JPH03292534A - Knowledge processing method - Google Patents

Abstract

PURPOSE:To attain the fast control of assurance by defining the assurance described in each rule as a variable called assurance name and setting separately the value of the variable. CONSTITUTION:The inference result of an inference engine 6 is confirmed by a CRT terminal 14. If the inference result is not coincident with the subject of an operator, the assurance is corrected toward the coincidence via CRT terminal 11. Then an inference is performed again by the engine 6. When the coincidence is obtained between the inference result and the operator's subjectivity with repetition of the inference, the inspection is finished as to the assurance. During the progress of this job, the correction input is applied to only the definition value of the ambiguity of the operator received from the terminal 11. Then no correction is required to other knowledge bases, etc. Thus the assurance is quickly controlled and the workability is extremely improved for the knowledge processing.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a knowledge processing method using inference based on rules, and in particular to a confidence level in which each rule is given a value based on the operator's subjective judgment. This paper relates to a knowledge processing method that determines rules with certainty and makes inferences using these rules.

[Conventional technology]

A conventional example of a knowledge processing system that performs inference by setting a certainty level for each rule is the one described in Japanese Patent Application Laid-Open No. 62-270712, in which the knowledge and experience of an operator is expressed as a production rule, and each rule is The CF value (CertaintyFacte) is used as an index to express the ambiguity of rules.
r: confidence level) is introduced to increase the reliability of inference. This confidence level is included in the rule description as a numerical value representing the confidence level, and is input to the inference system together with the rule input.

[Problem to be solved by the invention]

Since the confidence level expresses the operator's subjective opinion, it is necessary to verify that the result matches the operator's feelings. To do this, it is necessary to execute the inference process while adjusting the confidence level and set the confidence level so that a result that the operator is satisfied with can be obtained. However, since modifying the confidence level involves modifying the rules, there is a problem in that the work becomes complicated.The purpose of the present invention is to solve this problem in the prior art and enable quick confidence adjustment work. The objective is to provide a knowledge processing method for

[Means to solve the problem]

In order to achieve the above object, in the present invention, the confidence described in each rule is not its value, but a variable with a confidence name corresponding to the rule, and the value of this variable is The man-machine mechanism provided to set and modify the values of each confidence base can be determined on the database, and when inference is executed, the values on the database for the confidence names written in the rules are used. I did it like that.

[For production]

The confidence level described in each rule is a variable called confidence level name,
By making it possible to set the value separately, the confidence value itself can be changed by simply changing it, and there is no need to modify the entire rule. Therefore, it is possible to greatly improve workability and maintainability. Furthermore, if a function is provided to check the value of a variable having a certainty name set in the man-machine mechanism, errors in entering the certainty factor can be prevented.

〔Example〕

Embodiments of the present invention will be described below based on the drawings. 1st
The figure is a block diagram showing an embodiment of a knowledge processing system to which the method of the present invention is applied. A process management control computer system 15 applying knowledge processing includes a process data collection process 2 that takes in various process information from sensor data 1 and converts it into a computer internal representation format, and a process data file 3 that stores the processing results. Knowledge processing data processing 4 that creates information used for knowledge processing judgment upon receiving a processing request after completion of process data collection processing 2
and a knowledge data file 5 that stores the processing results, an inference engine 6 that refers to the file 5 and executes inference according to a request from the data processing 4, and stores knowledge (rules) used by the inference engine 6. A knowledge base 7 (generally, this differs depending on the target process, and various types are prepared), a confidence base 8 corresponding to the confidence variable in the rule, and these knowledge bases 7 and confidence It consists of a knowledge editor 9 for creating a degree base 8, a certainty setting process 10, and a man-machine process 13 for displaying inference results and confirming information such as process data used for inference. In addition, CR is used for the purpose of inputting knowledge, confirming confidence, and confirming and displaying inference results.
T terminals 11.12°14 are provided.

The process data collection process 2 processes information from the sensor data 1 in chronological order and accumulates raw process information. The storage unit is determined based on the characteristics of the target process, and information is collected in this determined unit and stored in the process data file 3. This stored information is instantaneous data R of the process as shown in FIG.

If this information R is used as is, inferences will be made using the noisy state of the process, which is not desirable in terms of system processing. For this.

Knowledge processing data processing 4 removes process noise and performs processing so that appropriate judgments can be made. Various methods can be considered for this processing, and as an example, the results of the averaging processing are shown by the polygonal line in FIG. Information in this form is stored in the knowledge data file 5.

On the other hand, the knowledge base 7 used for inference is stored on the CRT terminal 12.
Knowledge information such as rules input from the computer is converted by the knowledge editor 9 into a form suitable for computer processing. When creating this knowledge base, the knowledge editor 9 detects the confidence variable described in the rule and stores the variable name in the confidence base 8. The certainty setting process 10 sets values for the certainty variables extracted by the knowledge editor 9. This process is performed via the CRT terminal 11, and the set value is stored in the certainty base 8 in a form corresponding to the variable name. In the certainty base 8, each variable corresponds to the knowledge base 7, so management of certainty variables is performed for each knowledge base. This means that confidence variables with the same name can be used in different knowledge bases, and the uniqueness of variables only needs to be maintained within the awareness base.

Knowledge base 7 and process data 5 configured in this way
The inference engine 6 executes inference using . When a reference to a confidence variable appears during this execution process, the inference engine 6 extracts the corresponding confidence from the confidence base 8 and uses it. Since a plurality of knowledge bases 7 are prepared corresponding to the target process contents, the knowledge processing data processing 4 judges the process status, determines which knowledge base to use for inference, and performs inference. Contact engine 6.

Once the inference is complete, the results are stored in the knowledge data file 5.
is stored in the CRT terminal 1 through man-machine processing 13.
4 is output. The operator refers to the results output to the CRT terminal 14 and uses them for system operation.

Here, it is necessary to verify whether the knowledge input from the CRT terminal 12 operates as expected.

For this purpose, the inference engine 6 executes inference, and the CRT
Refer to the results output to the terminal 14 and proceed with the work. In this work, it is first necessary to verify the validity of the rule itself, but since the procedure is not directly related to the present invention, a description thereof will be omitted. Once the validity of the rule has been verified, the next step is to match the answer obtained as a result of the inference with the subjective judgment of the operator.

This is a verification of the confidence value that specifies the ambiguity, which represents the operator's subjective judgment. The method of proceeding with this work is to check the inference result by the inference engine 6 on the CRT terminal 14, and if it does not match the subjective opinion of the operator, the confidence level is corrected via the CRT terminal 11 in the direction of matching, and then the inference result is checked again using the inference engine. Execute the inference according to 6. When a result that matches the operator's subjective opinion is obtained by repeating this process, the verification of the confidence level is completed.

As this work progresses, correction input will be made on the CRT terminal 11.
The main point of the present invention is that it is only the operator's definition value of ambiguity from , and does not require modification of other knowledge bases. This improves work efficiency in return.

In the above explanation, the purpose was to modify the confidence level to match the subjectivity of the operator, but the present invention is useful when the process status changes and it becomes necessary to change the set value of the confidence level. is also valid. FIGS. 3 to 5 show examples of how to utilize this reliability correction function. FIG. 3 shows an example in which the temperature and pressure as process data are each classified into three levels, and the degree of risk of the judgment result is expressed by the degree of certainty based on the combination thereof. The figure shows that the reliability classification is changed depending on the period from the start of operation of the target process, with (a) for the early stage of operation and (b) for the middle stage of operation.

The final stage of operation is expressed in (c). That is, at the beginning of operation, when temperature < T3 and pressure < PL, confidence = c
fl When temperature < T3 and PL < pressure < P3, confidence = c f2 When other than the above, confidence = cfe In the middle of operation, when temperature < T2 and pressure < P2, confidence = c
fl T2<temperature<T3 and pressure<P3. or temperature <
When T3 and P2<pressure<P3, confidence = c
f2 When other than the above, confidence = cfe At the end of operation, when temperature < Tl and pressure < P3, confidence = c
This means that when fL TI<Temperature<T3 and Pressure<P3, confidence level = c f 2 When other than the above, confidence level = c f e .

In this way, since it is necessary to treat the confidence level as changing depending on the driving situation, the confidence level in the rule description is classified as shown in Figure 4 and handled using the variables CFij (i, j = 1 to 3). do. If this is done, the values of the confidence variables corresponding to each driving situation will be as shown in FIG. If the present invention is used when it is desired to perform inference by changing the confidence level as the driving situation changes, it is only necessary to change the value of the corresponding confidence variable as an input from the CRT terminal 11, and the driving situation This makes it easier to build a system that can flexibly respond to changes.

〔Effect of the invention〕

As described above, the present invention improves the workability when adjusting inference results by making it possible to handle certainty, which represents the operator's subjective judgment criteria, independently of rules in a knowledge processing system. has a big effect.

[Brief explanation of the drawing]

Fig. 1 is a functional block diagram showing one embodiment of the present invention;
The figures are explanatory diagrams of data processing for knowledge processing, Fig. 3, Fig. 4,
FIG. 5 is a detailed explanatory diagram of the present invention. 5...Data file for knowledge processing, 6...Inference engine, 7...Knowledge base, 8...Confidence level Heas, 9
...Knowledge editor, 10...Confidence level setting process, 11
．． 12.14...CRT terminal, 15...computer system.

Claims (1)

[Claims] 1. Knowledge using a system that has rules for making inferences and in which each rule is given a degree of certainty based on the operator's subjective evaluation of the result of applying the rule. In the processing method, on the knowledge storage means in which rules are written, the above-mentioned confidence corresponding to each rule is described as a variable having a unique variable name, and each variable name representing the above-mentioned confidence is associated with its value. A confidence storage means for storing the confidence value, and an input means for inputting the confidence value corresponding to each variable name to the means are provided.
When one rule is retrieved from the knowledge storage means, the value corresponding to the variable name representing the certainty in the retrieved rule is read from the certainty storage means and used for executing the rule. Knowledge processing methods. 2. The knowledge processing method according to claim 1, further comprising display means for displaying the confidence value inputted from said input means.