JP2002116912A - Fuzzy inference arithmetic processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a fuzzy inference arithmetic processing method for easily executing the temporary invalidation of a certain rule or the change of the validity of the rule. SOLUTION: In this fuzzy inference arithmetic processing method, weighting Wi is appropriately set beforehand corresponding to the respective rules Ri, the pertinent weighting Wi is performed to an antecedent part adaptation rate αi calculated by MIN arithmetic operation in the respective rules, a truncated consequent part membership function is obtained by using a corrected antecedent part adaptation rate βi instead of the conventionally used antecedent part adaptation rate αi, MAX arithmetic operation and centroid calculation are performed to the respective truncated consequent part membership functions and an inferred result is obtained. Thus, the effect similar to the one of changing of the rules by weighting is obtained without changing the respective rules. Also, a threshold is set corresponding to the respective rules and the invalidation of the rule when the corrected antecedent part adaptation rate is equal to or lower than the threshold is easily executed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fuzzy inference operation processing method, and more particularly, to a method of processing an antecedent in each rule.
The antecedent part conformance ratio is calculated from the IN operation, the consequent part membership function obtained by truncating by the antecedent part conformity ratio is obtained, and the consequent part membership function obtained by each rule is truncated. The present invention relates to a fuzzy inference operation processing method of performing a MAX operation, calculating a center of gravity of the result, and obtaining an inference result.

[0002]

2. Description of the Related Art Conventionally, in this type of fuzzy inference operation processing method, it is common to describe control knowledge cultivated over many years of experience as a set of fuzzy control rules and store it in a fuzzy controller. It is. When the antecedent part of the fuzzy inference is composed of a plurality of items, calculating the overall antecedent part conformity is equivalent to calculating the firing degree of the control rule (rule). It is an important factor that directly influences. Conventionally mainly used methods of calculating the antecedent part fitness include a MIN operation, an algebraic product operation, an arithmetic average operation, and a geometric average operation. There is a method of calculating the area centroid by performing a MAX operation after truncating the consequent part membership function using the fitness obtained from the above operation (Algebraic product-addition-centroid method, simplified There is a calculation method such as a method).

[0003] A set of fuzzy control rules is generally
The following formula (1)

(Equation 1) It is described in the form. in this case,

(Equation 2) Is a fuzzy set of the m-th item in the antecedent part of the i-th rule (rule),

(Equation 3) Is a fuzzy set of the nth item of the consequent part of the i-th rule (rule).

For example, when M = 2, N = 1, and I = 2, the rule corresponding to the equation (1) is expressed by the following equation (2).
a), (2b), ie, the rule _{R1: IF x 1 is NB &} x 2 is ZR THEN y 1 is PB ··· (2a) rule _{R2: IF x 1 is ZR &} x 2 is PB THEN y 1 is It is assumed that ZR (2b) can be described (however, NB is a negative and large, ZR is zero (in a fuzzy sense), and PB is a positive and large fuzzy set).

In this case, at time t, input observation information is a1 (t) and a2 (t), and MIN-M
According to the AX-center-of-gravity method, regarding the rule R1, the antecedent (T
From the MIN operation of the first item (before HEN) (FIG. 4 (A)) and the second item (FIG. 4 (A)), the truncated fuzzy set of FIG. R2
For the consequent part, the head-cut fuzzy set of FIG. 4 (K) is obtained from the MIN operation of the first item (FIG. 4 (F)) and the second item (FIG. 4 (G)) of the consequent part. The fuzzy set of FIG. 4 (S) is obtained by the MAX operation on these results, and the control value C1 (t) of FIG. 4 (S) as the inference result is obtained by the calculation of the center of gravity.

[0006]

As described above, in the conventional fuzzy inference operation processing method, the consequent part membership function is calculated using the antecedent part conformity, but temporarily. It is not easy to disable a rule or change the validity of a rule.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problem, and there is provided a fuzzy inference calculation processing method which can easily execute a rule temporarily invalidating or changing the validity of the rule. The purpose is to provide.

[0008]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention provides a MIN of an antecedent part in each rule.
The antecedent part conformance ratio is calculated from the calculation, the consequent part membership function obtained by truncating the antecedent part by the antecedent part conformity ratio is obtained, and the consequent part membership function obtained by each rule is MAX. In the fuzzy inference operation processing method of performing an operation and calculating an inference result by performing a centroid calculation on the result, when obtaining the membership function of the consequent part after the truncation, the antecedent calculated in accordance with each rule. To the part conformity ratio, a weight is set in advance corresponding to each rule to calculate the modified antecedent part conformance ratio, and the modified antecedent part conformity ratio is used instead of the antecedent part conformity ratio. Used.

According to such a configuration, in order to easily change the validity of a rule which has not been easy until now,
Weighting is appropriately set in advance in accordance with each rule, and a corresponding weight is applied to the antecedent part conformance ratio calculated by the MIN operation in each rule, and the antecedent part conformity ratio conventionally used is calculated. Instead, the consequent part membership function is obtained by truncating using the modified antecedent part conformity ratio.

Further, in the present invention, when any of the corrected antecedent part conformance ratios calculated corresponding to each rule is equal to or less than a threshold value set in advance corresponding to each rule, The antecedent part conformity ratio is set to zero, and the corresponding rule is invalidated.

[0011]

Embodiments of the present invention will be described below with reference to the accompanying drawings. FIG. 1A is a diagram showing a membership function of a fuzzy set of an antecedent part selected by a MIN operation in the embodiment of the fuzzy inference operation processing method of the present invention, and FIG.
FIG. 2 shows a membership function of a consequent part corresponding to (a). FIG. 2 shows the contents of a memory storing weights for changing the validity of each rule used in the embodiment of FIG. FIG. 3 and FIG. 3 are diagrams showing the contents of a memory storing a threshold value for determining whether or not each rule used in the embodiment of FIG. 1 is valid. In FIG. 1A, the display of membership functions for items that are not directly needed for explanation is omitted, and only the membership functions of the fuzzy set of the antecedent part selected by the MIN operation are displayed.

In FIG. 1A, the horizontal axis represents the input range, and the vertical axis represents the ratio of the degree of matching belonging to a certain set.
Therefore, with respect to the i-th rule Ri, as shown in FIG. 1B, the consequent consequent is truncated as shown in FIG. A department membership function is determined. In this case, weighting Wi (0 ≦ Wi ≦ 1.0)
Can be considered to be 1.0, and the fitness ratio βi used in FIG. 1A is βi = αi × 1 = αi, and FIG. 1 shows αi itself. Therefore, if the weighting Wi is 0.5, the fitness ratio βi is set to half of the height of the fitness ratio αi shown in FIG.
In accordance therewith, the weighted and truncated consequent part membership function (the dashed line in FIG. 1B) is obtained.
A MAX operation is performed on the weighted and truncated membership function of such a plurality of consequent parts for each rule, and the inference result is obtained by calculating the center of gravity of the result.

As shown in FIG. 2, the weight Wi applied to each rule Ri (i = 1 to M) is set corresponding to each rule Ri and stored in a memory.
Then, as described above, the fitness ratio αi in the membership function of the antecedent part corresponding to a certain rule Ri
Is selected, the control unit of the fuzzy controller determines the rule R
The weighting Wi corresponding to i is read from the memory, βi (= Wi × αi) corrected by the weighting Wi is calculated, and the consequent part membership function truncated is calculated using the calculated βi. As described above, the MAX operation is performed on the weighted and truncated membership functions of the plurality of consequent parts of each rule, and the inference result is obtained by calculating the center of gravity of the result. As described above, even if each rule is not changed, the same effect as the rule is substantially changed can be obtained by the easy means of weighting. Therefore, for example, in the case of abnormal or emergency control, when it is desired to perform control with a higher effectiveness than the rule of normal control, the control can be executed only by changing the weight, so that the rule is changed as in the conventional case. This makes the operation extremely easy.

Incidentally, the fitness ratio βi as described above
When the consequent membership function corresponding to the rule Ri is obtained based on
It is also possible to invalidate (or temporarily invalidate) the rule Ri when it is smaller. That is, rule R
i, the threshold value γi is set as shown in FIG.
It is set and stored in a memory in advance as a table. In this case, as described above, the control unit of the fuzzy controller calculates the βi equal to or less than the threshold γi before obtaining the consequent part membership function corresponding to the rule Ri, as described above. If the calculated βi is not less than or equal to the threshold value γi, the membership function corresponding to the rule Ri is determined based on the fitness ratio βi, as described above. Ask. However, when the calculated conformity ratio βi is equal to or smaller than the threshold γi, the antecedent part conformity ratio βi is set to zero, and the rule Ri is invalidated.

[0015]

The fuzzy inference operation processing method of the present invention is configured as described above, and in order to change the effectiveness of a rule which has not been easy so far, weighting is performed in advance for each rule. Appropriately set, the corresponding antecedent part ratio calculated by the MIN calculation in each rule is weighted, and the modified antecedent part conformance ratio is used instead of the antecedent part conformance ratio conventionally used. Calculate the consequent membership function by truncating using, perform MAX operation on the consequent membership function obtained in each rule and calculate the inference result by calculating the center of gravity of the result Thus, even if each rule is not changed, an effect similar to that obtained by substantially changing the rule by weighting is obtained. Also, a threshold value is set in advance for each rule, and if the corrected antecedent part conformance ratio (or antecedent part conformance ratio) of each rule is equal to or less than the threshold value, the rule is invalidated. It can be done easily.

[Brief description of the drawings]

FIG. 1A is a diagram showing a membership function of a fuzzy set of an antecedent part selected by a MIN operation in the embodiment of the fuzzy inference operation processing method of the present invention. (B) It is a figure which shows the membership function of the consequent part corresponding to said (a).

FIG. 2 is a diagram showing contents of a memory which stores weights for changing the validity of each rule used in the embodiment of FIG. 1;

FIG. 3 is a diagram showing contents of a memory which stores a threshold value for determining whether each rule used in the embodiment of FIG. 1 is made valid or not.

FIGS. 4A and 4B are diagrams showing membership functions of an antecedent part with respect to rule R1. (E) is a diagram showing a membership function of the consequent part of rule R1.
(F) and (g) are diagrams showing membership functions of the antecedent part regarding rule R2. () Is a diagram showing a membership function of the consequent part of rule R2. (SO)
FIG. 9 is a diagram showing a membership function and an inference result obtained by performing a MAX operation on the truncation membership functions of (E) and (K).

[Explanation of symbols]

αi Antecedent part conformance ratio for i-th rule Ri βi Corrected antecedent part conformance ratio obtained by applying weighting Wi to αi Wi Weighting for i-th rule Ri γi Threshold value for i-th rule Ri

Claims (2)

[Claims] 1. An antecedent part conformance ratio is calculated from a MIN operation of an antecedent part in each rule, and a consequent part membership function truncated based on the antecedent part conformity ratio is determined. In the fuzzy inference operation processing method of performing a MAX operation on the truncated consequent part membership function, calculating the center of gravity of the result, and obtaining an inference result, the fuzzy consequent part membership function is obtained. The modified antecedent part conformance ratio is calculated by giving a predetermined weight corresponding to each rule to the antecedent part conformance ratio calculated corresponding to each rule, and the antecedent part conformance ratio is calculated. A fuzzy inference calculation processing method characterized by using a modified antecedent part conformity ratio instead of a degree ratio. 2. If any one of the modified antecedent part conformance ratios calculated corresponding to each rule is equal to or less than a threshold set in advance corresponding to each rule, the modified antecedent part conformity is determined. 2. The fuzzy inference operation processing method according to claim 1, wherein the degree ratio is set to zero and the corresponding rule is invalidated.