JP2757039B2 - Self tuning controller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial Application Field> The present invention relates to a self-tuning controller for automatically adjusting at least proportional (P) and integral (I) operation parameters to optimal values. The present invention relates to a self-tuning controller having a process model in which a process to be controlled is modeled and determining P and I operation parameters by using a signal from the process model.

<Conventional Technology> As a process controller used for feedback control, a self-tuning controller that automatically tunes PI calculation parameters has been put to practical use.

The main self-tuning controllers proposed so far are as follows.

(1) Connect the auxiliary controller to the main controller in parallel, raise the gain of the auxiliary controller, cause vibration, and calculate PI from the amplitude and frequency based on the desired ZN limit sensitivity method by Ziegler and Nichols. Determining parameters (Showa 45, Transactions of the Society of Instrument and Control Engineers Vol.6
Research on adaptive control system using marginal sensitivity method, Toshiyuki Kitamori).

(2) A limit cycle is generated by using an on / off generator, and the optimum PI calculation parameter is determined from its amplitude and the like.
Proceedings of the 12th Academic Lecture Meetings P617-624 PID automatic setting adaptive controller Sumi, Fukuda).

(3) A pattern recognition means for observing the behavior of the control amount is provided. Here, without giving a disturbance to the process, a disturbance of the control system such as a disturbance generated at random is recognized, and a PI calculation parameter is determined from the recognition result. (US
PAT No. 4,602,326).

(4) Prepare a process model, change the parameters of the process model so as to minimize the error signal between the signal from the actual process and the signal from the process model, and determine the PI calculation parameter (ISA Transactions Vol. .2
2, No. 3 P. 50, 51, US PAT No. 4, 385, 362).

In the prior art described above, in the items (1) and (2), it is necessary to bring the process into an oscillating state or to forcibly apply a disturbance (identification signal) to the control system when determining the PI operation parameter. However, this has a problem that the process is considerably affected.

Item (3) is excellent in that it does not affect the process because it does not use an identification signal, but it is necessary to recognize a response pattern due to disturbance of the control system for a long time. Therefore, there is a problem that it takes time to determine the optimum PI calculation parameter.

Item (4) is excellent in eliminating these problems.

FIG. 5 is a configuration block diagram of a process control system disclosed in US Pat. No. 4,385,362.

In the figure, 1 is a control object (process), 2 is a deviation signal E between the process amount PV from the process 1 and the control target value SV.
PI control means for performing at least proportional (P) and integral (I) operations and outputting operation signals obtained to the process 1;
A signal from the control means 2 is a process model in which the process applied through the high-pass filter 41 is modeled and its parameters can be changed.

5 is a difference calculating means for calculating a difference between the signal from the process 1 having passed through the high-pass filter 42 and the signal from the process model 3, and 6 is searching for a parameter which minimizes the signal from the difference calculating means 5, and searching for the parameter. This is a model / parameter searching means for setting the obtained parameters as parameters of the process model 3.

The PI control unit 2 sets its PI calculation parameters based on the result searched by the model / parameter search unit 6.

<Problems to be Solved by the Invention> In the process control system having such a configuration, the signal applied to the process model 2 is
Since the signal from the PI control means has passed through, the DC component and noise are removed, and the operation using the process model can be performed relatively easily. However, it has the following disadvantages.

(A) It is desirable that the signal applied to the process model is originally the same as the signal applied to the actual process 1. However, the signal level becomes smaller and the waveform becomes different by passing through the high-pass filter. And the calculation accuracy in the process model decreases.

(B) Looking at the path of the signal reaching the difference calculation means 5, the signal from the PI calculation means 2 indicates that the signal from the real process
On the process model side, high-pass filter → process model → difference calculation means, whereas the order of the high-pass filter and the process model is interchanged. Therefore, in order for the model / parameter search means 6 to correctly perform the calculation for searching for the parameter that minimizes the difference signal from the difference calculation means, it is assumed that the process model 3 has a linear characteristic. That is, as a process model,
It is not possible to set upper and lower limits on the output or to use a non-linear output.

The present invention has been made in view of these problems,
An object of the present invention is to realize a self-tuning controller capable of reducing the amount of calculation in a process model and the amount of calculation for searching for a parameter.

It is another object of the present invention to provide a self-tuning controller capable of accurately performing an operation in a process model corresponding to an actual process and using a process model having a non-linear characteristic. .

<Means for Solving the Problems> According to the present invention for solving the above-described problems, at least proportional (P) and integral (I) calculations are performed on a deviation signal between a process amount from a control target (1) and a control target value. A PI control means (2) for performing and outputting the obtained operation signal to the control object;
A process model (3) configured to model the control target to which a signal from the control means is applied and its parameters are changeable, and a first band-pass filter (4) to which a signal from the process model is applied. And a second band-pass filter to which a signal from the control target is applied and which has the same characteristics as the first band-pass filter.
And a signal from the control target and a signal from the second band-pass filter are input, and it is determined whether the fluctuation of any of these signals is greater than a predetermined value for each identification cycle. The trigger generation means (6) for outputting a trigger signal if the above applies, and the signals from the first and second band pass filters are respectively inputted, and the first and second signals are received upon receiving the trigger signal from the trigger generation means. Search for a parameter that minimizes the difference between the signals from the two bandpass filters;
A model / parameter search means for changing the parameters of the process model based on the searched parameters; and a PI parameter of the PI control means based on a result searched by the model / parameter search means. And a PI parameter calculating means (8) for setting, wherein the process model, the first band-pass filter, the model / parameter searching means and the PI parameter calculating means are activated upon receiving a trigger signal from the trigger generating means. It is characterized by that.

<Operation> The first bandpass filter 4 removes stationary components, noise, and the like from the model output signal MO sent from the process model 3. The second bandpass filter 5 removes a stationary component, noise, and the like from the process amount PV signal of the control target 1 and has the same characteristics as the first bandpass filter 4 so that the process model and the control target 1 are removed. Improve the integrity.

The trigger generating means 6 receives a signal related to the process amount PV of the control target 1 and a signal from the second band-pass filter, and a variation of any of these signals is set to a predetermined value every identification cycle. Judgment is made as to whether or not it is larger, and if so, a trigger signal is output. If the fluctuation is smaller than a predetermined value, no trigger signal is generated.

The process model, the first bandpass filter, the model / parameter searching means, and the PI parameter calculating means are activated upon receiving a trigger signal from the trigger generating means. If the trigger signal is not received from the trigger generating means, the calculation is not performed, so that the amount of calculation as a whole is reduced.

<Example> Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a configuration block diagram of one embodiment of the present invention. In the figure, reference numeral 1 denotes a control object (process) whose dynamic characteristic changes due to a change in a production amount, a change in a control target value, a disturbance or the like. 2 is a PI control means for outputting to the process 1 an operation signal MV obtained by performing at least proportional (P) and integral (I) operations on a deviation signal DV between the process amount PV from the process 1 and the control target value SV. It is.

Reference numeral 3 denotes a process model which models the control target 1 to which the operation signal MV from the PI control means 2 is applied, and whose parameters can be changed.

The process model 3 receives an operation signal MV and performs an operation for obtaining a model output MO. This calculation differs depending on the process model.
If MOn, for example, the calculation is performed in accordance with the operation expression of Expression (1).

MOn = βm · MOn-1 + (1-βm) · Km · MV (n-Lm)
… (1) Km: gain of the process model Lm: dead time of the process model βm: first order delay coefficient of the process model MV (n−Lm): value of the MV before the time Lm 4 is the signal MO from the process model 3 Applied,
A first bandpass filter 5 for removing a stationary component, noise, and the like is a second bandpass filter to which a signal from the control target 1 is applied and which has the same characteristics as the first bandpass filter 4.

These first and second band-pass filters 4 and 5 are calculated by the equation (2), assuming that the time series data of the applied signal is represented by, for example, PVF and the stationary part is represented by PVo. .

PVF = (1−α) · PVF + α · (PV−PVo) (2) α: The filter constant 6 receives the signal PV from the control target 1 and the signal PVF from the second band-pass filter 5. When any of the signals change, we need to do various calculations,
This is a trigger generating means for outputting a trigger signal TRG as a calculation command.

7 is a signal M from the first and second bandpass filters 4 and 5
OF and PVF are input, respectively, and a trigger signal TRG from the trigger generating means 6 is received to search for a parameter that minimizes the difference between the signals from the first and second band-pass filters, and a process is performed based on the searched parameter. Model / parameter search means for making the parameters of the model 3 variable.

Reference numeral 8 denotes a PI parameter calculation unit that calculates and sets the PI parameter of the PI control unit 2 based on the result searched by the model / parameter search unit 7.

The trigger signal TRG from the trigger generating means 6 is
In addition to being applied to the parameter search means 7, it is also applied to the process model 3, the first bandpass filter 4, and the PI parameter calculation means 8 at the same time, and is configured to start calculation in response to a trigger signal. .

 The operation of the device configured as described above will be described below.

FIG. 2 is a flowchart showing a tuning operation of a PI calculation parameter started at a constant period.

The second band-pass filter 5 receives the process value PV from the process 1 and calculates the filtering as shown by the equation (2) (step 1).

The trigger generating means 6 reads the time-series data of the process value PV and the time-series data PVF of the output from the second band-pass filter 5 and monitors whether or not the fluctuation is larger than a predetermined value for each identification cycle. (Step 4).

If the fluctuation is smaller than a predetermined value, it is determined that the control is good, and the process ends without generating a trigger signal. Such a judgment is made as shown in FIG.
By integrating the deviation of the process amount PV and the output value PVF of the bandpass filter 5 from the steady state for a predetermined time,
The area (shaded area) is calculated, and the judgment is made based on the size.

If these values fluctuate greatly, a trigger signal TRG is output. Process model 3 receiving this trigger signal TRG
Reads the time-series data of the operation signal MV (the same signal as the operation signal output to the process 1) collected in the step 2, and calculates the model output according to the equation (1) (step 5).

The output MOF of the process model 3 is applied to the first band pass filter 4, where a filtering operation is performed according to the equation (2) (step 6). Through this filtering operation, stationary components, noise, and the like included in the model output MOF are removed.

The model parameter search means 7 receives the time-series data MOF from the first band-pass filter 4 and the time-series data PVF from the second band-pass filter 5, and searches for the parameters of the process model 3. Perform the operation.

This calculation is, for example, a calculation for obtaining the gain Km and the evaluation function CRIT represented by the equation (3).

Km = {MOF (n) * PVF (n)} / {MOF (n)} CRIT = 1- {MOF (n) * PVF (n)} 2 / {MOF
(N) 2 * {PVF (n) 2} (3) Then, an operation of searching for a combination of the time constant Tm and the model dead time Lm of the model that minimizes the evaluation function CRIT expressed by the equation (3) is performed.

That is, the output from the process model 3 is the process 1
The gain Km, the dead time Lm, and the first-order lag coefficient βm of the process model 3 are adjusted until the output of the process model sufficiently approaches (until the error of the process model becomes smaller than a predetermined value). repeat.

FIG. 4 shows an example in which the model parameter search means 7
The time series data of the output MOF from the first band-pass filter 4 and the output PVF from the second band-pass filter 5 is shown. The model parameter searching means 7 moves the output MOF from the first bandpass filter 4 closer to the output PVF from the second bandpass filter 5, that is, moves the output MOF indicated by the broken line in the direction of arrow A. Then, the gain Km, the dead time Lm, and the first-order lag coefficient βm of the process model 3 are adjusted so that the difference between the two becomes small.

If it is determined in step 8 that the model error has become smaller, that is, the output MOF from the first band-pass filter 4 is changed to the output PVF from the second band-pass filter 5,
When it is determined that the values substantially match with each other, the PI calculation parameter calculation means 8 calculates the proportional (PB), the integral (TI), and the derivative (TD) from the obtained parameters of the process model 3 by, for example, the Ziegier-Nichols method. The parameters are calculated according to equation (4), and the obtained calculation parameters are set in the PI control means 2 (steps 10 and 11).

(PB / Km) = a (Lm / Tm) 2 + b (Lm / Tm) + c (TI / Tm) = d (Lm / Tm) 2 + c (Lm / Tm) + fTD = 0.2 * TI (4) PI The control means 2 calculates the operation signal MV using the newly set calculation parameters.

It should be noted that the process model 3 does not necessarily have to be the first-order lag system represented by the equation (1). Process 1
When there is a measurable disturbance such as a load fluctuation, this disturbance may be used for the model calculation.

In addition, the calculation for obtaining the process model output is performed by repeatedly changing the values of the gain Km, the dead time Lm, and the primary delay coefficient βm, which are internal parameters, little by little. Here, for example, the gain Km
May be first obtained from the integration ratio of the output of the process model when the process amount and the process gain are set to “1”, and then the dead time Lm and the first-order lag coefficient βm may be obtained using an iterative search method.

As a repetitive search method for obtaining the dead time Lm and the primary delay coefficient βm, for example, the simplex method can be used. This method is described in, for example, “Nonlinear Programming” published by Nikkagiren, written by Hiroshi Konno and Hiroshi Yamashita.
It is described on pages 4 to 287.

Note that such an operation is performed only when the trigger signal TRG is issued from the trigger generating means 6, so that it does not take much operation time as a whole.

<Effects of the Invention> As described in detail above, according to the present invention, there are the following features as compared with the conventionally known control system shown in FIG.

(A) The same operation signal as the operation signal applied to the actual process 1 is simultaneously applied to the process model. Therefore, the accuracy of the model output calculation in the process model can be improved.

(B) The path of the signal from the PI control means 2 to the model parameter search means 7 is a process → hand path filter on both the real process side and the process model side. Can be provided, or one having a non-linear characteristic can be used.

(C) The calculation in the process model and the search for the parameter are performed in accordance with the trigger signal from the trigger generation means, and the amount of calculation as a whole can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention, FIG. 2 is a flowchart showing a tuning operation of a PI calculation parameter started at a constant period, FIG. 3 is a waveform diagram showing an operation of a trigger generating means, FIG. 4 is a waveform diagram showing the operation of the model parameter searching means, and FIG. 5 is a block diagram showing an example of a conventional apparatus. 1 Process 2 PI control means 3 Process model 4 5 Band pass filter 6 Trigger generation means 7 Model parameter search means 8 PI parameter calculation means

Claims (1)

(57) [Claims] The present invention relates to a deviation signal between a process amount from a control target (1) and a control target value, which is at least proportional (P) and integral (I).
PI control means (2) for performing an operation and outputting the obtained operation signal to the control object; and modeling the control object to which a signal from the PI control means is applied, so that parameters thereof can be changed. A process model (3), a first band-pass filter (4) to which a signal from the process model is applied, and a signal from the control target to apply the same characteristics as the first band-pass filter. A second band-pass filter (5) having a signal from a control target and a signal from the second band-pass filter, and a variation of any of these signals at a predetermined period is determined by a predetermined value. Judge whether it is greater than
A trigger generating means (6) for outputting a trigger signal if the magnitude is larger, and a signal from the first and second band-pass filters, respectively, and receiving a trigger signal from the trigger generating means, Model parameter searching means (7) for searching for a parameter that minimizes the difference between the signals from the second bandpass filter and varying the parameters of the process model based on the searched parameter. A PI parameter calculating means (8) for calculating and setting a PI parameter of the PI control means based on a result searched by the parameter searching means, wherein the process model, the first band-pass filter, Search means and PI parameter calculation means,
A self-tuning controller activated by receiving a trigger signal from the trigger generating means.