Nonlinear intermittent process iterative learning fault-tolerant prediction control method based on dynamic output feedback

By introducing dynamic output feedback and T-S fuzzy Rosser model in iterative learning control, combining the robust positive and fixed invariant set and terminal constraint set, the problem of actuator failure and unpredictable state during nonlinear intermittent processes is solved, and the system is fast Stable tracking and robustness improvements.

CN118915463APending Publication Date: 2024-11-08UNIV OF SCI & TECH LIAONING
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411059298.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-03
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Traditional iterative learning control methods are difficult to meet high-precision control requirements when dealing with nonlinear intermittent processes, especially in the case of actuator failure and unpredictable state, and existing methods are difficult to achieve real-time feedback and rapid convergence.

Method used

采用基于动态输出反馈的迭代学习预测控制方法,通过构建T-S模糊Rosser模型,结合鲁棒正定不变集和终端约束集,设计出实时最优控制法则,确保系统在执行器故障时稳定性和容错性。

Benefits of technology

The system stability and rapid convergence in the case of actuator failure and uncertainty is achieved, which shortens the tracking time of batch direction, enhances robustness and fault tolerance, and reduces the amount of calculation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915463A_ABST
    Figure CN118915463A_ABST
Patent Text Reader

Abstract

A nonlinear intermittent process iterative learning fault-tolerant predictive control method based on dynamic output feedback belongs to the field of advanced control of industrial processes, and comprises the following steps: step 1, establishing a state space model for an intermittent process with actuator faults; step 2, designing a T-S fuzzy Rosser model containing a time dimension and a batch dimension; 3, based on the T-S fuzzy Rosser model, designing an iterative learning robust model prediction controller based on dynamic output feedback; step 4, a Lyapunov-Krasovskii function is constructed, and the Lyapunov-Krasovskii function is constructed; 5, calculating gains Ac, Bc and Cc of a robust model prediction controller based on dynamic output feedback by using the robust positive definite invariant set and the terminal constraint set; according to the method, through a robust positive definite invariant set and a terminal constraint set, it is ensured that the state of the system can be converged to a safe area when an actuator breaks down, and therefore the stability and fault tolerance of the system are guaranteed. Meanwhile, an online solving mode is adopted, the gain of the controller is optimized in real time, and the response speed of the system is remarkably improved through the method.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The proposed method belongs to the field of advanced control of industrial processes and involves an iterative learning fault-tolerant predictive control method for nonlinear intermittent processes based on dynamic output feedback. Background Art

[0002] Over the past three decades, batch process control theory has been widely used in many fields, such as semiconductor manufacturing, plastic processing, and pharmaceuticals. Iterative learning control (ILC) methods have become one of the key methods for handling batch processes because they can quickly adapt to changes in product types and quantities. However, traditional ILC methods lack a closed-loop structure in the time domain and are difficult to meet the increasing control accuracy requirements. To overcome these limitations, a two-dimensional (2D) ILC method was proposed, which suppresses non-repetitive interference while ensuring stability in the batch and time dimensions. However, in actual industrial processes, this method has shortcomings in dealing with physical constraints.

[0003] Model predictive control (MPC) is widely used in batch process control because of its ability to handle constraints. Researchers have combined 2D-ILC with robust MPC and proposed a series of iterative learning MPC methods to improve the control performance and stability of the system. However, most of these methods are based on linear models and treat nonlinear characteristics as uncertainties or disturbances, making it difficult to achieve high-precision control. The fuzzy iterative learning control method alleviates the impact of non-repetitive disturbances on system stability to a certain extent by establishing a more accurate model, but since the control law gain is solved offline, it is difficult to effectively handle the real-time uncertainty of the system and the convergence speed is slow.

[0004] In actual industrial production environments, states are often unmeasurable. Most current research is based on the assumption that states are measurable, which makes it difficult to apply theory to practice. Output feedback methods have been proposed to address the problem of unavailable states. However, these methods only give a global optimal solution for each batch and cannot provide real-time feedback for system output, causing the output to deviate from the set value.

[0005] In addition, actuator failures cannot be ignored in batch processing, and their potential impact may cause serious damage to stability and control performance. Fault-tolerant control (FTC) strategies are divided into active fault tolerance and passive fault tolerance. Active fault tolerance has better adaptability, but due to its complex structure, it is difficult to debug in real time. Passive fault tolerance maintains control performance at an acceptable level when a fault occurs, which is more suitable for fault problems in batch processing. Under the ILC framework, some new fault-tolerant strategies have been proposed to deal with actuator failures and external interference. Although they are effective, their complex optimization process increases the difficulty of design and implementation. Summary of the invention

[0006] An iterative learning predictive fault-tolerant control method based on dynamic output feedback is developed for nonlinear intermittent processes with actuator faults. Firstly, a new TS fuzzy Rosser model is constructed, which integrates the information of time dimension and batch dimension. Considering the problem of unmeasurable state, dynamic output feedback is combined with iterative learning model predictive control for controller design. Then, the stability conditions are given by using robust positive definite invariant sets and terminal constraint sets, so that the state of the system can converge to the safe region when the actuator fails, thereby ensuring stability and fault tolerance. By solving these conditions, the real-time optimal control law gain can be obtained. Finally, simulation verification is carried out, and the results show that the proposed method is effective.

[0007] The proposed method is implemented by the following method:

[0008] Step 1: Establish a state-space model for a nonlinear intermittent process with actuator failure:

[0009] The model of the nonlinear intermittent process is as follows:

[0010]

[0011] In the formula, s and κ represent the time dimension and batch dimension respectively. Represent the system state, control output and control input respectively; represents Euclidean space, n x ,n y ,n u Represents an appropriate dimension, T and δ are nonlinear functions; using the TS fuzzy method and taking into account the external bounded interference, the nonlinear intermittent process can be expressed as follows:

[0012] Rules q:

[0013]

[0014] Then:

[0015]

[0016] In the formula, ω(s,κ) represents the external bounded disturbance, q represents the fuzzy rule, l represents the fuzzy set, Z→{Z1,…,Z} is the antecedent variable, represents a fuzzy set, A q (s,κ) represents the state uncertainty matrix of the qth fuzzy rule at the discrete time s in the κth batch, and satisfies A q (s,κ)=A q +Δ a (s,κ), where Δ a (s,κ)=N q Δ(s,κ)Ha ; A q , B q and C represent the system state constant matrix, control input constant matrix and output constant matrix of the qth fuzzy rule respectively; N and H a is a constant matrix with appropriate dimension; Δ(s,κ) represents a bounded perturbation and satisfies Δ(s,κ)Δ(s,κ)≤I, where I represents an identity matrix of appropriate dimension; the nonlinear intermittent process (2) is further rewritten as:

[0017]

[0018] In the formula, μ q (Z) represents the membership function, A μ (s,κ) represents the weighted sum matrix of the uncertainty of the state at discrete time s of the κth batch, satisfying B μ , C μ Denote the control input constant weighted sum matrix and output constant weighted sum matrix respectively, satisfying represents the number of fuzzy rules; q satisfies represents a set of positive integers; u F (s,κ) is the control input with partial actuator failure at discrete time s in the κth batch, satisfying the following expression:

[0019] u F (s,κ)=αu(s,κ) (4)

[0020] Step 2: Design a TS fuzzy Rosser model that includes time dimension and batch dimension:

[0021] In this section, we will design a TS fuzzy Rosser model that includes time dimension and batch dimension, and reconstruct the nonlinear model (3) to provide a strong guarantee for controller design. First, define an output error:

[0022] e(s,κ)=y(s,κ)-y r (5)

[0023] In the formula, y r Indicates the set value;

[0024] Next, define the following control law:

[0025]

[0026] Where r(s,κ) represents the iterative learning control input at discrete time s in the κth batch;

[0027] Therefore, a state-enhanced TS fuzzy Roesser model is obtained as follows:

[0028]

[0029] In the formula, represents the weighted sum matrix of state uncertainty expanded at discrete time s in the κth batch, represents the expanded state uncertainty matrix of the qth fuzzy rule at discrete time s in the κth batch, The extended system state constant matrix of the qth fuzzy rule satisfy, is an extended constant matrix of appropriate dimension, satisfying and represents the weighted sum matrix of control inputs expanded at discrete time s in the κth batch, satisfying represents the control input constant matrix expanded at discrete time s for the κth batch, D = [IC μ ] T is the interference matrix, represents the expanded output matrix, represents the extended state of the κth batch at discrete time s+1, satisfying represents the extended state of the κth batch at discrete time s, satisfying represents the extended unknown bounded interference, satisfying

[0030] Step 3: Based on the TS fuzzy Rosser model, design an iterative learning robust model predictive controller based on dynamic output feedback:

[0031] Based on formula (7), when the state is unmeasurable, a robust model predictive controller based on dynamic output feedback is designed as follows:

[0032]

[0033] In the formula, represents the controller state at discrete time s of the κth batch, represents the appropriate dimension, A c ,B c ,C c is the controller gain;

[0034] The iterative learning robust model predictive controller based on dynamic output feedback is further expressed as:

[0035]

[0036] In the formula, is the gain matrix of the control input, satisfying l and m are the prediction steps in time and batch direction respectively;

[0037] Combining equations (7), (8) and (9), the following closed-loop TS fuzzy Roesser feedback error prediction model is obtained:

[0038]

[0039] In the formula, Represents the closed-loop system state of the κ+mth batch at the s+1+lth time, satisfying represents the output matrix of the closed-loop system, represents the disturbance matrix of the closed-loop system, Represents the closed-loop system state uncertainty matrix, satisfying

[0040] Then, the finite optimal cost function of the robust model predictive control is given as follows, and the robust model predictive control optimization problem is described:

[0041]

[0042] Subject to the following constraints: In the formula, is the terminal cost, Ω t is the terminal constraint set; the input perturbation satisfy Iterative learning control input r, satisfying η is a known constant, τ is a positive scalar, r K is the Kth element of the control input;

[0043] Step 4: Construct the Lyapunov-Krasovskii function:

[0044] The Lyapunov-Krasovskii function with time and batch direction information is given as follows:

[0045]

[0046] Where P μ is a positive definite matrix of appropriate dimension;

[0047] Step 5: Using the robust positive definite invariant set and the terminal constraint set, calculate the gain A of the robust model predictive controller based on dynamic output feedback c ,B c ,C c :

[0048] System control law gain A c ,B c ,C c It is obtained by solving the robust positive definite invariant set and the collective conditions of the terminal constraints, referring to equations (25)-(28) in the specific implementation method, to ensure the stability and effectiveness of the system in the case of unmeasurable state and partial failure of the actuator.

[0049] Compared with the existing methods, the advantages of the proposed method are as follows:

[0050] The main advantages of the proposed method include

[0051] (1) By solving the control law gain in real time, the defect of the initial batch lacking historical data for optimization learning is compensated, the convergence speed along the batch direction is shortened, and stable tracking of the set point is achieved in fewer operating cycles.

[0052] (2) When dealing with faults and uncertainties, this method does not introduce additional performance indicators, but instead uses robust positive definite invariant sets and terminal constraint sets to derive stability conditions under actuator faults and interferences to ensure that the system is always in the safe region. This improvement not only enhances the robustness and fault tolerance of the system, but also reduces the amount of computation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The output response diagram of the proposed method;

[0054] Figure 2 The output response diagram of the system for the two-dimensional offline iterative learning method;

[0055] Figure 3 The control input diagram for the proposed method;

[0056] Figure 4 It is the control input map of the two-dimensional offline iterative learning method;

[0057] Figure 5 DTI index diagram of the proposed method;

[0058] Figure 6 It is the DTI index map of the two-dimensional offline iterative learning method;

[0059] Figure 7 A flowchart of the steps of the proposed method is shown in Figure 2. DETAILED DESCRIPTION

[0060] The proposed method is further explained below in conjunction with the accompanying drawings and embodiments.

[0061] The nonlinear intermittent process iterative learning fault-tolerant predictive control method based on dynamic output feedback is characterized by: the specific steps are as follows:

[0062] Step 1: Establish a state-space model for a nonlinear intermittent process with actuator failure:

[0063] The model of the nonlinear intermittent process is as follows:

[0064]

[0065] In the formula, s and κ represent the time dimension and batch dimension respectively. Represent the system state, control output and control input respectively; represents Euclidean space, n x ,n y ,n u Represents an appropriate dimension, T and δ are nonlinear functions; using the TS fuzzy method and taking into account the external bounded interference, the nonlinear intermittent process can be expressed as follows:

[0066] Rules q:

[0067]

[0068] Then:

[0069]

[0070] In the formula, ω(s,κ) represents the external bounded disturbance, q represents the fuzzy rule, l represents the fuzzy set, Z→{Z1,…,Z l} is the antecedent variable, represents a fuzzy set, A q (s,κ) represents the state uncertainty matrix of the qth fuzzy rule at the discrete time s in the κth batch, and satisfies A q (s,κ)=A q +Δ a (s,κ), where Δ a (s,κ)=N q Δ(s,κ)H a ; A q , B q and C represent the system state constant matrix, control input constant matrix and output constant matrix of the qth fuzzy rule respectively; N and H a is a constant matrix with appropriate dimension; Δ(s,κ) represents a bounded perturbation and satisfies Δ(s,κ)Δ(s,κ)≤I, where I represents an identity matrix of appropriate dimension; the nonlinear intermittent process (18) is further rewritten as:

[0071]

[0072] In the formula, μ q (Z) represents the membership function, A μ(s,κ) represents the weighted sum matrix of the uncertainty of the state at discrete time s of the κth batch, satisfying B μ , C μ Denote the control input constant weighted sum matrix and output constant weighted sum matrix respectively, satisfying represents the number of fuzzy rules; q satisfies represents a set of positive integers; u F (s,κ) is the control input with partial actuator failure at discrete time s in the κth batch, satisfying the following expression:

[0073] u F (s,κ)=αu(s,κ) (20)

[0074] Step 2: Design a TS fuzzy Rosser model that includes time dimension and batch dimension:

[0075] In this section, we will design a TS fuzzy Rosser model that includes time dimension and batch dimension, and reconstruct the nonlinear model (19) to provide a strong guarantee for controller design. First, define an output error:

[0076] e(s,κ)=y(s,κ)-y r (twenty one)

[0077] In the formula, y r Indicates the set value;

[0078] Next, define the following control law:

[0079]

[0080] Where r(s,κ) represents the iterative learning control input at discrete time s in the κth batch;

[0081] Therefore, a state-enhanced TS fuzzy Roesser model is obtained as follows:

[0082]

[0083] In the formula, represents the weighted sum matrix of state uncertainty expanded at discrete time s in the κth batch, represents the expanded state uncertainty matrix of the qth fuzzy rule at discrete time s in the κth batch, The extended system state constant matrix of the qth fuzzy rule satisfy, is an extended constant matrix of appropriate dimension, satisfying and represents the weighted sum matrix of control inputs expanded at discrete time s in the κth batch, satisfying represents the control input constant matrix expanded at discrete time s for the κth batch, D = [IC μ ] T is the interference matrix, represents the expanded output matrix, represents the extended state of the κth batch at discrete time s+1, satisfying represents the extended state of the κth batch at discrete time s, satisfying represents the extended unknown bounded interference, satisfying

[0084] Step 3: Based on the TS fuzzy Rosser model, design an iterative learning robust model predictive controller based on dynamic output feedback:

[0085] Based on formula (23), when the state is unmeasurable, a robust model predictive controller based on dynamic output feedback is designed as follows:

[0086]

[0087] In the formula, represents the controller state at discrete time s of the κth batch, represents the appropriate dimension, A c ,B c ,C c is the controller gain;

[0088] The iterative learning robust model predictive controller based on dynamic output feedback is further expressed as:

[0089]

[0090] In the formula, is the gain matrix of the control input, satisfying l and m are the prediction steps in time and batch direction respectively;

[0091] Combining equations (23), (24) and (25), the following closed-loop TS fuzzy Roesser feedback error prediction model is obtained:

[0092]

[0093] In the formula, Represents the closed-loop system state of the κ+mth batch at the s+1+lth time, satisfying represents the output matrix of the closed-loop system, represents the disturbance matrix of the closed-loop system, Represents the closed-loop system state uncertainty matrix, satisfying

[0094] Then, the finite optimal cost function of the robust model predictive control is given as follows, and the robust model predictive control optimization problem is described:

[0095]

[0096] Subject to the following constraints: In the formula, is the terminal cost, Ω t is the terminal constraint set; the input perturbation satisfy Iterative learning control input r, satisfying η is a known constant, τ is a positive scalar, r K is the Kth element of the control input;

[0097] Step 4: Construct the Lyapunov-Krasovskii function:

[0098] The Lyapunov-Krasovskii function with time and batch direction information is given as follows:

[0099]

[0100] Where P μ is a positive definite matrix of appropriate dimension;

[0101] Step 5: Using the robust positive definite invariant set and the terminal constraint set, calculate the gain A of the robust model predictive controller based on dynamic output feedback c ,B c ,C c :

[0102] Linear matrix inequalities satisfy the following robust positive definite invariant set:

[0103]

[0104]

[0105] In the formula, Z3=βC c N T , S2=XS1, S1 and S2 are intermediate variable matrices, X and X -1is a positive definite matrix, M and N are full rank square matrices, ε1,ε2, are unknown scalars;

[0106] Terminal constraint set:

[0107]

[0108] In the formula,

[0109]

[0110] By solving the linear matrix inequality conditions (29)-(32) online, the system control law gain A is obtained c ,B c ,C c , ensuring the stability and effectiveness of the system in the event of unpredictable states and partial actuator failures.

[0111] Example:

[0112] Proposed method A nonlinear intermittent process iterative learning fault-tolerant predictive control method based on dynamic output feedback is proposed. This method can effectively solve the problems of actuator failure and system stability.

[0113] For a nonlinear batch reactor with partial actuator failure, the expression is as follows:

[0114]

[0115] Where T and T c Represent the reactor temperature and coolant temperature respectively. Other related parameters are 200≤T c ≤450(K),0.01≤C A ≤1(mol / L),250≤T≤500(K), C Af =1(mol / L),ρ=1(kg / L),C p =1(J / gK),ΔH=-2×10 5 (J / mol),UA=1×10 5 (J / minK), k0=4.71×10 8 min -1 ,E / R=8000K.

[0116] Let x(s,κ)=[x 1 (s,κ) x 2 (s,κ)] T =[T(s,κ) C A (s,κ)] T ,u(s,κ)=Tc (s,κ). The membership function is:

[0117]

[0118] Therefore, the expression for a nonlinear batch reactor with partial actuator failure is transformed into:

[0119]

[0120] Where q=3,Δ(s,κ)∈[-1 1],

[0121] Next, the system setpoint is defined as

[0122]

[0123] In order to evaluate the tracking performance of the system in the time direction, the dynamic performance index (DTI) is defined as follows:

[0124]

[0125] Where T s Indicates the operation time.

[0126] exist Figure 1 and Figure 2 In the figure, the output tracking effects of the proposed method and the two-dimensional offline iterative learning method are shown. When the random fault is α=0.8+0.1Δ1, Figure 1 It exhibits a stable output response and is able to quickly track the set point in the third batch. Figure 2 The output response can only track the set value in the ninth batch, and shows large fluctuations and slow response speed. This shows that the proposed method has better performance and stability in dealing with random faults.

[0127] Figure 3 and Figure 4 The control inputs of the proposed method and the two-dimensional offline iterative learning method are shown respectively. Figure 3 The control input of the system shows higher sensitivity and stability in the face of faults and can adjust quickly to keep the system running normally. Figure 4 The control input of the proposed method fluctuates greatly under random fault conditions, which is not conducive to practical industrial applications. Since the fluctuation of the control input may lead to increased equipment loss, the proposed method has more advantages in practical applications.

[0128] Figure 5 and Figure 6The DTI of the proposed method and the two-dimensional offline iterative learning method are shown respectively. Under random fault conditions, the DTI of the proposed method gradually decreases, indicating that the tracking error of the system gradually decreases and the system performance gradually improves. Figure 6 The DTI shown has a large overshoot in the early stage, which decreases over time but remains high. This indicates that the proposed method has better tracking performance and lower error under random fault conditions.

[0129] In summary, the proposed method develops an iterative learning predictive fault-tolerant control method based on dynamic output feedback. Taking the temperature control process of a nonlinear batch reactor as an example, the stability and effectiveness of the designed controller are verified. By solving the control law gain in real time, the defect of the initial batch lacking historical data for optimization learning is made up, the convergence speed along the batch direction is shortened, and the stable tracking of the set point is achieved in fewer operating cycles. At the same time, when dealing with faults and uncertainties, this method does not introduce additional performance indicators, but uses robust positive definite invariant sets and terminal constraint sets to derive stability conditions under actuator faults and interferences to ensure that the system is always in a safe area. This improvement not only enhances the robustness and fault tolerance of the system, but also reduces the amount of calculation. Finally, the simulation results prove the effectiveness and feasibility of the proposed method and demonstrate its application potential in actual industrial processes.

Claims

1. Iterative learning fault-tolerant predictive control method for nonlinear intermittent processes based on dynamic output feedback, Features: The specific steps are as follows: Step 1: Establish a state-space model for a nonlinear intermittent process with actuator failure: The model of the nonlinear intermittent process is as follows: In the formula, s and κ represent the time dimension and batch dimension respectively. Represent the system state, control output and control input respectively; represents Euclidean space, n x ,n y ,n u Represents an appropriate dimension, T and δ are nonlinear functions; using the TS fuzzy method and taking into account the external bounded disturbance, the nonlinear intermittent process can be expressed as follows: Rules q: Then: In the formula, ω(s,κ) represents external bounded interference, q represents fuzzy rules, l represents fuzzy sets, is the antecedent variable, represents a fuzzy set, A q (s,κ) represents the state uncertainty matrix of the qth fuzzy rule at the discrete time s in the κth batch, and satisfies A q (s,κ)=A q +Δ a (s,κ), where Δ a (s,κ)=N q Δ(s,κ)H a ; A q , B q and C represent the system state constant matrix, control input constant matrix and output constant matrix of the qth fuzzy rule respectively; N and H a is a constant matrix with appropriate dimension; Δ(s,κ) represents a bounded perturbation and satisfies Δ(s,κ)Δ(s,κ)≤I, where I represents an identity matrix of appropriate dimension; the nonlinear intermittent process (2) is further rewritten as: In the formula, μ q (Z) represents the membership function, A μ (s,κ) represents the weighted sum matrix of the uncertainty of the state at discrete time s of the κth batch, satisfying B μ ; , C μ ; represent the control input constant weighted sum matrix and output constant weighted sum matrix respectively, satisfying represents the number of fuzzy rules; q satisfies represents a set of positive integers; u F (s,κ) is the control input with partial actuator failure at discrete time s in the κth batch, satisfying the following expression: you F (s,κ)=αu(s,κ) (4) Step 2: Design a TS fuzzy Rosser model that includes time dimension and batch dimension: In this section, we will design a TS fuzzy Rosser model that includes time dimension and batch dimension, and reconstruct the nonlinear model (3) to provide a strong guarantee for controller design. First, we define an output error: e(s,κ)=y(s,κ)-y r (5) In the formula, y r Indicates the set value; Next, define the following control law: Where r(s,κ) represents the iterative learning control input at discrete time s in the κth batch; Therefore, a state-enhanced TS fuzzy Roesser model is obtained as follows: In the formula, represents the weighted sum matrix of state uncertainty expanded at discrete time s in the κth batch, represents the expanded state uncertainty matrix of the qth fuzzy rule at discrete time s in the κth batch, The extended system state constant matrix of the qth fuzzy rule satisfy, is an extended constant matrix of appropriate dimension, satisfying ;and ; represents the weighted sum matrix of the control input expanded at discrete time s of the κth batch, satisfying represents the control input constant matrix expanded at discrete time s for the κth batch, D = [IC μ ] T is the interference matrix, represents the expanded output matrix, represents the extended state of the κth batch at discrete time s+1, satisfying ; represents the extended state of the κth batch at discrete time s, satisfying represents the extended unknown bounded interference, satisfying Step 3: Based on the TS fuzzy Rosser model, design an iterative learning robust model predictive controller based on dynamic output feedback: Based on formula (7), when the state is unmeasurable, a robust model predictive controller based on dynamic output feedback is designed as follows: In the formula, represents the controller state at discrete time s of the κth batch, represents the appropriate dimension, A c ,B c ,C c is the controller gain; The iterative learning robust model predictive controller based on dynamic output feedback is further expressed as: In the formula, is the gain matrix of the control input, satisfying l and m are the prediction steps in time and batch direction respectively; Combining equations (7), (8) and (9), the following closed-loop TS fuzzy Roesser feedback error prediction model is obtained: In the formula, ; represents the closed-loop system state of the batch κ+m at the s+1+lth time, satisfying represents the output matrix of the closed-loop system, represents the disturbance matrix of the closed-loop system, Represents the closed-loop system state uncertainty matrix, satisfying Then, the finite optimal cost function of the robust model predictive control is given as follows, and the robust model predictive control optimization problem is described: Subject to the following constraints: In the formula, is the terminal cost, Ω t is the terminal constraint set; the input perturbation satisfy Iterative learning control input r, satisfying η is a known constant, τ is a positive scalar, r K is the Kth element of the control input; Step 4: Construct the Lyapunov-Krasovskii function: The Lyapunov-Krasovskii function with time and batch direction information is given as follows: Where P μ is a positive definite matrix of appropriate dimension; Step 5: Using the robust positive definite invariant set and the terminal constraint set, calculate the gain A of the robust model predictive controller based on dynamic output feedback c ,B c ,C c : Linear matrix inequalities satisfy the following robust positive definite invariant set: In the formula, S 2 =XS 1 , S 1 and S 2 is the intermediate variable matrix, X and X -1 is a positive definite matrix, M and N are full rank square matrices, ε 1 ,ε 2 , is an unknown scalar; Terminal constraint set: In the formula, By solving the linear matrix inequality conditions (13)-(16) online, the system control law gain A is obtained c ,B c ,C c , ensuring the stability and effectiveness of the system in the event of unpredictable states and partial actuator failures.