Signal compensation robust prediction real-time optimization control method for outlet temperature of ethylene cracking furnace

Abstract

The invention discloses a signal compensation robust prediction real-time optimization control method for the outlet temperature of an ethylene cracking furnace, and belongs to the field of advanced control of industrial processes. Comprising the following steps: step 1, establishing a bottom layer model of a discrete state space with parameter uncertainty and unmodeled dynamics and a lower layer control loop model based on bottom layer model extension for the outlet temperature control system of the ethylene cracking furnace; 2, designing a lower loop control layer signal compensation robust prediction controller; 3, expanding and establishing a discrete state space upper layer operation optimization layer model with parameter uncertainty and unmodeled dynamics based on a lower layer control loop; 4, designing an upper operation optimization layer signal compensation robust prediction controller; according to the method, a brand new scheme is provided for the real-time optimization design of the ethylene cracking furnace outlet temperature control system with uncertainty and strong nonlinearity, so that the system can give a more accurate optimized set value in real time under the condition near an economic optimal state.

Description

Technical Field

[0001] The invention belongs to the field of advanced control of industrial processes and relates to a signal compensation robust prediction real-time optimization control method for the outlet temperature of an ethylene cracking furnace. Background Art

[0002] In recent years, with the continuous development of industrialization, competition in the field of industrial processes has also intensified. In order to maintain competitiveness, modern factories must operate in a cost-optimal manner and respond to changes that affect their economic efficiency and control performance. As a major field in the actual industrial production process, temperature control systems are widely used in industrial process control. The uncertainty and strong nonlinearity contained in this type of system make it a difficult problem in the control field to accurately and quickly give the optimal set value for complex industrial processes near the economic optimal state and enable the system to track it.

[0003] This patent studies the real-time optimization control of the set value of the ethylene cracking furnace outlet temperature control system. On the one hand, it is difficult to establish an accurate control system model. As the external conditions change during the control process, the control system will also change accordingly, which makes the control system have certain uncertainties, which may cause the overall system to not work near the economic optimal state or fail to give the current optimal set value. On the other hand, the control system often has nonlinear characteristics. It is impossible not to simplify the modeling of such a complex system. Therefore, there will inevitably be links that are not modeled. The dynamic characteristics of these unmodeled links may have a great impact on the system, which is called unmodeled dynamics in the control model established in this patent. When dealing with these unmodeled dynamics, if they are simply ignored, when modeling and controller design are performed under ideal conditions, the actual control often does not achieve the expected effect. Therefore, designing an advanced control set value real-time optimization method for such problems has become an inevitable requirement in the field of actual industrial process control.

[0004] At present, the mainstream set value optimal control method for the ethylene cracking furnace outlet temperature control system is real-time optimization control. However, with the increasing complexity of industrial process control objects and people's increasing requirements for control economic performance indicators, it is difficult for traditional real-time optimization methods to achieve ideal control effects for operating process control objects with uncertainty and strong nonlinearity. Therefore, it is of great significance to study equipment with such industrial characteristics and integrate advanced control algorithms into them for the research of industrial process control systems. Summary of the invention

[0005] In order to solve the above technical problems, this patent proposes a signal compensation robust prediction real-time optimization control method for the outlet temperature of an ethylene cracking furnace. This method adopts a combination of robust predictive control and signal compensation to address the uncertainty and strong nonlinearity of the control system, so as to better set the outlet temperature of the ethylene cracking furnace near the economically optimal steady state and control it within the required range.

[0006] In order to make the overall system give the optimal setting value in the current situation in real time near the economic optimal state and enable the system to track, it is necessary to consider and deal with the uncertainty and strong nonlinearity of the system during system modeling and controller design, which brings considerable challenges to system modeling and controller design. This method studies this problem. First, the ethylene cracking furnace outlet temperature control system model, that is, the controlled object simulation model, is established. On this basis, the lower control loop model is constructed according to the expansion latitude of the discrete state space model and the controller is designed: a conventional robust predictive controller is designed for the uncertainty part in the lower control loop model, and the uncertainty of the system is taken into account at the beginning of the design, so that the entire control system can still show the expected control performance when facing the uncertainty of the object parameters in actual control. Then, a linear performance index evaluation function is introduced and the upper operation control layer model is constructed based on the lower control loop model, and the controller is designed: a robust predictive controller is also designed for the uncertainty part in the upper operation control model, and then the unmodeled dynamic and data-driven method is used to virtualize the unmodeled dynamic data information of the upper operation control model at the previous moment, and the unmodeled dynamic compensation controller of the upper operation control model at the previous moment is designed to eliminate its influence on the closed-loop system. Finally, the upper-level operation control model does not model the dynamic incremental compensation controller to eliminate the influence of the rate of change of high-order nonlinear terms.

[0007] The method is implemented through the following technical solutions:

[0008] Step 1: Establish a bottom-level model of discrete state space with parameter uncertainty and unmodeled dynamics for the ethylene cracking furnace temperature control system and expand the lower-level loop control model based on the bottom-level model;

[0009] Set: x(k) = [Δy(k) Δu(k-1) Δu(k-2) Δ(k-3) y(k)], where x(k) represents the underlying system state at discrete time k, Δy(k) is the underlying system output increment of the system at discrete time k, Δu(k-1) is the underlying control input increment of the system at discrete time k-1, Δu(k-2) is the underlying control input increment of the system at discrete time k-2, Δu(k-3) is the underlying control input increment of the system at discrete time k-3, and y(k) represents the underlying system output at discrete time k. The discrete underlying state space model with parameter uncertainty and unmodeled dynamics is as follows:

[0010]

[0011] In the formula, x(k+1) represents the underlying system state at discrete time k+1, u(k) represents the underlying control input at discrete time k, and v(k) represents the unmodeled dynamics between the discretized underlying model and the original underlying model at discrete time k. A(k)=A+Δ a (k) is the underlying state matrix of the system at discrete time k, B(k) = B + Δ b (k) is the underlying input matrix of the discrete k-time system, A is the underlying state matrix to be identified, B is the underlying input matrix to be identified, C is the underlying output matrix, Δ a (k) is the uncertain perturbation of the underlying state matrix at discrete time k, Δ b (k) is the uncertainty perturbation of the underlying input matrix at discrete time k, satisfying:

[0012] [Δ a (k) Δ b (k)] = NΔ(k)[H a H b ] (2)

[0013] Where N is the unified matrix of the underlying system uncertainty, H is the state matrix of the underlying system uncertainty, and H b is the input matrix of the underlying system uncertainty, is the uncertain perturbation that depends on the discrete time k, where Δ1, Δ2 are random numbers between [-1, 1].

[0014] From formula (1), we know that v(k) = x(k+1)-A(k)x(k)-B(k)u(k), and x(k+1) is unknown at time k, so v(k) is expressed as follows:

[0015] v(k)=v(k-1)+Δv(k) (3)

[0016] Where v(k-1) is the unmodeled dynamics between the discretized underlying model and the original underlying model of the system at discrete time k-1, and Δv(k) is the unmodeled dynamic increment of the system at discrete time k.

[0017] Establish the extended loop control layer state space vector X u (k) = [x(k), E(k), e(k-1)] T , where X u (k) is the state space vector of the loop control layer at time k, e(k-1) is the relationship between the bottom system output y(k-1) and the set value y at time k-11 * (k-1), E(k) is the cumulative error from time k=0 to time k; the state space equation of the new extended loop control layer established at the same time is as follows:

[0018]

[0019] In the formula, is the system state of the loop control layer at discrete time k, K d1 is the control law gain of the controller at discrete time k in the loop control layer, is the system state input matrix of the loop control layer at discrete time k, C u =[C 0 0] is the system state output matrix of the loop control layer, is the unmodeled dynamics of the loop control layer at discrete time k, and I is the identity matrix of appropriate dimension.

[0020] Step 2: Design a robust predictive controller for signal compensation at the lower loop control layer;

[0021] First, the incremental conventional robust predictive controller u(k) is designed as follows:

[0022]

[0023] In the formula is the control law gain K of the robust predictive controller based on data-driven and unmodeled dynamics at discrete time k d1 The expansion matrix.

[0024] If there is a known scalar θ d >0, is the unknown positive definite symmetric matrix 1 of the loop control layer that needs to solve the linear matrix inequality (P is the corresponding dimension of the corresponding matrix), is the unknown positive definite symmetric matrix 2 of the loop control layer that needs to solve the linear matrix inequality, is the unknown matrix 3 of the loop control layer that needs to solve the linear matrix inequality, ε d1 >0 is the unknown scalar of the loop control layer that needs to be solved by the linear matrix inequality 1, εd2 >0 is the unknown scalar 2 of the loop control layer that needs to be solved by the linear matrix inequality, so that the following linear matrix inequality condition holds, thereby making the controlled system asymptotically stable.

[0025]

[0026]

[0027] in I is the identity matrix of appropriate dimension, Q d is the state compensation matrix of the loop control layer robust predictive controller, R d Input compensation matrix for the loop control layer robust predictive controller, is the unified matrix of the uncertainty of the extended loop control layer system, is the state matrix of the extended loop control layer system uncertainty, is the input matrix of the system uncertainty of the extended loop control layer, and * is the transpose of the elements in the symmetrical position.

[0028] Step 3: Expand and establish the upper operation optimization layer model in discrete state space with parameter uncertainty and unmodeled dynamics based on the lower control loop;

[0029] The upper operation control layer uses a larger sampling period T in one cycle. u Sampling, the lower loop control layer uses a smaller sampling period T in one cycle d To sample, where T u =aT d The sampling rate a is a positive integer. At the same time, the system state equation of the upper operation control layer considering the sampling period is established based on the expansion of the lower control loop as follows:

[0030]

[0031] In the formula, is the system state matrix of the operation control layer, A is the system state input matrix of the operation control layer. r (ak) is the base system state matrix of the operation control layer, B r The base system state input matrix for the operation control layer, is the system status of the operation control layer, where e u (ak) = r * -r(ak) is the performance error of the performance index evaluation function, r * is the ideal performance index of the evaluation function of the operation control layer, r(ak) is the actual performance of the performance evaluation function of the operation control layer, To run the virtual unmodeled dynamics of the control layer, is the benefit coefficient matrix of the positive benefit of the performance index evaluation function after the operation control layer is expanded, where M u =(M e -D e )[C 00]+N e [K d1 0 0],M u is the benefit coefficient matrix of the positive benefit of the evaluation function of the base performance index of the operation control layer, M e is the benefit coefficient matrix of the positive benefit of the basic performance index evaluation function of the operation control layer, N e is the benefit coefficient matrix of negative consumption of the basic performance index evaluation function of the operation control layer, D e It is the error benefit coefficient matrix of the basic performance indicator evaluation function of the operation control layer.

[0032] Step 4: Design a robust predictive controller for signal compensation at the upper operation control layer;

[0033] This method deeply integrates the advantages of feedback, feedforward, data-driven control and robust predictive control. The proposed signal compensation robust predictive controller is expressed as:

[0034] y * (ak)=u u1 (ak)+u u2 (ak)+u u3 (ak) (9)

[0035] where y * (ak) is the controller output of the operation control layer and also the set point input of the lower loop control layer, u u1 (ak) is the robust predictive controller of the operational control layer, u u2 (ak) is the unmodeled dynamic compensation controller at the previous moment of the running control layer, u u3 (ak) is the unmodeled dynamic incremental compensation controller of the operational control layer.

[0036] First, design a robust predictive controller u at the operational control layer u1 (ak) as follows:

[0037] u u1 (ak) = K u1 (ak)X u (ak) (10)

[0038] Where K u1 (ak) is the control law gain of the robust predictive controller at the discrete ak moment of the operating control layer.

[0039] If there is a known scalar θ u1 >0, is the unknown positive definite symmetric matrix 1 of the operation control layer that needs to solve the linear matrix inequality, is the unknown positive definite symmetric matrix 2 of the operation control layer that needs to solve the linear matrix inequality, is the unknown matrix 3 of the operation control layer that needs to solve the linear matrix inequality, ε u1 >0 is the unknown scalar 1 of the operation control layer that needs to solve the linear matrix inequality, ε u2 >0 is an unknown scalar 2 in the operation control layer that needs to be solved by the linear matrix inequality, so that the following linear matrix inequality condition holds, thereby making the controlled system asymptotically stable.

[0040]

[0041]

[0042] in I is the identity matrix of appropriate dimension, Q u is the state compensation matrix of the robust predictive controller of the operation control layer, R u Enter the compensation matrix for the operational control layer robust predictive controller, is the unified matrix of uncertainty of the operational control layer system, is the state matrix of the uncertainty of the operation control layer system, where the base state matrix of the uncertainty of the operation control layer system is Secondly, according to the virtual unmodeled dynamic data information of the previous moment of the operation control layer, the unmodeled dynamic compensation controller u of the previous moment of the operation control layer is designed. u2 (ak), that is:

[0043]

[0044] Where K u2 (ak) is the control law gain of the compensation controller at the discrete ak moment of the operation control layer, The dynamics are not modeled for the discrete ak-a moments of the operational control layer.

[0045] In order to perform dynamic and static compensation for the unmodeled dynamics of the control layer running at the previous moment, ensure that the controller u u2 (ak) control performance, the following performance indicators are introduced:

[0046]

[0047] Where z -1 is the backshift operator, P u (z -1 ) is X u (ak+a) about z -1 The weighted polynomial, Q u (z-1 ) is y * (ak) About z -1 The weighted polynomial of for About z -1 The weighted polynomial of -1 )=1-z -1 For about z -1 The backshift compensation parameter, u u3 (ak) is the unmodeled dynamic incremental compensation controller output at the discrete ak moment of the operating control layer.

[0048] Minimize J2 with the unmodeled dynamics of the previous moment The one-step optimal feedforward control law for compensation is:

[0049]

[0050] The dynamics of the previous moment are not modeled For dynamic and static compensation, the unmodeled dynamic The one-step optimal feedforward control law for compensation is:

[0051]

[0052] Design the unmodeled dynamic incremental compensation controller at the operational control level; define r * is the ideal performance index of the performance index evaluation function of the operation control layer, e u (ak) is the performance error of the performance index evaluation function at discrete ak moment, then:

[0053] e u (ak) = r * -r(ak) (17)

[0054] The following performance indicators are introduced:

[0055] J3=min[e u (ak+a)] 2 (18)

[0056] Where e u (ak+a) is the performance error of the performance indicator evaluation function at the discrete ak+a moment of the operation control layer.

[0057] At this time, the one-step optimal prediction of the performance error of the performance index evaluation function of the running control layer is e u (ak+a|ak) is:

[0058]

[0059] The unmodeled dynamic incremental compensation controller of the operating control layer is obtained as:

[0060]

[0061] In the formula, is the unmodeled dynamic increment of the system at the discrete ak-a moment of the operational control layer.

[0062] Compared with the prior art, the method has the following beneficial effects: a signal compensation robust prediction real-time optimization control method is proposed for the ethylene cracking furnace outlet temperature control system. On the one hand, based on the traditional real-time optimization method, the method combines the operating data information of the ethylene cracking furnace outlet temperature control system, regards the high-order nonlinear terms in the controlled object model as unmodeled dynamics, and designs a feedforward compensator based on the unmodeled dynamics, thereby reducing the conservatism of the traditional robust predictive control method in the system with model uncertainty and strong nonlinearity. On the other hand, unlike the traditional real-time optimization method, the method considers the influence of the unmodeled dynamics of the underlying model on the upper optimization setting value, and uses the signal compensation Robust predictive control method again in the upper model for processing, while also considering the difference between the nonlinear model and the linear model, combining the one-step optimal control idea, and effectively using the dynamic error information of the system to design a compensator that eliminates the error signal, thereby designing a nonlinear controller with fewer requirements, strong universality, and weak conservatism.

[0063] Instruction Manual

[0064] Figure 1 This is a simulation diagram of a random noise signal;

[0065] Figure 2 The simulation comparison diagram of the set point output tracking response between the proposed method and the traditional real-time optimization method is shown;

[0066] Figure 3 The comparison chart of the linear evaluation function tracking performance of the overall system performance index of the proposed method and the traditional real-time optimization method;

[0067] Figure 4 The simulation comparison diagram of the lower-level input response between the proposed method and the traditional real-time optimization method is shown;

[0068] Figure 5 A flowchart of the steps. DETAILED DESCRIPTION

[0069] For further explanation, a detailed description is given below in conjunction with the accompanying drawings and simulation examples.

[0070] Implementation Cases:

[0071] The present invention studies the ethylene cracking furnace outlet temperature control system to solve this problem. Through repeated tests, the state space model parameters of the ethylene cracking furnace outlet temperature control system are: A(k) = A + Δa (k), B(k)=B+Δ b (k), C = [00 0 0 1], [Δ a (k) Δ b (k)] = NΔ(k)[H a H b ], Where Δ1, Δ2, Δ3, Δ4, Δ5 are random numbers between [-1 1].

[0072] The estimated values ​​of the model parameters in equation (1) obtained by alternating recursive least squares and neural network identification are:

[0073]

[0074] The benefit coefficient matrix M of the positive benefit of the basic performance index evaluation function of the operation control layer e =100, the benefit coefficient matrix N of the negative consumption of the basic performance index evaluation function of the operation control layer e =-0.1, which is the error benefit coefficient matrix D of the basic performance index evaluation function of the operation control layer e =0.25, initial test setting value y of lower loop control layer * =25, performance index evaluation function ideal performance index r * =10, the lower loop sampling period T d =1, upper loop sampling period T u =5.

[0075] Since the controller parameters designed by this method will be optimized in a rolling manner as the system runs, and the controller parameters will only fluctuate within a small range, only the controller parameters when k = 100s are given here for reference, and the values ​​are as follows:

[0076]

[0077] Therefore, when k = 100s, the robust predictive controller outputs u u1 (k) = 849.1030, the output u of the dynamic compensation controller without modeling at the previous moment u2 (k) = 1.3873 × 10 -18 , the unmodeled dynamic incremental compensation controller output u u3 (k)=0.7101, signal compensated robust predictive controller u(k)=849.8131.

[0078] The present invention simulates the real-time optimization control system for the outlet temperature of the ethylene cracking furnace. The simulation adopts the traditional real-time optimization control method and the proposed signal compensation robust prediction real-time optimization control method. The set point output tracking response, the overall performance index linear evaluation function tracking performance and control effect, and the image of the input response of the proposed method are as follows: Figure 2 , 3 , as shown in Figure 4.

[0079] Depend on Figure 2 It is known that under the influence of the same external noise, the proposed method can better control the outlet temperature of the ethylene cracking furnace to the set value near the economic optimal state, while the traditional real-time optimization control method has several large fluctuations in the process of tracking the set value generated by the upper layer. The occurrence of these situations greatly reduces the controller accuracy of the outlet temperature of the ethylene cracking furnace.

[0080] Depend on Figure 3 It is known that the absolute value of the tracking performance of the linear evaluation function of the overall performance index of the traditional real-time optimization control method exceeds the specified error upper limit, and the time exceeding the upper limit accounts for 2.8% of the total simulation time. The proposed method controls the linear evaluation function of the overall performance index of the ethylene cracking furnace within the control requirement range, and the time exceeding the upper limit accounts for 0% of the total simulation time. The tracking performance of the system is improved.

[0081] Depend on Figure 4 It is known that the input response of the traditional real-time optimization control method fluctuates more when it approaches the final stable state than that of the proposed method.

[0082] In addition, the performance evaluation indicators mean square error (1MSE) and integrated 1 absolute 1 error (1IAE) as shown in equations (23) and (24) are used to compare the control effects. The results are shown in Table 1.

[0083]

[0084]

[0085] Table 1 Comparison of performance evaluation of tracking and controlling outlet temperature of ethylene cracking furnace

[0086]

[0087] As shown in Table 1, when the traditional real-time optimization control method is used, the MSE of the performance index of the ethylene cracking furnace is 9.6904, and when the proposed control method is used, the MSE of the outlet temperature of the ethylene cracking furnace is 4.5676, which is reduced by 52.8649% by the proposed method; when the traditional real-time optimization control method is used, the IAE of the outlet temperature of the ethylene cracking furnace is 257.3521, and when the proposed control method is used, the IAE of the outlet temperature of the ethylene cracking furnace is 172.6306, which is reduced by 32.9205% by the proposed method.

[0088] Table 2 Performance comparison of linear evaluation function for tracking control of ethylene cracking furnace

[0089]

[0090] It can be seen from Table 2 that when the traditional real-time optimization control method is used, the MSE of the performance index of the ethylene cracking furnace is 0.0218, and when the proposed control method is used, the MSE of the outlet temperature of the ethylene cracking furnace is 0.0196, which is reduced by 10.2590% by the proposed method; when the traditional real-time optimization control method is used, the IAE of the outlet temperature of the ethylene cracking furnace is 6.3815, and when the proposed control method is used, the IAE of the outlet temperature of the ethylene cracking furnace is 5.0287, which is reduced by 21.1983% by the proposed method.

[0091] In summary, the signal compensation robust prediction real-time optimization control method can be effectively applied to the ethylene cracking furnace outlet temperature control system. Through the analysis of simulation results, this method has a good control effect compared with the traditional real-time optimization control method. Therefore, the signal compensation robust prediction real-time optimization control method can better meet the actual control needs, and puts forward a new solution for the controller design of industrial control systems with uncertainty and strong nonlinearity. It has good application value and application prospects, and is of great significance to achieving the ultimate goal of my country's industry leading the global technology system.

Claims

1. A signal compensation robust prediction and real-time optimization control method for the outlet temperature of an ethylene cracking furnace is provided. The specific steps are as follows: Step 1: Establish a bottom-level model of discrete state space with parameter uncertainty and unmodeled dynamics for the ethylene cracking furnace temperature control system and a lower-level control loop model extended based on the bottom-level model; Where x(k+1) represents the underlying system state at discrete time k+1, u(k) represents the underlying control input at discrete time k, v(k) represents the unmodeled dynamics between the discretized underlying model and the original model at discrete time k, and A(k) = A+Δ a (k) is the underlying state matrix of the system at discrete time k, B(k) = B + Δ b (k) is the underlying input matrix of the system at discrete k time, A is the underlying state matrix to be identified, B is the underlying input matrix to be identified, C is the underlying output matrix, y(k) represents the underlying system output at discrete k time, Δ a (k) is the uncertain perturbation of the underlying state matrix at discrete time k, Δ b (k) is the uncertainty perturbation of the underlying input matrix at discrete time k, satisfying: [Δ a (k) Δ b (k)] = NΔ(k)[H a H b ] (2) Where N is the unified matrix of the underlying system uncertainty, H a is the state matrix of the underlying system uncertainty, H b is the input matrix of the underlying system uncertainty, is the uncertain perturbation that depends on the discrete time k, where Δ1, Δ2 are random numbers between [-1, 1]; Establish the extended loop control layer state space vector X u (k) = [x(k), E(k), e(k-1)] T , where X u (k) is the state space vector of the loop control layer at time k, e(k-1) is the relationship between the bottom system output y(k-1) and the set value y at time k-11 * (k-1), E(k) is the cumulative error of the bottom system output error e(k) from time k=0 to time k; the state space equation of the new extended loop control layer established at the same time is as follows: In the formula, is the system state of the loop control layer at discrete time k, K d1is the control law gain of the controller at discrete time k in the loop control layer, is the system state input matrix of the loop control layer at discrete time k, C u =[C 0 0] is the system state output matrix of the loop control layer, is the unmodeled dynamics of the loop control layer at discrete time k, and I is the identity matrix of appropriate dimension; Step 2: Design a robust predictive controller for signal compensation at the lower loop control layer; The signal compensation robust predictive controller u(k) in the loop control layer is expressed as: In the formula is the control law gain K of the signal compensation robust predictive controller at discrete time k d1 The expansion matrix of If there is a known scalar θ d >0, is the unknown positive definite symmetric matrix 1 of the loop control layer that needs to solve the linear matrix inequality (P is the dimension corresponding to the corresponding matrix), is the unknown positive definite symmetric matrix 2 of the loop control layer that needs to solve the linear matrix inequality, is the unknown matrix 3 of the loop control layer that needs to solve the linear matrix inequality, ε d1 >0 is the unknown scalar of the loop control layer that needs to be solved by the linear matrix inequality 1, ε d2 >0 is the unknown scalar 2 of the loop control layer that needs to be solved by the linear matrix inequality, so that the following linear matrix inequality conditions are established, and thus the controlled system is asymptotically stable; in I is the identity matrix of appropriate dimension, R is the state compensation matrix of the robust predictive controller for loop control layer signal compensation, d Input compensation matrix for loop control layer signal compensation robust predictive controller, is the unified matrix of the uncertainty of the extended loop control layer system, is the state matrix of the extended loop control layer system uncertainty, is the input matrix of the system uncertainty of the extended loop control layer, * is the transpose of the elements in the symmetrical position; Step 3: Expand and establish the upper operation optimization layer model in discrete state space with parameter uncertainty and unmodeled dynamics based on the lower control loop; The upper operation control layer uses a larger sampling period T in one cycle. u Sampling, the lower loop control layer uses a smaller sampling period T in one cycled To sample, where T u =aT d The sampling rate a is a positive integer. At the same time, the system state equation of the upper operation control layer considering the sampling period is established based on the expansion of the lower control loop as follows: In the formula, is the system state matrix of the operation control layer, A is the system state input matrix of the operation control layer. r (ak) is the base system state matrix of the operation control layer, B r The base system state input matrix for the operation control layer, is the system status of the operation control layer (where e u (ak) = r * -r(ak) is the performance error of the performance index evaluation function, r * is the ideal performance index of the evaluation function of the operation control layer, r(ak) is the actual performance of the performance evaluation function of the operation control layer, To run the virtual unmodeled dynamics of the control layer, is the benefit coefficient matrix of the positive benefit of the performance index evaluation function after the operation control layer is expanded, where M u =(M e -D e )[C 0 0]+N e [K d1 0 0],M u is the benefit coefficient matrix of the positive benefit of the evaluation function of the base performance index of the operation control layer, M e is the benefit coefficient matrix of the positive benefit of the basic performance index evaluation function of the operation control layer, N e is the benefit coefficient matrix of negative consumption of the basic performance index evaluation function of the operation control layer, D e The error benefit coefficient matrix of the basic performance index evaluation function of the operation control layer; Step 4: Design a robust predictive controller for upper layer operation optimization layer signal compensation; The signal compensation robust predictive controller of the operation control layer is expressed as: y * (ak)=u u1 (ak)+u u2 (ak)+u u3 (ak) (8) where y * (ak) is the controller output at the discrete time ak of the operation control layer and is also the set point input of the lower loop control layer. u1 (ak) is the robust predictive controller at discrete time ak of the operational control layer, uu2 (ak) is the unmodeled dynamic compensation controller at the previous discrete ak moment of the operating control layer, u u3 (ak) is the unmodeled dynamic incremental compensation controller at the discrete ak moment of the operating control layer; First, design a robust predictive controller u at the operational control layer u1 (ak) as follows: u u1 (ak) = K u1 (ak)X r (ak) (9) Where K u1 (ak) is the control law gain of the robust predictive controller at the discrete ak moment of the operating control layer; If there is a known scalar θ u1 >0, is the unknown positive definite symmetric matrix 1 of the operation control layer that needs to solve the linear matrix inequality, is the unknown positive definite symmetric matrix 2 of the operation control layer that needs to solve the linear matrix inequality, is the unknown matrix 3 of the operation control layer that needs to solve the linear matrix inequality, ε u1 >0 is the unknown scalar 1 of the operation control layer that needs to solve the linear matrix inequality, ε u2 >0 is the unknown scalar 2 of the operation control layer that needs to be solved by the linear matrix inequality, so that the following linear matrix inequality conditions are established, and thus the controlled system is asymptotically stable; in Q u is the state compensation matrix of the robust predictive controller of the operation control layer, R u Enter the compensation matrix for the operational control layer robust predictive controller, is the unified matrix of uncertainty of the operational control layer system, is the state matrix of the uncertainty of the operation control layer system, where the base state matrix of the uncertainty of the operation control layer system is Secondly, according to the virtual unmodeled dynamic data information of the previous moment of the operation control layer, the unmodeled dynamic compensation controller u of the previous moment of the operation control layer is designed. u2 (ak), that is: Where K u2 (ak) is the control law gain of the compensation controller at the discrete ak moment of the operation control layer, The dynamics are not modeled for the discrete ak-a moments of the operational control layer; In order to perform dynamic and static compensation for the unmodeled dynamics of the control layer running at the previous moment, ensure that the controller u u2 (ak) control performance, the following performance indicators are introduced: Where z -1 is the backshift operator, P u (z -1 ) is X r (ak+a) about z -1 The weighted polynomial, Q u (z -1 ) is y * (ak) About z -1 The weighted polynomial of for About z -1 The weighted polynomial of -1 )=1-z -1 For about z -1 The backshift compensation parameter, u u3 (ak) is the unmodeled dynamic incremental compensation controller output at discrete ak time of the operation control layer; Minimize J2 and obtain the unmodeled dynamics of the previous moment The one-step optimal feedforward control law for compensation is: The dynamics of the previous moment are not modeled For dynamic and static compensation, the unmodeled dynamic The one-step optimal feedforward control law for compensation is: Design the unmodeled dynamic incremental compensation controller at the operational control level; define r * is the ideal performance index of the performance index evaluation function of the operation control layer, e u (ak) is the performance error of the performance index evaluation function at discrete ak moment, then: e u (ak) = r * -r(ak) (16) The following performance indicators are introduced: J3=min[e u (ak+a)] 2 (17) Where e u (ak+a) is the performance error of the performance index evaluation function of the operation control layer at discrete time ak+a; At this time, the one-step optimal prediction of the performance error of the performance index evaluation function of the running control layer is e u (ak+a|ak) is: The unmodeled dynamic incremental compensation controller of the operating control layer is obtained as: In the formula, is the unmodeled dynamic increment of the system at the discrete ak-a moment of the operational control layer; By combining the designed robust predictive controller, the unmodeled dynamic compensation controller at the previous moment of the operating control layer and the unmodeled dynamic incremental compensation controller, a signal compensation robust predictive controller is finally formed.

Images