JPS63268001A - Foreknowledge type learning control method - Google Patents

Abstract

PURPOSE:To automatically perform the optimum control of a process where a set pattern is repeated by estimating a dynamic model formula of the process with the preceding process operation by means of a sequential most-likely estimating method. CONSTITUTION:An arithmetic processing part 10 supplies the temperature within a reactor and the jacket temperature set in a preceding batch operation as the operation data of a time series. Then the temperature change caused in the reactor by the change of the jacket temperature is supposed as an ARX (self-recursive input/output) of a linear differential model and a dynamic characteristic model is estimated for the operation data after a parameter is decided by a sequential most-likely estimating method. Based on this dynamic characteristic model, a jacket temperature pattern is obtained and outputted to a PID controller 11 as the initial value of a batch operation of this time. The controller 11 performs the open/close control of the control valves 5a and 5b based on the output received from the part 10 and also gives the PID control to the difference between an output (dynamic characteristic model) and an actual fact.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a process predictive learning control method for predictively controlling current process operation based on previous process operation data.

[Prior Art] Conventionally, in the control of batch processes where control is repeatedly performed according to a set pattern, for example, reaction temperature control in a batch reactor, it is necessary to: (1) prevent overshoot during temperature increase; (2) increase temperature in the shortest possible time; ■ Stability during steady state is an important point in control, and in particular, overshoot during temperature rise, which causes uneven quality and abnormal reactions, is the most important point that must be prevented. Ta.

On the other hand, the temperature control of a batch polymerization reactor has been carried out using a program setting device that instructs a target temperature increase/decrease pattern, a cascade feedback control system in which the internal temperature of the reactor is used as a master controller, and yJ is controlled depending on the jacket temperature.

That is, as shown in Fig. 4, a jacket 2 for temperature 2 g and 1 section is provided around the reactor 1, and a thermometer 21 for detecting the internal temperature of each reactor 1 is provided, and the temperature detected from this thermometer 21 is A signal based on the difference between the temperature and the set temperature from the setting device 22 is sent from the controller 23 to the external temperature controller
Furthermore, the external temperature controller 25 corrects the set temperature based on the detected temperature from the thermometer 24 that detects the temperature of the Shakerat 2, and sends a control signal according to the correction result to the control valve. Output to 26a and 26b,
The internal temperature was controlled by adjusting the amount of steam and cooling water, thereby controlling the jacket temperature.

In this case, since PID control alone cannot provide sufficient control, it is necessary to switch from heating to forced cooling as the polymerization reaction progresses. Therefore, the control valve 26 of the steam supplied to the jacket 2 is turned on and the temperature is increased to 3rd speed, and the output of the controller 23 is adjusted at the stage when the internal temperature of the reactor 1 reaches the upper limit setting value of the operating output, that is, at the turning point. This was done by lowering the temperature by the bias setting value and then performing PID control.

[Problems to be Solved] The conventional method for controlling the internal temperature of a reactor described above has a time delay of several minutes or more before responding by changing the jacket temperature, and is prone to over-control due to feedback control. As a result, the jacket temperature fluctuated rapidly, causing phenomena such as overshoot and runaway, and the internal temperature of the reactor tended to become unstable.

Therefore, control has been carried out by adjusting the rate of change in jacket temperature or the setting time of the turning point for switching the temperature from heating to cooling, based on the operator's experience.

As a result, long operating experience is required to determine the turning point, the level of skill of the operator determines whether the control is good or bad, and in the case of an unskilled operator, the controllability becomes extremely poor. The more you become, ■
(2) It is not possible to respond to changes in process characteristics caused by a decrease in activity due to differences in catalyst lots, and a decrease in heat transfer due to contamination on the inner surface of the reactor.

The present invention has been made in view of the above-mentioned problems. For example, in a batch polymerization reactor, etc., the turning point can be automatically determined without requiring long operating experience, and the process The purpose of this study is to provide a predictive learning control method that can adapt to changes in the environment.

[Means for Solving Problems] In order to achieve the above object, the predictive learning control method of the present invention uses a sequential maximum likelihood estimation method based on the previous process operation in a batch process that repeatedly controls according to a set pattern. This is a control method that estimates the dynamic characteristic model equation of the process, determines a control pattern that matches the target pattern of the process based on this dynamic characteristic model equation, and performs the next process operation using this control pattern as an initial value. be.

[Example 7] Hereinafter, an example in which the present invention is applied to control a batch polymerization reactor will be described.

When the predictive learning control method of the present invention is applied to the control of a hatch polymerization reactor, a dynamic characteristic model equation between the reactor internal temperature and heater temperature is estimated from the batch polymerization operation data using the sequential maximum likelihood estimation method. , and based on this dynamic characteristic model equation, a heater temperature pattern that matches the reactor internal temperature target pattern is determined, and this heater temperature pattern is used as an initial value to perform the next operation of the hatch polymerization reactor. .

Next, the method of the embodiment will be explained in detail based on FIGS. 1 to 4.

Figure 51 shows an example of an apparatus configuration for carrying out the method of this embodiment.

In FIG. 1, l is a reactor, and around it is provided a jacket 2 for controlling the internal temperature of the reactor. 3
is a heating or cooling medium supply pipe that is supplied to the jacket 2, and heat exchange is performed with heated steam or cooling water in a heat exchanger 4 to control the temperature. Reference numerals 5a and 5b are temperature control flow rate regulating valves provided in the heating steam or cooling water supply pipes. 6 is a thermometer for detecting the internal temperature of the reactor 1;
7 is a thermometer for detecting the temperature of the jacket 2, and 8 is a thermometer for detecting the temperature of the heating or cooling medium.

10 is an arithmetic processing unit which records operating data of the reactor internal temperature and jacket temperature in the previous hatch operation using a thermometer 6.7.
The model equation for the dynamic characteristics between the reactor internal temperature and jacket temperature is estimated using the sequential maximum likelihood estimation method. Then, based on this dynamic characteristic model, a jacket temperature pattern is determined and outputted to the PID controller 11 as an initial value for the hatch operation to be performed this time. The PID controller 11 is
Based on the output from the arithmetic processing unit 1O, the control valves 5a, 5
Control is performed by opening and closing b, and the difference between the output (dynamic characteristic model) and the actual one is also controlled by PID.

Next, we will discuss the flowchart in Figure 2 and Figures 3 (a) and (b).
) The method of controlling the hatch polymerization reactor will be explained using a graph of the temperature curve.

■ Operational data of the reactor internal temperature T7 and jacket temperature TJ obtained by PID control only during the initial hatch operation are input to the processing unit 10 via the thermometer 6.7, and the time series data (TR(k), T J (k))k -1-1
(201).

If the operating data at this time is expressed as a temperature curve graph, it will be rough as shown in FIG. 3(a).

(2) Based on the above operating data, a dynamic characteristic model of the reactor internal temperature and jacket temperature is determined (202).

It is assumed that the change in the reactor internal temperature T8 when the jacket temperature TJ is changed is a linear difference model ARX (autoregressive input/output) model T, (k+1)=aT++(k)+bT,+(k).

Then, the above operating data (T R (k), T J (k
)) Using the data of k = 1-1, determine the parameters a and b using the sequential maximum likelihood estimation method, and estimate the dynamic characteristic model.

■The manipulated variable jacket temperature T J ” (k) to match the reactor internal temperature T Ll (k+1) or the reactor internal temperature target pattern Tsp (k+1), k −
Obtained from 1-ff.

■Reactor internal target pattern TJ obtained in this way
(k) is stored (204).

■ Output the reactor internal temperature target pattern TJ'(k) as an initial value to the PID controller 11 and perform the current batch operation (205).The actual deviation from the target pattern TJ'(k) is determined by the PID controller Control is performed by 11.

When the operating data at this time is expressed as a temperature curve graph, it becomes a graph as shown in FIG. 3(b).

■The operational data of the reactor internal temperature T7 and jacket temperature TJ during the current hatch operation are input to the arithmetic processing unit 10, and the time series data (T R (k)) is input.

TJ(k)) Save as k= 1− p (20
6), and repeat steps (1) to (6) above.

When the hatch polymerization reactor is controlled in this way, 1
It becomes possible to determine an appropriate turning point based on ~2 hatch operations, ensuring overshoot prevention during temperature rise, temperature rise in the shortest time, and stability during steady state. Furthermore, this makes it possible to easily handle a wide variety of different types of hatch polymerization reactions, as well as automatically adapt to changes in process characteristics.

The predictive learning control method of the present invention can be applied not only to the control of a hatch polymerization reactor, but also to general control systems of processes in which a set pattern is repeated, such as molding control of an injection molding machine.

[Effects of the Invention] As described above, according to the method of the present invention, it is possible to automatically and optimally control the process of turning the edges of a set pattern.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of a device that implements the method of one embodiment of the present invention, Fig. 2 is a flowchart of the method of this embodiment, and Figs. 3 (a) and (b) show the method of this embodiment. FIG. 4 is a diagram showing the operating data before and after the adoption as a temperature curve graph when the conventional method is adopted. 1: Reactor 2/heater (jacket) 6.7.8: Thermometer 10: Arithmetic processing section 11: PID controller

Claims (1)

[Claims] In a batch process that is repeatedly controlled according to a set pattern, the dynamic characteristic model equation of the process is estimated using the sequential maximum likelihood estimation method from the previous process operation.
A predictive learning control method is characterized in that a control pattern that matches the target pattern of the process is determined based on this dynamic characteristic model formula, and the next process is operated using this control pattern as an initial value.