JPH08297574A - Method and device for constructing instance base, and method and device for instance base inference - Google Patents

Abstract

PURPOSE: To retrieve a past problem solution instance flexibly at a high speed when a solution for a new problem is generated by making good use of part problem solution instances. CONSTITUTION: A similar instance network generating device 107 is provided which inspects the property value of an arbitrary property of a new instance when the instance is inputted to an instance base 101 storing past problem solution instances, retrieves a conventional instance having an attribute value similar to the attribute value from the instance base 101, and records information showing that the new instance and conventional instance are similar in the arbitrary attribute as data on the respective instances. Consequently, data on the similar instance regarding the arbitrary attribute X of the arbitrary instance A in the instance base can be retrieved and referred to fast.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a case-based reasoning method and apparatus for generating a solution of a new problem case by referring to past problem solution cases, and is particularly suitable for machine design, plant diagnosis and medical diagnosis. A case-based reasoning method and apparatus thereof.

[0002]

2. Description of the Related Art In recent years, a case-based reasoning system has been attracting attention as a problem solving method based on knowledge processing represented by an expert system. Case-based reasoning is that when a problem to be solved is given, the past case similar to the problem is searched and the solution of the found similar case is appropriately corrected to obtain the solution to the problem. It is a framework of inference methods.

For an overview of case-based reasoning, see Shigenobu Kobayashi, "Present situation and prospects of case-based reasoning", Journal of Japan Society for Artificial Intelligence, Vo.
For details, see l.7, No.4, pp559-566, 1992. The case-based reasoning generates a solution of a new problem based on similar cases in the past, and therefore, when a problem is given like the rule-based reasoning that is generally used in the past, It is said that it is superior in the efficiency of obtaining a solution as compared with the method of generating a solution based on a rule from an unprecedented state. In terms of engineering, machine adjustment support systems (for example, Kotaro Nakamura, Shigenobu Kobayashi "Interactive case-based reasoning model and its application to machine adjustment support", Journal of Japan Society for Artificial Intelligence, Vol.4, No.6, pp704-713. , 1989) and various diagnostic problems, and its effects have been confirmed.

Incidentally, as related arts, there are JP-A-2-195473, JP-A-4-302027, JP-A-6-4290 and the like.

[0005]

[Problems to be Solved by the Invention] According to the above-mentioned paper (Shigenobu Kobayashi: Present state and perspective of case-based reasoning), case-based
A general framework of reasoning is written as shown in FIG. This conventional case-based reasoning apparatus includes a case base 201 which is a collection of characterized problem solving cases, and a problem analyzer 202 which characterizes a problem by a characterizing rule.
And a case searcher 20 that searches the case base 201 for a case that best matches the given feature of the problem.
3 and the domain knowledge is used to correct the solution of the case,
A case corrector 204 that is a solution to a given problem and a case repairer 205 that corrects cases and characterization rules using domain knowledge or another knowledge if the application of the search case to the problem fails. And a case store 206 for characterizing the obtained cases and storing them in the case base 201. Conventionally, the problem is characterized by giving each case an attribute name and an attribute value attached to the case in a frame format format. Basic data of conventional cases
The data format is shown in FIG.

Generally, in order to improve the accuracy of the solution generated by the case-based inference, it is preferable that the number of cases stored in the case-based is large. However, as the case base becomes larger, the time required to search for the case becomes a big problem.

In order to solve this problem, parallelization of case retrieval and hierarchization of case bases can be considered. Parallelizing case retrieval is not an essential solution in the sense that if the scale of the case base doubles, the search time also doubles. As shown in FIG. 12, the case-based hierarchization is to store the cases in the case-based in order in the attribute hierarchy of attribute 1, attribute 2, and attribute 3, and all attribute values are stored. It is efficient in retrieving similar cases of problems that have been decided.

However, when a search condition such as "search for all cases in which the value of attribute 3 is 3A" occurs in this example, the contents of attribute 3 need not be evaluated in all cases. However, there is a problem that the efficiency of the search is deteriorated. This is a problem that occurs when the search conditions do not match the pre-determined hierarchical structure.

It is an object of the present invention to provide a case-based reasoning method and a device therefor capable of performing case retrieval efficiently and at high speed.

[0010]

The above-mentioned object is to inspect the attribute name X and the attribute value of an arbitrary case A for the case stored in the case base, and to set the attribute value similar to the attribute value to the attribute name. Search for another case B in X, and case A and case B are attribute names X
This is achieved by associating a similar relationship with respect to each case.

[0011]

[Operation] Since a case is associated with a case similar to the case, in all cases stored in the case base, another case similar to each attribute of the case can be easily identified. it can. Therefore, for example, in the case where a condition such as “attribute X must be W” is given in the case search, if one case in which the attribute X is W is found, the case similar to that case and the attribute X is found. The name can be immediately obtained, and the case search can be speeded up.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a knowledge processing device to which a case-based reasoning method according to an embodiment of the present invention is applied. In this embodiment, a medical diagnosis system is constructed using this knowledge processing device.

In FIG. 1, the knowledge processing apparatus includes a case base 101, a problem analyzer 102, and a case searcher 10.
3, case correction device 104, case restoration device 105, and case storage device 106. These devices perform the same functions as the devices of the same name shown in FIG. Similar case network creation device 107 newly added by the device of FIG.
When a case A to be stored in the case base 101 is given, another case is examined in which the attribute name X and the attribute value of the case A are checked and the attribute name X has an attribute value similar to the attribute value. Case B
Searching from the table 101, case A and case B are attribute name X
The process of recording in each case that there is a similar relationship with respect to.

In the present embodiment, it is assumed that patient data for medical diagnosis is stored as an example in the case base 101. Here, the format of the patient data is a format as shown in FIG. 2, and attribute names such as sex, age, height, weight, body temperature, blood pressure, etc., attribute values of each attribute, and similar case names. , A set of a disease name which is a diagnosis result and a treatment method thereof.

Now, it is assumed that the patient data shown in FIG. 3 is given to the similar case network creating device 107. At this stage, the columns of the similar case names of this data are all blank.

FIG. 4 shows a similar case network creating apparatus 1.
It is a flowchart which shows the processing procedure which 07 performs. In step 401, the input case is set to the case A, and the attributes of the case A are set to the attributes 1 to n. In FIG. 3, the case name of case A is patient number ID: 1234. Step 40
In step 2, paying attention to the i-th attribute i, the attribute value Ai is obtained from the input data. In FIG. 3, for example, when the attribute i is an age attribute, Ai = 38.

In step 403, the set of cases stored in the case base is set as S. Step 404 is S
Is a branch to go to step 408 to end the process for the attribute i when it is an empty set. Step 405 is S
This is a process of extracting the case B from and obtaining the attribute value Bi of the attribute i. For example, as Case B, consider the data of patient number KA5678 shown in FIG. In FIG. 5, when the age attribute of the patient number KA5678 is checked, it is “39”. That is, Bi = 39.

If it is determined in step 406 that Ai and Bi have a similar relationship, similarity relationship processing 4
Execute 07. Details of the similarity processing 407 will be described later. Whether Ai and Bi have a similar relationship is
Generally, the judgment is made by using the domain knowledge depending on the attribute. In this example, patient ID 1234 and patient KA567
It is assumed that 8 has a similar relationship in the sense that they belong to a group of 35 to 40 years old.

At step 408, the attribute number is incremented by 1. In step 409, if the attribute number is larger than n, it is assumed that all the attributes of case A have been checked, and the process ends. If there are unprocessed attributes, the process returns to step 402.

FIG. 6 shows a detailed procedure of the similarity relation processing 407. In step 601, the patient KA is used in this embodiment.
In the example shown in FIG. 5, for example, a set R of similar cases regarding the age attribute of 5678 is R = {HJ9720, JS620.
1}. In step 602, case A (patient ID 1
234), case B (patient KA5678) and R are recorded in the similar case regarding the age attribute. By this processing, as shown in FIG. 7, similar cases regarding the age attribute of the case A (patient ID 1234) are KA5678, HJ9720, J.
It becomes S6201.

In step 603, case B (patient KA5
678) Case A (Patient I)
D1234) is added. That is, as shown in FIG. 8, similar cases regarding the age attribute of case B (patient KA5678) are HJ9720, JS6201, and ID1234. That is, it is recorded in the respective case data that the case A and the case B are similar cases regarding the age attribute.

Steps 604 and 605 correspond to R =
For each case of {HJ9720, JS6201}, the case A is added to the similar case regarding the age attribute of the case.
This is a process of adding (patient ID 1234).

The above-described similar case network creating apparatus 10
By the processing performed by 7, it becomes possible to immediately specify, for all the cases in the case base, the similar cases regarding each attribute of the case. In this case, ID1234, K
Since A5678, HJ9720, and JS6201 are similar cases regarding the age attribute, these cases constitute a similar case network regarding the age attribute.

Now, let us consider a case where a patient having a remarkable characteristic with respect to a blood pressure value is given to this medical diagnosis system as a problem case. The case searcher 103 is a case base 1
It is assumed that the case 01 is searched, and as a result, the case A similar to the problem case and the blood pressure value attribute is found. Then, by referring to the similar case data regarding the blood pressure value attribute of the case A, all the similar cases regarding the blood pressure value attribute in the case base 101 can be immediately specified. Therefore, the case searcher 103 does not need to search the case base 101 again to find a similar case related to the problem case and the blood pressure value attribute. That is, in this embodiment, case retrieval can be processed at high speed.

Further, according to the present embodiment, not only the above-mentioned blood pressure value attribute, but also similar high-speed case retrieval can be provided not only for the blood glucose level attribute, the white blood cell count attribute but also for any attribute serving as a medical index. That is, the present embodiment is flexible in case retrieval.

There is no limitation that the case name of the similar case must be held by the case data. For example, when the case A, the case B, and the case C are similar cases with respect to a certain attribute X, as shown in FIG. 9, the case base 101 or other storage device indicates that these cases are similar cases. If it is stored in, the same effect can be achieved with a smaller storage amount as compared with the above-mentioned embodiment. In FIG.
For the similar cases relating to the attribute X of the case A in the case base, referring to the set data of the similar cases, the cases A, B, and C are 3
It turns out that this is a case. Similarly, from cases B and C,
It can be seen that similar cases regarding the attribute X are three cases A, B, and C. That is, a set data of similar cases
The data can be stored independently of the case data.

Further, in the above-described embodiment, when a new case is added to the case base, the similar case network creating device performs the similar case network creating process.
This processing may be performed after an appropriate time has elapsed since the new case was added to the case base. For example, 10 new cases
It is conceivable to execute the similar case network creation processing at the stage of collecting the individual cases.

Next, an embodiment in which the knowledge processing device is applied to a power plant will be described. The device configuration of the knowledge processing device according to the present embodiment is similar to that of FIG. 1 described above. The following description also refers to FIG. 1.

The purpose of the knowledge processing device (abnormality diagnosis device) in this embodiment is to measure state quantity data of a plant at a certain point of time by using a plurality of detectors, and if the data has some abnormality, By referring to the anomaly cases accumulated in, the cause of the present plant anomaly and its response are shown. That is,
The case here is a set of measurement data representing the state of the plant to be diagnosed.

In this embodiment, what is stored in the case base 101 of the abnormality diagnosing device is the abnormality case base written in the format shown in FIG. In this embodiment, the date and time when the abnormal case is obtained is given as the case name in FIG. Further, the identification numbers of the detectors are written in the attribute names 1 to n. In the attribute values 1 to n, the values of the data measured by the corresponding detectors are written. In the column of the diagnostic result, the diagnostic result of the plant abnormality such as "turbine trip" or "steam pipe leak" is written. In the measure column, measures to deal with a plant abnormality that has occurred, such as "operation stop" and "steam pipe replacement", are written.

The processing of the similar case network creation device 107 of this abnormality diagnosis device is also similar to that of the above-mentioned medical diagnosis system. That is, when a new abnormal case A is input, an abnormal case B similar to the case with respect to the attribute value of the attribute name i (i = 1, ..., N) is searched from the case base 101, and the abnormal case A is searched. The abnormal case B is additionally recorded as a similar case name related to the attribute name i of, and the abnormal case A is additionally recorded as a similar case name related to the attribute name i of the abnormal case B.

The above similar case network creation device 10
By building a case base 101 using
It enables high-speed search for plant abnormality cases. That is, at some point (for example, periodically), the state quantity data of the plant is collected from the detector corresponding to the attribute name i (i = 1, ..., N) and recorded in the attribute value i. These data are input to the problem analyzer 103 as a problem case. Here, the problem analyzer 103 examines the attribute value i of the attribute name i of the problem case, and if it is outside the predetermined normal range of the attribute value, the set of this attribute name and attribute value Is sent to the case search unit 102. now,
It is assumed that the problem analyzer 103 has determined that the attribute value v of the attribute name s is outside the normal range.

The case retriever 102 receives the problem case data (set of attribute name and attribute value) from the problem analyzer 103, and determines whether there is an abnormal case similar to the problem case and the attribute s in the case base 101. Are searched by successive comparison. As a result, when it is found that the attribute value of the attribute name s of the abnormal case A is similar to the attribute value v of the attribute name s of the problem case, by referring to the similar case column regarding the attribute name s of the abnormal case A, If there is a similar case regarding the attribute name s of the abnormal case A in the case base 101, the case name (for example, abnormal case B) can be extracted. That is, it can be seen that the abnormal case B is also a similar case with respect to the problem case and the attribute name s. Further, by referring to the similar case column regarding the attribute name s of the abnormal case B, similar cases other than the abnormal case A regarding the abnormal case B and the attribute name s can be found in the case base 1
In case of 01, the case name can be taken out.

As described above, by tracing the relationship between similar cases, the attribute s of the problem case is selected from the case base 101.
All similar cases for can be retrieved more efficiently than prior art search methods. As a result, the time required for the case search is shortened, and the performance of the abnormality diagnosis apparatus as a whole is improved.

[0035]

According to the present invention, since the data of the similar case relating to the arbitrary attribute X of the arbitrary case A in the case base can be referred to at high speed, the case existing in the case base. Similar cases for A and attribute X can be identified immediately without further searching. This enables fast and flexible case inspection.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a knowledge processing device according to an embodiment of the present invention.

FIG. 2 is a configuration example of case data according to an embodiment of the present invention.

FIG. 3 is a diagram showing an example of case data of a new case.

FIG. 4 is a flowchart showing a processing procedure in the similar case network creation device shown in FIG.

FIG. 5 is a diagram showing an example of case data in a case base.

FIG. 6 is a flowchart showing a procedure of similarity relationship processing.

FIG. 7 is a diagram showing an example of case data of a new case after the similar case network creation processing.

FIG. 8 is a diagram showing an example of the case data in the case base after the similar case network creation processing.

FIG. 9 is a diagram showing a configuration example of a case base according to an embodiment of the present invention.

FIG. 10 is a general configuration diagram of a knowledge processing device by case-based reasoning.

FIG. 11 is a diagram showing a configuration example of case data in a general case-based reasoning apparatus.

FIG. 12 is a diagram showing an example of hierarchization of case bases.

[Explanation of symbols]

101 ... Case base, 102 ... Problem analyzer, 103 ... Case searcher, 104 ... Case corrector, 105 ... Case repairer,
Reference numeral 106 ... Case storage, 107 ... Similar case network creation device.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Takeshi Yokota, 7-2-1, Omika-cho, Hitachi-shi, Ibaraki Hitachi Energy Co., Ltd. (72) Inventor Yasuhiro Kobayashi 2-chome, Omika-cho, Hitachi-shi, Ibaraki No. 1 Inside the Hitachi Energy Research Laboratory

Claims (10)

[Claims] 1. In a case-based construction method used in a case-based reasoning apparatus for generating a solution of a new problem by using a past problem-solving case, a new problem is stored in a case-base storing past problem-solving cases. When adding a solution case, when at least one past problem solution case similar to the value of the attribute of the new problem solution case exists, data associating the past and the new problem solution case as feature data. A case base constructing method, characterized in that the above case is added and stored in the case base. 2. A method of performing case base inference using accumulated data of a case base constructed by the case base construction method according to claim 1, when an attribute of a problem case to be solved and its attribute value are given, For any attribute, it is searched whether or not the problem case and one of the past problem solution cases stored in the case base have similar attribute values, and the problem has a similar relationship for any attribute. A case-based reasoning method characterized by extracting a set of solution cases and performing inference processing. 3. The case-based reasoning method according to claim 2, wherein a set of cases having similar relations with respect to attributes is taken out, and a solution is given by using knowledge given in advance so as to match a problem case to be solved. A case-based inference method, which comprises correcting, and when the correction fails, using a case different from the above case to repair the solution of the failure case. 4. A case base storing a past problem solution case in a case base construction apparatus of a case base reasoning apparatus for generating a solution of a new problem by utilizing a past problem solution case.
When a new problem solving case is added to a database, when the new problem solving case and the past problem solving case have similar attribute values with respect to an arbitrary attribute, these problem solving cases with respect to the arbitrary attribute are performed. A case-based construction apparatus characterized in that a similar case network creating means is provided for recording the fact that cases are in a similar relationship as characteristic data of each problem solving case. 5. A case-based reasoning apparatus comprising the case-based construction apparatus according to claim 4, means for inputting an attribute of a problem case to be solved and its attribute value, and the problem case and the past problem for an arbitrary attribute. A set of problem solution cases having a similar relationship with respect to the arbitrary attribute is referred to by referring to the characteristic data recorded in the past problem solution case, which has an attribute value similar to one of the solution cases. A case-based reasoning apparatus comprising case search means for performing a process of extracting. 6. The case-based reasoning apparatus according to claim 5, wherein the problem solving means for characterizing the problem case to be solved and the solution of the similar case retrieved by the case searching means are matched with the problem case. And a case correction means for correcting the failure case using the knowledge different from the knowledge when the case correction means fails to correct the solution of the similar case. A case-based reasoning apparatus comprising: a case repairing means for repairing a case and a case storing means for characterizing a new case to be stored in a case base. 7. The case-based reasoning apparatus according to claim 6, wherein the similar case network creating means inputs the case data processed by the case storing means as an input. 8. The case-based reasoning apparatus according to claim 6, wherein the processing of the similar case network creating means is activated at an arbitrary time interval. 9. In the case-based reasoning apparatus according to claim 5, it is possible to refer from each case that a new case and a conventional case have a similar relationship regarding arbitrary attributes. Case-based reasoning device characterized by storing in a different storage device. 10. A case in which a past problem solution case data having an attribute value similar to that of the new problem case data is obtained by searching a case base and a solution of the new problem case is inferred based on the problem solution case. In the base inference method, characteristic data indicating other case data having similar attribute values is added as case data and stored in the case base, and case data having an attribute value similar to the attribute value of the new problem case is taken as an example. When searching from the base, the case data indicated by the characteristic data added to the retrieved case data is extracted, and the case data indicated by the characteristic data of the case data is repeatedly extracted to find related past problem solving case data. A case-based reasoning method characterized by seeking.