Double-event triggering model predictive control method and system for multi-working-condition industrial system

Abstract

The invention discloses a double-event triggering model prediction control method and system for a multi-working-condition industrial system, and the method comprises the steps: enabling a working condition change trigger to judge whether a new working condition occurs in the industrial system or not according to the data in an identification data buffer, and triggering the online updating of a prediction model if the new working condition occurs in the industrial system; the disturbance trigger judges whether a disturbance event triggering condition is achieved or not according to the data in the identification data buffer, and if the disturbance event triggering condition is achieved, optimization solution of the optimization controller is triggered; the optimization controller calculates a control sequence required by the industrial system in a next period of time based on the current prediction model; the control sequence buffer is used for storing the control sequence and the prediction output generated by the optimization controller. According to the method, the control precision can be improved, and meanwhile, the computing resource overhead is greatly saved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial control, and particularly relates to a dual-event-triggered model predictive control method and system for multi-condition industrial systems. Background Art

[0002] Process manufacturing is an important part of the manufacturing industry and is of great significance to economic and social development. The long-term stable operation of the process is an important guarantee for producing high-quality and low-cost products. Therefore, factories are committed to the control of key process indicators. However, most process manufacturing processes have strong nonlinearity and many constraints in the manufacturing process, and traditional control methods cannot handle the control problems with constraints. In this context, model predictive control (MPC) has been proposed by scholars and has received attention from the academic and industrial communities. MPC is a model-based closed-loop optimization control strategy that can be used to solve multi-variable control problems with constraints and has advantages such as good control effect and strong robustness. It has been widely applied to process manufacturing at present.

[0003] In order to improve the real-time performance of control, MPC is often deployed at the edge in practical applications. However, edge controllers often cannot meet the computing power requirements of MPC rolling optimization. With the progress of modern information technology, the powerful computing power of cloud computing and distributed computing makes it possible to apply complex control algorithms to industrial process control. However, both of these methods require additional equipment, and the transmission of real-time data during operation will consume a large amount of communication resources and energy. To address the above problems, some experts and scholars have adopted an event-triggered method to adjust the control strategy at the software level and combined it with MPC to introduce the event-triggered model predictive control (ET-MPC) strategy. This strategy can reduce the consumption of communication resources and computing resources without adding hardware costs.

[0004] Although the above method solves the problem of high computational cost of online optimization faced by traditional MPC to a certain extent, in the actual process manufacturing, due to changes in the operating environment, raw material components, as well as fluctuations in the external environment and production strategies, the equipment often operates under different working conditions during operation. Since the target operating point, stability, controllability, etc. of the system will change under different working conditions, when the above method is applied to a multi-condition industrial system, the prediction model is mismatched with the actual industrial system, and the control accuracy of the traditional event-triggered MPC will also be greatly reduced. Multi-model predictive control (MMPC) is considered an effective method to solve the multi-condition problem. However, this method can only achieve stable control of known working conditions. In fact, we may not be able to predict all operating conditions in the industrial process in advance, and MMPC will fail when unknown working conditions occur. Summary of the Invention

[0005] The present invention provides a dual-event-triggered model predictive control method and system for a multi-condition industrial system, which can improve the control accuracy while greatly saving the computational resource overhead.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] A dual-event-triggered model predictive control method for a multi-condition industrial system includes:

[0008] The working condition change trigger determines whether a new working condition occurs in the industrial system according to the output data of the prediction model and the actual output data of the industrial system. If so, it triggers the online update of the prediction model;

[0009] The disturbance trigger determines whether the disturbance event trigger condition is met according to the output data of the prediction model and the actual output data of the industrial system. If so, it triggers the optimization solution of the optimization controller;

[0010] The optimization controller calculates the control sequence required for the industrial system in the next period of time based on the current prediction model;

[0011] The control sequence is applied to the industrial system.

[0012] Further, the working condition change trigger determines whether a new working condition occurs in the industrial system, including:

[0013] (1) Filter the most recent actual output sequence Y(k) of the industrial system to obtain the sequence after disturbance weakening where k is the moment in the sequence, k = k0:k0 + LEN - 1, k0 is the starting moment of the sequence, and LEN is the detection window length;

[0014] (2) Calculate the prediction output sequence of the prediction model With the filtered sequence Error at each moment:

[0015]

[0016] Where n is the order of the industrial system output variable y, is the predicted output at time k, is the output obtained by filtering at time k;

[0017] (3) Statistical prediction output sequence The error exceeds the error threshold k th,2 The number of sample points number, if Greater than the threshold δ th , it is determined that a new operating condition has occurred in the industrial system.

[0018] Furthermore, the weighted mean filtering method is used to filter the actual output sequence Y(k), which can be expressed as:

[0019]

[0020] Among them, W L is the length of the sliding filter window, and W L =2w+1, LEN≥W L , α i is the weight coefficient of the filter window, satisfying

[0021] Furthermore, the subspace identification method is used to update the prediction model online, which is expressed as:

[0022]

[0023] Where k represents the sampling time, is the manipulated variable, is the controlled variable, is the process state variable; n y 、n u 、n x are the dimensions of controlled, manipulated, and state variables, respectively; are the system matrices of the prediction models respectively.

[0024] Furthermore, the disturbance event triggering conditions are:

[0025]

[0026] Among them, k i+1 is the time when the i+1th trigger event occurs, k iis the time of the i-th trigger event, k th,1 > 0 is the given error threshold parameter, N c is the control sequence length; means the minimum k when variable k satisfies k > k i and meets the condition exp in the following parentheses; min means to find the minimum value.

[0027] Furthermore, the optimization problem of the optimization controller for calculating the control sequence is:

[0028]

[0029] where Q Nc , Q, and R are respectively weighted positive matrices, R(k) is the given reference output sequence, and r(k) is the reference output at time k in the reference output sequence R(k); A, B, C, D are the system matrices of the prediction model, and ΔU(k) represents the increment sequence of the optimal control quantity of the industrial system obtained by optimization solution, is the prediction output sequence; N c is the control sequence length, represents predicting the value at time k + i with the value at time k as the initial condition, represents the increment of the control quantity at the (k + i)-th time obtained by solving with the actual state at time k as the initial state; The N in c represents the length of the prediction sequence obtained by a single solution; represents the initial state increment during the optimization solution; Δx(k) represents the actual state increment of the system at time k; represents the initial output increment during the optimization solution; Δy(k) represents the actual output increment of the system at time k; Δu min represents the minimum value constraint of the control input increment per unit time; Δu max represents the maximum value constraint of the control input increment per unit time; represents the state increment at the (k + i)-th time predicted with the actual state at time k as the initial condition; represents the output increment at the (k + i)-th time predicted with the actual output at time k as the initial condition.

[0030] A dual-event-triggered model predictive control system for multi-condition industrial systems includes: an identification data buffer, an identification data preprocessor, a working condition change trigger, a disturbance trigger, a prediction model identification module, an optimization controller, and a control sequence buffer;

[0031] The identification data buffer is used to collect the actual output data of the industrial system and the output data of the prediction model within a continuous period of time;

[0032] The identification data preprocessor is used to filter the actual output data of the industrial system;

[0033] The operating condition change trigger determines whether a new operating condition occurs in the industrial system according to the output data of the prediction model and the filtered data of the actual output of the industrial system. If so, it triggers the online update of the prediction model;

[0034] When triggered, the prediction model identification module performs an online update of the prediction model;

[0035] The disturbance trigger determines whether the disturbance event trigger condition is met according to the output data of the prediction model and the actual output data of the industrial system. If it is met, it triggers the optimization solution of the optimization controller;

[0036] The optimization controller calculates the control sequence required for the industrial system in the next period based on the current prediction model;

[0037] The control sequence buffer is used to store the control sequence and prediction output generated by the optimization controller.

[0038] Beneficial effects

[0039] The dual-event-triggered model predictive control method and system for multi-condition industrial systems of the present invention first constructs a metric for monitoring the operating state of an industrial system, and monitors the predicted output and actual output of the system in real time during system operation. When the prediction error is large, it accurately identifies the reasons for the degradation of the system control performance: model mismatch or disturbance. On this basis, for the control performance degradation problems caused by disturbances and operating condition changes respectively, an event trigger for disturbance suppression and an event trigger method based on operating condition re-identification are proposed, while ensuring the control accuracy and real-time performance of multi-condition industrial systems. Description of the drawings

[0040] Figure 1 is the framework of the dual-event-triggered model predictive control method in the embodiment of the present application;

[0041] Figure 2 An accurate monitoring mechanism for operating state based on weighted mean filtering;

[0042] Figure 3 An example of a disturbance event trigger mechanism;

[0043] Figure 4 is a comparison of the simulation output results of different control methods without noise: where (a) is the first-dimensional output y1; (b) is the second-dimensional output y2; (c) is the RMSE of the system output and the predicted output;

[0044] Figure 5Optimization triggering moments and intervals of different control methods without noise: where (a) represents the control method ET-MPC; (b) represents the dual-event triggering control method DET-MPC of the present invention.

[0045] Figure 6 Comparison of simulation output results of different control methods with noise: where (a) is the first-dimensional output y1; (b) is the second-dimensional output y2; (c) is the RMSE between the system output and the predicted output.

[0046] Figure 7 Optimization triggering moments and intervals of different control methods with noise: where (a) represents the control method ET-MPC; (b) represents the dual-event triggering control method DET-MPC of the present invention. Detailed implementation manner

[0047] The embodiments of the present invention will be described in detail below. Based on the technical solutions of the present invention, the detailed implementation manners and specific operation processes are given, and the technical solutions of the present invention are further explained and illustrated.

[0048] This embodiment provides a dual-event triggering model predictive control method for multi-condition industrial systems, which includes five parts: an identification data buffer, a condition change trigger, a disturbance trigger, an optimization controller, and a control sequence buffer. Specifically, the identification data buffer is used to collect the input and output data of the industrial process within a continuous period of time to prepare for the re-identification of the prediction model; the condition change trigger judges whether a new condition appears in the system according to the data in the identification data buffer, and if so, triggers the online update of the prediction model; the disturbance trigger is used to judge whether the disturbance event triggering condition is met, and if so, triggers the optimization solution of the optimization controller to calculate a new control sequence; the optimization controller can calculate the control sequence required for the actual process in the next period of time based on the prediction model; the control sequence buffer is used to store the control sequence and prediction output generated by the optimization controller. The five parts respectively solve problems such as data storage, model identification, and solution of optimization problems, and constitute the overall framework of the DET-MPC method, as Figure 1 shown, realizing the high-precision real-time control problem of multi-condition industrial systems.

[0049] I. Condition change trigger

[0050] When applying MPC in an industrial system, a prediction model needs to be obtained through mechanism-driven or data-driven methods to approximate the dynamic characteristics of the system. However, the various operating conditions and complex disturbances experienced by industrial systems during operation pose challenges to control based on fixed prediction models. Therefore, when the disturbance trigger identifies that the specified threshold is exceeded, the predictive control of the system may be in the model mismatch stage, or it may simply be due to excessive process noise, and the prediction model still matches the actual system. For the prediction model Equation (1) indicates that the excessive prediction error is caused by model mismatch, and Equation (2) indicates that the excessive prediction error is caused by disturbances.

[0051]

[0052] ψ(ω(k))>δ2 (2)

[0053] where ξ represents the prediction error caused by the difference between the prediction model and the actual system, ψ represents the prediction error caused by system disturbances, ω(k) represents the system disturbance at time k, and δ1 and δ2 respectively represent the error thresholds that make the control effect of the system unacceptable. In order to select appropriate control strategies for different operating conditions of the system, an embodiment of the present invention proposes an accurate operating state monitoring mechanism based on weighted average filtering. This mechanism involves real-time and historical data analysis to accurately identify changes in operating modes and disturbances within the system. The basic process of this mechanism is as Figure 2 shown.

[0054] Since the degradation of control performance can only be caused by disturbances and changes in operating conditions, if the prediction error remains at a high level after excluding disturbance factors, it indicates that the reason for the degradation of control performance is model mismatch. The weighted mean filter can adjust the weights according to the influence of different points on the filtered point to obtain the system output with weakened disturbance influence. Let the length of the sliding filter window be W L = 2w + 1, then the output after filtering is shown in Equation (3).

[0055]

[0056] where α i is the weight coefficient of the filter window, satisfying is the output obtained after filtering. Suppose there is currently data with a length of LEN for detecting changes in operating conditions (LEN ≥ W L ), the predicted output of the system can be obtained from the identification data buffer Calculate and the prediction errors corresponding to the outputs. If the proportion of the filtered prediction errors greater than k th,2 in the detection time window exceeds δ th, it indicates that the failure of the prediction model is the reason for the decline in control performance, and the system has now entered a new operating condition. The specific implementation steps are shown in Algorithm 1.

[0057]

[0058] In Algorithm 1, the initialization part includes setting the prediction error threshold k th,2 , the error overrun ratio δ th , the detection window length LEN, and the weighted filtering parameter W L and α. To achieve adaptive update after the change of operating condition, when it is recognized that the operating condition has changed, Algorithm 1 needs to generate a trigger signal for operating condition change TRIGG to update the prediction model of the optimization controller.

[0059] II. Prediction Model Update

[0060] When the operating condition change trigger recognizes that the current system operating condition has changed, it indicates that the original prediction model is no longer applicable and the prediction model needs to be adaptively updated. Algorithm 2 is the specific execution process of adaptive identification for model mismatch.

[0061]

[0062] In Algorithm 2, after the operating condition change detector recognizes that the system operating condition has changed, it will send a trigger signal to re-identify the prediction model according to the data U(k) and Y(k) in the identification data buffer, and send the new prediction model parameters A, B, C, D to the optimization controller. From the current moment on, the optimization controller can perform optimization and solution according to the new prediction model to obtain a more reliable control quantity.

[0063] In the embodiment of the present invention, the subspace identification method is adopted to update the prediction model, which can achieve high-precision system identification under small sample data, and the model obtained by the subspace identification method is in the form shown in formula (4).

[0064]

[0065] Where is the manipulated variable, is the controlled variable, is the process state variable. n y 、n u 、n x are the dimensions of the controlled, manipulated, and state variables respectively. are the system matrices of the state space model respectively.

[0066] III. Disturbance Trigger

[0067] When the working condition remains unchanged, the decline in control performance is largely due to the disturbances during the system operation. To address this issue, the present invention also adopts a disturbance event triggering mechanism. The optimization trigger sends an optimization trigger signal to the optimization controller only when the prediction error is too large or the control sequence in the prediction data buffer is used up, causing the generation of a new control sequence. The next trigger moment of the disturbance trigger can be determined by the condition shown in formula (5).

[0068]

[0069] where k th,1 > 0 is a given error threshold parameter, k i is the moment of the previous trigger event, n is the order of y, and inf represents. N c is the control sequence length; represents the minimum k when the variable k satisfies the condition k > k i and reaches the condition exp in the following parentheses; min represents finding the minimum value. Formula (5) indicates that starting from the moment of the previous trigger event, within the next N c time range, if the prediction error is greater than the threshold k th,1 when using the original control sequence for control, it means that the disturbance causes the state to deviate from the prediction trajectory, and a new control sequence needs to be solved; in addition, if the prediction error does not exceed the threshold for N c consecutive moments, the control sequence obtained from the previous trigger optimization is fully used up, and since there is no control amount to be used, a new control sequence also needs to be solved.

[0070] Figure 3 is an example of a disturbance event triggering mechanism. At the moment of k1 + N c , although the prediction error has not reached the set threshold k th,1 , the control sequence generated by the previous optimization has been used up, so a new control sequence needs to be solved. At the moment of k n+1 , the result obtained according to the prediction error calculation formula exceeds the threshold k th,1 , so the optimization solution is triggered.

[0071] IV. Optimization Controller

[0072] To enable the output of the system to track the reference trajectory as much as possible after triggering the preset event, the present invention designs the optimization problem shown in formula (6).

[0073]

[0074] where Q and R are weighted positive matrices respectively, R(k) is the given reference output sequence, and A, B, C, D are the system matrices of the prediction model. ΔU(k) represents the increment sequence of the optimal control quantity obtained by optimization and solution. is the predicted output sequence. In this patent, the optimization problem is solved by the sequential quadratic programming algorithm. Compared with traditional MPC, the addition of the disturbance event trigger mechanism enables the control sequence generated by MPC to be more fully utilized, effectively reducing the consumption of computing resources.

[0075] V. Detection and Experimental Verification Data

[0076] To verify the effectiveness of the DET-MPC method in control, a two-input two-output numerical simulation system is designed in this experiment for the performance test of the proposed scheme in the present invention. The specific form is shown in Equation (7).

[0077]

[0078] For the above system, the operating conditions of the system can be represented by the vector When the vector changes, it can be considered that the operating conditions of the system have changed. The operating conditions are set as follows:

[0079] Operating condition 1: c 1 = [0.8 -0.5 0.4 0.2 0.3 0.6 0.1], and the setpoint is y = [1 1] T

[0080] Operating condition 2: c 2 = [0.8 -0.025 0.4 0.2 0.375 0.6 0.1], and the setpoint is y = [2 2] T

[0081] First, under operating condition 1, a random input with a magnitude of u = [0.875 1.75] T ±20% is applied to the system, the output data of the system is collected, and through the subspace identification method, we can obtain the initial state-space model.

[0082] Set the simulation experiment length T = 500. In the first 50 steps, the system operates under operating condition 1, and after 50 steps, the system switches from operating condition 1 to operating condition 2. Let N c = N p = 10, the optimization trigger threshold k th,1 = 0.2, the model update trigger threshold k th,2 = 0.2, the maximum error detection length LEN1 = 15, and the maximum outlier length LEN2 = 2. With this setting, the ET-MPC and DET-MPC methods are respectively applied for control, and the control effects are shown in Figure 4 、 Figure 5 .

[0083] From Figure 4 the experimental results in, it can be seen that when ET-MPC is dealing with a large change in working conditions, due to the too large difference between the prediction model and the actual process, stable control can no longer be achieved. In contrast, the dual-event-triggered control method DET-MPC of the present invention can accurately identify the change in working conditions and adaptively update the prediction model, enabling the system to reach the desired output in a short time. Figure 5 It is a schematic diagram of the trigger interval when switching from working condition 1 to working condition 2 under the control of different methods. It can be seen from this that after the system working conditions change, the event trigger mechanism of ET-MPC completely fails. Even if the optimization controller performs optimization and solution control at each step, the system still cannot stabilize or reach the desired output. Under the control of the dual-event-triggered control method DET-MPC of the present invention, the trigger interval of the optimization trigger still remains equal to the control step size, making full use of the control sequence generated by the optimization controller, and enabling the system output to stabilize to the desired value. In addition, Table 1 gives the performance index comparison of different control methods under noise-free conditions.

[0084]

[0085] Table 2 gives the control results of numerical simulation when transitioning from working condition 1 to working condition 2 under different control step sizes. It can be seen that when the working conditions change significantly, regardless of the selected control step size, the overall prediction accuracy and control accuracy of the dual-event-triggered control method DET-MPC of the present invention are significantly improved. Compared with the ET-MPC method, the number of triggers is reduced by more than 70%.

[0086] Table 2 Numerical simulation control results under different control step sizes

[0087]

[0088] Since in the actual industrial production process, the system often operates in a noisy environment, in order to be closer to the actual process, Gaussian noise with a standard deviation of 0.02 and a mean of 0 is added in the experiment. In the first 300 steps, the system operates under working condition 1, and Gaussian noise with a standard deviation of 0.48 and a mean of 0 is superimposed at the 140th, 180th, 220th, and 260th steps to observe the performance of the ET-MPC and DET-MPC methods under large disturbances. After the 300th step, the system switches from working condition 1 to working condition 2, and other experimental parameters remain unchanged. The obtained control results are shown in Figure 6 、 Figure 7 .

[0089] From Figure 6 、 7It can be seen that after adding noise interference, both the ET-MPC and DET-MPC methods can achieve good control effects before the working condition switching. Although there are certain fluctuations in the system, it can still be stabilized near the set value. Despite the presence of disturbances in the system, the disturbance event triggering mechanism can avoid the accumulation of prediction errors by re-performing MPC optimization. After the working condition switching, the ET-MPC cannot adjust the prediction model according to the change of the working condition, and the triggering mechanism completely fails. In contrast, the dual-event triggering control method DET-MPC of the present invention can accurately identify that the system switches from working condition 1 to working condition 2, and adjusts the prediction model to quickly stabilize the system output to the set point. Thanks to the higher prediction accuracy, the number of optimization solutions under the dual-event triggering control method DET-MPC of the present invention is less.

[0090] The dual-trigger model predictive control method of the present invention can be applied to the control of multi-condition industrial processes, and has significant effects in improving the prediction accuracy, control accuracy, and reducing the computational amount during online operation. Specifically, compared with the traditional event-triggered model predictive control method, the prediction accuracy and control accuracy of the technology of the present invention during online operation are respectively increased by 82.9% and 78.9%, and at the same time, the number of MPC solutions is saved by 77.7%.

[0091] The above embodiments are the preferred embodiments of the present application. Those of ordinary skill in the art can also make various transformations or improvements based on this. Without departing from the general concept of the present application, these transformations or improvements should all fall within the scope of protection required by the present application.

Claims

1. A dual-event-triggered model predictive control method for multi-operation industrial systems, characterized in that, Including: The operating condition change trigger determines whether a new operating condition occurs in the industrial system according to the output data of the prediction model and the actual output data of the industrial system. If so, it triggers the online update of the prediction model; The disturbance trigger determines whether the disturbance event trigger condition is met according to the output data of the prediction model and the actual output data of the industrial system. If it is met, it triggers the optimization solution of the optimization controller; The optimization controller calculates the control sequence required for the industrial system in the next period of time based on the current prediction model; Apply the control sequence to the industrial system.

2. The dual-event-triggered model predictive control method for multi-condition industrial systems according to claim 1, wherein The operating condition change trigger determines whether a new operating condition occurs in the industrial system, including: (1) Filter the most recent actual output sequence Y(k) of the industrial system to obtain a sequence with weakened disturbances. Where k is the time in the sequence, k = k0:k0+LEN-1, k0 is the starting time of the sequence, and LEN is the length of the detection window. (2) Calculate the predicted output sequence of the prediction model and the filtered sequence The error Error at each moment: where n is the order of the output variable y of the industrial system, is the predicted output at time k, is the output obtained by filtering at time k; (3) Statistical prediction output sequence The number of sample points where the mean error exceeds the error threshold k th,2 , if is greater than the threshold δ th , it is determined that a new working condition has occurred in the industrial system.

3. The dual-event-triggered model predictive control method for multi-condition industrial systems according to claim 2, wherein Filter the actual output sequence Y(k) using the weighted mean filtering method, expressed as: Among them, W L is the length of the sliding filter window, and W L = 2w + 1, LEN ≥ W L , α i is the weight coefficient of the filter window, satisfying 4. The dual-event-triggered model predictive control method for multi-operating-condition industrial systems according to claim 1, wherein Update the prediction model online using the subspace identification method, expressed as: where k represents the sampling time, is the manipulated variable, is the controlled variable, is the process state variable; n y and n u and n x are the dimensions of the controlled, manipulated, and state variables, respectively; are the system matrices of the prediction model, respectively.

5. The dual-event-triggered model predictive control method for multi-operation industrial systems according to claim 1, wherein The disturbance event trigger condition is: where k i+1 is the time of the (i + 1)-th triggering event, k i is the time of the i-th triggering event, k th,1 > 0 is a given error threshold parameter, N c is the control sequence length; denotes the minimum k for variable k to satisfy the condition exp in the following parentheses when k > k i ; min represents finding the minimum value.

6. The dual-event-triggered model predictive control method for multi-operating-condition industrial systems according to claim 1, characterized in that The optimization problem for the optimization controller to calculate the control sequence is: Among them, Q and R are respectively weighted positive matrices, R(k) is a given reference output sequence, and r(k) is the reference output at the k-th moment in the reference output sequence R(k); A, B, C, and D are system matrices of the prediction model, and ΔU(k) represents the incremental sequence of the optimal control quantity of the industrial system obtained by optimization and solution. is the predicted output sequence; N c is the control sequence length, represents predicting the value at the (k + i)-th moment with the value at the k-th moment as the initial condition, represents the increment of the control quantity at the (k + i)-th moment obtained by solving with the actual state at the k-th moment as the initial state; The N in c represents the length of the prediction sequence obtained by a single solution; represents the initial state increment during optimization and solution; Δx(k) represents the actual state increment of the system at the k-th moment; represents the initial output increment during optimization and solution; Δy(k) represents the actual output increment of the system at the k-th moment; Δu min represents the minimum value constraint of the control input increment per unit time; Δu max represents the maximum value constraint of the control input increment per unit time; represents the state increment at the (k + i)-th moment predicted with the actual state at the k-th moment as the initial condition; represents the output increment at the (k + i)-th moment predicted with the actual output at the k-th moment as the initial condition.

7. A dual-event-triggered model predictive control system for multi-condition industrial systems, characterized in that, Including: An identification data buffer, an identification data pre-processor, an operating condition change trigger, a disturbance trigger, a prediction model identification module, an optimization controller, and a control sequence buffer; The identification data buffer is used to collect the actual output data of the industrial system and the output data of the prediction model within a continuous period of time; The identification data pre-processor is used to filter the actual output data of the industrial system; The operating condition change trigger determines whether a new operating condition occurs in the industrial system according to the output data of the prediction model and the filtered actual output data of the industrial system. If so, it triggers the online update of the prediction model; When triggered, the prediction model identification module updates the prediction model online; The disturbance trigger determines whether the disturbance event trigger condition is met according to the output data of the prediction model and the actual output data of the industrial system. If it is met, it triggers the optimization solution of the optimization controller; The optimization controller calculates the control sequence required for the industrial system in the next period of time based on the current prediction model; The control sequence buffer is used to store the control sequence and prediction output generated by the optimization controller.

Images