JPH0410133A - Fuzzy inference system - Google Patents

Abstract

PURPOSE:To increase the inference speed by using a maximum value and a minimum value of all adaptabilities to rules and detecting unnecessary parts included in plural rules to eliminate an unnecessary inference time. CONSTITUTION:An inference control part 1 outputs all adaptabilities of data x1 - xn to plural preliminarily set rules. An adaptability buffer part 2 receives this output to store the adaptability of each rule. a maximum/minimum buffer part 3 receives the output from the adaptability buffer part 2 to tentatively store the maximum value and the minimum value of adaptabilities in the range of the input variance. These maximum value and the minimum value of adaptabilities are inputted to an unnecessary rule discriminating block 5, and the maximum value and the minimum value are referred to discriminate unnecessary rules. Thus, unnecessary rules are deleted to increase the inference speed even if many unnecessary rules exist.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a fuzzy inference system that infers a result by quantifying the ambiguity of a condition.

(Prior Art) Conventional fuzzy inference systems include the following. That is, the system is configured to input data to an inference control unit that performs fuzzy inference, process the information based on a plurality of preset rules and membership functions, and output the results of fuzzy inference. are doing.

In this case, the rules and membership functions are created based on the experience and intuition accumulated by experts in the specialized technical field, that is, qualitative human know-how.

(Problem to be solved by the invention) However, in such conventional fuzzy inference systems, multiple rules and membership functions are created based on the qualitative know-how of experts; Depending on the situation, there may be unnecessary Lulu. Therefore, when actually using a fuzzy inference system for control, etc., even if some of the rules are unnecessary, the fuzzy inference is continued without identifying the unnecessary rules. This is the reality.

As a result, there is a problem in that inference time is wasted by an amount corresponding to the number of unnecessary rules, and inference time is delayed.

This invention was made by focusing on such conventional problems, and it outputs all the degrees of conformance to a rule when fuzzy inference is performed, and uses that degree of conformity to create a system that is included in multiple rules. The purpose is to solve the above problems by detecting unnecessary parts and eliminating wasted inference time.

(Means for Solving the Problems) In order to achieve the above-mentioned object, the present invention includes a suitability output unit that outputs all the suitability of each rule, and a suitability output unit that outputs all the suitability of each rule. The present invention is characterized by comprising a buffer section that stores maximum and minimum values of the degree of suitability of rules, and an unnecessary rule determination block section that determines unnecessary rules based on the maximum and minimum values of the degree of suitability.

(Operation) In the above configuration, when performing fuzzy inference, all the degrees of conformity for a plurality of rules are first outputted by the degree of conformity output section. Then, all the output degrees of conformity are stored in the buffer section, and the maximum value of the degree of conformity for each rule is
Temporarily store the minimum value. Next, the maximum value and minimum value of the second degree of conformance are manually entered into the unnecessary rule determination block section, and unnecessary rules are determined here while referring to the maximum value and minimum value of the degree of conformity. As a result, even if there are many unnecessary rules,
Since these can be deleted and fuzzy inference can be performed efficiently, the inference speed can be increased.

(Example) An example of the present invention will be described below based on the drawings.

FIG. 1 is a block diagram showing the overall configuration of a fuzzy inference system according to this embodiment.

In this figure, 1 represents a plurality of data X1lX21...
, xn and performs fuzzy inference manually.
It has a function as a suitability output section.

This inference control unit 1 generates data Xl, X2 . . . , all the fitness degrees of xn are displayed. A fitness buffer section 2 and a maximum/minimum buffer section 3 are sequentially connected to one output side of the inference control section 1, and a decision block section for each group is connected to the other output side of the inference control section 1. 4
is connected.

Among these, the suitability buffer section 2 receives the output from the inference control section 1 and stores the suitability of each rule.
The maximum/minimum buffer section 3 receives the output from the suitability buffer section 2 and stores the maximum and minimum suitability values within the range of the input fluctuation.

Further, the group determination block section 4 receives the output from the inference control section 1 and performs grouping based on the maximum value and minimum value of the degree of fitness in the consequent part. The decision block unit 4 for this group includes an unnecessary rule determination block unit 5 that determines unnecessary rules within each group, and a determination result buffer unit 6 that stores and accumulates the determination results of this unnecessary rule determination block unit 5. are connected sequentially.

Next, the algorithms in each component of the fuzzy inference system will be explained.

First, regarding the inference control unit 1, it is assumed that a plurality of rules are set as follows.

Rule 1 if x,=A, and x2=A2...t
hen yI=B1-・Rule 2 if x3=A3 and X4=A4・-t
hen yI =B1 ・-Here, human horn data X
I + X2+ . . . , xn has a variation range determined in advance depending on the intended control content. These input data XII X21 ..., x
n is varied within a range that requires human effort, and the maximum and minimum values in the antecedent part for each rule at this time are stored in the suitability buffer section 2.

FIG. 2 shows an example of the storage state in the suitability buffer section 2, where max uz and min ui represent the maximum value and minimum value of the suitability for the second rule, respectively.

For rules with the same consequent, the maximum value max ui of their fitness.

The minimum value min ui is summarized for each consequent part.

For example, organize and summarize as follows.

Rules regarding V+=B+ Rule 1 max ul min ul Rule 2 max u2 min u2Y+=B
Rule 2 Regarding Rule 2 Now, it is assumed that among the rules regarding the consequent part Y+=B+, the compatibility of Rule 1 and Rule 2 has changed as shown in FIG. That is, the maximum value maX ul and the minimum value min ul of the fitness for rule 1 are respectively 0.
8. 0. 5, and the maximum value max u2 and minimum value min u2 of the fitness for rule 2 are respectively 0. Suppose that it changes to 3.0°1. Then, within the range in which the human power value fluctuates in the consequent part 5'1=B1, as shown in FIG. 4, the maximum and minimum values are 0.
It becomes 8°0.5. Therefore, in a human-powered state like '-', Rule 2 is as far as the consequent y+=B+ is concerned.
This means that it will not have any effect within the range in which the human power value fluctuates. In other words, the part related to the consequent part VI=82 of rule 2 no longer needs to be used when outputting the inference result, so there is no problem even if this part is removed.

Below, the algorithms for each main component will be explained in detail using diagrams.

In the inference control unit 1, as shown in FIG. 5, first, a plurality of data XI+X2+ . . .
Perform fuzzy inference based on xn. In this case, the degree of suitability of each rule is determined using the membership function, and this degree of suitability is temporarily stored in the suitability buffer section 2.

Thereafter, the maximum and minimum values of suitability for each rule are determined and stored in the maximum/minimum buffer section 3. This will be explained in detail with reference to FIG. First, to find the maximum value, start from when the rule number is 1 in step 61. In step 62, it is determined whether the rule number is greater than the total number of rules N. It ends at that point. On the other hand, if the number of rules is smaller than the total number N of rules, it is determined whether the suitability ul of the L-th rule is larger than a predetermined maximum value u1-(max) (step 63). . the result,
If the fitness uL is larger than the maximum value uL(max),
On the condition that the fitness degree uL in step 64 is equal to the maximum value uL(max), the process proceeds to step 65; otherwise, the step 65 is directly performed without going through step 64.
Proceed to.

Next, in step 65, a predetermined minimum value uL(
As a result, if the fitness degree ul is smaller than the minimum value uL(min), step 66 determines whether the fitness degree u1- is smaller than the minimum value uL(m
in), the process proceeds to step 67; otherwise, the process proceeds directly to step 67 without passing through step 66. In step 67, the rule number is L.
On the condition that the number increases from L+1 to L+1, the process returns to step 62 and the above-described procedure is repeated again.

In addition, when calculating the maximum value uL (may) and minimum value ul (min) according to the above operation, XIS≦x1≦Xl
e, X2 S≦X2≦x2e.

-xn S≦xn≦xne The data XI+ x2+ . Here, X, S, X2
S.

..., xns is the minimum value of each human data, xle, x2
e, . . . , xne are the maximum values of each human power data.

The maximum MuL(maX
) and the minimum value uL (min) are stored in the maximum/minimum buffer unit 3 for each rule, as shown in FIG.

Next, for the groups that receive pressure from the maximum/minimum buffer section 3 and the inference control section 1, the decision block section 4 calculates the maximum value uL (max) and minimum value uL (min) of the fitness according to the type of the consequent. ) into groups.

FIG. 8 is a flowchart showing this procedure, where m is the total number of outputs in the consequent part, ! (i) is the output yi
is the total number of labels. When this group is divided into groups by the decision block section 4, first step 81
Starting from the output of the first consequent, it is determined in step 82 whether the output number i is greater than the total number m of outputs, and if so, the grouping ends at that point. On the other hand, if the output number i is smaller than the total number m of outputs, the process starts from the first label in the output yi in step 83, and the label number j is equal to the total number of labels! (i) Determine whether it is larger than (i). As a result, label number j is the total number of labels! If it is larger than (i), the process returns to step 82 on the condition that the output number increases from i to i+1 in step 85, and conversely, label number j is the total number of labels! If it is smaller than (i),
yi=Bj (i) The maximum and minimum values of the fitness for the rule having the consequent part are selected.

FIG. 9 is a flowchart showing this procedure. First, in step 91, the first rule is started, and in step 92, it is determined whether the rule number k is greater than the total number N of rules. Then move on to step 86.
The process returns to step 84 on the condition that the label number increases from j to j+1. On the other hand, if the rule number k is smaller than the total number N of rules, it is determined in step 93 whether or not the th Lulu has yi=Bj (i) in the consequent, and if so, the l(th The maximum and minimum fitness values for the rule are put into the group of the consequent part yi-Bj (i) (step 94).Then, in step 95, the rule number is increased from 1 (from 1 to +1). Returning to step 92, the same procedure as described above is repeated again.

In this way, for each group, the maximum and minimum values of the degree of suitability for each rule grouped in the decision block section 4 are organized for each type of consequent part yi=Bj (i). For example, FIG. 10 shows an example. Regarding the consequent part y+=B+(1), the maximum and minimum values of the fitness for rules 1 and 3 are summarized, and the consequent part y1=B+ (2 ), the maximum and minimum conformance values for Rule 4.6.7 are summarized.

Next, the group part is determined in the decision block part 4 by the consequent part y
After the maximum and minimum values of the degree of suitability for each rule are divided into groups for each type of i=Bj (i), unnecessary rules within each group are determined.

When specifically determining unnecessary rules, the unnecessary rule determination block unit 5 calculates the degree of suitability for the rules grouped into groups.

Figure 11 shows this calculation procedure. First, step 1
Step 11 starts from the first rule for the first output y1, and in step 112 it is determined whether the rule number i is greater than the total number N of rules, and if it is, the calculation ends at that point. On the other hand, if the rule number i is smaller than the total number N of rules, in step 113 the output y
On the condition that the label for j is 1, the process proceeds to step 114, where the label number j is the total number of labels! (
i) Determine whether it is larger than i). If the label number j is thicker than the total number of labels, e(i), step 1
In step 15, the process returns to step 112 on the condition that the rule number increases from i to i+1.

The label number j is the total number of labels! If it is smaller than (i), among the rules related to the consequent part yi=Bj (i), the one with the maximum fitness degree u (min) is determined by the following procedure.

FIG. 12 shows this procedure. First, step 121
In step 122, the consequent part yi=B starts from the first rule when the variable i related to the fitness is 0.
The total number of rules belonging to the group j (i) P(
It is determined whether the rule number k is larger than i, j). As a result, the rule number k is the total number of rules P(i,
j), in step 123 the consequent part yi=
The minimum fitness value u min (i, j,
k) is larger than the variable r. If the minimum value u min (i, j, 10 is larger than the variable r, the minimum value u m1n (
i, j, k) and the variable r are equal to each other, the process proceeds to step 125, and conversely, the minimum value u min
If (i, j, lc) is smaller than the variable r, then
The process directly proceeds to step 125 without passing through step 124. In step 125, the rules are numbered from k to 1 (+
On the condition that the number increases to 1, the process returns to step 122 and the above-described procedure is repeated again.

Then, in step 122, if the rule number l (is larger than the total number of rules P(i, j), it is determined whether the rule is unnecessary. In other words, as shown in FIG. 13, first, in step 131, the first Starting from Lulu, in step 132, the rule number k is larger than the total number of rules P(i, j) belonging to the group with the consequent part yi=Bj (i).
If the rule number I (is larger than the total number of rules P(i, j), the process moves to step 116 in FIG. 11, and the rule number is increased from j to j+1. As a condition, the process moves to step 114.

On the other hand, rule number 1 (is the total number of rules P (i.

j), it is determined in step 133 whether the variable r is larger than the maximum value u max (i, j, k) of fitness. As a result, if the variable r is larger than the maximum fitness value u max (i, j, k), then the l(th rule for the group of consequents yi = Bj (i) is not necessary for the consequents). The next rule number is then stored in the determination result buffer section 6 (step 136), and the process then proceeds to step 134.

Further, in step 133 above, the variable r is set to the maximum value u ma
If x is smaller than (i, j, Ic), directly perform step 134 without passing steps 135 and 136.
Proceed to. In this step 134, on the condition that the rule number increases from k to +1, the process returns to step 132 and repeats the same procedure as described above.

As described above, in the fuzzy inference system according to this embodiment, the inference control section 1 first outputs all the degrees of suitability for each rule, stores this in the suitability buffer section 2, and then calculates the suitability for each rule. The maximum and minimum values of are stored in the maximum/minimum buffer section 3. Then, the maximum value of this fitness,
The minimum value for each group is manually input to the decision block section 4, and the maximum value and minimum value of the degree of fitness are grouped according to the consequent part. Then, the result of this grouping is manually input to the unnecessary rule determination block section 5, and unnecessary rule numbers are stored in the determination result buffer section 6. As a result, when performing fuzzy inference, unnecessary rules can be deleted, so that the inference time can be shortened by the number of deleted rules, and the inference speed can be increased.

(Effects of the Invention) As explained above, according to the present invention, by outputting all the degrees of conformance to rules when fuzzy inference is performed on the configuration, and using the maximum and minimum values of these degrees of conformity, Since unnecessary parts of multiple rules are determined, in fuzzy inference using software or hardware, it is possible to improve the inference speed by dividing the unnecessary rules and reducing the inference time. have

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of the fuzzy inference system according to the present invention, Fig. 2 is an explanatory diagram showing a state in which the maximum and minimum values of fitness are stored, and Fig. 3 is a rule 1.2.
4 is an explanatory diagram showing a state in which the fitness of 6 is a flowchart showing the algorithm in the maximum/minimum buffer section, FIG. 7 is an explanatory diagram showing the storage state of information in the maximum/minimum buffer section, and FIGS. 8 and 9 are for determining the groups, respectively. A flowchart showing the algorithm in the block section, FIG. 10 is an explanatory diagram showing the state of the group section in the decision block section, and FIGS. 11, 12, and 13 are the unnecessary decision block section and the judgment results, respectively. 3 is a flowchart showing an algorithm in a buffer unit. 1... Inference control section (fitness output section) 2... Suitability buffer section 3... Maximum/minimum buffer section 4... Decision block section for groups 5... Unnecessary rule judgment block section

Claims (1)

[Claims] 1. a suitability output unit that outputs all the suitability of each rule; a buffer unit that receives the output from the suitability output unit and stores the maximum and minimum suitability values of each rule; and the maximum suitability value. and an unnecessary rule determination block unit that determines unnecessary rules based on the minimum value.