JPH0319030A - Production system and its inferring method - Google Patents

Abstract

PURPOSE:To omit the duplication of processes even if the inferences are carried out for each input of the attribute value of a variable fact and to perform these inferences at a high speed and with high efficiency by eliminating the part of an original rule where the attribute value of a fact covers a fixed value and maintaining only the part the covers the variable attribute value of the fact to use it as a new rule. CONSTITUTION:A fixed/variable state deciding means 6 extracts the fixed attrib ute value of a fact from a working memory 2 and sends it to a partial calcula tion mechanism 7. The mechanism 7 calculates partly the rule of a rule base based on the received attribute of the fact and erases the true part of a condi tional part of the rule to replace the original rule with the remaining part. Thus the original rule is reduced by an amount equal to the erased part and finally only a rule related to the variable attribute value of the fact is obtained. As a result, no duplication of processes occurs even if the inferences are carried out every time the variable attribute value is changed. Then the inferences are performed at a high speed and with high efficiency.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a production system, and more particularly to a production system that can avoid redundant processing and perform inference at high speed and with high efficiency.

[Conventional technology] An example of a conventional production system is shown in FIG.

This production system 51 includes a working memory 52 consisting of a set of facts, a rule base 53 consisting of a set of rules, an inference mechanism 54 that applies the rules based on the facts and makes inferences, and an input/output An interface 55 is provided.

The facts in the working memory 52 have a structure as shown in FIG. That is, one fact has one fact name, and one fact has one or more attributes. One attribute has one attribute name, and one attribute has one attribute value. Such facts are input to the working memory 52 via the interface 55, but when the attribute values are not determined, provisional initial values are set for the attribute values of the input facts. In the chart shown in FIG. 9, attribute values other than ``0'' are determined values, and attribute values ``O'' are set to ``0'' as temporary values.

The rules in the rule base 53 have a structure as shown in FIG. That is, one rule has one rule name, and one rule has a condition part derived from IF and a conclusion part derived from THEN. One or more sentences can be written in both the conditional part and the conclusion part. A statement consists of an attribute name, an operator, and a value. The multiple statements written in the condition part become an AND condition.

The inference mechanism 54 operates as shown in the flowchart shown in FIG. That is, the attribute value of one fact is read from the working memory 52 (S1), and a rule having a condition part that matches the attribute value of that fact is extracted from the rule base 53 (S2). If attribute values of other facts are needed to judge the condition part of the rule, they are read out from the working memory 52.
2, an external input is requested via the interface 55 (S3). If all the sentences in the conditional part of one rule become true (S4), the conclusion part of that rule is added to the working memory 52 (S5). Such a cycle is repeated until there is no rule in which all of the statements in the condition part are true, and the operation is ended (S4). At this time, the result of the inference has been obtained in the working memory 52.

Techniques for speeding up production systems such as those described above are described in the Journal of the Information Processing Society of Japan [Information Processing J V o
I, 29. No5 (May 1988) No. 46
Techniques are known, such as those disclosed on pages 7 to 477, for improving algorithms for extracting rules in which all sentences in the condition part are true.

On the other hand, if you have a program that requires input of multiple pieces of data, and only some of the data is given, the part of the program that can be processed with only that part of the data is processed first, and the remaining parts are created anew. A theory of partial calculation is known in which a program is set as a new program, and when data of another part is given, processing is performed using the new program. Duplicate calculations can be avoided by creating a new program by giving fixed value data, and giving variable value data to the new program every time the value changes.

In reality, such partial calculations have traditionally been performed by programmers, but a study on how to perform this on a computer was published in a special edition of BIT magazine [New Era Software J (1984), pp. 81-91. Disclosed.

[Problems to be Solved by the Invention] In the conventional production system 51 described above, there may be cases where the attribute values of some facts are frequently changed while the attribute values of other facts are always constant. The mechanism 54 performs inference using all rules that match the attribute values of facts, so every time the attribute values of some facts are changed, it also makes inferences using rules that match the attribute values of facts that do not change. will be carried out.

In other words, the same process is repeated every time the attribute value of some facts is changed, which is wasteful.

Therefore, the inventor of this invention focused on the theory of partial calculations that aims to avoid duplicate calculations, and thought that if this theory was applied to production systems, the above waste could be prevented, and as a result of intensive research. , has completed this invention.

Therefore, an object of the present invention is to provide a high-speed, highly efficient production system equipped with partial calculation functions and an inference method thereof.

[Means for Solving the Problems] The production system of the present invention includes a working memory consisting of a set of facts, a rule base consisting of a set of rules, and an inference mechanism that applies the rules based on the facts and makes inferences. A production system comprising: fixed/variable discriminating means for discriminating between fixed and variable attribute values of facts in working memory; and a partial calculation mechanism that applies a rule-based rule to the original rule, deletes the true part of the conditional part of the rule, and replaces at least the conditional part of the original rule with the remaining part. is a structural feature.

Furthermore, from the point of view of Y, the production system of the present invention includes a working memory consisting of a set of facts, a rule base consisting of a set of rules, and an inference dedicated to the rule base that makes inferences based on the facts. A production system comprising: a rule compiler that applies the rules to create a rule base from the rule base; A structural feature further comprises: a partial calculation mechanism that partially calculates the inference mechanism based on the attribute value of the fact determined to be fixed by the fixed/variable determination means to create a new inference mechanism. It is something to do.

[Operation] In the production system of the present invention, the fixed/variable determining means extracts facts with fixed attribute values from the working memory and passes them to the partial calculation mechanism. The partial calculation mechanism partially calculates the rule based rule based on the attribute value of the passed fact, deletes the part of the condition part of the rule that is true, and replaces the original rule with the remaining part.

Therefore, it becomes smaller by the amount that the original rule is deleted,
In the end, there are only rules for facts whose attribute values are variable. Therefore, even if inference is performed every time there is a change in these variable attribute values, there is no duplication of processing, resulting in high speed and high efficiency.

In another aspect, the fixed/variable determining means extracts facts with fixed attribute values from the working memory and passes them to the partial calculation mechanism. The partial calculation mechanism partially calculates the dedicated inference mechanism based on the attribute value of the passed fact to create a new dedicated inference mechanism.

Therefore, the new dedicated inference mechanism is only concerned with facts whose attribute values are variable, eliminating redundant inferences with respect to facts whose attribute values are fixed, resulting in high speed and efficiency.

[Example] Hereinafter, the present invention will be explained in more detail with reference to Examples shown in the drawings. Note that this invention is not limited to this.

A production system 1 according to an embodiment of the present invention shown in FIG. 1 includes a working memory 2, a rule base 3,
In addition to comprising an inference mechanism 4 and an interface 5,
It further includes a fixed/variable discriminator 6 and a partial calculation mechanism 7.

The facts in the working memory 2 have a structure as shown in FIG. That is, one fact has one fact name, and one fact has one or more attributes. One attribute has one attribute name,
One attribute has one attribute value. Further, fixed and variable information indicating whether it is "fixed" or "variable" is held for each attribute value. The fixed variable information is given by an operator (for example, a fact input person), and is also given by the partial calculation mechanism 7, as will be described later.

The facts are entered into the working memory 2 via the interface 5. In addition, in the chart shown in Figure 2, attribute values other than rOJ are fixed values, and are ``0''.
" is one in which rOJ is set as a temporary attribute value.

The rules in the rule base 3 have a structure as shown in FIG. 3 or 4. That is, one rule has one rule name, and one rule has a condition part derived from IF and a conclusion part derived from THEN.

One or more sentences can be written in both the conditional part and the conclusion part. A statement consists of an attribute name, an operator, and a value. The multiple statements written in the condition part become an AND condition. Such a structure is the first
This is the same structure as the conventional rule shown in FIG. In addition,
FIG. 3 shows the original rule, and FIG. 4 shows the new rule after being replaced by a partial calculation mechanism 7, which will be described later.

The inference mechanism 4 basically operates in the same manner as before.

That is, as shown in FIG.
Read the attribute value of one fact in (Sl),
A rule having a condition part that matches the attribute value of the fact is extracted from the rule base 3 (S2). If attribute values of other facts are necessary for determining the condition part of the rule, they are read out from the working memory 2, but if they are not present in the working memory 2, input is requested from the outside via the interface 5 (S3). If all the sentences in the first part of the condition of one rule become true (S4), the conclusion part of that rule is added to the working memory 2 (S5). Such a cycle is repeated until there is no rule in which all of the statements in the condition part are true, and the operation is terminated (S5).

At this time, the result of the inference will be obtained in the working memory 2.

The fixed/variable discriminator 6 receives the fifth
It operates as shown in the figure. That is, check whether there is an attribute value of the fact to be retrieved in the working memory 2 (Dl), read the attribute value of that fact (D2), and check the fixed variable information of the attribute value of that fact (D3). ) If it is "fixed", the attribute value of that fact is passed to the partial calculation mechanism 7 (D4). If "variable", the process returns to step D1. In step D1, if there is no attribute value of the fact to be retrieved in the working memory 2, this is notified to the partial calculation mechanism 7 (D5).

The partial calculation mechanism 7 operates as shown in FIG. 6(a) in response to instructions from the operator.
It operates as shown in (b). In other words, the fixed/variable determination unit 6 is requested to retrieve the fact whose attribute value is "fixed" from the working memory 2 (Pl), and the attribute value of the fact is "fixed".
If there is an attribute value of the fact (P2), it is received from the fixed/variable determining unit 6 (P3). Next, extract the rule that matches the attribute value of that fact from Rule Base 3 (P4), delete the statement that is true in the conditional part of that rule, and delete the statement whose value is confirmed in the conditional part of that rule. The value is written to the working memory 2 as "fixed" and then erased (P5). Then, check whether there are any remaining sentences that cannot be deleted (P6), and if there are, create a new rule with a conditional part and a conclusion part made up of the remaining sentences, and replace the original rule with this ( Pl). On the other hand, if there are no remaining sentences in the conditional part, the process proceeds to the conclusion part of the rule, and for sentences whose values have been determined in the conclusion part of the rule, that value is written to the working memory 2 as "fixed" and then deleted (
P8). Then, create a new rule with a conclusion consisting of the remaining sentences, and replace the original rule with this (Page 9).
).

Next, the operation of this partial calculation mechanism 7 will be specifically explained using the facts shown in FIG. 2 and the rules shown in FIG. 3.

First, the partial calculation mechanism 7 starts operating according to an instruction from the operator, and requests the fixed/variable determination section 6 to determine that the attribute value of the fact is fixed.

The fixed/variable determining section 6 extracts the attribute value "1" of the attribute name a of the fact name A from the facts shown in FIG. 2, and since this is "fixed", it passes it to the partial calculation mechanism 7.

The partial calculation mechanism 7 extracts a rule name I having a condition part that matches the attribute name a of the fact name A from among the rules shown in FIG. This conditional part consists of two sentences, and the operator "=" in the first sentence represents "the left side and the right side are equal", and this sentence is true. Therefore, the partial calculation mechanism 7 deletes this sentence.

The second statement requires the attribute name of fact name A, so
A request is made to the fixed/variable determining section 6.

The N fixed variable determination section 6 extracts the attribute value "5" of the attribute name of the fact name A from the rules shown in FIG. 2, but since this is also "fixed", it passes it to the partial calculation mechanism 7.

The operator “→” in the second sentence of rule name " is written into the working memory 2, and this second sentence is erased.

Since all the conditional parts of the rule name ■ have been deleted, proceed to the conclusion part, and the operator "-" here "substitutes the right side to the left side"
Since this can also be executed, the value of the attribute name C of the fact name B, which is the value on the left side, is written as "fixed" in the working memory 2, and this conclusion part is erased. In other words, all rule names I have been deleted.

Since the value of attribute name C of fact name B is "fixed",
The fixed variable determination unit 6 passes the value of the attribute name C of the fact name B to the partial calculation mechanism 7, and the partial calculation mechanism 7 thereby extracts the rule name ■. Both the rule part and the conclusion part are deleted. In other words, the rule name ■ has also all been deleted.

In the conclusion part of rule name ■, the value of attribute name a of fact name C is "
The fixed/variable determination unit 6 passes the value of the attribute name a of the fact name C to the partial calculation mechanism 7, and the partial calculation mechanism 7 thereby extracts the rule name ■. In the rule name ■, the first sentence of the condition part is true, so it is deleted, but the second and third sentences are not deleted because it is unclear whether they are true or not. Therefore, the condition of the original rule name ■ is deleted. A new rule α consisting of the second and third sentences of the part and the conclusion part will be created and will be replaced with the rule name ■. Thus, the rules of rule base 3 are as shown in FIG.

As can be seen by comparing Figures 3 and 4, the rules have been converted to extremely simple rules. Therefore, it will be understood that even if inference is performed each time attribute name d of fact name C and attribute name d of fact name are given, inference will be performed quickly and efficiently without unnecessary duplication. .

3 Right: In addition to creating a new rule and then performing interpreted inference as described above, you can also obtain the same effect by creating a dedicated inference mechanism from the new rule using a rule compiler and performing inference using this. I can do it.

Next, FIGS. 7(a) and 7(b) show another embodiment of the present invention. In the above embodiment, the rules were reduced by partial calculation, but in this embodiment, the rules were not reduced. Create a dedicated inference mechanism using the rule compiler in C
3), a new dedicated inference mechanism is created by applying partial calculation to the dedicated inference mechanism (C5). Since a new dedicated inference mechanism is obtained that is reduced in size compared to the original dedicated inference mechanism, it is necessary to input it every time the attribute value of the "variable" fact is changed (C9), and when inference is performed, the same effect as above can be obtained. What is obtained will be easily understood.

Still another example is to store the update history of the attribute values in the working memory corresponding to each attribute of the fact, and to determine that the attribute values are fixed and the others to be variable. One example is one using a fixed/variable discriminating means. Items that have not been updated for a specified period of time are
This is because it can be assumed to be "fixed."

In addition, the above production system and its inference method can not only be used in expert systems, but also provide high-speed and highly efficient processing, so they can be used in expert database systems that handle huge amounts of data, and knowledge bases that perform advanced processing. It can also be suitably used in systems.

[Effect of the invention: According to the production system and inference method of this invention, the part of the original rule that applies to facts with fixed attribute values is deleted, and only the part that applies to facts with variable attribute values is deleted. Since it is left as a new rule,
Even if inference is performed each time a variable fact attribute value is input, there is no duplication of processing, and inference can be performed at high speed and with high efficiency.

Also, in production systems where a dedicated inference mechanism is created from the original rules using a rule compiler,
The part of the dedicated inference mechanism that applies to facts whose attribute values are fixed is deleted, and only the part that applies to facts whose attribute values are variable is left as the inference mechanism for Niizuma, so it is still fast and Inference can be performed with high efficiency.

[Brief explanation of the drawing]

Figure 1 is a block diagram of a production system according to an embodiment of the present invention, Figure 2 is a conceptual diagram showing the structure of facts in the production system shown in Figure 1, and Figure 3 is a rule for the production system shown in Figure 1. FIG. 4 is a conceptual diagram showing the new rules in the production system shown in FIG. 1; FIG. 5 is a flowchart showing the main parts of the operation of the fixed variable determination section in the production system shown in FIG. 1; Figure 6(a) (
b) is a main part flowchart showing the operation of the partial calculation mechanism in the production system shown in Fig. 1, and Fig. 7 (
a) and (b) are schematic flowcharts showing the operation of a production system according to another embodiment of the present invention, FIG. 8 is a block diagram of a conventional production system, and FIG.
A conceptual diagram showing the structure of facts in the production system shown in the figure, a conceptual diagram showing the structure of rules in the production system shown in FIG. (Explanation of symbols) 1...Production system 2...Working memory 3...Rule base 4...Inference mechanism 5...Interface 6...Fixed variable determination unit 7...Partial calculation mechanism.

Claims (1)

[Claims] 1. In a production system comprising a working memory consisting of a set of facts, a rule base consisting of a set of rules, and an inference mechanism that applies the rules based on the facts and makes inferences. , a fixed-variable discriminator for determining whether the attribute value of a fact in working memory is fixed or variable, and a rule-based rule based on the attribute value of the fact determined to be fixed by the fixed-variable discriminator. 1. A production system comprising: a partial calculation mechanism for deleting a part of a conditional part of a rule that becomes true by application and replacing the conditional part of the original rule with the remaining part. 2. A working memory consisting of a set of facts, a rule base consisting of a set of rules, and a dedicated inference mechanism for the rule base that makes inferences based on the facts, and rules that are created from the rule base by applying the rules. a compiler; fixed/variable determining means for determining whether attribute values of facts in working memory are fixed or variable; and attribute values of facts determined to be fixed by the fixed/variable determining means. A production system further comprising: a partial calculation mechanism that partially calculates the inference mechanism based on the above to create a new inference mechanism. 3. The working memory holds information as to whether the value is fixed or variable corresponding to each attribute of the fact, and the fixed/variable determining means refers to the fixed/variable information held in the working memory to determine whether the attribute value is fixed. 3. The production system according to claim 1 or 2, wherein the production system determines whether or not the variable is variable. 4. The production system according to any one of claims 1, 2, and 3, wherein the fixed variable information is given according to an operator's designation. 5. The working memory stores an update history of attribute values corresponding to each attribute of the fact, and the fixed variable determining means refers to the attribute value update history held in the working memory and updates the attribute values after a predetermined period. The production system according to any one of claims 1, 2, 3, and 4, wherein attribute values that are not specified are determined to be fixed, and other attribute values are determined to be variable. 6. Any one of claims 1, 2, 3, 4, and 5, wherein the partial calculation mechanism writes the attribute value of the fact determined in the condition part or the conclusion part as a fixed value in the working memory. One production system. 7. Determine whether the attribute value of the fact is immutable or variable, and as a result of the determination, refer to the immutable attribute value of the fact and delete the part of the conditional part of the rule that is true; A high-speed inference method characterized in that the condition part of the rule is replaced with the part that remains as a result of deletion. 8. Referring to the update history information of the attribute values of the fact, if the variable attribute value of the fact has not been updated for a predetermined period, the variable attribute value is determined to be immutable. 7 fast inference methods.