JPS62241002A - Auto-tuning controller - Google Patents

Abstract

PURPOSE:To attain an automatic control of various controlled system without identifying the controlled systems, by storing previously the human experience or perception as an inference rule and inferring optimum parameters from the feature value of waveforms like the controlled variable, etc. based on said inference rule. CONSTITUTION:A position type fuzzy inference unit 10 receives a feature value Si from a feature value extractor 8 and then outputs an optimum parameter through fuzzy inference to a PID controller 4 according to the inference rule Rj obtained from the human experience or perception stored in an inference rule storage device 9. The controller 4 receives a deviation e(k) between a target value signal r(k) and a controlled variable y(k) and then outputs a manipulated variable u(k) to a controlled system 3 based on the set control parameters, i.e., the gain KC, the integration action time TI and the derivative action time TD. The system 3 feeds the value y(k) back to the controller 4. Thus the controller 4 controls continuously the system 3 under an optimum condition.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention is directed to an auto-tuning controller used for example in process control, which is equipped with a function to automatically adjust control parameters according to the characteristics of a controlled object. The present invention relates to controllers, and in particular to automatic adjustment of their control parameters.

[Conventional technology]

Conventionally, an auto-tuning controller as shown in FIG. 5, for example, has been proposed. This is A
, B., Colivio, P., M., Tombkins “Industrial Application Day of a Self-Tuning Feedback Control Algorithm”,
l5A) Ranzakshins vol, 20.1981.
Pages 3 to 10 (A, B, Corripio+P
, M. Tompkins, “rndustr.
ial Application of a Self
-Tuning Feedback Control A
lgorithm” + ISA Transacti
ons, vol.

20, No, 2. 1981. pp, 3-10
). In the figure, 1 is a target signal generator, 2 is an auto-tuning controller,
3 is a controlled object, 4 is a PID controller, 5 is a mathematical model calculator, 6 is an identifier, and 7 is an adjustment calculator.

Conventional auto-tuning configured as above
The operation of the controller will be explained.

The auto-tuning controller 2 inputs the target value signal (r) given from the target value signal generator 1 and the control amount Y (k) which is the output of the controlled object 3, and serves as an input to the controlled object 3. Output the manipulated variable u(k). here,
Values in parentheses represent discretized times at each sampling interval T.

At this time, the following operations are performed inside the auto-tuning controller 2.

First, the target value signal r (kl and the control amount y(k)
From this, the deviation e(k) is calculated.

e (k) = r Ocl −y (k)
(1) The PID controller 4 inputs the deviation e (k), calculates and outputs the manipulated variable u (k) according to the set control parameters. In the PID controller 4, the control parameters are gain Kc,
Integral time 'r+, m minute time T, and according to these parameters, the manipulated variable u(k) is calculated as follows.

'l゛The above-mentioned manipulated variable u (k) becomes an input to the controlled object 3 and also becomes an input to the mathematical model calculator 5 and the identifier 6.

When the mathematical model calculator 5 receives the manipulated variable u (k), it calculates the output v(k) using a mathematical model such as the following formula.

v(k)ma, v(k-1) +at v(k-2)
+ b r u (k-s-1) + b t
u (km-2) (3) Here, m is an integer greater than or equal to 0, and means the dead time of the controlled object 3.

The identifier 6 includes the controlled object 3 and the mathematical model calculator 5.
The coefficients a1 m ! ! + bl + bl is determined.

For this purpose, the identifier 6 inputs the manipulated variable u(k), the controlled variable y(k), and the output v(k) of the mathematical model.

In order to explain the operation of the identifier 6, the following vectors x(k), z(k), φ(t)) are defined here.

Xte (k-1) = (y(k-1),
y(k-2), u(k-gg-1).

u (k-1-2))
...(az" (k-1) = (v(k-1)
, v(k-2), u(ks-1).

u(k-m-2) ) ・-(5)φ(
k) = (at + as # bt + bt
) ”・(6) Here, the subscript T on the right shoulder of the vector
means transpose of a vector.

The identifier (6) executes the following algorithm.

G(k)=(1+z”(k)P(k)x(k))
-', z' (k)P(k) = (7)φ(kl1
)−φ伽)+(y(kl1)−φ(k)x(k))
G(k)=(8)P(kl1) −P(k)−P(k)
x(k)G(k) = (9) By this algorithm, the vector φ), that is, the coefficient of the mathematical model (j equation "1 + at 1 1) I + b" is sequentially obtained.

The vector φ (kl) obtained in the above manner is
The output from the identifier 6 is sent to the mathematical model calculator 5 for use in modifying the mathematical model, and is also sent to the adjustment calculator 7 to determine the control parameter, that is, the gain. It is used to obtain the integral time Ti+differential time T. In order to obtain these control parameters, the adjustment calculator 7 executes the following calculations.

Kc ” (at + 2 am) Q/ tz
""(ulHere, Q appearing in the equation (to) and equation (2) is Qm l -e-""
...defined in Qll. where B is an adjustment parameter and is the desired time constant in the closed group.

Gain Ken integral time T obtained as above,
The l differential time T0 is sent to the PID controller 4 and used to calculate the manipulated variable u(kl) again from the deviation e(k) using equation (2).

[Problem that the invention seeks to solve]

Since the conventional auto-tuning controller is configured as described above, it was necessary to identify the object to be controlled. However, this identification has the following problems.

(1) Calculations are complicated.

(2) The amount of calculation is large.

(3) It takes time to converge.

(4) Nonlinearity of the controlled object cannot be handled.

(5) Since the type of the mathematical model is limited to, for example, equation (3), it is inappropriate for identifying other types of controlled objects.

(6) In order to obtain the three control parameters of Kp, TI, Tt+, a, + at + I)
There is no use in identifying four coefficients of l*b2.

The above-mentioned identification problem has become a problem of its own in conventional auto-tuning controllers.

This invention was made to solve the above-mentioned problems. Instead of identifying the controlled object, human experience and intuition are memorized as inference rules, and these are used to calculate the control amount. Deviation from test signal.

Another purpose is to obtain an auto-tuning controller that can infer optimal control parameters for a controlled object from control variables, test signals, and the like.

[Means for solving problems]

The auto-tuning controller according to the present invention uses a deviation or target value signal instead of having a mathematical model calculator, an identifier, and an adjustment calculator.

a feature extractor that inputs a control amount, etc. and outputs a feature indicating a control characteristic; an inference rule memory that stores an inference rule for inferring an optimal control parameter from the feature; and an inference rule memory that inputs the feature. and a position-type fuzzy inference device that infers optimal control parameters according to the above estimation rules and outputs them to the PfD controller.

[Effect]

The auto-tuning controller of this invention uses human experience and time as an inference rule.

According to this, optimal control parameters are phenomenologically inferred from waveform features such as deviations, target value signals, and control amounts, and automatic adjustment is performed.

〔Example〕

The following is a block diagram showing an auto-tuning controller according to an embodiment of the present invention.

In FIG. 1, 1 is a target value signal generator, which generates a target value signal r(k). Reference numeral 2 denotes an auto-tuning controller, which inputs the target value signal r (k) and the controlled variable y(k) which is the output of the controlled object 3, and outputs the manipulated variable u(k). 3 is a controlled object, which inputs the manipulated variable u(k) and outputs the controlled variable 7 (k).

As mentioned above, this control amount Y (k) is
It is fed back to the tuning controller 2.

Next, the internal configuration of the above auto-tuning controller will be explained. 4 is a controller, and in this embodiment a PID controller is used. This PID controller 4 inputs the deviation e(k) between the target value signal r(g)) and the control amount y(g)), and inputs the set control parameters, that is, the gain KCI integral time T1
The manipulated variable u(k) is output according to one differential time TI.

Reference numeral 8 denotes a feature quantity extractor, which inputs the above-mentioned deviation e al) * or the above-mentioned target value signal r(k), the above-mentioned operation amount 7 (k), etc., and extracts a feature quantity Si;
-1, 2, ..., n are output. Reference numeral 9 denotes an inference rule storage device, in which inference rules R are used to infer optimal control parameters from the feature amount Sl.

5j-1, 2, . . . , - are stored. Reference numeral 10 denotes a positional fuzzy inference device, which inputs the feature amount S1 and calculates optimal control parameters according to the inference rule R,
That is, the gain Kc, integral time T+, and differential time T are inferred and output. Above Kc, TI.

TI is given to the PID controller 4 and used again to calculate the manipulated variable u(k).

FIG. 2 is a flowchart showing the operation of the auto-tuning controller.

First, in step 11, the value of K is set to O.

Next, in step 12, control parameters are initialized.

Its value is as follows. Set the gain KC to a smaller value. In this case, the integral time T1 and the differential time T0 are set to infinity and zero, respectively, or are set to the maximum value and the minimum value, respectively. In steps 13 to 16, the PID controller 4 calculates equation (2) using the above initial parameters and controls the controlled object 3. At this time, the deviation eIl or the target value signal ro<)+control fly(k), etc. are recorded.

After collecting these data for N samples, in step 17, the feature extractor 8 calculates the feature S4; i-1+2+..., i from these data i. The above-mentioned feature amount is, for example, as follows.

52 = (average rate of change of error) In step 18, the positional fuzzy inference device 1° inputs the feature amount St, and infers the inference rules Rt stored in the inference rule storage device 9: J-1, 2, .・According to 1m, fuzzy inference and output the optimal control parameters,
It is given to the PID controller 4.

Thereafter, in steps 19 to 21, the PID controller 4 performs the (second
) formula and continues to control the controlled object 3.

Now, the inference rules R stored in the inference rule storage 9 and the operation of the positional fuzzy inference device 10 will be explained below.

First, the above-mentioned inference rule Rj is a rule based on the empirical rules and intuition that humans use when adjusting control parameters, and is, for example, as follows.

Rt: rIf S is large and the average rate of change of error S8 is large, set the gain Kc to a medium value.'' Rt: ``If the average error SI is large and the average rate of change of error S2 is small. , set the gain Kc large.

” In this way, the inference rule R, is ``If... then...
It has the shape of 廿YO. The part that says "if..." is called the antecedent proposition, and the part that says "do..." is called the consequent proposition.

The form of this consequent proposition is, for example, the above inference rule R1゜Rt
As in, "Take it to..." or "Set it to..."
When the value itself is specified, this fuzzy inference is particularly called positional fuzzy inference.

On the other hand, when the form of the consequent proposition instructs the amount of change in value, such as "make it larger by..." or "make it longer by...", the fuzzy inference is particularly called speed-type fuzzy inference.

The present invention includes a positional fuzzy inference device 10 that executes positional fuzzy inference. Its operation will be explained.

In fuzzy inference, first, the extent to which the current state satisfies the conditions of the antecedent proposition is evaluated using a membership function and expressed as a numerical value between 0 and 1.

FIG. 3 shows an example of evaluation using membership functions. Here, we have proposed the proposition that "the average error S is large." Assume that the average error actually calculated by the feature extractor 8 is 3 + -3+''.In this case, as shown in Figure 3, the degree to which the antecedent proposition holds true is 0.
．． Rated 75.

FIG. 4 shows the inference mechanism performed by the positional fuzzy inference device 10. Here, Sl and S! are used as features. is selected and R3 and R1 are used as inference rules. Further, although only gain adjustment will be explained here, the same applies to integration time and differentiation time adjustment.

First, as described above, the fuzzy inference rules R+, R
Evaluate the degree to which the antecedent proposition of t holds true. Here, R
,, Rt, the antecedent proposition consists of multiple terms,
If it is in the form "If, then, and then", the lowest probability of each term is taken as the probability of the entire antecedent proposition. In the example of FIG. 4, the values of the actual average error Sl and the average change rate of error S2 are 5I, respectively.
=s,”,s,=s!”, then the rule R+
The degree to which “if the average error S1 is large” holds true is 0.
．． 75, and “if the average rate of change of error S is large”
The degree to which is true is 0.2. Therefore, the degree to which the entire antecedent proposition holds true is 0.2. -The consequent proposition is also the fourth
It is represented by a membership function as shown in the figure. Now, the degree to which the antecedent proposition of rule RI is satisfied is 0.
．． 2, the membership function of the consequent proposition is reduced by a factor of 0.2.

Similarly, calculation is performed for rule R1, and the consequent proposition is reduced by 0.75 times.

Finally, the membership functions of the reduced consequent propositions of each rule are superimposed and their centroids are determined. This center of gravity is used to obtain the optimum gain.

In the same way, the optimal integration time and suitable differentiation time are also deduced.

In the manner described above, the rank 1 type fuzzy inference device 1° infers optimal control parameters.

In addition, in the above embodiment, a PID controller is used as the controller, and its gain and integration time are determined.

Although the auto-tuning controller that automatically adjusts the differential time has been described, the present invention can also be applied to other types of auto-tuning controllers. It is also effective to apply it to an auto-tuning controller that automatically adjusts the width of the dead zone, and an auto-tuning controller that uses an optimal controller based on modern control theory as the controller and automatically adjusts its evaluation function noharameter. It is also effective when applied to a controller.

〔Effect of the invention〕

As described above, according to the present invention, human experience and intervals are memorized as inference rules, and control parameters are fuzzy inferred from waveform features such as deviations, target value signals, and control amounts in accordance with the inference rules. It's complicated,
There is no need to perform identification that would limit the type of control target, and automatic adjustment based on human experience and intuition can be performed through simple membership function calculations, resulting in a light calculation load.

Automatic adjustment in a short time 1 This has the effect of enabling automatic adjustment of a wide variety of control objects.

[Brief Description of the Drawings] Fig. 1 is a block diagram showing an auto-tuning controller according to an embodiment of the present invention, Fig. 2 is a flowchart showing its operation, and Fig. 3 is an evaluation using a membership function. FIG. 4 is a diagram showing a mechanism of fuzzy inference, and FIG. 5 is a diagram showing an example of a conventional auto-tuning controller. In the figure, 1 is a target value signal generator, 2 is an auto-tuning controller, 3 is a controlled object, and 4 is a controller; here, as an example, a PID controller, 8 is a feature extractor, and 9 is an inference rule. Memory device, 10
is a positional fuzzy reasoner. Note that the same reference numerals in the figures indicate the same or equivalent parts.

Claims (1)

[Claims] (1) A controller that inputs the deviation between the target value signal output by the target value signal generator and the controlled variable that is the output of the controlled object, and outputs the manipulated variable that is the input of the controlled object; a feature extractor that inputs the signal or the control amount and outputs a feature indicating the state of the control system; an inference rule storage that stores an inference rule for inferring optimal control parameters of the controller; The automatic system is characterized by comprising a position-type fuzzy inference device that inputs feature quantities, infers optimal control parameters according to the above-mentioned inference rules, and provides them to the above-mentioned controller.
tuning controller.