JPH07129402A - Expert system inference control method - Google Patents

Abstract

(57) [Summary] [Purpose] It is rich in expressiveness and enables fine and rapid activation control of rules. [Structure] In the main program, select rules (S
1) is performed, and it is determined whether or not there is a selected rule (S
2) As a result, if there is a rule, that rule is executed (S
3) If there is no rule, the process waits for an event (step S5) and then returns before selecting a rule (S1). In the execution of the rule (S3), processing of an event procedure that performs state evaluation based on a predetermined logical condition for an element selected from a combination of events necessary for activation, processing of occurrence of an event that performs activation control, and event The processing of other procedures, including the display of logical values, takes place.
After the execution of the rule (S3), it is determined whether or not it is finished (S3).
4) Then, as a result, if it is finished, the operation is finished, but if it is not finished, it returns before the rule selection (S1).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inference chain control system in an inference control unit which is a part of an expert system construction support tool.

[0002]

2. Description of the Related Art Conventionally, in an expert system of the type that makes inferences using ordinary rules, all rules have a condition part and an execution part. The inference control unit, which is a part of the expert system construction support tool, refers to the condition part of each rule as data and executes one or all execution units of the rules that satisfy the condition. By repeating it, the reasoning is advanced. At this time, if there is a change in the data performed by the execution unit of the specific rule, it becomes a factor that determines the chain from the specific rule to another rule. Therefore, the execution of rules that do not change data does not cause a chain of inference.

FIG. 4 shows an example of an inference chain control method in a conventional inference control section. In the case of this inference chain control method, first, in rule 1, the condition part 1a collates the data a generated in the execution part in the preceding rule, and if the condition is "true", the execution part 1b changes the data a. To generate data b. Next, in rule 2, condition part 2
If "a" collates the data "b" and the condition is "true", the execution part 2b changes the data "b" to generate the data "c" which contributes to the condition part in the rule at the subsequent stage. That is, in this inference chain control method, the execution of the rule appears as a change in data, and as described above, a chain of inference is caused when the data is changed.

FIG. 5 shows another example of the inference chain control method in the conventional inference control section. In the case of this inference chain control method, the rule has an event logical expression triggered by an event generated as an arbitrary true / false value by the condition section in the preceding rule in addition to the condition section and the execution section. First, in rule 1, the event α generated in the condition part in the preceding rule is evaluated based on the event logical expression 1α, and if the evaluation value is “true”, then the condition part 1β is subsequently evaluated based on the evaluation value. Generate event β.
Next, in rule 2, the event β is evaluated based on the event logical expression 2α, and if the evaluation value is “true”, then the conditional unit 2β is changed to the event logical expression in the subsequent rule based on the evaluation value. Generate a contributing event γ. The execution units 1γ and 2γ are provided as needed, and when they are provided, the condition units 1β and 2β are provided.
The predetermined processing is executed with reference to the result of the true / false evaluation of.

That is, in this inference chain control method, the evaluation of the condition part appears as the occurrence of an event, and when the events are chained, the chain of inference is caused. By the way,
The occurrence of the event is automatically performed by the inference control unit.

[0006]

However, in the case of the inference chain control method described with reference to FIG. 4, since the chain of inference does not occur unless the data is changed, the knowledge representation is unnatural in a medical examination type expert system, for example. turn into. In addition, in the chain of inference that generally requires data change, every time one rule is executed, all rules are re-referenced, so that a sufficient inference speed cannot be obtained.

On the other hand, in the case of the inference chain control method described with reference to FIG. 5, there is an advantage that the problem of the inference chain control method described above can be solved, but on the other hand, the event generated from the conditional part of the rule is a true / false logic. Since it is a value, there is a drawback that it lacks expressiveness. Further, the event logical expression corresponds only to the logical operation, and the activation control of the rule (by the way, "activation" means that the condition part is waiting for evaluation) becomes simple. Further, the event logic expression cannot obtain an evaluation value unless all the event values to be referenced are determined, so that there is a problem that the activation timing cannot be finely controlled.

The present invention has been made to solve the above problems, and its technical problem is to provide an inference control system of an expert system which is rich in expressive power and which can finely and quickly activate and control rules. Especially.

[0009]

According to the present invention, a combination of a plurality of events required for activation is determined, an arbitrary element is selected from the combination, and the combination is determined based on a predetermined logical condition. After the state evaluation of the selected element, an event logical value is generated according to the result of the state evaluation, and knowledge data having a plurality of mutually different rules each including an execution unit for performing a procedure for displaying the event logical value is obtained. Based on this, an inference control method for expert systems that links inferences can be obtained. In this inference control method of the expert system, arbitrary data can be used for a plurality of events.

Further, according to the present invention, in the inference control method of the expert system, the execution unit evaluates the necessary event designation in the rule associated with the event logical value as the event logical value occurs. And the inference control method of the expert system in which the execution unit evaluates the activation condition of the rule according to the evaluation result of the event designation.

In addition, according to the present invention, in the inference control method of the expert system, the execution unit can generate an event other than the evaluation of the activation condition of the rule during the inference execution, and the inference control method of the expert system, The execution unit is an expert system inference control method that includes the rule in a selectable set according to the evaluation result of the activation condition, or the execution unit sets the evaluation value of the rule to the selectable set of rules. An inference control method of an expert system that is automatically generated as an event logic value can be obtained.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The following is a detailed description of the inference control system of the expert system according to the present invention with reference to the drawings.

First, the outline of the inference control system of the present invention will be briefly described. This inference control method links inferences based on knowledge data having a plurality of different rules. Each rule defines a combination of a plurality of events necessary for activation in advance, and an arbitrary combination is selected from the combinations. An execution unit that selects an element, evaluates the state of the selected element based on a predetermined logical condition, generates an event logical value according to the result of the state evaluation, and executes a procedure for displaying the event logical value. Contains. Here, arbitrary data can be used for a plurality of events.

That is, the rule firstly combines the events required for activation. For example, if the required events are a, b, c, and d, then the combination can be expressed as [aAND (bORc) ANDd]. This condition means that there are a and d, and that there is b or c.

Second, the rules require a procedure to test the state of the event needed for activation. This logical condition is to evaluate the state by designating an element necessary for activation from the combination, and for example, the element is [a + BOU
ND (b)? It can be expressed as b: c] ≧ 0. In this case, a
It means that the value obtained by adding the value of b (or c if there is no b) is zero or more.

Third, the rule requires a procedure to obtain the event logic value that would occur if it were adopted by itself. This condition can be expressed as RETURN [d> 100], for example. In this case, a logical value of true is generated when d is larger than 100, and false when 100 is less than 100.

Fourth, the rule requires some other procedure to display the previous event logic value. This condition is, for example, P
RINT: f [d =% d, d] can be expressed. In this case, it means to indicate the value of d.

FIG. 1 is a simplified flowchart showing a main program for performing inference control based on the knowledge data constructed as described above.

In this control main program, when the operation is started, a rule is selected (step S1). Next, it is determined whether there is a selected rule (step S
2) As a result, if there is a rule, the rule is executed (step S3), but if there is no rule, the process waits for an event (step S5) and then returns before selecting a rule (step S1). In addition, after executing the rule (step S3), it is determined whether or not to end (step S4).
Then, as a result, if it is finished, the operation is finished, but if it is not finished, it returns before the selection of the rule (step S1).

FIG. 2 is a flow chart showing a subprogram in which the rule execution part included in the control main program shown in FIG. 1 is subdivided. In the sub-program in this rule execution (step S3), first, an event procedure process (step S31) including state evaluation based on a predetermined logical condition for an element selected from a combination of events required for activation is performed, and then, After performing the event generation process including the activation control (step S32), other procedures including the display of the event logical value (step S33) are performed. After that, the process returns before selecting the rule (step S1).

FIG. 3 is a flowchart showing the rule activation subprogram relating to the event generation processing (step S32) shown in FIG.

This event generation process (step S32)
In the sub-program, first, it is determined whether or not there is a rule whose state is evaluated (step S321). As a result, if there is no rule, the process returns before the selection of the rule in the control main program (step S1). If there is, the event is set (step S322), and then the required event is designated and evaluated (step S323).
To do. After that, it is checked whether the evaluation value is true or false (step S324), and if false, the process returns before the determination of whether there is a rule (step S321), but if true, the activation condition continues. Is evaluated (step S325), and then whether the evaluation value is true or false is checked (step S3).
26) As a result, if the result is false, the process returns before determining whether or not there is a rule (step S321), but if the result is true, the rule is subsequently activated (step S32).
After 7), it is determined whether there is a rule (step S3
Return before 21). In the rule activation (step S327), the corresponding rule is included in the selectable set according to the evaluation result of the activation condition.

That is, in the case of this inference control method, the execution unit evaluates the necessary event designation in the rule related to the event logical value with the occurrence of the event logical value, and according to the evaluation result of the event designation. The activation condition of the applicable rule can be evaluated. In addition, the execution unit includes the corresponding rule in the selectable set according to the evaluation result of the activation condition, and automatically evaluates the rule evaluation value as the event logical value for the rule of the selectable set. Can occur. In such an inference control method, it is easy for the execution unit to cause an event other than the evaluation of the rule activation condition during the inference execution.

By chaining inferences based on knowledge data by such an inference control method, rules that refer to an event that has occurred are repeatedly processed. Therefore, unless the events necessary for a particular rule are available, the activation condition is not satisfied. Will not be tested, and a rule that does not satisfy the activation condition will not be selected, and control for the desired rule will be performed quickly.

[0025]

As described above, according to the inference control method of the expert system of the present invention, the activation condition is not tested unless the events required for a specific rule are prepared, and the activation condition is not satisfied. Since a desired rule is activated and finely controlled without activating the rule as a selection target, it is possible to handle from simple processing to complicated advanced processing. In particular, since unnecessary rules are filtered out and only relevant rules are executed, the time required for the processing can be extremely short. In addition, any data can be used for the event, and both the event that is automatically generated according to the rule evaluation and the event that other users can freely generate can be handled. It is possible to support any description format of an automatic inference chain based on the above, or an explicit inference invocation chain by writing data or the like.

[Brief description of drawings]

FIG. 1 is a simplified flowchart showing a control main program related to an inference control method of an expert system of the present invention.

FIG. 2 is a flow chart showing a subprogram of a rule execution part included in the control main program shown in FIG.

FIG. 3 is a flowchart showing a rule activation subprogram relating to the event occurrence shown in FIG.

FIG. 4 is an explanatory diagram showing an example of an inference chain control method in a conventional inference control unit.

FIG. 5 is an explanatory diagram showing another example of the inference chain control method in the conventional inference control unit.

[Explanation of symbols]

 1a, 2a, 1β, 2β ... Conditional part 1b, 2b, 1γ, 2γ ... Execution part 1α, 2α ... Event logical expression a, b, c ... Data α, β, γ ... Event

Claims (7)

[Claims] 1. A combination of a plurality of events required for activation is defined, an arbitrary element is selected from the combination, and the selected element is state-evaluated based on a predetermined logical condition. An expert characterized by generating an event logical value in accordance with the result of the above and chaining inferences based on knowledge data having a plurality of mutually different rules each including an execution unit for performing a procedure for displaying the event logical value. Inference control method of system. 2. The inference control system for an expert system according to claim 1, wherein arbitrary data is used for the plurality of events. 3. The inference control method for an expert system according to claim 1, wherein the execution unit evaluates a necessary event designation in a rule associated with the event logical value as the event logical value occurs. An inference control method for expert systems characterized by the following. 4. The inference control method of the expert system according to claim 3, wherein the execution unit evaluates the activation condition of the rule according to the evaluation result of the event designation. method. 5. The inference control method for an expert system according to claim 4, wherein the execution unit can generate an event other than the evaluation of the activation condition of the rule during the inference execution. control method. 6. The expert system inference control method according to claim 4 or 5, wherein the execution unit includes the rule in a set that can be selectively executed according to an evaluation result of the activation condition. Inference control method of system. 7. The inference control method for an expert system according to claim 6, wherein the execution section automatically generates an evaluation value of the rule for the rule of the selectively executable set as the event logical value. An inference control method for expert systems characterized by the following.