Data-driven hybrid vehicle platoon robust control method

Abstract

The application discloses a data-driven robust control method for a mixed vehicle platoon, and a linear model of the mixed longitudinal vehicle platoon is established by using an extended dynamic mode decomposition (EDMD) algorithm to approximate an infinite-dimensional Koopman operator. Unlike existing related work, a model mismatch error is introduced in the data-driven linear model. Secondly, a tube model predictive control (TubeMPC) method is used to design a robust controller for an autonomous vehicle, so that the actual state trajectory of all vehicles in the platoon is constrained in a tube with the nominal trajectory as the center, and bounded stability is achieved. Finally, the effectiveness of the designed data-driven modeling method and the TubeMPC controller is verified through numerical simulation, and the results show that the method can overcome the influence of uncertainties such as model mismatch errors and achieve fast and efficient closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of vehicle networking application technology, specifically to a data-driven robust control method for hybrid vehicle platoons, used to achieve robust control of intelligent connected and autonomous vehicles in hybrid vehicle platoons, so as to ensure traffic safety and improve traffic efficiency. Background Technology

[0002] The increasing number of vehicles places higher demands on road capacity, traffic safety, and traffic efficiency. With technological advancements, the emergence of intelligent connected vehicle (HAV) technology offers new solutions to various traffic problems, serving as a crucial means to achieve road safety and energy conservation. The emergence and rapid development of HAVs can significantly alter the inherent characteristics of traffic flows composed entirely of human-driven vehicles, improving road utilization and effectively preventing collisions. For a considerable period in the future, roads are likely to be dominated by mixed vehicle platoons consisting of conventionally driven vehicles (HDVs) and conventionally driven vehicles (CAVs). Considering the random driving characteristics of humans and the uncertainties in the interaction between HDVs and CAVs, the traffic environment will become more complex, posing significant challenges to platoon modeling and control. Therefore, how to control CAVs and improve the safety of mixed vehicle platoons in situations where conventionally driven and conventionally driven vehicles coexist will be a critical research topic that urgently needs to be addressed.

[0003] As the controllable component of a mixed vehicle platoon, existing research largely employs Model Predictive Control (MPC) to replan the CAV trajectory at each time step (e.g., 0.1 seconds). MPC is widely used in CAV motion control, enabling real-time prediction of vehicle states and handling various constraints. However, many studies on CAV control utilize traditional MPC methods, neglecting the impact of uncertainties such as model mismatch caused by various real-world conditions. While uncertainties are typically assumed to be bounded within a time step, they can accumulate as the prediction time range increases. Controlling the CAV without suppressing accumulated uncertainties can lead to deviations in CAV states (e.g., spacing and speed) from their planned trajectories, resulting in slow, overly conservative control performance and potential violations of system constraints such as safety and stability. Therefore, designing a robust controller for the CAV is crucial in this context.

[0004] This invention proposes a data-driven Tube Model Predictive Control (Tube MPC) method, innovatively combining the Extended Dynamic Mode Decomposition (EDMD) algorithm with Tube MPC to solve the modeling and control problem of mixed vehicle platoons. Its goal is to control the CAVs (Carrier Aerial Vehicles) to achieve the desired state of the entire platoon, making the mixed vehicle platoon stable and robust. The most significant difference in the control method proposed in this invention is that the data-driven robust vehicle control strategy can effectively cope with unexpected situations during platoon operation, suppress the negative impacts of various uncertainties, and achieve fast and efficient closed-loop control. Summary of the Invention

[0005] This invention proposes a data-driven robust control method for hybrid vehicle platoons. This method designs a control framework incorporating feedforward and feedback control. Feedforward control plans the optimal trajectory for the nominal system, while feedback control keeps the state tracking error between the actual system and the nominal system within the minimum robust positive invariant set. Ultimately, this drives the actual state of the vehicles to approach the desired state, enabling the hybrid vehicle platoon to achieve the desired driving effect. Compared to traditional MPC control methods, this invention suppresses uncertainties during platoon operation while ensuring traffic safety and improving road occupancy.

[0006] This invention aims to model a hybrid vehicle platooning system based on the extended dynamic pattern decomposition algorithm, and then design a TubeMPC controller to control the CAVs in the hybrid vehicle platoon. The specific implementation steps are as follows:

[0007] Step 1: Collect traffic data

[0008] Because data-driven modeling algorithms are based on vehicle state data from mixed vehicle queues, therefore... Figure 1 The mixed traffic scenario shown was simulated using MATLAB to obtain the required mixed vehicle queue state data. The HDVs in the mixed vehicle queue employ an IDM driving strategy, while the CAVs are controlled using a simple feedback controller to meet safety requirements. Sufficient vehicle state data was obtained through simulation on the MATLAB platform. The resulting set of sequential, time-evolving traffic queue data snapshots can be represented as:

[0009] {x 1,1 ,x 2,2 ,…,x n,m}(1)

[0010] Where x i,k =[p 1,k …p n,k ] T ∈R 2n p i,k =[h i,kv i,k ] T h i,k v represents the distance between the i-th car and the car in front at time k. i,k represents the speed of the i-th car at time k, and m represents the time series.

[0011] Step 2: Establish a mathematical model for the mixed vehicle queuing system.

[0012] This invention employs the Extended Dynamic Pattern Decomposition (EPD) algorithm to model a mixed vehicle queuing system. To utilize the traffic data obtained in the previous step for system modeling, the mixed vehicle queuing data is first represented as a matrix:

[0013] X = [x1, x2, ..., x K-1 ] T

[0014] X + = [x2,x3,…,x K ] T (2)

[0015] U = [u1, u2, ... u] K-1 ] T

[0016] First, a lifting function Ψ(x) is constructed to lift the original state of the system to a higher-dimensional space. The construction of the lifting function is best based on a known fundamental model of hybrid vehicle dynamics. According to the driving strategy of a traditional driver controlling a vehicle in a hybrid vehicle platoon, the lifting function is constructed as follows:

[0017]

[0018] By projecting the original state of the system into a higher-dimensional space using a lifting function, we obtain the corresponding data snapshot:

[0019]

[0020] Then, the system matrices A and B can be obtained by solving the optimization problem.

[0021]

[0022] The analytical solution to this problem is:

[0023]

[0024] Matrix C can be obtained using the least squares method:

[0025]

[0026] Solving the above algebraic problem easily yields the system matrices A, B, and C. However, the number of vehicles in the mixed vehicle queues studied is limited, typically ranging from a few to several dozen. Even with increased dimensionality, X... lift It remains a short and wide matrix, meaning the number of columns exceeds the number of rows. To reduce computational resource consumption, it is more beneficial to calculate the normal equations to solve for matrices A and B. The normal equations are:

[0027] M=[AB]V (8)

[0028] in

[0029]

[0030] The approximate system matrices A and B can then be easily obtained, leading to the linear model of the mixed vehicle queue with zero uncertainty:

[0031]

[0032] Considering that the derivation of (9) from (3) may introduce modeling errors, it will lead to:

[0033]

[0034] Where ε i It is explicitly or implicitly bounded, i = 1, ..., N+2n. Let [ε1 ε2 … ε N+2n ] T Defined as w, the linear model of the mixed vehicle queue considering model mismatch is then obtained as follows:

[0035]

[0036] Since the controller designed in this invention is based on an incremental model, it is necessary to convert the physical model (12) into an incremental model. The vehicle's balance control quantity is known (0), and the vehicle's balance state cannot be predetermined, i.e., u * =0, As * -s * ≠0, the total modeling error is As * -s * +w=[μ1 μ2 … μ N+2n ] T , recorded as Where μ i Explicitly or implicitly bounded. For all i∈{1,2,...,2n}, W N+2n It is compact and includes the origin. Therefore, the linear incremental model of the hybrid vehicle queuing system with model error is:

[0037]

[0038] Where Δs k =s k -s * ,Δu k =u k . W is compact and contains the origin.

[0039] Step 3: Design Feedback Control

[0040] In this step, we design the feedback control section of the Tube MPC controller for the mixed vehicle platoon system. This feedback control dynamically reduces the deviation between the actual and nominal states of the vehicles. Specifically, we first determine the nominal system of the mixed vehicle platoon, then obtain the error system, and finally obtain the feedback matrix K by establishing and solving a discrete linear quadratic regulator problem. The specific solution process is as follows:

[0041] set up and Let the state and control inputs of the nominal model of the hybrid vehicle queue at time k be denoted as follows. Ignoring bounded uncertainty disturbances, the nominal system of the hybrid vehicle queue can be obtained as follows:

[0042]

[0043] Because system (13) has a continuous and bounded unknown disturbance. The actual state of a system is not always equal to its nominal state. Define the actual lifted state Δs of the system relative to its nominal lifted state. The deviation is e s , Accordingly, the deviation of the actual state Δx of the system from the nominal lifted state Δx is defined as e. x , The error system for the mixed vehicle queue is as follows:

[0044]

[0045] in As the control input of the error system, in order to prevent the cumulative propagation of error from causing system divergence, a feedback auxiliary controller K is designed for the error system of (15) to suppress the propagation of error. The auxiliary feedback control rate is:

[0046]

[0047] The feedback matrix K is solved by establishing and solving a discrete linear quadratic regulator problem:

[0048]

[0049] Where Q and R are the weight matrices for the boosting state and control input, respectively.

[0050] Step 4: Design feedforward control

[0051] The purpose of feedforward control is to ensure that the nominal state trajectory satisfies the tightened constraints, and ultimately to make the nominal state strictly move toward the target terminal trajectory at each time step. To tighten the constraints, it is necessary to first find a path containing all e... s Robust positive invariant set of mixed vehicle queues. Satisfy all and All To reduce conservatism, we hope The smaller the better, the smallest robust positive invariant set is:

[0052]

[0053] Although it is usually impossible to obtain Explicit representation, but can be obtained The estimation of robust positive invariant sets is presented in this paper. The inner approximation method is used to estimate the convex set of the robust positive invariant set. The internal approximate perturbation invariant set

[0054]

[0055] in If A k It is a nilpotent function of α (A α =0), as α→∞, we have

[0056] Then the nominal boost state and nominal control satisfy:

[0057]

[0058] The tightened constraints satisfied by the nominal trajectory can ensure that the actual vehicle state and control constraints are satisfied, thus ensuring the safety of the hybrid vehicle platoon system.

[0059] Traditional finite-time design criteria for model predictive control cannot guarantee asymptotic stability; closed-loop stability can only be achieved by appropriately adjusting design parameters. Therefore, the Tube MPC controller designed in this invention adds terminal constraints. After the system state enters the terminal constraint set within a finite number of steps, it reaches the system's stable point under the action of the state feedback control law, thus ensuring the stability of the closed-loop system.

[0060] Define S f It is the terminal constraint invariant set, terminal constraint set S f The following requirements must be met:

[0061]

[0062] To facilitate the calculation of the system's terminal state constraint set, the system's constraints are processed as follows, and formula (20) can be modified as follows:

[0063]

[0064] (14) The maximum terminal constraint invariant set of the defined dynamics and constraints (22) can be expressed as:

[0065]

[0066] Where v is the value that makes The smallest positive integer, for all satisfy Because A k It is strictly stable and (A) k If F+GK is observable, then v must be a finite value. Therefore, calculating v can be transformed into solving the following linear programming problem:

[0067]

[0068] j is the j-th row of F+GK, j = 0,...,n, m = 1,...,v. Clearly, the value of v does not depend on the system's state variables. Therefore, v can be obtained offline. Formula (23) defines a polyhedron set, and this maximally invariant set is used as the terminal state constraint set S. f .

[0069] Therefore, the feedforward control design for a hybrid vehicle platoon can be formulated as the following optimization problem:

[0070]

[0071] In the above optimization of the quadratic cost function These are the decision variables, and Q and R are the weight matrices for the state and control variables, respectively. It is relative to the nominal elevation state. terminal cost, A symmetric positive definite matrix P is a solution to the Lyapunov equations:

[0072]

[0073] in

[0074] Useful Explanation:

[0075] This invention proposes a data-driven robust control method for hybrid vehicle platoons. By constructing a set of lifting functions and using the EDMD algorithm to establish a linear model of the hybrid vehicle platoon, it addresses the problem that mechanistic models may not accurately explain vehicle following behavior in real-world scenarios. Considering the uncertainties in the system, a Tube MPC controller incorporating feedforward and feedback loops is designed based on this linear model. This controller plans the optimal trajectory for the nominal system and restricts the deviation between the nominal and actual systems within a robust invariant set. This control method can explicitly handle uncertainties with relatively low communication and computational burden, providing dual guarantees of feedforward and feedback control, and offering a new approach for the development of intelligent connected vehicles. Attached Figure Description

[0076] Figure 1 The diagrams show a ring road traffic scenario; the left image shows a circular lane. The right image is a simplified diagram of a network system, where CAVs are input nodes and HDVs are uncontrolled nodes.

[0077] Figure 2 The actual and nominal states of CAV2 and related constraints are shown in the diagram when k = 60;

[0078] Figure 3 The actual and nominal states of CAV2 and related constraints are shown in the diagram when k = 160;

[0079] Figure 4 The state component response curves of CAV2 during the first 400 simulation steps are shown.

[0080] Figure 5 The graph shows the vehicle spacing and speed response curves of the five vehicles under the robust model predictive control method when W = 0.05. Detailed Implementation

[0081] The following will provide a more detailed description of the data-driven robust control method for hybrid vehicle queuing proposed in this invention, with reference to the accompanying figures.

[0082] The traffic scenario in this invention is a single-lane, one-way circular road with a perimeter of 50 meters. Five vehicles are unevenly distributed on the road, labeled 1 to 5, where 1, 3, and 5 are HDVs (High-Density Vehicles), and 2 and 4 are CAVs (Concurrent Vehicles). Simulation was performed on the MATLAB platform, collecting 100 seconds of vehicle state data with a collection period of 0.1 seconds. The data includes information such as vehicle speed, spacing, and acceleration. This invention uses this data to model a mixed vehicle queue, thereby verifying the effectiveness of the proposed Tube MPC control method.

[0083] Step 1: Construct the lifting function

[0084] This invention employs a data-driven EDMD algorithm to model the system. The core of the EDMD algorithm is constructing a suitable boost function, the optimal choice of which depends on the underlying dynamic system and the sampling strategy used to acquire data. Since the HDVs in the hybrid vehicle platoon of this invention utilize an Intelligent Driver Model (IDM), the construction of the boost function draws upon the HDV's control strategy. The dynamic model of the HDV is as follows:

[0085]

[0086] In the formula, a represents the vehicle's maximum acceleration, and b represents the vehicle's comfortable deceleration. s0 represents the maximum speed the vehicle can reach, s0 represents the minimum stopping distance, and T represents the maximum stopping distance. i δ represents the minimum following distance of vehicle i, and δ represents the acceleration exponent, with the value of δ usually being 4. By adjusting the above parameters, the differences in following dynamics of different HDVs can be characterized.

[0087] The lift function is constructed by referring to the dynamic model of HDVs in a mixed vehicle platoon. Expanding equation (27) and combining like terms, we get the following equation:

[0088]

[0089] Where b1, b2, b3, b4, b5 are constant coefficients, i = 1, ..., m, (28) has 5 nonlinear forms. Since the state variable x of the mixed vehicle platoon system only includes the speed and spacing of each vehicle, it is based only on These four nonlinear terms form the lifting function. The lifting function is constructed from the system state itself and the four nonlinear terms mentioned above:

[0090]

[0091] The choice of boosting function is crucial to the accuracy of modeling. It is advantageous to build a data-driven model based on a physical model and to consider prior knowledge.

[0092] Step 2: Parameter Initialization

[0093] We use the following method for initialization. First, we collect sufficient data sequences of the mixed vehicle queue, calculate the system matrices A, B, and C of the mixed vehicle queue system, and obtain the system model of the mixed vehicle queue. The boundary of the perturbation is set as w. h =w v =0.05. See Table 1 for specific vehicle attributes.

[0094] Table 1 Vehicle Parameters

[0095]

[0096]

[0097] The sampling time of the Tube MPC controller is set to 0.1s, and the weight ratio of state variables and control variables in the multi-objective cost function (25) is set to 1:1. p The value was set to 15 to balance the feasibility and solution complexity of the optimization problem in the Tube MPC controller. A lifting function was constructed based on equation (29), and the lifted dimension N... φ The value is 20. The terminal penalty matrix P, gain matrix K, and robust invariant set are calculated according to formulas (26), (17), (19), and (23). and terminal constraint S f We used the open-source MATLAB toolboxes MPT3 and YALMIP, and the QP solver to solve the linear quadratic programming problem.

[0098] Step 3: Simulation Experiment

[0099] like Figure 1 As shown, this example uses a one-way, single-lane circular road as the traffic scenario for the simulation experiment. The length of the circular road is 50m. The control frequency for each vehicle is 10Hz, and the initial speed of all vehicles is 1m / s. The vehicles are unevenly distributed on the circular road, resulting in an unbalanced convoy. To verify the Tube MPC controller designed in this invention, the effectiveness of the tube invariant set, the feedback control, and the feedforward control are verified below.

[0100] First, we take CAV2 as an example to evaluate the effectiveness of the tube invariant set, and then verify the convergence and closed-loop robustness of the system. Figure 2 and Figure 3 The images show the actual and nominal states of CAV2 at k=60 and k=160, respectively. The green area represents the robust invariant set, the dark blue area represents the constraints on the actual state, the light blue area represents the constraints on the nominal state, and the dark gray area is the terminal invariant set CS of the actual state Δx. f The light gray area represents the nominal condition. terminal invariant set The true state is marked with a pink square, and the actual state is marked with a green square. From the two graphs, it can be seen that the true state Δx of the system lies within the range of [missing information - likely a specific area or region]. In the tube centered on, that is Furthermore, when k=160, the nominal state tends to 0, and the actual system tracking the nominal system also tends to 0, verifying the robustness of the closed-loop control system.

[0101] Figure 4The response curves of CAV2's state components over time within the first 400 simulation steps are shown. The nominal system trajectory converges to 0 at the end, and the actual system trajectory fluctuates within the range of the tube, centered on the nominal system trajectory. This indicates that the Tube MPC controller ensures that the nominal error state of CAV2 tends to 0 through feedforward control and that the actual trajectory of CAV2 can track the optimal trajectory of its corresponding nominal system through feedback control, thus achieving asymptotic stability.

[0102] The safety of mixed vehicle platoons is a control objective of the controller; ensuring that the vehicle states within the platoon meet set constraints is crucial for road safety. Figure 5 It can be seen that the TubeMPC controller can keep the vehicle states in the mixed vehicle queue within a safe range, that is, satisfy the safety constraint that the distance between vehicles is not less than 2m. Furthermore, Tube MPC can control the queue so that its states eventually converge to a robust invariant set centered on the expected value.

[0103] Experimental Results: This invention proposes a data-driven robust control method for hybrid vehicle platoons, which can suppress the negative impacts of uncertainties such as model mismatch and drive the actual vehicle state to approach the desired vehicle state. The innovative combination of the Tube MPC method and the Extended Dynamic Pattern Decomposition algorithm not only establishes a linear model that conforms to the actual traffic scenario, facilitating subsequent controller design, but also effectively suppresses uncertainties such as model mismatch. Experiments using vehicle trajectory data verify that the controller designed in this invention can achieve fast and efficient closed-loop control, improving the safety and robustness of the hybrid vehicle platoon system.

Claims

1. A data-driven hybrid vehicle platoon robust control method, characterized in that: Includes the following steps, Step 1: Construct a lifting function. Based on a data-driven modeling algorithm, the nonlinear characteristics of the hybrid vehicle platoon system are lifted to a high-dimensional linear space, resulting in a global high-dimensional linearized representation of the hybrid vehicle platoon system, i.e., a global linearized model of the hybrid vehicle platoon system. Step 2: Design the Tube MPC controller based on the global linearization model of the hybrid vehicle platoon system. First, design the feedback control part of the TubeMPC control framework. By selecting the coefficients of the weight matrix, establish and solve a discrete linear quadratic problem to solve the feedback matrix. The deviation between the actual state and the nominal state is dynamically reduced through feedback control. Step 3: Design the feedforward control part of the Tube MPC control framework, tighten the state constraints and control constraints of the hybrid vehicle platoon system. First, find the robust positive invariant set containing the hybrid vehicle platoon system, shrink the value range of the original state variables and control variables of the hybrid vehicle platoon system, and use the shrinkage result as the value range of the state variables and control variables of the nominal system. Rolling optimization is performed based on the nominal system model and state variables in a high-dimensional linear space to learn the approximate optimal predictive control sequence of the hybrid vehicle platoon system at the current moment. Step 4: Obtain the actual control rate based on the approximate optimal predictive control sequence at the current moment; apply the actual control rate at the current moment to the hybrid vehicle queuing system to update the state of the hybrid vehicle queuing system, obtain the state quantity of the hybrid vehicle queuing system at the next moment, and then perform mapping processing on the state quantity to update the state variable in the high-dimensional linear space, and repeat steps 3-4. A lifting function is constructed based on the vehicle control strategy of human-driven vehicles in a mixed vehicle platoon. A linear model is established for the mixed vehicle platoon system based on the extended dynamic mode decomposition algorithm. The dynamic model of the human-driven vehicles is as follows: ； In the formula, Indicates vehicle The derivative of the following distance, Representative vehicle The derivative of the velocity, This indicates the vehicle's maximum acceleration. Indicates the vehicle's comfortable deceleration. This indicates the maximum speed the vehicle can reach. Indicates the minimum parking distance. Indicates vehicle Minimum following distance, The acceleration index is used to characterize the differences in car-following dynamics among different HDVs by adjusting the above parameters; The lift function is constructed by referencing the dynamic model of HDVs in a hybrid vehicle platooning system; Expanding equation (1) and combining like terms, we get the following equation: ； in It is a constant coefficient. Equation (2) has five nonlinear forms, due to the state variables of the mixed vehicle queuing system. It only includes the speed and spacing of each vehicle, based solely on... , , , These four nonlinear forms are used to compose the lifting function; The lifting function is constructed from the system state itself and the four nonlinear terms mentioned above. 。 2. The data-driven hybrid vehicle platoon robust control method of claim 1, wherein: Modeling a hybrid vehicle platoon system based on the Extended Dynamic Mode Decomposition (EMD) algorithm. Data-driven modeling refers to collecting data from the hybrid vehicle platoon system to elevate its nonlinear characteristics to a high-dimensional linear space, and then approximating the global linearized model through system identification. When the dynamic characteristics are unknown, the EMD algorithm is used to approximate the system model to a finite extent, and the global linearized model of the control system is calculated from the data sample set of the hybrid vehicle platoon system, thus establishing the global linearized model. Obtaining the linear model of the system using the Extended Dynamic Pattern Decomposition algorithm involves three steps: First, constructing a lifting function to elevate the measurement states of each vehicle in the mixed vehicle queue to a high-dimensional space, without elevating the state input; second, using classical linear identification to obtain the approximate matrix of the mixed vehicle queue system. , The third step is to use least squares regression to solve for the mapping matrix. This involves mapping the system back from the high-dimensional space to the original state space, thereby obtaining the state evolution of the original state space; based on this, the following linear model is established for the mixed vehicle queue: ； wherein ; Furthermore, the linear incremental model of the hybrid vehicle queuing system with model error is derived as follows: ； wherein ; , , compact and containing the origin; Let and be the state and control input of the nominal model of the mixed vehicle platoon, ignoring the bounded uncertainty disturbance, the nominal system of the mixed vehicle platoon is obtained as 。 3. The data-driven hybrid vehicle platoon robust control method of claim 2, wherein: When designing the feedback control section, the discrete linear quadratic problem of solving for the feedback matrix and establishing and solving it is as follows: ； in , These are the weight matrices for the boosting state and the control input, respectively. The actual upgrade status of the system Relative nominal lift status deviation, The robustness of the system can be improved by designing a feedback controller to dynamically adjust the deviation between the actual system and the nominal system.

4. The data-driven hybrid vehicle platoon robust control method of claim 3, wherein: The range of values ​​for the original state variables and control variables in the hybrid vehicle platooning system is narrowed, and the narrowed range is used as the nominal range of values ​​for the system's state variables and control variables. This ensures that the hybrid vehicle platooning system satisfies the original range of values ​​for the state variables and control variables even with approximation residuals, and maintains the robustness of closed-loop control. To narrow the range of values ​​for the original state variables and control variables, it is necessary to first find a range that includes all... Robust positive invariant set of mixed vehicle platoons; Satisfy all and They all ,in The smallest robust positive invariant set is: ； An inner approximation method is used to estimate the convex set of the smallest robust invariant set, Robust invariant set of inner approximation : ； wherein ; if is a nilpotent function of , when there is ; then the nominal boost state and the nominal control satisfy: ； Satisfaction of the nominal trajectory with the tightening constraints guarantees satisfaction of the actual vehicle state constraints and control constraints and thus the safety of the hybrid vehicle platoon system.

5. The data-driven hybrid vehicle platoon robust control method of claim 4, wherein: The long-term rolling optimization model based on feedforward model predictive control is as follows: (1) Objective function; The control time interval of the long-scale rolling optimization model is 0.1s, and the optimization objective is to minimize the state increment of the nominal system of the hybrid vehicle queue in each prediction time domain under tightened constraints. and control increment That is, the objective function is expressed as: ； In the above optimization of the quadratic cost function , It is a decision variable. and These are the weight matrices for boosting the state and controlling the inputs, respectively. It is relative to the nominal improvement state. terminal cost, Symmetric positive definite matrix It is a solution to the Lyapunov equation: ； wherein ; (2) Constraints The constraints of the long-term rolling optimization model are vehicle state constraints, control quantity constraints, and terminal state constraints, as detailed below: ； The designed Tube MPC controller adds terminal constraint condition, the system state enters the terminal constraint set in finite steps under the action of state feedback control law to reach the stable point of the system, ensuring the stability of the closed-loop system.

Images