General dynamic matrix control system and control method based on data drive

Abstract

A data-driven general dynamic matrix control system and a control method thereof in the field of industrial process control technology include a sensor, a memory, and a prediction controller module, wherein the sensor is connected to a controlled object to detect an actual output signal, and the output end of the sensor is compared with the predicted output of a prediction model, and the error is transmitted to the prediction control module, the memory is connected to the prediction control module, and the stored historical data is sent to the prediction control module to solve the optimization problem, the prediction control module is connected to the memory to update the historical data, and is connected to the controlled object to output a control strategy, and the sensor is connected to an industrial computer via an analog input channel. The method of the present invention effectively simplifies system modeling, reduces costs, and improves real-time performance, while introducing data drive to ensure the control performance of the controller, and has strong practicality and application value.

Description

Technical Field

[0001] The invention relates to industrial process control, in particular to a data-driven general dynamic matrix control system and a control method. Background Art

[0002] Based on the general framework of predictive control, different model forms, optimization strategies and feedback methods will form different predictive control algorithms. Among them, the dynamic matrix control (hereinafter referred to as DMC) algorithm is an optimization control algorithm that uses the object step response as a prediction model and combines rolling optimization with feedback correction. Since the rise of predictive control, the algorithm has been widely used in industrial process control such as metallurgy and chemical industry due to its low complexity and strong robustness.

[0003] The DMC algorithm needs to establish a step response model, which takes a long modeling cycle, especially for systems with slow dynamics, which will lead to higher modeling costs. Therefore, how to simplify the modeling steps, shorten the modeling cycle, and effectively solve the errors caused by inaccurate models to establish a general dynamic matrix control method is a problem that needs to be solved. Summary of the invention

[0004] The purpose of the present invention is to propose a general dynamic matrix control system and control method based on data drive to simplify the modeling process, reduce costs, and use the idea of ​​data drive to effectively solve the impact of inaccurate modeling on control performance.

[0005] The present invention is implemented by the following scheme:

[0006] A general dynamic matrix control system based on data drive includes a sensor, a memory, and a prediction controller module, wherein: the sensor is connected to a controlled object to detect an actual output signal, and the output end of the sensor is different from the predicted output of a prediction model, and the error is transmitted to a prediction control module, the memory is connected to the prediction control module, and the stored historical data is sent to the prediction control module to solve an optimization problem, the prediction control module is connected to the memory to update the historical data, and is connected to the controlled object to output a control strategy, and the sensor is connected to an industrial computer via an analog input channel.

[0007] The present invention utilizes a control method based on a data-driven general dynamic matrix control system, which is characterized in that the method comprises the following steps:

[0008] 1) Establish a simplified prediction model: Determine the sampling value f of the unit step response of a simplified model of a system with similar dynamics i =f(iT), where T is the sampling period. Assume that when t N =NT, the step response will tend to be stable, so the model vector f = [f1 f2…fN ] T Describe the dynamic information of the simplified model. Set the prediction time domain to P, the control time domain to M, and the historical data length to N D , F is the step response coefficient f of the simplified model with similar dynamics i The dynamic matrix composed of the following is used to establish the prediction model in the industrial computer.

[0009]

[0010] Among them, Y PM (k) is the output prediction value at the next P moments, Y P0 (k) is the initial prediction value, Y P0 (ki) is the historical value of the stored initial prediction value, is the historical predicted output sequence after the actual output feedback correction stored, where the first subscript of the vector Y represents the number of predicted future outputs, and the second subscript represents the number of times the control amount changes, δ u,M (k) = [δ u (k)δ u (k+1)…δ u (k+M)] is the residual increment term of the control vector. P For Y P0 (ki), a vector function related to the adjustable historical data parameter a.

[0011] 2) The industrial computer issues a sampling command at time k according to the sampling period. The sensor detects the output variable y(k) of the controlled object. The detected signal is converted by A / D through the analog input channel and then transmitted to the industrial computer.

[0012] 3) The industrial computer solves the optimization problem based on the system output obtained from the second step of detection. The objective function of the optimization problem is selected as follows:

[0013]

[0014] The input and output satisfy the corresponding physical constraints and

[0015] w P (k)=[w(k+1)w(k+2)…w(k+P)],

[0016] Q=diag(q1,...q P ),R=diag(r1,...r M ).

[0017] The first step Y PM Substituting (k) into the objective function, solving the optimization problem yields the residual increment term of the control quantity at time k as δu (k), and the control increment is further obtained as

[0018]

[0019] Where a is an adjustable historical data parameter, and Δu(ki) is the historical control increment stored in the memory. Calculate the actual control amount u(k) = u(k-1) + Δu(k) at time k, apply u(k) to the controlled object, and store it in the memory to update the stored historical data for future solutions. At the next moment, replace k with k+1 to obtain u(k+1) acting on the object, and store it in the memory.

[0020] 4) Use real-time information for feedback correction: At time k, after adding an input with an amplitude of Δu(k) to the input end of the object, the output prediction value for the next N moments under its action is:

[0021]

[0022] Among them G N is the historical data Y N0 (ki)and And the vector function related to the adjustable historical data parameter a. Detect the actual output y(k+1) of the controlled object at time k+1 and the output at time k+1 predicted at time k Compare the output error Using e(k+1) to correct the prediction of future output where h=[h1 h2…h N ] T As the correction vector, The historical data stored in the memory is updated for future solutions. The initial prediction value at time k+1 can be obtained by shifting in

[0023]

[0024] is the shift matrix. Similarly, Y N0 (k+1) is stored in the memory to update the historical data stored, so that based on steps 1)-3) the optimization calculation at time k+1 is continued to obtain u(k+1) acting on the controlled object.

[0025] 5) Input the result of the optimization calculation into the controlled object, and select parameter a to adjust the control performance of the system: parameter a∈[0,1]. When a=0, the control result is the same as DMC. When a>0 and gradually increases, the control performance of the system gradually improves. However, when a is too large, jitter will occur. Therefore, the value of parameter a should be adjusted reasonably, that is, the value of a should be gradually increased within the range of [0,1]. When a better control effect is obtained, the current value can be selected. If jitter occurs, the maximum value of a with better control effect is taken.

[0026] The present invention proposes a general dynamic matrix control system and control method based on data-driven performance guarantee, which can effectively solve the problem of time-consuming DMC modeling. The introduced data-driven method can effectively guarantee the control performance of the system and has high practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present application. The illustrative embodiments of the present invention and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application.

[0028] Figure 1 is a control system block diagram of the present invention;

[0029] Figure 2 is a block diagram of the control method of the present invention;

[0030] Figure 3 is a response curve of the output y of the system 1 of the embodiment of the present invention;

[0031] Figure 4 is a response curve of the output y of the system 2 of the embodiment of the present invention; DETAILED DESCRIPTION

[0032] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0033] like Figure 1 As shown, the data-driven dynamic matrix control system of this embodiment includes a sensor, a memory, and a prediction controller module, wherein: the sensor is connected to the controlled object to detect the actual output signal, and the output end of the sensor is compared with the predicted output of the prediction model, and the error is transmitted to the prediction control module, the memory is connected to the prediction control module, and the stored historical data is sent to the prediction control module to solve the optimization problem, the prediction control module is connected to the memory to update the historical data, and is connected to the controlled object to output the control strategy, and the sensor is connected to the industrial computer via an analog input channel.

[0034] The control method of the present invention comprises the following steps:

[0035] For this embodiment, the prediction time domain of the predictive control is P=10, the control time domain is M=1, the sampling period is T=0.1, N=40, and the historical data length is N. D = 1, the N-dimensional feedback correction vector is h = [11…1] T .

[0036] This embodiment includes the following steps:

[0037] The first step is to establish a simplified prediction model: consider two first-order inertial controlled objects and give the corresponding dynamically similar simplified system models as follows:

[0038]

[0039] The model vectors of the two prediction models are determined as follows:

[0040]

[0041] Let the sampling period be T = 0.1, and take the first 40 sampling values ​​of the step response to form the model vector. From P = 10, M = 1, the prediction model step response matrix F1 is a vector composed of the first 10 components of f1, and F2 is a vector composed of the first 10 components of f2. The prediction model is established as follows:

[0042]

[0043] Among them, Y PM (k) is the output prediction value at the next P moments, Y P0 (k) is the initial prediction value, Y P0 (ki) is the historical value of the stored initial prediction value, is the historical predicted output sequence after the actual output feedback correction stored, where the first subscript of the vector Y represents the number of predicted future outputs, and the second subscript represents the number of times the control amount changes, δ u,M (k) = [δ u (k)δ u (k+1)…δ u (k+M)] is the residual increment of the control vector. According to the value range of parameter a, the vector function G P The parameter a in is taken as 0.5;

[0044] In the second step, the industrial computer issues a sampling command at time k according to the sampling cycle. The sensor detects the actual output variable of the controlled object (i.e., the industrial control system), and transmits the detected signal y(k) to the industrial computer after A / D conversion through the analog input channel.

[0045] Step 3: The industrial computer solves the optimization problem based on the actual system output y(k) obtained from the second step:

[0046] The objective function of the optimization problem is selected as follows:

[0047]

[0048] Among them, the input and output satisfy the corresponding physical constraints and

[0049] w P (k) = [1 1…1],

[0050] Q=diag(1,1,...1), R=diag(0.1,0.1,...0.1).

[0051] The first step Y PM Substitute (k) into the objective function and solve the optimization problem to obtain the residual increment term of the control quantity at time k as δ u (k), the control increment is further obtained as:

[0052]

[0053] Among them, the historical data parameter a is 0.5, and Δu(ki) is the historical control increment stored in the memory. Calculate the actual control amount at time k: u(k) = u(k-1) + Δu(k), apply u(k) to the controlled object, and store it in the memory to update the stored historical data for future solutions. At the next moment, replace k with k+1 to obtain u(k+1) acting on the object, and store it in the memory.

[0054] Step 4: Use real-time information for feedback correction:

[0055] At time k, after adding an input with an amplitude of Δu(k) to the input end of the object, the output prediction value for the next 40 moments under its action is:

[0056]

[0057] Among them, the vector function G N The parameter a in is taken as 0.5, and the actual output y(k+1) of the controlled object at time k+1 is detected, which is different from the output at time k+1 predicted at time k. Compare the output error: Using e(k+1) to correct the prediction of future output where h = [11…1] T As the correction vector, The historical data is stored in the memory and updated for future solutions. The initial prediction value at time k+1 is obtained by shifting in

[0058]

[0059] is the shift matrix. Similarly, Y N0 (k+1) is stored in the memory to update the historical data stored, so that based on steps 1)-3) the optimization calculation at time k+1 is continued to obtain u(k+1) acting on the controlled object.

[0060] Through the above steps, the data-driven general dynamic matrix control method of the present invention can use the data-driven method to update and feedback the historical data of the system in real time for a dynamically similar simplified model, thereby reducing the impact of model errors, and can change the proportion of historical data by adjusting the parameter a of the historical data, thereby improving the control performance of the system. Therefore, the method can effectively shorten the modeling cycle and achieve better control effects, which is of great significance for industrial process control systems.

[0061] Let the initial values ​​be zero and show the control results through the output response of the actual controlled object. Figure 3 is the output response of system 1, that is, the time constant of the actual controlled object is greater than the time constant of the prediction model (the response speed of the actual controlled object is slower). Figure 4 is the output response of system 2, that is, the time constant and gain of the actual controlled object and the prediction model are different (that is, the error between the object and the model is large). It can be seen from the two figures that when the response speed of the actual controlled object is relatively slow, the parameter a of the historical data is set to 0.5, and the application of the method of this embodiment is faster than the DMC response speed. When the error between the prediction model and the actual controlled object is large, the effect of the error on the control performance can still be effectively compensated by adjusting the parameters of the historical data. When a is 0.5, the application of the method of this embodiment is also faster than the DMC response speed. Therefore, the general dynamic matrix control system and control method based on data drive proposed in this embodiment can improve the system performance more effectively by adjusting the parameter a compared with DMC.

Claims

1. A general dynamic matrix control system based on data drive, characterized in that: It includes a sensor, a memory and a prediction controller module, wherein the sensor is connected to the controlled object, the other input end of the sensor is connected to the output end of the prediction model, the output end of the sensor is connected to the input end of the prediction control module, the prediction control module is connected to the memory, the output end of the prediction control module is respectively connected to the controlled object and the input end of the prediction control module, and the sensor is connected to the industrial computer via an analog input channel; The method comprises the following steps: 1) Establish a simplified prediction model: Determine the sampling value f of the unit step response of a simplified model of a dynamically similar system i =f(iT), where T is the sampling period. Assume that when t N =NT, the step response will tend to be stable, so the model vector f = [f1 f2…f N ] T Describe the dynamic information of the simplified model, set the prediction time domain to P, the control time domain to M, and the historical data length to N D , F is the step response coefficient f of the simplified model with similar dynamics i The dynamic matrix composed of the above mentioned prediction model is established in the industrial computer as follows: Among them, Y PM (k) is the output prediction value at the next P moments, Y P0 (k) is the initial predicted value stored, Y P0 (ki) is the historical value of the stored initial prediction value, is the historical predicted output sequence after actual output feedback correction, where the first subscript of the vector Y represents the number of predicted future outputs, and the second subscript represents the number of times the control amount changes, δ u,M (k) = [δ u (k)δ u (k+1)…δ u (k+M)] is the residual increment term of the control vector, G P For Y P0 (ki) is a vector function related to the adjustable historical data parameter a; 2) The industrial computer issues a sampling command at time k according to the sampling period, and the sensor detects the output variable y(k) of the controlled object. The detected signal is transmitted to the industrial computer after A / D conversion through the analog input channel; 3) The industrial computer solves the optimization problem based on the system output detected in step 2): The objective function of the optimization problem is selected as follows: Among them, the input and output satisfy the corresponding physical constraints and w P (k)=[w(k+1)w(k+2)…w(k+P)], Q=diag(q1,...q P ),R=diag(r1,...r M ). The Y in step 1) PM Substitute (k) into the objective function and solve the optimization problem to obtain the residual increment term of the control quantity at time k as δ u (k), the control increment is calculated as follows: Wherein, a is an adjustable historical data parameter, Δu(ki) is the historical control increment stored in the memory, the actual control amount u(k) at time k is calculated according to the formula u(k)=u(k-1)+Δu(k) and stored in the memory, the actual control amount u(k) at time k is applied to the controlled object, and the stored historical data is updated: At the next moment, k is replaced by k+1 to obtain u(k+1) to act on the controlled object and store it in the memory; 4) Use system real-time information for feedback correction: At time k, after adding an input with an amplitude of Δu(k) to the input end of the controlled object, the output prediction value at the next N moments under its action is: Among them, G N is the historical data Y N0 (ki)and and the vector function associated with parameter a, The sensor detects the actual output y(k+1) of the controlled object at time k+1 and inputs it into the prediction controller module. The prediction controller module compares the actual output y(k+1) at time k+1 with the output at time k+1 predicted at time k. Compare and get the output error: Use e(k+1) to correct the prediction of future output as follows: where h=[h1h2…h N ] T As the correction vector, storing the updated historical data in the memory; The initial prediction value at time k+1 obtained by shifting is: in is the shift matrix. Similarly, Y N0 (k+1) is stored in the memory to update the historical data stored, so that based on steps 1)-3) the optimization calculation at time k+1 is continued to obtain u(k+1) acting on the controlled object; 5) Input the result of the optimization calculation into the controlled object, and select parameter a to adjust the control performance of the system: parameter a∈[0,1]. When a=0, the control result is the same as DMC. When a>0 and gradually increases, the control performance of the system gradually improves. However, when a is too large, jitter will occur. Therefore, the value of parameter a should be adjusted reasonably, that is, the value of a should be gradually increased within the range of [0,1]. When a better control effect is obtained, the current value can be selected. If jitter occurs, the maximum value of a with better control effect is taken.

Images