JPH0293939A - Fuzzy inference machine - Google Patents

Abstract

PURPOSE:To simplify constitution and to execute rapid processing by using the center position of a membership function outputted from a consequent part and adaptation to calculate a determined output. CONSTITUTION:When inputs (a) to (c) corresponding to language values (A) to (C) are inputted, the adaptation of a membership function corresponding to an input value is outputted from a condition part membership function generator MFC and the output value h1 of a condition part AND is outputted as the output value of an inference part 3 on the basis of the adaptation. Simultaneously, the output value h1 is inputted to a multiplier 4 and multiplied by a value omega1 expressing the center position of the membership function corresponding to a conclusion part language value Y and the multiplied value is outputted as the other output value omega1.h1 of the inference part 3. All values omega.h and (h) expressing the center positions outputted from respective inference parts 3 are added and added values are inputted to a defuzzy fire part 5 respectively as SIGMAomega.h and SIGMAh. The defuzzy part 5 divides the inputted SIGMAomega.h by SIGMAh to find out and output a determined value output H. Thus, the constitution can be simplified.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention This invention relates to a fuzzy inference machine that performs fuzzy logic.

(Summary of the Invention) This invention provides a fuzzy ltt theory machine that uses a plurality of rules and membership functions to perform fuzzy inference by the M in -M a x method, in which the normally used membership functions are left and right. By focusing on symmetry and calculating the deterministic output using the center position and goodness of fit of the membership function output from the consequent, the configuration has been simplified compared to the conventional centroid method. At the same time, the processing speed is increased.

(Prior art and its problems) Conventionally, when performing fuzzy tit theory in hardware, the Min-Max method is used as the inference method ll1f[ to synthesize a plurality of discretized output values obtained from the consequent. A known method is to use the center of gravity method on the obtained output to determine the final value.

FIG. 5 conceptually shows this method, and the inference section 3 provided for each rule has a conditional language value
, 13. C is given. When the human powers a, b, and c corresponding to rg and c are input to this linguistic hani, the fitness of the membership function corresponding to +il directly; W is output from the condition part membership function generation 23yFc, The output value of the condition part conjunction is sent to the conclusion part conjunction based on its fitness.

On the other hand, from the conclusion part membership leap number output 2SM F G, a membership function corresponding to the set conclusion part language (/IY) is sent as a conclusion part logical product.

In the logical product of the conclusion section, the membership function can be obtained from the function generator M1. Send to section 2.

In the defuzzifier section 2, the other ;1[4
The resulting figure is displayed, its center of gravity is calculated, and it is determined (/j) and output as CG.

Furthermore, FIG. 6 shows a detailed explanation of the inference part I (C), and the conditional part function (p×1, input?, x3 is synthesized to obtain the °C conditional logical product W. By converting the triangular membership function m, <y) obtained from the partial membership function generator MFG into stroke numbers and performing an AND operation with W for each element, membership functions m', (y) is obtained. Here, calculations are performed for the number of elements, and output terminals for that number are required.Furthermore, the logical sum and f4 calculation of definite values in the defuzzifier 112 are also discretized. As many operations as there are elements are required.

Furthermore, in conventional methods, when executing production rules, each rule requires the same number of outputs as the discretized elements of the output function, which increases the scale of the circuits that make up these rules. , there is a problem of high cost.

Furthermore, when attempting to perform fuzzy HL 2 remainders using software, the same processing as described above is performed, resulting in large-scale calculations that hinder speed-up of the processing.

(Invention 1-1) This invention was made to solve the problem described in 1-1. 7H7, the purpose is to provide a fast Noasi inference machine.

(Structure and Effects of the Invention) In order to achieve the two objects of this invention, a plurality of rules and member top functions are used to create
In a fuzzy JLT theory machine that performs fuzzy inference using the law, there is a means for setting the center position of a bilaterally symmetrical membership function in the consequent part of each rule, and a means for setting the center position of a bilaterally symmetrical membership function in the consequent part of each rule. Next, calculate the means for outputting the fitness of the meyuha ship function, the center ( ) 7 of the consequent membership function indicated by the distance, and the corresponding fitness for each rule, and calculate the sum of the products by ( A means for calculating the sum of the fitness for each rule obtained from the fitness output 1 and the stage, dividing the sum of the products of the center position and the fitness by the sum of the fitness; The present invention is characterized in that it includes means for outputting the obtained value as a final value.

The fuzzy inference machine according to the present invention uses the center position of the symmetrical membership function output from the consequent part and its fitness to calculate the final output value, so that the output from each rule is 1t- 1, and the configuration is simpler than the conventional centroid method.

As a result, when a fuzzy inference machine is configured with hardware, the circuit becomes smaller, making it possible to save space and reduce costs.

Furthermore, when the process is performed using software, the number of calculations is reduced, and high-speed processing becomes possible.

Roughly speaking, only one setting is required for setting the membership function in the consequent part, and the setting becomes a simple IP.

(Description of Embodiments) FIG. 1 is an explanatory diagram of the height method used as a method of definite output calculation in this invention.

In Figures a and 2, there is a large chi 1αx6 that was manually applied to the lit theory department.
+y0 gives the goodness of fit for each corresponding membership function. The logical product is represented as a trapezoidal membership function Ω with a diagonal line downward from the height h, as shown in FIG.

The inference result Ω of this trapezoid is the product ω of the center position ω and the height h
・It can be expressed by h.

In this way, the inference result Ω obtained for each rule is
Assuming that ω1·h respectively and the determined output to be obtained is 11, H can be obtained by the following equation.

Σh.

However, i: Rule number The final output H obtained here is evaluated based on the degree of fitness. In particular, if the consequent membership function is bilaterally symmetric, the center position ω is determined by the height There is no audience response to h.

In other words, this height method can be applied to all fuzzy inference machines in which the consequent membership functions used in fuzzy inference are symmetrical.

FIG. 2 is a diagram showing the configuration of the fuzzy inference machine of the present invention.

In the figure, the inference section 3 provided for each rule is given condition section language values A, B, and C, respectively. Input a corresponding to this language value A, B, C.

When b and c are manually input, the conditional membership function generator MFC outputs the degree of fitness of the membership function corresponding to the human value, and based on the degree of fitness, the output value of the logical product of the conditional part is output. h1 is output as the output value of the inference section 3. At the same time, the output value h1 is input to the multiplication 234, and the value ω representing the center position of the member subfunction corresponding to the conclusion language value Y
, and the other output value of the public opinion section 3 is ω,・h,
is output as

The values ω·h and h representing the center position outputted from each inference unit 3 are all added together and input to the defuzzifier unit 5 as Σω·h1Σh, respectively.

The defuzzifier unit 5 divides the input Σω·11 by Σh to obtain and output a fixed value output H.

Comparing this illustrated example with the conventional example, in the inference section, the conclusion section logical product section and the condition section membership function generation 2SMFG are undesirable, and moreover, the number of output terminals and wiring is two each. Further, the defuzzifier section 5 can also be constructed from the previous divider, and the entire fuzzy inference machine can be simplified.

FIG. 3 is a circuit diagram showing the main parts when this invention is realized by an analog circuit.

In the figure, the inference unit 31 outputs the output value as a voltage h1 from the logical product of the conditional part (not shown). This voltage R is applied to the tit resistor 71 with a resistance value R, and a current h
+/R is output. At the same time, the voltage and the resistance value are
/ω1(l(also added to resistor 81, current is generated,・ω+/
r is output. Note that the resistance value of the resistor 81 "/ω,"
The linguistic value set in this consequent, that is, does not represent the center position of the membership function (directly ω), but is a value inversely proportional to ω.

In this way, the currents output from the C8 inference section are summarized and sent to the divider 6 as a total current Ik representing the degree of conformity and a total sum 1.1 of currents representing the product of the degree of conformity and the center position. Man-powered. The divider 6 divides the input current Ink by the current 1 and outputs the result as a final value.

FIG. 4 is a circuit diagram showing another embodiment of the lit logic section. The figure shows a case where the linguistic value set in the consequent part is variable, and the linguistic value Y is set to P Ll) M, PS, Z
r1. To set NS, NM, NL, tittJL PL- is inversely proportional to the value representing their center position.
r□, . . . rNL are connected in parallel and further connected to the analog multiplexer 9. When the language value Y is input to the analog multiplexer 9, the set value can be made variable by selectively connecting a resistor according to the content.

With the above configuration, this embodiment has an equivalent fuzzy Ilk circuit with approximately one-third the circuit scale compared to the conventional one.
It was possible to construct a fuzzy inference machine, and the size and cost of the fuzzy inference machine itself could be reduced.

In addition, in the example, the fuzzy 11 configured by hardware was only explained as a logic machine, but even if configured by software using the same processing method, the number of processing operations will be reduced, and the processing speed will be increased accordingly. I was able to do that.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of the height method used in this invention, Fig. 2 is a diagram showing the configuration of a fuzzy inference machine according to this invention, and Fig. 3 is a main part when this invention is realized with an analog circuit. 4 is a circuit diagram showing another embodiment of the inference section, and FIGS. 5 and 6 are diagrams showing the configuration of a conventional example. 3... Inference section 4... Multiplier 5... Defuzzifier section 6... Divider 9... Analog multiplexer 31.32・・・
...Inference section 71.72...Resistance 81.82...Resistance

Claims (1)

[Claims] 1. Using multiple rules and membership functions, Min
- In a fuzzy inference machine that performs fuzzy inference using the Max method, there is a means for setting the center position of a symmetrical membership function in the consequent part of each rule, and based on the conditions and input values of the antecedent part, a means for outputting the degree of fitness of the membership function; a means for multiplying the center position of the consequent membership function indicated by the distance by the corresponding degree of fitness for each rule; and a means for calculating the sum of the products; and the degree of fitness. means for calculating the total sum of fitness degrees for each rule obtained from the output means; and dividing the sum of the product of the center position and the fitness degree by the sum of fitness degrees;
A fuzzy inference machine characterized by comprising means for outputting an obtained value as a definite value.