JPH04329439A - Method and device for grouping rules and method and device for fuzzy inference according to grouped rules - Google Patents

Abstract

PURPOSE:To divide set rules into groups while paying attention to input/output variables and to operate fuzzy inference for each group as well. CONSTITUTION:Concerning plural rules, (1) the input variable contained in the antecedent part of one rule is extracted (step 1), (2) all the rules having the extracted input variables in the antecedent parts are extracted (step 3), (3) all the input variables containing the antecedent parts of the extracted rules are extracted (step 4), and (4) the processing is repeated from (2) concerning these all input variables. Therefore, one group is formed by the rules containing all the input variables related mutually at the antecedent parts.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for grouping rules that can be used for fuzzy inference systems and expert systems, and a fuzzy inference method and apparatus that follow the grouped rules.

0002

[Prior Art] Conventional fuzzy inference systems and expert systems rely on the experience and intuition of experts.
Rules are written in the format IF~THEN~. Fuzzy inference calculations were always performed with reference to all set rules.

0003

[Problem to be solved by the invention] However, in conventional fuzzy inference systems, calculations are performed with reference to all rules in each fuzzy inference, so when there are a large number of rules, it takes a very long time to obtain an inference result. It takes a lot of time, and calculations that follow unnecessary rules are performed, which is inefficient.

[0004] Furthermore, since hardware and software for input and output interfaces are required for each type of input and output variables used for all rules, the circuit size becomes large when there are many input and output variables. It turns into

[0005] According to the present invention, set rules are grouped by focusing on input and output variables, and fuzzy inference calculations can also be performed for each group.

[0006]

[Means for Solving the Problems] A method for grouping rules focusing on input variables according to the present invention is such that each antecedent part includes a pair of at least one input variable and its corresponding membership function, and each antecedent part includes a pair of at least one input variable and a membership function corresponding to it. In a plurality of rules whose consequent part includes at least one pair of an output variable and its corresponding membership function, (1) extract the input variables included in the antecedent part of one rule, and (2
) Extract all rules that have the extracted input variables in their antecedent parts, (3) extract all input variables included in the antecedent parts of the extracted rules, and (4) extract all of these input variables. By repeating step (2), one group is formed by rules that include all mutually related input variables in the antecedent part.

[0007] The rule grouping device according to the present invention focuses on input variables, and each antecedent part has at least one
a device for grouping a plurality of rules including a pair of one input variable and a membership function corresponding thereto, each consequent part including a pair of at least one output variable and a membership function corresponding thereto; A first means for extracting an input variable included in the antecedent part of a rule, a second means for extracting all rules having the extracted input variable in the antecedent part, and a second means for extracting an input variable included in the antecedent part of the extracted rule. A third means for extracting all of the included input variables, and repeating the processing by the second and third means for all these input variables, thereby including all of the input variables that are mutually related in the antecedent part. It is characterized by having a means for forming a group according to rules.

A method for grouping rules focusing on output variables according to the present invention is as follows. each antecedent is at least 1
In a plurality of rules that include a pair of one input variable and its corresponding membership function, and each consequent part includes a pair of at least one output variable and its corresponding membership function, (1) one rule Extract the output variable included in the consequent part of (2) Extract all rules that have the extracted output variable in the consequent part, (3)
By extracting all the output variables included in the consequent part of the extracted rule, and (4) repeating step (2) for all these output variables, all the output variables that are related to each other in the consequent part are extracted. A group is formed by the rules it contains.

[0009] The rule grouping device according to the present invention focuses on output variables, and each antecedent part has at least one
a device for grouping a plurality of rules including a pair of one input variable and a membership function corresponding thereto, each consequent part including a pair of at least one output variable and a membership function corresponding thereto; A first means for extracting an output variable included in the consequent part of one rule, a second means for extracting all rules that have the extracted output variable in the consequent part, and a second means for extracting all rules that have the extracted output variable in the consequent part; A third means for extracting all of the included output variables, and repeating the processing by the second and third means for all these output variables to include all of the output variables that are mutually related in the consequent part. It has a means to form a group based on rules.

The fuzzy inference device according to the grouped rules according to the present invention performs a fuzzy inference operation in which a group of rules is set for each rule that includes all mutually related input variables in the antecedent part of the rule. a means for checking the presence or absence of changes in input signals for all input variables and detecting input signals that have changed; and performing fuzzy inference operations based on the rules of a group that includes the input variables of the input signals that have changed. The fuzzy inference calculation means is further provided with means for controlling the fuzzy inference calculation means.

[0011] In the fuzzy inference method according to the grouped rules according to the present invention, a group of rules is set for each rule that includes all mutually related input variables in the antecedent part of the rule. The system checks whether there is a change in the input signal with respect to the input variable, detects the input signal that has changed, and uses the changed input signal to perform fuzzy inference operations based on the rules of the group that includes the input variable of that input signal. It is something to do.

The fuzzy inference device according to the grouped rules according to the present invention performs a fuzzy inference operation in which a group of rules is set for each rule that includes all mutually related input variables in the antecedent part of the rule. The means and the input signals of the input variables belonging to each group are taken in, and based on the rules of the corresponding group,
Means is provided for controlling the fuzzy inference calculation means to perform fuzzy inference calculation for each group.

[0013] In the fuzzy inference method according to grouped rules according to the present invention, a group of rules is set for each rule that includes all mutually related input variables in the antecedent part of the rule, and each It takes input signals of input variables belonging to a group, and performs fuzzy inference calculations for each group based on the rules of the corresponding group.

[0014]

[Operation] A large number of rules are grouped by focusing on the input variables of the antecedent part so that rules including all mutually related input variables in the antecedent part form one group.

[0015] Also, a large number of rules are grouped by focusing on the output variables in the consequent part, so that rules including all mutually related output variables in the consequent part form one group.

Further, in a fuzzy inference device in which grouped rules are set in this way, fuzzy inference calculations are performed in accordance with the rules included in the group related to the input variable of the input signal that has changed.

Further, in the fuzzy inference device in which grouped rules are set in this manner, input signals are taken in for each input variable belonging to each group, and fuzzy inference is performed for each group.

[0018]

[Effects of the Invention] According to this invention, since a large number of rules are grouped, it is very easy to understand when examining rules, and it is easy to discover omissions in rule creation, duplication of rules, etc. Analysis can be performed efficiently.

Furthermore, since fuzzy inference can be performed based only on groups related to input signals that have changed as described above, the fuzzy inference calculation time can be shortened.

Furthermore, since input signals can be taken in groups, the number of types of input variables for which input signals are taken in at once is reduced, and the input/output interface hardware or software is simplified.

[0021]

[Example] As an example, consider the following eight rules.

Rule 1: If x1 =A11,
then y1 =B11
and y2 =B12 and y4 =
B14 Rule 2: If x1 = A21 an
d x2 = A22 and x3 = A2
3, then y
1 =B21, Rule 3: If x1 =A31
and x3 = A33,
then y1 =B31 and
y3 = B33, Rule 4: If x3
=A43 and x4 =A44,
then y3 =B43
and y4 =B44, Rule 5: If
x5 = A55,
then y5 =B55, Rule 6: If
x5 = A65 and x6 = A66,
then y6 =B
66 and y7 =B67, rule 7:
If x5 = A75 and x7 = A7
7, then y
5 =B75 and y7 =B77, Rule 8: If x5 =A85 and x7
=A87, the
n y6 =B86 and y7 =B87
, where the consequent and and simply means that the antecedents are the same.

When these rules are expressed in a matrix focusing on the input variables x1 to x7 of the antecedent part, the result is as shown in FIG. In Figure 1, the circles indicate the input variables xi for the input variables x1 to x7 arranged vertically.
This shows that all rules having (i=1 to 7) in the antecedent part include input variables xi arranged horizontally. For example, input variable x1 is included in rule 1, rule 2, and rule 3. These rules 1 to 3 include input variables x1, x2, and x3. Therefore, the input variable x1 is the input variable x1, x2, and x3
It will only be related to.

If a similar matrix is created focusing on the output variables y1 to y7 of the consequent part, it will be as shown in FIG. 2.

It can be seen from such a matrix that rules can be grouped so that mutually related input variables and mutually related output variables form a group, respectively.

In this embodiment, the following four groups are formed.

That is, regarding the input variables, (1) rule group Gi1 related to x1 to x4
Gi1={Rule 1, Rule 2
, rule 3, rule 4}
(2) Rule group Gi2 related to x5 to x7
Gi2={Rule 5, Rule 6, Rule 7, Rule 8} Regarding the output variables, (3) y1 ~
Rule group Go1 related to y4
Go1={Rule 1, Rule 2, Rule 3, Rule 4}
(4) Rule group Go2 related to y5 to y7
Go2={Rule 5, Rule 6, Rule 7, Rule 8} can be grouped.

This is considered to be due to the following reasons.

In fuzzy reasoning and expert systems, know-how such as experience and intuition possessed by experts is expressed as rules in the form of If~, Then~.

[0030] It is extremely difficult for a person's general ability to draw a certain conclusion by considering all of the many conditions. Therefore, when describing experience and intuition using If~, then~ format rules, If, th
en rule, If, then when assuming another case
If you roughly classify the cases, such as rules,
, then rules are often created.

[0031] For this reason, if and then rules for a certain system become more grouped as the number of rules increases.

[0032] When performing control, fault diagnosis, etc. using fuzzy inference or an expert system, only rule groups related to changed inputs or desired outputs may be subjected to calculations.

The case where the above-described grouping algorithm is focused on input variables will be explained. This grouping process is performed by a programmed computer. This computer has a survey parameter storage section (A) and a grouping parameter storage section (B).
and a grouping rule storage section (C). These stores can be implemented with registers or memories. Parameters here refer to input variables. The grouping algorithm is shown in the form of a flow chart in Figure 3.
is shown.

Step 1: Take out the first parameter x1 of the first rule 1, and store the rule and parameters in storage units (A) and (B).
and (C) respectively (see Figure 4).

Step 2: Give a group name. In this example, the group is Gi1.

Step 3: Examine the parameters stored in the investigation parameter storage section (A) in order from the top. Then, it is checked which rules include those parameters, and if there are any rules that are not included in the grouping rule storage area (C), they are stored.

In this example, a parameter x1 is stored in the storage section (A). In addition to rule 1, rules including parameter x1 in the antecedent part include rule 2 and rule 3, so these rules 2 and 3 are stored in the storage section (C) (see FIG. 5).

Step 4 Grouping rule storage section (C
) Among the parameters included in the newly stored rule, check whether there are any parameters that are not included in the grouping parameter storage area (B), and if there are, check them and store them in the inspection parameter storage area (A) and the grouping parameter storage area (A). Store in the parameter storage section (B).

In this example, newly stored rules 2 and 3 include input variables x2 and x3, so these input variables x2 and x3 are stored in the storage section (A),
(B) (see FIG. 6).

Step 5 The parameters investigated this time are deleted from the investigation parameter storage section (A).

In this example, input variable x1 is deleted (see FIG. 7).

Step 6 Investigation parameter storage section (A)
Does the parameter still exist? If it is, return to step 3. If not, move on to step 7.

In this example, the process returns to step 3, and next, the input variable x2 in the storage section (A) becomes the object of investigation. Since the input variable x2 is included only in rules 2 and 3, the contents of the storage section (C) remain unchanged (see FIG. 8). Then, through step 4, the input variable x2 is deleted in step 5 (see FIG. 9).

From step 6, the process returns to step 3 again. Input variable x3 is the subject of investigation. Input variable x3 is also included in rule 4, so rule 4 is the storage part (C)
(See FIG. 10). Since rule 4 includes input variable x4, this input variable x4 is stored in storage units (A) and (B) in step 4 (
(See Figure 11).

Then, in step 5, the input variable x3 is deleted (see FIG. 12), and from step 6,
Returning to , input variable x4 becomes the subject of investigation. After step 4, input variable x4 is deleted in step 5 (
(See Figure 13). Since there are no longer any input variables in the storage section (A), proceed to step 7.

Step 7 Parameters in the grouping parameter storage section (B) and grouping rule storage section (
C) Give the group name determined in step 2 to the rule and indicate which group these parameters and rules belong to (see Figure 14).

Step 8 Check whether there are any rules left other than the grouped rules. If there are any remaining, the process returns to step 1, and if there are no, the process ends.

In this example, the process returns to step 1 and the investigation of input variable x5 of rule 5 is started. The above-described process is repeated to form group Gi2.

Groups Go1 and Go2 are formed by the same process for output variables.

The rules thus grouped are set in a fuzzy inference device and used for fuzzy inference.

Next, an example of a fuzzy inference device having grouped rules will be explained.

The fuzzy inference device shown in FIG. 15 includes an input processing section 11 and a fuzzy calculation section 12.

Normally, in fuzzy control, all inputs are taken in every certain control period and fuzzy inference is performed on all rules. However, according to this method, the inference calculation takes time.

In FIG. 15, the input processing section 11 detects a changed input, and only that input signal is given to the fuzzy operation section 12. The fuzzy calculation unit 12 performs inference calculations using only the rules of the group related to the input that has changed. In this way, reduction of inference time can be achieved.

FIG. 16 shows another fuzzy inference device. This fuzzy inference device also includes an input processing section 21 and a fuzzy calculation section 22. The number of possible inputs to the fuzzy calculation unit 22 is limited to 10 inputs a to j. The input processing section 21 takes in only inputs related to input variables belonging to each rule group from among many inputs (for example, 100 types), and supplies them to the fuzzy operation section 22 . The fuzzy calculation unit 22 performs calculation processing using the input according to only the rules of the applicable rule group. When the computation for one group of rules is completed, the input processing unit 21 takes in input related to the next group of rules.

For example, suppose that there are 100 inputs from input 1 to input 100 for a certain controlled object. The number of inputs to the fuzzy operation unit 22 is limited to 10 inputs a to j due to software and hardware constraints. If the rules are not grouped, inputs 1 to 100 are sequentially taken in 10 at a time, and inference is performed after inputs 1 to 100 are all collected in the fuzzy calculation unit 22. In this case, it takes time to import the data, and a section for storing the input data is required in the calculation section 22.

By grouping, only inputs related to group 1 can be taken in and inferred by, for example, bus switching. After the inference of group 1 is completed, if group 2 is then processed in the same manner, there is no need to cause the input stage of the fuzzy calculation unit 22 to perform any special processing. Furthermore, if the input processing section 21 and the fuzzy inference section 22 are operated in a pipeline manner, the overall throughput can be improved.

Currently, fuzzy rules are created by experts based on their experience and intuition. Therefore, some of the rules end up being duplicated or missing. If you try to investigate them, you need to look at all the rules and check the relationships between them.

[0059] By displaying grouped input and output variables and rules so that the user can understand them at a glance, it becomes possible to hierarchize the rules (all rules can be organized in what kind of rule block). It has advantages such as being able to check whether the rules are configured properly), and being efficient as it allows you to examine each block when investigating problems with rules. In other words, it can be used as a rule analysis tool.

[Brief explanation of drawings]

FIG. 1 is a matrix showing grouping of input variables.

FIG. 2 is a matrix showing grouping of output variables.

FIG. 3 is a flow chart showing a rule grouping process procedure.

FIG. 4 shows the contents of each storage unit in the rule grouping process.

FIG. 5 shows the contents of each storage unit in the rule grouping process.

FIG. 6 shows the contents of each storage unit in the rule grouping process.

FIG. 7 shows the contents of each storage unit in the rule grouping process.

FIG. 8 shows the contents of each storage unit in the rule grouping process.

FIG. 9 shows the contents of each storage unit in the rule grouping process.

FIG. 10 shows the contents of each storage unit in the rule grouping process.

FIG. 11 shows the contents of each storage unit in the rule grouping process.

FIG. 12 shows the contents of each storage unit in the rule grouping process.

FIG. 13 shows the contents of each storage unit in the rule grouping process.

FIG. 14 shows the contents of each storage unit in the rule grouping process.

FIG. 15 is a block diagram showing the configuration of a fuzzy inference device.

FIG. 16 is a block diagram showing the configuration of a fuzzy inference device.

[Explanation of symbols]

11, 21 Input processing section 12, 22 Fuzzy operation section

Claims (8)

[Claims] Claim 1: Each antecedent part includes a pair of at least one input variable and its corresponding membership function, and each consequent part includes a pair of at least one output variable and its corresponding membership function. In multiple rules including (1) extract the input variable included in the antecedent part of one rule, (2) extract all rules that have the extracted input variable in the antecedent part, (3) By extracting all the input variables included in the antecedent part of the extracted rule, and (4) repeating step (2) for all these input variables, all the mutually related input variables in the antecedent part are extracted. A method of grouping rules in which the rules they contain form a group. [Claim 2] Each antecedent part includes a pair of at least one input variable and its corresponding membership function, and each consequent part includes a pair of at least one output variable and its corresponding membership function. It is a device for grouping multiple rules including multiple rules, and it is a first means for extracting input variables included in the antecedent part of one rule, and extracts all rules that have the extracted input variables in the antecedent part. The second means, the third means for extracting all the input variables included in the antecedent part of the extracted rule, and repeating the processing by the second and third means for all these input variables. a rule grouping device, comprising: means for forming a group by rules containing all mutually related input variables in a subject; 3. Each antecedent part includes a pair of at least one input variable and its corresponding membership function, and each consequent part includes a pair of at least one output variable and its corresponding membership function. In multiple rules including (1) extract the output variable included in the consequent part of one rule, (2) extract all rules that have the extracted output variable in the consequent part, (3) By extracting all the output variables included in the consequent part of the extracted rule, and (4) repeating step (2) for all these output variables, all the output variables that are related to each other in the consequent part are extracted. A method of grouping rules in which the rules they contain form a group. 4. Each antecedent part includes a pair of at least one input variable and its corresponding membership function, and each consequent part includes a pair of at least one output variable and its corresponding membership function. It is a device for grouping multiple rules including multiple rules, and it is a first means for extracting an output variable included in the consequent part of one rule, and extracts all rules that have the extracted output variable in the consequent part. The second means, the third means for extracting all the output variables included in the consequent part of the extracted rule, and repeating the processing by the second and third means for all these output variables. a rule grouping device comprising: means for forming a group by rules containing all interrelated output variables in a subject; [Claim 5] A fuzzy inference calculation means set with a group of rules grouped by rules including all mutually related input variables in the antecedent part of the rule, and whether or not there is a change in the input signal regarding all the input variables. and means for controlling the fuzzy inference calculation means to perform fuzzy inference calculation based on the rules of the group including the input variables of the changed input signal. fuzzy inference device. [Claim 6] A rule group is set for each rule that includes all mutually related input variables in the antecedent part of the rule, and the presence or absence of a change in the input signal regarding all input variables is checked, A fuzzy inference method that detects a changed input signal and uses the changed input signal to perform fuzzy inference operations based on the rules of a group that includes the input variables of that input signal. [Claim 7] A fuzzy inference calculation means configured with a set of rules grouped by rules including all mutually related input variables in the antecedent part of the rule, and an input signal of the input variables belonging to each group. and means for controlling the fuzzy inference calculation means to perform fuzzy inference calculation for each group based on the rules of the group corresponding to the rules of the group. [Claim 8] In the antecedent part of the rule, a group of rules is set for each rule that includes all mutually related input variables, and the input signals of the input variables belonging to each group are taken in and handled accordingly. A fuzzy inference method that performs fuzzy inference operations for each group based on the rules of the group.