JPH02204835A - Rule type program executing method - Google Patents

Abstract

PURPOSE:To increase the response speed of an expert system, etc., by dividing the condition part of a rule into a limited condition which necessitates the retrieval of an element of a data assembly, and other fundamental condition, and reserving one set of data as for the limited condition. CONSTITUTION:A central processing unit 11, an input/output device 12, a memory device 13 and an arithmetic unit 14 are provided. In this state, the condition part of the rule is divided into the limited condition by which an element in a data assembly must be retrieved, and other fundamental condition, and as for the limited condition, one set of data for satisfying the condition is reserved, and as for the fundamental condition, the result of calculation is not reserved. Accordingly, all sets of data for satisfying the condition of each rule are not derived, and only one set is derived at the most. In such a manner, the execution speed is brought to high speed, and the response time to a user can be shortened.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] This method relates to a method for executing a rule-based program such as an expert system using a computer.

[Conventional technology]

In production systems that write and execute software such as expert systems in a rule-based format, the conventional execution method is ``Technology for accelerating production systems 2 Ishida and Kuwabara, Information Processing.

29, No. 5 (1988), pp. 467-477, there are RETE methods and the like.

The origin of production systems was to increase the amount of information in a static world that is not related to the flow of time, such as automatic proofs of theorems. A program in a production system is written as a set of rules that have a conditional part and an execution part.

When creating an expert system that uses a computer to carry out the thinking of an expert in a certain field, a production system is used because by creating a program using the expert's knowledge as rules, there is no need to be aware of algorithms as in the past. ing.

By the way, if we describe a dynamic world in which the logical value of a condition changes as a rule is executed, there is a possibility that the result will differ depending on which rule is executed when there are multiple rules for which the conditional part is true at a certain point in time. be. In many production systems, this is called conflict resolution and is given as a strategy for specifying the order in which rules are selected.

In the present invention, it does not matter which rule is executed based on the rules that hold true, and if any inconvenience occurs due to the way the rules are selected, we take the position that this is an error in the program.
Although the present invention also has a strategy for specifying the selection order of rules,
This is not used to compensate for program imperfections, but rather to shorten program execution time. For the above reasons, the present invention will not be referred to as a production system, but will be expressed as a rule-based formal description program.

[Problem to be solved by the invention]

The RETE method saves intermediate results and recalculates only the changes caused by rule execution, with the aim of improving execution performance compared to a method that recalculates all conditions. However, when the amount of change caused by the execution of a rule is large, the saved intermediate results become invalid, and additional processing is required to update the intermediate results.

Therefore, by not saving some of the results between TREATRETE expressions, the burden of updating intermediate results when the amount of change is large is reduced.

However, in order to find all sets of data that satisfy the conditional part of a rule, if the conditional part of another or the same rule is no longer satisfied when a certain rule is executed, it is necessary to search for those rules. All data sets that were previously saved will be deleted. For this reason, since the processing required to obtain these data sets in advance and the processing to delete those data sets are required, there is still a problem that storing intermediate results has a negative effect on improving execution performance.

The purpose of the method of the present invention is to reduce the rate at which stored information becomes invalid and improve the execution performance of programs written in rule-type formal descriptions by devising the calculation method for the condition part of rules and the format and management method of stored information. An object of the present invention is to provide a method for executing a rule-based program that increases the response speed of an expert system or the like.

[Means to solve the problem]

In order to achieve the above object, the present invention divides the condition part of a rule into a limiting condition that requires a search for an element of a data set and other basic conditions, and saves one set of data for the limiting condition. , avoiding recalculation every time after rule execution.

In addition, in the present invention, the logical value of the condition part of the rule is "true" r
In addition to ``false,''``unknown'' is added to delay the condition determination process until it is necessary.

[Effect]

By finding at most one set of data that satisfies the limiting conditions, execution performance is improved by reducing the number of searches for data sets.

Furthermore, by delaying the condition determination process until it is necessary, the total number of condition determination processes is reduced and execution performance is improved.

The impact of rule execution on the limiting conditions is due to the internal clock which increases by one unit due to the execution of the rule, or the clock attached to the data set which increases by one unit due to addition to or deletion from the data set.The final time of addition to or deletion from the data set. The last time when the former value was checked is stored, and the execution performance is improved because it is only necessary to check when the former value is larger.

Furthermore, if the effect of rule execution on limiting conditions can be analyzed in advance, this information can be used to speed up condition determination during execution.

〔Example〕

An embodiment of the present invention will be described below with reference to FIG. 1st
The figure is a flowchart showing the procedure for executing a rule set.

A rule set is a collection of rules having a condition part and an execution part. Executing a rule set means selecting a rule whose conditional part is true among the rules in the rule set, and repeating the process of executing that execution part until there are no more rules whose conditional part is true. If there are multiple rules, any of them may be executed, and the selection of the rule is left to the processing system. When a rule is executed, a rule whose conditional part was previously satisfied no longer holds true, or a rule whose conditional part was previously held becomes true.

In addition to logical expressions such as comparing variable values, rule conditions include logical expressions that require a search of a data set, such as the presence of data that satisfies a certain condition in a data set. Here, the former is called the basic condition, and the latter is called the limiting condition.

In this embodiment, basic conditions are calculated every time a rule is executed, but for limiting conditions, one set of special data such as examples or exceptions of the limiting conditions is saved.

Furthermore, by allowing the logical value of the limiting condition to be "unknown", the data set is not searched until it is needed. Since "true", "false", and "unknown" are added to the logical values of the limiting conditions to create three-valued logic, the condition part of the rule also becomes three-valued logic.6 The three-valued logic described above is fundamentally different from fuzzy logic. different.

1, which is either true or false in fuzzy.
In contrast to the binary logic of 0, which allows a value between O and 1 as the probability of being satisfied, the 3-value logic used in this method has a probability of 0 or 1 because the calculation has not yet been completed. This allows for an unknown state of not knowing what is going on.

FIG. 3 is a three-value logic calculation table used in one embodiment of the present invention. This is used when performing the three-value evaluation of the logical value of the rule condition part in step 3 of FIG.

(, a) is an operation table of logical product "and"", and A a
In the logical expression ndB, if either A or B is false, it is false; if both are true, it is true, and the others are unknown. (b) is the logical sum “or#(o
) is negated #n0tII, (d) is equivalent a=” (6
) is the negation of (d), and A≦B in (g) is [If A then BJ
The implication of (f) is that A + B is an operator for the negation of A2B.

In addition, if the value of the argument or subscript is unknown for a function that takes a logical value as an argument or for an array that uses a logical value as a subscript, the value of that function or array is assumed to be unknown.

In step 1 of FIG. 1, a limiting condition table is initialized.

The limiting condition table is an array with the limiting condition number as an index, and its elements are the three-value logical value of the limiting condition, examples of the limiting condition or special data that are exceptions, partial data sets for which the limiting condition has been checked, and the limiting condition. This is the last time you checked.

For each data set, a natural number called a frame number is given to the data it contains, and the data is managed in an array with the frame number as a subscript. The special data under the above-mentioned limiting conditions and the checked partial data set are indicated by this frame number.

The final time of the above-mentioned limiting condition check is set using an internal clock that increases by one unit when a rule is executed, or a clock attached to a data set that increases by one unit when additions to or deletions from the data set occur. Furthermore, the final time of data addition and the final time of data deletion are recorded in each data set.

In step 2, selection priorities are initialized for all rules.

In step 3, the logical values of the condition parts for all rules are evaluated using three values.

That is, for the limiting condition, the binary logical value of the limiting condition table is used without performing a condition search.

In step 4, it is determined whether there is a rule whose condition part is not "false", and if there is no rule, the execution of the rule set is terminated.

In step 5, the rule with the highest priority is selected from among the rules whose condition part is not "false".

In step 6, the logical value of the condition part of the selected rule is set to [
If "unknown," the necessary limiting conditions are calculated and the logical value is evaluated as a binary value of true or false.

It is determined whether the condition part of the rule selected in step 7 is "true". When 0 is selected, it is "unknown" and the condition part of the rule selected in step 7 is "true".
If the result of the calculation is "false", the selected rule is not executed and the process proceeds to step 3.

In step 8, the execution portion of the selected rule is executed.

Step 9 is adding data to the data set.

Investigate the impact on the limited conditions when deletions or changes are made,
Update the limited condition table. That is, the checked partial data set of the limiting condition, the special data of the limiting condition, and the three-value logical value of the limiting condition are updated.

However, at this time, no data search is performed to calculate the limiting conditions.

In the execution of rules, changing data is considered to be a continuous process of deleting and adding data.

In step 9, when considering changes in the limiting condition, whether or not there has been an addition to or deletion from the data set depends on the magnitude relationship between the last data addition time, the last data deletion time, and the last time the limiting condition was checked. .

In step 10, the rule selection priority is updated for all rules.

After step 10, the process advances to step 3, and the above process is repeated.

FIG. 2 shows the general configuration of a computer for realizing the program execution method of the present invention described above, in which the central processing unit 11. Input/output device 12° memory bag W113. It includes an arithmetic unit 14 and the like.

〔Effect of the invention〕

The present invention does not require all sets of data that satisfy the conditions of each rule, but only one set at most, and also uses three-valued logic for the rule conditions and delays all condition judgments until they are needed. The execution speed has been increased by more than 10 times.

Therefore, by using the present invention in executing a program such as an expert system, the response time to the user can be shortened.

In addition, the rule-based program execution method of the present invention takes the position that if there are multiple rules that hold true in the condition part at a certain point, it does not matter which one is selected.
Since programming that assumes a specific conflict resolution strategy is not allowed, rule-based programs are highly versatile and maintainable.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing the procedure for executing a rule set according to an embodiment of the present invention. FIG. 2 is a block diagram illustrating a computer on which the present invention is implemented.

Claims (1)

[Claims] 1. A method for executing a rule-based program in which a rule having a conditional part and an execution part is written, and calculation is performed by executing the execution part of the rule for which the conditional part is satisfied, wherein the conditional part of the rule is is divided into a limiting condition that requires searching for an element in a data set and other basic conditions, and a set of data that satisfies the condition is saved for the limiting condition, and a calculation result is saved for the basic condition. A rule-based program execution method characterized by not saving. 2. In a rule-based program execution method in which a rule with a conditional part and an execution part is written and calculations are performed by executing the execution part of the rule where the conditional part is satisfied, the logical value of the conditional part of the rule is set to "true". , using three-valued logic that includes the value "unknown" due to the condition not being determined in addition to "false", and if the logical value of the selected rule is "unknown", it is clarified as "true" or "false". , a rule-based program execution method characterized in that the rule is executed when the rule is "true". 3. In a rule-based program execution method in which a rule with a conditional part and an execution part is written and calculations are performed by executing the execution part of the rule where the conditional part is satisfied, the time required to judge the condition in the conditional part of the rule. A method for executing a rule-based program, characterized in that calculation results are saved for conditions for which the period is long and the rate of change due to rule execution is small. 4. In a rule-based program execution method in which a rule with a conditional part and an execution part is written and calculations are performed by executing the execution part of the rule where the conditional part is satisfied, the time required to judge the condition in the conditional part of the rule. A rule-based program execution method characterized by setting the logical value of a long condition to "unknown" and calculating the condition to clarify whether it is "true" or "false" after the condition is needed. 5. A rule-based program execution method, characterized in that the logical value of the limiting condition as defined in claim 1 is "unknown" and a logical value is determined in the condition part of the rule using the three-value logic as defined in claim 2. 6. It has an internal clock that increases by one unit when a rule is executed, and it records the time when data is added, deleted, changed, etc. for each data set, the logical value to be saved for each limiting condition, and the time when a set of data is obtained. 6. The rule-based program execution method according to claim 5, wherein it is determined whether the saved information of the limiting condition is valid by storing the limiting condition and checking the magnitude relationship. 7. Has an internal clock that increases by one unit when data is added, deleted, or changed for each data set, records the last data addition time, and stores a logical value for each limiting condition;
6. The rule-based program according to claim 5, wherein the rule-based program stores the time of the corresponding data set at the time when the data set is obtained, and determines whether the storage information of the limiting condition is valid by checking the size relationship. Execution method. 8. The rule according to claim 1 or claim 2, characterized in that the influence of execution of the rule on the conditions of each rule in the rule set is calculated in advance, and this information is used to determine conditions at the time of execution. How to run a type program. 9. An expert system in which knowledge is described in a rule-based format and whose purpose is control and acquisition of knowledge by executing rules, using the rule-based program execution method described in each claim above.