JPH0210432A - Inference system having contention resolving function for expert system - Google Patents

Abstract

PURPOSE:To simplify the action of inference and to easily express rules by resolving contention among rules to select a rule at the time of existence of plural successful rules and deciding whether contention resolution in the selected rule is designated or not and resolving contention at the time of deciding that it is designated. CONSTITUTION:Working memories 41-4m are initialized to start a rule matching processing system. After the matching processing, the number of successful rules is decided and the processing is terminated when it is zero. When there are plural successful rules, contention among rules is resolved and a rule is extracted by contention resolution. The extracted rule is transferred to a judgement processing part which decides whether contention resolution in the rule is designated or not. When it is designated, the intra-rule contention resolution processing is performed to perform the execution processing of the extracted rule.

Description

DETAILED DESCRIPTION OF THE INVENTION A. Field of Industrial Application This invention relates to an inference method having a conflict resolution function in an expert system.

B1 Overview of the Invention This invention uses an inference method with a conflict resolution function to select a rule by conflict resolution between δ rules when there are multiple established rules, and check whether there is a conflict resolution specification in the selected rule. By resolving conflicts when making a judgment, it is possible to perform horizontal search inference only on the necessary parts of the rule, making it easier to understand the inference movement and making rule expression easier. It was made so that it would become so.

C1 Conventional technology Figure 2 is a schematic configuration diagram of an expert system.
2 is an inference system. The inference system 2 includes an inference engine 3 and a plurality of working memories 41 to 4o. The operation of this expert system will now be briefly described. First, the inference engine 3 searches for a rule that holds the IF part.

The establishment of the IF section means that working memory 4, ~
This is whether there is a rule that has the finding written in 4□ in the condition part (TIIEN part). T I of the next existing rule
I Execute the EN section. In this case, the contents predicted in the T I-(E N section are added to the working memories 41 to 4n. In the state of the additional working memories 41 to 4n, the inference engine 3 searches again for a rule for which the IF section is established. The same applies thereafter. Repeat this process until there are no rules that hold true.

In expert system processing, the rules that hold true in a certain state of working memory are not necessarily limited to rules such as -. Therefore, when there are multiple applicable rules, it is necessary to select one of them by some means. This means is usually called a strategy.

Production systems used in expert systems generally repeat cycles of matching, conflict resolution, and execution. In other words, one rule is selected from among the rules established by conflict resolution according to a specified strategy, and its execution part is executed. By executing the rule, a change occurs in the state of the working memory (changes from state i to state FJ i ), and in the next cycle, rules that hold in state i + t are extracted (in the matching process). , perform conflict resolution again. This is a method of boat fishing, and there is only one rule to be executed in a certain state, which is the speed of sound. Also,
Rules that hold true at a certain point in time (state @) may be more efficient if executed simultaneously, and the inference movement itself may be easier to understand. This depends on the nature of the problem you are trying to solve, but if such a mechanism is realized,
Expressing rules becomes easier.

Next, describe a specific example.

(1) ``Perform process X for all those who meet a certain condition.'' Next, ``...''. For example, "My grade is 8.
If the score is 0 or less, the wage rate will be 1.5. Examples of rules and working memory in that case are shown below.

○Rule j″xyz” l r (A) (?X (grades > 80))) ○Rule hen (modiry (?
I "UVW" if (?
5)...t -1-1)(c (grade 90)...L+
2) (A t+3) In the above, a, b, c... are famous songs, and L is ta.
gN O. For example, (grade 85) indicates the person's attributes and their values, and in the above example, if an expression frame is used in a list, it becomes a slot and its value.

A represents any pattern. However, in working memory (A
It is assumed that the pattern t+3) is asserted by the execution of the rule established in state j-1. Here, for simplicity, it is assumed that only rule j is established.

But at this point? There are two candidates for X, namely t
Two of agNo, t, and t+3 are established at the same time.

01 Invention or problem to be solved 4'' Conventional conflict resolution strategy, for example, if only one rule is executed with max-tag, with rule-j?X=C
is executed, and the state of working memory is as follows.

State i + 1 (a (grade 85)...L) (b (grade 65)...t -+-1) (A
t −+−3) (c (grade 90) (pay rate 1.5) = t +4) In the next cycle, use rule -j? X=a and rule -j-1-
At 1? Two rules are established: X=C.

The rule extracted from this state is rule -j+1, and its execution changes the state to state j+2. The movement is a vertical search.

The characteristics of max-tag are Event Driven (
It is convenient for problems in which inference is proceeded with tllll light), but as in this example, when you want to go to state 1 (at -1, the following state) by horizontal search inference. I couldn't control it.

State i + 1 (b (grade 65)...t + 1) (A t + 3) (c (grade 90) (wage rate 1.5) t + 4) (21 (
Grade 85) (Wage rate 1.5)...t+5)" - In the case of "-", set LagNO of C to t+4. L+ tagN O of a
I set it to 5, but the opposite is also possible.

-As mentioned in L, although it is originally intended to be mastered by horizontal search, inference proceeds by vertical search, and the inference movement differs from the movement that is easy for people to think of, making it extremely difficult to understand.

(2) Furthermore, it has been extremely difficult to implement such an example with conventional production systems.

``Among those that do not meet a certain condition,'' perform a certain process on some of the third-ranked (or lower-ranked) items. For example, among the employees who ranked 2nd with a score of 80 points or higher, the pay rate for the 5 third-ranked employees was set at 1.
This is a case of 1 to 5, etc.

However, if the number of applicable people is less than 5 (for example: 3 people), two-legged processing is performed for up to 3 people.

The purpose of this invention is to make it possible to specify conflict resolution strategies within rules, to advance horizontal search inference, to make inference movements easier to understand, to facilitate rule expression, and to enable fine-grained control of inference. An inference method with a conflict resolution function is provided.

E. Means for Solving the Problems This invention uses an Exverter system that infers and controls rules stored in a knowledge base using an inference system.If there are multiple established rules, a rule is selected by resolving conflicts between each rule. After that, it is determined whether there is a specification for conflict resolution in the selected rule, and if there is, the conflict is resolved before the rule is executed.

F1 effect As mentioned above, a rule is selected and it is determined whether conflict resolution is specified in the selected rule, so if all the rules that meet a certain condition are executed at the same time, the top ( or subordinate) at the same time.

G. Example An embodiment of the present invention will be described below based on the drawings.

In FIG. 1, first, the working memory is initialized in step S1, and the process proceeds to the rule matching processing system in step S. When the matching process is completed, the process proceeds to step S3, where it is determined how many rules have been established, and if it is zero, the process ends. If there is a plurality of rules established in step S3, conflict resolution between rules is performed in step S4.
Rules are extracted by conflict resolution. The extracted rule proceeds to step S, a processing unit that determines whether or not intra-rule conflict resolution is specified. By making it possible to specify this intra-rule conflict resolution for each rule, we can achieve the following (a) and (b)
).

(2) to execute all the rules that meet a certain condition at the same time; (b) to execute several higher (or lower) rules that satisfy a certain condition at the same time. There are various variations in where to specify each rule. For example, in the case of a rule with the following syntax, (rule name, comment, jf condition part then execution part) (
1) Write after the rule name (2) Write after comment (3) Write after if (4) Write after then.

The above conflict resolution strategy is referred to as intra-rule conflict resolution.

In the case of (1) above, the specification method is, for example, a symbol such as act-all or a list such as (act-all).

An example of this is given below.

Rule-j (act-all) “X Y Z”1
f(A) (?Z (grade > 80)) then (modify (? , (Bt-
minnn variable name).

'' double series means that the values of the specified variable names are sorted in ascending (or descending) order, and only the specified n values are executed. An example of this is given below.

Rule-j (act-max-n5?Y) ”XY
Z”1f(A) (? Among them, for the top five, th
This means that the process specified in the en part is performed.

In addition, if you want to extract (select) only one rule, use (
In the case of 3), for example, (max-tag), or when following a conflict resolution strategy between rules, (4)
), specify it in a format like ().

As a method of specifying internal conflict resolution in step S described above, (act -all), (act -max
-n -), (act-min-n -), (max-
tag), (max-Cf), etc. If it is determined in step S5 that there is a designation, the process proceeds to step S8, in which an intra-rule conflict resolution process is performed, and in step S7, the extracted rule is executed.

Next, we will discuss each designation case.

(a) In the case of act-all, all instantiations may be extracted, even if they are true, and the execution order may be determined by the inference engine.

For example, the order of instantiation max-tag, max-cr It can also be specified by the user.

For example (act-all max-tag) (act
-all max-c[) (b) (act-max -n n variables) (a
ct-min -n n variables) Sort the variables from the largest (smallest) and extract the L-ranked n variables.

(c) (max-tag) (max-cr) Extract one item with the largest Lag (or cf'). Or (max-tagn), (max-
crn ), it is also possible to extract n pieces in descending order of tag (or (cf)).

” In addition to the two-layer example, there are also the following intra-rule conflict resolution strategies.

(1) (act-max -n n variable) Here, the value assigned to the variable may be only a numerical value, or there may be cases where it is desired to process the values in alphabetical order, international order, etc. in addition to numerical values. In that case, be able to determine the data type and process it.

(act-max-n n variable processing method) The processing method can be, for example, Li5p's S formula or a function name, and the conflict set (certainty) and the value of the variable can be used as arguments.

(2) If the value of the variable is a lift symbol or string, n values can be extracted in order of length. In this case as well, you can specify the processing method.

This embodiment can be used not only for production systems but also for hybrid inference systems that are organically combined with frames.

II. Effects of the Invention As described above, according to the present invention, since it is determined whether a conflict resolution strategy is specified in a rule, it is possible to perform horizontal search inference only on the necessary parts.

This makes it easier to understand the movement of the inference, facilitates rule expression, and enables fine-grained control of the inference.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing an embodiment of the present invention, and FIG. 2 is a schematic diagram of the configuration of an exvert nostem. 1...Knowledge Heath,...Inference System, 3. Enojin, 4. ~4o...working memory. Reasoning Figure 1 2 Reasoning system 4, ~4m...working memory

Claims (1)

[Claims] (1) In an expert system that infers and controls rules stored in a knowledge base using an inference system, if there are multiple established rules, a rule is selected by resolving conflicts between each rule, and then conflicts within the selected rules are selected. An inference method having a conflict resolution function in an expert system, characterized in that it determines whether there is a specification for resolution, and if there is a conflict, the conflict is resolved and then the rule is executed.