JPH02273833A - Fuzzy inference device and fuzzy rule evaluating device - Google Patents

Abstract

PURPOSE:To acquire quantitatively the influence agree to the final inference result of each rule and to grasp the change of validity of each rule by obtaining the moment around the fixed value for the fuzzy inference result of each rule. CONSTITUTION:An input signal is applied to each of fuzzy computing means 1 - 4 consisting of an antecedent part membership function generator, an antecedent part AND circuit, a consequent part membership function generator, a consequent part AND circuit, etc. Then the inference result is outputted from the fuzzy computing means for each rule. A defuzzy means 5 consists of an OR circuit which applies an OR to each inference rule, a fixed value computing circuit which computes the fixed value from the output of the OR circuit. A moment computing means 6 obtains the moment around the fixed value F as a degree of contribution for the inference result of each inference rule based on the inference results of means 1 - 4 and the fixed value received from the means 5. Thus the correction is possible to improve the degree of contribution of each rule. Then a fuzzy inference device having high inference accuracy can be obtained with use of a small number of rules.

Description

Detailed Description of the Invention (a) Industrial Application Field The present invention relates to a fuzzy inference device and a fuzzy rule evaluation system that are capable of evaluating the effectiveness of each set inference roof L/ in the fuzzy inference device. Regarding equipment.

Summary of the invention A fuzzy inference device and a fuzzy rule evaluation device according to the present invention actually operate a fuzzy logic system and evaluate the degree of influence (determined value) of the set fuzzy inference rules on the fuzzy inference results (determined value). This method evaluates the dynamics of each rule in order to determine its degree of contribution.

(C) Prior Art Conventionally, when using a fuzzy inference device, membership functions and rules to be set have been determined mainly by human operators based on experience.

However, it is difficult to set optimal rules and membership functions from the beginning, so it is necessary to evaluate the effectiveness of the rule itself or the membership function used in the rule by actually operating it, and then modifying the rule. A reconfiguration is being done. Additionally, an accurate model of the controlled object is created, each rule is evaluated in advance, and the optimal rule is set.

(d1 Problem to be solved by the invention) However, the former method cannot quantitatively express the effectiveness of each rule and can only be evaluated empirically, and the latter method Accurate rule evaluation was not possible because accurate modeling was difficult.

Note that some fuzzy inference devices have a monitor function that shows the inference results, but this is insufficient for evaluating rules.

The purpose of this invention is to make it possible to quantitatively determine how effective a set rule is on the overall inference, and to modify the rule or the membership function used in the rule. An object of the present invention is to provide a fuzzy inference device and a fuzzy rule evaluation device that can easily improve inference accuracy.

(a) Means for Solving yA8 A fuzzy inference device according to claim (1) of the present invention comprises an antecedent membership function generator, a consequent membership function generator, and an antecedent membership function generator. In a fuzzy inference device comprising a fuzzy calculation means for obtaining an inference result for each inference rule between a label and a consequent part label, and a defuzzifying means for obtaining a definite value from the inference result for each rule, The present invention is characterized by the provision of moment calculation means for calculating moments around the value.

Further, the fuzzy rule evaluation device according to claim (2) of the present invention includes an antecedent part membership function generator, a consequent part membership function generator, and an inference rule between an antecedent part label and a consequent part label. An evaluation device for a fuzzy inference device comprising a fuzzy operation means for obtaining an inference result for each rule, and a defuzzifying means for obtaining a definite value from the inference result for each rule, the evaluation device for evaluating a fuzzy inference device, which calculates the moment around the definite value for the inference result of each inference rule The present invention is characterized in that it is provided with a means for calculating the moment to be determined.

(f) Effect A block diagram is shown in FIG. 1 as an example of the configuration of this invention.

In FIG. 1, reference numerals 1 to 4 are fuzzy calculation means comprising an antecedent membership function generator, an antecedent AND circuit, a consequent membership function generator, a consequent AND circuit, and the like. Input signals are given to these fuzzy calculation means 1 to 4, respectively, and inference results are output for each rule.

The form of the consequent membership function provided in the fuzzy calculation means 1 to 4 is as shown in FIG. 2, for example. Here, each label means ZR: make the output almost 0, S: make the output small (do it), M: make the output medium, L: make the output large.

Such a consequent membership function is subjected to so-called head truncation by a consequent logical product circuit or the like, and a trapezoidal output as shown in FIG. 3, for example, is obtained.

The defuzzifying means 5 is composed of an OR circuit for ORing these inference results, a definite value calculation circuit for calculating a definite value from the output thereof, and the like. This logical sum circuit logically adds the respective trapezoidal outputs to form an area as shown by hunting in FIG. 3. Further, the determined value calculation circuit outputs the center of gravity position -tF of the hatched portion as a determined value. In addition,
In FIG. 3, Xl, X2. X3 and X4 are ZR,S
, M and L. Also, Sl, S2.
S3 and S4 are the inference results of each rule whose consequent membership function labels are ZR, S, M, and L (
area).

The moment calculating means 6 shown in FIG. 1 calculates the degree of contribution of each rule in the following manner, for example.

If the contribution of rules 1 to 4 is 81 to A4, then AI-IF
-XI $31 ... (1) A2= F-
X21 *S2 ... (2) A3-IF-X3
1 *S3 ... (3) A4-F-Xi
*S4 ... Obtained as (4).

That is, the moment calculation means 6 is the fuzzy calculation means 1 to
4 and the definite value output from the defuzzifying means 5, the moment around the definite value F for the inference result of each inference rule is determined as the degree of contribution.

The integrating means 7 integrates the moment of each inference rule based on the elapsed time from the input of the integral recent signal, and the display means 8 displays the integral output for each rule.

In this way, it is possible to know the degree of influence (degree of contribution) each rule has on the final value. By this, it can be determined, for example, that rules with a high degree of contribution are working effectively, and rules with an extremely low degree of contribution are unnecessary. Therefore, by making modifications to increase the degree of contribution of each rule, it is possible to configure a fuzzy inference device with high inference accuracy using a small number of rules.

(First Embodiment FIG. 4 shows a block diagram of a fuzzy inference device and a fuzzy rule evaluation device that are an embodiment of the present invention.

In Fig. 4, 1.2.4 are respectively fuzzy operation units,
5 is a defuzzify section, which constitutes a fuzzy inference device. Further, numeral 100 in the figure is a fuzzy rule evaluation device, which evaluates each fuzzy rule by being connected to the fuzzy inference device. By incorporating this fuzzy rule evaluation device 100 into a fuzzy inference device, a fuzzy inference device equipped with a fuzzy rule evaluation function is constructed.

In this example, two input signals Vl and V2 are input to each of the fuzzy calculation units 1, 2, . . . , 4, respectively. In the fuzzy operation unit 1, 10.1) is an antecedent membership function generator, and 101) is input with the variable v1 and label of the first term of the antecedent, and the variable v2 and label of the second term. . As a result, the antecedent membership function generator 10 generates the membership value of ``1'' and the membership value of v2, respectively. The antecedent logical product circuits 1 and 2 select the smaller membership value according to the so-called m l n L -max rule. Further, in the fuzzy operation unit 1, 13 is a consequent part membership function generator, and a consequent part AND circuit 1 generates a consequent part membership function by inputting a label of the consequent part.
4 performs so-called head truncation on the consequent membership value number generated from 13 based on the output of the antecedent part AND circuit 12, and outputs the inference result X1. In the same way, the other fuzzy calculation units 2...4 calculate the inference result X2.
...outputs Xi.

In the defuzzifier 5, the OR circuit 15 calculates the max rule of the minlmax x rule for each inference result Xi, X2, . . . Xi. r+1 constant value calculation circuit 16
calculates the center of gravity position from the output of the OR circuit 15 and outputs the determined value F.

Now, the fuzzy rule evaluation device 100 is configured as follows. In the figure, 61, 62, . . . , 64 are circuits corresponding to moment calculation means, respectively. The center of gravity/area calculation circuit 17 is an output from the fuzzy calculation section 1. What is the center of gravity of the inference result x1? i! (Xi) and surface m (Sl) are determined. The moment calculation circuit 18 calculates (1) from the final value F output from the defuzzifier 5 and the center of gravity position and area of the inference result.
The moment around the determined value is obtained by performing the calculation shown in the formula. Integrating circuits 71, 72, . . . , 74 integrate each moment over time, and display units 81, 82, . . . , 84 display the values in graphic ink.

(h1 Effect of the invention According to this invention, it becomes possible to quantitatively determine the degree of influence by determining the moment around the determined value for the fuzzy inference result of each rule, and to understand changes in the effectiveness of each rule.) This makes it possible to perform highly accurate processing with fewer rules by, for example, deleting rules with low effectiveness, reducing the number of fuzzy boards or fuzzy chips, and reducing the number of fuzzy boards or chips. It is also possible to add a learning function by applying feedback to automatically modify the rules according to the contribution of each rule.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of the configuration of the present invention. FIGS. 2 and 3 are diagrams showing examples of consequent membership functions and inference results. FIG. 4 is a block diagram of a fuzzy inference device and a fuzzy rule evaluation device that are embodiments of the present invention. 10.1)-Antecedent Membership Function Generator, 13-Consequent Membership Function Generator.

Claims (2)

[Claims] (1) An antecedent part membership function generator, a consequent part membership function generator, a fuzzy calculation means for obtaining an inference result for each inference rule between an antecedent part label and a consequent part label, and a fuzzy calculation means for each rule. 1. A fuzzy inference device comprising a defuzzify means for obtaining a definite value from an inference result, characterized in that the fuzzy inference device is provided with a moment calculation means for determining a moment around the definite value for the inference result of each inference rule. (2) An antecedent part membership function generator, a consequent part membership function generator, a fuzzy calculation means for obtaining an inference result for each inference rule between an antecedent part label and a consequent part label, and a fuzzy calculation means for each rule. An evaluation device for a fuzzy inference device comprising a defuzzifying means for calculating a definite value from an inference result, the fuzzy rule comprising a moment calculation means for calculating a moment around the determined value for the inference result of each inference rule. Evaluation device.