JPH06282436A - Approximate inference device - Google Patents

Abstract

PURPOSE:To eliminate the need to calculate the center of gravity of output estimation by decreasing the man-hours required to adjust the membership function of an input variable when an output variable is approximately estimated from the input variable by using plural inference rules. CONSTITUTION:This device is equipped with inference arithmetic parts 111, 112,..., 119 each consisting of a rule truth value storage part 130 which stores the truth values of plural inference rules, plural input variable truth value function parts 120 and 125, a fidelity arithmetic part 150 which calculates the limit difference between their outputs, an output variable truth value inverse function part 160 which receives the output of the rule truth value storage part 130 and outputs an output variable representative value, and a multiplication part 170, addition parts 180 and 185 and a division part 190; when an optimum output variable is identified by using the relation between the input variable and output variable, truth values corresponding to the inference rules are corrected to find the representative value of an output estimated value.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is an approximation for inferring an optimum output variable by using a plurality of inference rules that describe input / output relationships when a value of an input variable is given in the control of a process, mechatronics or the like. Use for inference.

The present invention relates to an approximate inference apparatus capable of performing approximate inference for approximating an output variable from an input variable with a small number of man-hours.

[0003]

2. Description of the Related Art Approximate inference aims to approximately estimate an output variable corresponding to an input variable when the input / output relationship is not grasped as a clear model such as a mathematical formula.

As shown in FIG. 6, the conventional method of approximate inference is based on fuzzy inference applying fuzzy set theory (Michio Sugano "Fuzzy Control" 1988 Nikkan Kogyo Shimbun) as follows. There is.

An input variable and an output variable in an input / output relationship are qualitatively expressed, and an input / output relationship is a qualitative expression of the relationship between the input variable and the output variable, which is described as a plurality of rules. For example, assuming that there are two input variables and one output variable, they are respectively written as x ₁ and x ₂ : y, and the respective quantities are fuzzy expressions A _1i ,
A _2i , B _i , (i is the rule number), and n rules are as follows.

[0006]

[Equation 1] For each of these rules, rule calculation units 1011 1012, ..., 1019 are provided as shown in FIG.

Although the fuzzy expression of the qualitative expression may be arbitrary, it is usually performed as follows. The variable area is divided into positive and negative, and the absolute value of each is divided into small, medium, and large classes. For example, the abbreviations for each class are PB (positive and large), PM (positive and medium), PS (positive and small),
Z (near zero), NS (negative and small), NM (negative and medium), and NB (negative and large).

As described above, the fuzzy expression can be represented by the membership function for each class. The membership function of each class is the degree of membership of a variable when the fuzzy representation of the variable is associated with the fuzzy set of each class. This is shown in the figure, for example, as shown in FIG. 7. Here, the input variables are [outer 1] and [outer 2], and as shown in FIG. 6, fuzzy inference operation units 1011 and 1012,
... Input variable membership function unit 112 for expressing inputs A _1i and A _2i with respect to 1019 by a membership function
0, 1125, and the membership functions [outside 3] and [outside 4] are sent to the fuzzy logic operation unit 1140 from the membership function in response to the inputs [outside 1] and [outside 2]. In the fuzzy logic operation unit 1140, a rule part conforming value w _i which determines how effective the i-th rule is for a plurality of variables [outer 1] and [outer 2] is calculated by the following equation.

[0009]

[Equation 2] Here ∧ indicates that the minimum is selected.

[0010]

[Outer 1]

[0011]

[Outside 2]

[0012]

[Outside 3]

[0013]

[Outside 4] According to the fuzzy theory, the output value in each rule is determined by using w _i as follows.

[0014]

[Equation 3] This can be shown by the fuzzy representation of the shaded portion a as shown in FIG. As shown in FIG. 6, the system stores the truth value of the output variable of the rule as a membership function in each of the rule calculation units 1011 1012 1019. A membership function is sent from the output membership function unit 1160, and a rule part conforming value is sent from the fuzzy logic operation unit 1140 to the output estimation unit 1170, and the operation according to the equation (3) is performed. Calculate the following formula based on fuzzy theory to integrate the output estimates from multiple inference rules

[0015]

[Equation 4] Here, ∨ indicates that the maximum is selected. The operation of the system when n = 2 is shown in FIG.

In the fuzzy inference system, an arithmetic unit 1180 is provided as shown in FIG.
The output estimated values from 1, 1012, ..., 1019 are sent to the fuzzy sum computing unit 1180 to perform the computation of the equation (3 ′).

B ⁰ (y), which is the result of equation (3 '), is represented by a membership function as a fuzzy set. When it is necessary to estimate the output value as one point y ⁰ , the centroid method using the following equation is generally well known.

[0018]

[Equation 5] This operation is called defuzzification. y ⁰ gives the coordinates of the center of gravity with respect to y of B ⁰ shown in FIG.

In the fuzzy inference system, as shown in FIG. 6, the result of the fuzzy sum calculation unit 1180 is given to the center of gravity calculation unit 1190, the calculation according to the equation (4) is performed, and y ⁰ is output.

When approximate inference is used for control of processes, mechatronics, etc., it is generally carried out as follows. In order to perform control, an optimum operation amount is determined corresponding to the state quantity indicating the state of the controlled system, and the operation is performed moment by moment. In the above approximate inference, the input variable is associated with the state variable and the output variable is associated with the manipulated variable, so that control can be performed even when the mathematical relationship between the input variable and the output variable is not known.

[0021]

According to such a conventional method, when the optimum output variable is identified by using the relationship between the input variable and the output variable, the membership must be obtained unless a satisfactory identification result is obtained. It is necessary to reset the function. For example w _i since w _i as shown in equation (2) to regulate the output B
Need to be adjusted. As shown in the equation (3), w _i finds the minimum value of the membership function of the input variable corresponding to the rule in the input values [outer 1] and [outer 2]. Show the degree.

To adjust w _i , it is necessary to change either or both of [Outer 3] and [Outer 4]. This change of the membership function also changes the output B corresponding to the rule that does not need to be changed, and the target adjustment is difficult.

For example, in order to increase the effect of the rule R ₁ shown in FIG. 9, if [outer 5] is increased and the membership function b of A ₂₁ is changed to PM as shown by the broken line, [outer 5] 6] also increases at the same time as c, and the effect of rule R _n also increases.

[0024]

[Outside 5]

[0025]

[Outside 6] As described above, according to the conventional method, since the center of gravity of the membership function is obtained as in the equation (4) for the one-point estimation of the output, a complicated calculation is required and a complicated calculation device is used. Or, there is a problem that the calculation must be done over a long period of time using a conventional calculation device.

The present invention solves such a problem, and an object of the present invention is to provide a system which eliminates complicated calculation and can perform calculation in a short time even with a commonly used calculation device. .

[0027]

A first object of the present invention is to provide a plurality of inference operation units for outputting output variable estimated values according to a plurality of inference rules, and to approximate from output variable estimated values respectively obtained by the inference operation units. A rule truth value storage unit that includes a calculation unit that obtains an inference value, the truth value storage unit that separately stores truth values of a plurality of inference rules, and a truth value that indicates the degree of credibility regarding the qualitative expression of the input variable in each rule. , A plurality of input variable truth value function units for outputting for each input variable, and an output variable truth value inverse function for obtaining a representative value of the output variables by the rule truth values from the rule truth value storage unit. A part, an input variable operation part that receives the output from the input variable truth value function part and obtains the complement of the representative value of the input variable truth value,
A fidelity calculation unit that calculates a limit difference between the output of the rule truth value storage unit and the output of the input variable truth value function unit;
A multiplication unit for multiplying the output variable representative value from the output variable truth value inverse function unit by the output of the fidelity calculation unit to output an output variable estimated value, and each of the multiplication units outputs to the calculation unit. An output variable estimated value adding unit for adding the output variable estimated value, a fidelity adding unit for adding the limit difference calculated by each of the fidelity calculating units, and an output of the output variable estimated value adding unit by the output of the fidelity adding unit. And a dividing unit for dividing.

The input variable truth value function section is an original function section that establishes a predetermined functional relationship with respect to the relationship between the input variable and its truth value, and the input variable when it is "very" in the description of the input variable of the rule. And the square function part for inputting the square of a predetermined function, and in the description of the input variable of the rule, when there is "somewhat", the input variable is input and the half power of the predetermined function is input 1 / Square function part and a complement function part for inputting the input variable when it is "not ..." in the description of the input variable of the rule and forming the complement of a predetermined function, and the output variable truth value inverse function part Is the original inverse function part that makes the relationship between the output variable and its truth value a predetermined functional relationship, and when there is "very" in the description of the output variable of the rule, the output variable is the square of the predetermined function. Squared inverse function part , In the description of the output variable of the rule, when saying "which do you say?", The half-power inverse function part that makes the output variable the 1/2 power of the predetermined function, and in the description of the output variable of the rule, "not ... , ", The output variable can be composed of a complement inverse function part which is the complement of a predetermined function.

A second aspect of the present invention comprises a plurality of inference operation units for outputting output variable estimated values according to a plurality of inference rules,
The inference operation section is provided with operation means for obtaining an approximate inference value from the output variable estimated value obtained in each of the inference operation section,
A rule truth value is set in the rule stored in the rule truth value storage unit, a means for setting the limit difference between the rule truth value and the input variable truth value complement as the fidelity of the representative value of the output variable, and the fidelity is positive. Sometimes the rule truth value is inferred to be the truth value of the output variable, the result is input to the output variable truth value inverse function part, and the representative value of the output variable is sent out; and the fidelity of the representative value of the output variable. Means for outputting an output variable estimated value by multiplying by, and means for obtaining the output variable estimated value according to a plurality of inference rules by arithmetically averaging the output variable estimated value with respect to all of the plurality of rules in the arithmetic means. It is characterized by having.

By the output variable truth value inverse function unit, the rule truth value is inferred as the truth value of the output variable, and the representative value of the corresponding output variable is calculated as the output variable truth value larger than the rule truth value. The truth value of the input variable and the output variable are increased by increasing or decreasing the plurality of rule truth values in the direction opposite to the direction in which the difference between the predetermined output value and the output estimated value is increased, including means for setting the median value of the interval. Including means for reducing the difference between the required output value and the output estimated value while fixing the truth value of
It is desirable to include means for matching the larger value of the truth value of the input variable and the truth value of the output variable with the truth value of the rule describing the input / output relationship.

[0031]

In contrast to the conventional fuzzy theory, recently, the classical logic handling only 0 or 1 is newly extended, and a new multi-value capable of handling the value of the interval [0, 1] in the logical expression. Logic is K. K. Announced by Sombre (FUZZ-1992 Preprint 271, sponsored by IEEE)
~ P. 8 March 1992). When this is expressed in a format corresponding to fuzzy inference, the expression of the rule is as shown in Expression (1).

A truth value (t _Ri ) is processed in each rule as a grade of its authenticity, and A ₁ .. in fuzzy reasoning is used. A ₂ . , B. If the membership function of is considered to match the truth values of the input variables x ₁ and x ₂ and the output variable y and the symbols are used as they are and the input variables are [outer 1] and [outer 7], the i-th rule , K. By the multivalued logic of K Sonber,

[0033]

[Outside 7]

[0034]

[Equation 6] R _i : if x ₁ is A _1i and x ₂ is A _2i , then y is B _i ; t _RI (6),

[0035]

[Equation 7] Is established.

The expression (8) is a necessary condition for the expression (7) to be satisfied (called modus ponence). B _i
Determined inverse function [External 8] of 6, rule R _i
From

[0037]

[Outside 8]

[0038]

[Equation 8] The output [outer 7] is inferred from the expression (9). Rule R _i
It is considered that the degree of establishment of the inference (according to fidelity and referred to as f _pi ) is the degree of establishment of the equation (8), and thus is defined as follows.

[0039]

[Equation 9] In this way, the output inference value [outer 9] and the fidelity f _p .are obtained from each rule.

[0040]

[Outside 9] When calculating the representative value by integrating these, the theoretical value of the total output variable is the same as the calculation of the expected value.

[0041]

[Equation 10] Calculated by

The truth value t _R. If there is an actual value in advance, t _R. = (1-A ₁ .∧A _2. ) ∨B. It can be determined in (12). If it can be theoretically determined in advance, that value is used. Without prior knowledge or knowledge, t _R. = 1 is used.

As a result, the labor and time required for adjusting the membership function of the fuzzy set of input variables and for calculating the center of gravity of the fuzzy set of output variables can be greatly reduced.

[0044]

Embodiments of the present invention will now be described with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention.

In the first embodiment of the present invention, a plurality of inference operation units 11 for outputting output variable estimated values according to a plurality of inference rules.
1, 112, ..., 119, and this inference operation unit 11
.., 119, and the inference operation units 111, 112, ..., 119 each have a rule for separately accumulating the truth values of a plurality of inference rules. The truth value storage unit 130 and two input variable truth value function units 120 and 125 that output for each input variable the value of the truth value indicating the degree of credibility regarding the qualitative expression of the input variable in each rule. When,
The output variable truth value inverse function unit 160 that obtains the representative value of the output variable from the rule truth value storage unit 130 and the representative value of the input variable truth value that receives the outputs from the input variable truth value function units 120 and 125. Input variable operation unit 1 for calculating the complement of
40, a fidelity calculation unit 150 that calculates a limit difference between the output of the rule truth value storage unit 130 and the outputs of the input variable truth value function units 120 and 125, and the output variable truth value inverse function unit 1
The fidelity calculation unit 150 is set to the output variable representative value from 60.
170 that multiplies the output of
And an output variable estimated value adding section 180 for adding the output variable estimated value output from each of the multiplying sections 170, and a fidelity adding section 185 for adding the limit difference calculated by each of the fidelity calculating sections 150 to the calculating means. And a division unit 190 that divides the output of the output variable estimated value addition unit 180 by the output of the fidelity addition unit 185.

Next, the operation of the first embodiment of the present invention thus constructed will be described with reference to the drawings.

Here, as shown in FIG. 1, the inference operation units 111 and 112 correspond to a plurality of rules for inference.
, 119 are provided, and in order to simplify the explanation, it is assumed that there are two input variables [outer 1] and [outer 2].

The input variable truth value function units 120 and 125 are
When inputs [outer 1] and [outer 2] are given, the input variable in each rule outputs the value of the truth value indicating the degree of credibility with respect to the qualitative expression of the input variable. The outputs are [outer 10] and [outer 11].

[0050]

[Outside 10]

[0051]

[Outside 11] The input variable calculation unit 140 uses the K. K. [External 12], which is the right-hand side of the equation (8) indicating the required condition of modus ponence in Somber's multivalued logic, that is, the representative value of the input variable logical value [External 10], [External 11], [External 13] Find the complement of. Further, the rule truth value storage unit 130 stores the rule truth value t _R. Memorize

[0052]

[Outside 12]

[0053]

[Outside 13] The fidelity calculation unit 150 includes a rule truth value storage unit 130.
To the rule truth value t _R. In addition to receiving K.K. K. Fidelity f _p ., Which is the degree to which the expression (8) showing the necessary condition of modus ponence in Somber's multivalued logic is satisfied. In order to obtain, the calculation of the above equation (10) is performed. That is,
Rule truth value t _R. And the representative value of the input variable truth value [External 13]
The limit difference calculation is performed. When the equation (8) indicating the necessary condition of modus ponence is satisfied, that is, the fidelity f _p . Only when is positive, the inference operation unit 11
1, 112, ..., 119 to a fidelity addition unit 185 provided outside the fidelity f _p . Is sent. The output variable truth value inverse function unit 160 has a function of an inverse function that outputs the output variable when the truth value of the output variable is given. FIG. 3 shows the rule truth value t _R. It is a figure which shows the relationship between an output variable [outer 9].

K. K. If the requirement of modus ponence in Somber's multivalued logic is satisfied, the modus ponence's conclusion (7) can be used, so that the rule truth value t _R. From the rule truth value storage unit 130 to the output variable truth value inverse function unit 160 to obtain the representative value [outer 9] of the output variables. According to the above equation (9), this [outer 9] is the estimation result of the output variable. [Outer 9] is calculated by the multiplication unit 170 as fidelity f _p . And the fidelity f _p . , The inference operation units 111, 112,
119 is sent to the output variable estimated value addition unit 180 provided outside. In order to comprehensively infer the estimation results of each rule, the above equation (11) should be calculated.
Fidelity adder 185, output variable estimated value adder 18
The output of 0 is divided by the division unit 190 to obtain the output variable inference value y ⁰ .

In the first embodiment, one inference operation unit 111, 112, ..., 119 is provided for each of a plurality of rules, but one inference operation is performed depending on the configuration of the system. It is also possible to use the parts in a time-sharing manner.
With such a new approximate reasoning system, it is possible to similarly control processes and mechatronics using fuzzy reasoning according to the related art. That is, by associating the state variables of the controlled system with the output variables and the manipulated variables with the input variables, respectively, even if the mathematical relationship between the state variables and the manipulated variables is not known, the operation for the momentary state variables is performed. You can decide the quantity.

The method will be described below in comparison with the conventional method.

When the input / output function shown in FIG. 4 is inferred by the approximate inference, in the conventional method, it is inferred by the procedure shown in FIG. 10 and is estimated as a point (black) shown in FIG.

On the other hand, according to the present invention, it can be inferred by the procedure shown in FIG. 10 and is estimated as shown by the points (white) in FIG.

However, as shown in FIG. 4, at x = 3, the inference error is Δy = 15-12 = 3, which is too large to ignore.

In order to reduce this error, according to the conventional method, the membership function of the region PS shown in FIG. 4 where x is positive and small is changed as shown by the broken line, and the effect of the rule R ₁ at x = 3 is obtained. Strengthening is one measure. However, as shown in FIG. 10, the inferred value is y = 14, and the error does not become small. Moreover, if there is another rule that uses PS, the effect of that rule is also affected.

On the other hand, according to the present invention, as in the procedure shown in FIG. 10, by changing the degree of reliability of the rule R ₂ from 1 to 0.6 in order to increase the weight of the rule R ₁ , x =
When 3, y = 11.6, and the error is slightly smaller (0.4).

By adjusting the credibility grade of the rule in this way, the degree of inference can be easily improved.
Further, in the conventional method, in order to output the output amount at one point, it is necessary to calculate the center of gravity of the membership function of the fuzzy set. Taking a microprocessor as an example, the time (T) required for the calculation is expressed as follows by the cycle time (C), the data density (N) expressing the membership function, and the number (n) of rules.

T≈ (Nn + α + β) C (13) where α and β are the number of cycles required for sum of products and division, respectively.

On the other hand, according to the present invention, since the membership function is not used, the required time (T ') corresponding to the equation (13) is as follows.

T′≈ (n + α + β) C (14) N is usually several to 256, n is several, α, β
Is 2-3. Therefore, in the present invention, the time required to obtain the output at one point is about an order of magnitude shorter than that in the conventional method. Furthermore, as an effect of the present invention, when the authenticity grade of the rule that describes the input / output relationship is predetermined, the K. K. Applying the new multi-valued logic by Sonber,
Formula (12) can be used to determine the actual data,
You can make inferences based on your achievements.

FIG. 5 shows a concrete example of the inference process in the first embodiment of the present invention.

(Second Embodiment) FIG. 2 is a block diagram showing the arrangement of the essential parts of a second embodiment of the present invention.

In the second embodiment of the present invention, as the input variable truth value function part in the first embodiment, the original function part 220 which establishes a predetermined functional relationship for the function of the input variable and its truth value.
When the description of the input variable of the rule is "very", the input variable is input and the square of a predetermined function is entered. 2
A power function unit 221, a half power function unit 222 for inputting the input variable to make a predetermined function 1/2 power when there is "somewhat" in the description of the rule input variable, and a rule input variable In the description, the input variable is input and the complement function unit 223 that complements the predetermined function is provided when the output variable and the truth thereof are used as the output variable truth value inverse function unit in the first embodiment. The original inverse function part 270 that makes a predetermined functional relationship with the value, and the squared inverse function part 271 that makes the output variable the square of the predetermined function when there is "very" in the description of the output variable of the rule.
When the output variable of the rule says "Which one", the output variable is 1/2 of the predetermined function.
It is provided with a ½-power inverse function unit 272 that is a power and a complement inverse function unit 273 that uses the output variable as a complement of a predetermined function when “not ...” In the description of the output variable of the rule.

In the second embodiment, the input variable truth value function part and the output variable truth value inverse function part in the first embodiment are grounded in parallel, and the operation is performed as follows.

[0070] If there is no modifier for qualitative representation A ₁ relates to the input variable x ₁ in the description of the rules, the switching unit 210, the input variables [External 1] is connected to the input to the original function unit 220, qualitative representation When the modifier in A ₁ is “very”, the modifier is connected to the square function unit 221 that sets a function that squares the truth value of the input variable, and the modifier is added to the qualitative expression A _1. Is "somewhat", it is connected so as to be input to the 1/2 power function unit 222 that sets a function that squares the truth value of the input change, and the qualitative expression A ₁ is not "... ", The connection is made to the complement function unit that sets the complement of the truth value of the input variable.

The qualitative expression A ₂ is similarly set for the variable x ₂ , and the original inverse function unit 27 is similarly set for the output variable.
The switching unit 250 switches the 0-square inverse function unit 271, the 1 / 2-power inverse function unit 272, and the complement inverse function unit 273, respectively, and sends the truth value to the modifier of the qualitative expression regarding the output variable of the rule. .

As described above, in the second embodiment of the present invention, when a modifier is attached to the qualitative expression of the input variable, a plurality of input variable truth value function units are provided in parallel in advance, so that the switching unit 210 is selected according to the modifier. Can switch the connection of the input variables to the plurality of parallel input variable truth value function units, and further, the plurality of output variable truth value inverse function units are also provided in parallel, so that the inference is performed by the switching unit 250 according to the modifier. You can switch the destination of the authenticity grade.

Thus, if the rule is loaded with modifiers for input and output variables as follows, for example, if R ₁ : x ₁ is “very” A ₁ and x ₂ is A _2, then y
Is “very” B ₁ R ₂ : x ₁ is “from which” A ₁ and x ₂ is A
₂ If y is B ₁ R _3: x ₁ is "not de" A _1, and if x ₂ is A _2,
When y is based on the three rules of B ₁ "not", it is operated as follows using one inference operation unit 200 shown in FIG.

Regarding the rule R ₁ , [outer 1] is set to 221 corresponding to “very”, and the rule truth value t _{R1 is set} to “very”.
To the inverse square function unit 271 corresponding to
Switch between 0 and 250.

Regarding the rule R ₂ , the switching unit 21 sends [outer 1] to the ½ power function unit 222 corresponding to “if anything” and the rule truth value t _R1 to the original inverse function unit 270.
Switch between 0 and 250.

Regarding the rule R ₃ , the switching unit 210 sends the [outer 1] to the complement function unit 223 corresponding to “not” and the rule truth value t _R1 to the complement inverse function unit 273 corresponding to “not”. Control 250.

As described above, the functions of the three inference operation units can be achieved by using only one inference operation unit 200 without setting the inference operation units for the rules R ₁ , R ₂ and R _3. You can As described above, in a plurality of rules, when a modifier is added to the qualitative expression of an input variable, the function characteristic of the characteristic corresponding to the modifier is defined for the input variable truth value function part and the output variable truth function part, and the modifier is defined. By selectively using it according to the word, the number of inference operation units can be reduced and the cost of the system can be reduced. The first and second embodiments have been described with reference to the block diagrams in the drawings. However, in order to realize this device, it is not always necessary to provide a circuit that can be divided by such a block diagram, Can be realized by executing the arithmetic processing corresponding to the above with one processor. Often, it is more rational as a design.

Therefore, each term in the claims which does not correspond to the drawing for each block has its meaning, and can be similarly understood from the description of the above two embodiments.

[0079]

As described above, according to the present invention, when the result of the fuzzy inference does not reach a desired output value, the operation requiring a calculation process such as modifying the membership function is unnecessary, and the inference is simply performed. Reasoning can be performed simply by modifying the truth value corresponding to the rule. Therefore, even if the membership function is adjusted, the desired output value was not obtained, and thus approximate reasoning can be performed even in the region where fuzzy reasoning cannot be used. There is an effect that can be utilized.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a main part of a second embodiment of the present invention.

FIG. 3 is a diagram showing a relationship between a rule truth value and an output variable in the first embodiment of the present invention.

FIG. 4 is a diagram for explaining a comparison of operations between the first embodiment of the present invention and a conventional example.

FIG. 5 is a diagram showing a specific example of an inference process in the first embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a conventional example.

FIG. 7 is a diagram showing an example of a membership function in a conventional example.

FIG. 8 is a diagram illustrating a method of inference in a conventional example.

FIG. 9 is a diagram illustrating an example of a problem to be solved in the present invention.

FIG. 10 is a diagram illustrating an inference process in a conventional example.

[Explanation of Codes] 111, 112, ..., 119, 200 Inference operation unit 120, 125 Input variable truth value function unit 130 Rule truth value storage unit 140 Input variable operation unit 150 Fidelity operation unit 160 Output variable truth value inverse function unit 170 Multiplying unit 180 Output variable estimated value adding unit 185 Fidelity adding unit 190 Dividing unit 210, 250 Switching unit 220 Original function unit 221 Square function unit 222 1/2 Exponential function unit 223 Complement function unit 270 Inverse inverse function unit 271 Squared inverse Function part 272 1/2 inverse function part 273 Complement inverse function part 1011 1012, ..., 1019 Fuzzy inference operation part 1120, 1125 Input variable membership function part 1140 Fuzzy logic operation part 1160 Output membership function part 1170 Output estimation part 1180 Fuzzy sum operation unit 1190 Centroid calculation Department

Claims (6)

[Claims] 1. An inference operation unit for outputting an output variable estimation value according to a plurality of inference rules, and an operation unit for obtaining an approximate inference value from the output variable estimation value obtained by each of the inference operation units. A rule truth value storage unit that separately stores the truth values of multiple inference rules in the operation unit, and the input value of the truth value indicating the degree of credibility regarding the qualitative expression of the input variable in each rule. A plurality of input variable truth value function units that output for each, an output variable truth value inverse function unit that obtains a representative value of the output variable by the rule truth value from the rule truth value storage unit, and the input variable truth value function unit An input variable calculation unit that receives the output of the input variable truth value and obtains the complement of the representative value of the input variable truth value; and a fidelity operation that calculates the limit difference between the output of the rule truth value storage unit and the output of the input variable truth value function unit. And a multiplication unit that outputs an output variable estimated value by multiplying the output variable representative value from the output variable truth value inverse function unit by the output of the fidelity calculation unit, and the calculation unit includes each of the multiplication units. Output variable estimated value adding section for adding the output variable estimated value output, a fidelity adding section for adding the limit difference calculated by each of the fidelity calculating sections, and the output of the output variable estimated value adding section for the fidelity adding section. An approximate inference apparatus, comprising: a division unit that divides by the output of. 2. The input variable truth value function part has an original function part that establishes a predetermined functional relationship with respect to the relationship between the input variable and its truth value, and when there is “very” in the description of the input variable of the rule, A square function part that inputs an input variable and takes the square of a predetermined function, and in the description of the input variable of the rule, when there is "somewhat", input the input variable and input 1/2 of the predetermined function.
The output variable truth is composed of a ½ power function part for raising the power and a complement function part for inputting the input variable and complementing a predetermined function when “not ...” Is described in the description of the input variable of the rule. The value inverse function part is the original inverse function part that makes a predetermined functional relationship for the relationship between the output variable and its truth value, and when the output variable of the rule is "very", the output variable is In the description of the squared inverse function part for making a square and the output variable of the rule, when it says "whichever", the output variable is 1/2 of the predetermined function.
2. An inverse function part for raising to the power of one half, and an inverse complement function part for using the output variable as a complement of a predetermined function when there is "not ..." In the description of the output variable of the rule. Approximate reasoning device. 3. An inference operation section for outputting an output variable estimation value according to a plurality of inference rules, and an operation means for obtaining an approximate inference value from the output variable estimation values respectively obtained by the inference operation section. The arithmetic unit sets a rule truth value to the rule accumulated in the rule truth value storage unit, and means for setting the rule difference between the rule truth value and the complement of the input variable truth value as the fidelity of the representative value of the output variable, When the fidelity is positive, the rule truth value is inferred to be the truth value of the output variable, and the result is input to the output variable truth value inverse function unit to output the representative value of the output variable, and the output variable Means for multiplying the representative value by fidelity to output an output variable estimation value, and the arithmetic means means arithmetically averaging the output variable estimation value for all of the plurality of rules to estimate the output variable by a plurality of inference rules. Approximate reasoning apparatus comprising the means for determining the value. 4. An output in which the output variable truth value is larger than the rule truth value by inferring the rule truth value as the output variable truth value by the output variable truth value inverse function unit. The approximate inference apparatus according to claim 3, further comprising means for setting a median value of a variable interval. 5. A truth value of an input variable and a truth value of an output variable are fixed by increasing or decreasing a plurality of rule truth values in a direction opposite to a direction in which a difference between a predetermined output value and an estimated output value is enlarged. The approximate inference apparatus according to claim 3, further comprising means for reducing a difference between a required output value and an estimated output value as it is. 6. The approximate inference apparatus according to claim 3, further comprising means for matching a larger value of the truth value complement of the input variable and the truth value of the output variable with the truth value of the rule describing the input / output relationship.