JPH01158509A - Verifying device for expert system - Google Patents

Abstract

PURPOSE:To execute the efficient, quantitative and qualitative verification by comparing the operation recording of an expert operator and the action recording of an expert system in the same conditions as this while the expert system of the verifying object is simultaneously operated in parallel to operating the plant. CONSTITUTION:The operation pattern recording of an expert 2 is transmitted to a verifying device 3, the data concerning a plant 4 are inputted from a super vising control device 5 to an expert system 1 and the data to operate the plant 4 are communicated to the verifying device 3 as an action recording. The opera tion pattern recording of the expert 2 and the action recording of the expert system 1 are inputted to the verifying device 3, respective types of operation are executed based on the evaluation object operation item and respective types of the parameter for evaluation and a verifying result 6 of the knowledge base of the expert system 1 is outputted. Thus, the verification of the knowledge base of the expert system can be executed efficiently, quantitatively and qualita tively.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to an expert system verification device for verifying an expert system for supporting plant operation or diagnosing abnormalities.

(Conventional technology) As social mechanisms have become more complex over the years, various specialists have emerged. For example, in a large hospital, there are many departments such as internal medicine, surgery, otorhinolaryngology, dentistry, obstetrics and gynecology, etc., and there are specialists in charge of each department. Design technology is divided into countless specialized fields, and as technology advances, it is expected that the number of specialists will increase. In these specialized fields, in addition to the knowledge organized as advanced academic knowledge such as laws, principles, and methods related to each field, there are also specializations that have deep specialized knowledge that includes know-how from experience. Home is working. By imparting the deep specialized knowledge of these experts to computers, it has the ability to solve problems in specific fields at a level close to that of experts, preventing misunderstandings and oversights, and improving work efficiency. A system that provides support, or, for non-experts, a system that consults on solving problems in the field is called an expert system.

This expert system basically consists of a knowledge base for registering expert knowledge in a predetermined format;
An inference mechanism that uses the knowledge in this knowledge base to guide questions (input data) from the user, analyze and understand the responses, and derive inference results, as well as human data and intermediate results. It consists of a database for storing information such as information, a knowledge acquisition support module for constructing and updating the knowledge base, an inference process explanation module for explaining the basis of inference results, and a user interface module. .

Now, as a conventional method for verifying the knowledge base of an expert system constructed by registering the knowledge of experts, the knowledge base la is shown in FIG.

The expert system 1 consisting of the database lb, the inference mechanism 1c, and the user interface 1d has an expert 2 who has provided knowledge artificially inputting various situations through the user interface 1d. The method used is to evaluate the output results.

However, since this method requires manual input by experts who have provided knowledge to the expert system, the effort required for verification is excessive, the efficiency is low, and the verification lacks objectivity. There is a problem. Furthermore, since the expert system is verified independently, there is a problem in that it is not possible to evaluate the behavior of the expert system when it is actually combined with a plant.

(Problems to be solved by the invention) As described above, conventional verification methods are not only inefficient in verification, but also cannot perform quantitative or qualitative verification, and furthermore, There was a problem in that it was not possible to evaluate the behavior of the Exverton stems when combined.

The purpose of the present invention is to be able to efficiently, quantitatively and qualitatively verify the knowledge base of an expert system without manual verification by experts;
Moreover, it is an object of the present invention to provide an expert system verification device capable of evaluating the behavior of an expert system when actually combined with a plant.

[Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the present invention is configured with a knowledge base, a database, and an inference mechanism, and is configured to support plant operation or detect abnormalities. A device for verifying an expert system that performs diagnosis includes an operation record importing means for importing the plant operation records of skilled operators, an operation record importing means for importing the operation records of the expert system, and evaluation target items. and a verification standard setting means for setting various evaluation parameters in advance, plant operation records and expert system operation records taken in by the operation record importing means and the operation record importing means, and the evaluation target items and various The apparatus is configured to include an arithmetic control means for performing various arithmetic operations based on evaluation parameters to obtain a verification result of the knowledge base of the expert system, and an output means for outputting the verification result by the arithmetic control means.

(Function) Therefore, in the present invention, by adopting the above-described means, the expert system to be verified can be operated in parallel with the operation of the plant by a skilled operator.
Driving records of experienced operators.

By comparing this with the operation record of the expert system under the same conditions, the knowledge base of the expert system is verified, and it is possible to check whether the expert system is operating correctly and can withstand practical use, or which items' knowledge base is It becomes possible to efficiently evaluate whether it is incomplete.

(Example) The present invention operates an expert system to be verified in parallel with a skilled operator operating a plant,
This test evaluates how faithfully the expert system can reproduce the driving of a skilled operator who is a knowledge provider by comparing the driving record of a skilled operator with the operating record of an expert system under the same conditions. be.

Hereinafter, the present invention will be described with reference to an embodiment shown in the drawings.

FIG. 1 is a diagram showing an example of the configuration of an expert system verification device according to the present invention. As shown in FIG. 1, the verification device 3 of the present embodiment has operations that are plant operation records of skilled operators (experts), and a first transmission system that is an operation record acquisition means for capturing turn records. A control device 31, a second transmission control device 32, which is an operation record capture means for capturing operation records of the expert system to be verified that are executed in accordance with plant conditions, and registration of operation items to be evaluated. A keyboard 33 serves as a verification standard setting means for presetting various evaluation parameters such as plant influence weight coefficients for each item, and a first transmission control device 31 and a second transmission control device 32. It is a calculation control means that receives the incorporated operation pattern records and expert system operation records as input, performs various calculations based on the evaluation target operation items and various evaluation parameters, and obtains verification results of the knowledge base of the expert system. Arithmetic control device 34 and operation items to be evaluated.

A CR7 display device 35 and a printer 36, which are output means for displaying and printing various evaluation parameters and verification results by the arithmetic and control device 34, and an auxiliary memory for storing all data in the verification device 3. It consists of a device 37.

FIG. 2 is a diagram showing an example of the overall configuration when the verification device of FIG. 1 is applied, and the same parts as in FIG. 1 are denoted by the same reference numerals.

In FIG. 2, the aforementioned expert system 3 is connected to a monitoring/control device 5 and an expert system 1 for operating the plant 4. In FIG. The expert 2 obtains data regarding the plant 4 from the monitoring/control device 5 and operates the plant 4 based on the expert's 2 knowledge and experience. The details of the driving are transmitted via the monitoring/control device 5 to the verification device 3 as a record of the expert's operation pattern. In addition, the expert system 1 inputs data regarding the plant 4 from the monitoring/control device 5, operates the inference mechanism IC, utilizes the knowledge base 1a, and transmits data for operating the plant 4 as an operation record to the verification device 3. It is supposed to be done.

Next, the operation of this embodiment will be explained.

First, in the verification device 3, various evaluation parameters such as registration of operation items to be evaluated and plant influence weight coefficients for each item are set in advance using the keyboard 33. Next, the expert 2 obtains data regarding the plant 4 from the monitoring/control device 5 and operates the plant 4 based on the knowledge and experience of the expert 2. After that, it is transmitted to the verification device 3 as a record of the expert's operation pattern. It has become. In addition, data regarding the plant 4 is inputted to the expert system 1 from the monitoring/control device 5, and the inference mechanism IC works and utilizes the knowledge base 1a to send data for operating the plant 4 to the verification device 3 as an operation record. transmitted. As a result, the operation pattern record of the expert 2 and the operation record of the expert system 1 are input to the verification device 3, and various calculations are performed based on the above-mentioned operation items to be evaluated and various evaluation parameters. The verification result 6 of the knowledge base 1a is output.

Specifically, in order to normalize the degree of influence of each operation item on the plant 4, a plant influence degree weighting factor K is set and registered for each operation item using the keyboard 33.

At this time, the number of the knowledge base 1a to which each operation item is related is also set and registered. Here, the operation items include person/cut, open/
There are digital operation items such as closing, and analog operation items that select a specific value from a series of values.

For digital operation items, as shown in FIG. 3, the difference between the operation pattern of specialization 2 and the operation pattern of expert system 1 is evaluated by the total operation mismatch time Δt. In the case of Fig. 3, there is a discrepancy of 0N10FF three times from time tl to tl, and Δtl and Δtl respectively.
t2. If Δt3, then Δ1−ΣΔ11−Δ1. +Δt2+Δt3 (however, n:
(number of mismatches). Then, the ratio of this Δt to the entire verification time is calculated, and the plant influence weighting coefficient Kd for that operation item is calculated.
The deviation value Dd of the digital type operation item is obtained by multiplying by .

Dd −Kd XΔt/(tl−tl) (0<K
(d≦100) Then, the operation item whose deviation value Dd is larger than the standard deviation value Dsd of the digital type operation item is restored as a poor parameter.

On the other hand, for analog type operation items, as shown in Figure 4,
The difference between the operation pattern of specialization 2 and the operation pattern of expert system 1 is evaluated by (difference in operation amount) x (operation time). In the case of FIG. 4, from time t1 to tl
There is a discrepancy in the amount of operation and operation timing between then, and the difference dL in the amount of operation at time t is d(t)=lfo(t) f(t)l (however, fo
(t): Operation ffi, f(t
) :Movement amount of the expert system 1 at time t). Then, the ratio of the cumulative operation difference during the entire verification time in the operation range Δf of the expert 2 is calculated, and the deviation value Da of the analog type operation item is obtained by multiplying it by the plant influence weighting coefficient Ka of the operation item. Desired.

(However, O<Ka≦100) Then, the operation item whose deviation value Da is larger than the standard deviation value Dsa of the analog type operation item is restored as a poor parameter.

The above calculations are performed for all operation items, and the completeness of the expert system 1 is determined as the overall fidelity F for the expert 2 from the following equation.

F-100-(ΣDd+ΣDa) (%) Therefore,
The value of F takes a larger value as the total operating deviation ΣD decreases,
This becomes an expert system that behaves more like the expert 2. Note that the maximum value of F is 100, and the unit is %. Further, as an index for improving the knowledge base 1a of the expert system 1 with a low F value, rules related to the restored poor parameters are searched and output in the order of the bad values of the poor parameters. Thereby, it is possible to change from the knowledge base 1a that has a high improvement effect. The flow of the above algorithm is shown in FIG.

As described above, the expert system verification device 3 according to the present embodiment qualitatively and quantitatively evaluates the difference in level between the expert 2 who provided the knowledge and the knowledge base 1a of the expert system 1. The completeness of the expert system 1 can be clearly and objectively verified.

In other words, qualitatively speaking, which operation items affect the plant 4
How much does the expert system 3
It is possible to understand in detail whether the This makes it possible to efficiently know which side of the knowledge base 1a is incomplete. In addition, quantitatively, the expert system 3
The overall degree of completeness can be concretely expressed using an evaluation function called total fidelity F. This makes it possible to clearly understand whether the performance of the expert system 3 has been improved after changing the knowledge base 1a. Furthermore, on the other hand, it is also possible to use the above-mentioned overall fidelity F as a major criterion when transferring the expert system 3 to practical use.

[1. Effects of the Invention] As explained above, according to the present invention, the knowledge base of an expert system can be efficiently and quantitatively and qualitatively verified without manual verification by experts. It is possible to provide an expert system verification device that can perform the following: and evaluate the behavior of the expert system when actually combined with a plant.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of an expert system verification device according to the present invention, FIG. 2 is a diagram showing an example of the overall configuration when the verification device of the same embodiment is applied, and FIG. 3 is a diagram showing the same embodiment. FIG. 4 is a conceptual diagram for explaining the deviation calculation of the digital type operation item in the same embodiment, FIG. 5 is a conceptual diagram for explaining the deviation calculation of the analog type operation item in the same embodiment, and FIG. FIG. 6 is a flowchart for explaining a conventional expert system verification method. 1...Expert system, 1a...Knowledge base, 1b...Database, IC...Inference mechanism, 1d
... User interface, 2... Expert, 3.
... Verification device, 31... First transmission control device, 32.
...Second transmission control device, 33...Keyboard, 34
... Arithmetic control device, 35 ... CRT display device, 36
...Printer, 37...Auxiliary storage device, 4...Plant, 5...Monitoring/control device, 6...Verification results. Applicant's agent Patent attorney Takekin Suzue 1 Figure 2 t,
t2 place r, 1 Fig. 5 Fig. 6

Claims (1)

[Claims] A device for verifying an expert system that is configured to include a knowledge base, a database, and an inference mechanism and performs plant operation support or abnormality diagnosis, which captures plant operation records of skilled operators. an operation record importing means for importing operation records of the expert system; a verification standard setting means for presetting evaluation target items and various evaluation parameters; The plant operation records and expert system operation records taken in by the importing means and the operation record importing means are input, and various calculations are performed based on the evaluation target items and various evaluation parameters to obtain the verification results of the knowledge base of the expert system. An expert system verification device comprising: arithmetic control means for determining the arithmetic and control means; and an output means for outputting a verification result by the arithmetic and control means.