JP2828768B2 - Fuzzy knowledge construction device - Google Patents

Description

The present invention relates to a fuzzy knowledge construction apparatus for constructing a fuzzy knowledge base for a fuzzy control apparatus for performing control based on fuzzy control rules, and particularly to optimization. PID
The present invention relates to a method of constructing a fuzzy control rule based on a constant and tuning the constructed fuzzy control rule to construct an optimal fuzzy control rule.

(B) Conventional technology In fuzzy control, based on fuzzy knowledge including a membership function and fuzzy control rules, for example, an optimal operation amount corresponding to a control target is calculated from a control (response) deviation and its difference information by calculation. Perform control. As a result, a non-linear and variable gain that cannot be obtained by conventional PID (proportional, integral, differential) control or the like can be easily realized, and high-precision control can be performed. For this reason, it is applied to a very large number of control systems.

In order to perform good fuzzy control, it is important to construct fuzzy knowledge suitable for the control target.

Therefore, for example, in “Design of Self-Adjusting Fuzzy Controller” (1989, 5th Fuzzy System Symposium Proceedings, pp. 89-94), control variables and the first order The second order difference of the difference and the control deviation is taken, the first order difference of the manipulated variable is taken as the consequent variable, and the three antecedent variables are N (negative: negative) and Z (zero:
In a fuzzy control device composed of fuzzy control rules obtained from the result of fuzzy division into zero (0) and P (positive), control is performed after adjusting a scaling factor for normalizing the input / output value of the fuzzy control device by learning. When the control response is obtained by sampling during the operation, the automatic tuning of the fuzzy control rule is performed so as to obtain the target response by correcting the conclusion part (the operation amount in the consequent part) of the fuzzy control rule.

In this way, the fuzzy knowledge once constructed (for the fuzzy control rules in the above example) is corrected to optimize the fuzzy knowledge control system.

(C) Problems to be Solved by the Invention However, in the above-described automatic tuning, the basic correction amount of the conclusion part of the fuzzy control rule is determined by a response deviation, which is a difference between a transfer response waveform and a control response waveform at the time of sampling, And the amount of change from the previous response deviation are determined only by information such as positive, negative or zero (the final correction amount is based on the degree of establishment of each fuzzy control rule as a basic correction amount). Multiplied), it was not possible to correct the conclusion part of the detailed fuzzy control rule according to the magnitude of the response deviation and the variation of the response deviation. For this reason, when the difference between the ideal response waveform and the control response waveform based on the fuzzy control rule before correction is large (an inappropriate value is set as an initial value of fuzzy knowledge), sufficient convergence is ensured. It was not possible to perform good control.

Further, in order to correct fuzzy knowledge, it is necessary to construct an initial fuzzy knowledge before correction.

Conventionally, the initial fuzzy knowledge has to be constructed by the designer giving an appropriate fuzzy division to the input variables and further considering the initial values of the membership function and the fuzzy control rules. However, it was not easy to build fuzzy knowledge from nothing.

The present invention has been made in view of such a point, and automatically creates fuzzy knowledge capable of performing good control.
An object of the present invention is to provide a fuzzy knowledge construction apparatus that can optimally tune generated fuzzy knowledge in accordance with a response deviation, which is a difference between an ideal response and a control response, and the magnitude of the change.

(D) Means for Solving the Problems The present invention relates to a fuzzy knowledge construction apparatus, wherein a parameter storage means for storing at least one PID constant for PID control, and a PID constant stored in the parameter storage means PID calculation means for controlling the control target in accordance with the control response from the control target based on the PID, and determining a characteristic amount for evaluation from the control response of the control target and controlling the difference between the characteristic amount and the control target value PID control evaluation means for outputting target deviation, fuzzy knowledge base for PID constant tuning storing fuzzy knowledge for tuning PID constant according to control target deviation, membership function and fuzzy control based on PID constant Conversion means for generating control fuzzy knowledge composed of rules; control fuzzy knowledge storage means for storing control fuzzy knowledge generated by the conversion means;
Fuzzy control means for controlling the control target in accordance with a control response from the control target based on the control fuzzy knowledge stored in the control fuzzy knowledge storage means; and a fuzzy control rule for the control response in the fuzzy control means. A rule satisfaction degree storage means for storing a satisfaction degree, an ideal response storage means for storing an ideal response of the controlled object, and a difference between a control response of the control object and an ideal response stored in the ideal response storage means. Fuzzy control evaluation means for outputting a certain response deviation and the magnitude of the change, and for fuzzy control rule tuning storing fuzzy knowledge for tuning the fuzzy control rule according to the response deviation and the magnitude of the change A PID control evaluator based on the fuzzy knowledge base and the fuzzy knowledge stored in the fuzzy knowledge base for tuning PID constants. Corrects the PID constant stored in the parameter storage means in accordance with the control target deviation output from the CPU, or is stored in the rule satisfaction degree storage means based on the fuzzy knowledge stored in the fuzzy knowledge base for fuzzy control rule tuning A tuning fuzzy inference means for modifying the fuzzy control rules stored in the control fuzzy knowledge storage means in accordance with the degree of establishment and the response deviation outputted from the fuzzy control evaluation means and the magnitude of the change. is there.

(E) Operation In the control process of the control object by the PID calculation means, the PID constant stored in the parameter storage means is changed by the tuning fuzzy inference means based on the fuzzy knowledge stored in the PID constant tuning fuzzy knowledge base. The PID control is modified so that the optimum PID control is performed in accordance with the control target deviation which is the difference between the control response characteristic amount and the control target value obtained by the PID control evaluation means.

Next, the converting means generates control fuzzy knowledge including a membership function and a fuzzy control rule based on the optimized PID constant, and stores the generated fuzzy knowledge in the control fuzzy knowledge storage means.

Then, in the control process of the control object by the fuzzy control means, the fuzzy inference means for tuning uses the fuzzy knowledge stored in the fuzzy knowledge storage means based on the fuzzy knowledge stored in the fuzzy knowledge base for fuzzy control rule tuning. Is corrected in accordance with the response deviation obtained by the fuzzy control evaluation means and the magnitude of the change and the rule fulfillment degree stored in the rule fulfillment degree storage means.

(F) Embodiment FIG. 1 is a schematic configuration diagram of an embodiment of the apparatus of the present invention.

(1) is a control target, (2) is a PID parameter memory as parameter storage means for storing and holding PID constants P, T _I , and T _D , and (3) is a control response y at a set value r and a current sampling time j. enter the control deviation e _j with _j, the operation amount based on the PID constants stored in the PID parameter memory (2) 1
A PID calculation circuit that calculates and outputs a floor difference du _j.

(4) a fuzzy control knowledge base as control fuzzy knowledge storage means for storing control fuzzy knowledge comprising a membership function and a fuzzy control rule, and (5) a control deviation e _j and a first order difference de of the control deviation. _This is a fuzzy control circuit that inputs _j , performs inference based on the control fuzzy knowledge stored in the fuzzy control knowledge base (4), and outputs a first-order difference du _j of the operation amount.

PID operation circuit (3) or fuzzy control circuit (5)
First-order difference du _j of the output operation amount from the adder (6)
In, the operation amount u _j is obtained by performing an addition operation of u _j = u _j-1 + du _j , and the operation amount u _j is added to the control object (1) to control the control object (1). Done.

The PID parameter memory (2), the PID operation circuit (3), the fuzzy control knowledge base (4), the fuzzy control circuit (5), the adder (6), and the PID operation circuit (3) ) Or fuzzy control circuit (5)
And the switches (7) and (7) for selectively connecting to the control device (8) for controlling the control target (1).

(9) is the P stored in the PID parameter memory (2).
A parameter / fuzzy knowledge conversion circuit as conversion means for generating control fuzzy knowledge comprising a membership function and a fuzzy control rule based on an ID constant, wherein the control fuzzy knowledge conversion circuit (9) generates the control fuzzy knowledge. The fuzzy knowledge is stored in the fuzzy control knowledge base (4).

(10) is an ideal response setting circuit (No. 1) that stores the ideal control response y _j゜ of the controlled object as a waveform or discrete values.
24, the response deviation e _j応 答 is also shown in FIG. 24), and is appropriately set by input means (not shown).

(11) calculates a control target value which is a characteristic amount such as an overshoot amount, an amplitude attenuation ratio (attenuation rate), and an arrival time from the ideal control response (waveform) stored in the ideal response setting circuit (10). Control target value calculation circuit, ideal response setting circuit (10)
When the control target value is set directly through the control target value calculating circuit (1), the control target value is selected.
There is no need to calculate the control target value from the ideal control response according to 1).

(12) is a control evaluation unit including a PID control evaluation circuit (13) as a PID control evaluation means and a fuzzy control evaluation circuit (14) as a fuzzy control evaluation means. enter the control response y _j from the circuit (10) or the control target value calculation circuit (11) output from the control target value and the control target (1) is its its features from the control response y _j over A chute amount, an amplitude attenuation ratio (attenuation rate), and an arrival time are calculated. Further, a control target deviation, which is a difference between each calculated characteristic amount and a corresponding control target value, is output. Further, a fuzzy control evaluation circuit (14) ) ideal control response y _j ° difference in a response error e _j ° and the response deviation e _j ° s sample difference of the previous e _js ° with control response y _j from the control object (1) (Change) de _j゜ is calculated and output.

(15) is a rule fulfillment degree storage means for storing the fulfillment degree of the fuzzy control rule established for the control response from the latest (current) sampling time j to s samples before in the fuzzy control circuit (5). The rule satisfaction degree stored in the rule satisfaction degree storage circuit (15) is output to a fuzzy control rule tuning input variable register described later as necessary.

(16) is the PID control evaluation circuit (13) of the control evaluation unit (12)
PID constant tuning input variable register that stores the control target deviation output from the controller. (17) is the control evaluator (12).
Response deviation output from the fuzzy control evaluation circuit (14)
e _j ° and response deviations e _j ° s previous sample e _js ° difference (variation) de _j °, and rules established level storage circuit (15)
Is a fuzzy control rule tuning input variable register that stores the rule fulfillment degree stored in.

(18) is the membership function of each control target deviation,
PI to eliminate the control target deviation according to the control target deviation
It is a fuzzy knowledge base for tuning PID constants in which fuzzy knowledge consisting of fuzzy rules for correcting D constants is stored.

(19), depending on the magnitude of the response error e _j ° and response deviations e _j ° s previous sample e _js ° variation de _j ° of, modifies the value of the consequent part of the fuzzy control rules Fuzzy knowledge base for fuzzy control rule tuning in which fuzzy rules and membership functions are stored.

(20) is stored in the PID parameter memory (2) according to the control target deviation output from the PID constant tuning input variable register (16) based on the fuzzy knowledge stored in the PID constant tuning fuzzy knowledge base (18). Corrects the stored PID constants, or based on the fuzzy knowledge stored in the fuzzy control rule tuning fuzzy knowledge base (19), the rule satisfaction degree output from the fuzzy control rule tuning input variable register (17), s This is a tuning fuzzy inference circuit that corrects the fuzzy control rules stored in the fuzzy control knowledge base (4) in accordance with the change de _j゜ in the response deviation from before the sample.

(21) is to control the control target (1) by the PID operation circuit (3) or the fuzzy control circuit (5), and to tune the PID constant when using the PID operation circuit (3). A switching circuit that controls the switches (7) (7) and (22) (22) (22) so that the fuzzy control rule is tuned by the fuzzy control circuit (5). The switching of control by the arithmetic circuit (3) or the fuzzy control circuit (5) is set.

First, correction of the PID constant in the process of controlling the control target (1) by the PID operation circuit (3) will be described. In this case, the switches (7) and (22) are all switched to the a side by the switching circuit (21).

Now, the PID operation circuit (3) calculates the proportional sensitivity P of the PID constant,
Given the integration time T _I and the differentiation time T _D , the set value r
Based on the control deviation e (t) = r (t) -y (t) defined by (t) and the control response output y (t), an operation amount u (t) represented by the following equation is calculated and output. I do.

In the control target (1), control is performed based on the given operation amount u (t), and a control response y (t) is output.

Now, a control target value including an overshoot amount OV ^* , an attenuation rate (amplitude attenuation ratio) DP ^* , and an arrival time RT ^* is set in the ideal response setting circuit (10), and these control target values are set in the control evaluation circuit (12). To the PID control evaluation circuit (13).

The control target value given to the PID control evaluation circuit (13) is, for example, as shown in FIG. 10 in consideration of stability near the target value (emphasis on external disturbance characteristics), quick response at the start of the plant, and transient characteristics. As shown, various types are conceivable, and a set of control target values is given as needed.

The PID control evaluation circuit (13) calculates the characteristic amounts of the control response overshoot amount OV, attenuation rate DP, and arrival time RT from the control response u (t). The overshoot amount is a ratio exceeding the set value r, the damping ratio is a ratio at which the amplitude of the control response attenuates with respect to the set value r, and the arrival time is a time from the start of the control to the time when the control response reaches the set value r. .

Taking the control response u (t) shown in FIG. 2 as an example, the characteristic amounts of the overshoot amount OV, the attenuation DP, and the arrival time RT are calculated as follows.

OV = 100 × | ov1 / r | (%) DP = 100 × | ov2 / ov1 | (%) RT (sec) Furthermore, the PID control evaluation circuit (13) uses the obtained feature and the ideal response setting circuit (10) A control target deviation, which is a difference from the input control target value, is obtained for each of the overshoot amount, the attenuation rate, and the arrival time. Each control target deviation E _OV , E _DP , E
_RT is obtained by the following equation.

E _OV = OV−OV ^* _EDP = DP−DP ^* _ERT = (RT−RT ^* ) / RT ^* PID control evaluation circuit (13) The control target deviation E _OV , E _DP , The _ERT is input to a PID constant tuning input variable register (16), and further input to a tuning fuzzy inference circuit (20) via a switch (22) as a fuzzy inference input variable.

In tuning the fuzzy inference circuit (20), the input control target deviation E _OV, E _DP, the E _RT as antecedent variables, PID
Variable tuning Fuzzy inference is performed based on the membership function stored in the fuzzy knowledge base (18) and the fuzzy rule so that the antecedent variable (input variable: control target deviation) is eliminated (becomes zero), and PID is performed. Outputs constant correction coefficients K _P , K _I , and K _D.

FIGS. 3, 4 and 5 show examples of membership functions of the antecedent variables (input variables) E _OV , E _DP and E _RT of the fuzzy rule for tuning, and FIGS. 6 and 7. , consequent variables in FIG. 8 (output variable: correction factor) K _P, K _I, an example of a membership function of K _D. In this example, the consequent variable is a real number (singleton), but may be fuzzy-divided and represented by a membership function like the antecedent variable.

Further, an example of a fuzzy rule for tuning is described in ninth.
Shown in the figure. In this figure, for example, the first fuzzy rule, means _{IF E OV = PB AND E DP} = PB THEN K P = PB AND K I = PB AND K D = NB.

Control target deviation E _OV , E _DP , E from PID control evaluation circuit (13)
_{When RT} is given, the fuzzy inference circuit for tuning (20) calculates the correction coefficients K _P , K _I , and K _D by the following equations in this example in which the consequent part is a real number.

Where n is the total number of fuzzy rules, control target deviation E _OV ,
When E _DP and E _RT are given, w _i is the degree of establishment of the i-th fuzzy rule, and h _Pi , h _Ii , and h _Di are K _P and K _I in the consequent part of the i-th fuzzy rule, respectively. a label (real value) regarding K _D. If the consequent is not a real value but a membership function, for example, MIN / MAX-
Output variable values (correction coefficients K _P , K _I , and K _D ) are calculated by an inference algorithm such as the centroid method.

When the correction coefficients K _P , K _I , and K _D are obtained, the tuning fuzzy inference circuit (9) corrects the PID constant using the correction coefficients.

The PID constants in the ^Nth control cycle are P ^N , T _I ^N , and T _D
And ^N, if the correction coefficient K _P ^N, K _I ^N, and ^{_{K D N, (N + 1}} )
PID constants P _{N + 1} , T _I ^{N + 1} , T _D in the second control cycle
^{N + 1} is calculated and modified as follows.

P ^{N + 1} = K _P ^N・ P ^N T _I ^{N + 1} = K _I ^N・ T _I ^N T _D ^{N + 1} = K _D ^N・ T _D ^N Modified PID constants P ^{N + 1} , T _I ^{N +1,} T _D ^{N + 1} is sent to the PID parameter memory (2), PID parameter memory (12)
From, while the tuning of the PID constant is not completed (the termination determination is not made), the corrected PID constants P ^{N + 1} , T _I ^{N + 1} , and T _D ^{N + 1} remain unchanged in the PID operation circuit (3). ).

Then, the PID operation circuit (3) calculates the given PID constant P
_{^{N + 1, T I N +}} 1, T D N + 1 using executes the (N + 1) th PID control.

As a result of the PID control, the control response x is, as described above, the PID
The control target deviation is obtained by input to the control evaluation circuit (13), and the tuning fuzzy inference circuit (20) corrects the correction coefficient and correction of the PID constant and proceeds with tuning.

The tuning cycle of the PID constant is the end determination constant ε
(> 0), the control target value deviation in the Nth cycle is
_{Assuming that} E _OV ^N , E _DP ^N , and E _RT ^N , (E _OV ^N , E _DP ^N |, 100 · | E _RT ^N |) <ε is satisfied until (E _OV ^N , E _DP ^N | ^N, 100 · either an absolute value of E _RT ^N until less than epsilon) is continued.

Then, when the tuning is completed (when the above equation is satisfied), a set of control target values (O
^{V *, DP *, RT *} ) set of optimal PID constants corresponding to the _{^{(P *, T I *,}} T D *) is determined (tuning is finished)
And, the PID constants _{^{P *, T I *, TD}} * is stored in the PID La meter memory (2). In FIG. 11, PID constant tuning is finished P ^*, T _I ^*, an example of a storage format of T _D ^*.
Here, the control deviation e is stored in accordance with the magnitude relationship with respect to the thresholds δ ₁ and δ ₂ , but at least a set of PID constants may be stored.

In this way, the PID constant for the control target (1) is optimized.

When the control target value is not directly given to the ideal response setting circuit (10) as described above, an ideal control response waveform is given to the ideal response setting circuit (10). The control response waveform is input by, for example, drawing coordinate axes and set values on a display (not shown), inputting representative coordinates of the response waveform from a keyboard (not shown), and performing interpolation processing (for example, spline interpolation), a mouse, or a mouse. This is performed by drawing an ideal control response waveform directly on the display screen using a light pen or the like, or by inputting a discrete numerical value.

Then, the set ideal control response waveform y ゜ (t)
From the ideal control response waveform, the control target value (ideal) is calculated by the control target value calculating circuit (11) using the same equation as that for calculating the characteristic amount from the control response in the PID control evaluation circuit (13). OV ^* , DP ^* , and RT ^* are calculated. The calculated control target value is stored in the PID control evaluation circuit (1
3), and the control target deviation is calculated as described above.

Next, based on the optimized PID constants, a fuzzy control rule consisting of a fuzzy control rule and a membership function for fuzzy control by a fuzzy control circuit (5) in a parameter / fuzzy knowledge conversion circuit (9) is used. Will be described.

FIG. 12 shows a parameter / fuzzy knowledge conversion circuit (9).
FIG.

(91) is a division information storage unit that stores the number of fuzzy divisions of an input variable serving as an antecedent variable of a fuzzy rule and a range thereof (for example, a maximum value and a minimum value that the variable can take) for each input variable. In the division information register (91), respective values are input and stored by operation of input means such as a keyboard (not shown).

(92) An input variable dividing means for fuzzy dividing each input variable into a set number of divisions based on the division information of the input variables stored in the division information register (91) and generating a standard memberive function. Is an input variable dividing circuit.

In (93), the representative value of each label of the input variable divided by the input variable dividing circuit (92) (the value of the degree of satisfaction of the membership function is 1) is input and stored in the PID parameter memory (2). A fuzzy rule generation circuit as a consequent part determining means for calculating an output value as a consequent part of the fuzzy control rule from the PID constant and generating a fuzzy control rule. The consequent part is calculated in accordance with the hyperplane feedback rule set based on this.

Now, ^* optimized PID constants P by tuning the fuzzy inference circuit ^(20), T _I ^*, T _D ^* is stored in the PID parameter memory (2), the feedback law P :( based on this PID constants e, de, d ² e) → du is represented by a hyperplane du = T _I ^* .e + P ^* .de + T _D ^* d ² e in a four-dimensional orthogonal space [e × de × d ² e × du]. .

PID by parameter / fuzzy knowledge conversion circuit (9)
When generating fuzzy control knowledge based on constants, the range of the input variables (all or part of e, de, d ² e) to be the antecedent variables of the fuzzy control rule from input means (not shown) Defined by the maximum and minimum possible values)
Then, the division number is input and set to the division information register (91).

When the division information of the input variables is stored in the division information register (91), the input variable division circuit (92) creates a membership function in accordance with the standard division of each input variable and the division.

For example, for e, de, d ² e, the range (maximum value, minimum value) is (e _max , −e _max ), (de _max , −de _max ),
_Assuming that (d ² e _max , −d ² e _max ) is set to 7 as the number of divisions, the range (domain) is divided into seven equal parts on the axis of the input variable as shown in FIG. Then, for each of the divided input variables, a label is attached to the divided partial area by the number of divisions, and a standard membership function is generated using each input variable axis as a set. The standard membership functions are:
For example, as shown in FIG. 13, the midpoint of the divided partial area on each input variable axis is defined as the vertex of the degree of establishment 1, and the midpoint (point of the degree of establishment 0) and the vertex of the two adjacent partial areas are defined as the vertex. Generate a connected isosceles triangle. However, the maximum and minimum membership functions of the partial areas are not isosceles triangles but trapezoidal ones. However, the present invention is not limited to this, and a bell-shaped membership function in which the middle point of the partial area is set as the vertex of the degree of establishment 1 may be used as a standard membership function.

Further, the input variable dividing circuit (92) generates a representative value of each label of the generated membership function, for example, when the degree of establishment of the membership function is 1, for each input variable serving as a variable of the antecedent of the fuzzy control rule. The value, that is, the value of the midpoint of each partial area on the input variable axis, is output to a fuzzy rule generation circuit (93) together with a standard type fuzzy control rule according to the number of divisions of the input variable (antecedent variable). .

The standard type of fuzzy control rule is that in the PID control system, L first antecedent variables e, M second antecedent variables de,
When the third antecedent variable d ² e is fuzzy divided into N,
The fuzzy control rule R _jk is given by R _jk : IF e = e _i AND de = de _j AND d ² e = d ² e _k THEM du = du _ijk i = 1,..., L, j = 1 _,. k = 1,..., N (each corresponding to a label).

In this case, the PID control system is composed of three antecedent variables e, de, and d ² e. However, in the PI control system and the PD control system, the antecedent variables are e, de, and de, respectively. And d ² e, and the fuzzy control rules R _ij and R _jk are as follows: R _ij : IF e = e _i AND de = de _j THEN du = du _ij i = 1,..., L, j = 1 _,. : IF de = de _j AND d ² e = d ² e _k THEN du = du _jk j = 1,..., M, k = 1 _,.

Now, in the hyperplane feedback rule generation unit (9), PI
PID constants P ^* and T _I stored in the D parameter memory (2)
^{From *} and T _D ^* , a feedback rule du = T _I ^* e + P ^* de + T _D ^* d ² e based on these constants is generated and held.

Then, when the representative value of each label of the antecedent variable and the standard type fuzzy control rule are input to the fuzzy rule generation circuit (93), the fuzzy rule generation circuit (93) sets each fuzzy control rule (i = 1, ..., L, j = 1, ..., M, k = 1
.., N) are substituted into the feedback rule for the representative value of the antecedent variable, and du is calculated. The calculated du is used as the real value of the consequent in the fuzzy control rule.

That is, the real value of the consequent part in each fuzzy control rule is du in accordance with the above-mentioned feedback rule, du _ijk = T _I ^* e _i + P ^* de _j + T _D ^* d ² e _k i = 1,. , L, j = 1,..., M, k = 1,.

The fuzzy rule generation circuit (5) calculates the real value du of the consequent part.
Is calculated, the state part (e _i , de _j , d ² e _k ) of the antecedent variable of the standard fuzzy control rule is replaced with the label of the representative value substituted as the antecedent variable, and A fuzzy control rule is generated by replacing the output value (du _ijk ) with the calculated real value.

Thus, the generated fuzzy control rule is used together with the membership function generated by the input variable dividing circuit (92),
It is stored in the fuzzy control knowledge base (4). This automatically builds a fuzzy control knowledge base that is very close to the hyperplane feedback rule based on the optimized PID constants.

In the PI control system or the PD control system instead of the PID control system, the real value du of the consequent part is calculated from the representative value of each antecedent part variable (input variable).

For example, in the PI control system, du _ij = T _I ^* · e _i + P ^* · de _ji = 1,..., L, j = 1,..., M, and in the PD control system, du _jk = P ^* · De _j + T _D ^* · d ² e _k i = 1,..., M, k = 1,.

Next, tuning of fuzzy control knowledge generated by the parameter / fuzzy knowledge conversion circuit (9) and stored in the fuzzy control knowledge base (4) will be described. In this case, the switches (7) and (22) are all switched to the b side by the switching circuit (21).

Here, for the sake of simplicity, the fuzzy control rule of the generated fuzzy control knowledge uses a simple inference such that the consequent part is a real value. The fuzzy control rule is based on the control deviation e _j and the control deviation 1
The floor difference de _j is taken, and the first-order difference du _j of the operation amount is taken as the consequent variable.

Although fuzzy division of the first-order difference de _j of the control deviation e _j and the control deviation is arbitrary, here, the control deviation e _j and the control deviation 1
As shown in FIG. 14, the floor difference de _j is seven (NB: ne
gative big, NM: negative medium, NS: negative smal
l, ZO: zero, PS: positive small, PM: positive mediu
m, PB: Positive big)
It is assumed that fuzzy control rules have been set. Of course, further rules may be added to the blank portion in FIG.

The fuzzy control rule shown in FIG. 15 is expressed as follows.

However, the sample point of time of the most recent (current) and _{j, e j = r-y} j, de j = e j -e j-1 du j = u j -u j-1 r: set value, y _j: Control Response, e _j : control deviation, de _j : first-order difference of control deviation, du _j = first-order difference of operation amount u _j .

The membership function of the antecedent part is as shown in FIG. 14, and the membership function of the consequent part is shown in FIG. In this embodiment, the membership function of the consequent is a real value h _i (i
= 1 to 7).

In the fuzzy control circuit (5), when inputs e _j and de _j are given, an output du _j is obtained by the following equation.

Here, w _i is the rule fulfillment degree of the i-th rule with respect to e _j and de _j .

Then, the first order difference du of the operation amount obtained in this manner is
As described above, _j is the sum of the manipulated variable u _{j-1 and the} sum at the previous sample time point j−1, and u _j = u _j−1 + du _j is performed by the adder (6). An operation amount u _j is obtained, and the operation amount u _j is added to the control object (1) to control the control object (1).

Such tuning of the fuzzy control rule is performed so that the control response y _J observed at the current sampling point j matches the ideal response y _j゜ that has been previously set and stored in the ideal response setting circuit (10). The correction value of the real value of the consequent part of the fuzzy control rule used s samples before the current sample time j considered to affect the current control state is the response deviation e _j゜ and the response between the s samples. Deviation change
It is obtained by fuzzy inference according to the sign of de _j゜ and its value, and is performed by increasing or decreasing the real value of the consequent part of the control rule.

Here, e _j゜ = y _j −y _j゜ de _j゜ = e _j゜ −e _js゜ e _j゜: response deviation, de _j゜: change in response deviation between s samples, y _j゜: ideal Response, y _j : Control response.

That is, the control target (1) by the fuzzy control circuit (5)
When the control is executed, the fuzzy control evaluation circuit (14) of the control evaluation unit (12) stores the control response y _J from the control target (1) and the ideal response y _j stored in the ideal response setting circuit (10). It is゜Ga input response deviation e _j ° and the response deviation e _j ° s previous sample e _js ° difference (variation) de _j゜Wo calculated.
These response deviations e _j ° and the response deviation e _j ° s previous sample e _js ° difference (variation) de _j ° is, is output to the fuzzy control rules tuning input variable register (17) is held .

Also, the feasibility of each fuzzy control rule at each sampling for fuzzy control in the fuzzy control circuit (5) is stored in the rule fulfillment degree storage circuit (15), and the fuzzy control rule tuning input variable register (17) is stored. ) Also contains the rule fulfillment degree from the rule fulfillment degree storage circuit (15). FIG. 23 shows an example of the stored contents of the rule satisfaction degree storage circuit (15).

Then, from the fuzzy control rule tuning input variable register (17), the response deviation e _j゜ and the response deviation
The difference (change) between e _j゜ and e _js前 before s samples
_j゜ and the rule fulfillment degree are input to the tuning fuzzy inference circuit (20) through the switch (22) as input variables for fuzzy inference.

Tuning the fuzzy inference circuit (20) is responsive deviation entered e _j ° and the response deviation e _j ° s previous sample e _js ° difference (variation) in accordance de _j °, fuzzy control rules Fuzzy reasoning is also performed to optimize.

For example, taking the Figure 20 A as an example, in FIG. 20 A, "the present time j in response deviation e _j ° is the NB, and s sample preparation response deviation compared to further decrease (de _j ° N), it tends to move away in the direction of a value smaller than the ideal response, so it is considered that the manipulated variable before s samples was too small, and s
The value of the consequent part of the control rule used before the sample must be greatly increased. "

Expressing this as a fuzzy control rule tuning fuzzy rule, at time j, IF e _j゜ is NB, de _j゜ is N, THEN dh _j is PB where dh _i : the consequent of the i-th fuzzy control rule Part value h
It is expressed as the correction amount of _j .

The fuzzy division of the variables in the antecedent and consequent parts of the fuzzy rule for fuzzy control rule tuning is also arbitrary, but FIG. 17 shows an example of the fuzzy division of the response deviation e _j 、 and FIG. An example of fuzzy division of _j゜ is shown in FIG. Further, membership functions of dh _i of the consequent part, as shown in FIG. 19, since using a simplified inference even in fuzzy inference for tuning, and replaced with a real value p _i.

In other words, PB because corresponds to p _7, above the tuning fuzzy rules, and IF e _j ° is NB, de _j ° is N, and THEN dh _i is p _7.

FIGS. 17 and 18 show how to generate a control response y _j corresponding to the ideal response y _j対 応 corresponding to all the fuzzy divisions shown in FIG.
20 Shown in FIGS. The fuzzy rules for fuzzy control rule tuning corresponding to FIG.
It looks like Figure 21.

This tuning fuzzy rule does not need to be described in all cases as shown in FIG.
As shown in FIG. 22, the description may be only for typical patterns.

The fuzzy inference circuit for tuning (20) includes a membership function (FIGS. 17 and 18) stored in the fuzzy knowledge base for fuzzy control rule tuning (FIGS. 17 and 18).
The response deviation e calculated by the fuzzy control evaluation circuit (14) of the control evaluation unit (12) based on the fuzzy control rule tuning fuzzy rule of FIG. 21 (or FIG. 22).
_j゜ and the change in response deviation (first-order difference in response deviation) d
When e _j゜ is given, by simple inference, the modification amount dh _i of the consequent part of the control rule is (However, the total number of tuning fuzzy rules is m, and the k-th tuning fuzzy rule is e _j 、,
The degree of rule establishment for de _j゜ is _defined as μ _k ).

Then, the value h _i of the consequent part of the fuzzy control rule is corrected by the following equation at the current sampling point j for all the fuzzy control rules used for calculating the manipulated variable before s samples. .

h _i ^NEW = h _i ^OLD + w ^(js) · dh _i where w ^(js) : the degree of satisfaction of the first fuzzy control rule at (j−s) h _i ^OLD : used at (j−s) The value of the consequent part of the modified fuzzy control rule h _i ^NEW : The value of the consequent part of the fuzzy control rule modified at the time point j.

w ^(js) is the first-order difference of the manipulated ^variable at the point (j−s)
When calculating du _(js) , the rule fulfillment degree obtained for each fuzzy control rule in the fuzzy control circuit (5) is used as a rule fulfillment degree storage circuit (15), and a fuzzy control rule tuning input variable register ( It is input to the fuzzy inference circuit for tuning (20) via 17).

The value of the consequent part of the fuzzy control rule used at the point (j−s) in the fuzzy control knowledge base (4) is corrected to the h _i ^NEW thus obtained.

Tuning of such fuzzy control rules, for example, During the control of the control target (1), until the end criterion indicated by is satisfied, the fuzzy control rule after the previous tuning is used as an initial rule until the criterion is satisfied.
Real-time tuning is repeated. Here, φ is the value of the end judgment criterion.

(G) Effect of the present invention As is apparent from the above description, the present invention provides a fuzzy correction of a PID constant using a control target deviation, which is a difference between a feature value obtained from a control response and a control target value, as an input variable. Even if different control targets are given according to different control situations by performing inference. Rather than preparing multiple tuning knowledge bases in advance for each control target as in the past, only one type of fuzzy knowledge base for tuning PID constants can be used for any control suited to the control target or its characteristic process. PID constants can be optimized to achieve goals.

Then, by the parameter / fuzzy knowledge conversion circuit,
According to the number of divisions and range of the antecedent variable that is arbitrarily set,
A fuzzy control rule that performs control that is very close to the feedback rule based on the optimized PID constant is automatically generated. That is, the construction of the fuzzy control rules, which has been performed manually, is automated. Also, at this time, a fuzzy control rule that performs a control that is very close to the feedback rule based on the optimized PID constant is constructed,
At least the same control as the PID control can be performed, and it is possible to perform better control than the fuzzy control rule constructed along the state trajectory of the conventional control amount.

Furthermore, by applying the fuzzy inference to the inference of the correction amount of the consequent part of the fuzzy control rule, the response deviation e _j゜ and the variation of the response deviation de _j゜ (the first-order difference of the response deviation) can be reduced. Accordingly, more precise tuning of the fuzzy control rules is performed. Then, tuning when the difference between the ideal response and the response according to the initial control rule (initial response deviation) is large to some extent becomes possible, and the robustness of tuning and the convergence efficiency can be improved.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of the device of the present invention, FIG. 2 is a schematic diagram showing an example of a response waveform of a control output according to an embodiment of the present invention, and characteristic amounts, and FIGS. FIG. 6 is a diagram showing an embodiment of a membership function of an input variable in fuzzy inference for tuning of a PID constant according to an embodiment of the present invention. FIGS. 6 to 8 show tuning of a PID constant according to an embodiment of the present invention. FIG. 9 is a diagram showing an example of a membership function of an output variable in fuzzy inference for the purpose, FIG. 9 is a diagram showing an example of a fuzzy rule for tuning a PID constant according to an embodiment of the present invention, and FIG. FIG. 11 is a diagram showing a control target value with respect to the magnitude of a control deviation in fuzzy inference for tuning of a PID constant according to an example. FIG. 11 is a diagram showing a PID constant with respect to the magnitude of a control deviation according to an embodiment of the present invention. FIG. 12 shows an embodiment of the present invention. FIG. 13 is a diagram showing a membership function of an antecedent variable in generating a fuzzy control rule according to an embodiment of the present invention, and FIG. FIG. 15 is a diagram showing a membership function of the antecedent part of the fuzzy control rule according to the present invention, FIG. 15 is a diagram showing a fuzzy control rule according to an embodiment of the present invention, and FIG. FIGS. 17 and 18 show membership functions (real values) of the consequent part, and FIGS. 17 and 18 show membership functions of the antecedent part of the fuzzy rule for fuzzy control rule tuning according to an embodiment of the present invention. FIG. 19 is a diagram showing a membership function (real value) of a consequent part of a fuzzy rule for fuzzy control rule tuning according to an embodiment of the present invention.
FIG. 21 is a diagram showing a state of a control response to an ideal response, FIGS. 21 and 22 are diagrams showing a fuzzy rule for fuzzy control rule tuning according to an embodiment of the present invention, and FIG. FIG. 24 is a diagram showing an example of contents stored in a rule satisfaction degree storage circuit, and FIG. 24 is a diagram showing an example of an ideal response set in an ideal response setting circuit according to an embodiment of the present invention. (1) ... controlled object, (2) ... PID parameter memory (parameter storage means), (3) ... PID operation circuit (PID
Calculation means), (4) fuzzy control knowledge base (control fuzzy knowledge storage means), (5) fuzzy control circuit (fuzzy control means), (9) parameter / fuzzy knowledge conversion circuit (conversion means) ), (10) ... ideal response setting circuit (ideal response storage means), (11) ... control target value calculation circuit (control target value calculation means), (12) ... control evaluation section, (13) ... PID control evaluation circuit (PID control evaluation means),
(14)… fuzzy control evaluation circuit (fuzzy control evaluation means), (15)… rule satisfaction degree storage circuit (rule satisfaction degree storage means), (18)… fuzzy knowledge base for PID constant tuning, (19) ... Fuzzy control rule tuning fuzzy knowledge base, (20)... Tuning fuzzy inference circuit (tuning fuzzy inference means), (21)... Switching circuit, (91)... Division information register (division information storage means) ), (92) ... an input conversion division circuit (input variable division means), (93) ... a fuzzy rule generation circuit (consequent part determination means), (94) ... a hyperplane feedback rule generation section.

Continuation of front page (56) References JP-A-2-186402 (JP, A) JP-A-62-143103 (JP, A) JP-A-62-135902 (JP, A) JP-A-4-76702 (JP, A) , A) JP-A-1-293401 (JP, A) JP-A-1-258003 (JP, A) Patent 2532967 (JP, B2) Patent 2532976 (JP, B2) Mikio Maeda, one other, "self-adjusting fuzzy Controllers, Transactions of the Society of Instrument and Control Engineers, February 1988, Vol. 24, No. 2, p. 191-197 (58) Field surveyed (Int. Cl. ⁶ , DB name) G05B 13/00-13/04 G06F 9/44 554

Claims (3)

(57) [Claims] A parameter storage means for storing at least one PID constant for PID control, and a control of a control target in accordance with a control response from the control target based on the PID constant stored in the parameter storage means. PID calculating means for performing, and a PID for obtaining a characteristic amount for evaluation from a control response of a control target and outputting a control target deviation which is a difference between the characteristic amount and a control target value.
A control evaluation means, a fuzzy knowledge base for tuning PID constants for tuning the PID constants according to the control target deviation, and a fuzzy control fuzzy comprising membership functions and fuzzy control rules based on the PID constants. Conversion means for generating knowledge, control fuzzy knowledge storage means for storing the control fuzzy knowledge generated by the conversion means, and a control object based on the control fuzzy knowledge stored in the control fuzzy knowledge storage means Fuzzy control means for controlling a control object in accordance with a control response from the control means, a rule satisfaction degree storage means for storing the degree of satisfaction of a fuzzy control rule for the control response in the fuzzy control means, and an ideal response of the control object. Ideal response storage means for storing the control response of the control object and the ideal response stored in the ideal response storage means. Fuzzy control evaluation means for outputting the response deviation as the difference and the magnitude of the change, and a fuzzy control rule storing fuzzy knowledge for tuning the fuzzy control rule according to the response deviation and the magnitude of the change A fuzzy knowledge base for tuning, and a PID constant stored in the parameter storage means are corrected according to the control target deviation output from the PID control evaluation means based on the fuzzy knowledge stored in the fuzzy knowledge base for PID constant tuning. Alternatively, based on the fuzzy knowledge stored in the fuzzy knowledge base for fuzzy control rule tuning, control is performed in accordance with the satisfaction degree stored in the rule satisfaction degree storage means, the response deviation output from the fuzzy control evaluation means, and the magnitude of the change. The fuzzy control rules stored in the fuzzy knowledge storage Fuzzy knowledge builder, characterized in that it comprises a tuning fuzzy inference means. 2. The division means stores division information and a range of an input variable serving as an antecedent variable, and an input variable according to the division number and the range stored in the division information storage means. Variable dividing means for fuzzy dividing the input variables to generate membership functions of the input variables, a representative value of each divided portion of the input variable divided by the input variable dividing means, and a value of the parameter stored in the parameter storing means. 2. A fuzzy knowledge construction apparatus according to claim 1, further comprising: a consequent part determining means for calculating a consequent part of the fuzzy control rule according to the above formula and generating a fuzzy control rule. 3. A control target value calculation means for obtaining a control target value from an ideal response stored in said ideal response storage means and outputting the control target value to said PID control evaluation means. The described fuzzy knowledge construction device.