JPS634333A - Expert system - Google Patents

Abstract

PURPOSE:To improve rule retrieval efficiency by calculating a frequency which is used so far for deduction for each rule in a rule data base and setting the order of rule retrieval according to the used frequency. CONSTITUTION:A rule retrieval efficiency increasing mechanism 5 stores the use frequencies (f) of all diagnostic rules (r). When a signal for the use of some diagnostic rule ri is received from a deduction mechanism 2, '1' is added to the use frequency fi of the diagnostic rule ri to update the use frequency fi. When the frequency is updated, the use frequency fj of a diagnostic rule rj having priority Pi-1 which is one higher than application priority Pi set currently for the diagnostic rule ri is read out and the read use frequency fj of the diagnostic rule ri is compared with the use frequency fi of the updated diagnostic rule ri; when fj<fi, the application priority of the rule ri is set to Pi-1 and the application priority of the rule rj is set to Pi.

Description

[Detailed Description of the Invention] 3. Detailed Description of the Invention [Object of the Invention] (Industrial Application Field) The present invention is applicable to tasks that require a high degree of specialized knowledge, such as equipment abnormality diagnosis in nuclear power plants. Ill rely on expert systems to assist on behalf of experts.

(Prior Art) For example, when an abnormality occurs in equipment in a nuclear power plant or the like, specialized knowledge regarding each component equipment is required to investigate the cause and take appropriate measures. However, experts with such knowledge are not always available, and even if an expert in one field starts investigating the cause, as the investigation progresses, experts in other fields may also be needed. . Especially in nuclear power plants)4
In a human-type plant, the number of devices that make up the project is enormous, and it includes 41 specialized devices, so it requires many experts and their specialized knowledge is highly advanced and specialized. be done. Moreover, since feminine integrity is strictly required, it is necessary to quickly determine the cause of the abnormality and take countermeasures.

Against this background, there has recently been active development of devices called expert systems that support abnormality diagnosis of plant component equipment using knowledge engineering techniques in place of experts. Such expert systems may also be used in other fields, such as medical diagnosis,
It is being used in various Q-type knowledge fields such as accounting conreel knowledge, work 41 strokes, mechanical equipment 31, etc.

An expert system basically consists of a rule database and an inference mechanism. Taking a plant abnormality diagnosis system as an example, the rule database stores diagnostic rules that express knowledge such as the diagnostic inference process and work procedures for performing diagnosis. The inference mechanism sequentially makes inferences according to the diagnosis rules while presenting the work procedure for investigating the cause to the operator via the input/output interface based on the machine height status information input from the operator or Plan 1-. This is a mechanism for determining the cause of an abnormality.

The most frequently used rule expression format in rule databases is called the °'IF-・,18EN-" format. Specifically, conditions are written in the "IF-" section, and "T HE N-" is written. ``This section describes the repeat operation that will be executed when the conditions are met.''
The IF-" part is called the condition part, and the "THEN-" part is called the action part.

Inference is generally carried out in the form of dialogue with the worker, and basically, a rule that has a condition in the condition part that matches the state information input by the worker is searched for in the rule database. [Proceeds sequentially] by repeating the steps of selecting a rule and executing the action part of the selected rule.

(Problem that the invention seeks to solve) In order for expert systems to be put to practical use in various fields in the future, it will be necessary to enrich the rule database, and as a result, the number of rules stored in the rule database will have to increase. Ru.

By the way, the time required for inference processing in an expert system is determined by the time required for rule matching and search.

Therefore, in the above-mentioned expert system that handles a huge number of rules at a practical level, it takes a lot of time to match and search the rules, and there is a problem that inference processing cannot be performed quickly.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide an expert system that can perform inference processing at high speed by shortening the time required for matching and searching rules in a rule database.

[Structure of the invention]

(Measures for solving the problem 82) To achieve the above object, the present invention calculates the frequency with which each rule in the rule database has been used for inference, and according to this usage frequency, This system is equipped with a rule search efficiency mechanism that sets the rule search order.

(Operation) The rule search efficiency mechanism sets a search order for each rule in the rule database in descending order of past usage frequency. Inference 1 follows this search order, that is, the most frequently used rule is first followed by a matching search.

As a result, the probability of discovering the rule to be used at an early stage increases, making the search more efficient.

(Example) The present invention will be explained below with reference to Examples.

FIG. 1 is a functional block diagram showing the configuration of an embodiment of an expert system according to the present invention. Taking the case of a plant abnormality diagnosis system as an example, the rule database 1 contains knowledge such as the inference process for diagnosing equipment and the work procedures such as human interaction for that purpose.
, THEN-" format is stored. The inference mechanism 2 extracts the corresponding diagnostic rules from the rule database 1 based on equipment status information input from workers or the plant regarding equipment abnormalities. By repeating the steps of reading and executing the operation part, the inference is progressed sequentially.The input/output interface mechanism 3t, the friend, and the inference mechanism 2 display the work procedures and diagnosis results regarding the diagnostic target equipment. 3a
is presented to the worker via the keyboard 3b.
The status information inputted from the inference mechanism 2 is transmitted to the inference mechanism 2. Also,
The input/output interface mechanism 3 transmits specialized knowledge inputted from the keyboard 3b to the rule editing mechanism 4 prior to diagnosis. The rule editing mechanism 4 creates diagnostic rules necessary for performing a diagnosis based on the specialized knowledge input via the input/output interface mechanism 3, and writes the diagnostic rules into the rule database 1, or writes the diagnostic rules into the rule database 1 that has already been written. Modify or delete diagnostic rules. The rule search efficiency mechanism 5 calculates the number of times the inference mechanism 2 reads a diagnostic rule from the rule database 1, that is, the frequency of use, for each rule by a training process, and assigns application priority to each diagnostic rule according to its frequency of use. This is to set.

FIG. 2 is a schematic diagram showing the configuration of the rule database 1. The rule database 1 is composed of a rule header 6 and a diagnostic rule storage section 7.

The rule header 6 contains the application priority P1.
The identification number 1d of the six diagnostic rules and the storage address a of the diagnostic rule in the diagnostic rule storage unit 7 are stored. The diagnostic rule storage section 7 stores diagnostic rules r corresponding to the contents of the rule header 6.

Next, the operation of this embodiment configured as described above will be explained.

First, referring to the flowchart of the inference process in Figure 3,
The operation of the inference machine J/I42 will be explained. When an abnormality occurs in the device to be diagnosed and the operator starts this system, questions and the like are presented on the display 3a. When the operator inputs the requested state information using the keyboard 3b in response to this question, this state information is transmitted from the input/output interface mechanism 3 to the reasoning mechanism 2 (step 100).
).

When the inference mechanism 2 receives the status information, the inference mechanism 2 stores the diagnostic rule r having a condition part that matches this information in the rule database 1.
In order to select from within, the highest application priority P (=1)
The condition part of the diagnostic rule r with the input status information starts to be compared with the input state information (steps 101, 102>. If the result of this comparison does not match, then the application priority P(
=P+1) and the status information (steps 103 and 104). In this way, the matching search for diagnostic rules is performed in descending order of application precedence P(7), and when a matching diagnostic rule r is found, the action part of that diagnostic rule ri is executed to advance the inference ( Step 1
05).

Then, using a certain diagnostic rule r. like this! When inference is executed, a signal indicating that the diagnostic rule r is used for inference is transmitted to the rule search efficiency improvement mechanism (step 106).

If the result of applying the read diagnostic rule r is not determining the cause (step 107), the inference mechanism 2 uses gay spray 3a to further advance the inference.
presents a message prompting the user to input new status information,
The above block 1 is executed again based on the new information input by the worker.
Execute coses.

In addition, if the diagnosis rule r, the application result is determined to be the cause, step 107), # end the disconnection step 10)
8), informs the rule search efficiency improvement mechanism 5 that the diagnosis is complete (
Step 109), and present the diagnosis result on the display 3a (Step 110).

Next, the operation of the rule search efficiency improvement mechanism 5 will be explained with reference to the flowchart of the rule search efficiency improvement process shown in FIG. Rule Search Efficiency [11i15 stores the frequency of use f for every diagnostic rule r. Reasoning mechanism 2
When a signal for the use of a certain diagnostic rule r is received from (
Step 200), the usage frequency f1 of its diagnostic rule r
1°° is added to and stored as a new usage class Flf (step 201).

Next, the diagnostic rule r that rewritten this frequency of use f.
, the usage frequency fj of the diagnostic rule rJ, which has a priority p, -i that is one higher than the application priority P1 currently set for , is read out, and rewritten as the usage frequency fj of the read diagnostic rule rJ. Usage frequency f1 of diagnostic rule ri
(Step 202). As a result, if the relationship f, <f holds (step 203), the application priority of diagnostic rule r is set to P, -1, and the application priority of diagnostic rule r is set to P, Direct V (step 204
). In other words, the application priorities of the diagnostic rule r and the diagnostic rule r are exchanged J. In this way, ]From RIIi rule ri^
The relationship J between the usage frequency f of diagnostic rule r, which has a high application priority, and the usage frequency f, of diagnostic rule r,
l
+ is f, >f.

The application priority P of the diagnostic rule r is increased until it becomes . As a result, the application priority P of each rule r is rearranged in descending order of usage frequency.

The above process is repeated every time a usage signal for a certain diagnostic rule r.

When the diagnosis and completion signal is received from the inference mechanism 2 (step 205), the contents of the rule header 6 in the rule database 1 are rewritten in accordance with the rearranged application priority P.

In this way, the application priority P8 is set for each rule r in descending order of use frequency f, and by performing a collation search of diagnostic rules r in the order of the application priority P, necessary information can be quickly identified. The probability of finding a diagnostic rule increases, and therefore the time required for matching and searching is shortened, making it possible to quickly provide diagnostic results. Therefore, compared to the conventional method, it is possible to perform detailed diagnosis using a large number of diagnostic rules, and it is also possible to quickly implement countermeasures against abnormalities, resulting in great practical effects. .

(Effects of the Invention) As explained above, according to the present invention, since the search is performed from the most frequently used rules, the number of unnecessary rule matching searches is reduced, and the time required for inference processing is reduced. Therefore, inference processing can proceed at high speed even in an expert system with a large-scale rule database at a practical level.

[Brief explanation of the drawing]

Fig. 1 is a functional block diagram showing a schematic configuration of an embodiment of the present invention, Fig. 2 is a schematic diagram showing the configuration of a rule database of the embodiment, and Fig. 3 shows the operation of the inference mechanism of the embodiment. FIG. 4 is a flowchart showing the operation of the search efficiency improvement mechanism of the same embodiment. 1... Rule database, 2... Reasoning mechanism, 3.
... Input/output interface mechanism, 4... Rule editing horizontal, 5... Rule search efficiency mechanism, 6... Rule header, 7... Diagnostic rule storage section. Applicant's agent Mr. Sato Figure 4

Claims (1)

[Claims] In an expert system equipped with a rule database storing rules necessary for inference processing, and an inference mechanism that searches for a rule corresponding to input information from this rule database and proceeds with inference using the selected rule, An expert system comprising a rule search efficiency mechanism that calculates the frequency of use by the inference mechanism for each rule in the rule database and sets the order of the rule search based on the frequency of use for each rule. .