JPH02100141A - Expert system - Google Patents

Abstract

PURPOSE:To realize an appropriate inference by holding a counted value obtained by counting the number of times of execution at every production rule, at the time of executing an inference, and executing most preferentially the rule corresponding to the largest counted value. CONSTITUTION:An operator operates a mouse 13 and designates a rule set 'trust' 40 as a top node from in a classification 30 displayed on a CRT 17. In this case, when a condition part of its started rule set is formed, a CPU (central processor unit) 19 judges whether a subordinate rule set which is executed in the next time is indicated clearly in the rule concerned or not. As a result, when the rule set is not indicated clearly, the execution is started from the rule set in which the number of times of execution in the past is the most, that is, the rule set in which a counted value of a reference count is the largest. In such a manner, an inference by an experience in the past and giving the priority between production rules can be executed simply, and an appropriate inference can be executed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to expert systems, and in particular to production rule-based systems.

[Conventional technology]

Conventionally, expert systems include production systems as one type of knowledge representation. This production system is a problem-solving system that makes inferences by executing production rules, which are pairs of premises and conclusions.

In an expert system that performs inference using such a production system, as a method for resolving conflicts between production rules during the inference process,
There are methods that resolve conflicts by comparing the number of condition elements (rules), and methods that resolve conflicts by comparing the newness of production rules.

[Problem to be solved by the invention]

However, the conventional expert system described above does not have a mechanism for inference (judgment) based on past experience, which is required in a diagnostic inference system, and a ms that can assign a priority between appropriate production rules. Conflict resolution could not be performed and appropriate inference could not be made.

The present invention has been made in view of the above circumstances, and aims to provide an expert system that can easily make inferences based on past experience and prioritize production rules, and can make appropriate inferences. .

〔Example〕

Embodiments of the expert system according to the present invention will be described below with reference to the drawings of FIGS. 1 to 6.

FIG. 1 is a block diagram showing one embodiment of a device, such as a workstation, for implementing an expert system according to the present invention. In the figure, a keyboard/mouse control section 11 controls input of various data, commands, etc. from a keyboard 12 and a mouse 13 connected to the keyboard 12.

The disk control unit 14 controls access to the disk device 15. This disk device 15 stores data groups such as programs including programs based on flowcharts showing processing procedures to be described later, dictionary files for kana-kanji conversion having a hierarchical structure, font configuration data, and bitmaps for displaying each of the above-mentioned fonts. It is stored. Further, the disk device 15 stores predetermined knowledge as rI necessary for inference.
It stores a knowledge base in which rules consisting of F...THENJ, that is, a production rule set (hereinafter referred to as a rule set) are expressed as a hierarchical structure and accumulated. Each rule set is counted each time its own rule set is executed, and the counted value, that is, a reference count value is set as attribute information of the rule set in each of the rule sets.

The display control unit 16 is a display device (hereinafter referred to as
Bitmap data corresponding to font configuration data and windows corresponding to various functions are displayed on the CRT (CRT) 17.

The real memory 18 consists of a RAM, which stores an operating program for operating the system, a keyboard 12 which stores commands and data input from the mouse 13, the conversion dictionary file, font configuration data, the display bitmap, etc. Stores part of the data group.

A central processing unit (hereinafter referred to as CPU) 19 comprehensively controls each of the control units and the real memory 18 in a time-sharing manner, and also activates a rule set based on the value of the reference count set in each of the rule sets. Perform inference. Further, the CPU 19 counts up the reference count of the executed rule set, and sets the count value in the rule set.

Next, the operation of the workstation will be explained based on the flowchart shown in FIG.

In addition, FIG. 3 is a browser 30 that shows the relationship between each rule set such as "Trust", "Bond and Bond Investment Trust", and "Bond and Bond Investment J". In this embodiment, this corresponds to the top node in the browser 30. Let us start the inference from the rule set "Trust".

Now, when the operator operates the mouse 13 to specify the ruleset "Trust" 40 as the top node from the browser 30 displayed on the CRT 17, as shown in FIG. Activate the ruleset as a node (step 100) - the conditional part (rule i.e. rlF) of the activated ruleset
...THENJ) is satisfied or not (step 101), and if the condition part is satisfied, it is determined whether the corresponding rule clearly specifies a lower rule set of the rule set to be executed next. Step 10 to decide
2) If the result of the judgment is clear, a list of the corresponding lower rule sets is created (step 103), and the contents of this created list (lower rule sets) are displayed on the CRT 17. .

In the embodiment, the rule set "Trust" 40 includes rIF...THEN of Rule 1 or Rule 2, as shown by arrows A and B in FIG.

Each rule specifies which lower rule set is to be executed next depending on the rule that holds true.

Here, rule 1 of rule set "Trust" 40 is established, and the corresponding lower rule set "Public and corporate bond investment trust J50" is established.
shall be executed. As shown in Figure 5, this rule set "Bond and Corporate Bond Investment Trust J50" clearly has four rule sets such as "Mid-term International Fant", "Bond Investment", etc., which are the next executable lower-order rule sets of the relevant rule set. However, when Rule 1 is established, it is not specified which lower-level rule set is to be executed next, and each rule set in the rule browser 30 is marked with arrow C in FIG. As shown, the reference count value is set.

In this way, when it is not specified which lower-level ruleset is to be executed next, in the inference according to the present invention, the ruleset that has been executed the most times in the past, that is, the count value of the reference count is the largest. The method is to execute from a ruleset.

Here, the above-mentioned rule set “Public and Corporate Bond Investment Trust J50
It is assumed that the rule set "Medium International Fant" 60 has the largest reference count value among the lower rule sets.

By the way, the CPU 19 investigates the count value of the reference count of each created rule set (step 104), lists the rule sets sorted in descending order of the count value based on the investigated result (step 105), and further , it is determined whether or not there is a rule set that has not been executed yet in the list in step 105). If there is a rule set, the rule set with the largest reference count value is selected first. For example, the ruleset '
Step 107) to take out “Medium International Fant” 60,
The extracted ruleset is activated (step 108).

Then, when the CPU 19 starts a ruleset extracted from the list, such as the ruleset "Medium International Fant" 60 shown in FIG.
Step 109)
, if the condition part is not satisfied, the process returns to step 106 and executes the steps from this step onwards. That is, the ruleset having the next largest count value of the reference count of the ruleset extracted in step 107 is activated. .

On the other hand, if the condition part is satisfied in step 109, the CPU 19 returns to step 102 and executes the steps after this step.
If there is a rule set lower than the rule set activated in step 08, steps 103 and subsequent steps are executed for the lower rule set, and step 102
If there is no lower-level rule set, such as the rule set "Medium International Fant" 60 shown in FIG. 6, the activated rule set is determined to be the lowest-level rule set, and Open the message card of the rule set and write the contents of the message card to C.
It is displayed on the RT 17 (step 110), and then the execution of the inference process is ended.

Note that in step 106, the CPU 19 executes all the rule sets in the list, and if there is no rule set (in this case, the condition part is not satisfied for all the rule sets). To do this, return to the higher level rule set for each lower level rule set listed (for example, each rule set for "Public and Corporate Bond Investment", "Medium Term International Fund", etc. shown in Figure 2) (step 1).
11), it is determined whether this upper-level ruleset is the highest-level ruleset 1- (ruleset corresponding to the top node, for example, the ruleset "trust" shown in FIG. 2) or not (step 112). , if it is not the top rule set, the process returns to step 106 and executes Stella 1 and subsequent steps, but in this case, since there is no rule set in the list, the process proceeds to step 111. Become.

On the other hand, if the CPU i 9 determines in step 112 that it is the highest rule set, it ends the inference process.

In addition, if the condition part does not hold in step 101, the CPU 19 determines whether there is a next condition part (step 113), and if the condition part exists,
Step 1: Determine whether the corresponding conditional part is satisfied or not.
14) As a result of this judgment, if the conditional part holds true, the process proceeds to step 102, and further, if the conditional part does not exist in step 113, or if the conditional part does not hold in step 114, inference processing is executed. end.

In the above example, the rule set "
Although the inference is started from the rule set "Public Corporation Investment Trust", the present invention is not limited thereto.
Inference may be started from any rule set such as .

According to the above embodiment, the ruleset that has been executed the most times in the past is given top priority for execution, making it easy to make decisions based on past experience and assign priorities between appropriate production rulesets. [Effects of the Invention] As explained above, according to the present invention, when inference is executed, the number of times each production rule is executed is counted and the count value is retained. As a result, the rule with the largest stored count value is executed with the highest priority, making it easy to make judgments based on past experience and prioritize appropriate production rules, making it possible to make appropriate inferences. The advantage is that it can be done.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of a device for implementing an expert system according to the present invention, and FIG.
FIG. 3 is a flowchart showing the operation of the inference processing of the device shown in FIG. 3. FIG. 3 shows an example of a browser showing the relationship between rule sets. FIGS. It is a figure showing an example. DESCRIPTION OF SYMBOLS 11... Keyboard/mouse control part, 12... Keyboard, 13... Mouse, 14... Disk control part, 15... Disk device, 16... Display control part, 1
7...Display device, 18...Real memory, 19...Central processing unit (CPU).

Claims (1)

[Claims] In an expert system that has a knowledge base that stores a plurality of production rules related to predetermined knowledge and that executes the stored rules to make inferences, the number of times each rule is executed is: and a control means that executes a corresponding rule according to the held count value, and a production that corresponds to the largest count value when performing inference.・An expert system that prioritizes and executes rules.