A data-driven self-triggering control method for unknown systems

Abstract

The present invention discloses a data-driven self-triggering control method for an unknown system. Before the system to be stabilized is operated online, sufficient input and output trajectories are collected offline. During the online operation, at each sampling moment, the sensor side sends a data packet containing the system output to the controller side via a network. The controller side generates a series of inputs according to the received output data packets by solving a data-based optimization problem. These inputs are sent to the system to be stabilized in turn as the control input of the system until the next output sampling moment, and then a new input sequence is recalculated. At the same time, the controller side calculates the next transmission moment according to the optimal solution of the optimization problem and sends it to the sensor side via the network. The present invention can directly use data to design a system self-triggering mechanism without pre-identification of the system when the system matrix is ​​unknown, thereby greatly reducing the resources and time consumed by system identification, and ensuring the stability of the system with only a small number of self-triggering times.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent manufacturing, and in particular relates to a data-driven self-triggering control method for an unknown system. Background Art

[0002] In order to promote the transformation and upgrading of my country's manufacturing industry to intelligent manufacturing, it is urgent to solve the problems of production line data collection, information processing, equipment control, job scheduling, production plan management and decision-making. The production line is a typical industrial Internet system. With the continuous development of communication level and network technology, this system is used by more and more modern engineering systems. In this system, information is transmitted through wired or wireless networks. The most basic problem in the study of industrial Internet systems is how to set the transmission time between different devices (such as sensors, controllers and actuators) so that the communication loss and system performance in the system can reach a balanced state.

[0003] Many industrial Internet systems use a periodic transmission mechanism to send data. This method is simple to operate, but it will cause a great degree of waste of transmission resources. For example, consider an open-loop unstable system. During the initial operation of the system, the state of the system changes greatly between sampling times, but as the system gradually reaches a stable state, the state of the system will change slowly. In other words, the sampling of adjacent moments of the system usually cannot provide much effective information. The non-periodic sampling method was born for this problem. Event triggering and self-triggering are two commonly used methods in non-periodic sampling. In addition, due to self-triggering control, the measured value of the system only needs to be sampled at each triggering moment, which enables the system's sensors and output channels to be completely closed at non-triggering moments. Therefore, this can greatly extend the service life of the sensor and reduce the resource usage of the output channel.

[0004] However, the design of self-triggering laws requires the predicted value of the system state, which requires knowing the exact model of the system in advance. However, when the dimension of the system is large, system identification often takes a lot of time and resources. The development of computer technology in recent years has made data-driven methods a reality. This method provides an end-to-end design method, that is, obtaining the control law of the system directly from the data. Different from the traditional method, this method is also called data-driven control. However, there is currently no literature that considers unknown systems and directly designs self-triggering laws for the system through pre-collected data to control the system. Therefore, it is urgent to fill the gap in this field. Summary of the invention

[0005] In view of this, the present invention provides a data-driven self-triggering control method for an unknown system, which can make use of data to directly design the system self-triggering mechanism without the need for prior system identification when the system matrix is ​​unknown, thereby greatly reducing the resources and time consumed by system identification and ensuring the stability of the system with only a small number of self-triggering times.

[0006] The technical solution for implementing the present invention is as follows:

[0007] A data-driven self-triggering control method for an unknown system is applied to a system to be stabilized. The system to be stabilized realizes closed-loop control through a data-driven controller, a self-triggering module, a sensor side and a controller side. The method is:

[0008] Before the system to be stabilized is run online, sufficient input and output trajectories are collected offline. When running online, at each sampling moment, the sensor side sends the data packet containing the system output to the controller side through the network. The controller side generates a series of inputs by solving a data-based optimization problem based on the received output data packets. These inputs are sent to the system to be stabilized in turn as the control input of the system until the next output sampling moment, when a new input sequence is recalculated. At the same time, the controller side calculates the next transmission time according to the optimal solution of the optimization problem and sends it to the sensor side through the network.

[0009] Furthermore, the specific steps of the method of the present invention are as follows:

[0010] Step 1: For an observable and stabilizable linear time-invariant system, set x t ,y t ,u t They are the state value, output value and control input value of the system at time t, and their dimensions are n x , n y , n u , the stable point of the system is (x e ,u e ,y e ), the system output is subject to bounded network noise n during transmission t , at each self-triggered transmission time t l , the sensor side of the system sends the most recent n outputs to the controller side On the other hand, the control input of the system controls the system in a periodic transmission manner;

[0011] Assume the dimension of the system is a known quantity n x , the observable index of the system is η, and set L≥n x + η; constant λ g >0, constant λ h>0; real positive definite symmetric matrix Q, R; network noise boundary

[0012] Step 2: At any initial value Next, select N control inputs Record the input data used by the system and the resulting output data

[0013] Step 3: Arrange the trajectory composed of the above N input data in the form of Hankel matrix H L+η (u s ), so that the row rank of the matrix is ​​(L+η)n u ; Similarly, construct output data The Hankel matrix H L+η (y s ) and construct the combined Hankel matrix of input-augmented state in For i∈[η, N-1], the superscript T represents the transpose of the matrix, Representing a collection The input vectors from i-n to i-1 in, Representing a collection The output vectors from i-n to i-1;

[0014] Step 4: Use the Hankel matrix of the output and input data constructed above to solve the observable parameter ρ of the system i , where i = 1, 2, ..., L-1;

[0015] Step 5: Define the augmented state of the system at time t as Similarly, we define the extended state stable point ξ of the system e , when the subscript is less than zero, that is, when t∈[0,η), it represents the input and output values ​​of the system initial setting; select the constant r>0, and solve the real positive definite matrix P r And the matrix K r , so that the augmented state Under the state feedback law u t =u e +K r (ξ t -ξ e ) in the time after the set r and converges, where n ξ = η(n u +n y ), Indicates dimension n ξ ×1 real vector set; then select a constant 0<∈≤r and construct the set Ξ according to the same rule ∈ , real positive definite matrix matrix P, and state feedback matrix K, and the constant ∈ and constant r need to satisfy the relationship in, Represents the matrix P r The maximum eigenvalue of λ (P) represents the minimum eigenvalue of the matrix P;

[0016] Step 6: Define the self-trigger time as Where N0 = {0, 1, 2, ...} is a set of natural numbers including 0, and the time interval between two adjacent self-triggering moments is defined as τ l :=t l+1 -t l , where l∈N0; if the system is in the self-triggering moment at this time, the system's output data packet will be transmitted to the controller side; the controller side solves the following optimization problem after receiving the output data packet:

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] Where h(t l ), g(t l ), is the physical quantity to be solved; g(t l ) represents the weight vector, yes R weight norm of the matrix; is a matrix The Q weight norm of ; represents the loss function; t l represents the lth self-triggering moment; represents the current time t to be predicted l The first n inputs and the last L-1 inputs; represents the current time t to be predicted lThe previous n outputs and the next L-1 outputs; u min and u max is the maximum and minimum value of the pre-given input; u [-η，-1] ,y [-η，-1] are the values ​​of n initial inputs and outputs given to the system;

[0023] represents the current time t l A column vector consisting of the actual inputs of the previous n systems;

[0024] Indicates that from time t l -η to time t l -1, the controller side of the system is disturbed by the noise nt in the network, and the current time t actually received l The previous n output values, where the variable

[0025] represents the current time t l i system inputs to be predicted, i∈[-η,L-1];

[0026] represents the current time t l The i system outputs to be predicted i∈[-η, L-1];

[0027] represents the current time t to be predicted l Input data of the previous n systems;

[0028] represents the current time t to be predicted l Output data of the previous n systems;

[0029] h i (t l ) represents the time t l The i h(t l ) value;

[0030] Indicates time t l The constructed current time t l To the augmented state of the subsequent L systems;

[0031] Indicates time t l The constructed i-th ξ(t l ) value;

[0032] Step 7: Substitute the above-solved input sequence Transmitted to the system side to control the system.

[0033] Furthermore, in step 2, each component of the control input is randomly selected from the interval [-1, 1].

[0034] Further, using the solution to the optimization problem in step 4, solve the following problem:

[0035]

[0036] in

[0037]

[0038]

[0039] Here, the parameter σ∈(0, 1) is a threshold parameter given artificially, and ||·|| represents the binary norm of the variable ·. Represents the matrix P r The maximum eigenvalue of represents the largest eigenvalue of the matrix Q, λ (R) represents the minimum eigenvalue of the matrix R, Denotes the matrix H uξ The left pseudo-inverse matrix, h * (t l ) indicates that at t l By solving the optimization problem in S4, we can get the variable h(t l ), Indicates time t l The i-th h to be predicted * (t l ), Indicates that at t l The variables are obtained by solving the optimization problem in S4 at all times The optimal solution of Indicates time t l The i-th The value in; use the calculation to get the interval τ between this transmission time and the next transmission time l , calculate the next self-triggering time t l+1 :=t l + τ l , and transmit this moment to the sensor side through the network.

[0040] Furthermore, at this moment t∈(t l , t l+1 ) is not the self-triggering moment, the sensor side does not transmit the output value to the controller side. At this time, the controller side uses t lThe control value corresponding to this moment in the control sequence solved at this moment Control the system, including Indicates that at t l The variables are obtained by solving the optimization problem in step 4 at any time. The optimal solution.

[0041] Beneficial effects:

[0042] 1. The present invention aims at unknown systems, does not require prior system identification, and directly provides a data-driven control and a data-based self-triggering mechanism by utilizing system trajectories.

[0043] 2. Different from the previous methods that require the use of system models to design controllers and self-triggering mechanisms, the method of the present invention only uses pre-collected system trajectories and can achieve stable operation of the system by designing a data-driven predictive control algorithm.

[0044] 3. The present invention designs a self-triggering mechanism by using variables obtained through a data-driven predictive control algorithm, thereby achieving stable operation of the system under the designed self-triggering mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the industrial Internet system structure of a data-driven self-triggering control method for an unknown system provided by the present invention.

[0046] Figure 2 It is an implementation flow chart of a data-driven self-triggering control method for an unknown system provided by the present invention.

[0047] Figure 3 It is an operating effect diagram of an embodiment of a data-driven self-triggering control method for an unknown system provided by the present invention; (a) is the trajectory of the system output changing over time, and (b) is a trigger time schematic diagram. DETAILED DESCRIPTION

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

[0049] The present invention provides a data driven self-triggering control method for an unknown system, such as Figure 1 As shown, by collecting the input and output trajectories of the system in advance, a data-driven predictive controller and a self-triggering mechanism are designed, so that the unknown system can remain stable under the designed self-triggering control mechanism.

[0050] The sensor to controller side of the system of the present invention adopts a self-triggering data transmission mechanism, and the output signal is transmitted only at the self-triggering moment, and the controller to system side adopts a periodic transmission mechanism.

[0051] S1: Set the dynamic equation of the system to be stabilized as:

[0052] x t+1 =Ax t +Bu t , t∈N0,

[0053] y t =Cx t +Du t

[0054] Among them, x t ,y t ,u t They are the state value, output value and control input value of the system at time t, and their dimensions are n x , n y , n u . A, B, C, D are all system matrices and are unknown, but the system is observable and stable, and the input and output data of the system are available. Set the observable index of the system to be η, which can be obtained by existing technology. The set N0 = {0, 1, 2, 3, ...} represents a set of natural numbers including 0, and the initial value of the system is an arbitrary value. The stable point of the system is (x e ,u e ,y e ). Considering that the output value of the system will be corroded by the noise caused by the network when the system is running online, that is,

[0055] ζ t =y t +n t

[0056] Among them, t is the output of the system actually collected by the controller side, n t is bounded network noise, i.e.

[0057] Define the augmented state of the above system at time t as Therefore, based on existing technology, we can transform the above system into an extended system

[0058]

[0059]

[0060] The matrix is the matrix of the appropriate augmented system, transformed from A, B, C, and D. Set N η ={η,η+1,…} represents the set of natural numbers greater than or equal to η.

[0061] At the same time, the system uses a data packet-based data transmission method from the sensor to the controller, and the sensor has a certain data storage function. Specifically, at each self-triggered transmission time t l , the sensor side of the system sends the most recent n outputs to the controller side This can be achieved in the following way: since the next self-triggering moment received is in the future, the sensor side only needs to set the sampling and storage to start at n time before the received self-triggering moment, and keep sampling until the moment before the self-triggering moment, and send the stored data packet at the self-triggering moment. On the other hand, the control input of the system controls the system in a periodic transmission mode.

[0062] Assume that the dimension of the system is known to be n x , given the prediction time domain constant L ≥ n x +η; constant λ g >0, constant λ h >0; real positive definite symmetric matrix Q, R; network noise boundary

[0063] S2: At any initial value Next, select N control inputs And each component of the control input is randomly selected from the interval [-1, 1], recording the input data used by the system and the output data generated When choosing, please pay attention to N≥(L+n x +1)n x +L+n x .

[0064] S3: Arrange the trajectory composed of the above N input data in the form of Hankel matrix H L+η (u s ), and calculate the row rank of this matrix. If this matrix is ​​full row rank, keep this matrix, otherwise continue to select control inputs according to the requirements in step S2 until the Hankel matrix composed of input data is full row rank. The construction of the Hankel matrix is ​​as follows:

[0065]

[0066] in Represents the input amount at the i-th moment in the process of collecting data. Similarly, construct the output data y s The Hankel matrix H L+η (y s ) and construct the combined Hankel matrix of input-augmented state in For i∈[η,N-1].

[0067] S4: Using the Hankel matrix of the output and input data defined above, select parameters i = 1, 2, ..., L-1 and solve the following optimization problems respectively:

[0068] p i =max y，α ||y i || ∞

[0069] st||y [0，η-1] || ∞ ≤1

[0070]

[0071] where ||·|| ∞ Represents the infinite norm of the variable, parameter is the weight vector, is the predicted output, y [0，η-1] represents the output value from time 0 to time η-1, H i+η (u s ) indicates that the row rank is (i+η)n u About The Hankel matrix, H i+η (y s ) indicates that the row rank is (i+η)n y About The Hankel matrix of . Record the solved ρ i , where i = 1, 2,…, L-1. This parameter can be regarded as the observable parameter of the system.

[0072] S5: Select a constant r>0 and solve the real positive definite matrix P r And the matrix K r The augmented state that satisfies the condition Under the state feedback law u t =u e + K r (ξ t -ξ e ) after the augmented state is always in the set Ξ r In, that is, τ ∈Ξ r For any τ ≥ t, the inequality holds, where the matrices R and Q are those defined above, and ||X|| Y represents the Y norm for any matrix X, specifically expressed as ||X|| Y =XT YX, the same below. Then select a constant 0<∈≤r and construct the set Ξ according to the same rule ∈ , real positive definite matrix matrix P, and state feedback matrix K. And the constant ∈ and constant r satisfy the relationship in represents the largest eigenvalue of the matrix P, λ (P) represents the minimum eigenvalue of the matrix P, Represents the matrix P r The maximum eigenvalue of .

[0073] S6: Define the self-triggering moment as Where N0 = {0, 1, 2, ...} is a set of natural numbers including 0, and the time interval between two adjacent self-triggering moments is defined as τ l :=t l+1 -t l , where l∈ N0. If the system is in the self-triggering moment, the system's output data packet will be transmitted to the controller side. After receiving the output data packet, the controller side solves the following optimization problem:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] Where h(t l ), g(t l ), is the physical quantity to be solved; yes R weight norm of the matrix; is a matrix The Q weight norm of ; represents the loss function; t l represents the lth self-triggering moment; represents the current time t to be predicted l The first n inputs and the last L-1 inputs; represents the current time t to be predicted l The previous n outputs and the next L-1 outputs; u min and u maxis the maximum and minimum value of the pre-given input; u [-η，-1] ,y [-η，-1] are the values ​​of n initial inputs and outputs given to the system;

[0080] represents the current time t l The actual input data of the previous n systems;

[0081] Indicates that from time t l -η to time t l -1, the controller side of the system is disturbed by the noise nt in the network, and the current time t actually received l The previous n output values; variable

[0082] represents the current time t l i system inputs to be predicted, i∈[-η,L-1];

[0083] represents the current time t l The i system outputs to be predicted i∈[-η, L-1];

[0084] represents the current time t to be predicted l Input data of the previous n systems;

[0085] represents the current time t to be predicted l Output data of the previous n systems;

[0086] h i (t l ) represents the time t l The i h(t l ) value;

[0087] Indicates time t l The constructed current time t l To the augmented state of the subsequent L systems;

[0088] Indicates time t l The constructed i-th ξ(t l ) value.

[0089] S7: The input sequence obtained by the above solution Transmitted to the system side to control the system.

[0090] S8: Using the solution of S4 optimization problem, solve the following problem

[0091]

[0092] in

[0093]

[0094]

[0095] Here, the parameter σ∈(0, 1) is a threshold parameter given artificially, and ||·|| represents the binary norm of the variable ·, which can be obtained through existing technology. Represents the matrix P r The maximum eigenvalue of represents the largest eigenvalue of the matrix Q, λ (R) represents the minimum eigenvalue of the matrix R, Denotes the matrix H uξ The left pseudo-inverse matrix of h can be obtained by existing technology * (t l ) indicates that at t l By solving the optimization problem in S4, we can get the variable h(t l ), Indicates time t l The i-th h to be predicted * (t l ), Indicates that at t l The variables are obtained by solving the optimization problem in S4 at all times The optimal solution of Indicates time t l The i-th The value in. Use the calculation to get the interval τ between this transmission time and the next transmission time l , calculate the next self-triggering time t l+1 :=t l +τ l , and transmit this moment to the sensor side through the network.

[0096] S9: At this moment t∈(t l , t l+1 ) is not the self-triggering moment, the sensor side does not transmit the output value to the controller side. At this time, the controller side uses t l The control value corresponding to this moment in the control sequence solved at this moment Control the system, including Indicates that at t l The variables are obtained by solving the optimization problem in S4 at all times The optimal solution.

[0097] like Figure 2 , which is an implementation flow chart of a data-driven self-triggering control method for an unknown system provided by the present invention;

[0098] like Figure 3 The effect diagram of the data-driven self-triggering control method for an unknown system provided by the present invention running for 200 unit times on an example. The state matrix of the system to be stabilized is

[0099]

[0100]

[0101] The observation index of the system is η = 2, and a system input and output trace of length N = 800 is collected in advance. Assume that the upper bound of the system network transmission noise is The matrix parameters and matrices in the data-driven prediction algorithm are selected as Q = I, R = 8·10 -3 I, The prediction time domain L = 11, the input limit of the system is [u min ,u max ]=[-2,2], set the equilibrium point of the system to (u e ,y e )=([1,1] T , [0.65, 0.77] T ), the selected parameter σ=0.88, the initial value of the system is x0= [0.4,0.4,0.5,0.5] T The horizontal axis in the simulation results represents the time step, and the trajectory of the system output changing with time is shown in Figure 3 The solid line and dashed line in (a) are shown. The dashed lines near 0.6 and 0.8 in the figure are the first component of the system output (solid line) and the second component of the system output (dotted line), which are the equilibrium points of the system. The sampling time of self-triggering is shown in Figure 3 The simulation results show the effectiveness of the invented data-driven self-triggering control method for unknown systems.

[0102] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data-driven self-triggering control method for an unknown system, applied to a system to be stabilized, wherein the system to be stabilized realizes closed-loop control through a data-driven controller, a self-triggering module, a sensor side and a controller side, characterized in that: The method is: Before the system to be stabilized is run online, sufficient input and output trajectories are collected offline. During the online operation, at each sampling moment, the sensor side sends the data packet containing the system output to the controller side through the network. The controller side generates a series of inputs by solving a data-based optimization problem based on the received output data packets. These inputs are sent to the system to be stabilized in turn as the control input of the system until the next output sampling moment, when a new input sequence is recalculated. At the same time, the controller side calculates the next transmission time according to the optimal solution of the optimization problem and sends it to the sensor side through the network. Step 1: For an observable and stabilizable linear time-invariant system, set x t ,y t ,u t They are the state value, output value and control input value of the system at time t, and their dimensions are n x ,n y ,n u , the stable point of the system is (x e ,u e ,y e ), the system output is subject to bounded network noise n during transmission t , at each self-triggered transmission time t l , the sensor side of the system sends the most recent n outputs to the controller side On the other hand, the control input of the system controls the system in a periodic transmission manner; Assume the dimension of the system is a known quantity n x , the observable index of the system is η, and set L≥n x +η; constant λ g >0, constant λ h >0; real positive definite symmetric matrix Q, R; network noise boundary Step 2: At any initial value Next, select N control inputs Record the input data used by the system and the resulting output data Step 3: Arrange the trajectory composed of the above N input data in the form of Hankel matrix H L+η (u s ), so that the row rank of the matrix is ​​(L+η)n u ; Similarly, construct output data The Hankel matrix H L+η (y s ) and construct the combined Hankel matrix of input-augmented state in For i∈[η,N-1], the superscript represents the transpose of a matrix, Representing a collection The input vectors from i-n to i-1 in, Representing a collection The output vectors from i-n to i-1; Step 4: Use the Hankel matrix of the output and input data constructed above to solve the observable parameter ρ of the system i , where i = 1, 2, ..., L-1; in‖·‖ ∞ Represents the infinite norm of the variable, parameter is the weight vector, is the predicted output, y [0,η-1] represents the output value from time 0 to time η-1, H i+η (u s ) indicates that the row rank is (i+η)n u About The Hankel matrix, H i+η (y s ) indicates that the row rank is (i+η)n y About The Hankel matrix of Step 5: Define the augmented state of the system at time t as Similarly, we define the extended state stable point ξ of the system e , when the subscript is less than zero, that is, when t∈[0,η), it represents the input and output values ​​of the system initial setting; select the constant r>0, and solve the real positive definite matrix P r And the matrix K r , so that the augmented state Under the state feedback law u t =u e +K r (ξ t -ξ e ) in the time after the set r and converges, where n ξ =η(n u +n y ), Indicates dimension n ξ ×1 real vector set; then select a constant 0<∈≤r and construct the set Ξ according to the same rule ∈ , real positive definite matrix P, and the constant ∈ and the constant r need to satisfy the relationship in, Represents the matrix P r The maximum eigenvalue of , λ(P) represents the minimum eigenvalue of the matrix P; Step 6: Define the self-triggering time as Where N0 = {0, 1, 2, ...} is a set of natural numbers including 0, and the time interval between two adjacent self-triggering moments is defined as τ l :t l+1 -t l , where l∈N0; if the system is in the self-triggering moment at this time, the system's output data packet will be transmitted to the controller side; the controller side solves the following optimization problem after receiving the output data packet: in is the physical quantity to be solved; g(t l ) represents the weight vector, yes R weight norm of the matrix; is a matrix The Q weight norm of ; represents the loss function; t l represents the lth self-triggering moment; represents the current time t to be predicted l The first n inputs and the last L-1 inputs; represents the current time t to be predicted l The previous n outputs and the next L-1 outputs; u min and u max is the maximum and minimum value of the pre-given input; u [-η,-1] ,y [-η,-1] are the values ​​of n initial inputs and outputs given to the system; Represents the column vector consisting of the actual inputs of the n systems before the current time t\; Indicates that from time t l -η to time t l -1, the system is affected by the noise n in the network t The intrusion controller side actually receives the current time t l The previous n output values, where the variable represents the current time t l i system inputs to be predicted, i∈[-η,L-1]; represents the current time t l The i system outputs to be predicted i∈[-η,L-1]; represents the current time t to be predicted l Input data of the previous n systems; represents the current time t to be predicted l Output data of the previous n systems; h i (t l ) represents the time t l The i h(t l ) value; Indicates time t l The constructed current time t l To the augmented state of the subsequent L systems; Indicates time t l The constructed i-th ξ(t l ) value; Step 7: Substitute the above-solved input sequence Transmit to the system side to control the system; Using the solution to the optimization problem in step 6, solve the following problem in Here, the parameter σ∈(0,1) is a threshold parameter given artificially, ‖·‖ represents the binary norm of variable ·, Represents the matrix P r The maximum eigenvalue of represents the maximum eigenvalue of the matrix Q, λ(R) represents the minimum eigenvalue of the matrix R, Denotes the matrix H uξ The left pseudo-inverse matrix, h * (t l ) indicates that at t l By solving the optimization problem in step 6, we can get the variable h(t l ), Indicates time t l The i-th h to be predicted * (t l ), Indicates that at t e The variables are obtained by solving the optimization problem in step 6 at any time. The optimal solution of Indicates time t e The i-th The value in; use the calculation to get the interval τ between this transmission time and the next transmission time l , calculate the next self-triggering time t l+1 :=t l +τ l , and transmit this moment to the sensor side through the network.

2. A data-driven self-triggering control method for an unknown system as claimed in claim 1, characterized in that: In step 2, each component of the control input is randomly selected from the interval [-1,1].

3. A data-driven self-triggering control method for an unknown system as claimed in claim 1, characterized in that: If the current time t∈(t l ,t l+1 ) is not the self-triggering moment, the sensor side does not transmit the output value to the controller side. At this time, the controller side uses t l The control value corresponding to this moment in the control sequence solved at this moment Control the system, including Indicates that at t l The variables are obtained by solving the optimization problem in step 6 at any time. The optimal solution.

Images