JP2767625B2 - Fuzzy inference apparatus and operation method thereof - Google Patents

Description

Description: BACKGROUND OF THE INVENTION The present invention relates to a fuzzy inference apparatus and an operation method thereof.

In a conventional fuzzy inference apparatus, the maximum value of the number of types of input variables in the antecedent part of applicable fuzzy inference rules is determined by its architecture. Therefore, there is a problem that inference cannot be performed according to a rule having many types of input variables.

SUMMARY OF THE INVENTION An object of the present invention is to provide a fuzzy inference apparatus capable of executing fuzzy inference even when the number of input variables in the antecedent part of a rule is large, and an operation method thereof.

A fuzzy inference apparatus according to the present invention is created by decomposing a plurality of individual inference means capable of executing a fuzzy inference of a rule having a predetermined number of input variables or less, and a rule in which the number of input variables exceeds the predetermined number. Executes fuzzy inference according to the new rules obtained by combining the inference results of the original rules from the inference results of a plurality of individual inference means, and executes fuzzy inference according to the new rules created by decomposition of the original rules And a total operation means for integrating the inference result of the individual inference means other than the individual inference means and the operation result of the combination operation means.

The operation method of the fuzzy inference apparatus according to the present invention checks whether or not the number of input variables in the antecedent part of each estimated rule exceeds the predetermined number, and if the number of input variables exceeds the predetermined number. By decomposing the existing rules, a new plurality of rules having the same consequent part as the original rule with the number of input variables within the above-mentioned predetermined number are created, and according to the created new rules, It is characterized in that the individual inference means, the joint operation means and the total operation means are controlled to operate.

According to the present invention, when a rule having many types of input variables in the antecedent is set, the rule is decomposed into a plurality of rules having a number of input variables that can be handled by the fuzzy inference apparatus, and each decomposed rule is decomposed. The fuzzy inference operation is performed and the results are combined, so that even complicated rules can be processed.

FIG. 1 shows an embodiment of a fuzzy inference apparatus according to the present invention.

The fuzzy inference device has N individual inference units R _{1 to} R _N
Contains. These individual inference units R _i (i = 1
2 to 2) are shown in FIG. Individual inference unit R _i are in principle performs the inference operation for one fuzzy inference rules. After these individual inference unit R _i As will become apparent are organized to perform the inference operation of one rule using the plurality.

First structure and operation of the individual inference unit R _i, each previously described in embodiments for performing inference operations in one rule. For simplicity, there are 3 types of input variables in the antecedent
This case will be described. The rules for fuzzy inference are expressed in If, then form as follows.

If A _i1 = L _i1 , A _i2 = L _i2 , A _i3 = L _i3 then y = M _i (1) where A _{i1 to} A _i3 are input variables, y is an output variable, and L _{i1 to} L _i3
Is a fuzzy set or a membership function of the consequent fuzzy sets or membership functions of the matter part before that correspond to each input variable, the M _i corresponding to the output variable.

Referring to FIGS. 2 and 3, inputs A _{i1 to} A _i3 (collectively A _i1 ) are applied to membership function circuits (hereinafter referred to as MFC) 41 to 43, respectively. MFC41,42,
43, membership functions L _i1 , L _i2 , and L _{i3 of the} antecedent part are set, respectively, and MFCs 41, 42, and 43 are membership functions L _i1 , L _i1 , and A _i3 corresponding to the inputs A _i1 , A _i2 , and A _i3 , respectively. Signals representing function values (grades) a _i1 , a _i2 , a _i3 of L _i2 , L _i3 are output, respectively. The signals representing these grades a _{i1 to} a _i3 are
It is input to the MIN circuit 44, and the smallest one (here, a _i3 ) is selected by the MIN circuit 44 and given to the truncation circuit 46.

On the other hand, the membership function generator circuit (hereinafter referred MFG) 45 are set membership function M _i of the consequent part. Membership function M _i in this embodiment is represented as a voltage distribution on the line of the large number (e.g. 32), MFG45 is to generate such a voltage distribution (JP 63-123177 see JP ). In the drawing, a set of transmission lines of each voltage constituting a voltage distribution representing a membership function is shown by a hatched bus expression. Voltage distribution representing the membership function M _i is applied to the truncation circuit 46, the output a of the MIN circuit 44
MIN operation was performed with _i3 . The output B _i of the truncation circuit 46 (the trapezoidal membership function indicated by hatching in FIG. 3) is the output of the individual inference unit R _i .

As described above, the individual inference unit R _i performs an inference operation on one inference rule, and outputs a result B _i . Generally, in the application of fuzzy inference including fuzzy control, a plurality of rules are set, so that fuzzy inference operations on these rules are performed by individual inference units R ₁ to R ₁ .
_RN respectively. Returning to FIG. 1, if a plurality of rules are mutually independent, the individual inference unit R
Through output B ₁ .about.B _N intact gate circuit 11 of the ₁ to R _N (the output of the gate circuit 11 represented by F ₁ to F _N), provided to cons a bonding MAX circuit (hereinafter referred to as CMAX circuit) 12 . The CMAX circuit 12 performs a MAX operation on the input signals of a plurality of line groups for each corresponding line (see the above publication). The output E of the CMAX circuit 12 represents the inference result integrated for a plurality of rules (see FIG. 3). Comprehensive inference result E, if necessary, in defuzzifier 13
For example, it is defuzzified by the center of gravity calculation and output (output y _w ).

Individual inference unit R _i in which the number of types of input variables of the antecedent part as described above to perform the inference operation of the three following rules. In the fuzzy inference apparatus shown in FIG.
A plurality of individual inference units are organized for one rule so that an inference operation of a rule having four or more types of input variables in the antecedent part is also possible.

Now, consider a rule having nine types of input variables in the antecedent part as follows.

_{If A 11 = L 11, A} 12 = L 12, A 13 = L 13, A 21 = L 21, A 22 = L 22, A 23 = L 23, A 31 = L 31, A 32 = L 32, A ₃₃ = L ₃₃ then y = B (2) Expression (2) can be expanded as in Expressions (3) to (5) using expansion rules of fuzzy inference rules.

If A ₁₁ = L ₁₁ , A ₁₂ = L ₁₂ , A ₁₃ = L ₁₃ then y = B ... (3) If A ₂₁ = L ₂₁ , A ₂₂ = L ₂₂ , A ₂₃ = L ₂₃ then y = B ... ( 4) If A ₃₁ = L ₃₁ , A ₃₂ = L ₃₂ , A ₃₃ = L ₃₃ then y = B (5) As is clear from the expansion rules, the rules of the formulas (3) to (5) The inference result is synthesized by the MIN operation.

In FIG. ₁ , if it is assumed that the inference according to the rule expressed by the equation (2) is executed using _three individual inference units R ₁ , R ₂ , R ₃ , the units R ₁ , R ₂ , R ₃ , The membership functions of the MFC and MFG are set so as to execute inference according to the rules expressed by the equations (3), (4), and (5), respectively. Individual inference unit R _1-
Output B ₁ .about.B ₃ of R ₃ are given to the correspondent loading MIN circuit (hereinafter referred to as CMIN circuit) 31 via a gate circuit 21,
These MIN operations are performed. The MIN operation result D ₁ passes through the gate circuit 11 (the output of the gate circuit 11 is represented by G ₁ ) C
Input to MAX circuit 12. The CMIN circuit performs a MIN operation on input signals of a plurality of line groups for each corresponding line (see the above publication).

The gate circuit 11 is not connected to the outputs B _{1 to} B _{3 of the} individual inference units R _{1 to} R ₃ , and the corresponding outputs F _{1 to} F ₃ of the gate circuit 11 are all zero (the minimum value is Output). Outputs B _{4 to} B _N excluding outputs B _{1 to} B ₃
And D ₁ will be provided to CMAX circuit 12 through gate 11. The gate circuit 21 (a C ₁ -C ₃ a corresponding output of the gate circuit 21) input B ₁ .about.B ₃ allow passage of only a non-connected state to the other input B ₄ .about.B _N,
And all the outputs C _{4 to} C _N corresponding to the power supply voltage V
Output as _CC (meaning maximum value).

In addition to gate circuit 21 and CMIN circuit 31, gate circuit and CMIN
One or more circuit combinations are provided as indicated by reference numerals 22 and 32. If there are further rules with four or more input variables in the antecedent part, the inference operation according to the rules is executed by the combination of a plurality of individual similarity units, gate circuits and CMIN circuits. Will be.

When the number of types of input variables in the antecedent part of the rule is 4 to 6, two individual inference units are used. When the number of types of input variables is 7 to 9, three individual inference units are used. The number of individual inference units corresponding to the number of types is organized as described above. Then, the gates 11, 21, and 22 are controlled according to the organization of the individual inference units.

The computer system 10 controls the operation of the fuzzy inference apparatus so as to perform arithmetic processing on the set rules. When there are rules including four or more types of input variables, the organization of the individual inference units,
It sets membership functions according to the rules for MFC and MFG, and controls the gates 11, 21, and 22.

FIG. 4 shows a configuration example of the gate circuit 21. Output B _i of the individual inference unit R _i is represented as a voltage distribution on 32 signal lines. Changeover switches 31~3N is provided for the output B ₁ ~B _N. Each selector switch group includes a 32 changeover switch is given from the computer system 10 via bus, is controlled by the data stored temporarily d _{i (i} = 1~N) in the register 30. For example, when the data d _i is 1, the changeover switch of the changeover switch group 3i is connected to the left side so as to output the input B _i01 .about.B _i32 as it is as the output C _i01 -C _i32, if the data d _i is 0 Outputs the power supply voltage V _CC and outputs C _i01 to C _i01.
Connected to the right side to output as _i32 .

Although an analog type circuit is used as the individual inference unit in the above embodiment, it goes without saying that a digital type circuit can be used.

Next, control processing by the computer system 10 will be described.

When four or more input variables are included in the rule set as shown in the equation (2), it is necessary to organize individual inference units. Therefore, the computer system 10 performs processing for expanding the given rule as shown in the equations (3) to (5). FIG. 6 shows the given (set) rules, FIG. 7 shows the state of rule expansion, FIG. 8 shows the finally obtained new rules, and FIG. 9 shows the rule expansion processing procedure. ing.

The set rules are stored in a magnetic disk, semiconductor memory, or other storage device in the form of data as shown in FIG. In FIG. 6, the rule data of the set rule number (I) represents the following If, then rules.

_{If x 11 = L 11, x} 12 = L 12, x 13 = L 13, x 14 = L 14, x 15 = L 15, x 16 = L 16, x 17 = L 17, x 18 = L 18 then y ₁ = K ₁ (6) In order to avoid confusion, the expression (6) uses a different symbol from the expression (2). x ₁₁ ~x ₁₈ code representing the input variables, L ₁₁ ~L ₁₈ code representing the membership functions of the antecedent, code y ₁ is representative of the output variables, K ₁ represents the membership functions of the consequent part Code.

Rules set as shown in FIG. 6 include those with a large number of types of input variables and those with a small number of types. On the other hand, as described above, the types of input variables in the antecedent part of the rule must be 3 or less due to the architectural restrictions of the fuzzy inference apparatus. Therefore, it is necessary to develop rules for rules having four or more types of input variables.

The rule expansion process is performed in the computer system 10 as follows. Referring to FIG. 9, one set rule stored in the storage device is fetched (step 51), and the number of input variables in the antecedent part of the rule is counted. If the number of types of input variables is four or more (NO in step 52), the rule antecedent is divided into groups of three or less input variables (step 53). The same is used for the consequent. For example, the following three new rules are created for the rule of the setting rule number (I) represented by Expression (6).

_{If x 11 = L 11, x} 12 = L 12, x 13 = L 13 then y 1 = K 1 ... (7) If x 14 = L 14, x 15 = L 15, x 16 = L 16 then y 1 = K ₁ ... (8) If x ₁₇ = L ₁₇ , x ₁₈ = L ₁₈ then _{1 1} = K ₁ (9) As expressed by the formula (9), the input variable of the antecedent part is 3 in the new rule. There may be less. FIG. 7 shows the state of such development and the developed rule data.

The number of input variables in the antecedent part of the set rule is 3
Rules are not expanded in the following cases (step
YES at 52).

Finally, a control code 1 is assigned to the new rule developed by the processing of step 53, and a control code 0 is assigned to the new rule that has not been developed (step 54).

The above processing is performed for all setting rules in the storage device (step 55), and new rule data as shown in FIG. 8 is created.

Based on these new rule data is as described above is the MFC, the membership functions of MFG is set in the individual inference unit R ₁ to R _N. Further, the gate circuits 11, 21, and 22 are controlled based on the control code.
That is, the gate circuit 11 allows the passage of the individual inference result based on the rule of the control code 0. The gate circuits 21 and 22 are controlled so as to permit the passage of the individual inference result for each new rule group whose control code is 1.

In the above example, different symbols from x ₁₁ to x _m3 are used for all input variables, but many of these input variables are common. For example, x ₁₁ and x ₂₁ and x _j1 and x _m1
Often it is representing the same input variables x _1.

Therefore, as shown in Fig. 5, each MF of the individual inference unit
Multiplexers 47, 48, 49 are provided on the input side of C41, 42, 43,
Based on the rule data shown in FIG. 8, this may be controlled so that only the corresponding input variables among all the input variables x _{1 to} x _l are input. Computer system 10
Indicates a code representing an input variable to be selected based on rule data in a register attached to each of the multiplexers 47 to 49.
Set to 47A to 49A. Each multiplexer selects the input of the set input variable and supplies it to the corresponding MFC.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an entire configuration of a fuzzy inference apparatus, FIG. 2 is a block diagram showing an example of a configuration of an individual inference unit, FIG. 3 is an explanatory diagram of a fuzzy inference process, and FIG. , FIG. 5 is a block diagram showing an input selection circuit provided in the individual inference unit, FIG. 6 is a diagram showing set rule data, and FIG. 7 is a diagram showing how rules are developed. FIG. 8 is a diagram showing the new rule data, and FIG. 9 is a flow chart showing the processing procedure of rule expansion. R _{1 to} R _N ... Individual inference units, 10.
System, 11, 21, 22 …… Gate circuit, 12 …… CMAX circuit, 31,
32 …… CMIN circuit, 47, 48, 49 …… Multiplexer.

Claims (3)

(57) [Claims] A plurality of individual inference means capable of executing fuzzy inference of a rule having a predetermined number of input variables or less, and a new inference means created by decomposing a rule in which the number of input variables exceeds the predetermined number. Combining operation means for deriving the inference result of the original rule from the inference results of a plurality of individual inference means executing fuzzy inference according to rules, and individual inference performing fuzzy inference according to new rules created by decomposition of the original rules A fuzzy inference apparatus comprising: a total operation means for integrating an inference result of the individual inference means other than the means and an operation result of the joint operation means. 2. A means for determining whether or not the number of input variables in the antecedent part of each set rule exceeds the predetermined number, and for a rule in which the number of input variables exceeds the predetermined number. Means for generating a plurality of new rules having the number of input variables within the predetermined number and having the same consequent part as the original rule by decomposing the input variables. Fuzzy inference device. 3. For each set rule, it is checked whether the number of input variables in the antecedent part exceeds the predetermined number, and if the number of input variables exceeds the predetermined number, it is checked. By decomposing, a plurality of new rules are created with the number of input variables within the above-mentioned predetermined number and having the same consequent part as the original rule. The operation method of the fuzzy inference device according to claim 1, wherein the operation means and the general operation means are controlled to operate.