JPS63191202A - Batch process control system - Google Patents

Abstract

PURPOSE:To improve the control performance as the processing frequency increases by providing a learning part which stores the control information obtained by the preceding processing and converts this control information so that it can be utilized for the next processing and carrying out the next processing based on said converted information. CONSTITUTION:The manipulated variable u(k) obtained in the preceding processing is stored and then filtered for production of uf(k) which is used as the reference value of the current manipulated variable. here an allowance range DELTAu of the current manipulated variable u(t) is set against the value uf(k). Then the manipulated variable u(t) calculated with use of the preceding parameter theta(t+dtau) is corrected so that the current manipulated variable is set within a range uf(t)+ or -DELTAu. The range DELTAu can be varied in accordance with the value of the deviation between the current control amount y(t) and the target value r(t). That is, the range DELTAu is reduced as said deviation is decreased. Thus it is possible to suppress the vibration of the variable u(t) as well as the vibration of controlled variable y(t). As a result, the batch process control performance is improved as the processing frequency increases.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an autotuning method for process control, and particularly to a control method for a batch process in which the same process is repeated.

[Conventional technology]

Conventionally, auto-tuning methods for process control have been designed for one-time processing, even in continuous processes or batch processes. There was no way to do it with the information.

[Problem that the invention seeks to solve]

Among conventional auto-tuning methods, there is a method in which processes are sequentially identified and controlled based on input/output information of manipulated variables and controlled variables. This is an effective method when the characteristics of the process are nonlinear and time-varying, and constant changes in the process state are expected. However, with this method, process identification inevitably takes some time immediately after process identification begins or when process characteristics suddenly change.

There were problems such as deterioration of control performance.

The purpose of the present invention is to eliminate the deterioration of controllability at the start of identification and at parts where process characteristics suddenly change in a lifting platform 2 controlled by an auto-tuning method for batch processes by repeating the same process, which is a characteristic of batch processes. The aim is to improve control performance in areas where controllability deteriorates by paying attention to the information from the previous process.

[Means for solving problems]

In order to achieve the above purpose, the information from the previous processing will be accumulated and utilized as follows.

First, the least squares method is generally used as a method for estimating the estimated parameter 0, which is sequentially estimated by the identification unit and sent to the manipulated variable calculation unit. The estimated value of the unknown parameter when can be described with i=1 is...
(4) Z is an observation vector expressed by equation (5).

z”(t)=(y(t-t)ty(at t-2)t-w
y(t-n τ)tu(t-d t), u(t-t-
d t), -, u(t-ms - d t))...(5
) Further, ρ is a forgetting coefficient that represents the proportion of past input/output information contributing to the estimated value, and P is called an adaptive gain that represents the relationship between input and output information.

On the other hand, the manipulated variable calculation section takes into account the process delay time d so that the controlled variable y(t+dτ) matches the target value r(t+dτ), that is, r(t+d t) y(t+d t)=O" (6) Operation M using θ estimated by the identification part so that 8 is satisfied.
Determine u (t). Concretely, from the formula (3), y(t+dτ) is as follows: y(t+d t)=-aly(t+d t-v)-az
y(t+d τ-2τ)---any(t+dx-
nt)+bou(t;)+btu(t-τ)ten...+
b, u(t-mτ) - (7), and the equation for calculating u (k) from equations (6) and (7) is as follows.

■ u(t)=(r(t+d τ)+a1y(t
+d v-t)+-・-b.

+an3'(t+d f-n y)-btu(t-y)
--bmu(t-mτ)) (8) Now, what should be noted here is the difference in input and output information between the identification section and the manipulated variable calculation section. In the manipulated variable calculation section,
The input/output information used to determine u (k) is (8
) from the formula (y(10dt), -, y(t+d τ-nt), u(
t+t), -u(t-mτ))...(9) Since 8, the estimated parameter 0 used here is (9)
Adding u (t) to the equation, (y(t+cl?)+・-・ty(t+df−n
t) * u (t), -u (at t-m)) = (
t0). However, the 0 actually sent from the identification Δ section is (y(t)+"'ty(t-n t), u(t-d t
)+...'u(t-dt-mτ))...(t1) is used. Therefore, in the present invention, in order to eliminate this time difference in 0, the 0 estimated at each time in the previous process is saved as data, and in the next process, the manipulated variable u (t) at time t is The calculation formula uses (?(t+dτ)) of the previous time t+dτ.As a result, if the process status is the same with respect to the time axis between the previous and next processing, there will be a time shift of △ 0. is eliminated and control performance is improved.

In this case, the manipulated variable u (t) at time t changes in value between the previous process and the next process, and naturally the controlled variable y(t) also changes. Therefore, it is possible that the state of the process is not the same between the previous and next time, but is time-shifted, so in order to bring Δ to the desired 0 through the above process, this time shift must be corrected. There is a need. For this reason, as will be shown in later examples, it is known from experience that the speed at which the process state changes depends on the size of control bty (t), so the difference between the previous and current control Ity (t) is seek,
Based on the difference, the time correction amount is read from a table created based on experience, and the time axis is corrected.

Next, a process for suppressing fluctuations in the control amount that occur in areas where identification is incomplete in the conventional method will be described.

First, accumulate the manipulated variable u (k) from the previous processing, apply a filter to it to create uf(k), and use this as the reference value for the current manipulated variable. Set the allowable range Δu of the current operation amount u (t) for the current operation amount u (t),
The previous value of 0(t+dτ) is adjusted so that the current △ operation amount is within the range of uf(t)±Δu.
Correct the operation 'Itu (t) calculated using . The allowable range Δu of the manipulated variable is made to change depending on the magnitude of the deviation between the current controlled variable y (t) and the target value r (t),
The smaller the deviation, the smaller the allowable range. Thereby, vibrations in the manipulated variable u (t) can be suppressed, and thereby vibrations in the controlled variable y (t) can be suppressed.

According to the processing described above, control performance improves as the number of processing increases.

(Effect) First, to determine the manipulated variable u (t), use θ(t+d
In the method using τ), if the process states in the previous and current processing are the same 8 with respect to the time axis, a desirable 0 is used to determine the operation Jtu(t). As a result, the control performance should improve compared to the previous processing, and will not deteriorate.

In addition, in the method of correcting the time axis when there is a time axis deviation between the previous and current process status, it is an empirical process and it is not possible to exactly match the time axis, but the time axis The deviation can be improved to some extent.

In the method of suppressing the vibration of the controlled variable y (t), the vibration of the controlled variable y (t) is suppressed by the manipulated variable u (t
) by suppressing the vibration of the controller, and since the filtered value of the previous manipulated variable is used as the reference value for the manipulated variable, operations M that would extremely disturb the control are avoided.
u (t) is never added to the process.

〔Example〕

An embodiment of the present invention will be described below.

FIG. 1 is a block diagram of a control device to which the present invention is applied. In FIG. 1, the control device 5 receives a control signal y
Identification unit 3 identifies the pulse transfer function of process 1 from each time series data of (t) and operation signal u (t) and the deviation e (t) between control target value r (t) and control signal y (t)
A manipulated variable calculation unit 2 generates an operation signal u (t) from the eight coefficients 0(t) identified by the identification unit 3, and stores various information from the previous processing for the batch process. It is comprised of a learning section 4 that processes the multiplied information and thereby assists in controlling the next patch process.

Process 1 to be controlled in this example used the polyvinyl chloride polymerization reaction (PVC) process shown in FIG. In this process, the temperature begins to rise, and when a certain temperature is reached, rapid self-heating occurs. It has highly nonlinear and time-varying characteristics, with the heat generation decreasing rapidly as the reaction approaches its end.

FIG. 2 shows a process for producing polyvinyl chloride (used in water and sewage pipes, window frames, etc.) by subjecting the monomers in the reactor 19 to a multilayer reaction. The monomer does not start reacting at room temperature, but in the initial stage of the reaction, the reaction starts by absorbing heat from the jacket 14. Hot water or cold water flows into the jacket 14 from the upper right of the 21g reaction can and flows out from the lower left.

Once the monomers start reacting, they will self-heat and the reaction will proceed rapidly, so cold water must be poured into the jacket to maintain the temperature of each bowl at a constant value. It will be necessary to adjust the temperature and amount of water.

Such control is essential because madder-quality polyvinyl is produced when the reaction humidity is kept at a constant value. Further, the reactant is stirred by a motor 11 to ensure uniformity of the product.

In the mixing tank 15, hot water 17 and cold water 18 can be mixed and allowed to flow into the jacket. By operating the valve 16 that regulates the amount of hot water, the jacket inflow temperature and amount can be adjusted.

The control device 5 measures the temperature y (t) inside the reactor obtained from the temperature sensor 13, and operates the valve 16 to supply hot water fi.
By adjusting ku (t), the temperature inside the reactor is controlled to be kept constant.

Identification unit 3 in the control device 5. Learning part 4. Operation amount calculation section 2
will be explained along the operating procedure.

The polymer produced by the batch reaction process is transferred to another step, and then the same process is repeated in which monomers are placed in a reaction vessel and reacted. Here, the number of repetitions is expressed as 0th time, and the time during one cycle of batch processing is generally expressed as time t.

The procedure of batch reaction processing during one cycle will be explained. As shown in block 71 of FIG. 7, the identification section first performs
An observation vector zT(t) is created (also expressed in equation (5) above). Here, y is the output from the process and U is the manipulated variable. Next, the adaptive gain P (t) is expressed as (2)
In addition, the estimated parameter Δ vector 0(t) is determined as shown in the equation (block 72 in FIG. 7) in the case of the first batch processing. However, after the 211th time, the estimated parameters are calculated as in block 81 of FIG. 8 or block 91 of FIG. 9, as will be described later. Here, the learning section stores the Δ parameter θm-x(t) estimated for the previous batch (Q-1) at all times t in one cycle.
is remembered about.

Next, in the manipulated variable calculating section, the manipulated variable u (t) is calculated as shown in equation (8) above. In the first batch, (8
) can be generalized and expressed as follows.

Δ ux(t)=f (0t(tL y(t), Ll(t)
)−・(t3) Here, uz(t) is the manipulated variable at the 7th time of the first batch, the function f is in the form of equation (8), and y(t) is the time change from 10dτ−nτ to t+dτ− Observations up to τ,
ul(t) is the manipulated variable from time t-mτ to t-τ.

In the case of batch processing from the second time onward, as described later,
In the method of using Δ 0□−, (t+dτ) from the previous processing to determine the operation tuD), the result is as follows.

8ua(j)=f(On-z(t+d'tLy
m(t), un(t))...(t4) Furthermore, in the second and subsequent cases, the method of correcting the time axis deviation of the process state is as follows.

8us(t)=f(θm−t(t+dτ0dτ)
, ym(t)+um(t))...(t5) Here, α will be described later.

The above-mentioned control device 5 is used for this process 1.

The effects of control according to the present invention are shown below.

First, a method of using 80a-t(t+dτ) from the previous processing to determine the manipulated variable u(t) will be described.

The calculation of the manipulated variable ur(t) during the first processing is performed using (Is(t)) estimated by the identification unit 3 using the conventional method. is the previous θm−t(t+
d) is used. In addition, in the identification unit 3, from the second time onwards when calculating θ(t) using equation (t)8, instead of using the previous a m-s (-τ) for the τ time, the learning unit 4 θm, −n(
t) is used. This corresponds to replacing block 73 in FIG. 7 with block 81 in FIG. 8. The results of control using this method are shown in FIG.

In this example, since the control performance of the start-up is markedly improved, Fig. 3 shows the part of the heat generation process from the initial temperature increase of the entire treatment process, that is, the temperature of the PVC process is maintained constant. Showing the control performance when controlling as follows, (a) is the first time, (b) is the second time, and (c) is the fifth time. As can be seen by comparing them, the second time using the method according to the present invention has a better start-up than the first time using the conventional method, and the fifth time is better than the second time, and the control performance is improved. I can see that.

Next, compensate for the time axis difference between the previous and current broth states iE
This section explains how to do this. In the case of the PvC process of this example, it is known that the higher the temperature of the reactant, which is the control amount, the faster the process state changes, and conversely, the lower the temperature, the slower the change. Therefore, find the difference between the control ftky (t) at time t between the previous time and this time, prepare a table like the one shown in Figure 6, and use this to advance the time axis if it is higher this time, or delay it if it is lower. In addition, the amount to be corrected is determined based on the magnitude of the difference in control amount.

Previous and next control amount y! (t), yz(t)
When comparing , we are considering the case where they are oscillatory, so we will compare the following filtered observed values.

yz(t)=hly(t)+(hz-hx)y(t-τ
)+()ta-hl)y(t-2τ)+(t,0-h
a) y (t −3τ) ...(t
6) Here, hl"ha calculates the difference after applying a filter like the method of constants less than or equal to 1.

yzt(t) yz(t) ”'
(t7) This value is the temperature difference and corresponds to the first column in FIG. Based on the table in FIG. 6, 91 in FIG. 9 and α in equation (t5) are determined as follows. here.

When yzl(t) - yz*(t) continues to be positive or negative, if the values shown in Figure 6 are added together and the value becomes ``1'' or more, change 1 time t to τ time. For example, van(t) y xz(t
) is positive and 2 degrees occurs twice in a row, then 0.67+0.67=1.34...(t
8), and one time can be found as α mentioned above.

The results using this method are shown in Figure 4.
As in the figure, the part from temperature rise to exothermic process.

FIG. 7 is a diagram showing control performance when using a method of correcting the time axis deviation between the previous and current process states.

(a) is the second time of processing, and (b) is the fifth time. Compared to the case where the time axis in Figure 3 was not corrected,
It can be seen that the control performance improved in both the second and fifth times, indicating that it was effective.

Finally, a method for suppressing oscillations when the control amount has an oscillatory behavior will be explained. uz(t), which is the reference value for the current operation range, is determined from the previous operation amount stored in the learning section using the following equation.

un(t)=htu(t)+(ha-ht)X(u (
t − τ) + u (t + τ))/ 2+・=”・
+(t,0-hn)X(u(t-n t)+u(t+
n ))/2.0...(t9) The allowable range Δu of the manipulated variable is the controlled variable y (
t) and the target value r(t) using equation (t3).

Δu=I r(t)−y(tN Xβ...
(20) (β is the conversion coefficient) Using uj(t) obtained from equation (t6) and △ Δu obtained from the notation, 0 (t + dτ
), we decided to modify u (t) as shown below. Using uz(t) in equation (t9) and Δu in equation (20), if ut(t)−Δu<:u(t)≦u, (t)+Δu, then u(t)=u1(t )u(t)<un(t)−Δu
If so, u(t)=uz(t)−Δuu(t)>uf(
t)+Δu, then -u(t)=ut(t)+Δu. In this way, a permissible range for the amount of operation is established.

The manipulated variable is determined within that range. The results of using this method are shown in FIG. 5. FIG. 5 shows the process from temperature rise to exothermic process, similar to FIG. 3, and is an enlarged view of the temperature on the vertical axis. (a) is the first time, (b) is the second time, and (C) is the fifth time. The vibrations that occurred during the first test were suppressed during the second test by using this method, and it can be seen that the vibrations were suppressed even during the fifth test and were effective.

〔Effect of the invention〕

According to the present invention, when controlling a batch process using an auto-tuning method, focusing on the fact that the same process is repeated for each batch, the information of the previous process is reflected in the next process by the above means, thereby making it possible to It is possible to improve the parts where the control was incomplete in the method described above, and the effect is that the control performance improves as the number of processing increases.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a control device to which the present invention is applied;
Figure 2 is a schematic diagram of the polyvinyl chloride polymerization reaction process used in the examples of the present invention, Figures 3 to 5 are the results when the process in Figure 2 was controlled by the control device in Figure 1, and
The figure shows the manipulated variable u (t) by 80 m-5c t + dτ) from the previous process.
FIG. 4 is a diagram showing the temporal change in control xi (temperature) when using the method of calculating v4.
FIG. 5 is a diagram illustrating control performance when using a method of suppressing vibration of the control amount y (t). In Fig. 5, each (a) is the first processing time, and (b) is the second processing number.
(C) is the fifth time, and each (a) in Figure 4
(b) is the second time, and (b) is the fifth time. FIG. 6 is an example of a table used for time axis correction. Figure 7 shows 1
A flowchart showing the operation procedure of the identification unit when performing the second control, Figure 8 is a diagram for calculating the current estimated parameters from Om-z (t + dτ) of the previous process, and Figure 9 is a diagram for correcting the 1 time axis. FIG.

Claims (1)

[Claims] 1. An identification unit that obtains an estimated value ■(t) of a process parameter from input/output information of a manipulated variable u(t) and a controlled variable y(t) of a process to be controlled, and the controlled variable An automatic device comprising a manipulated variable calculating unit that calculates a manipulated variable u(t) using the deviation e(t) between y(t) and the target value r(t) of the process and the estimated parameter ■(t). In a tuning-type process control device, a learning section is provided to store control information from the previous process, convert the control information into a form that can be used for the next process, and perform the next process based on the converted information. A batch process control method featuring: 2. In the next processing after the second processing number, the above operation amount u(t) is calculated using the process parameter ■( t+dτ). 3. It is characterized by correcting the time difference between the process states of the previous process and the current process based on the magnitude of the deviation of the control amount y(t) between the previous process and the current process using a table given in advance. The batch process control method of the first term. 4. Set the amount u_f(t) obtained by filtering the manipulated variable u(t) in the previous process as the reference value of the manipulated variable in the current process, and set the control amount y(t) in the current process and the target value r(t) Calculate the allowable range Δu of the manipulated variable for this process from the reference value u(t) from the deviation of , and set the above process parameter ■(t+dτ ) The batch process control method according to item 2, characterized in that the manipulated variable determined by the method is corrected.