JPS62274345A - Fuzzy inference system - Google Patents

Abstract

PURPOSE:To improve the throughput with good man-machine properties by performing inference only by known event data, and inquiring of a system operator and handling the known and unknown events so that maximum effect is obtained only when a sufficient likelihood ratio is not obtained. CONSTITUTION:An inference is performed by using events which are known to the system so as to improve the likelihood ratio of events that the operator of the system inputs up to a target value. Then it is decided whether or not the likelihood ratio of events to be decided about truth is sufficient. If the likelihood ratio is not sufficient, events which contribute to the truth decision making most among events whose likelihood ratios are not known are calculated and when there is an unknown even which contributes to the inference, the operator of the system is inquired about its likelihood ratio. The inference is carried on based on input values and the processing is repeated. Then, the inference result is outputted on a monitor and the processing is completed when there is no event which contributes to the inference or when the specified likelihood ratio is reached.

Description

[Detailed Description of the Invention] 3. Detailed Description of the Invention [Field of Industrial Application] The present invention relates to an inference method in which vague knowledge is used to determine the establishment of a specific event in cooperation with humans, and in particular, This paper relates to a fuzzy inference method using fuzzy knowledge suitable for knowledge-based systems.

[Conventional technology]

MY as a system that makes inferences using ambiguous knowledge
CIN (E, l(,5hortliffe: Medical computer system: Bunkoei), 5PERIL
(Mitsuru Ishizuka: Expert system for building damage assessment: Transactions of the Information Processing Society of Japan, 24.3).

In all of these systems, when the truth value of an unknown event is needed during inference regarding the establishment or failure of a specific event, an inquiry is immediately made to the system operator.

[Problems to be solved by the invention]

In this method, although the system can infer the specific event that it wants to determine whether it holds true or not using only known events, if an unknown fact is needed during the inference, the system operator is contacted. There are drawbacks.

Additionally, in order to avoid unnecessary and troublesome questions, 5PER provides a skip path for rules that are unlikely to be activated, but this is completely unnecessary. It only omits the questions that are difficult to answer.

In this way, in conventional systems that perform fuzzy inference, when an unknown event appears during inference even though sufficient confidence can be obtained using only known events, this is immediately queried. , This is caused by inferences being made without any prediction of how important a certain event is to the inference that determines whether a particular event is true, resulting in inefficient inferences that are troublesome for system operators. There were problems such as being bothered by questions, which caused the system's low processing efficiency and poor man-machine performance.

An object of the present invention is to provide an ambiguous inference method that has high processing efficiency and excellent man-machine performance in a system that uses ambiguous knowledge to determine whether a specific event holds true or not.

[Means for solving problems]

The problems with conventional fuzzy inference systems are: (1) During inference, when a truth value regarding an unknown event is needed, the operator of the system is immediately asked, which is troublesome.

(2) Some of the inquiries made to the system operator are unnecessary, and the results of the inquiries have little effect on the inference, so the inquiries are not effective.

There are two points.

Regarding the first problem, even if the truth value of an unknown event is needed during inference, the inference can be made without using knowledge related to that event, and the confidence of the obtained result is sufficient. If this is not the case, the problem can be solved by restarting the inference when the truth value of the unknown event is needed and proceeding with the inference while making inquiries to the system operator.

Regarding the second problem, we ask the system operator the truth value of an unknown event, assume that the value is obtained, and then predict how much it will affect the certainty of the event that we want to judge as true. However, it can be resolved by consulting the system operator based on the event that will yield the greatest effect.

Therefore, in the present invention, the above-mentioned 1. It is characterized by a method that combines solutions for the second problem and uses vague knowledge to perform efficient reasoning.

[Effect]

The present invention makes inferences as automatically and effectively as possible, and by effectively querying system operators when automatic inferences are difficult.

A fuzzy inference method with high processing efficiency and excellent man-machine performance can be obtained.

(Example〕 An embodiment of the present invention will be described below with reference to FIG.

FIG. 1 is a diagram showing an overview of the processing of the fuzzy inference system 11 of the present invention.

The ambiguous inference system 1 of the present invention uses the ambiguous knowledge base 2 to say, ``If OO, then ΔΔ, and the confidence level of this knowledge is XX''. The ambiguous rule is read out, and this knowledge, the event data stored in the event database 3 such as "the confidence level of this event is oo", and the "confidence level of this event" obtained by querying the system operator 4 Based on the data related to the event such as 071, the system operator 4 determines whether the event specified by the system operator 4 has occurred, and outputs the determination result to the monitor 5.

FIG. 2 shows the processing flow of the fuzzy inference system of the present invention.

First, initialize the device and initialize the program data area (process 6), then read the ambiguous rules from the ambiguous knowledge base (process 7), and read data regarding the certainty of the event from the event database. (Processing 8
).

In process 9, the system operator is asked about the name of the event for which the event is to be determined to be true and the degree of certainty required to determine that it is true.

Therefore, in order to increase the confidence level of events input by the system operator to the target value, we first make inferences using events that are generally known to the system (events whose confidence level is known).
Process 10) As a result, it is determined whether the certainty of the event to be determined to be true is sufficient. Process 11), if it is not sufficient,
Calculatively process unknown events with a high degree of certainty that contribute most to the determination of validity 12). If there is an unknown event that contributes to the inference, ask the system operator about its degree of certainty (process 14) and input it. Based on the value, inference is continuously repeated from process 15) and process 11. If there is no contributing event or if the specified certainty is satisfied, the inference result is output to the monitor (step 16), and the process ends.

The features of this system include an inference method that first performs ambiguous inferences using only known event data, and then queries the system operator only if sufficient confidence cannot be obtained; It deals with the certainty of events and the certainty of unknown events. In order to explain this characteristic processing method, the rule expression used in the fuzzy inference method of the present invention will be explained first, and then the inference method will be explained.

FIG. 3 is a diagram showing a rule expression format.

/lz-ru4L IF-THEN-WITH~ (7) Represented in the format.

If the condition part 21 is satisfied, it is the conclusion part 22, which indicates that the correctness of this rule is a confidence level of 23. For example, rule 24 is XI, Xs, and...xn.
If (condition part 25) is satisfied, it is Y (conclusion part 26), and the confidence level 27 of this rule is.

It shows that it is 0.9. xl...Xn, Y indicate events, and each has a degree of certainty.

As explained above, a rule consists of a condition part consisting of the logical product of multiple events, a conclusion part consisting of one event, and a rule consisting of 0 or more and 1
It is composed of confidence factors that take the following values, and this rule is used to determine the occurrence of a specific event.

With reference to FIG. 4, a description will be given of how the confidence level is calculated in the inference process.

Rule 31 has 1 condition part: XI, xz,...Xn
, the conclusion part is T, and the confidence level is 2. We will show how the confidence level of Y is calculated using this rule.

Step 1: Find the confidence level of each event in the conditional part.

It is assumed that the confidence of Xi, ~xIl is known, and its value can be obtained using the function CF as CF (XI), .=, CF (Xn).

Step 2: Find the confidence level of the conditional part (the degree to which the conditional part is satisfied). The confidence level of each event in the conditional part is obtained in step 1 as CF (Xz), ..., CF (X-), and the confidence level of the conditional part is determined based on these. Each event in the conditional part is ANDed using a function f and that takes this into account.

As s fund to find the certainty of the conditional clause, sin
Use functions.

Step 3: From the certainty of the condition part obtained in the previous step and the certainty of the rule itself, calculate the certainty of the conclusion part using the function frule.
frul, e as frule Cxe y)
=x-y is used.

Step 4 Up to step 3, event X1. X2. .

..., the confidence level of event Y when inferred using -Xn is obtained. Using this value and the confidence factor CF (Y) of Y, which has already been obtained using another rule, as a combination function f comb, the confidence factor of Y is updated. l
As fcomb, f comb(xn y)=x+
Use y-xjy.

In the above calculation procedure, the confidence level of Y before inference using this rule is raised by using the rule based on the new events x1 to Xn.

This confidence level further influences the confidence level of higher-ranking events.

Next, the method of predicting the maximum confidence value of each event in the system of the present invention and the search method using the same will be explained using the graph shown in FIG. 6, assuming that the specific event whose occurrence is to be determined is A.

Each hypothesis has a confidence value and an expected maximum confidence value, which are called an af value and an up-to-af value, respectively.

Which rule should be selected for inference depends on this af value, u
Determine based on the p-to-af value.

For example, in order to infer A, there are rule 1 and rule 2, and when each rule is used.

Calculate the maximum increase in the confidence level of A.

The inference is proceeded by preferentially using the rule with the maximum value.

This initial value is set to 1, which explains how to manage up-to-af values. Up-to-up of events in the event database and events inquired of the operator
The af value is made equal to the af value. In other words, since these events are root events, there can be no increase in confidence, u
The p-to-af value is updated by 1, for example, if the af values of events F and H are each 0.4 . 0.5, and a of events G and T
When the f value is unknown, that is, it is an unknown event, the maximum confidence when inferring B using Rule 3 is 0.7Xmin (
up-to-cf (F), up-to cf (G))
= 0.28, and the maximum confidence when inferring B using Rule 4 is 0.8Xmin(up-to-cf(H), up-to
Since -cf(I))=0.40, it becomes 0.588, and up-to-af(11)=0.588.

This up-toaf management, other than setting the initial value to 1, suffices if it is performed during the search, and there is no need to calculate it for all nodes on the graph at once.

In the first stage of inference using only known phenomena, the above
Inference is made using the rule selection method as a rule selection criterion for the vertical search method.

In the second stage of reasoning while inquiring about unknown events, unlike the first stage search method, the first stage search method is redone every time the confidence level of A is obtained.

According to this embodiment, it is possible to reduce the number of unnecessary questions asked to the system operator, and to make inquiries more effective, it is possible to improve man-machine performance. In addition, processing efficiency can be improved because the rule that is predicted to increase the reliability the most is used preferentially.

〔Effect of the invention〕

According to the present invention, in a system that makes inferences using ambiguous knowledge, it is possible to reduce unnecessary inquiries and perform effective searches and inquiries, thereby improving system processing efficiency and man-machine processing. It has effects such as improving sex.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of a processing outline of a fuzzy inference system, and FIG. 2 is a diagram showing an example of a processing flow. Figure 3 is a diagram showing an example of the rule description method, Figure 4 is a diagram showing an example of the method of calculating the certainty of an event, Figure 5 is a diagram showing an example of the rule, and Figure 6 is a diagram showing an example of the method of calculating the certainty of an event. This is a related graph of the illustrated rules. 1... Ambiguous reasoning system, 2... Ambiguous knowledge Be No. Z Figure: l: 8 To 111 -1

Claims (1)

[Claims] 1. In a system that uses ambiguous knowledge to infer whether a specific event holds true or not, the inference mechanism for determining whether a specific event holds true or not holds with sufficient certainty is divided into two stages, and the first stage Then, the system automatically makes inferences using only known events, and if the first stage of inference does not provide sufficient confidence, the second stage is to ask the system operator about unknown events. Inquiry, using it to continue the first stage inference and obtain a sufficient degree of certainty, and in the above first and second stage inferences, the ambiguity of disjunctions, logical products, and connection relationships in ambiguous knowledge Based on the calculation method, we predict the event that can most increase the confidence that a specific event will occur or not, and in the first stage of reasoning,
An ambiguous inference method characterized by providing a means for automatically inferring the above information preferentially and, in the second stage of inference, for preferentially inquiring the operator of the system.