JPH0981206A - Fuzzy control device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically set an adjusting parameter to a suitable value. SOLUTION: The fuzzy control device 10 provided with a preprocessing part 12, a normalizing part 14, a fuzzy inference part 16, an output part 18 for outputting manipulated variable to a controlled target 30 is also provided with a parameter adjusting part 20 consisting of a waveform feature value calculating part for calculating the feature value of a response waveform based upon a target value and a controlled variable, a fuzzy inference part for finding out the reference updating quantity of adjustment gain from the feature value of the waveform, an waveform evaluating part for evaluating the score of the waveform, a fuzzy inference part for finding out the correction coefficient adjustment gain from the evaluation score, and a gain variation operation part for finding out the adjustment gain from the reference updating variable and the correction coefficient and the gain of at least either one of the normalizing part 14 and the output part 18 is calculated by the parameter adjusting part 20.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fuzzy controller suitable for controlling by a servo controller.

[0002]

2. Description of the Related Art Generally, a fuzzy control device formulates a high-level idea of human beings and a qualitative judgment method based on a fuzzy theory, and controls advanced and complicated control of machines, devices and robots with a built-in computer. Make it even more natural.

In this case, the performance of the fuzzy control device largely depends on the adequacy of the setting of the adjustment parameter (adjustment gain) such as the membership function for the fuzzy reasoning. However, the setting of this parameter requires repeated manual thought and error. However, since it is selected to have the optimum value, it requires a lot of labor and time. Alternatively, there are some that automatically set the adjustment parameters, but in this case,
The condition is that the same control operation is repeated many times, and the optimum parameter is calculated by this iterative learning. Therefore, when the frequency of operation repetition under the same condition is low, the control result is not always optimal.

On the other hand, Japanese Laid-Open Patent Publication No. 4-15705 proposes a method in which optimum parameters can be automatically set without the need for repeated learning.

This will now be described with reference to FIG.

The fuzzy controller comprises a preprocessing unit 3, a normalization unit 5, an inference unit 7, an output unit 9, an integration unit 11, and a holder 1.
3, a gain adjusting unit 15, a signal generating unit 17, and a rule adjusting unit 19, and a controlled object 21 is connected to the holder 13.

The target value r (k) and the control amount y (t) which is the output of the controlled object 21 are input to the preprocessing unit 3, and the control amount y (t) is converted into a sample signal y (k) by a sampler. Then, the control deviation e (k) and the one-time difference (differential value) de (k) of this control deviation are calculated by the following equation.

E (k) = r (k) -y (k) de (k) = de (k) -ek (k) The normalization unit 5 has these control deviations e calculated by the preprocessing unit 3.
(K) and this control deviation one-time difference de (k) are input, and the control deviation normalized value E and the control deviation one-time difference normalized value DE are calculated by the following equations.

E = c1 * e (k) DE = c2 * de (k) where c1 and c2 are adjustment gains. The inference unit 7 has a normalized value E of the control deviation which is the output of the normalization unit 5 and control. The normalized value DE of the one-time difference of the deviation is input, and the normalized value DU for the change of the manipulated variable is calculated by fuzzy inference using the control rule shown in the following expression, for example.

IF (EisNBandDEisNB) THEN (DUisNB) IF (EisNBandDEisNM) THEN (DUisNB) ...... IF (EisPBandDEisPB) THEN (DUisPB) However, NB, NM ... The normalized value DU of the change amount of the manipulated variable calculated in step 3 is input, and the change amount du of the operation amount is calculated by the following equation.
Calculate (k).

Du (k) = c3 * DU, where c3 is the adjustment gain. The amount of change d in the manipulated variable calculated by the output unit 9 in the integrating unit 11.
u (k) is input, and the manipulated variable u (k) (sample value) is calculated by the following equation.

U (k) = u (k-1) + du (k) where u (k-1) is the manipulated variable before the sampling time. The gain adjusting section 15 controls the manipulated variable u (t) and the controlled variable y.
(T) is input and the manipulated variable u converted by the sampler
From (t) and the controlled variable y (k), the dead time L and the slope R when the controlled object is expressed in the form of the following equation are calculated.

G (s) = (R / s) e ^-Ls Normalization part gains c1 and c2 and output part gain c3 are calculated by the dead time L and the slope R calculated when expressed in the form of this equation. Are calculated and output to the normalization unit 5 and the output unit 9, respectively.

The signal generator 17 inputs the gain adjustment signal d (k) to the adder 23 according to the command signal from the gain adjuster 15. The adder 23 adds the manipulated variable u (k) and the gain adjustment signal d (k) and outputs the result to the holder 13.

On the other hand, the rule adjusting unit 19 has a target value r
(K) and the control amount y (t) are input, and the sample value signal y (t) of the control amount y (t) converted by the sampler,
The control evaluation amount is calculated from the target value r (k), and the control rule of the inference unit 3 is modified by fuzzy inference.

The control evaluation amount includes, for example, a rising time when the target value is changed, an overshoot amount, a setting time, a maximum deviation with respect to a disturbance during the constant value control, a setting time, and the like.

In this way, the gain adjustment signal is added to the controlled object to adjust the gains c1 and c2 of the normalization section and the gain c3 of the output section, and the control rule is adjusted from the control evaluation amount, thereby fuzzy. In order to improve the control accuracy, the need for repetitive learning is eliminated, and fuzzy control can be applied to plant equipment, etc., where operations are rarely repeated.

[0018]

However, in this fuzzy controller, the control target is controlled by the operation amount and the control amount.
(S) = (R / s) e ^−Ls The dead time L expressed in the form of
And the slope R are calculated, and the normalization unit gains c1 and c2 and the output unit gain c3 are calculated based on these,
When trying to realize control by the servo controller,
Responsiveness is fast, the dead time L and the slope R cannot be calculated correctly, and in general, the control target is G (s) = (R
Since it cannot be expressed in the form of / s) e- ^Ls , it was not suitable for control by the servo controller.

Further, in order to calculate the normalization section gains c1 and c2 and the output section gain c3, the gain adjustment signal is output from the signal generation section to the adder, and the operation amount and the gain adjustment signal are added in the adder. However, since it is output to the control target via the holder, the gain adjustment signal acts as a disturbance on the control target.

That is, in order to set the normalization part gains c1 and c2 and the output part gain c3, an originally unnecessary gain adjustment signal is added as an operation amount of the controlled object until the control is converged and stabilized (automatic adjustment is performed. Until the end), the control performance will be reduced.

An object of the present invention is to automatically set adjustment parameters such as a gain of a normalization unit and a gain of an output unit to appropriate values without obtaining a dead time of control and a slope.

Another object of the present invention is to set an adjustment parameter without inputting a gain adjustment signal that acts as a disturbance to a controlled object.

A further object of the present invention is to reduce the number of times and time required for automatic adjustment when setting the adjustment gain.

[0024]

A first aspect of the present invention is to provide a preprocessing unit for calculating a control deviation and its differential value from an input target value and control amount, and a control deviation output from the preprocessing unit and its control value. A normalization unit that calculates a normalized value of these control deviations and differential values based on the differential value, and a fuzzy inference unit that calculates the manipulated variable by a fuzzy inference operation according to a predetermined control rule based on the input of these normalized values. In a fuzzy control device including an output unit that converts the calculated operation amount into an operation amount according to a control target and outputs the operation amount, a waveform for calculating a characteristic amount of a response waveform based on the target value and the control amount. A feature amount calculation unit, a basic update amount for finding the basic update amount of the adjustment gain by fuzzy inference from the waveform feature amount, a fuzzy inference unit, a waveform evaluation unit for scoring the response waveform, and a basic update amount for the adjustment gain from the evaluation score. A correction coefficient fuzzy inference unit for determining by fuzzy inference the correction coefficient, and the gain change amount calculation unit for obtaining the adjustment gain based on the basic updating amount and the correction coefficient,
The parameter adjusting unit is configured to calculate the adjusting gain of at least one of the normalizing unit and the output unit.

In a second aspect based on the first aspect, the parameter adjusting section adjusts each adjustment gain of the normalizing section and the output section.

In a third aspect based on the first or second aspect, the waveform characteristic amount calculating section calculates the waveform characteristic amount from the rise time of the response waveform, the overshoot amount, and the vibration amount.

In a fourth aspect based on the first or second aspect, the waveform characteristic amount calculating section calculates the waveform characteristic amount from the overshoot amount, the vibration amount, and the steady deviation of the response waveform.

A fifth aspect of the present invention is a control device that determines an operation amount based on a proportional component, an integral component, and a derivative component of a control deviation from an input target value and control amount. Based on the waveform feature amount calculation unit that calculates the response waveform feature amount based on the waveform feature amount, the basic update amount fuzzy inference unit that obtains the basic update amount of each constant of proportional, integral, and derivative from the waveform feature amount by fuzzy inference, and the response waveform A waveform evaluation unit for evaluating the score,
From the evaluation score, the correction coefficient of the basic update amount of each constant of proportional, integral, and differential is obtained by fuzzy reasoning. The correction coefficient fuzzy inference unit, and the constants of proportional, integral, and derivative are obtained based on the basic update amount and the correction coefficient. A parameter adjustment unit including a gain change amount calculation unit is configured, and the parameter adjustment unit calculates each constant of proportionality, integration, and differentiation.

In a sixth aspect based on the fifth aspect, the waveform characteristic amount calculating section calculates the waveform characteristic amount from the rising time of the response waveform, the overshoot amount, and the vibration amount.

In a seventh aspect based on the fifth aspect, the waveform characteristic amount calculation section calculates the waveform characteristic amount from the overshoot amount, the vibration amount, and the steady deviation of the response waveform.

[0031]

In the first aspect of the invention, in calculating the adjustment gain for the normalizing section or the output section in the parameter adjusting section, the characteristic quantity of the response waveform is calculated from the control quantity which is the output of the controlled object and the target value. Then, the basic update amount of the gain is calculated from the characteristic amount of the waveform by fuzzy inference. Further, the score of the waveform is evaluated from the characteristic amount of the waveform, and the correction coefficient of the basic update amount is calculated by fuzzy inference based on this evaluation.

Then, the adjustment gain is calculated using the results of these two fuzzy inferences, that is, from the basic update amount and the correction coefficient.

Therefore, the adjustment gains of the normalizing section and the output section can be calculated without obtaining the dead time and the slope of the response waveform, which is extremely effective in the case of performing servo control with a fast response by the servo controller.

Further, the parameter adjusting unit can automatically perform the parameter adjustment based on the experience and knowledge of the expert by using the fuzzy reasoning. Further, in the parameter adjustment unit, the gain can be changed largely in the initial stage of adjustment by the two fuzzy inferences, and thereafter the gain can be finely adjusted. Even when the control target is changed, the gain adjustment can be performed with a smaller number of adjustments and time. .

For this reason, it is not necessary to input a gain adjustment signal to the control target so as to cause a disturbance, so that there is no problem that the control during adjustment becomes unstable.

The adjustment gain is adjusted by at least one gain adjustment of the normalizing section or the output section,
It is also possible to reduce the feature amount of the waveform to be evaluated and simplify the adjustment.

In the second invention, all the gains for the normalizing section and the output section are adjusted, so that the performance of fuzzy control is improved accordingly.

In the third aspect of the invention, since the waveform characteristic amount calculating section calculates the waveform characteristic amount from the rising time, the overshoot amount, and the vibration amount of the response waveform of the controlled object, the gain adjustment can be performed more accurately and quickly.

According to the fourth aspect of the invention, the waveform characteristic amount calculation unit calculates the waveform characteristic amount from the overshoot amount, the vibration amount, and the steady deviation of the response waveform of the controlled object. Therefore, the gain can be accurately and quickly adjusted according to the control content. Can be adjusted.

In the fifth aspect of the present invention, in calculating the constants of proportionality, integral and derivative in the parameter adjusting section, the characteristic amount of the response waveform is calculated from the control amount which is the output of the controlled object and the target value, The basic update amount of each constant of proportionality, integral, and derivative is calculated from the feature quantity of the waveform by fuzzy reasoning. Further, the score of the waveform is evaluated from the characteristic amount of the waveform, and the correction coefficient of the basic update amount is calculated by fuzzy inference based on this evaluation.

Then, using the results of these two fuzzy inferences, that is, from the basic update amount and the correction coefficient, the proportional, integral, and derivative constants are calculated.

Therefore, the constants of proportionality, integral, and derivative can be calculated without obtaining the dead time and slope of the response waveform.
It is extremely effective when performing quick response servo control by the servo controller.

Further, the parameter adjusting unit can automatically perform the adjustment based on the experience and knowledge of the expert by using the fuzzy reasoning, and the number of adjustments can be reduced even when the controlled object changes. You can adjust each constant in time.

According to the sixth aspect of the invention, the waveform characteristic amount calculating section calculates the waveform characteristic amount from the rising time, the overshoot amount, and the vibration amount of the response waveform of the controlled object, so that the adjustment can be performed more accurately and promptly.

According to the seventh aspect of the invention, the waveform characteristic amount calculating section calculates the waveform characteristic amount from the overshoot amount, the vibration amount and the steady deviation of the response waveform of the controlled object, so that the adjustment can be made accurately and promptly according to the control content. Can be done.

[0046]

BEST MODE FOR CARRYING OUT THE INVENTION The embodiment of the present invention will be described. First, in FIG. 1, a fuzzy controller 10 includes a preprocessing unit 12, a normalization unit 14, a fuzzy inference unit 16, and an output unit 1.
8. The parameter adjusting unit 20 is provided, and the operation amount is output from the output unit 18 to the controlled object 30.

The preprocessing unit 12 controls the target value r and the controlled object 30.
The control amount y, which is the output of the above, is input, and the control deviation err, which is the deviation between them, and its differential value errdot are calculated.
The normalization unit 14 outputs the control deviation er which is the output of the preprocessing unit 12.
Based on r and its differential value errdot, a normalized value E of the control deviation and a normalized value DE of its differential value are calculated by the following equations.

E = c1 * err DE = c2 * errdot where c1 and c2 are adjustment gains (adjustment parameters) The fuzzy inference unit 16 inputs the normalized values E and DE which are the outputs of the normalization unit 14, and, for example, The normalized value u of the manipulated variable is calculated by fuzzy inference using the control rules shown below.

IFEisNBandDEisNBTHENuisNB IFEisNBandDEisNMTHENuisNB ... IFEisPBandDEisPBTHENUuisPB However, NB, NM ... PB cannot output the corresponding operation amount by the fuzzy variable, because the operation amount is a variable amount that cannot be directly changed by the servo controller. Output.

The output unit 18 is the fuzzy inference unit 16
The normalized value u of the manipulated variable calculated in step 3 is input, and converted into the manipulated variable U according to the controlled object 30 by the following equation, and the controlled object 3
Output to 0.

U = c3 * u c3 is an adjustment gain (adjustment parameter) In order to calculate the adjustment parameters c1, c2 and c3,
A parameter adjustment unit 20 that calculates each gain of the normalization unit 14 and the output unit 18 based on the control amount y and the target value r obtained from the controlled object 30 is provided.

FIG. 2 shows the configuration of the parameter adjusting unit 20. The parameter adjusting unit 20 calculates the waveform characteristic amount based on the control amount y and the target value r obtained from the controlled object 30. A unit 21, a basic update amount fuzzy inference unit 22 for obtaining a basic update amount of gain from a waveform feature amount by fuzzy inference, a waveform evaluation unit 23 for scoring a response waveform, and a correction coefficient for a basic update amount of gain from a score. Correction coefficient fuzzy inference unit 24 for obtaining
And a change amount calculator 25 that calculates the change amount of the gain from the basic update amount and the correction coefficient.

Here, referring to the flowchart of FIG. 3, the adjustment operation in the parameter adjusting section 20 will be mainly described. First, a small step input is added to the controlled object 30 as a target value r (step 31). . As a characteristic amount of the waveform response from the target value r and the control amount y obtained from the controlled object 30, the time required for the rising of the control waveform as shown in FIG. 4 (90% of the target displacement after the step input is performed.
The amount of vibration in which the waveform periodically moves up and down after the rise time elapses (the value obtained by dividing the integrated value of the difference between the maximum value and the minimum value of the vibration waveform during the predetermined period by the number of vibrations). ), After the step input, the waveform feature amount calculation unit 21 calculates the magnitude when the control waveform exceeds the target value and overshoots (step 32).

Then, in step 33, the basic update amount of the adjustment parameter is calculated by fuzzy inference in the basic update amount fuzzy inference unit 22 based on these three waveform feature amounts.

For example, an adjustment rule when the basic update amount for the adjustment gain (adjustment parameter) c3 of the output section 18 is c3 'is shown below. The adjustment rules are created based on the experience and knowledge of experts.

IF Fast Rise Time and Few Oscillation and Small Overshoot THEN C3 'Negative Very Sma
ll Here, Rise Time is the rise time, Osi
"llation" is the vibration amount, "Overshoot" is the overshoot, and "Fast", "Few", "Small", "Ne"
“Gathering Very Small” means a fuzzy variable.

In this case, all of the adjustment gains c1 and c2 for the normalization unit 14 and the adjustment gains c3 for the output unit 18 do not have to be adjusted, but some of them, for example, c1.
It is also possible to reduce the feature amount to be used by setting only c3 and c3.

Of course, the feature amount is not limited to these three, and a steady deviation can be used instead of another feature amount serving as an index for response waveform evaluation, for example, the rise time of the response waveform.

As shown in FIG. 5, a weight (score) is given to each label of each membership function of the feature quantity of the waveform input to the basic update amount fuzzy inference unit 22. The basic update amount fuzzy inference unit 22 uses the label label weight and the label conformance used in the fuzzy inference, and the waveform evaluation unit 23 evaluates the response waveform from the three feature amounts, that is, the waveform is scored. (Step 34).

Here, the adjustment operation is terminated by any one of the point having the extreme value, the evaluation point reaching the threshold value (threshold value), and the designated maximum number of tuning (adjustment) times. It is when the conditions are satisfied.

At step 35, it is judged from the above conditions whether the tuning is completed, and if it is not completed,
In step 36, the correction coefficient fuzzy inference unit 24 fuzzy infers the correction coefficient of the basic update amount by inputting the amount of change between the present score and the previous score. Now, an example of a fuzzy inference adjustment rule when the correction coefficient of the basic update amount c3 ′ of the adjustment gain c3 to the output unit 18 is set to c3D is shown below.

IF Rating is Good and Rating change is Positive
Medium THEN C3D Negative Small Here, Rating is the evaluation score of the response waveform, Ratin
g change represents the amount of change in the evaluation score of the response waveform from the previous time, and Good, Psitive Media
m and Negative Small are fuzzy variables.

Using the basic update amount obtained in step 33 and the correction coefficient obtained in step 36, the change amount calculation unit 2
In step 5, a new output section adjustment gain c3 is calculated (step 37). Next, an example of the adjustment gain c3 is calculated as follows: c3 = c3old + c3 ′ × c3D. However, c3old is the previous c3 value,
c3 'is a basic update amount fuzzy inferred from the characteristic amount of the response waveform, and c3D is a correction coefficient fuzzy inferred from the evaluation score of the response waveform and its change amount.

This correction coefficient corrects the basic update amount which is fuzzy inferred from the characteristic amount of the waveform. Since the evaluation score of the waveform is low in the initial stage of adjustment and the change amount of the score is large, the next adjustment gain is basically updated. It changes more than the amount. After that, as the adjustment progresses, the evaluation score of the waveform increases and the amount of change decreases, so that the next adjustment gain changes smaller than the basic update amount. As a result, the gain is largely changed in the initial stage of adjustment, and then fine adjustment is performed, which serves to prevent the number of adjustments from increasing. After that, the step response is performed again with a new gain, and this adjustment operation is repeated until the tuning end condition is satisfied.

When the tuning end condition is satisfied, the routine proceeds to step 38, where the adjustment gain at that time is determined as the final gain, and the adjustment operation is ended.

Next, another embodiment shown in FIG. 6 will be described.

This is an example in which the present invention is applied in the case of controlling a controlled object by PID control (proportional / integral / derivative control) instead of fuzzy control. A PID control unit 26 is provided instead of the fuzzy inference unit 16. , Parameter adjusting unit 20
Therefore, instead of the adjustment gains of the normalization unit 14 and the output unit 18, the constants of P, I, and D of the PID control unit 11 are similarly adjusted.

The PID control unit 26 determines the target value r and the control amount y.
The operation amount u is determined based on the proportional component, the integral component, and the derivative component of the deviation between and, and the operation of the controlled object 30 is controlled. In this case, the proportional, integral, and derivative constants are used as the parameter adjustment unit 20.
Thus, by performing adjustment in the same manner as described above, the control result can be quickly and accurately converged to the target value after the control is started.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a block diagram showing a parameter adjusting unit of the same.

FIG. 3 is a flowchart showing a parameter adjustment operation.

FIG. 4 is an explanatory diagram showing control waveforms.

FIG. 5 is an explanatory diagram showing a membership function.

FIG. 6 is a block diagram showing another embodiment of the present invention.

FIG. 7 is a block diagram showing a conventional example.

[Explanation of symbols]

 10 fuzzy control device 12 pre-processing unit 14 normalization unit 16 fuzzy inference unit 18 output unit 20 parameter adjustment unit 21 waveform feature amount calculation unit 22 basic update amount fuzzy inference unit 23 waveform evaluation unit 24 correction coefficient fuzzy inference unit 25 change amount calculation Part 30 Control target

Claims (7)

[Claims] 1. A pre-processing unit for calculating a control deviation and its differential value from an input target value and control amount, and a control deviation and a differential value based on the control deviation and its differential value output from the pre-processing unit. A normalization unit that calculates a normalized value of, a fuzzy inference unit that calculates an operation amount by a fuzzy inference operation according to a predetermined control rule based on the input of these normalized values, and the calculated operation amount as a control target. In a control device provided with an output unit that converts and outputs the operation amount according to the operation amount, a waveform feature amount calculation unit that calculates the feature amount of the response waveform based on the target value and the control amount, and adjusts from the waveform feature amount A basic update amount fuzzy inference unit that obtains the basic update amount of gain by fuzzy inference, a waveform evaluation unit that scores and evaluates the response waveform, and a correction coefficient that obtains the correction factor of the basic update amount of the adjustment gain from the evaluation score by fuzzy inference. A fuzzy inference unit and a gain change amount calculation unit that obtains an adjustment gain based on the basic update amount and the correction coefficient constitute a parameter adjustment unit, and the parameter adjustment unit configures at least one of the normalization unit and the output unit. A fuzzy controller characterized in that an adjustment gain is calculated. 2. The fuzzy control device according to claim 1, wherein each of the adjustment gains of the normalization unit and the output unit is adjusted by the parameter adjustment unit. 3. The fuzzy control apparatus according to claim 1, wherein the waveform characteristic amount calculation unit calculates the waveform characteristic amount from the rise time, the overshoot amount, and the vibration amount of the response waveform. 4. The fuzzy control apparatus according to claim 1, wherein the waveform characteristic amount calculation unit calculates the waveform characteristic amount from an overshoot amount, a vibration amount, and a steady deviation of the response waveform. 5. A control device for determining a manipulated variable based on a proportional component, an integral component, and a derivative component of a control deviation from an input target value and control amount, and a response waveform based on the target value and the control amount. A feature quantity calculation unit that calculates the feature quantity of the waveform, a basic update quantity fuzzy inference unit that obtains the basic update quantity of each constant of proportional, integral, and derivative from the feature quantity of the waveform by fuzzy inference, and a waveform that evaluates the response waveform An evaluation unit and a correction coefficient fuzzy inference unit that obtains a correction coefficient of the basic update amount of each constant of proportionality, integral, and derivative from the evaluation score by fuzzy inference,
A parameter adjustment unit consisting of a gain change amount calculation unit that obtains each constant of proportionality, integration, and differentiation based on the basic update amount and the correction coefficient, and this parameter adjustment unit calculates each constant of proportionality, integration, and differentiation. A fuzzy control device characterized by being calculated. 6. The fuzzy control apparatus according to claim 5, wherein the waveform feature amount calculation unit calculates the waveform feature amount from the rise time, the overshoot amount, and the vibration amount of the response waveform. 7. The fuzzy control apparatus according to claim 5, wherein the waveform feature amount calculation unit calculates the waveform feature amount from the overshoot amount, the vibration amount, and the steady deviation of the response waveform.