JPH0425901A - Optimum value retrieval system for control parameter - Google Patents

Abstract

PURPOSE:To execute automatic tuning and to shorten tuning time by extracting an inference characteristic quantity from the width of a complicated signal and sequentially converging it to a parameter value and an optimum value by means of fuzzy inference. CONSTITUTION:A characteristic extraction part 51 detects peak values 13, and 14 from a control target value 11 and a process value 12 and both are compared. Then, a follow-up excess quantity 15 or a follow lack quantity 16 are calculated and it is outputted to an inference part 53 as the inference characteristic quantity 52. The inference part fuzzy-infers the transmitted inference characteristic quantity 52. A fuzzy evaluation value calculated in an inference antecedent part 55 is outputted to an inference precedent part 56 and the alteration quantity of a parameter is decided from the fuzzy evaluation value of the other rule and knowledge for deciding the parameter accumulated from the past, whereby the more satisfactory parameter value is outputted to PID operation parts 4a and 4b. Then, the operation is executed at every peak detection and it is sequentially converged to the optimum parameter value.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an auto-tuning method for automatically determining control parameters for PID control.

(Prior art) Figure 9 is an example of Yokogawa Electric, Expert Self-Tuning Controller, Technical Informat.
ion.

This is a functional block diagram of conventional auto-tuning shown in TT IB4co-01 (1987, 6). In the figure, (1) indicates the control target value (S, (
2) is the process amount detected from the process, and (3) is P
ID control operating pressure value, (4) is PID calculation unit, (5)
is the process, (6) is the auto tuning section, (7)
is a control parameter of PID control.

Next, the operation will be explained with reference to FIGS. 10 and 11. FIG. 11 shows a typical response of the process quantity (12) when the control target value (1 process) is changed.

As shown in FIG. 11, after the control target value (11) is transformed into a rectangular shape, the process amount gradually approaches with a time delay. This transient waveform shows various shapes, but the first
As shown in Figure 0, "If the waveform is oscillatory and the overshoot is small, increase the proportional gain (P parameter) and decrease the integral constant (■ parameter).", "If there is no vibration,
If the convergence is slow, decrease the P parameter and decrease the ■parameter. ”, “If the overshoot is large and the convergence is fast, increase the P parameter! Increase the parameter. ”, “If convergence is slow due to long-period vibration, decrease the P parameter and increase the ■ parameter. ′, the auto-tuning unit (6) calculates the amount of parameter change according to the inference rules, and uses the new control parameter example as the PI.
Set for the D calculation section (4) and set the appropriate operation output value (3
) was given to process (5).

(Problem to be Solved by the Invention) As described above, in the conventional auto-tuning method, the inference rules were structured mainly around step changes that occur due to setting value changes, etc. Rules cannot be applied to signals with complex shapes such as control target values, and inference may not work and effective tuning may not be performed. In such a case,
Tuning must be performed by intentionally inputting a step signal as a disturbance and applying rules, and when starting up a plant, on-site coordinators spend time inputting disturbances many times.

In addition, when there are process fluctuations during normal plant operation, there are problems in that the on-site operator inputs disturbances in the same way or manually performs tuning without activating auto-tuning.

This invention was made to solve the above-mentioned problems, and allows appropriate tuning even with complex inputs such as PID cascade control lower PID.
The purpose of this study is to obtain an optimum value search method for control parameters that can shorten the time required to start up a plant and perform appropriate tuning for process fluctuations during normal plant operation.

[Means to solve the problem]

The control parameter optimum value search method according to the present invention compares the control target value input to the control unit with the peak value of the process quantity measurement value, and determines whether the process quantity measurement value is overshooting or following the control target value. A feature detection unit calculates the amount of deficiency each time a peak value is inspected, and a fuzzy evaluation of the magnitude of the current control parameter given to the control unit is performed based on the amount of tracking overshoot or amount of tracking shortage calculated above. and an inference section that converges to the control parameters.

(Operation) The feature quantity extraction unit in this invention detects a peak value from among the signal changes of the control target value and the process value, compares the two peak values that appear with a time delay, and determines the amount of overshooting or undertracking. The inference section calculates a fuzzy result such as "The current value of the P parameter is too large.'" or "The current value of the P parameter is too small" based on the amount of tracking overshoot and amount of tracking undershoot calculated above. By performing evaluation and gradually bringing the values closer to the optimal parameter values, it is possible to deal with complex signal waveforms, and to make inferences based on signal disturbances that appear every time there are changes in the signal system such as process fluctuations even during normal plant operation. The parameters can be converged to new optimal values.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a system configuration diagram in which an auto-tuning function is added to the IJD cascade control applied to the control parameter optimum value search method in this embodiment.

In the figure, (la) and (1b) are the control target values of the upper PID calculation unit (4a) and the lower PID calculation unit (4b),
(2a) and (2b) are the upper processor of the controlled object (5) (
5a), the operation amount input from the lower processor (5b), (3a) and (3b) are input from each auto-tuning unit (6) from the upper PID calculation unit (1a) and the lower PID calculation unit (1b).
) is the manipulated variable output to (E).

In the auto-tuning section (5), (5) added to the cascade control system having the above configuration,
(51) is a feature extraction unit, and (52) is a feature extraction unit (5
1) is a feature amount for inference outputted from, (53) is a fuzzy inference unit, and (54) is a knowledge base of rules, parameters, and data necessary for inference. (7a).

(7b) is the output of the autotuning section, that is, the PI
These are control parameters for the D calculation units (4a) and (4b).

As is well known, the fuzzy inference section (53) is composed of a knowledge base (54), an inference antecedent section (55), and an inference consequent section (56), as shown in FIG.

Next, the operation of this embodiment according to the above configuration will be explained by taking as an example the complicated control signals shown in FIGS. 3 to 5. In each figure, (11) is the control target value, (12) is the process value, (13) is the peak value of the control target value, (14) is the peak value of the process value, and (15) is the tracking overshoot amount (EVO).
V) and (16) are the follow-up shortage amount (EVON).

Furthermore, to explain the operating function of the auto-tuning section according to Fig. 2, the control target value (11), the process value (12)
) are constantly observed by the feature extraction unit (51) as 1 and 2 in FIG. The feature extraction unit (51) detects peak values (13) and (14) from the control target value (11) and process value (12), respectively, and compares both values to determine the amount of tracking overshoot (15) or the amount of tracking The missing amount (16) is calculated and outputted to the inference section (53) as an inferred feature amount (52).

The inferred feature quantity (52) is expressed as positive when it is an overshooting amount (15), and negative when it is an insufficient amount (16). In the reasoning section,
Fuzzy evaluation is performed on the inferred feature quantity (52) that is sent. For the evaluation, membership functions (21a) and (21b) as shown in Figs. 6 and 7 are used in the knowledge base ( If the inference feature value (52) in 54) is an excessive amount, the membership function (21a) in Figure 6 is used for evaluation, and if it is an insufficient amount, the membership function (21b) in Figure 7 is used for evaluation.
) membership functions are used for evaluation, and their evaluation is performed as shown in Figure 8, and the amount of overshoot detected (15)
By comparing the membership function (21a) with
Calculate the fuzzy evaluation value (17). The fuzzy evaluation value is calculated in the range (0, 1), and for example, "the current P parameter is too large" is expressed in the range (0, 1). (In the case of a shortage, it is expressed as "The current P parameter is too small.") The fuzzy evaluation value calculated in the inference antecedent section (55) in this way is output to the inference consequent section (56), The PID calculation unit (4a) or (4b) determines the amount of parameter change based on fuzzy evaluation values of other rules and knowledge for determining parameters accumulated from the past, and calculates better parameter values.
Output to. Perform these operations every time a beak is detected,
Sequentially converge to the optimal parameter values.

In addition, in the above embodiment, the rules for peak detection are shown, but in addition to the above rules, for example, ``If the process amount does not change even though the control target value fluctuates greatly, the current P parameter and "increase the I parameter." or "If the manipulated variable is applied to the limiter even though the control target value is not close to the signal upper limit value and lower limit value, reduce the P parameter." By configuring the inference unit with the following, convergence to the optimum value of the parameter becomes more reliable when the signal has a complicated shape.

Fuzzy inference generally has a large number of rules and operates these rules in parallel (simultaneously) to produce an inference output. This group of rules is relatively easy to add or delete, so as long as you check the logical consistency between the rules, for example, a typical setting change (
Rules such as step disturbance, ramp disturbance) can also be added, allowing better control.

(Effects of the Invention) As described above, in this invention, inference features are extracted from the width of a complex signal, and they are sequentially converged to parameter values and optimal values by fuzzy inference.
Cascade control lower PID and many other target PIDs
This means that even signals with complex waveforms can be autotuned, making it possible to improve controllability during plant startup, shorten tuning time, and respond to process fluctuations in the controlled process during normal plant operation. This has the effect of

[Brief explanation of drawings]

Fig. 1 is a system configuration diagram of an optimum value search method for control parameters using an auto-tuning method according to an embodiment of the present invention, Fig. 2 is an inference flow diagram of the auto-tuning section, and Figs. 6 to 8 are diagrams showing examples of fuzzy evaluation in the inference section. FIG. 9 is a functional block diagram of the conventional auto-tuning method. FIG. 10 is a diagram showing an example of a conventional inference rule, and FIG. 11 is a diagram showing an example of a signal targeted by the inference rule. (4) is the pH) calculation unit, (13) is the peak value of the control target value, (14) is the peak value of the process amount, (25) is the 4f follow-up overshoot amount, (16) is the follow-up shortage amount, and (17) is the Fuzzy evaluation value, (51) is a feature extraction unit, (52) is a feature, and (53) is a fuzzy inference unit. In addition, in each figure, the same reference numerals refer to the same or corresponding parts. Grand diagram procedural amendment (voluntary)

Claims (1)

[Claims] Feature value detection that compares the control target value input to the control unit and the peak value of the process quantity measurement value, and calculates the amount of overshooting or undertracking of the process quantity measurement value with respect to the control target value each time a peak value is detected. and an inference unit that performs fuzzy evaluation of the magnitude of the current control parameter given to the control unit based on the calculated tracking overshoot amount or tracking undershoot amount and converges to the optimal control parameter. Optimal value search method for control parameters.