

[0063] Step 201, take the sampling time as The next sampling time is marked as t k+1 =t k +Φ M The sampling interval time is T represents the prediction time domain, and M represents the preset value of the control sequence; adjacent sampling times t k and t k+1 The m-th control sequence interval σ between i Represented as σ m The control sequence interval number m∈[1,M].

[0064] Step 202, at sampling time t k The total cost function is designed as follows:

[0065]

[0066] Wherein, E(ξ) e (t k +l|t k ),u e (t k +l|t k )) and F(ξ e (t k +T|t k ),u e (t k +T|t k ()) represent the stage cost function and the terminal cost function, respectively. In this invention, unless otherwise specified, parentheses are added after a physical quantity to indicate the time of its value, such as ξ. e (t k (t) represents the sampling time. k Vehicle tracking error ξ e Accordingly, (t) k +l|t k ) represents the sampling time t k The predicted information of the corresponding quantity in the next l steps, (t) k +T|t k ) represents the sampling time t k After predicting the corresponding quantity in the prediction time domain T, other similar time representations are described similarly. Based on dissipation theory, the cost function for the rotation stage is designed as follows:

[0067]

[0068] Wherein, the storage function γ(·): It is bounded, and the storage function satisfies the inequality γ(f) in dissipative theory. e (ξ e ,u e ))-γ(ξ e )≤α(ξ e ,u e The support function is α(ξ). e ,u e )=E(ξ e ,u e )-E s E s The optimal value function is the function corresponding to the optimal steady state and the input, which are defined as follows:

[0069]

[0070] According to dissipation theory, the cost function of the rotating terminal is:

[0071]

[0072] Therefore, the total rotation cost function is:

[0073]

[0074] Step 203: To ensure that the total rotation cost function decreases at adjacent transmission times, a self-triggering condition is designed. in, Indicates t k+1 The optimal value of the total rotational cost function corresponding to the optimal error state and optimal error control input at time t. Indicates t k The optimal value of the total rotational cost function corresponding to the optimal error state and the optimal error control input at time t is obtained. According to the classical stability theorem of model predictive control, we can obtain:

[0075]

[0076] The above inequality can also be written as:

[0077]

[0078] Will The upper bound is defined as G(σ1,…,σ M If ), then inequality (10) can be rewritten as:

[0079]

[0080] Among them, L J The total rotation cost function The Lipschitz constant is used to design the self-triggering condition as follows:

[0081]

[0082] Where β∈(0,1) represents the triggering factor, which can be further simplified to:

[0083]

[0084] That is, the designed self-triggering condition ensures that the difference in the rotation cost function between adjacent time steps decreases.

[0085] This invention designs a rotating economic cost function and control optimization triggering conditions using dissipative theory. By constructing and judging the control optimization triggering conditions, it not only ensures the safety and effectiveness of the method under all conditions, but also achieves maximum optimization of vehicle communication and computing burden, reduces the burden on the automotive electronic system, and does not negatively affect the vehicle's tracking performance.

[0086] Step 204: Calculate the actual error and nominal error The difference norm between The upper bound is defined as the upper bound value function. in For intermediate parameter variables, The optimal control sequence interval is defined. The value of the number M of the optimal control sequence interval is analyzed. When M=1, the optimal control sequence interval is set... Upper bound function Actual error and nominal error The difference norm between The upper bound; when M≥2, set At this time, the upper bound function Actual error and nominal error The difference norm between The upper bound is set when M=2. Follow the above steps until M optimal control sequence intervals are obtained. That is, for 2 ≤ m < M, when the self-triggering condition is met, the upper bound of the control sequence interval is obtained. Determine the upper bound function The value of is then determined by maximizing the upper bound function. and The optimal control sequence interval is obtained by using the difference between them. Right now When m = M, set

[0087] Step 205: For 2 ≤ m ≤ M, according to Lipschitz's theorem and the Gronwall-Bellman inequality, It can be recursively defined as:

[0088]

[0089] in, τ∈[0,T], define the intermediate function s(ξ) e (t k ),u e (t k ))=f e (ξ e (t k ),u e (t k ))-ξ e (t k ), intermediate function s(ξ e (t k ),u e (t k )) represents the error model Sum of error ξ e The difference between them, L s The intermediate function s(ξ) e (t k ),u e (t k The intermediate Lipschitz constant is the Lipschitz constant representing the difference between the vehicle's error model and the error itself. This is achieved by maximizing... and The difference between them yields the optimal control sequence interval, which is:

[0090]

[0091] Furthermore, by simplification, we obtain:

[0092]

[0093] Define function The function Z(σ) m ) relative to σ m Perform differentiation, and let The final form of the explicit solution for the optimal control sequence interval is:

[0094] Determine the sampling time t kIf the self-triggering condition is not met, then the set equal time interval is used as the control sequence interval; if it is met, then the optimal control sequence interval is solved based on the explicit solution of the optimal control sequence interval.

[0095] The present invention provides an economic model predictive control method for autonomous electric vehicles. In order to reduce communication and computational burden, the optimal control sequence interval time is optimized and an explicit solution is obtained. By parameterizing the self-triggered EMPC algorithm, the method has the ability to balance computational efficiency and control performance, and shows superiority over existing time-triggered MPC and self-triggered algorithms in terms of computational burden and triggering frequency.

[0096] Step 3: Considering ideal tracking performance and economic efficiency, design the corresponding economic cost function based on the kinematic model, and construct an economical MPC optimal control problem. This includes the following steps:

[0097] Step 301: Considering the ideal tracking effect and economic performance of electric vehicles, the stage cost function E(ξ) in cost function (3) is... e (t k +l|t k ),u e (t k +l|t k The design is as follows:

[0098] E(ξ e (t k +l|t k ),u e (t k +l|t k ))=E t (p e (t k +l|t k ),u e (t k +l|t k ))+E e (a h (t k +l|t k ),υ h (t k +l|t k (17)

[0099] The stage cost function consists of two parts, namely E t (p e (t k +l|t k ),u e (t k +l|t k)) represents the tracking cost function, which uses the position error p e (t k +l|t k ) and input error u e (t k +l|t k To assess the tracking capabilities of electric vehicles, E e (a h (t k +l|t k ),υ h (t k +l|t k )) represents the economic cost function, which uses acceleration a. h (t k +l|t k ) and its relationship with velocity υ h (t k +l|t k The product of ) reflects energy consumption, and the expression for the two parts of the cost function is as follows:

[0100]

[0101]

[0102] Where, matrices P and Q are positive definite symmetric matrices, μ a and μ b Let p be the coefficient of the economic cost function, and p be the location error. e (t k +l|t k ) = p h (t k +l|t k )-p r (t k +l|t k ), (t k +l|t k (t) represents time t k Predictive information for the next step, p e (t k +l|t k (t) represents time t k The position error in the next l steps, u e (t k +l|t k (t) represents time t k The input error in the next l steps, a h (t k +l|t k (t) represents time t k The actual acceleration in the next l steps, υ h (tk +l|t k (t) represents time t k The actual speed of the next l steps, p r (t k +l|t k ) is the newly generated reference trajectory.

[0103] Step 302, at time t k For l = 0, 1, ..., T-1, the economical MPC optimization problem for electric vehicles tracking economical driving is described as the following optimization problem: The optimization problem has a set of constraints, namely:

[0104] ξ h (t k |t k )=ξ h (t k ), ξ e (t k |t k )=ξ e (t k Constraint (20a)

[0105] u h (t k +l|t k )=u e (t k +l|t k )+u r (t k Constraint (20b)

[0106] Constraint (20c)

[0107] Constraints (20d)

[0108] ξ e (t k +l|t k )=f e (ξ e (t k +l|t k ),u e (t k +l|t k Constraint (20e)

[0109] E(ξ e (t k +T|t k ),u e (t k +T|t k))∈Ω; Constraint (20f)

[0110] Where, ξ h (t k |t k ) for t k At time ξ, the predicted initial position of the electric vehicle; h (t k ) for t k Location information of electric vehicles at any time; ξ e (t k |t k ) for t k At time ξ, the initial prediction error position of the electric vehicle; e (t k ) for t k Position error information of electric vehicles at any time; u h (t k +l|t k (t) represents time t k The control input for the next step, u r (t k +l|t k (t) represents time t k The reference control input for the next step; ξ e (t k +l|t k (t) represents time t k The state error in the next l steps; ξ e (t k +T|t k (t) represents time t k The state error in the next T steps; E(ξ) e (t k +T|t k ),u e (t k +T|t k Let Ω be the terminal cost and Ω be the terminal domain. The first constraint (20a) represents the predicted trajectory and prediction error starting from the initial time; the second constraint (20b) represents the control input associated with the predicted trajectory; the third constraint (20c) represents the control input constraint associated with the predicted trajectory; the fourth constraint (20d) represents the reference control input constraint associated with the trackable trajectory; the fifth constraint (20e) represents the state equation associated with the error trajectory; and the sixth constraint (20f) guarantees the feasibility and stability of the system.

[0111] Step 303, Terminal cost is:

[0112]

[0113] Wherein, the terminal cost weighting matrix Z is a positive definite symmetric matrix, and the terminal domain Ω is:

[0114]

[0115] The economic model predictive control method for autonomous electric vehicles provided by this invention ensures the recursive feasibility of the optimization problem, the convergence of the closed-loop system, and the asymptotic average performance by constructing an economic cost function and an economic MPC optimal control problem. Compared with existing EMPC and other self-triggered control algorithms, the self-triggered EMPC control algorithm designed in this invention has better economic performance. By setting a set of constraints, not only is the solution to the economic MPC optimal control problem realized, but also the factors that need to be optimized can be specifically set, improving the flexibility of the method's application and subsequent improvement and upgrade, and possessing universal applicability.

[0116] Step 4: Determine the sampling time t k The system checks whether the self-triggering condition is met. If not, it generates a corresponding control sequence based on the state trajectory of the autonomous electric vehicle at the previous moment and outputs the control sequence to the vehicle's control unit at set time intervals. If the condition is met, it solves the economical MPC optimal control problem constructed in step 3. A suitable nonlinear solver is used for the solution; in this embodiment, the IPOPT solver is used to obtain the current state trajectory of the autonomous electric vehicle and the corresponding control sequence. The current control sequence is transmitted to the vehicle's control unit at the optimal control sequence interval obtained in step 2, thereby reducing the communication and computational burden on the vehicle while achieving economical driving during the tracking process.

[0117] In summary, the flow of an economical model predictive control algorithm for autonomous electric vehicles can be described as follows: Offline design: number of optimal control sequence intervals M, trigger factor β∈(0,1), prediction time domain T, sampling time T p The weight matrices P, Q, and Z are used to determine the terminal set Ω according to step 303. The algorithm flow includes:

[0118] Procedure 1: At any update time, measure the current error state quantity ξ. e (t k Given ξ e (t k |t k )=ξ e (t k );

[0119] Step 2: Determine if the current self-triggered condition is met; if not, proceed to Step 3; if met, proceed to Step 4.

[0120] Step 3: Generate the control sequence and state trajectory of the current moment based on the control sequence and state trajectory of the previous moment, and output them to the vehicle unit at set equal time intervals, and proceed to Step 7;

[0121] Step 4: Solve the optimization problem to obtain the optimal control sequence. and optimal state trajectory

[0122] Step 5: For a given M, calculate the optimal control sequence interval using (16). Time to obtain the next transmission control input

[0123] Step 6, in The optimal control input sequence The output is sent to the vehicle's infotainment system, and this sequence is used as the current control input to solve the next optimization problem.

[0124] Process 7, Update Time Return to process 1.

[0125] This embodiment further utilizes the YALMIP toolbox and IPOPT solver in MATLAB for simulation experiments to verify the effectiveness and feasibility of the algorithm of this invention. Simulation parameters are selected as follows: the initial states of the virtual vehicle and the autonomous electric vehicle are set to... and The vehicle wheelbase is upper bound of perturbation w max =0.05, prediction time domain is T=10, sampling time is T p =0.1s, the predefined optimal control sequence interval number M=3, Lipschitz constant L s =1,L J =2.41, the weighting matrices in the tracking cost function and terminal cost function are P = diag{0.5, 0.5}, Q = diag{0.08, 0.08}, Z = diag{0.5, 0.5}, and the coefficient μ in the economic cost function. a =1×10 -6 μ b =1×10 -5 .

[0126] The simulation environment is a space with a length of 25m and a width of 15m. The front of the autonomous electric vehicle is a virtual vehicle. The simulation results show that the autonomous electric vehicle can improve economic performance while reducing the computational and communication burden, and at the same time ensure control performance in the tracking control process.

[0127] In the second embodiment of the present invention, compared with the first embodiment, the relatively complex self-triggering condition is not used as the control optimization trigger condition. Instead, a tracking error upper limit is directly selected based on the actual operating environment of the vehicle, and the range of this tracking error upper limit is used as the control optimization trigger condition. This tracking error upper limit is a preset preferred value, which is selected and set by the controller according to the specific operating conditions and the expected accuracy of autonomous driving. When the current tracking error of the vehicle is less than or equal to the tracking error upper limit, in order to reduce the computational burden on the vehicle's control unit, no optimization calculation is performed on the control trajectory, and the control sequence is still output to the vehicle's control unit using the control trajectory of the previous moment. When the current tracking error of the vehicle is greater than the tracking error upper limit, the vehicle's control unit performs optimization calculation on the control trajectory based on the current trajectory state, and outputs the updated control trajectory to the vehicle's control unit at the optimal control sequence interval in Embodiment 1. Since the optimal control sequence interval is longer than the equal time interval used in general control and can maintain the same tracking accuracy, the communication burden of the vehicle's control unit is reduced.

[0128] This embodiment simplifies the design of the control optimization triggering condition. Compared with the self-triggered condition in Embodiment 1 where the difference in the total rotation cost function between adjacent time points decreases, this further reduces the computational burden on the autonomous vehicle's infotainment system at the expense of the vehicle's tracking accuracy.

[0129] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An economical model predictive control method for autonomous electric vehicles, characterized in that, At each sampling time, the vehicle's tracking error is used to determine whether the vehicle meets the control optimization trigger condition; if so, the control trajectory is updated based on the current sampling data; a control sequence interval longer than the set time interval is determined, and the updated control trajectory is output to the vehicle at the control sequence interval. No, the control trajectory will not be updated, and the control trajectory will be output to the vehicle at set time intervals; Design a total cost function based on the vehicle's tracking error; design a rotational total cost function based on the total cost function using dissipation theory; design the control optimization trigger condition based on the rotational total cost function, wherein the control optimization trigger condition is: ensuring that the difference between the rotational total cost function at adjacent time points decreases.

2. The economic model predictive control method for autonomous electric vehicles as described in claim 1, characterized in that, The control sequence interval is: The control sequence interval is obtained based on Lipschitz's theorem and Grangell's inequality. for: ； （I） in, To control the upper bound of the sequence interval, m is the sequence interval number. This is the intermediate Lipschitz constant.

3. The economic model predictive control method for autonomous electric vehicles as described in claim 1, characterized in that, The tracking error is: Construct a vehicle kinematic model with the center point of the rear axle as the origin. ,for: in, Vehicle status. For vehicle location, For vehicle yaw angle, The control inputs received by the vehicle, For vehicle speed, Let ω be the yaw rate of the vehicle, and w be the disturbance experienced by the vehicle. For nonlinear kinematic model functions; Constructing a vehicle reference trajectory model ,for: ; (III) in For reference only. For reference position, For reference yaw angle, For reference control input, For virtual vehicle speed, The yaw rate of the virtual vehicle; The vehicle status With the reference state The difference between them is the tracking error. ; According to the tracking error Constructing an error model ,in, To control input error, This is a nonlinear error model function.

4. The economic model predictive control method for autonomous electric vehicles as described in claim 1, characterized in that, The total design cost function is: The k-th sampling time is The next sampling time adjacent to it is , ≤ T is the sampling interval, T is the prediction time domain, and M is the preset value of the control sequence; adjacent sampling times are... and The m-th control sequence interval between Represented as The sequence interval number m∈[1, M] is controlled according to the sampling time. The tracking error and control input error Design the total cost function ,for: in, For stage cost function, For terminal cost function, Indicates the sampling time After Predictive information for steps, Indicates the sampling time Then, the prediction information is obtained from the prediction time domain T.

5. The economic model predictive control method for autonomous electric vehicles as described in claim 1, characterized in that, The design rotation total cost function is: Based on the total cost function and dissipation theory, design the rotational total cost function. ,for: in, The cost function for the rotation stage. For the cost function of the rotating terminal, Indicates the sampling time After Predictive information for steps, Indicates the sampling time Then, the prediction information is obtained from the prediction time domain T.

6. The economic model predictive control method for autonomous electric vehicles as described in claim 1, characterized in that, The updated control trajectory is as follows: Based on the total cost function, an economic cost function is designed and an economic MPC optimal control problem is constructed; a set of constraints is constructed for the economic MPC optimal control problem to optimize the vehicle's tracking effect and economic performance; based on the optimization constraints of the set of constraints, the economic MPC optimal control problem is solved, and the vehicle's control trajectory is updated.

7. The economic model predictive control method for autonomous electric vehicles as described in claim 6, characterized in that, The constraint set includes: The first constraint reflects the predicted trajectory and prediction error starting from the initial moment; The second constraint reflects the control inputs related to the predicted trajectory; The third constraint reflects the range of control inputs associated with the predicted trajectory; The fourth constraint reflects the range of reference control inputs related to the tracking trajectory; The fifth constraint reflects the state equation related to the error trajectory; The sixth constraint reflects the feasibility and stability of the system.

Images