JPS6347839A - Inference control system - Google Patents

Abstract

PURPOSE:To improve the efficiency of an inference processing, by using a knowledge module in a low-order level first, shifting a processing to the processing by the knowledge module of order higher by one level when no solution is obtained, and returning the processing to a low level after obtaining the solution when it is obtained. CONSTITUTION:An interruption processing mechanism 23 selects the knowledge module 14 from the lowest order level based on an initial condition held in a working memory 22, and an inference engine 21 executes an inference processing, and when no conclusion is obtained, an interruption request is issued to the interruption processing mechanism 23. The interruption processing mechanism 23 selects the knowledge module 15, and the inference engine 21 executes the inference processing. Thus, when the inference processing is completed at a certain level, the inference engine 21 calls the interruption processing mechanism 23, and the interruption processing mechanism 23 returns the inference engine 21 to the level at time of interruption based on a bit of holding return information, and starts up the inference processing again. In this way, it is possible to generate the hierarchical constitution of the knowledge module, and to improve the efficiency of execution by expanding a rule retrieving range in order, to a high order level only when the no conclusion is obtained at the low order level.

Description

[Detailed Description of the Invention] [Summary] This is an improved control method for inference processing by a knowledge module in a computer.

The knowledge modules are grouped into a plurality of levels and used starting from the lower level, and if a solution to inference processing cannot be obtained, an interrupt request is generated and processing is started by one knowledge module. In this way, if there is a solution at a higher level, the result is returned to the lower level.

With this method, it becomes easy to effectively limit the scanning target in inference processing and improve the efficiency of the processing by layering the knowledge modules.

[Industrial application field]

The present invention relates to an inference control method for a computer system that processes knowledge information.

Systems that proceed with inference by sequentially scanning rules constructed from knowledge are called expert systems, and are being applied to various fields.

[Conventional technology]

FIG. 3 is a block diagram showing an example of the configuration of the inference system.

For example, when proceeding with inference processing from a given initial condition to reach a conclusion, the inference engine 1 of the inference device 10 receives an inference request including the initial condition from the outside, and sets the condition in the working memory 2. do.

The knowledge base 3 stores knowledge modules 4 necessary for inference, and each knowledge module 4 includes an activation condition section 5 and a rule section 6.

The activation condition section 5 indicates conditions to be satisfied when selecting the knowledge module 4, and the rule section 6 consists of one or more rules.

Each rule consists of a conditional part and an execution part. For example, according to the format of the IF statement and THEN statement, which are well-known in the field of programming languages, the conditional part is 1F condition, ANII condition 2^ND ---The execution part is THEN execution procedure.

It's an execution procedure. It is expressed as follows.

The inference engine 1 selects one of the knowledge modules 4 by comparing the activation condition part 5 of the knowledge module 4 of the knowledge base 3 with the conditions held in the working memory 2.

When the rules of the knowledge module 4 are scanned and one of the rules having a condition part that is satisfied by the condition is selected, the process specified by the execution part is executed.

As a result, the conditions that are stored in working memory 2 generally change,
The inference engine 1 searches for rules again using new conditions in the working memory 2 obtained as an intermediate inference result.

As a result of proceeding with the inference process in this manner, the rule execution unit indicates the end, and at that time, for example, data indicating the conclusion is obtained in the working memory 2.

In other cases, for example, because there are no rules that are satisfied by the conditions held in the working memory 2, the inference process does not proceed, and the conclusion cannot be drawn depending on the knowledge stored as rules in the knowledge module 4. It is detected that the result is not obtained.

[Problem that the invention seeks to solve]

If a conclusion cannot be reached as described above, it is necessary to add more rules, and by adding more rules, the knowledge base will be enriched, but at the same time, the search target will become larger and the efficiency of inference processing will deteriorate. . Furthermore, addition/modification of rules causes problems such as difficulty in maintaining the knowledge module.

[Means for solving problems]

FIG. 1 is a block diagram showing the configuration of the present invention.

The figure shows the configuration of the inference mechanism, 20 is an inference device, 21 is an inference engine, 22 is a working memory, 23 is an interrupt processing mechanism, 1
3 is a knowledge base, and 14 to 16 are knowledge modules constituting the knowledge base 13.

[For production]

The knowledge modules making up the knowledge base 13 are grouped into multiple levels, such as knowledge bases 14, 15, and 16, for example.

Upon receiving the inference request, the inference engine 21 requests the interrupt processing mechanism 23 to perform processing.

The interrupt processing mechanism 23 sets the initial lowest level, and
Depending on the initial conditions held in the working memory 22, knowledge modules 14 from the lowest level are selected and specified to the inference engine 21.

The inference engine 21 uses the specified knowledge module to
Inference processing is executed in the conventional manner, and if a conclusion is obtained, the processing ends; however, if a conclusion is not obtained, an interrupt request is issued to the interrupt processing mechanism 23.

The interrupt processing mechanism 23 saves the information necessary to restore the current level and the knowledge module, etc., sets the level on the rubel, and selects, for example, the knowledge module 15 from that level in the same manner as described above. Specify it to the inference engine 21.

The inference engine 21 executes inference processing using the designated knowledge module in the same manner as described above.

In this way, when the inference process ends at a certain level,
The inference engine calls the interrupt processing mechanism 23, and the interrupt processing mechanism 23 uses the return information held to return the inference engine 2I to the level at the time of the interrupt, and restarts inference processing.

The above method makes it possible to have knowledge modules in a hierarchical structure, and when performing inference processing, the rule search range is expanded to higher levels only when a conclusion cannot be reached at lower levels. , execution efficiency can be improved.

〔Example〕

FIG. 2 is a block diagram showing the configuration of one embodiment of the interrupt processing mechanism 23. As shown in FIG.

The inference engine 21 generates an interrupt request at the start of inference processing and when it detects that a conclusion cannot be obtained at a certain level of inference processing.

When the interrupt reception unit 30 receives an interrupt request from the inference engine 21, it sets the current level value + 1 or the lowest level value in the first case in the level memory 31, and stores the return information stack 3.
? After stacking information necessary for restoration, such as the identification name of the knowledge module currently being processed, the knowledge module selection unit 33 is activated.

The knowledge module selection unit 33 refers to the level values held in the level memory 31 and selects the knowledge module search list 34.
, the knowledge modules belonging to the corresponding level are known, the activation condition sections of these knowledge modules are examined, and a knowledge module that can be satisfied by the condition information held in the working memory 22 is selected.

The activation control unit 35 passes the identification name of the selected knowledge module to the inference engine 21 and activates inference processing.

When the inference engine 21 obtains the conclusion of the inference process at a certain level and finishes it, it generates a return process request.

The interrupt processing mechanism 2 receives a return processing request from the inference engine 21.
3, the current level value minus 1 is set in the level memory 31. The return control unit 37 sends the last level of information held in the return information stack 32 to the inference engine 21 to restart the inference process of the previous level.

By using the above method, it is possible to configure a knowledge base in such a way that knowledge modules are hierarchically divided into multiple levels, and the rule search range is sequentially expanded to higher levels only when a conclusion cannot be reached at a lower level when performing inference processing. Therefore, the processing efficiency of inference processing can be improved.

〔Effect of the invention〕

As is clear from the above description, the present invention has a significant industrial effect in that it is possible to hierarchize the knowledge base of a computer system and improve inference processing efficiency.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of the present invention, FIG. 2 is a block diagram of a configuration example of an interrupt processing mechanism, and FIG. 3 is a block diagram of a conventional configuration example. In the figure, 1.21 is the inference engine, 2.22 is the working memory, and 3.1
3 is the knowledge base, 4.14.15.16 is the knowledge module, 5 is the activation condition part, 6 is the rule part, 10.20 is the inference device,
23 is an interrupt processing mechanism, 30 is an interrupt reception unit, 31 is a level memory, 32 is a return information stack, 33 is a knowledge module selection unit, 34 is a knowledge module search list, 35 is a startup control unit, 36 is a return reception unit , 37 indicates a return control section.

Claims (1)

[Claims] A knowledge module consisting of rules each having a condition part and an execution part is selected, and a rule having a condition part that matches a predetermined input condition is scanned from the knowledge module and designated as the execution part of the rule. In the inference processing device (20), the knowledge modules are grouped into a plurality of levels, and by the inference processing by the knowledge modules belonging to the lower level, means (21) for generating a predetermined interrupt request when it is detected that a solution cannot be obtained; means (23) for activating inference processing by the selected knowledge module; and, when a solution is obtained by the inference processing by the knowledge module belonging to the higher level, returning to the inference processing at the source of the interrupt request to the level; An inference control method comprising means (23).