JP2605965B2 - Fuzzy inference rule rearranging method and apparatus, fuzzy inference rule coding method and apparatus, fuzzy inference processing method and apparatus according to the fuzzy inference rule, rule rearranging method and apparatus, rule coding method and apparatus, and Inference processing method and apparatus according to the rule - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a method and apparatus for rearranging fuzzy inference rules, a method and apparatus for coding fuzzy inference rules, and performs fuzzy inference processing in accordance with rearranged and coded fuzzy inference rules. The present invention relates to a method and apparatus for performing rules, a method and apparatus for rearranging rules, a method and apparatus for coding rules, and a method and apparatus for performing inference processing in accordance with rearranged and coded rules.

BACKGROUND ART For the purpose of utilizing knowledge that cannot be described by mathematical formulas such as know-how of a skilled technician or a skilled technician for automatic control of a highly non-linear controlled object such as automatic control, decision making, and fault diagnosis. The application of fuzzy theory has been widely practiced.

Knowledge for fuzzy inference is often expressed by multiple (sets of) rules described in IF, THEN format. The antecedent part generally contains a plurality of fuzzy propositions (pairs of input variables and membership functions), which are connected by AND. The consequent may also include multiple fuzzy propositions (pairs of output variables and membership functions).

Such a set of fuzzy inference rules is stored in a general-purpose digital computer programmed to perform fuzzy inference operations or a fuzzy processor having an architecture dedicated to fuzzy inference operations. A digital computer or a fuzzy processor performs a fuzzy inference operation by applying given input data to rules. This fuzzy inference operation is often executed by so-called sequential processing, in which rules for which arithmetic is set are sequentially processed one by one.

The same fuzzy proposition may be included in a set of fuzzy inference rules. This is because a part of the rule content is allowed to be duplicated in the rule creation. When the number of rules is extremely large, the number of overlapping fuzzy propositions naturally increases, and may become redundant.

As described above, fuzzy inference operations are often performed by sequential processing, so if some rules overlap, the same operation will be repeated multiple times, resulting in poor processing efficiency and reduced operation speed. descend.

An object of the present invention is to enable high processing efficiency to be ensured even if a part of a set of rules is duplicated.

DISCLOSURE OF THE INVENTION The present invention provides a method and apparatus for rearranging fuzzy inference rules, a method and apparatus for coding fuzzy inference rules,
And a fuzzy inference processing method and apparatus.

The rearrangement method of the fuzzy inference rules according to the present invention is as follows.
For a given set of fuzzy inference rules, each consisting of an antecedent and a consequent, (a) a rule that extracts the antecedent proposition with the highest occurrence frequency and that includes the extracted antecedent proposition in common (B) In the above group, if there is a rule in which the antecedent ends with a common antecedent proposition, the rule is
(C) extracting the antecedent proposition with the next highest occurrence frequency for the remaining rules in the group, and including the extracted antecedent proposition in common A subgroup is formed by a rule, and (d) the steps (b) and (c) are repeated for the above subgroup, so that the common antecedent propositions are arranged in descending order of appearance frequency of the common antecedent proposition. The rules are stored in the rule memory in the order in which the rules are included in a larger proportion. (E) After the processes of steps (a) to (d) are completed for one group, the remaining rules excluding the rules included in the group are removed. The processing of steps (a) to (d) is repeated for a rule until there is no more rule.

The rearrangement device for fuzzy inference rules according to the present invention
For a given fuzzy inference rule consisting of an antecedent part and a consequent part, one antecedent proposition with the highest occurrence frequency is extracted, and one group is formed by rules containing the extracted antecedent proposition in common. Grouping means to form
If there is a rule in the group where the antecedent ends with a common antecedent proposition, the first storage control means for storing the rule in the rule memory and excluding the rule from the group; Subgrouping means for extracting the antecedent proposition having the next highest appearance frequency for the rule, and forming a subgroup by a rule including the extracted antecedent proposition in common; and the first storage control for the subgroup By repeating the processing by the means and the subgrouping means, the rules are stored in the rule memory in the order of the appearance frequency of the common antecedent proposition and in the order in which the common antecedent proposition is included in a larger proportion. After the processing from the grouping means to the second storage control means for the storage control means and one group is completed, The remaining rules except rule included in the group of treatment with the said grouping means to the second storage control unit, in which is provided a control means for controlling so as to repeat until the rule is eliminated.

The coding method of the fuzzy inference rule according to the present invention comprises the steps of: (a) extracting a proposition having the highest appearance frequency with respect to a given plurality of fuzzy inference rules each having an antecedent part and a consequent part; A group is formed by rules that include the extracted antecedent proposition in common. (B) In the above group, if there is a rule that ends the antecedent with a common antecedent proposition, the rule is coded by coding the rule. (C) extracting the antecedent proposition with the next highest occurrence frequency for the remaining rules in the group, and including the extracted antecedent proposition in common A subgroup is formed by the rule, and (d) the processing of steps (b) and (c) is repeated for the subgroup, whereby the frequency of appearance of the common antecedent part proposition. The rules are coded and stored in the rule memory in descending order and in the order that the common antecedent proposition is included in a larger proportion. (E) The processing of the above steps (a) to (d) is performed for one group. After the end, the processes of steps (a) to (d) are repeated for the remaining rules excluding the rules included in the group until there are no more rules.

In one embodiment of the fuzzy inference rule coding method according to the present invention, a rule is decomposed for each proposition of the antecedent part and the consequent part to form a subrule, and variables and membership functions included in the subrule are encoded. Become

In another embodiment of the fuzzy inference rule coding method according to the present invention, a command is added to each sub-rule code depending on whether the sub-rule is included in the preceding part or the consequent part, and according to its order position. Is stored in the rule memory.

The fuzzy inference rule coding apparatus according to the present invention extracts and extracts the antecedent part proposition with the highest frequency of occurrence for a given plurality of fuzzy inference rules each consisting of an antecedent part and a consequent part. Grouping means for forming a group by rules containing the antecedent proposition in common. In the above group, if there is a rule that ends the antecedent with a common antecedent proposition, the rule is coded and stored in the rule memory. First encoding means for removing the rule from the group, extracting the antecedent proposition having the next highest frequency of occurrence from the remaining rules of the group, and including the extracted antecedent proposition in common Sub-grouping means for forming a sub-group according to a rule, and the sub-group is provided to the first coding means and the sub-grouping means Coding means for coding rules and storing them in a rule memory in the order in which the frequency of appearance of the common antecedent proposition is high and in the order in which the proposition of the common antecedent proposition is included in a larger proportion. , And after the processing from the grouping means to the second coding means is completed for one group, the remaining rules excluding the rules included in the group are processed from the grouping means to the second coding means. The control means controls to repeat the process according to (1) until there are no more rules.

According to the present invention, a plurality of rules are arranged by paying attention to overlapping fuzzy propositions described therein, and overlapping rules are rearranged so that inference processing can be completed by a single fitness calculation. This eliminates the need to repeat the same operation many times, shortens the processing time, and increases the fuzzy inference efficiency.

The fuzzy inference processing method according to the present invention executes an arithmetic processing in accordance with the rules coded and rearranged by the above-described fuzzy inference rule coding method, and executes a sub-rule code including a command from a rule memory. Read and determine the type of command of the read sub-rule code, and if the above command indicates that it is the first sub-rule in the antecedent, use the given input data to determine the conformity according to that sub-rule. Is calculated, and the calculated conformance is temporarily stored. If the above-mentioned command indicates that the sub-rule is located at the second or subsequent position in the antecedent part, the sub-rule is obtained using the input data provided. Is calculated in accordance with the following formula, and the calculated fitness is calculated and temporarily stored. A predetermined calculation is performed with respect to the degree of conformity, and the result of the calculation is temporarily stored. If the command indicates that it is a subrule of the consequent part, the rule is calculated first and temporarily stored. A predetermined consequent part operation is performed using the matching degree.

A fuzzy inference processing apparatus according to the present invention is an apparatus for executing fuzzy inference processing in accordance with rules coded by the above-described fuzzy inference rule coding apparatus and stored in a rule memory. Means for reading the code of the included sub-rule, means for determining the command type of the read sub-rule code, and when the command indicates that it is the first sub-rule in the antecedent part, the given input data is First calculating means for calculating the conformity according to the sub-rule and temporarily storing the calculated conformity, indicating that the command is a sub-rule located at the second or subsequent position in the antecedent part Uses the given input data to calculate the fitness according to its subrules, A second calculating means for performing a predetermined calculation between the calculated degree of conformity and the previously calculated and temporarily stored degree of conformity, and temporarily storing the result of the calculation; In the case of indicating that the rule is a sub-rule, a third calculating means for performing a predetermined consequent part calculation using the previously calculated and temporarily stored matching degree for the rule is provided. .

According to the present invention, overlapping fuzzy propositions appear intensively, and the fitness calculated for the fuzzy proposition is temporarily stored and used in the processing of another rule. Since there are not so many of the calculated conformances used in the processing of other rules, the temporary storage area of the conformity can be small. This makes it possible to save registers for storing the degree of conformity. This is particularly effective when constructing a digital fuzzy inference IC dedicated to fuzzy inference.

The present invention also provides a method and apparatus for rearranging rules, a method and apparatus for coding rules, and a method and apparatus for processing inference used in inference operation according to rules, including fuzzy inference.

The rule rearranging method according to the present invention includes the following steps: (a) extracting a proposition having the highest appearance frequency for a given plurality of rules each consisting of an antecedent part and a consequent part; A group is formed by rules that include the proposition proposition in common. (B) In the above group, if there is a rule that ends the antecedent part with a common proposition proposition, that rule is
(C) extracting the antecedent proposition with the next highest occurrence frequency for the remaining rules in the group, and including the extracted antecedent proposition in common A subgroup is formed by a rule, and (d) the steps (b) and (c) are repeated for the above subgroup, so that the common antecedent propositions are arranged in descending order of appearance frequency of the common antecedent proposition. The rules are stored in the rule memory in the order in which the rules are included in a larger proportion. (E) After the processes of steps (a) to (d) are completed for one group, the remaining rules excluding the rules included in the group are removed. The processing of steps (a) to (d) is repeated for a rule until there is no more rule.

The rule rearrangement apparatus according to the present invention extracts an antecedent proposition having the highest appearance frequency for a given plurality of rules each including an antecedent part and a consequent part, and extracts the extracted antecedent proposition. Grouping means for forming a group by a rule including in common in the above group, if there is a rule in which the antecedent part ends with a common antecedent part proposition, store the rule in a rule memory and store the rule in the group First storage control means for eliminating the remaining rules of the above group, extracting the antecedent proposition with the next highest occurrence frequency, and forming a subgroup based on the rule including the extracted antecedent proposition in common By repeating the processing by the first storage control means and the subgrouping means for the subgrouping means and the subgroup, the common antecedent part The second storage control means for storing rules in the rule memory in the order of the frequency of appearance of the titles and the order in which the common antecedent proposition is included in a larger proportion, and the second grouping means for the one group After ending the processing by the storage control means, control means for controlling the processing from the grouping means to the second storage control means with respect to the remaining rules excluding the rules included in the group so as to be repeated until there are no more rules. It is provided with.

The rule coding method according to the present invention comprises the steps of: (a) extracting a proposition with the highest appearance frequency for a given plurality of rules, each rule having an antecedent part and a consequent part; A group is formed by rules that include the proposition proposition in common. (B) In the above group, if there is a rule in which the antecedent part ends with a common antecedent proposition, the rule is coded and stored in the rule memory. And (c) extracting the antecedent proposition with the highest frequency of occurrence from the remaining rules of the group, and sub-grouping the rule by a rule including the extracted antecedent proposition in common. And (d) repeating the processing of the above steps (b) and (c) for the above subgroup, in the order of the appearance frequency of the common antecedent proposition and the common antecedent The rules are coded and stored in the rule memory in the order in which the propositions are included in a larger proportion. (E) After the processing of steps (a) to (d) is completed for one group, the rules are included in the group. The processes of steps (a) to (d) are repeated for the remaining rules excluding the rule to be executed until there are no more rules.

In one embodiment of the rule coding method according to the present invention, a rule is decomposed for each proposition of the antecedent part and the consequent part to form a subrule, and the subrule is coded for each subrule.

In another embodiment of the rule coding method according to the present invention, a command is added to the code of each sub-rule in accordance with whether the sub-rule is included in the preceding or consequent part, and in accordance with its order position. And store it in the rule memory.

The rule coding apparatus according to the present invention extracts an antecedent proposition having the highest appearance frequency for a given plurality of rules each including an antecedent part and a consequent part, and extracts the extracted antecedent proposition. Grouping means for forming a group by a rule that includes in common,
First encoding means for encoding a rule which terminates the antecedent part with a common antecedent part proposition, storing the rule in a rule memory and excluding the rule from the group, the remaining rules of the group , A subgrouping means for extracting the antecedent part proposition with the next highest frequency and forming a subgroup by a rule including the extracted antecedent part proposition in common, and the first coding means for the subgroup By repeating the processing by the sub-grouping means, the common antecedent part propositions appear in the descending order of appearance frequency and in the order in which the common antecedent part propositions are included in a larger proportion.
Second to encode rules and store them in rule memory
After the processing by the coding means and the processing from the grouping means to the second coding means for one group is completed, the remaining rules excluding the rules included in the group are subjected to the second processing from the grouping means. Control means is provided for controlling the processing up to the coding means to be repeated until there are no more rules.

According to the present invention, a plurality of rules are arranged by paying attention to overlapping antecedent propositions described therein, and reordering is performed so that inference processing is completed by a single antecedent part processing operation. Have been. This eliminates the need to repeat the same operation many times, shortening the processing time and increasing the inference efficiency.

The inference processing method according to the present invention executes arithmetic processing in accordance with the rules rearranged and coded by the above-described rule coding method, and reads and reads the code of a sub-rule including a command from a rule memory. Judge the type of command of the sub-rule code, and if the command indicates that it is the first sub-rule in the antecedent part, perform the first operation in accordance with the sub-rule using the given input data. Executed, and temporarily stores the first calculation result. If the command indicates that the sub-rule is located at the second or subsequent position in the antecedent part, the sub-rule is provided using the given input data. A first operation is performed in accordance with the sub-rule, and the first operation result and the first calculated and temporarily stored A predetermined second operation is performed with the result of the operation, and the second operation result is temporarily stored. If the command indicates that the rule is a subrule of the consequent part, the rule is first executed. A predetermined consequent part operation is performed using the calculated and temporarily stored operation results.

An inference apparatus according to the present invention is an apparatus for executing inference processing in accordance with rules coded by the above-described rule coding apparatus and stored in a rule memory. Reading means, means for determining the type of command of the read sub-rule code, and when the command indicates that it is the first sub-rule in the antecedent part, the sub-rule is determined using the input data given. Therefore, the first operation means for executing the first operation and temporarily storing the result of the first operation, when the command indicates that the command is a sub-rule at the second or subsequent position in the antecedent part Performs the first operation using the given input data in accordance with the sub-rule, and executes the first operation A second calculation means for performing a predetermined second calculation between the first calculation result and the first calculation result temporarily stored, and temporarily storing the second calculation result; A third calculating means for performing a predetermined consequent part operation using the operation result previously calculated for the rule and temporarily stored when the sub-rule indicates that the rule is a sub-rule of the subject part Things.

According to the present invention, duplicate antecedent propositions appear intensively, and the first calculation result calculated for the antecedent proposition is temporarily stored and used in processing of another rule. Since there are not so many of the calculated first calculation results used in the processing of other rules, the temporary storage area of the first calculation result can be small. By this,
It is possible to save registers for storing the first operation result. This is particularly effective when constructing a digital inference IC dedicated to inference.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an outline of an electrical configuration of a fuzzy inference processing device.

FIG. 2 shows an example of a given fuzzy inference rule.

FIG. 3 shows the fuzzy inference rules rearranged according to the method of the present invention.

FIG. 4 shows a tree structure of the rearranged fuzzy inference rules.

FIG. 5 is a flow chart showing a rule rearrangement and encoding process according to the present invention.

FIG. 6 shows a part of the rule memory in which the rearranged and coded rules are stored.

7a shows the input data area, FIG. 7b shows the output data area, and FIG. 7c shows the consequent data area.

 FIG. 8 shows the fitness register.

FIG. 9 is a flowchart showing the fuzzy inference operation processing according to the present invention.

BEST MODE FOR CARRYING OUT THE INVENTION FIG. 1 shows an outline of a configuration of a fuzzy inference processing apparatus. This fuzzy inference processor is a general-purpose digital computer programmed to perform fuzzy inference operations.
It is realized by a computer or a fuzzy processor having an architecture suitable for fuzzy inference operations.

The fuzzy inference processing device includes a CPU 10 that performs rearrangement and coding of rules, which will be described later, and an operating system.
ROM11 that stores the system, fuzzy inference execution program, fuzzy inference rule rearrangement and coding execution program, RAM12 that stores fuzzy inference rules, membership functions, etc. A fitness register 13 that is stored temporarily, a keyboard 14 for inputting fuzzy inference rules, etc., a port unit 15 for input / output of input / output data, etc., and displays input rules, inference results, etc. And a display device 16 such as a CRT or a liquid crystal display device.

To make the explanation concrete and easy to understand, it is assumed that rules (IF, THEN rules) as shown in FIG. 2 are set.

x1, x2, x3, x4 are input variables. Input variable type is 4
Not limited to y1 and y2 are output variables. The types of output variables are not limited to two.

PL to NL are labels indicating language information that the membership function means. There are seven kinds of labels, PL is just large (Positive Large), PM is just middle (Positive Medi)
um), PS is positively small (Positive Small), ZR is almost zero (Zero), NS is negatively small (Nagative Small), NM is negatively medium (Nagative Medium), and NL is negatively large (Nlgative L)
arge).

It is assumed that there are seven types of membership functions in the antecedent part for each input variable. Of course, the shape and position of the membership function may be different if the input variables are different even if the labels are the same.

It is also assumed that the above seven types of membership functions in the consequent part are set for each output variable. The consequent membership function can also be represented by a singleton to simplify the operation.

These membership functions are created in advance,
Data representing them is stored in the RAM 12.

The antecedent and consequent parts of each rule shown in FIG. 2 are decomposed for each input variable or output variable. This decomposition is called a subrule. One sub-rule contains one fuzzy proposition (pair of input and output variables and membership functions). For example, rule 1 is decomposed into the following four sub-rules.

If x1 = PL and x2 = PM and x4 = PS then y1 = PM A group of rules as shown in FIG. 2 is input from the keyboard 14 and stored in the RAM 12. The input order of these rules is arbitrary. It is assumed that rule numbers are assigned in the order of input.

According to the present invention, such a group of rules is analyzed and reorganized. The rule group shown in FIG. 2 is rearranged as shown in FIG.

This analysis and reorganization processing are performed according to the procedure shown in FIG.

First, the proposition (sub-rule) that appears most frequently in the antecedent part of all rules is found (step
twenty one). In the rule group shown in FIG. 2, the proposition x1 = PL appears most frequently.

Next, one group is formed by a rule including the proposition x1 = PL extracted in step 21 in the antecedent part (step 22). Since x1 = PL is included in rules 1, 2, 6, 8, and 9, these rules form one group.

In the rules belonging to this group, it is checked whether there is a rule having only the proposition x1 = PL extracted in step 21 in the antecedent part, and if so, the rule is extracted and the rule is coded (coded Is stored in the rule memory (RAM12) and this rule is excluded from the above group (step
23,24). Since the antecedent part of rule 8 consists only of x1 = PL, this rule 8 is coded and eliminated.

Among the above-mentioned groups, the antecedent proposition that appears most frequently among the rules other than the rule 8 excluded in step 24 is found (step 26). The most frequently appearing of the remaining rules 1, 2, 6, and 9 is proposition x2 = PM.

A subgroup is formed by a rule having the proposition x2 = PM in the antecedent part (step 27). Rules 1, 6, and 9 form a subgroup.

By repeating the same processing as in steps 23 to 27 for this subgroup, the rules are rearranged in the order of the appearance frequency of the common antecedent proposition and the order including the common antecedent proposition in a larger proportion, and the code Be transformed into The rearranged rules are excluded from the unordered rules.

The rules 1, 6 and 9 both include the antecedent proposition x1 = PL and x2 = PM, but there is no end of the antecedent proposition with these two antecedent propositions (NO in step 23). Therefore, these rules are further sub-grouped (step 26). Since there is no longer a common antecedent proposition, these rules 1, 6, and 9 are separate subgroups (1
Sub-groups containing only rules). Then, these rules 1, 6, and 9 are arranged in a certain order (in this embodiment, the order of other input variables and output variables included in the rules).

Since rule 2 has not been arranged yet in the above group, rule 2 is arranged after these rules 6, 9, and 1 (repetition of steps 23 to 27).

When the rearrangement of the rules in the group including x1 = PL is completed in this way (YES in step 25), if there are any remaining rules (YES in step 28), the above-described steps 21 to 27 are performed on the remaining rules. The process is repeated.

In the example shown in FIG. 2, the above processing is repeated for the remaining rules 3, 4, 5, 7, and 10.

Decomposition and coding of rules into sub-rules may be performed after rearrangement of all rules has been completed.

In any case, these rules are decomposed into sub-rules in parallel with the rule rearrangement or after all rules have been rearranged, and the sub-rules are arranged in an order suitable for sequential processing of fuzzy inference operations. And a command is given to each sub-rule.

For example, rules 1,2,6,8,9 belonging to the above group
Are organized so as to form a tree structure as shown in FIG.

The coding of the sub-rules is done by assigning a code representing them for each input variable and membership function. However, if the rule is already coded at the time of inputting the rule, the coding process is not necessary, and only the process of giving the command is performed.

The command assignment processing is as follows. The command IF is assigned to the most frequent proposition x1 = PL.

The command THEN is added to the consequent part of Rule 8 including only the proposition x1 = PL in the antecedent part.

The command AND1 is given to the proposition x2 = PM which is the second most frequent.

Rules 1, 6, which contain both these propositions x1 = PL and x2 = PM
In 9 the third proposition x3 = PM, x4 = PS command AND2
Is added, and the command THEN is added to the consequent part.

Proposition x1 = Second proposition x4 = ZR of other rule 2 including PL
To the command AND1, and then to the subject part the command THEN
Is given.

These sub-rules are arranged in the order of the rearranged rules and stored in the rule memory as shown in FIG. At the end of the sub-rule, a code DEF indicating the definite calculation process and information on the definite calculation process (indicating the type of output variable to be used for the definite calculation process and the definite calculation method)
Are arranged. An end code END and end processing information are arranged at the end of these rule data.

 Next, the fuzzy inference processing will be described.

The RAM 12 has a work area in addition to the rule memory area shown in FIG.
As shown in FIGS. 7a, 7b and 7c, the input data
An area, an output data area and a consequent part data area are provided.

The input data area (FIG. 7a) stores input data for each input variable, and has a storage location (for example, 1 byte each) for the number of types of input variables.

The output data area (FIG. 7b) stores fuzzy inference results for each output variable as output data, and has a storage location for the number of types of output variables.

The consequent part data area (Fig. 7c) stores the fitness to be applied to the membership function for each output variable. The number of types of membership functions in the consequent part (the number of types of output variables) X7 label).

The fitness register 13 is provided with a plurality of registers as shown in FIG. These registers temporarily store the fitness calculated as a result of the fitness calculation or the MIN calculation result, and each of the registers has a storage location for one byte. Register these registers with register Ri
(I = 1, 2, 3, 4,...).

FIG. 9 shows the flow of the fuzzy inference operation processing. This processing is started after input data for all input variables is set in the work area of the RAM 12.

The sub-rule codes are read one by one from the rule memory (FIG. 6) in accordance with the order of the arrangement (step 31), and the command associated with the read sub-rules is
, ANDi, THEN, or DEF is determined (steps 32, 35,
38,40).

If the command is an IF, since it is the antecedent, the input data for the input variable of the subrule is read from the input data area, and the degree of conformity of the input data to the membership function of the subrule is calculated (step 32,33). The degree of fitness is a membership function value (grade) obtained when input data (input value) of an input variable included in a subrule is given to a membership function of the same subrule. This degree-of-fit calculation may be performed by a known method.

Subsequently, the calculated conformity in the sub-rule is stored in the first register R1, and the contents of all other registers are cleared (step 34).

Then, an address indicating the address of the rule memory
The counter is incremented (step 42), and the process returns to step 31 to read the next sub-rule.

If the code is ANDi (i = 1, 2,...), The fitness in the sub-rule is calculated (steps 35 and 36), and the calculated fitness and the fitness stored in the register Ri are calculated. The MIN operation is performed, and the result of the MIN operation is stored in the register R
It is stored in (i + 1) (step 37). Register R
The number (i + 1) of (i + 1) is stored in the register number counter. This MIN operation is performed between all the antecedent subrules included in one rule. Thereafter, the address counter is incremented (step 42).

If the code is THEN, it is stored in the storage location corresponding to the conformity stored in the register of the number stored in the register number counter and the consequent part membership function in the consequent part data area. Value (initial value is 0)
Is performed, and the result of the MAX operation is stored in the storage location of the corresponding consequent part membership function in the corresponding consequent part data area (steps 38 and 39).
This MAX operation is used to obtain the only fitness that should be applied to the common consequent part membership function between rules that have the same consequent part (both y1 = PM), such as Rule 1 and Rule 6. Done. Thereafter, the address counter is incremented, and the process returns to step 31.

If the read code is DEF, the consequent part data
Using the fitness data stored in the area, a definite computation process is performed for each output variable (steps 40 and 4).
1). When the consequent part membership function is a singleton, the conformity of the consequent part data area directly indicates the height of the corresponding singleton. The definite operation processing includes a centroid method for finding the position of the center of gravity of the membership function (singleton) of the consequent part function subjected to such a goodness-of-fit, a maximum height method for selecting a membership function having the highest height, and the like. . The result of the definite calculation processing is stored in the output data area as output data for each output variable.

Thereafter, the address counter is incremented and the end code is read (steps 42 and 31), so that end processing such as output of output data is performed and one fuzzy inference operation is completed (step 43). .

This fuzzy inference processing method has the advantage that it is not necessary to prepare a fitness register equal to the number of rules, and it is sufficient to provide several fitness registers as shown in FIG.

INDUSTRIAL APPLICABILITY Fuzzy inference rule rearranging method and apparatus, fuzzy inference rule coding method and apparatus, and fuzzy inference processing method and apparatus, rule rearranging method and apparatus, rule coding method according to the present invention A method and apparatus for performing inference processing according to rearranged and coded rules is provided for improving processing efficiency, particularly in fuzzy inference processing units, fuzzy control units, and other rule-based inference or control units. It is preferably used for

(54) [Title of the Invention] Fuzzy inference rule rearranging method and apparatus, fuzzy inference rule coding method and apparatus, fuzzy inference processing method and apparatus according to the fuzzy inference rule, rule rearranging method And apparatus, coding method and apparatus for rules, and inference processing method and apparatus according to the rules

Claims (20)

(57) [Claims] For a given plurality of fuzzy inference rules each consisting of an antecedent part and a consequent part, (a) extracting the antecedent proposition with the highest appearance frequency and extracting the extracted antecedent proposition (B) If there is a rule in the above group that ends the antecedent part with a common antecedent proposition, the rule
(C) extracting the antecedent proposition with the next highest occurrence frequency for the remaining rules in the group, and including the extracted antecedent proposition in common A subgroup is formed by a rule, and (d) the steps (b) and (c) are repeated for the above subgroup, so that the common antecedent propositions are arranged in descending order of appearance frequency of the common antecedent proposition. The rules are stored in the rule memory in the order in which the rules are included in a larger proportion. (E) After the processes of steps (a) to (d) are completed for one group, the remaining rules excluding the rules included in the group are removed. A method for rearranging fuzzy inference rules, wherein the processing of steps (a) to (d) is repeated for rules until there are no more rules. 2. With respect to a plurality of given fuzzy inference rules each consisting of an antecedent part and a consequent part, (a) extracting the antecedent part proposition having the highest occurrence frequency, and extracting the extracted antecedent part proposition (B) In the above group, if there is a rule in the group where the antecedent part ends with a common antecedent part proposition, the rule is coded, stored in a rule memory, and the rule Is removed from the above group. (C) For the remaining rules in the above group, the antecedent proposition with the next highest frequency is extracted, and a subgroup is formed by rules including the extracted antecedent proposition in common. (D) By repeating the processing of steps (b) and (c) for the subgroup, the common antecedent proposition is more frequently ordered in the descending order of appearance frequency of the common antecedent proposition. The rules are coded and stored in the rule memory in the order in which the rules are included in the same order. (E) After the processes of steps (a) to (d) are completed for one group, the rules included in the group are removed. A method of coding a fuzzy inference rule, wherein the processing of steps (a) to (d) is repeated for the remaining rules until there are no more rules. 3. The method according to claim 2, wherein the rule is decomposed for each proposition of the antecedent part and the consequent part to form a subrule, and variables and membership functions included in the subrule are encoded. How to code fuzzy inference rules. 4. A method according to claim 1, wherein a command is added to a code of each sub-rule and stored in a rule memory in accordance with whether the sub-rule is included in the preceding or consequent part, and in accordance with the order position. 3. A method for coding a fuzzy inference rule according to claim 3. 5. A method for performing fuzzy inference processing according to rules coded by the method according to claim 4 and stored in a rule memory, comprising: Reads the code, determines the command type of the read sub-rule code, and if the command indicates that it is the first sub-rule in the antecedent part, uses the given input data to follow the sub-rule. The fitness is calculated, and the calculated fitness is temporarily stored. If the above command indicates that the subrule is located at the second or subsequent position in the antecedent part, the given input data is used. The fitness is calculated according to the sub-rule, and the calculated fitness is compared with the previously calculated and temporarily stored fitness. , And temporarily stores the result of the calculation. If the above command indicates that the rule is a subrule of the consequent part, the matching rule previously calculated and temporarily stored for that rule is used. A fuzzy inference processing method that performs a predetermined consequent part operation using degrees. 6. A plurality of rules each having an antecedent part and a consequent part: (a) extracting the antecedent proposition with the highest appearance frequency and sharing the extracted antecedent proposition (B) If there is a rule in the above group that ends the antecedent with a common antecedent proposition, the rule is
(C) extracting the antecedent proposition with the next highest occurrence frequency for the remaining rules in the group, and including the extracted antecedent proposition in common A subgroup is formed by a rule, and (d) the steps (b) and (c) are repeated for the above subgroup, so that the common antecedent propositions are arranged in descending order of appearance frequency of the common antecedent proposition. The rules are stored in the rule memory in the order in which the rules are included in a larger proportion. (E) After the processes of steps (a) to (d) are completed for one group, the remaining rules excluding the rules included in the group are removed. A rule rearrangement method in which the processing of the above steps (a) to (d) is repeated for rules until there are no more rules. 7. With respect to a plurality of rules each having an antecedent part and a consequent part, (a) extracting an antecedent proposition with the highest appearance frequency and sharing the extracted antecedent proposition (B) In the above group, if there is a rule in which the antecedent part ends with a common antecedent part proposition, the rule is coded and stored in the rule memory, and the rule is stored in the rule memory. (C) For the remaining rules in the above group, extract the antecedent proposition with the next highest frequency, and form a subgroup by using the rule that includes the extracted antecedent proposition in common. d) By repeating the processing of the steps (b) and (c) for the subgroup, the common antecedent proposition is included in the descending order of the appearance frequency of the common antecedent proposition, and the common antecedent proposition is included in a larger proportion. In order, the rules are coded and stored in the rule memory. (E) After the processing of the above steps (a) to (d) is completed for one group, for the remaining rules excluding the rules included in the group, A rule coding method which repeats the above steps (a) to (d) until there are no more rules. 8. The rule coding method according to claim 7, wherein the rule is decomposed for each proposition of the antecedent part and the consequent part to form sub-rules, and the sub-rules are coded. 9. A method according to claim 1, wherein a command is added to a code of each sub-rule and stored in a rule memory in accordance with whether the sub-rule is included in a preceding part or a consequent part, and according to its order position. A method for coding rules according to claim 8. 10. A method for performing inference processing in accordance with rules coded by the method according to claim 9 and stored in a rule memory, wherein a code of a sub-rule including a command from the rule memory is provided. Is read, and the type of the command of the read sub-rule code is determined. Perform the operation of 1,
And temporarily storing the first calculation result, and if the command indicates a sub-rule located at the second or subsequent position in the antecedent part, using the given input data, in accordance with the sub-rule. Executing the first operation, performing a predetermined second operation between the first operation result and the first operation result previously calculated and temporarily stored, and executing the second operation result; Is temporarily stored, and if the above command indicates that the rule is a subrule of the consequent part, a predetermined consequent part operation is performed using the operation result previously calculated and temporarily stored for the rule. , Inference processing method. 11. With respect to a given plurality of fuzzy inference rules each consisting of an antecedent part and a consequent part, an antecedent proposition with the highest occurrence frequency is extracted, and the extracted antecedent proposition is shared. A grouping means for forming a group by a rule including a rule, in the above-mentioned group, if there is a rule whose antecedent part ends with a common antecedent part proposition, store the rule in a rule memory and exclude the rule from the group First storage control means, for the remaining rules of the above group, extracting the antecedent proposition having the next highest frequency, and forming a subgroup by a rule including the extracted antecedent proposition in common Means, by repeating the processing by the first storage control means and the subgrouping means for the subgroup,
Second storage control means for storing rules in a rule memory in order of appearance frequency of common antecedent propositions and in an order of including common antecedent propositions in a larger proportion; and grouping means for one group After the processing by the first storage control means to the second storage control means is completed, the processing from the grouping means to the second storage control means is repeated for the remaining rules excluding the rules included in the group until there are no more rules. A fuzzy inference rule rearrangement device comprising: a control means for controlling. 12. With respect to a plurality of given fuzzy inference rules each consisting of an antecedent part and a consequent part, the antecedent proposition with the highest occurrence frequency is extracted, and the extracted antecedent proposition is shared. A grouping means for forming a group based on the included rule. If there is a rule in the group where the antecedent ends with a common antecedent proposition, the rule is coded and stored in a rule memory, and the rule is stored in the group. A first coding means for eliminating from the rule, for the remaining rules of the above group, extract the antecedent proposition with the next highest occurrence frequency, and form a subgroup by rules including the extracted antecedent proposition in common Subgrouping means, by repeating the processing by the first coding means and the subgrouping means for the subgroup,
A second coding means for coding rules and storing them in a rule memory in order of appearance frequency of common antecedent propositions and in order of including common antecedent propositions in a larger proportion; After the processing from the grouping means to the second coding means is completed, the processing from the grouping means to the second coding means for the remaining rules excluding the rules included in the group is eliminated. A coding means for fuzzy inference rules, comprising: 13. The first and second encoding means include:
13. The fuzzy inference according to claim 12, wherein the rule is decomposed for each proposition of the antecedent part and the consequent part to form a subrule, and a variable and a membership function included in the subrule are encoded. Rule coding device. 14. The first and second encoding means include:
A command is attached to a code of each sub-rule and stored in a rule memory according to whether the sub-rule is included in the preceding part or the consequent part, and according to its order position. 13. An apparatus for coding a fuzzy inference rule according to item 13. 15. An apparatus for performing a fuzzy inference process in accordance with rules coded by the apparatus according to claim 14 and stored in a rule memory, wherein the sub-rule includes a command from the rule memory. Means for reading the code, means for determining the type of command of the read sub-rule code, and if the command indicates that it is the first sub-rule in the antecedent part, use the given input data to A first calculating means for calculating the fitness according to the sub-rule and temporarily storing the calculated fitness, when the command indicates that the command is a sub-rule located at the second or subsequent position in the antecedent part, Using the given input data, calculate the fitness according to the sub-rules, and calculate the fitness and the previously calculated fitness. A second calculating means for performing a predetermined calculation with the temporarily stored matching degree and temporarily storing the calculation result; and when the command indicates that it is a subrule of the consequent part, A third calculating means for performing a predetermined consequent part operation using the previously calculated and temporarily stored matching degree for the rule. 16. With respect to a plurality of given rules each consisting of an antecedent part and a consequent part, a rule that extracts the antecedent part proposition having the highest appearance frequency and that includes the extracted antecedent part proposition in common A grouping means for forming a group by the above-mentioned group. In the above-mentioned group, if there is a rule whose antecedent part ends with a common antecedent part proposition, the rule is stored in a rule memory and the rule is excluded from the group. A sub-grouping means for extracting the antecedent proposition having the next highest occurrence frequency from the remaining rules of the group, and forming a sub-group by a rule including the extracted antecedent proposition in common; By repeating the processing by the first storage control means and the subgrouping means for the subgroup,
Second storage control means for storing rules in a rule memory in order of appearance frequency of common antecedent propositions and in an order of including common antecedent propositions in a larger proportion; and grouping means for one group After the processing by the first storage control means to the second storage control means is completed, the processing from the grouping means to the second storage control means is repeated for the remaining rules excluding the rules included in the group until there are no more rules. A rule rearrangement device comprising a control means for controlling. 17. A rule which extracts the antecedent proposition having the highest frequency of appearance for a given plurality of rules each including an antecedent part and a consequent part, and includes the extracted antecedent part proposition in common. Grouping means for forming a group by the above group. If there is a rule in the above group that ends the antecedent with a common antecedent proposition, the rule is coded and stored in the rule memory, and the rule is excluded from the group. A first coding means for extracting the antecedent proposition having the next highest occurrence frequency from the remaining rules of the above group, and forming a subgroup by a rule including the extracted antecedent proposition in common By repeating the processing by the first encoding means and the sub-grouping means for the sub-group,
A second coding means for coding rules and storing them in a rule memory in order of appearance frequency of common antecedent propositions and in order of including common antecedent propositions in a larger proportion; After the processing from the grouping means to the second coding means is completed, the processing from the grouping means to the second coding means for the remaining rules excluding the rules included in the group is eliminated. A rule coding device comprising a control means for controlling the repetition of the rule. 18. The first and second encoding means include:
18. The rule coding apparatus according to claim 17, wherein the rule is decomposed for each proposition of the antecedent part and the consequent part to form sub-rules, and the sub-rules are coded. 19. The first and second encoding means include:
A command is attached to a code of each sub-rule and stored in a rule memory according to whether the sub-rule is included in the preceding part or the consequent part, and according to its order position. An apparatus for coding rules according to clause 18. 20. An apparatus for executing an inference process in accordance with rules coded by the apparatus according to claim 19 and stored in a rule memory, wherein a code of a sub-rule including a command from the rule memory is provided. Means for reading the sub-rule code read out, means for judging the command type of the read-out sub-rule code, and when the command indicates the first sub-rule in the antecedent part, the sub-rule using the given input data. Performs the first operation according to
And a first operation means for temporarily storing the first operation result. If the command indicates a sub-rule located at the second or subsequent position in the antecedent part, the given input data is used. The first operation is executed according to the sub-rule, and a predetermined second operation is performed between the first operation result and the first operation result previously calculated and temporarily stored. A second calculating means for temporarily storing a second calculation result; and, if the command indicates a subrule of a consequent part, a calculation previously calculated and temporarily stored for the rule. A third calculating means for performing a predetermined consequent part calculation using the result.