JPH0486921A - Method and device for preparing knowledge base for approximate inference - Google Patents

Abstract

PURPOSE:To enable an efficient approximate inference by storing temporarily the knowledge of an expert which indicates a relation between set phenomena and conclusions, calculating the legibility of each phenomenon at every conclusion by using this, and checking the value. CONSTITUTION:The knowledge of the expert is temporarily stored in a knowledge storing device 11, and synthesized in a knowledge synthesizing device 12. Then, a legibility Clij of each phenomenon at every conclusion is calculated by both a still information content arithmetic unit 19 and a legibility arithmetic unit 20. This calculated legibility Clij is applied to a knowledge base restructing device 31, in order to rearrange the value Clij of the legibility in order of the small one after relating it with the phenomenon at every conculusion cj and prepare a legibility table. The process is repeated in order of the small value of the legibility about entire phenomena concerning the resolution cl, and this process is operated for entire resolutions. Like this, the relation between the conculusion and the phenomenon whose legibility is very low (that is, they are not operated effectively for the approximate inference), can be released, and the knowledge stored in the knowledge storing device 11 is arranged.

Description

[Detailed Description of the Invention] A method and apparatus for reconstructing a knowledge base for approximate inference, which is made up of expert knowledge data representing the relationship between summary events and conclusions of an invention, once set. . By using the clarity value of each event for each conclusion or the static information amount value between two conclusions for each event to remove the association between events and conclusions that have very low correlation, knowledge Make the knowledge base concise by removing unnecessary data from the base. This makes it possible to minimize the size of the knowledge base file and obtain approximate inference results in a short time.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for forming a knowledge base for approximate reasoning that uses expert knowledge representing the relationship between events and conclusions.

Prior Art Approximate inference is known in which the inference result is modified or changed depending on the amount of information about the event used to derive the inference result (for example, ZhangHongIIin
AN EXPERT SYSTEM WITH
(INKING INIMAGES', Prepr
ints of' 5econd IFSA Cong
ress.

Tokyo, July 20-25. 1987.
P, 785).

This approximate inference method calculates the amount of information for each event (i.e., information identification ability of the event) using a membership function given to each conclusion for the event, and calculates the inference result (i.e., the probability that the conclusion holds true). The aim is to improve the discriminative ability of inference results by modifying or changing the information based on the amount of information of the event used to draw the conclusion (by multiplying the probability and the amount of information).

In such an approximate reasoning method, a knowledge base consisting of knowledge data representing a relationship between an event and a conclusion is set in advance. Then, by inputting event data and applying this event data to the knowledge base, possible conclusions are derived.

In such a knowledge base, if a certain number of events with a high amount of information are related to a certain conclusion, a highly reliable conclusion can be obtained by inputting several event data,
Efficient event data input is possible. However, if many events with low information content are included and are related to a certain conclusion, it is necessary to enter data for all events, and even if this is the case, it is necessary to enter data for all events, so that a reliable conclusion can be drawn. is not necessarily obtained. For example, this is very inconvenient when it is desired to reach an accurate conclusion in a short period of time, such as when diagnosing a fault.

Furthermore, in the case of a knowledge base created from data input from experts' past experiences, there is a problem in that the input knowledge data varies depending on the individuality and level of experience of the expert.

Summary of the Invention Purpose of the Invention This invention aims to enable efficient approximate inference.
- Aims to provide a method and apparatus for reconstructing a knowledge base once established.

Structure of the Invention 1 Operation and Effect The knowledge base formation method for approximate reasoning according to the present invention temporarily stores expert knowledge representing the relationship between a set event and a conclusion, and uses the stored knowledge to create a knowledge base for each conclusion. Calculate the clarity of the event, check the clarity value of each event for each conclusion, detect the event with a small value of clarity, and disassociate the detected event from the corresponding conclusion. It is characterized by

It is advisable to delete conclusions that are no longer associated with all related events.

The knowledge base forming device for approximate reasoning according to the present invention includes a means for storing expert knowledge representing the relationship between a set event and a conclusion, and a means for storing expert knowledge representing the relationship between a set event and a conclusion. The apparatus includes a clarity calculation means for calculating the clarity of an event, and a means for disassociating an event having a clarity of a small value from a corresponding conclusion for each conclusion.

As an interface with humans, it is preferable to provide a display means for displaying the clarity calculated by the clarity calculating means and an input means for inputting a cancellation confirmation or a cancellation command. In this case, the canceling means operates to cancel the association between the event and the conclusion in response to the input from the input means.

It is also possible to automatically cancel the association between an event and a conclusion. In this case.

The canceling means checks the clarity value calculated by the clarity calculation means, detects an event with clarity equal to or less than a predetermined value, and cancels the association of the detected event with the corresponding conclusion. configured. In addition to the above release means,
It would be good to have a function to delete conclusions that are no longer associated with all related events.

The knowledge base formation method for approximate reasoning according to the present invention temporarily stores expert knowledge representing the relationship between set events and conclusions, and uses the stored knowledge to determine the relationship between two conclusions for each event. For each event, calculate the amount of static information between two conclusions, extract the one with the smallest amount of static information between two conclusions, and calculate the amount of static information between two conclusions related to the amount of static information between the two conclusions. The method is characterized in that the conclusion with the smaller amount of information is disassociated from the corresponding event.

It is a good idea to delete events that are no longer associated with all related conclusions.

The knowledge base forming device for approximate reasoning according to the present invention includes means for storing expert knowledge representing the relationship between set events and conclusions, and using the knowledge stored in the storage means, for each event. , an inter-conclusion static information amount calculation means for calculating the amount of static information between two conclusions, and a conclusion between two conclusions related to a small value of inter-conclusion static information amount for each event with another conclusion. A means is provided for disassociating the conclusion with the smaller amount of static information from the corresponding event.

As an interface with humans, it is preferable to provide display means for displaying the inter-conclusion static information amount calculated by the inter-conclusion static information amount calculation means, and input means for inputting cancellation confirmation or cancellation command. In this case, the canceling means operates to cancel the association between the conclusion and the event in response to the input from the input means.

It is also possible to have a configuration that automatically releases the association between an event and a conclusion. In this case, the canceling means checks the value of the inter-conclusion static information amount calculated by the inter-conclusion static information amount calculation means, and selects two conclusions related to the inter-conclusion static information amount less than a predetermined value. It has the function of extracting.

The canceling means may further have a function of deleting an event whose association with all related conclusions has been canceled.

As described above, according to the present invention, inferences can be made in conclusions that have many associations with events with a low amount of information, and in knowledge that has many associations with events but include many events with a low amount of information. By removing the association between conclusions and events that are not valid, the knowledge base is rebuilt and the minimum necessary relationship between events and conclusions is formed.

Therefore, efficient and reliable approximate inference can be performed in a short time. Furthermore, by performing the reconstruction, the knowledge storage data capacity can be kept to a minimum, and the capacity of the data storage device can be reduced.

Description of Embodiments (1) Overall configuration of approximate inference device FIG. 1 shows the overall configuration of an approximate inference device according to a first embodiment. FIG. 8 shows the overall configuration of an approximate inference device according to the second embodiment. As a common element in both of these embodiments, the approximate reasoning device includes a knowledge storage device 11.
Knowledge synthesis device 121 Post-synthesis knowledge storage device 13. Event value input device 14. Fitness calculation device 15. Dynamic information amount calculation device 1B, possibility calculation device 17. Possibility display device 18. Static information amount calculation device 19, clarity calculation device 20. Clarity storage 21. Clarity addition device 22 and clarity display device 2
It has 3.

The approximate inference device of the first embodiment (FIG. 1) further includes a knowledge base reconstruction device 811, a display device 32, and an input device 33. The approximate inference device of the second embodiment (FIG. 8) further includes a knowledge base reconstruction device 41.display device 42. It includes an input device 43 and an inter-conclusion static information amount calculation device 44.

First, elements common to both embodiments will be explained, and then configurations and operations specific to each embodiment will be explained.

(2) Knowledge storage device The knowledge storage device 11 is a device that stores knowledge manually created by experts and the like in a format that shows the relationship between events and conclusions. This device can store the knowledge of multiple experts.

Two experts exl stored in the knowledge storage device 11
, ex2 knowledge example is shown below in the form of a rule.

Expert exl: if'20≦f'l≦60. 0≦f2≦40゜-, a≦fi≦b,・then cltl
il...(1) if 40≦f1≦80. 60≦f2≦100゜”
” ai2≦f'i≦b12゜then...(2) if-, a, -≦f'i≦b, -then c
jlko 1j...(3) Expert ex2 + H30≦f1≦50.10≦f2≦30°-, a≦
fi≦b, -thenll I...(4) if 50≦fL≦70. 70≦f2≦90゜-
, a ≦f'f≦b12''' then c2 ... (5) if ... , a 0.≦fi≦ b 1., ...13
1 then Cj...(6) The above rules are expressed in the form of a table for each expert as follows.

Table 1 Expert exl fl, f'2. fi is an event, and these are respectively event 1. Event 2. Sometimes called event i. The event index i takes a value in the range 1 to m. cl, c2. cj
are the conclusions, and these are respectively the conclusions 1. It is sometimes called conclusion 2゜conclusion j. The subscript j of the conclusion takes a value in the range of 1 to n.

Also, al, ≦fi≦b1. al expressed as,
.

1.1 1,1
1Jb, are called the minimum value and maximum value, respectively.

J Expert ex2 Table 2 (3> Knowledge synthesis device The knowledge synthesis device 12 is a device that synthesizes the knowledge of a plurality of experts stored in the knowledge storage device 11 and combines it into one knowledge.

There are various ways to synthesize knowledge, but here we calculate the average value and standard deviation of multiple experts for the maximum and minimum values of each event involved in each conclusion.

The knowledge synthesis process will be described below, taking as an example the knowledge of the two experts described above that leads to the conclusion c1 from the event f1.

From the above rules (Equations (1) and (4)), event 1
The rule for obtaining conclusion 1 (cl) from (fl) is extracted as follows.

Expert ext: if 20≦f1≦80 th
en cl-(7) expert ex2: if 30
≦rl≦50 then cl-(8) Average value m of the minimum value and average value 1 of the maximum value! In m is calculated.

+5aX 20+ 30 = 25. ,, (9゜m,
■iln 2 ” n+ax 80+50 (10) Standard deviation σ of the minimum value, and.

111 mouths The deviation σ is calculated.

a + a
For equations (1) to (6), if all the minimum and maximum values of each event involved in each conclusion are examined, a table as shown below is obtained.

(Left below) Table 3 Generally, in approximate reasoning, a membership function is given to an event. Here, as an example, a method for determining membership functions using Gaussian distribution using expert knowledge synthesized as described above will be described.

Average value of the minimum value m ,, average value of the maximum value mmax'l1
Using the standard deviation ln of the minimum value σ, and the standard deviation of the maximum value 11n σ, the membership function is calculated by the following formula IIa
x

...(13) TachiX: Value of input data to the event Φ(X) Degree to which two-person data X fits the event (degree of fit) Gauss(x): Value of Gaussian distribution at input X .

Figure 2 shows an example of a Gaussian distribution. Only the left half of this Gaussian distribution is used to create the membership function. The position of X at Φ(x)-0,5 is m,
or m determined by ff1ln
IIaX, the slope is σ or σ
determined by 111n 1
NaXru.

As an example, the membership function for determining the conclusion c1 from the event f1 is created as shown in FIGS. 3a to 3c using the values calculated by equations (9) to (12). In this case, equation (13) becomes as follows.

...(14) Figure 3a is the first right-hand side of equation (13) or equation (14).
Figure 3b shows the second term on the right side of equation 13) or equation (14), and Figure 3C shows the result of subtracting the second term from the first term, that is, equation (13) or equation (14). It represents the membership function expressed by equation (14).

Each event fl, f'2. created based on the synthesized knowledge shown in Table 3. For fi, the conclusion cL, c2
．． Examples of membership functions for determining cj are shown in FIGS. 4a, 4b and 4c.

(4) Post-synthesis knowledge storage device The post-synthesis knowledge storage device 13 stores the average value and standard deviation calculated by the knowledge synthesis device 12 in the format shown in Table 3. Since it is not necessary to synthesize knowledge every time an inference is made, the results calculated in advance in this way are memorized. By reading out and using the values in the storage device 13 each time an inference is made, it is possible to speed up the inference process.

(5) Event value input device The event value input device 14 is a device that reads input data input for each event from a keyboard, communication interface device, memory, file, etc. The input data is given to the fitness computing device 15, and the data of each event or information on whether or not it has been input is given to the clarity adding device 22.

The event value may be given not only as a definite value but also as a linguistic value or a membership function.

(6) Fitness calculation device The fitness calculation device 15 calculates the fitness of the data input from the event value input device 14 for each membership function (or conclusion). Specifically, when the input data is a definite value, the goodness of fit can be calculated by substituting the input data as the variable X on the right side of equation (13).
It is required as. Of course, such an arithmetic expression does not necessarily have to be used. If the input data is a linguistic value or a membership function, the goodness of fit may be calculated using, for example, the old N-MAX operation.

(7> Dynamic information calculation device and static information calculation device event r
The event value of event r2 (input data) is Xl, and the event value of event r2 is x2. Let the event value of event rj be xi. These data are input from the event value input device 14.

As shown in FIGS. 5a and 5b, each degree of fitness 1112
° lj 21 22°m2j is determined as follows.

m11: Degree of suitability of input data x1 to the membership function of conclusion c1 m12: Degree of suitability of input data x1 to the membership function of conclusion c2 mLj ' Degree of suitability of input data x1 to the membership function of conclusion cj m21: Human data x2 The goodness of fit of the input data x2 to the membership function of the conclusion C1 m22: The goodness of fit of the input data x2 to the membership function of the conclusion C2 m2j' The goodness of fit of the input data x2 to the membership function of the conclusion cj Generally, as shown in FIG. 5c, goodness of fit m1. ]J is the fitness of the input data (event value) xl to the membership function of the conclusion cj.

These fitness degrees, i, are input data xi, x2. xi
It is calculated by the fitness computing device 15 when a given value is given.

Here we consider the concept of fuzzy skewer entropy.

Fuzzy entropy Ef when input xi is given
Define i as follows.

Eft--,: + (m, / M, ) IOg (
m, /M, ) 1...(15) where M, -Σ m,,
...(16) 1 no + lj This fuzzy ψ entropy is a type of index of information discrimination ability, and when input data xi is given, the value is small enough that the conclusion can be clearly identified, and the conclusion is only ambiguous. The value is so large that it cannot be identified.

The range of values that the fuzzy entropy Ef can take is as shown below.

0≦Ef≦log(n) n: number of conclusions on the event Next, using this fuzzy entropy Eft, the dynamic information amount If'i when input data xi is given
Find (xi). Here, the dynamic information amount 1fiD(y;
i) is the ability to identify events that determine a conclusion when making inferences.

Therefore, the dynamic information amount 1ftp(xt
) from the maximum fuzzy entropy, input data x1
is defined as minus the fuzzy entropy Eft when is given.

IfD(xi)=log(n)+,X l(mlj/
M, ) log (mij/M1) 1...(1
7) The dynamic information amount of event f1 to which input data x1 is not given becomes zero.

The dynamic information amount calculating device 16 calculates the dynamic information amount for each event according to equation (17) using the degree of suitability obtained by the degree of suitability calculating device 5.

The amount of dynamic information depends on the input data xi as described above.

On the other hand, the amount of static information does not depend on the input data, and the static information of the entire event is calculated by subtracting the average fuzzy Φ entropy within the range of the event from the maximum fuzzy entropy. Quantity. The amount of static information about event i is expressed by the following formula % mij (xik): Input data Xl about event fi
When calculating the fitness static information amount for the membership function of the conclusion cj of k, as shown in Figure 4c,
The range width x1□~X l p of the event fl is set at intervals of δ (p
-1) It is divided into pieces. (1/p) of equation (18) [
>Th , , ] represents an operation for calculating the fuzzy entropy for each xik and finding their average.

As can be seen from equations (18) and (19), the greater the overlap between the event membership functions.

The amount of static information of an event is small, and the smaller the overlap between the membership functions of an event, the larger the amount of static information of an event.

In other words, the amount of static information indicates the ability of an event's membership function to identify a conclusion.

The static information amount calculating device 19 calculates and stores the amount of static information for each event from the membership function obtained from the two combined knowledges according to the above-mentioned equations (18) and (19). do. Since the static information amount does not depend on input data, it only needs to be calculated once.

(8) Possibility calculation device For each conclusion, calculates the amount of information of events such that the sum of the amount of information of events involved in that conclusion is 1, and the relative strength of the amount of information of those events does not change. . This calculated amount of information is called a weight.

For example, if the amount of dynamic information described above is used, the weight wi for each event fi will be as follows.

wi −1f'i (xi)/ , )RIffD(xi
) - (20) Next, calculate the product of these weights and fitness.

The total sum for each conclusion is calculated as the probability Pr of each conclusion cj.

For example, in the above example, Pr, −Σ (vix m,,)
−(21)j +−+
The IJ possibility calculation device 17 performs the above-mentioned calculations to calculate the probability for each conclusion.

(9) Possibility Display Device This possibility display device 18 displays the possibilities calculated by the possibility calculation device 17 for each conclusion. The possibility may be displayed for all conclusions, or one or more highly probable conclusions may be displayed. Also, the possibility may be transmitted to another device through one communication, or the possibility may be stored in a memory or a file.

(10) Clarity calculation device The clarity calculation device 20 is a device that calculates the clarity of each event for each conclusion. Here, the clarity of each event for each conclusion indicates the relative discernment ability of each event when evaluating the possibility of a certain conclusion. Therefore, this clarity makes it possible to compare the discriminative ability of multiple events in order to establish a certain conclusion, and to understand which event has a high discriminative ability (contains a large amount of information). Ru.

The clarity of each event for each conclusion is determined by the following formula.

however.

Ao, -Σ　m, (xi) 1 piece k-113k Namely.

...(23) If the membership function of the conclusion cj exists in the event fi, A,, > O j If the membership function of the conclusion cj does not exist in the event ri, A,, -0 J Each event for each conclusion The denominator of equation (22) expressing the clarity of is the membership function of the conclusion cj or the existing event fi
The total sum of the static information amount 1fxs is calculated only for .

Since clarity is obtained by normalizing the amount of static information using equation (22), it is easy to intuitively understand the relative discrimination ability.

In this way, in the clarity calculation device 20.

The clarity of each event is calculated for each conclusion.

(11) Clarity Storage Device The clarity storage device 21 is a device that stores the clarity of each event for each conclusion calculated by the clarity calculation device 20.

Clarity calculations do not need to be performed every time an inference is made. Therefore, the clarity calculated when knowledge is synthesized is stored in the clarity storage device 21.

The value stored in the clarity storage device 21 is read out every time an inference is made. This makes it possible to speed up the inference processing.

(12) Clarity Addition Device The clarity addition device 22 is a device that calculates the clarity of an event to which data is actually input for each conclusion. here,
for the inferences that are actually made.

Take the summation for each conclusion of the clarity of the events for which data has been entered. The sum of this clarity indicates the clarity of the inference result. It can be said that the higher the clarity, the greater the amount of information for deriving the inference result (conclusion). Therefore, clarity can be used as an indicator to judge the reliability of the inference result itself.

The clarity for the inference results (that is, for each conclusion) is calculated according to the following formula.

cn, -1 child + C11J...
(24) However, in equation (24), input data xi
The clarity of the event fi for which Cρ5. is treated as O. That is, the addition of equation (24) is performed in row J only for the input data xj or the clarity cn of the given event.

What is the possible range of clarity C1 of the inference result?

0.0≦Cg, ≦1.0. In other words, if you make an inference by inputting data about all the events that can be used to draw a certain conclusion within the knowledge given before making the inference,
The clarity of the conclusion will be 1.0. Also, only some of the events that can be used to draw a certain conclusion.

If the data is entered, clarity will be a value between 0.0 and 1.0. At this time, it can be said that if a large number of events with high clarity are used among the events that can be used, the clarity of the conclusion will also be increased, and a highly reliable inference result can be obtained.

(13) Clarity Display Device The clarity display device 23 is a device that displays the clarity of the inference result (-the possibility mentioned above as an example) calculated by the clarity addition device 22. The clarity may be displayed along with the inference results, or the clarity may be transmitted to another device or stored in memory or a file.

This clarity display displays all conclusions of the inference results. Therefore, if there are multiple conclusions, the clarity corresponding to each conclusion is displayed.

In this way, every time data is entered.

By calculating the information amount of the event to which the input data belongs and displaying the clarity of the inference result, the user can judge the reliability of the inference result.

(14) First Embodiment Referring to FIG. 1, the approximate inference device of the first embodiment includes a knowledge base reconstruction device 311, a display device 32, and an input device 33.
It is equipped with In the first embodiment, the knowledge of experts is once stored in the knowledge storage device 11, and after being synthesized in the knowledge synthesis device 12, the static information amount calculation device 19 and the clarity calculation device 20 are used to calculate the results for each conclusion. Clarity of each event Cρ1j
is calculated. This calculated clarity Cg1. By giving J to the knowledge base reconstruction device 31, unnecessary pieces of knowledge stored in the knowledge storage device 11 and knowledge synthesized by the knowledge synthesis device 12 are deleted by the knowledge base reconstruction device 31, and these The knowledge base will be rebuilt.

The knowledge base reconstruction device 31 receives the clarity value cn, , of each event for each conclusion from the clarity calculation device 20, and converts these into a clarity value for each conclusion cj, as shown in FIG. CI1. Create a clarity table by associating and arranging J with events in descending order.

In FIG. 6, the horizontal alignment represents the conclusion.

The vertical arrangement represents the order of magnitude J of the clarity values CD, . In each column, the upper fi(cj) represents the event f1 related to the conclusion cj, and the lower number (0
, 034, 0,076, etc.) indicates the clarity value CD, . The number of events in the second column from the bottom represents the number of events associated with each conclusion. The bottom column shows the sum of clarity values, and since clarity is normalized for each conclusion, the total value for each conclusion is always 1.

The knowledge base reconstruction device 31 refers to the clarity table shown in FIG. 6 and performs the processing shown in FIG. 7.

First, the following processing is performed for the first conclusion C1. It is determined whether the smallest clarity value (0.006 in FIG. 6) is less than a predetermined value (for example, set to about 0.01) (step 53), and if so, in the knowledge storage device 11, The association between the event having a clarity value less than or equal to a predetermined value (fl(cl) in FIG. 6) and the conclusion c1 is canceled (step 54).

To disassociate event r1 from conclusion CI means, for example, in the above-mentioned equation (1) or equation (4), knowledge regarding event f1 regarding conclusion c1 is 20≦r1≦6.
This means to delete 0 or 30≦f1≦50.

In equation (1) or equation (4), conclusion CI and other events f2. The relationship with f'i, etc. is maintained as is. If the event f1 is related only to the conclusion C1 and has no relation to other events, the event f1 itself may be deleted. Unless the clarity value is less than or equal to a predetermined value, the association between the conclusion and the event is not removed.

The processes of steps 53 and 54 are repeated for all events related to conclusion c1 in descending order of clarity (
Step 52).

When the processing of steps 52 to 54 is completed for the conclusion C1, the same processing is performed for the conclusion C2.

In this way, steps 52 to 54 are performed in order for all conclusions (step 51).

By repeating the above-mentioned process of canceling the association between the conclusion and the event, there may be a conclusion in which all the related events are disassociated from the conclusion. Conclusions that are no longer associated with all such events are then deleted from the knowledge storage 11 (steps 56 and 57).

This process is performed for all conclusions (step 55).

As described above, the association between a conclusion with very low clarity (that is, it does not work effectively for approximate reasoning) and an event is canceled, and the knowledge stored in the knowledge storage device 11 is organized.

In the process of canceling the association between a conclusion and an event.

Displaying an event name having a clarity value smaller than a predetermined value on the display device 32 together with its clarity value, and/or displaying a message that the event should be canceled because its clarity value is smaller than the predetermined value. displayed on the device 32.

The association between the event and the conclusion in the knowledge storage device 11 may be canceled in response to a human input of cancellation confirmation or cancellation instruction from the input device 33.

The same applies to the conclusion deletion process.

In the process shown in FIG. 7, the cancellation of the association between an event and a conclusion or the deletion of a conclusion is carried out for the knowledge in the knowledge storage device 11, but it is also done for the post-synthesis knowledge stored in the post-synthesis knowledge storage device 13. You can.

When the knowledge (knowledge base) stored in the knowledge storage device 11 is organized as described above, the knowledge base reconstruction device 31 starts the knowledge synthesis device 12 to resynthesize the organized knowledge. Let it process. This recombining process is performed in the same manner as the above-described combining process. The knowledge resynthesis command may be output automatically after the knowledge base reconstruction device 31 completes the processing shown in FIG. 7, or may be output in response to human input from the human-powered device 33. .

The resynthesized knowledge is stored in the post-synthesis knowledge storage device 13 and is also provided to the static information amount calculation device 19, where the static information amount and clarity are recalculated.

Then, if necessary, the clarity after the recalculation is given again to the knowledge base reconstruction device 31, and the knowledge base reconstruction process shown in FIG. 7 is performed again. It goes without saying that this knowledge base reconstruction process can be repeated as many times as necessary.

In this way, a reliable knowledge base is finally obtained.

After the knowledge base is determined by the knowledge base reconstruction process, approximate inference calculations including relevance, dynamic information amount, possibility of conclusion, clarity addition, etc. are performed, or in response to input event data.

(15) Second Embodiment Referring to FIG. 8, the approximate inference device of the second embodiment includes the knowledge base reconstruction device 412 display device 42. It includes an input device 43 and an inter-conclusion static information amount calculation device 44. In the second embodiment, the concept of inter-conclusion static information amount is used for knowledge base reconstruction.

As mentioned above, the amount of static information regarding event 1 is
), and is calculated using a membership function to derive all conclusions related to the event 1.

On the other hand, the inter-conclusion static information amount Iri is calculated according to a 9th order equation using a membership function for deriving two conclusions cs and ct related to a certain lt event l. Here, s and t are subscripts specifying the conclusion, and take values of 1 to n.

...(25) If there are three or more conclusions related to event i, the amount of inter-conclusion static information regarding that event i is equal to half the number of combinations of two conclusions among those conclusions. can get.

Also the amount of static information between conclusions. The larger the overlap between the membership functions of two conclusions, the smaller the value (the smaller the amount of information), and the smaller the overlap, the larger the value (the larger the amount of information).

The inter-conclusion static information amount calculation device 44 uses the synthesized knowledge obtained from the knowledge synthesis device 12 to calculate the inter-conclusion static information amount for each event and creates an information amount table as shown in FIG. Create. FIG. 9 is an information amount table for event 1, and conclusions cl, c3. In the case where cs and et are related, it shows the amount of inter-conclusion static information when two of these conclusions are combined (and the inter-conclusion static information between conclusion C3 and eS). (The static information amount Ifi3S and the inter-conclusion static information amount If1S between the conclusions c1 and eS are not shown). Such an information amount table is given to the knowledge base reconstruction device 41.

The knowledge base reconstruction device 41 refers to the information amount table and executes the process shown in FIG. 10.

The inter-conclusion static information amount for a certain event fi is taken out from the information amount table, and it is determined whether each inter-conclusion static information amount is less than a predetermined value (steps 62, 63). If there is an inter-conclusion static information amount that is less than a predetermined value, it is determined which of the two conclusions related to the inter-conclusion static information amount should be deleted. To do this, for each of the two conclusions, calculate the total value of the inter-conclusion static information amount that the conclusion is related to.Step B4) 1) For the conclusion with the smaller total value, associate it with the event fi in the knowledge storage device. 11
(step 65.86).

For example, in FIG. 9, if the static information amount Iff between the conclusions cs and ct is determined to be smaller than a certain value, then the sum IF of the static information amounts between other conclusions to which the conclusion cs is related, The total IP of other inter-ring static information amounts to which the conclusion ct is related is calculated. That is, the conclusion cs
The amount of static information between other conclusions related to is Ifi and l r
] 3 S, so IF8- If, + lf
It becomes f3Ss. Similarly, IF -lN1t+1fi3t. If IP > IP, the association s t of the conclusion et with the event "i" is canceled, and if IP < IFt, the association of the conclusion aS with the event fi is canceled.

The above processing is performed sequentially for events r1 to rol (step 61).

Finally, it is checked whether there are any events that are no longer associated with all the conclusions that they were associated with, and if so, they are deleted from the knowledge storage 11 (
Step 68.69).

As described above, the association between one of the two conclusions with a very low inter-conclusion static information amount (that is, it does not work effectively for approximate reasoning) and the event is released, and the knowledge storage device 11 The knowledge stored within is organized.

In the process of canceling the association between an event and a conclusion.

Two conclusions with an inter-conclusion static information amount smaller than a predetermined value and the sum of other inter-conclusion static information amounts to which these two conclusions are respectively related are displayed on the display device 42, and/or a certain conclusion's The amount of static information between the conclusions and the above-mentioned total are small, so the display device 42 displays that the cancellation should be made, and in response to two people confirming the cancellation or inputting a cancellation command from the input device 43, the knowledge storage device 11 The association between the conclusion and the event may be canceled.

The same applies to event deletion processing.

In the process shown in FIG. 10, the cancellation of the association between a conclusion and an event or the deletion of an event is performed for the knowledge in the knowledge storage device 11, but it is now done for the post-synthesis knowledge stored in the post-synthesis knowledge storage device 13. You can.

When the knowledge (knowledge base) stored in the knowledge storage device 11 is organized as described above, the knowledge base reconstruction device 41 starts the knowledge synthesis device 12 to resynthesize the organized knowledge. Let it process. This recombining process is performed in the same manner as the above-described combining process. The knowledge resynthesis command may be output automatically after the knowledge base reconstruction device 41 completes the processing shown in FIG. 10, or may be output in response to human input from the input device 43. .

Then, if necessary, the synthesized knowledge is given again to the inter-conclusion static information calculation unit 44, and in the same way, the inter-conclusion static information amount is calculated for each event, and this information amount data is used to reconstruct the knowledge base. The knowledge base reconstruction process shown in FIG. 10 is performed again. It goes without saying that this knowledge base reconstruction process can be repeated as many times as necessary.

In this way, a reliable knowledge base is finally obtained.

After the knowledge base is determined by the knowledge base reconstruction process, the reconstructed (resynthesized) knowledge is stored in the post-synthesis knowledge storage device 13 and given to the static information amount calculation device 19, and the static information amount and A clarity calculation is performed.

Also, based on the knowledge stored in the post-synthesis knowledge storage device 13, the degree of fitness is determined by the devices 15, 16, 17, 22, etc.
Approximate inference operations including dynamic information content, conclusion probability, clarity addition, etc. are performed in response to input event data.

It goes without saying that each of the above-mentioned devices 11-23, 31-33, 41-44 can be realized by a computer including a memory and a display device. For example, the knowledge synthesis device 12
．． Various calculation devices 15.16.17.19.20.223
1, 41.44 is a CP that operates according to the program
This is suitably realized by U.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the overall configuration of an approximate inference device according to a first embodiment. FIG. 2 is a graph showing a Gaussian distribution. Figures 3a to 3c are graphs showing how membership functions are formed. Figures 4a, 4b and 4c are graphs showing the membership functions obtained for each event. FIGS. 5a, 5b, and 5c are graphs showing how the goodness of fit is determined. FIG. 6 is a diagram showing a clarity table. FIG. 7 is a flow chart showing the knowledge base reconstruction process. FIG. 8 is a block diagram showing the overall configuration of an approximate inference device according to the second embodiment. FIG. 9 is a diagram showing an information amount table. FIG. 10 is a flow chart showing the knowledge base reconstruction process. 11...Knowledge storage device. 12...Knowledge synthesis device. 13... Post-synthesis knowledge storage device. 14...Event value input device. 15...Fitness calculation device. 16...Dynamic information amount calculation device. 17... Possibility calculation device. 18... Possibility display device. 19... Static information amount calculation device. 20... Clarity calculation device. 21... Clarity memory. 22... Clarity addition device. 23... Clarity display device. 31, 41...Knowledge base reconstruction device. 32, 42...Display device. 33, 43... Input device. 44... Inter-conclusion static information amount calculation device. Fig. 2mmim m1nsu Fig. 4a Event f Fig. 4b Event f2 Fig. 4c Fig. East f Fig. 5a Akizuki Fig. 5b Fig. Image f2 Fig. 5C Event f Fig. Fig. ce] D Fig. 9

Claims (14)

[Claims] (1) Temporarily memorize the expert knowledge that represents the relationship between set events and conclusions, use the memorized knowledge to calculate the clarity of each event for each conclusion, and calculate the clarity of each event for each conclusion. A knowledge base formation method for approximate reasoning that detects events with a small value of clarity by checking the values, and disassociates the detected events from the corresponding conclusion. (2) The method for forming a knowledge base for approximate inference according to claim (1), wherein a conclusion that has been disassociated from all related events is deleted. (3) means for storing expert knowledge representing the relationship between set events and conclusions; and clarity calculation means for calculating the clarity of each event for each conclusion using the knowledge stored in the storage means. , and means for disassociating, for each conclusion, an event with a small value of clarity with a corresponding conclusion. (4) A display means for displaying the clarity calculated by the clarity calculation means, and an input means for inputting a cancellation confirmation or a cancellation command, and the cancellation means responds to the input from the input means. The knowledge base forming device for approximate inference according to claim (3), wherein the association between an event and a conclusion is canceled by using the method. (5) The cancellation means checks the clarity value calculated by the clarity calculation means, detects an event with clarity equal to or less than a predetermined value, and releases the association of the detected event with the corresponding conclusion. The knowledge base formation device for approximate inference according to claim (3). (6) Claim (5), wherein the canceling means further deletes the conclusion whose association with all related events has been canceled.
A knowledge base formation device for approximate inference described in . (7) A claim further comprising means for synthesizing multiple pieces of knowledge regarding the same event and the same conclusion (
The knowledge base forming device for approximate inference described in 3) to (6). (8) Once the expert's knowledge representing the relationship between set events and conclusions is memorized, the memorized knowledge is used to calculate the amount of static information between two conclusions for each event, and for each event, Extract the two conclusions that have a smaller amount of static information between them, and respond to the conclusion that has a smaller amount of static information between them and the other conclusion among the two conclusions related to the amount of static information between the conclusions. A knowledge base formation method for approximate inference that removes associations with events. (9) The knowledge base formation method for approximate inference according to claim (8), further comprising deleting events that are no longer associated with all related conclusions. (10) A means for storing expert knowledge representing the relationship between a set event and a conclusion, using the knowledge stored in the storage means to calculate the amount of static information between two conclusions for each event. The inter-conclusion static information amount calculation means to be calculated, and for each event, the 2
A knowledge base forming device for approximate inference, comprising means for disassociating a conclusion with a smaller amount of inter-conclusion static information from a corresponding event among two conclusions. (11) A display means for displaying the inter-conclusion static information amount calculated by the inter-conclusion static information amount calculating means, and an input means for inputting a cancellation confirmation or a cancellation command, and the cancellation means is configured as described above. The knowledge base forming device for approximate reasoning according to claim 10, wherein the knowledge base forming device for approximate reasoning cancels the association between the conclusion and the event in response to an input from the input means. (12) The canceling means checks the value of the inter-conclusion static information amount calculated by the inter-conclusion static information amount calculation means, and selects two conclusions related to the inter-conclusion static information amount less than a predetermined value. The knowledge base forming device for approximate inference according to claim 10, wherein the knowledge base forming device extracts. (13) The canceling means further deletes the event whose association with all related conclusions is canceled.
2) A knowledge base forming device for approximate inference. (14) The knowledge base forming device for approximate inference according to claims (10) to (13), further comprising means for synthesizing a plurality of pieces of knowledge regarding the same event and the same conclusion.