Safe Prediction Tracking Control Method for Unmanned Vehicles in Complex Network Environments

Abstract

The present invention discloses a method for safe predictive tracking control of an autonomous vehicle in a complex network environment. The method includes: S1 sending a desired trajectory to the autonomous vehicle; S2 determining whether the autonomous vehicle is running for the first time. If so, proceed to S3; otherwise, proceed to S6; S3 calculating the optimal control quantity of the autonomous vehicle and sending it to the state estimator to estimate the estimated state at the next moment; S4 transmitting the optimal control quantity to the autonomous vehicle for execution through a channel with potential network attacks; S5 determining whether the current moment is the last moment of the tracking control period. If so, end the control method; otherwise, proceed to S6; S6 collecting the untriggered output and calculating its function value using the event-triggered function expression; S7 determining whether the function value is greater than zero. If so, proceed to S8; otherwise, keep the triggered output unchanged and return to S3; S8 updating the triggered output to the untriggered output, calculating the attacked signal received by the model predictive controller, and then returning to S3.

Description

Technical Field

[0001] The present invention relates to tracking control technology, and in particular to a safe prediction tracking control method for an unmanned vehicle in a complex network environment. Background Art

[0002] With the rapid development of information network transmission technology and driverless technology, unmanned vehicles are expected to greatly improve traffic safety and relieve traffic congestion, and thus have received extensive attention in the industrial field and academic field. Trajectory tracking control under a networked system is one of the key technologies in the field of unmanned vehicle control. The goal of tracking control is to achieve precise tracking of a desired trajectory by eliminating the deviation between the actual trajectory and the reference trajectory during vehicle driving and maintain good stability. The intervention of a communication network brings great convenience to the unmanned vehicle system, and various components of the system can transmit information through the network, reducing the information transmission cost and reconstruction difficulty of the system. However, the dynamic nonlinearity and uncertainty of the unmanned vehicle model in a network environment, the complex and changeable driving road conditions, and the security of data transmission in the network system make the trajectory tracking control of unmanned vehicles challenging.

[0003] Lateral control directly affects the handling stability and ride comfort of an unmanned vehicle. Compared with other traditional lateral control methods, such as proportional-integral-derivative (PID) control, pure pursuit, linear quadratic regulator (LQR), and H ∞ robust control, model predictive control (MPC) has the advantage of being able to control multi-input multi-output systems with constraints, and thus is widely used in the lateral control of unmanned vehicles. A unique feature of a general unmanned vehicle tracking MPC method is that it can ensure the recursive feasibility of the optimization problem. Therefore, adding any type of complexity may lead to a loss of recursive feasibility.

[0004] General unmanned vehicle tracking MPC methods all assume that the longitudinal speed of the unmanned vehicle is a fixed value, and in robust MPC technology, the impact of various network attacks on the data transmission channel on the trajectory tracking accuracy is rarely studied, which cannot guarantee the safety of actual unmanned vehicle control. The linear parameter varying (LPV) model has a linear model structure and good ability to describe complex nonlinear systems, providing a good method for dealing with the nonlinear problems of unmanned vehicle models. Using polytope sets to describe the varying speed in the lateral dynamics model of the vehicle has been widely applied. With the development of information technology and computer technology, the means of attackers are becoming increasingly diverse and complex, making it possible to be simultaneously subjected to different types of network attacks in actual application scenarios. Simple problem modeling may lead to large deviations in the trajectory tracking of unmanned vehicles. In addition, the actual performance of unmanned vehicles is usually subject to various constraints, such as input-output constraints and communication constraints of the system. If these factors are not fully considered or cannot be met during actual operation, the actual movement of the unmanned vehicle will not conform to the expected situation, and the state and control of the system being at the constraint boundary for a long time will also damage the structure of the unmanned vehicle itself, seriously affecting the safety of the unmanned vehicle. Summary of the Invention

[0005] Aiming at the above deficiencies in the prior art, the unmanned vehicle safety predictive tracking control method provided by the present invention solves the problems that the safety and stability of the unmanned vehicle system in a complex network environment are easily affected and the trajectory tracking does not conform to the expectation.

[0006] To achieve the above invention purpose, the technical solution adopted by the present invention is as follows:

[0007] Provide an unmanned vehicle safety predictive tracking control method in a complex network environment, which includes

[0008] S1. Obtain the desired trajectory of the unmanned vehicle and send it to the unmanned vehicle;

[0009] S2. Determine whether the unmanned vehicle is running for the first time. If so, go to step S3; otherwise, go to step S6;

[0010] S3. According to the attacked signal and the estimated state at the current moment, use a model predictive controller to calculate the optimal control quantity of the unmanned vehicle and send it to the state estimator to estimate the estimated state at the next moment;

[0011] S4. Pass the optimal control quantity through a channel with potential network attacks, transform it into an attacked signal and then transmit it to the unmanned vehicle for execution; according to the attacked signal, calculate the state quantity of the unmanned vehicle deviating from the desired trajectory;

[0012] S5. Determine whether the current moment is the last moment of the tracking control period. If so, end the control method; otherwise, enter step S6 for the control of the next moment.

[0013] S6. Collect the untriggered output through the sensor, and calculate its function value using the event trigger function according to the untriggered output.

[0014] S7. Determine whether the function value is greater than zero. If so, enter step S8; otherwise, keep the triggered output unchanged and return to step S3.

[0015] S8. Update the triggered output to the untriggered output, calculate the attacked signal received by the model predictive controller according to the triggered output, and then return to step S3.

[0016] Furthermore, the method for calculating the optimal control quantity of the unmanned vehicle using the model predictive controller includes:

[0017] S31. Use the LMI technique to solve the optimization problem at the k-th moment online to obtain the optimal The initial value of k is zero, and the optimization problem is:

[0018]

[0019]

[0020] where ρ is a non-negative variable; α(k) is a positive scalar at the k-th moment; P1(k) and P2(k) are both positive definite symmetric matrices at the k-th moment; and are both matrices at the k-th moment; is the estimated state at the k-th moment; is a matrix related to the estimated state error e(k), and needs to satisfy e(k) T is the transpose of e(k); E is the identity matrix; ★ represents the symmetric part; is the control constraint; P(k) = diag{P1(k), P2(k)} is the concatenated matrix, and diag represents the diagonal matrix; y max is the output constraint; λ is a constant; and are the probabilities of θ = 1 and β = 1 respectively; θ and β are the flag signals of the FDI attack and the DOS attack respectively; and are both matrices of a preset dimension; Ξ2, and are both intermediate variable matrices of the linear matrix inequality; A 02 、A 03 、B 02 、B03 , C 03 , D 02 and D 03 are all coefficient matrices;

[0021] Ξ2, and The expressions of are as follows:

[0022]

[0023] In the formula, A c (k), κ(k) are both gain matrices to be optimized at the k-th moment; B c and H are both constant matrices; and B are both coefficient matrices; A m is the known matrix in each direction of the convex polyhedron, and m is a variable; is the estimated state at the (k + 1)-th moment; is the matrix related to the estimated state error e(k + 1);

[0024] A 02 , A 03 , B 02 , B 03 , C 03 , D 02 and D 03 The expressions of are as follows:

[0025]

[0026] In the formula, D m is the known matrix in each direction of the convex polyhedron;

[0027] S32. According to the optimal and P1 * (k), update the gain matrices A c (k), κ(k):

[0028]

[0029] where * represents the optimal solution;

[0030] S33. Calculate the optimal control quantity according to the gain matrix and the estimated state:

[0031]

[0032] where, u p (k) is the optimal control quantity at the k-th moment.

[0033] Furthermore, the expression for calculating the state quantity of the unmanned vehicle deviating from the desired trajectory is:

[0034]

[0035] Among them, is the attacked signal received by the model predictive controller at the current triggering moment k t ; k t and k t+1 are the current triggering moment and the next triggering moment respectively.

[0036] Furthermore, the LPV system is used to calculate the state quantity of the unmanned vehicle deviating from the expected trajectory and the untriggered output. The expression of the LPV system is:

[0037]

[0038] Among them, x(k) and x(k + 1) are the state quantities at the k-th moment and the (k + 1)-th moment respectively; A δ (k) and D δ (k) are both time-varying coefficient matrices; d(k) is the continuous external disturbance; y(k) is the untriggered output from the sensor to the controller at the k-th moment; u(k) is the attacked signal transmitted to the unmanned vehicle for execution at the k-th moment; the expression of u(k) is:

[0039] u(k) = (1 - β(k))(u p (k) + θ(k)φ C2A (k)) + β(k)u(k - 1)

[0040] Among them, θ(k) and β(k) are the flag signals of the FDI attack and the DOS attack at the k-th moment respectively; φ C2A (k) is the tampered information injected into the controller-actuator channel at the k-th moment; u(k - 1) is the attacked signal transmitted to the unmanned vehicle for execution at the (k - 1)-th moment;

[0041] The expression for calculating the attacked signal received by the model predictive controller is:

[0042]

[0043] Among them, and are the attacked signals received by the model predictive controller at the current triggering moment k t and the next triggering moment k t-1 respectively; y(k t ) is the triggered output from the sensor to the controller at the current triggering moment k t ; φ S2C (k) is the tampered information injected into the sensor-controller channel at the k-th moment.

[0044] Furthermore, the tampered information φ C2A(k) and φ S2C The expressions of (k) are respectively:

[0045] φ C2A (k) = -u p (k) + μ1(k)

[0046] φ S2C (k) = -y(k t ) + μ2(k)

[0047]

[0048] where μ1(k) and μ2(k) are respectively the attacks suffered by the sensor - controller channel and the controller - actuator channel at the k - th moment; is the upper bound of the energy of the attack suffered by the sensor - controller channel; is the upper bound of the energy of the attack suffered by the controller - actuator channel; ‖·‖ 2 is the square L2 norm.

[0049] Furthermore, the time - varying coefficient matrices A δ (k) and D δ (k) have the following expressions respectively:

[0050]

[0051] where A m , D m are all known matrices in each direction of the convex polyhedron; δ m is the weight coefficient;

[0052] The expressions of matrices A1, A2, A3, A4, D1, D2, D3, D4 and B are respectively:

[0053]

[0054] where T is the sampling period; C αf and C αr are respectively the cornering stiffness of the front wheels and rear wheels of the unmanned vehicle; M is the mass of the unmanned vehicle; l f and l r are respectively the length of the front axle and the length of the rear axle of the unmanned vehicle; I z is the moment of inertia of the unmanned vehicle about the z - axis; v x and are respectively the lower bound and the upper bound of the longitudinal speed v x of the unmanned vehicle.

[0055] Furthermore, the expression of the weight coefficient δ m is:

[0056]

[0057] Among them, δ1, δ2, δ3, and δ4 are the weight coefficients when m takes 1, 2, 3, and 4 respectively.

[0058] Furthermore, the input-output constraints satisfied by the LPV system are:

[0059]

[0060] Among them, (i + k|k) is the predicted value at the k-th moment for the i-th moment in the future; is the mathematical expectation; u(i + k|k) is the prediction of the control quantity at the k-th moment for the (k + i)-th moment in the future; ‖·‖ 2 is the square L2 norm; y(i + k|k) is the prediction of the output at the k-th moment for the (k + i)-th moment in the future; y(i + k|k) T is the transpose of y(i + k|k).

[0061] Furthermore, the expression of the event-triggering function is:

[0062]

[0063] Among them, g(y, ζ) is the event-triggering function; ζ is the proportionality coefficient; Λ is a diagonal matrix; represents the transmission error caused by the event-triggering condition, y(k t ) and y(k t-1 ) are the triggered outputs from the event-triggering mechanism to the controller at the current triggering moment k t and the previous triggering moment k t-1 respectively, and y(k) is the untriggered output from the sensor to the event-triggering mechanism at the k-th moment; is 's transpose; y T (k t-1 ) is the transpose of y(k t-1 ); T is the sampling period; inf represents the infimum; is the set of natural numbers; is the ideal next triggering moment obtained according to the event-triggering mechanism; y is the lateral position of the unmanned vehicle.

[0064] Furthermore, the expression of the lateral dynamics model of the unmanned vehicle in the ground coordinate system:

[0065]

[0066] Among them, M is the mass of the unmanned vehicle; and are the first derivative and the second derivative of the yaw angle ψ of the unmanned vehicle respectively; is the second derivative of the lateral position of the driverless vehicle; v x and v y are the longitudinal speed and lateral speed of the driverless vehicle respectively; C αf and C αr are the cornering stiffness of the front wheel side and rear wheel side of the driverless vehicle respectively; l f and l r are the length of the front axle and the length of the rear axle of the driverless vehicle respectively; I z is the moment of inertia of the driverless vehicle about the z-axis; σ is the front wheel steering angle.

[0067] Advantages of the present invention: The driverless vehicle safety prediction and tracking control method provided by this solution realizes accurate tracking control of the desired trajectory by the driverless vehicle in a complex network environment, and improves the tracking accuracy and the smoothness of the control quantity compared with the traditional model prediction method. By using linear matrix inequalities to solve the optimization problem and comprehensively considering multiple factors such as longitudinal speed fluctuations, input-output constraints, potential cyber attacks, communication constraints, and the unmeasurability of the state of the driverless vehicle LPV system, it not only improves the trajectory tracking performance and cyber attack resistance ability of the driverless vehicle in a complex environment, but also improves the ride comfort and stability of the driverless vehicle. Description of the Drawings

[0068] Figure 1 is a flowchart of an embodiment of the driverless vehicle safety prediction and tracking control method in a complex network environment.

[0069] Figure 2 is the tracking effect diagram of the algorithm driverless vehicle for the target trajectory (circular trajectory).

[0070] Figure 3 is the estimated state error curve graph.

[0071] Figure 4 is the driverless vehicle tracking error curve graph. [[ID=3D]]

[0072] Figure 5 is the curve graph of the controller output control quantity and the actual control quantity of the driverless vehicle.

[0073] Figure 6 is the schematic diagram of the event trigger moment.

[0074] Figure 7 is the comparison graph of the moments when the data transmission channel of the driverless vehicle is under DOS attack and FDI attack. (a) is the schematic diagram of the DOS attack moment, and (b) is the schematic diagram of the FDI attack moment.

[0075] Figure 8 is the 2 trajectory curve graph. Detailed Implementation Manner

[0076] The specific embodiments of the present invention will be described below to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0077] In this solution, the controller mentioned in some places is an abbreviation for the model predictive controller.

[0078] Refer to Figure 1 , Figure 1 which shows a flowchart of an embodiment of the safe prediction and tracking control method for an unmanned vehicle in a complex network environment; as Figure 1 shown, this method S includes steps S1 to S8.

[0079] In step S1, the desired trajectory of the unmanned vehicle is obtained and sent to the unmanned vehicle; in implementation, the expression of the lateral dynamics model of the unmanned vehicle in the ground coordinate system is preferably:

[0080]

[0081] where M is the mass of the unmanned vehicle; and are the first derivative and the second derivative of the yaw angle ψ of the unmanned vehicle respectively; is the second derivative of the lateral position of the unmanned vehicle; v x and v y are the longitudinal speed and the lateral speed of the unmanned vehicle respectively; C αf and C αr are the cornering stiffness of the front wheel side and the rear wheel side of the unmanned vehicle respectively; l f and l r are the length of the front axle and the length of the rear axle of the unmanned vehicle respectively; I z is the moment of inertia of the unmanned vehicle about the z-axis; σ is the front wheel steering angle.

[0082] In step S2, it is judged whether the unmanned vehicle is running for the first time. If so, go to step S3; otherwise, go to step S6;

[0083] In step S3, according to the attack signal and the estimated state at the current moment, the model predictive controller is used to calculate the optimal control quantity of the unmanned vehicle and send it to the state estimator to estimate the estimated state at the next moment.

[0084] In an embodiment of the present invention, the method for calculating the optimal control quantity of the unmanned vehicle by using the model predictive controller includes:

[0085] S31. Use the LMI technology to solve the optimization problem at the kth moment online to obtain the optimal The initial value of k is zero, and the optimization problem is as follows:

[0086]

[0087]

[0088] where ρ is a non - negative variable; α(k) is a positive scalar at the k - th moment; both P1(k) and P2(k) are positive definite symmetric matrices at the k - th moment; and are both matrices at the k - th moment; is the estimated state at the k - th moment; is a matrix related to the estimated state error e(k), and needs to satisfy e(k) T is the transpose of e(k); E is the identity matrix; ★ represents the symmetric part; is the control constraint; P(k)=diag{P1(k),P2(k)} is the concatenation matrix, and diag represents the diagonal matrix; y max is the output constraint; λ is a constant; and are the probabilities of θ = 1 and β = 1 respectively; θ and β are the flag signals of FDI attack and DOS attack respectively; and are both matrices of a preset dimension; Ξ2, and are all intermediate variable matrices of linear matrix inequalities; A 02 、A 03 、B 02 、B 03 、C 03 、D 02 and D 03 are all coefficient matrices;

[0089] Ξ2, and The expressions of are as follows:

[0090]

[0091] where A c (k) and κ(k) are both gain matrices to be optimized at the k - th moment; B c and H are both constant matrices; and B are both coefficient matrices; HCP1(k) and HCP2(k) are H*C*P1(k) and H*C*P2(k) respectively; A m is the known matrix in each direction of the convex polyhedron, and m is a variable; is the estimated state at the (k + 1)-th moment; is a matrix related to the estimated state error e(k + 1);

[0092] A 02 and A 03 and B 02 and B 03 and C 03 and D 02 and D 03 The expressions of are as follows:

[0093]

[0094] In the formula, D m is the known matrix in each direction of the convex polyhedron;

[0095] S32. According to the optimal and update the gain matrices A c (k) and κ(k):

[0096]

[0097] where * represents the optimal solution;

[0098] S33. Calculate the optimal control quantity according to the gain matrix and the estimated state:

[0099]

[0100] where u p (k) is the optimal control quantity at the k-th moment.

[0101] In the above optimization problem of this scheme, the first and second inequalities ensure that the current augmented state belongs to the invariant set, that is, E{ξ(k)} ∈ Ω k holds, and Ω k is the robust positive invariant set at the k-th moment. The third and fourth inequalities ensure the constraints on the input and output of the unmanned vehicle LPV system. The last inequality ensures the quadratic bounded property of the unmanned vehicle LPV system under the event-triggered mechanism, and it can be known that the state quantity and the estimated state error will finally converge in the neighborhood of zero, ensuring the stability of the unmanned vehicle LPV system.

[0102] In step S4, the optimal control quantity is transmitted to the unmanned vehicle for execution after being transformed into an attacked signal through a channel with potential network attacks; according to the attacked signal, calculate the state quantity of the unmanned vehicle deviating from the desired trajectory:

[0103]

[0104] where is the current trigger moment k tThe attacked signal received by the model predictive controller; k t and k t+1 are the current triggering moment and the next triggering moment respectively.

[0105] In step S5, it is judged whether the current moment is the last moment of the tracking control period. If so, the control method ends; otherwise, it enters step S6 for the control of the next moment.

[0106] In step S6, the untriggered output is collected by the sensor, and according to the untriggered output, the function value is calculated by using the event-triggering function.

[0107] In implementation, the preferred expression of the event-triggering function in this solution is:

[0108]

[0109] where g(y, ζ) is the event-triggering function; ζ is the proportionality coefficient; Λ is the diagonal matrix; represents the transmission error caused by the event-triggering condition, y(k t ) and y(k t-1 ) are the triggered outputs from the event-triggering mechanism to the controller at the current triggering moment k t and the previous triggering moment k t-1 respectively, and y(k) is the untriggered output from the sensor to the event-triggering mechanism at the k-th moment; is the transpose of; y T (k t-1 ) is the transpose of y(k t-1 ); T is the sampling period; inf represents the infimum; is the set of natural numbers; is the ideal next triggering moment obtained according to the event-triggering mechanism; y is the lateral position of the unmanned vehicle.

[0110] In implementation, this solution preferably uses the LPV system to calculate the state quantity and the untriggered output of the unmanned vehicle deviating from the desired trajectory. The expression of the LPV system is:

[0111]

[0112] where x(k) and x(k + 1) are the state quantities at the k-th moment and the (k + 1)-th moment respectively; A δ (k) and D δ (k) are both time-varying coefficient matrices; d(k) is the continuous external disturbance; y(k) is the untriggered output from the sensor to the controller at the k-th moment; u(k) is the attacked signal transmitted to the execution of the unmanned vehicle at the k-th moment; the expression of u(k) is:

[0113] u(k) = (1 - β(k))(u p (k) + θ(k)φ C2A (k)) + β(k)u(k - 1)

[0114] where θ(k) and β(k) are the flag signals of FDI attack and DOS attack at the k-th moment respectively; φ C2A (k) is the tampered information injected into the controller-actuator channel at the k-th moment; u(k - 1) is the attacked signal transmitted to the unmanned vehicle for execution at the (k - 1)-th moment;

[0115] The expression for calculating the attacked signal received by the model predictive controller is:

[0116]

[0117] where and are the attacked signals received by the model predictive controller at the current triggering moment k t and the next triggering moment k t-1 respectively; y(k t ) is the triggered output from the sensor to the controller at the current triggering moment k t ; φ S2C (k) is the tampered information injected into the sensor-controller channel at the k-th moment.

[0118] The u(k) and the dual-channel hybrid attack model when the sensor-controller channel (S2C) and the controller-actuator channel (C2A) are simultaneously subjected to independent FDI attacks and DOS attacks initiated by an attacker.

[0119] Currently, most networked control methods only consider the system being subjected to a single network attack or a single-channel network attack. Such control methods can make the designed LPV system have a certain degree of robustness, but the consideration is not comprehensive enough. Based on this, this scheme considers the scenario of a hybrid network attack where both channels of the LPV system are subjected to denial-of-service attacks (DOS) and false information attacks (FDI) initiated by an attacker, so as to improve the accuracy of the vehicle control quantity during vehicle tracking control, that is, it basically matches the desired trajectory.

[0120] The above-mentioned tampered information φ C2A (k) and φ S2C (k) are expressed as:

[0121] φ C2A (k) = -u p (k) + μ1(k)

[0122] φ S2C(k) = -y(k t ) + μ2(k)

[0123]

[0124] where μ1(k) and μ2(k) are the attacks suffered by the sensor - controller channel and the controller - actuator channel at the k - th moment, respectively; is the upper bound of the energy of the attack suffered by the sensor - controller channel; is the upper bound of the energy of the attack suffered by the controller - actuator channel; ‖·‖ 2 is the square L2 norm.

[0125] In step S7, it is judged whether the function value is greater than zero. If so, go to step S8; otherwise, keep the triggered output unchanged and return to step S3;

[0126] In step S8, update the triggered output to the un - triggered output, and calculate the attacked signal received by the model predictive controller according to the triggered output, and then return to step S3.

[0127] In an embodiment of the present invention, the time - varying coefficient matrices A δ (k) and D δ (k) in the expression of the LPV system are respectively:

[0128]

[0129] where A m , D m are all known matrices in each direction of the convex polyhedron; δ m is the weight coefficient;

[0130] The expressions of matrices A1, A2, A3, A4, D1, D2, D3, D4 and B are respectively:

[0131]

[0132]

[0133] where T is the sampling period; C αf and C αr are the cornering stiffnesses of the front and rear wheels of the unmanned vehicle respectively; M is the mass of the unmanned vehicle; l f and l r are the lengths of the front and rear axles of the unmanned vehicle respectively; I z is the moment of inertia of the unmanned vehicle about the z - axis; v x and are the lower and upper bounds of the longitudinal speed v x of the unmanned vehicle respectively.

[0134] When implemented, the weight coefficient δ of this solution is preferably m expressed as:

[0135]

[0136] where δ1, δ2, δ3, and δ4 are the weight coefficients when m takes 1, 2, 3, and 4 respectively.

[0137] The input-output constraints satisfied by the LPV system are:

[0138]

[0139] where (i + k|k) is the predicted value at the k-th moment for the next i moments; is the mathematical expectation; u(i + k|k) is the prediction of the control quantity at the k-th moment for the future (k + i)-th moment; ‖·‖ 2 is the square L2 norm; y(i + k|k) is the prediction of the output at the k-th moment for the future (k + i)-th moment; y(i + k|k) T is the transpose of y(i + k|k).

[0140] In this solution, the system corresponding to the method for safe prediction and tracking control of an unmanned vehicle in a complex network environment can refer to the framework Figure 2 .

[0141] Next, combined with MATLAB R2022a software, the effect of this solution is verified by simulation:

[0142] This solution obtains the control quantity by using steps S1 to S8, and then the network transmits the signal to the specific actuator of the unmanned vehicle. During the network transmission process, it is transformed into an attacked signal through a channel containing potential network attacks and then transmitted to the unmanned vehicle for execution. The event-triggered dynamic output feedback robust model predictive control method mentioned in this solution is verified through indicators such as the tracking error curve and the tracking effect of the unmanned vehicle on the target trajectory (desired trajectory).

[0143] Considering a circular trajectory as the target trajectory to be tracked, the unmanned vehicle starts from an arbitrary point not on the circle. To better prove the advantages of this control solution in terms of tracking accuracy and driving stability, the control method proposed in this solution is compared and simulated with the event-triggered model predictive control (ETMPC) under the same conditions; the specific implementation process is as follows:

[0144] Table 1 shows the selection of each parameter of the unmanned vehicle.

[0145] Table 1

[0146]

[0147] First, without loss of generality, let matrix B c = [0, 0.78004, 0, 0.72856] T , ζ = 0.05, λ = 0.005, matrix H = [0.0045, 0.012; 0.0046, 0.0130; 0047, 0.0140; 0048, 0.015]. At k = 0, x(0) = [4.8524, -3.0889, -0.73929, -0.44386] T , and the sampling time T = 0.01 s. The simulation step is recorded up to 500. For ETMPC, the selected prediction horizon is 50 steps.

[0148] Figure 2 shows the tracking effects of the control method of this scheme and the ETMPC method on the circular trajectory under the same conditions. Without loss of generality, a point not on the circle is selected as the starting point. It can be seen that the control method of this scheme tracks the target trajectory faster and without large fluctuations, indicating that the control method proposed in this scheme can enable the unmanned vehicle to quickly and accurately track the target trajectory in a complex environment and has strong robustness.

[0149] Figure 3 plots the trajectory curve of ||e i (k)|| 2 , i ∈ {1, 2, 3, 4}. It can be seen that compared with ETMPC, the control method of this scheme can estimate the LPV system faster and more smoothly by the designed observer, that is, the difference between the true state x(k) of the unmanned vehicle LPV system and the estimator state is very small, which means that the feedback tracking controller designed in this scheme works better than ETMPC in the presence of bounded disturbances and hybrid cyber attacks. At the same time, it can be seen that the LPV system state converges to the neighborhood of the equilibrium point, verifying the reliability of the designed controller.

[0150] Figure 4 shows the tracking error curve of the unmanned vehicle. From Figure 5 it can be seen the change of the true state x(k) of the unmanned vehicle LPV system. The state starts from x(0) and gradually converges to the neighborhood of the equilibrium point under the control action, so as to track the target trajectory. Due to the presence of cyber attacks, the LPV system state is not particularly smooth, but all converge to the neighborhood of the desired equilibrium point relatively quickly.

[0151] Figure 5 shows the curves of the controller output control quantity and the true control quantity of the unmanned vehicle. From Figure 6 it can be seen that the control input u solved by the designed feedback tracking controllerp (k) is very gentle and has low energy consumption. Due to the existence of hybrid network attacks, there are some distortions in the control input received by the actuator, and even the positive and negative signs change in extreme cases (for the autonomous vehicle, it means a rapid change in the left and right directions during driving), but the control of this scheme enables the system to still maintain high robustness, which means that the control method of this scheme can work well in a complex network environment.

[0152] Figure 6 Shows the moment when the output signal is transmitted to meet the event-triggering mechanism, demonstrating the aperiodic sampling mechanism. Only when the transmission error exceeds the dynamic threshold at each moment, the output signal is sent. Accordingly, compared with the continuous communication method, less data is transmitted, significantly reducing the communication burden of the autonomous vehicle.

[0153] Figure 7 Shows the moments when the data transmission channels of the autonomous vehicle are under DOS attacks and FDI attacks. Figure 7 In (a) and (b) of it, the moments when the attacker launches a denial-of-service / spoofing attack on the communication network are shown. The hybrid attack has a negative impact on both channels, reducing the performance of sensors and actuators. Figure 8 Plots the ||x(k)|| 2 trajectory curve, further verifying the tracking effect of the control method proposed in this scheme. The simulation results show the effectiveness of the design method.

[0154] In summary, the tracking control scheme of this scheme realizes the good tracking of the autonomous vehicle for the reference trajectory, thus improving the tracking accuracy of the autonomous vehicle's autonomous decision-making for the reference trajectory; at the same time, it also improves the safety of the autonomous vehicle's operation.

Claims

1. A method for safe predictive tracking control of an unmanned vehicle in a complex network environment, characterized in that, Including the steps: S1. Obtain the desired trajectory of the driverless vehicle and send it to the driverless vehicle; S2. Determine whether the driverless vehicle is running for the first time. If so, proceed to step S3; otherwise, proceed to step S6; S3. According to the attacked signal and the estimated state at the current moment, use a model predictive controller to calculate the optimal control quantity of the driverless vehicle and send it to the state estimator to estimate the estimated state at the next moment; S4. Transform the optimal control quantity through a channel with potential network attacks into an attacked signal and then transmit it to the driverless vehicle for execution; calculate the state quantity of the driverless vehicle deviating from the desired trajectory according to the attacked signal; S5. Determine whether the current moment is the last moment of the tracking control period. If so, end the control method; otherwise, proceed to step S6 for the control of the next moment; S6. Collect the untriggered output through a sensor and calculate its function value using an event trigger function according to the untriggered output; S7. Determine whether the function value is greater than zero. If so, proceed to step S8; otherwise, keep the triggered output unchanged and return to step S3; S8. Update the triggered output to the untriggered output, calculate the attacked signal received by the model predictive controller according to the triggered output, and then return to step S3.

2. The method for safe prediction and tracking control of an unmanned vehicle in a complex network environment according to claim 1, characterized in that The method for calculating the optimal control quantity of the driverless vehicle using a model predictive controller includes: S31. Use the LMI technique to solve the optimization problem at the k-th moment online to obtain the optimal The initial value of k is zero, and the optimization problem is: min ρ s.t. where ρ is a non - negative variable; α(k) is a positive scalar at the k - th moment; both P1(k) and P2(k) are positive definite symmetric matrices at the k - th moment; and are both matrices at the k - th moment; is the estimated state at the k - th moment; is a matrix related to the estimated state error e(k), and needs to satisfy e(k) T is the transpose of e(k); E is the identity matrix; ★ represents the symmetric part; is the control constraint; P(k)=diag{P1(k),P2(k)} is the concatenation matrix, where diag represents the diagonal matrix; y max is the output constraint; λ is a constant; and are the probabilities of θ = 1 and β = 1 respectively; θ and β are the flag signals of FDI attack and DOS attack respectively; and are both matrices of a preset dimension; Ξ2, and are all intermediate variable matrices of the linear matrix inequality; A 02 、A 03 、B 02 、B 03 、C 03 、D 02 and D 03 are all coefficient matrices; Ξ2、 and The expressions of are as follows: where A c (k) and κ(k) are both gain matrices to be optimized at the k-th moment; B c and H are both constant matrices; and B are both coefficient matrices; A m is the known matrix in each direction of the convex polyhedron, and m is a variable; is the estimated state at the (k + 1)-th moment; is a matrix related to the estimated state error e(k + 1); A 02 、A 03 、B 02 、B 03 、C 03 、D 02 and D 03 The expressions are: where D m is the known matrix in each direction of the convex polyhedron; S32. According to the optimal and P1 * (k), update the gain matrix A c (k) and κ(k): where * represents the optimal solution; S33. Calculate the optimal control quantity according to the gain matrix and the estimated state: Among them, u p (k) is the optimal control quantity at the k-th moment.

3. The method for safe prediction and tracking control of an unmanned vehicle in a complex network environment according to claim 2, wherein, The expression for calculating the state quantity of the driverless vehicle deviating from the desired trajectory is: Among them, is the attacked signal received by the model predictive controller at the current triggering moment k t ; k t and k t+1 are the current triggering moment and the next triggering moment respectively.

4. The method for predicting, tracking and controlling the safety of an unmanned vehicle in a complex network environment according to claim 2, characterized in that, Use an LPV system to calculate the state quantity of the driverless vehicle deviating from the desired trajectory and the untriggered output. The expression of the LPV system is: where x(k) and x(k + 1) are the state variables at the k-th and (k + 1)-th instants respectively; A δ (k) and D δ (k) are both time-varying coefficient matrices; d(k) is a continuous external disturbance; y(k) is the untriggered output from the sensor to the controller at the k-th instant; u(k) is the attacked signal transmitted to the unmanned vehicle for execution at the k-th instant; the expression of u(k) is: u(k) = (1 - β(k))(u p (k) + θ(k)φ C2A (k)) + β(k)u(k - 1) where, θ(k) and β(k) are the flag signals of the FDI attack and DOS attack at the k-th moment, respectively; φ C2A (k) is the tampering information injected into the controller-actuator channel at the k-th moment; u(k - 1) is the attacked signal transmitted to the unmanned vehicle for execution at the (k - 1)-th moment; The expression for calculating the attacked signal received by the model predictive controller is: Among them, and are the attacked signals received by the model predictive controller at the current triggering moment k t and the next triggering moment k t-1 respectively; y(k t ) is the triggered output from the sensor to the controller at the current triggering moment k t ; φ S2C (k) is the tampering information injected into the sensor-controller channel at the k-th moment.

5. The method for safe prediction tracking control of a driverless vehicle in a complex network environment according to claim 4, wherein Tampered information φ C2A (k) and φ S2C The expressions of (k) are respectively: φ C2A (k) = -u p (k) + μ1(k) φ S2C (k) = -y(k t ) + μ2(k) where, $\mu_1(k)$ and $\mu_2(k)$ are the attacks suffered by the sensor-controller channel and the controller-actuator channel at the $k$-th moment, respectively; is the upper bound of the energy of the attack suffered by the sensor-controller channel; is the upper bound of the energy of the attack suffered by the controller-actuator channel; $\|\cdot\|$ 2 denotes the squared $L_2$ norm.

6. The method for predicting, tracking and controlling the safety of a driverless vehicle in a complex network environment according to claim 4, wherein Time-varying coefficient matrix A δ (k) and D δ (k) are expressed as follows: Among them, A m , D m are all known matrices in various directions of the convex polyhedron; δ m is the weight coefficient; The expressions of matrices A1, A2, A3, A4, D1, D2, D3, D4, and B are respectively: where T is the sampling period; C αf and C αr are the cornering stiffnesses of the front and rear wheels of the driverless vehicle respectively; M is the mass of the driverless vehicle; l f and l r are the lengths of the front and rear axles of the driverless vehicle respectively; I z is the moment of inertia of the driverless vehicle about the z-axis; v x and are the lower and upper bounds of the longitudinal speed v x of the driverless vehicle respectively.

7. The method for predicting, tracking and controlling the safety of an autonomous vehicle in a complex network environment according to claim 6, wherein Weight coefficient δ m The expression is as follows: where δ1, δ2, δ3, and δ4 are the weight coefficients when m takes 1, 2, 3, and 4 respectively.

8. The method for predicting, tracking and controlling the safety of an unmanned vehicle in a complex network environment according to claim 4, characterized in that The input-output constraints satisfied by the LPV system are: Among them, (i + k|k) is the predicted value for the next i time instants at the k-th time instant; E{u(i + k|k)} is the mathematical expectation; u(i + k|k) is the prediction of the control quantity for the (k + i)-th future time instant at the k-th time instant; ‖·‖ 2 denotes the squared L2 norm; y(i + k|k) is the prediction of the output for the (k + i)-th future time instant at the k-th time instant; y(i + k|k) T is the transpose of y(i + k|k).

9. The method for predicting, tracking and controlling the safety of an unmanned vehicle in a complex network environment according to claim 1, wherein The expression of the event trigger function is: Among them, g(y, ζ) is the event-triggering function; ζ is the proportionality coefficient; Λ is a diagonal matrix; denotes the transmission error caused by the event-triggering condition, where y(k t ) and y(k t-1 ) are the triggered outputs from the event-triggering mechanism to the controller at the current triggering time k t and the previous triggering time k t-1 respectively, and y(k) is the untriggered output from the sensor to the event-triggering mechanism at the k-th time; is transpose; y T (k t-1 ) is the transpose of y(k t-1 ); T is the sampling period; inf represents the infimum; is the set of natural numbers; is the ideal next triggering time obtained according to the event-triggering mechanism; y is the lateral position of the unmanned vehicle.

10. The method for predicting, tracking and controlling the safety of an unmanned vehicle in a complex network environment according to any one of claims 1-9, characterized in that, The expression of the lateral dynamics model of the driverless vehicle in the ground coordinate system: where M is the mass of the driverless vehicle; and are the first derivative and the second derivative of the yaw angle ψ of the driverless vehicle respectively; is the second derivative of the lateral position of the driverless vehicle; v x and v y are the longitudinal speed and the lateral speed of the driverless vehicle respectively; C αf and C αr are the cornering stiffness of the front wheel side and the rear wheel side of the driverless vehicle respectively; l f and l r are the front axle length and the rear axle length of the driverless vehicle respectively; I z is the moment of inertia of the driverless vehicle about the z-axis; σ is the front wheel steering angle.

Images