JP3177995B2 - Fuzzy controller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fuzzy controller that outputs a determined value according to an input variable according to a predetermined fuzzy rule.

[0002]

2. Description of the Related Art With respect to handling of membership functions in fuzzy inference, there are a MIN-MAX operation method and an algebraic operation method. The former selects a fuzzy rule having the label to which the input variable belongs in the antecedent part, and changes the membership function of the label of the consequent part of the selected fuzzy rule to the minimum value of the membership of the input variable in the membership function of the antecedent part The area of the membership function after the truncation is obtained for all the selected fuzzy rules, and these are added.
The position of the center of gravity of this area range is determined as a definite value.

On the other hand, in the algebraic operation method, the product of the membership of the input variables in the antecedent membership function is obtained for each fuzzy rule, and the value obtained by adding the product of the membership for each consequent label is high. As a matter of fact, the consequent part membership functions are modified, and these are added for all the consequent part labels of the selected fuzzy rule, and then the center of gravity of the range is obtained. In the MIN-MAX operation method, the center of gravity position is obtained after deleting the overlapping part of the adjacent label in the consequent part membership function truncated.

Regardless of whether the MIN-MAX arithmetic method or the algebraic arithmetic method is used, when a fuzzy controller is constituted by an analog circuit, a truncation part for obtaining the membership of the consequent part membership function, In the defuzzification unit that performs addition and calculation of the position of the center of gravity, the horizontal axis of the membership function of the consequent part is divided by a considerable number according to the control accuracy, and the value of the vertical axis of the membership function at each division position is defined as a voltage value. A circuit configuration that can be set and held is generally used.

A digital fuzzy controller is used because it is suitable for setting a large-scale rule, further facilitates the correction of the rule, and is suitable for being incorporated into a computer network. In order to realize real-time inference in such a digital fuzzy controller, there is a fuzzy inference defuzzification unit using an area equalization method that does not require a product-sum calculation. However, if the sum of products is not calculated in the defuzzification unit, an accurate inference result may not be obtained in some cases, and depending on a controlled object, it may not be possible to converge.

[0006] Therefore, in advance calculating all the fuzzy inference offline and stored in ROM2 area of the one-chip microcomputer 11 indicating the result of the operation in FIG. 5, for example, the input variables from the sensor S ₁ to S _n A fuzzy ROM method has been proposed in which a value written in a ROM area is read out and output using the data as an address. By using this fuzzy ROM method, it is possible to output an inference value with sufficient accuracy to converge the control target in real time.

[0007]

However, in the conventional analog type fuzzy controller, it is necessary to increase the number of divisions on the horizontal axis of the consequent membership function in order to improve control accuracy. Causes an increase in the number of constituent transistors in the truncation section and the defuzzification section, resulting in a problem that the overall circuit configuration becomes complicated and large.

In addition, the digital fuzzy controller using the fuzzy ROM method has a problem that the storage capacity of a required ROM increases with an increase in input variables and a fuzzy rule, and a high-precision inference. There is a problem that it is necessary to execute complicated arithmetic processing such as interpolation calculation in order to obtain a result.

SUMMARY OF THE INVENTION An object of the present invention is to provide a semi-real-time inference type fuzzy controller with high accuracy and quick response without increasing the number of constituent transistors and the storage capacity of a ROM.

[0010]

According to a first aspect of the present invention, there is provided a fuzzy controller for determining a moment value and an area value of a membership function of a consequent part truncated according to a degree of affiliation obtained from an input variable. The first and second storage areas stored for each label of the consequent part and for each overlapping part of adjacent labels using the degree as a parameter, the label of the consequent part of the fuzzy rule including the input variable in the antecedent part, and Reading means for reading, from the first storage area and the second storage area, a moment value and an area value corresponding to the degree of affiliation for an adjacent label; and a first storage area for each of the moment value and the area value corresponding to the degree of affiliation. Subtraction means for subtracting the value of the second storage area from the value of the second storage area, and division means for dividing the moment value obtained by the subtraction means by the area value to obtain a definite value. Characterized in that was.

According to a second aspect of the present invention, there is provided a fuzzy controller comprising: a moment value and an area value of a membership function of a consequent part deformed by a membership obtained from an input variable; Reading means for reading, from the storage area, a storage area for storing the label of each part, a moment value and an area value corresponding to the degree of belonging of the consequent part of the fuzzy rule including the input variable in the antecedent part, and the reading means. Dividing means for dividing the obtained moment value by the area value to obtain a definite value.

According to a third aspect of the present invention, there is provided a fuzzy controller for generating an antecedent label to which an input variable belongs in an antecedent membership function and a degree of belonging of the input variable in the antecedent label. A fuzzy rule including the consequent part label is selected, and an antecedent part arithmetic circuit that sets the degree of conformity for the consequent part label based on the degree of affiliation, and a weight value unique to the consequent part label. The weight value output part to be output, and the consequent part membership relation of a triangle whose total area set for each consequent part label is known.
An area calculation unit that calculates an area value using the number and the degree of conformity as parameters, and a volume that calculates a sum of volume values that are products of the area value obtained by the area calculation unit and the weight value output from the weight value output unit. An arithmetic unit, and a consequent part arithmetic circuit having a division unit that obtains a definite value by dividing the sum of the volume values obtained by the volume value arithmetic unit by the sum of the area values, wherein the area arithmetic unit is ,
Calculating the area value by the following procedure.
You. (1) Calculate the complement value of the above-mentioned goodness of fit, and square this value to a constant
Consequence consequently truncated at the fitness by multiplying by K
Find the area of the triangle above the part membership function. (2) Determine the area of the upper triangle by the membership of the consequent part.
Is subtracted from the total area of the
The area of the trapezoid on the lower side of the function is obtained as the area value.

According to a fourth aspect of the present invention, there is provided a fuzzy controller, wherein the antecedent operation circuit includes a number of fitness operation units corresponding to the number of fuzzy rules that can be simultaneously selected, and an input unit according to the selected fuzzy rule. Analog tomato that dynamically switches between the fitness calculation unit and the output unit
Rix, and

[0014]

According to a fifth aspect of the present invention, in the fuzzy controller, the consequent part operation circuit outputs a weight value unique to the superposed part of the adjacent consequent part label, and the overlap area weight value output section; And an overlap section subtracting means for subtracting a value of an overlap section of an adjacent consequent part label in the volume operation section.

[0016]

According to the first aspect of the present invention, the moment value and the area value of the membership function of the consequent part truncated based on the degree of affiliation determined from the input variables are used as the parameters of the consequent part with the degree of affiliation as a parameter. Each label is stored in the first storage area. In addition, the moment value and the area value of the consequent part membership function, which are truncated by the degree of affiliation determined from the input variables for each overlapping part of the adjacent label of the consequent part, are assigned to each overlapping part using the degree of affiliation as a parameter It is stored in the second storage area. The subtraction means reads a moment value and an area value corresponding to the degree of belonging from each of the first storage area and the second storage area, and reads the moment value and the area value of the second storage area from the moment value and the area value of the first storage area. Subtract the moment and area values. The dividing means divides the result of the moment value subtraction obtained by the subtracting means by the result of the area value subtraction to obtain a definite value. Therefore, the moment value and the area value are read from each of the first storage area and the second storage area using the truncation value obtained from the input variable as a parameter, and the definite value is obtained by subtracting and dividing these values. can get. In this case, there is no need to store data corresponding to all combinations of input variables in the first and second storage areas, and when performing fuzzy inference by the MIN-MAX operation method, a so-called fuzzy ROM system is used. It is possible to obtain a highly accurate inference result while reducing the storage capacity in comparison.

According to the second aspect of the present invention, the moment value and the area value of the membership function of the consequent part deformed by the degree of affiliation obtained from the input variables are used as labels of the consequent part using the degree of affiliation as a parameter. Each time it is stored in the storage area. The reading means reads from the storage area the moment value and the area value corresponding to the degree of belonging for the label of the consequent part of the selected fuzzy rule, and the dividing means divides the moment value obtained by the reading means by the area value. Obtain a fixed value. Therefore, a definite value can be obtained by an algebraic operation method based on the degree of affiliation obtained from the input variables. In this case, there is no need to store data corresponding to all combinations of input variables in the storage area, and the storage capacity is not limited to the so-called fuzzy ROM method, but also to the digital method based on the MIN-MAX operation method. It is possible to obtain a highly accurate inference result while reducing the size.

According to the third aspect of the present invention, in the antecedent part membership function, the antecedent part label and the degree of affiliation to which the input variable belongs are generated by the fuzzification circuit, and the antecedent part generated by the fuzzification circuit is generated. The selection of the fuzzy rule including the label and the setting of the conformity for the consequent part label are performed in the antecedent part operation circuit. The consequent label and the set conformity of the fuzzy rule selected in the antecedent operation circuit are input to the consequent operation circuit. The consequent part operation circuit includes a weight value output part, an area operation part, a volume operation part and a division part. The weight value output unit outputs a unique weight value for each consequent part label. The area calculation unit calculates an area value for each consequent part label using the degree of conformity as a parameter. The volume calculation unit obtains a volume value by multiplying the area value obtained by the area calculation unit by the weight value output from the weight value output unit.

The divider divides the volume value obtained by the volume calculator by the area value obtained by the area calculator, and outputs the result of the division as a definite value.

Therefore, by setting the membership function of the consequent part to the same triangular wave shape for each label and giving the total area value for one consequent part label, the conformity can be reduced in the area calculation after the truncation calculation. The area value after truncation can be easily calculated for each consequent part label as a parameter, and the integral processing which is a general area calculation method in an analog circuit is unnecessary. In the defuzzification operation, a weight value is set for each consequent part label, a weight value is added to each consequent part label from the area value after truncation to obtain a volume value, and the obtained volume value is calculated as an area. It is only necessary to divide by a value, and the horizontal axis of the membership function is divided into a number of division points as is performed in a fuzzy rule composed of a general analog circuit, and the setting and holding of the vertical axis value at each division point Becomes unnecessary.

Further, according to the configuration described in claim 4,
Input part, conformity according to the selected fuzzy rule
Arithmetic unit and output unit are based on analog gate matrix
Dynamic switching can simplify the configuration of the arithmetic circuit.
You.

[0022]

According to the above configuration, when the MIN-MAX operation is performed on the consequent part, the area value and the volume value of the overlapping part of the adjacent consequent part label can be easily obtained. In addition, a circuit for setting and holding a vertical axis value at a division point on a horizontal axis of a membership function of a fuzzy controller constituted by a general analog circuit becomes unnecessary.

[0024]

DESCRIPTION OF THE PREFERRED EMBODIMENTS a. The embodiment of the invention described in claim 1 (MI
FIGS. 1A and 1B show the contents stored in the first and second storage areas of a fuzzy controller according to an embodiment of the present invention. Is shown. The ROM 12 of the microcomputer 11 constituting the fuzzy controller shown in FIG. 5 is divided into first and second storage areas. In the first storage area, as shown in FIG. 1A, for each label of the consequent part, the head cutoff value (degree of membership) μ _i is used as a parameter, and the moment of the truncated consequent part membership function is used. The value S _i and the area value A _i are stored. That is, M
In the fuzzy inference by the IN-MAX operation method, the label of the membership function of the consequent part is selected according to the label of the input variable of the antecedent part, and the selected label is the minimum value of the degree of membership of the antecedent part. is head cut in Atamasetsuchi μ _i.
The area A _{i of the} trapezoid part truncated by the truncation value μ _i and the moment value S _i obtained by multiplying the area value A _i by the distance x from the reference position of the center of gravity of the trapezoid part at each unit interval Is calculated in advance offline with respect to the truncated value μ _i of
Is stored in the first storage area. Here, the height of the membership function of the consequent part is divided into 256.

In the second storage area, as shown in FIG. 1B, the moment value S _j and the area value A _j of the overlapped portion of the membership function truncated at the adjacent label are truncated. The value μ _j is stored as a parameter for each overlapping portion. That is, in the second storage area, the moment value and the area value of the portion indicated by hatching in the label i and the label i 'shown in FIG. It is remembered.

FIG. 3 is a flowchart showing the processing procedure of the fuzzy controller.

The microcomputer 1 that constitutes the fuzzy controller reads as input variables the detection result of the single or multiple sensors S ₁ ~S _n (n1), performs calculation of the antecedent adaptation degree based on the input variables At the same time, a truncation value is calculated for the membership function of the consequent part (n3). The processing of n1 to n3 is the same as that of a known digital fuzzy controller. Then, for the label of the consequent part of the set fuzzy rule, the moment value S _i corresponding to the truncated value is read from the first storage area shown in FIG. Are added (n
4). Further, similarly to the label of the consequent portion reads the area value A _i corresponding to the head switching value, adds the area value of the consequent part membership function is off head for all fuzzy that has been set (n5) .

When the above processing of n4 and n5 is completed for all the labels (n6), the head cutoff value μ _j of the label selected in the consequent part which has the smaller head cutoff value of the labels adjacent to each other is selected. reads out the moment value S _j and the area value a _j corresponding to the head Setsuchi mu _j for the central portion of the label adjacent the second storage area shown in FIG. 1 (B), adds these for all polymerization unit (N8).

The moment value S _{i of the} membership function of the truncated consequent part obtained by the processing of n4 to n6.
And from the area value A _i, subtracting the moment value S _j and the area value A _j of the polymerization unit of each label obtained in the process at n7~n10 (n11, n12). Further, the calculation result S of the moment sum is divided by the calculation result A of the area sum to obtain a definite value D of fuzzy inference (n13). In the above processing, n
11, n12 correspond to the subtraction means of the present invention, and n13 also corresponds to the division means.

By the above processing, for example, as shown in FIG. 4, the labels NS, ZR,
When PS is selected, each label is truncated by a truncation value μ _NS , μ _ZR , μ _PS , and the truncated consequent part membership function range abcdcef is set. -Gh-
The position of the center of gravity in the range of “a” is a definite value that is an inference result of fuzzy inference. In this embodiment, in order to obtain the final value, the moment values and the area values of the labels NS, ZR, and PS are added, and the overlap result of the adjacent label, that is, the area c-n-lm- c and the region fkl-
The moment value and the area value of off are each subtracted,
The moment value after the overlap is subtracted is divided by the area value. The moment value and the area value in each label of the truncated consequent part membership function are the first.
The head value μ _i is stored as a parameter in the storage area, and the corresponding moment value S _i and area value A _i are read out using the label of the consequent part in the fuzzy rule as the upper address and the head value as the lower address. It is. Also,
For the overlapping part of the adjacent labels, the moment value S _j and the area value A _j are stored in the second storage area using the truncation value as a parameter. Is used as a lower address, and the corresponding moment value _Sj is read. Therefore, only the moment value and the area value are stored in the ROM of the fuzzy controller with the head cutoff value as a parameter for each label and the overlapping portion of the adjacent label. When all the inference values for the combination of the input variables are stored, R
The storage capacity of the OM is reduced. Further, only a subtraction and a division process are performed in order to obtain a definite value, and a complicated operation process is not required.

B. Embodiment of the Invention According to the Second Embodiment (Digital System Using Algebraic Arithmetic Method) FIG. 6 shows storage contents of a storage area of a fuzzy controller which is an embodiment of the second embodiment of the invention. In this storage area, similarly to FIG. 1A, the membership value μ _i is used as a parameter, and the moment value S _i and the area value A of the membership function of the consequent part deformed by the membership μ _i are stored. _i and are stored. In fuzzy inference using algebraic arithmetic, the label of the membership function of the consequent part is selected by the label of the input variable of the antecedent part, and the label of the membership function of the consequent part is the minimum of the membership of the antecedent part. Deformed by the value μ _i . This deformation is the area saving deformation, for example if the membership functions of triangular shape while maintaining its base, is deformed to a triangle of the height of the appertaining mu _i. The area value A _i of the deformed shape and the moment value S _i obtained by multiplying the area value A _i by the distance x from the reference position to the position of the center of gravity of the triangular shape are represented by the membership μ _i per unit interval. Is calculated in advance offline and stored in the storage area of the ROM 2.

In the fuzzy inference of the algebraic operation method, even when a plurality of labels adjacent to each other are selected in the consequent part, the area of the membership function of each label is simply algebraically added. As in the case of using the MIN-MAX operation method, there is no overlapping portion between adjacent labels. For this reason, the storage contents of the second storage area shown in FIG. 1B are not required, and the capacity of the storage area may be half that of the MIN-MAX operation method.

FIG. 7 is a flowchart showing the processing procedure of the fuzzy controller.

The fuzzy controller obtains the antecedent part conformity based on the read input variables, and sets the minimum value as the belonging degree μ _i of the consequent part (n21 to n23). Next, FIG.
Each read the moment value S _i and the area value A _i corresponding to the appertaining mu _i adds the storage area shown in (n24~n26). This addition result S of the moment value is divided by the addition result A of the area value, and the final value D of the fuzzy inference is obtained.
Is obtained (n27).

By the above processing, for example, as shown in FIG. 8, the labels L-1, L,
When L + 1 is selected, each label is assigned a membership degree μ
_{L-1, μ L,} deformed triangle having the height mu _{L + 1.} This modified membership function range abcd-
The position of the center of gravity of ea is the inference result of fuzzy inference (determined value)
D. In this range, the area abcca
Is equal to the area abma, and the area cdem-
c is equal to the area deemd. Therefore, the area a
-Bcdcea is an area abma, an area cokc, and an area d, which are areas after deformation of each label.
−emd. Therefore, the range abc-
In order to find the center of gravity of dea, the labels L-1,
The moment value S _i and the area value A _i for each of L and L + 1 are added, and the added moment value is divided by the area value.

When the membership function in the consequent part has the same shape for each label, the value of the area value for the degree of belonging is the same for all labels, and the contents of the area value table are common to all labels. Can be. The calculation of the barycenter position W for the three selected labels S, M, and B is performed by assigning the degree of affiliation of each label to μ.
_{S, μ M, μ B,} the distance from the reference position Z _S, Z _M, Z
When _B, W = _{[A (μ S) * Z S} + A (μ M) * Z M + A (μ B) * Z B ) / _{(A (μ S) + A (} μ M) + A (μ B) ] in = _{[A (μ S) * (Z} M -Z S) + A (μ B) * (Z B -Z M) ] / _{[A (μ S) + A (} μ M) + A (μ B) ] + Z _M As shown in FIG. 9, a table is not required for the label M, and the configuration of the storage area can be further simplified.

C. Embodiment of the Invention According to Claims 3 to 6 (Analog Fuzzy Controller) FIG. 10 is a block diagram showing the configuration of a fuzzy controller according to an embodiment of the present invention. As shown in FIG. 1, the fuzzy controller includes a fuzzification module 1, an antecedent part operation module 2, and a consequent part operation module 3. The input variables XA and XB are input to the fuzzification module 1, and the fuzzification circuit 1 outputs the membership 155 of the input variables to the antecedent module 2. The antecedent part operation circuit 2 outputs the degree of conformity 256 based on the membership 155 of the input variable to the consequent part operation circuit 3. In addition, the fuzzification module 1 sends to the antecedent part operation module 2 and the consequent part operation module 3 for identifying an area defined by a combination of two adjacent labels whose affiliation degree is not 0 for each input value. Hexadecimal-coded area data 156 is supplied. The consequent part operation circuit 3 performs a definite value operation based on the area data 156 and the fitness 256 output from the antecedent part operation module 2.

FIG. 11 is a block diagram showing the configuration of the fuzzification module. The fuzzification module 1 includes a reference reference potential supply unit 101. The reference reference potential supply unit 101 outputs the degree of membership of the triangular wave vertex of the membership function of the antecedent part, identifies an input value area, and generates a grade value. Apply to The reference reference potential 152 supplied from the reference reference potential supply unit 101 is input to the multiple comparison unit 102.
The multiple comparison unit 102 receives the input variable XA as shown in FIG.
To each label of the antecedent part membership function (NL to PL)
Is compared with the reference potential 152, which is the degree to which the triangle wave vertex belongs.
Label containing input variable XA (here PS and ZR)
Is output.

This area identification output 153 is
Is encoded into the above-described area data 156. Also, from the multiple comparison unit 102 to the differential amplification unit 103,
The minimum reference reference potential that is not smaller than the input variable XA is output as the region boundary value 154. The differential amplifier 103 calculates the difference between the area boundary value 154 and the input variable XA, and outputs the degree of membership 155 of the larger label. The above-mentioned multiple comparison unit 102, differential amplification unit 103, and encoding unit 1
04 is provided for each input variable.

FIG. 13 is a diagram showing the structure of the antecedent operation module of the fuzzy controller. In the antecedent part operation module 2, based on the membership 155 (155a, 155b) input from the fuzzification module 1,
An arithmetic processing for obtaining the antecedent part suitability vector 256 is performed by applying the MIN-MAX arithmetic method or the algebraic arithmetic method. Membership degree 15 input from fuzzification module 1
5 is directly and complemented by the complementing circuit 201, and then input to the logical product combination unit 202, where variables are aggregated.

The logical product system coupling section 202 is a MIN-M
In the AX operation method, an E-MIN operation is performed, and in the algebra operation method, an algebraic product operation is performed. Here, the complementing circuit 201 calculates the degree of affiliation in the antecedent label adjacent to the antecedent label to which the affiliation 155 corresponds in FIG. Therefore, in the logical product system combination unit 202 to which the MIN-MAX operation method is applied, the smaller value of the belonging degrees of the two labels to which the input variables belong is selected. In addition, in the logical product system combining unit 202 to which the algebraic operation method is applied, the product of the affiliations of the two labels is selected as the conformity of the consequent part.

The fitness selected by the logical product system coupling unit 202 is input to the analog gate matrix 210. The analog gate matrix 210 includes an input unit constituted by the complementing circuit 201 and the AND unit 20.
2 is switched in accordance with the fuzzy rule selected in the fuzzification module 1 between the fitness calculation unit constituted by 2 and the output unit connected to the consequent unit calculation module 3. This analog gate matrix 210
Is constituted by a unit analog gate 211 shown in FIG. That is, the unit analog gate 211 is turned on by a signal input to the gate terminal from the analog gate control unit 212, and connects the S terminal and the D terminal. The analog gate control unit 212 outputs a signal to a predetermined unit analog gate 211 according to a fuzzy rule selected by a label to which the input variable belongs. Actually, the area data 1 of the input value output from the fuzzification module 1
56 is input to the address control unit 221. The address control unit 221 includes a table for referring to a pointer to an fuzzy rule in an execution format from the area data 156. The pointer of the fuzzy rule represented by the area data 156 is input to the control memory unit 222.
The control memory unit 222 stores an execution form rule derived from the primitive description form of the fuzzy rule, that is, a netlist uniquely corresponding to the net pattern, and stores a netlist that conforms to the selected fuzzy rule into an analog gate. Output to the control unit 212. Accordingly, the analog gate control unit 212 decodes a netlist that conforms to the selected fuzzy rule, generates a gate on / off bitmap, and opens and closes the unit analog gate 211. The processing of the analog gate control section 212 is the same as the configuration of the common control section of the well-known space-division communication channel type electronic exchanger. In the area data 156 input to the address control unit 221, LA0 to LA2 are area data output from the fuzzy module 1 based on the input variable XA, and LB0 to LB2 are fuzzified modules based on the input variable XB. This is the area data output from No. 1.

The antecedent part operation module 2 includes an OR gate 203 connected to the AND gate 202 in the analog gate matrix 210. The logical sum system combining unit 203 uses E- in the MIN-MAX operation method.
MAX operation is performed, and algebraic sum operation is performed in the algebraic operation method. That is, the logical sum system combining unit 203 to which the MIN-MAX arithmetic method is applied selects the larger value of the plurality of fitness values input through the analog gate matrix 210, and the logic to which the algebraic arithmetic method is applied. The sum coupling unit 203 outputs the sum of a plurality of matching degrees input via the analog gate matrix 210. That is, the input variables XA, X
Depending on the value of B, the consequent part labels may match in a plurality of selected fuzzy rules. When a plurality of fuzzy rules having the same consequent part label are selected, the larger fitness is selected in the MIN-MAX algorithm, and the sum of the fitness in the multiple fuzzy rules is determined in the algebraic algorithm. Ask for.

The above fuzzification module 1 and MI
Antecedent operation module 2 to which N-MAX operation method is applied
Will be described with reference to FIG. 15 and FIG. In the fuzzification module 1, the affiliation 155a of the antecedent label NS and the labels ZR and NS to which the input variable belongs are input to the antecedent calculation module 2 as area data for the input variable XA, and the affiliation 155b of the input variable XB and Antecedent label PS, ZR to which this input variable XB belongs
Is input to the antecedent part calculation module as area data. A fuzzy rule is selected in the address control unit 221 based on the area data 156a and 156b input from the fuzzification module 1. Here, for example, if (XA = ZR and AB = ZR) then (y = PS) -formula 1 if (XA = ZR and XB = PS) then (y = PM) -formula 2 if (XA = NS and XB = ZR) then (y = PM) -expression 3 if (XA = NS and XB = PS) then (y = PL) -fuzzy rule of expression 4 is selected.

The control memory unit 222 outputs exchange connection data representing a net list corresponding to the selected fuzzy rule to the analog gate control unit 212.
Turn on the unit analog gate shown in. Since the consequent labels represented by the U and X columns in the analog gate matrix are both PM, the four rows and the four rows are respectively input so that these two degrees of fitness are input to the logical sum system combining unit 203. The unit analog gate located on the fifth row is turned on. The larger OR or the sum of the two ORs is obtained by the logical OR system combining unit 203 and output from the MUC2.

FIG. 17 is a diagram for explaining the operation of the antecedent operation module to which the algebraic operation method is applied. In the fuzzy arithmetic by the algebraic arithmetic method, the membership function of the consequent part is limited to a triangular waveform having a high degree of fitness. That is, for the fuzzy rule expressed by the above equation 2, the logical product conjunction unit 202 sets the membership PM of the consequent part to the input variable XA in the label NS of the antecedent part and the affiliation of the input variable XB to the label ZR. The product of 155b and 155b is height, and the base is transformed into the same triangle. The same applies to the fuzzy rule represented by the above equation (3). On the other hand, in the logical product system coupling unit 203, the triangle is limited to a triangle whose height is the sum of the two fitness levels.

However, the truncation of the triangular waveform of the consequent part membership function in the MIN-MAX operation and the limitation in the algebraic operation are for explaining the theoretical operation method. , 203, only numerical values representing the degree of conformity (height of a trapezoid or a triangle) are handled in any of the arithmetic methods. Therefore, it is also possible to apply the MIN-MAX operation method to one of the combining units 202 and 203 and apply the algebraic operation method to the other.

FIG. 18 is a block diagram showing the configuration of the consequent part operation circuit 3 of the fuzzy controller. The figure shows
4 shows a configuration of a consequent part operation module to which the MIN-MAX operation method is applied. The consequent part calculation module 3 receives the fitness data 256 from the antecedent part calculation module 2. The consequent part operation module 3 calculates the conformity data 25
6 to perform an area calculation process and a volume calculation process, and output a fixed value D. The consequent part operation module 3 includes four product-sum operation parts 310,
The conformity for each consequent part label constituting 56 is input to each product-sum operation unit 310. The product-sum operation unit 3
To 10, a label unique weight value 357 predetermined for each consequent part label is input from the label-specific weight distribution analog gate matrix 331. The analog gate matrix for label-specific weight distribution 331 has the same configuration as the analog gate matrix 210 included in the antecedent part operation module 2 described above, and the weight value for each label output from the weight value reference potential generation unit 340 is analog gated. Control unit 334
Is supplied to the product-sum operation unit 310 based on the bitmap data output from the. This analog gate control unit 334
The processing for generating an on / off bitmap from a netlist according to a fuzzy rule via an address control unit 338 and a control memory unit 339 based on the area data 156a and 150b supplied from the fuzzification module 1 is also described above. Is the same as the analog gate matrix 210 of FIG.

The product-sum operation unit 310 includes a complementing unit 311, a square calculating unit 312, a complementing unit 313, and a multiplying unit 314, as shown in FIG. The complementing unit 311 obtains a complement value of the fitness 256 for calculating the area by the analog circuit. For example, a complement value (1-MU)
Ci) is obtained. The complement value (1-MUCi) obtained in the complementing unit 311 is input to the square calculating unit 312.
The square calculator 312 multiplies the square of the complement value by a constant K to obtain a triangle area dbe in the consequent part membership function abc. Here, since the shape of the consequent part membership function is a predetermined triangular waveform,
This area calculation can be easily performed by fixing the constant K.

The area of the triangular part obtained in the square calculator 312 is input to the complementer 313 and subtracted from the range abc of the consequent part membership function with 1 as the height.

In this way, the area S of the trapezoid part truncated by the degree of conformity in the consequent part membership function is obtained.
As described above, since the product-sum operation unit configured by the analog circuit obtains the area of the membership function of the consequent part that is cut by the degree of conformity for each label, the conventional circuit configuration for the integration operation is unnecessary. The circuit configuration can be made much simpler than this.

The area S of the truncated range obtained by the complementing unit 313 is input to the multiplying unit 314 together with the weight value 357. The multiplication unit 314 multiplies the area value S by a weight 357 unique to the label to obtain a volume value 361. The fitness data 256 output from the antecedent part operation module 2 is also input to the overlap correction product-sum operation part 320. The overlap-correction product-sum operation unit 320 obtains the area value and the volume value for the overlapped portion of the adjacent consequent part label, and supplies them to the correction value in the subsequent OR operation. The overlap-correction product-sum calculation unit 320 is adjacent to the product-sum calculation unit 310 provided in the above-described case in which the product-sum calculation unit 310 is provided with four in accordance with the number of consequent labels generated in the 2-input 7-label 1-output fuzzy controller. Three are provided in accordance with the number of overlapping portions of the consequent label. Each overlap correction product-sum operation unit 320 is shown in FIG.
0, the minimum value calculating unit 321 and the complementing unit 32
2, square calculator 323, complementer 324, and multiplier 3
25. The minimum value calculating unit 321 obtains the smaller value of the fitness levels of two adjacent consequent labels.

The calculated value of the smaller degree of conformity is input to the complementing unit 322, and the complement operation in which the degree of conformity of the triangular wave vertex of the overlapping part of two adjacent labels in the consequent part membership function is 1 is performed. Done. Based on the complement value obtained by the complementing unit 322, the product-sum operation unit 310
In the same manner as described above, the square calculator 323 and the complementer 3
24 and the multiplier 325 in the complement value range b'-e '
−d ′, the area ISi of the overlapped part a′-d′-e′-c ′ truncated by the smaller fitness MUCl, and the overlapped area ISi multiplied by the weight IWT specific to the overlapped part. The combined polymerization part volume value IPS is determined. At this time, the weighting reference potential generating section 340 is supplied to the multiplying section 325 via the overlap correction weight distribution analog gate matrix 332.
A unique weight value is supplied to each overlapping section output from.
In distributing the overlap correction weight value, the analog gate control unit 334 outputs an on / off bitmap. This is the same as the operation of the analog gate 331 for distributing the label specific weight described above.

The overlap correction product-sum operation unit 320 has an overlap correction strobe analog gate matrix 333.
, An overlap correction strobe signal is input. The overlap correction strobe signal is a signal for distributing 0 as the suitability of the overlapping unit when the two suitabilities input to the overlap correction product-sum operation unit 320 are those of the consequent labels that are not adjacent to each other. is there. The overlap correction strobe analog gate matrix 333 is also controlled by the analog gate control unit 334 in the same manner as the other analog gate matrices 331 and 332.

The volume value data 361 of each consequent part label and the volume value data 362 of the overlap part obtained by the product-sum operation unit 310 as described above are input to the numerator addition unit 341 and the consequent part members In the ship function, the volume value IV of the overlapping portion is subtracted from the added value of the volume value V of the label whose fitness is not 0. Further, the area value data 363 of each consequent part label obtained in the product sum operation unit 310 is input to the denominator addition unit 342 together with the area value data 364 of the overlapping unit obtained in the overlap correction product sum operation unit 320. You. In the denominator adding unit 342, the area value IS of the overlapping part is subtracted from the sum of the area values S of the consequent part labels whose conformity is not 0 in the consequent part membership function.

The numerator adder 341 and the denominator adder 342
Sum data 365 and area sum data 3 obtained by
66 is input to the division unit 343. The division unit 343 divides the volume sum calculated by the numerator addition unit 341 by the area sum calculated by the denominator addition unit, and outputs the result as a final value 367.

With the above processing, the consequent part calculation module 3 shown in FIG. 18 is truncated by the conformity of the label whose definite value is not 0 in the consequent part membership function in accordance with the MIN-MAX operation method. Range center of gravity value D
Is obtained as final value data 367. In this calculation process, the consequent part calculation module 3 first obtains an area value for each consequent part label, multiplies the area value by a weight value unique to each label to obtain a volume value, and calculates the area value and the volume value. Are added to the values of all the labels, and then the value of the overlapping portion is subtracted to obtain the area sum S and the volume sum V shown in FIG. Then, the centroid value is obtained by dividing the obtained volume sum V by the area sum S.

FIG. 22 is a block diagram showing the structure of the consequent part operation module to which the algebraic operation method is applied. In the configuration of the consequent part operation module 3 'to which the algebraic operation method is applied, the analog gate matrix 3 for distributing the label-specific weights
31, the address control unit 338, the control memory unit 339, and the division unit 343 are common to the consequent part operation module 3 to which the above-described MIN-MAX operation method is applied. The goodness-of-fit data 256 is input to the four product-sum calculation units 310 '. Further, the product-sum calculation unit 310 ′ is supplied with a weight 357 unique to the label output from the weight-value reference potential generation unit 340 ′ via the analog gate 331. The analog gate matrix 331 forms a wiring in accordance with an on / off bitmap output from the analog gate control unit 334 '. This analog gate control unit 33
The processing of 4 ′ is basically performed by the analog gate control unit 33 described above.
4, but the configuration is simplified by one analog gate matrix to be controlled.

As shown in FIG. 23, the product-sum operation unit 310 'includes a constant unit 316 and a multiplication unit 314. The constant section 316 is formed of, for example, a variable gain operational amplifier, and a constant unique to the waveform of each label of the membership function of the consequent section is set. This constant part 31
In step 6, the area value Si for each label is obtained by multiplying the fitness data 256 output from the antecedent part operation module 2 by the set constant. In the algebraic calculation method, as shown in FIG. 17, the area of each label used in the calculation of the consequent part is the area of a triangle in which a triangular waveform is limited with the degree of conformity being high, and the area value Si to be obtained is The value of the consequent part membership function is proportional to the degree of fit for the entire area of one label. Therefore, by setting a constant corresponding to the area of the entire range of one label of the consequent part membership function in the constant part 316, the area value Si according to the degree of conformity can be easily obtained. If the value set in the constant section can be changed by, for example, control data output from the analog gate control section, the same effect as when a different function waveform is given to each label can be obtained.

The area value Si obtained by the constant section 316 is output as it is and is also input to the multiplication section 314. The multiplication unit 314 includes the analog gate matrix 3
From 31, a weight value WTi specific to the label is input. In the multiplication section 314, the volume value Vi is obtained by multiplying the area value Si by the weight value WTi.

The volume value data 361 'obtained by the product-sum operation unit 310' is input to the numerator addition unit 341 '.
In the consequent part membership function, the sum of the volume values of the labels whose definite values are not 0 is obtained. The product-sum operation unit 31
The area value data 363 'output from 0 is input to the denominator adding section 342', and the area sum of the label whose definite value is not 0 is obtained in the consequent part membership function. The operation results of the numerator addition unit 341 ′ and the denominator addition unit 342 ′ are input to the division unit 343. The division unit 343 divides the volume sum by the area sum to obtain a center-of-gravity value D.
Is output as final value data 367.

As described above, the consequent part operation module 3
24, as shown in FIG. 24, the area value S of the consequent label (PS, PM, PL as an example) whose final value is not 0 is obtained, and the volume value V is obtained by multiplying each label by a unique weight value. . Next, the area value S and the volume value V of each label are added, and the added volume value V (PS) + V (PM) + V (P
L) (S (PS) + S (PM) + S
(PL)) to obtain a definite value. As described above, in the consequent part operation module 3 'to which the algebraic operation method is applied, it is not necessary to calculate the area value and the volume value for the overlapping portion of the adjacent label, and the MIN-MA
The configuration can be simplified as compared with the consequent part operation module 3 to which the X operation method is applied.

FIG. 25 is a diagram showing a schematic configuration when the fuzzy controller is used in a local area network. In the fuzzy controller having the antecedent operation module and the consequent operation module configured as described above, the fuzzy rule is stored in the control memory of the antecedent operation module and the consequent operation module by digital communication means. The fuzzy rules can be transferred from another digital controller 411 connected to the network 412 as shown in FIG. 25 and used as a server on the network.

In the example shown in FIG.
Reference numeral 11 denotes a controller which is incorporated in, for example, a processing machine and performs numerical control in real time. Control software developed at the production management workstation 431 is downloaded via the communication unit 432. The fuzzy inference server 420 connected to the network 412 includes a fuzzy inference unit 400, a control memory unit 421, and an I / O
An O switch 423 is provided. The fuzzy inference unit 400 is a fuzzy controller according to the above-described embodiment shown in FIG. This fuzzy inference unit 400 includes I /
Detector 41 such as a video camera via O-switch 423
6 and the data in the effector 417 such as a servo mechanism for driving the robot arm are input as input variables.

The control memory unit 421 is provided with the fuzzy inference unit 40
Digital data such as a fuzzy rule is transmitted to 0.

When numerical control is performed by one digital controller 411 in the above configuration, the data of the detector 416 and the effector 417 corresponding to the processing machine in which the digital controller 411 is incorporated are selectively used by the fuzzy inference unit 400. And supplies a fuzzy rule suitable for the digital controller 411 previously stored in the control memory unit 421 to the fuzzy inference unit 400. The determined value obtained in the fuzzy inference unit is transmitted to the digital controller 41 via the local network 412.
1, and output as a control output to the processing machine, for example, an analog signal for motor control.

[0067]

According to the first aspect of the present invention, the MI
When performing fuzzy inference by the N-MAX operation method, the moment value and the area value of the membership function of the consequent part truncated by the degree of affiliation are used as parameters of the degree of affiliation, and each label of the consequent part and adjacent labels This is advantageous in that the storage capacity of the ROM can be considerably reduced as compared with the case where the fuzzy inference fixed value is stored for each input variable. Further, only a subtraction process and a division process are performed to obtain a definite value, and a complicated operation process is not required, and fuzzy inference with excellent responsiveness can be executed.

According to the second aspect of the present invention, the moment value and the area value of the membership function of the consequent part transformed by the degree of belonging are stored for each label of the consequent part using the degree of belonging as a parameter. This is advantageous in that the storage capacity of the ROM can be considerably reduced as compared with the case where a definite value of fuzzy inference is stored for each input variable. Further, since only a division process is performed to obtain a definite value, a complicated operation process is not required, and fuzzy inference by an algebraic operation method excellent in responsiveness can be executed. In addition, there is an advantage that a high-precision inference result can be obtained using a small-capacity fuzzy ROM without performing an interpolation operation because the interval of the membership degrees can be made sufficiently small.

According to the third aspect of the present invention, since the area value is calculated using the degree of conformity as a parameter for each consequent part label, an integral operation is not required, and the obtained area value is added to the consequent part label. The volume value is obtained by multiplying by the unique weight value, and this volume value is divided by the area value to obtain a definite value, so that each of the plurality of division points on the horizontal axis of the consequent part membership function is determined. There is no need for a circuit for setting and holding the value, and there is an advantage that the circuit configuration in the operations of the truncation unit and the defuzzification unit can be extremely simplified.

According to the selected fuzzy rule,
Input section, fitness calculation section and output section
Can be switched dynamically by the
The configuration of the road is simplified.

Further, in the area operation unit and the volume operation unit of the consequent part operation circuit, the value of the overlapping part of the adjacent consequent part label is subtracted to execute the fuzzy inference to which the MIN-MAX operation method is applied. Can be.

[Brief description of the drawings]

FIG. 1 is a diagram showing stored contents of first and second storage areas constituting a ROM of a fuzzy controller according to an embodiment of the present invention;

FIG. 2 is a diagram illustrating the contents stored in a ROM of the fuzzy controller.

FIG. 3 is a flowchart showing a processing procedure of the fuzzy controller.

FIG. 4 is a diagram showing calculation of a final value in the fuzzy controller.

FIG. 5 is a diagram showing a configuration of a general fuzzy controller including an embodiment of the present invention.

FIG. 6 shows storage contents of a storage area of a fuzzy controller according to an embodiment of the present invention.

FIG. 7 is a flowchart showing a processing procedure of the fuzzy controller.

FIG. 8 is a diagram showing calculation of a final value in the fuzzy controller.

FIG. 9 is a diagram showing storage contents of a storage area of a fuzzy controller according to another embodiment of the present invention.

FIG. 10 is a diagram showing a configuration of a fuzzy controller according to an embodiment of the invention described in claims 3 to 6;

FIG. 11 is a diagram showing a configuration of a fuzzification module of the fuzzy controller.

FIG. 12 is a diagram for explaining an operation in the fuzzification module.

FIG. 13 is a diagram showing a configuration of an antecedent operation module of the fuzzy controller.

FIG. 14 is a diagram showing a unit analog gate forming an analog gate matrix forming a part of the antecedent part operation module.

FIG. 15 is a diagram illustrating an operation in the MIX-MAX operation method.

FIG. 16 is a diagram illustrating an operation in the MIN-MAX operation method.

FIG. 17 is a diagram illustrating the operation of the antecedent part operation module when the algebraic operation method is applied.

FIG. 18 is a diagram showing a configuration of a consequent part operation module using the MIN-MAX operation method of the fuzzy controller.

FIG. 19 is a diagram illustrating the configuration and operation of a product-sum operation unit forming a part of the consequent-part operation module.

FIG. 20 is a diagram illustrating the configuration and operation of an overlap-product-sum operation unit that forms part of the consequent-part operation module.

FIG. 21 is a diagram for explaining the operation of the defuzzification operation unit in the consequent part operation module.

FIG. 22 is a diagram showing a configuration of a consequent part operation module to which an algebraic operation method is applied in the fuzzy controller.

FIG. 23 is a diagram illustrating a configuration of a product-sum operation unit provided in a consequent part operation module to which the algebraic operation method is applied.

FIG. 24 is a diagram illustrating the operation of the defuzzification process in the consequent part operation module to which the algebraic operation method is applied.

FIG. 25 is a diagram showing a configuration of a local area network to which the fuzzy controller is applied.

[Explanation of symbols]

1-Fuzzy circuit 2-Antecedent operation circuit 3-Consequent operation circuit 11-Microcomputer 12-ROM 210-Analog gate matrix (Antecedent exchange means) 212-Analog gate control section (Antecedent exchange means) 340-weight value reference potential generation section (weight value output section) 310-product sum calculation section (area calculation section, volume calculation section) 331-analog gate matrix for label-specific weight distribution (consequent part exchange means) 334-analog Gate control unit (consequent part exchange means) 343-division unit

Claims (5)

(57) [Claims] 1. A moment value and an area value of a membership function of a consequent part truncated based on a degree of affiliation obtained from an input variable. The first and second storage areas stored for each overlapped part, and the moment value and area value corresponding to the degree of affiliation for the label of the consequent part of the fuzzy rule including the input variable in the antecedent part and the adjacent label are stored in the Reading means for reading from the first storage area and the second storage area; subtraction means for subtracting the value of the second storage area from the value of the first storage area for each of the moment value and the area value corresponding to the degree of belonging; A fuzzy controller provided with division means for dividing the moment value obtained by the subtraction means by the area value to obtain a definite value. 2. A storage area for storing a moment value and an area value of a membership function of a consequent part deformed by the degree of affiliation obtained from an input variable for each label of the consequent part using the degree of affiliation as a parameter. Reading means for reading a moment value and an area value corresponding to the degree of affiliation from a storage area in a consequent part of a fuzzy rule including an input variable in the antecedent part; and dividing the moment value obtained by the reading means by the area value. A fuzzy controller, comprising: division means for obtaining a definite value. 3. A fuzzy circuit for generating an antecedent label to which an input variable belongs in an antecedent part membership function and a degree of belonging of the input variable in the antecedent label, and a fuzzy rule including the consequent part label. An antecedent part operation circuit for selecting and setting the degree of conformity with respect to the consequent part label based on the affiliation degree; a weight value output part for outputting a weight value unique to the consequent part label; A triangle with a known total area set for each label
An area calculating unit that calculates an area value using the consequent part membership function and the degree of conformity as parameters, a volume value that is a product of the area value obtained by the area calculating unit and the weight value output from the weight value output unit. A volume calculation unit for calculating the sum, and a consequent part calculation circuit having a division unit for obtaining a definite value by dividing the sum of the volume values obtained by the volume value calculation unit by the sum of the area values, The area calculation unit calculates the area value according to the following procedure.
A fuzzy controller characterized by: (1) Calculate the complement value of the above-mentioned goodness of fit, and square this value to a constant
Consequence consequently truncated at the fitness by multiplying by K
Find the area of the triangle above the part membership function. (2) Determine the area of the upper triangle by the membership of the consequent part.
Is subtracted from the total area of the
The area of the trapezoid on the lower side of the function is obtained as the area value. Dynamic wherein said antecedent arithmetic circuit, the number of fitness calculation unit corresponding to the number of fuzzy rules which may be selected simultaneously, the input unit according to the selected fuzzy rules, a fitness calculation unit and an output Analog gater that switches dynamically
4. The fuzzy controller according to claim 3, comprising: a trix. 5. The consequent part processing circuit according to claim 1, wherein
Stack weight that outputs a unique weight value for the bell stack
An output value output unit, the area calculation unit and the volume calculation unit.
Weight to subtract the value of the overlapping part of the adjacent consequent label
5. The filer according to claim 3, further comprising:
J controller.