JPH05181986A - Non-fuzzy device - Google Patents

Abstract

PURPOSE:To improve variation in final determination output due to a noise and an error in inference grade by providing a means which applies an optical bias value to each inference output grade and a means which calculates the value of center of gravity of each inference output grade after the bias application. CONSTITUTION:An adaptability computing element 1 finds the adaptability between an input and a antecedent part. A minimum value computing element 2 performs minimum value arithmetic between the found adaptability and a consequent part to find inference results by inference rules. Their integrated inference result is inputted to non-fuzzy hardware 9. The inference result applied with the bias value is inputted to a weighted adder 6 and a simple adder 7, whose outputs are inputted to a divider 8 to calculate a value of center of gravity as its output. This value of center of gravity is the final determination output.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to control, pattern recognition,
The present invention relates to a defuzzification device for realizing fuzzy reasoning based on ambiguous information and knowledge such as decision making as hardware.

[0002]

2. Description of the Related Art The basics of fuzzy inference are as follows: The inference rule is described as a so-called "if then rule" in the form of "if ... if ...". here,
The "if-" part is called the antecedent part, and the "-is" part is called the consequent part.

The above "-" part is expressed by a membership function as a fuzzy set.

That is. By adopting such a method, ambiguous knowledge and information can be treated numerically.

The fuzzy reasoning algorithm is divided into the following four processes.

The degree of matching between a given input and the antecedent part of each inference rule is obtained.

An inference result for each inference rule is obtained from the consequent part in accordance with the obtained conformity. Integrate multiple inference results.

A finalized output value is obtained from the integrated inference result and is used as the final output. Obtaining this definite output value is called defuzzification.

As specific operations of the ,,, various methods such as minimum value operation, maximum value operation, algebraic product operation, algebraic sum operation, etc. have been proposed. However, as a defuzzification means in any of the methods, a method called a barycentric method is generally used, in which a definite output value is obtained from the barycentric value of the inference result expressed on the base set of the consequent part.

FIG. 7 shows the minimum-maximum-often used in practice.
The concept of fuzzy inference algorithm by the center of gravity method is shown.
In this method, the minimum value calculation is adopted as the process calculation method, and the maximum value calculation is adopted as the process calculation method. Furthermore, as a means to defuzzify the process,
The above-mentioned centroid method is used to calculate the definite output.

A specific calculation for calculating the center of gravity value is expressed by the following equation.

[0012]

[Equation 1] In formula (1), k represents the number of divisions of the consequent copy set, and i represents each divided address, as shown in FIG. μ _i is a grade showing the inference result of the address i, and w _i is a weighting coefficient showing the position of the address _i in the horizontal axis direction.

[0013]

As a characteristic of the centroid method as a means of defuzzification, there is a case where the centroid value does not change even if the grade of the inference result changes, as shown in FIG. That is, the centroid value is the same as in FIG. 3a when the inference result has a large grade and when the inference result has a small grade as in FIGS. As a result, the following problems are pointed out.

When implemented as hardware, an abnormal definite output value is output even if there is a slight noise or error in the inference result. For example, as shown in FIG. 4a, the centroid value due to noise is output even if there is originally no grade of the inference result.

In the conventional method, the grade of the inference result is 0 if the input does not match any inference rule at all. At this time, the numerator and denominator of the equation (1) become 0, and the center of gravity cannot be calculated. Therefore, when implemented as hardware, an abnormal value is output unless the input conforms to any inference rule.

Therefore, the problem to be solved by the present invention is to provide the following defuzzification hardware when implementing fuzzy inference as hardware.

Anomalies in the deterministic output caused by noise and errors that occur in the grade of the inference result are improved.

It is possible to prevent an abnormality in the deterministic output that occurs when the input value does not match any inference rule at all.

The definite output is calculated according to the grade size of the inference result.

[0020]

In order to solve the above-mentioned problems, the defuzzification apparatus of the present invention is such that, in the defuzzification apparatus of the fuzzy reasoning apparatus, the consequent part of the fuzzy reasoning is represented by a fuzzy set. It is characterized by comprising means for applying a bias value of an arbitrary magnitude to the output grade and means for calculating the centroid value of each inferred output grade after the bias value is applied.

[0021]

By applying a bias value to the grade of the inference result and calculating the center-of-gravity value to obtain the final output value, the influence of noise or error on the final-output value (center-of-gravity value) is reduced as shown in FIG. 4b. .. Even when the inference result grade is 0 as shown in FIG. 5, the center of gravity value is controlled to an appropriate value by the applied bias value. Further, as shown in FIG. 6, a definite output value according to the size of the inference grade can be calculated. The magnitude and shape of the bias value can be freely adjusted according to the purpose.

[0022]

EXAMPLES The present invention will be specifically described below based on examples.

FIG. 1 is a block diagram showing an example in which the present invention is implemented as hardware. In this embodiment, the so-called minimum value calculation-maximum value calculation-centroid method is adopted as the inference method. Explaining with reference to the drawing, first, the conformity calculator 1 calculates the conformance between the input and the antecedent part for each of the n inference rules. Next, the minimum value is calculated between the conformity determined by the minimum value calculator 2 and the consequent part, and the inference result is obtained for each inference rule.

The respective inference results are input to the maximum value calculator 3 having an input terminal equal to the number k of divisions of the consequent copy set and integrated. The integrated inference result is input to the defuzzification hardware 9.

First, the bias value is added by the bias applying unit 5. The applied bias value can be set to have a size and shape according to the purpose by the bias generator 4. The bias value may be a negative value. The inference result to which the bias value is applied is input to the weighted adder 6 and the simple adder 7, the output thereof is further input to the divider 8, and the centroid value is calculated as the output. This value of the center of gravity becomes the final definite output, and this value is expressed by the following equation (2).

[0026]

[Equation 2] Here, Δμ _i represents the bias value applied to the address _i of the consequent unit set.

FIG. 2 shows the concept of the inference algorithm according to this embodiment.

According to the present embodiment, if a constant is used as Δμ _i , a function of pulling the center of gravity value to the center of the consequent part occurs as shown in FIG. The smaller the size of the inference result grade, the stronger the effect. Therefore, when the grade of the inference result is small and the influence of noise and error cannot be denied, the center of gravity value is located near the center of the consequent part.

In fuzzy control, which is a typical application example of fuzzy reasoning, in particular velocity type fuzzy control, the central part of the consequent part means that the control amount is 0, that is, the present state is maintained, and the present invention is used. This can improve erroneous control due to noise and errors.

Further, as shown in FIG. 5, even when the input does not match any inference rule at all, the numerator of the equation (2),
Both denominators do not become 0, and no abnormal output from the divider 8 occurs. Further, since the output at this time is in the center of the consequent part, the control amount is 0 in the example of the fuzzy control.
This means that when there is no matching inference rule,
By not controlling anything, it is in line with human natural judgment.

Further, if Δμ _i is set to various sizes and shapes instead of a constant, the effect of attracting the center of gravity value not only at the center of the consequent part but also at an appropriate position according to the grade of the inference result is produced. Can be made

In the present embodiment, as a concrete calculation method of the bias applying unit 5, for example, maximum value calculation can be performed in addition to addition by the algebraic sum. In the inference operation before non-fuzzy, the minimum value calculator 2 may be replaced with an algebraic product calculator and the maximum value calculator 3 may be replaced with an algebraic sum calculator.

[0033]

As described above, the present invention has the following effects.

When the fuzzy inference is implemented as hardware, it is possible to improve the abnormality of the deterministic output caused by the noise or the error generated in the grade of the inference result.

It is possible to prevent an abnormality in the deterministic output that occurs when the input value does not match any inference rule at all.

The definite output can be calculated according to the grade size of the inference result.

Since the margin for noise and error becomes large, the hardware can be easily developed and eventually a low cost product can be realized.

Even if the output precision of the inference result of the inference operation block in the front is changed, it can be dealt with by adjusting the size and shape of the bias value, and one kind of defuzzification hardware can be used for various purposes.

[Brief description of drawings]

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

FIG. 2 is a conceptual diagram of an inference algorithm according to an embodiment of the present invention.

FIG. 3 is an explanatory diagram of a centroid value according to a grade of an inference result.

FIG. 4 is an explanatory diagram showing an abnormal output due to noise and an effect of the present invention.

FIG. 5 is an explanatory diagram showing an effect of applying a bias value when an input does not conform to an inference rule.

FIG. 6 is an explanatory diagram showing a change in the center of gravity value depending on the grade of the inference result when a bias value is applied.

FIG. 7 is a conceptual diagram of fuzzy inference based on minimum value calculation-maximum value calculation-centroid method.

FIG. 8 is an explanatory diagram of an integrated inference result.

[Explanation of symbols]

1 fitness calculator, 2 minimum value calculator, 3 maximum value calculator, 4 bias generator, 5 bias applier, 6 weighted adder, 7 simple adder, 8 divider, 9 defuzzification hardware

Claims (1)

[Claims] 1. In a defuzzification device of a fuzzy inference device in which a consequent part of fuzzy inference is represented by a fuzzy set, means for applying a bias value of an arbitrary magnitude to each inference output grade, and after applying the bias value. And a means for calculating the barycentric value of each inference output grade.