JP2767603B2 - Fuzzy inference apparatus, antecedent part membership function processing apparatus, consequent part membership function processing apparatus, fuzzy inference method, antecedent part membership function processing method, and consequent part membership function processing method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a fuzzy inference apparatus that outputs a fuzzy inference result according to the state of an input signal.

(B) Summary of the Invention The present invention relates to a highly flexible fuzzy inference apparatus that can automatically define an appropriate membership function by adapting to an input signal.

(C) Prior Art Generally, when controlling an automatic control system or the like using a fuzzy inference device, a membership function according to a control rule and characteristics of the control system is set. These rules and membership functions are mainly determined by the operator based on experience.

Even for such a control system that is not represented by a simple mathematical expression or the like, a so-called expert system can be configured by a fuzzy inference apparatus using rules and membership functions based on empirical knowledge possessed by a human operator.

(D) Problems to be Solved by the Invention However, when rules and membership functions are determined based on empirical knowledge possessed by human operators, knowledge is hierarchized and grouped, but it is troublesome to describe all of them. And it was difficult to acquire knowledge. In order to automatically perform this hierarchy of knowledge, it is necessary to make judgments corresponding to various situations into knowledge and input them. However, such macro knowledge is advanced, and it has conventionally been difficult to describe it naturally. Even if a fuzzy inference device is used, there is a problem that a large-scale system using a large number of rules and a large number of membership function generators cannot be formed and a compact control device cannot be configured.

The present invention focuses on the fact that the scaling of the range in which the membership function is defined has a great effect on the control performance, together with the rules based on the language value, and introduces the knowledge of this scaling to make the rules more macroscopic. An object of the present invention is to provide a fuzzy inference device which can be described as knowledge.

(E) Means for Solving the Problems The invention according to claim (1) of the present invention provides an antecedent basic membership function generating means for generating a basic antecedent basic membership function for each input signal. An antecedent scale determining means for determining the scale of the variable axis of the antecedent basic membership function or the scale of the input signal for the antecedent basic membership function according to the history of the input signal during the past predetermined time; Antecedent part membership function, antecedent part membership value calculation means for obtaining antecedent part membership value based on antecedent part scale and input signal, and consequent part basic membership function for each input signal The means for generating the basic membership function of the consequent part, and the scale of the variable axis or the scale of the final value of the consequent part basic membership function are determined based on the scale of the antecedent part. And consequent scale determining means, characterized by comprising the consequent basic membership function, a fuzzy relation computing section for obtaining the determined value based on the consequent scale and antecedent membership values, the.

The invention according to claim (2) provides an antecedent part basic membership function generating means for generating an antecedent part basic membership function for each input signal, and according to a history of the input signal during a predetermined time in the past. Means for determining the scale of the variable axis of the antecedent basic membership function or the scale of the input signal to the antecedent basic membership function; antecedent basic membership function; antecedent scale; And antecedent part membership value calculating means for obtaining antecedent part membership value based on the input signal.

According to a third aspect of the present invention, there is provided an antecedent basic membership function generating means for generating a basic antecedent basic membership function for each input signal, and according to a history of the input signal during a predetermined time in the past. Antecedent scale determining means for determining the scale of the variable axis of the antecedent basic membership function or the scale of the input signal to the antecedent basic membership function; and the consequent basic membership as a basis for each input signal Consequent part basic membership function generating means for generating a function; consequent part scale determining means for determining the scale of the variable axis or the fixed value scale of the consequent part basic membership function based on the antecedent part scale; And a fuzzy relation operation unit for obtaining a final value based on the consequent part basic membership function, the consequent part scale and the antecedent part membership value.

The invention according to claim (4) is characterized in that the antecedent scale determining means is constituted by fuzzy inference means for performing fuzzy inference from the history of the input signal during a predetermined past time to determine the antecedent scale. And

The invention according to claim (5) is characterized in that the antecedent part scale determining means is constituted by fuzzy inference means for performing fuzzy inference from the history of the input signal during the past predetermined time to determine the antecedent part scale. And

The invention according to claim (6) is that the scale of the variable axis of the antecedent basic membership function or the scale of the input signal for the antecedent basic membership function for each input signal during the past predetermined time is determined. After it is determined according to the history of the input signal, the antecedent membership value is determined based on the antecedent basic membership function, antecedent scale, and the input signal. Determine the scale of the variable axis or the final value scale of the basic membership function based on the antecedent scale, and determine the final value based on the consequent basic membership function, consequent scale and antecedent membership value Is obtained.

The invention according to claim (7) is that the scale of the variable axis of the antecedent basic membership function or the scale of the input signal to the antecedent basic membership function for each input signal during the past predetermined time is determined. It is determined according to the history of the input signal, and then the antecedent membership value is determined based on the antecedent basic membership function, antecedent scale and the input signal.

(F) Operation FIG. 1 shows a block diagram of a fuzzy inference apparatus which is a configuration example of the present invention.

In FIG. 1, V1 and V2 are input signals, respectively, and the antecedent scale determining means 1 and 2 determine the antecedent scale, that is, the variable axis of the antecedent basic membership function according to the history of the input signals V1 and V2. Determine the scale or scale of the input signal to the antecedent membership function. The antecedent basic membership function generating means 3 and 4 generate a basic antecedent membership function for each of the input signals V1 and V2. The antecedent part membership value calculating means 5 and 6 are the antecedent parts from the antecedent part basic membership function generating means 3 and 4, the antecedent part scales SV1 and SV2 and the input signals V1 and V2. Find the membership value. The consequent part scale determining means 7 determines the consequent part scale, that is, the variable axis or the fixed value scale of the consequent part basic membership function, by the antecedent part scales SV1 and SV2. The consequent part basic membership function generating means 8 generates a membership function that is the basis of the consequent part. The fuzzy relation calculation unit 9 outputs a final value Y from the consequent part basic membership function, the consequent part scale SD and the antecedent part membership value. The fuzzy relation operation unit 9
In the figure, there is provided a consequent part scale changing means for changing the scale of the variable axis or the fixed value scale of the consequent part basic membership function.

Means for generating a basic membership function of the antecedent part shown in FIG.
3, 4 or the consequent part basic membership function generating means 8 is defined as a ratio (%) to the variable domain, for example, as shown in FIGS. 2 and 3, respectively.

Membership value is y, variable is x, membership function is μ, negative scale is scale1, positive scale is sc
Assuming ale2, the scale can be expressed by the following relationship:

Here, a is a set point, and when there is no set point, a = 0.

Next, a specific example of the relationship between the input signal and the scale using the fuzzy inference apparatus configured as described above will be described.

FIG. 4 shows an example in which the liquid temperature is quickly raised from 0 ° C. to 100 ° C., and then process control is performed to keep the temperature at 100 ° C. FIG. 5 shows a rule representing the state of the output signal Y according to the state of the input signals V1 and V2.

In FIG. 5, V1 is the liquid temperature, V2 is the differential value of the liquid temperature,
The symbols for the liquid temperature V1 are: NB: significantly below the set temperature NS: slightly below the set temperature ZR: almost set temperature.

PS: The temperature is slightly higher than the set temperature. PB: Indicates that the temperature is significantly higher than the set temperature.

 The symbols for temperature change V2 are NB: temperature is dropping significantly NS: temperature is dropping slightly ZR: temperature is hardly changing PS: temperature is rising slightly PB: temperature is rising Means rising.

In addition, for the output (Y), which is the content of the matrix, each symbol is: PB: Increase the amount of heating PS: Increase the amount of heating slightly ZR: Maintain the amount of heating NS: Decrease the amount of heating slightly NB: Decrease the amount of heating Means let.

Note that NB, NS, ZR, PS,
PB is called a label.

In FIG. 4, the scale (−100% to + 100%) of the input signal V1 is set to So, that is, 0 ° C. to 200 ° C., during the period of t0−t1, and after t1, the scale is set to S1, ie, 90 ° C. to 110 ° C. You. As described above, during the period in which the temperature as the input signal greatly changes, the scale of the antecedent part is widened to control the transition period. Since the temperature stabilizes and the temperature change decreases during the subsequent regular period, the fuzzy inference rule is applied on a scale based on the change width. Therefore, during the settling period, more stable control is performed near the set temperature (100 ° C.). Thereafter, for example, when the input value becomes larger than the previous scale due to disturbance at t2, the scale is expanded so that the input signal falls within the scale as shown in S2. Thus, the disturbance can be quickly converged.

The scale of the consequent part is determined from the scale of the antecedent part. To change the scale of the consequent part is to change the target area (the range of the heating output), but if the target has some degree of linearity, change the scale of the consequent part. And use the same rules.

As described above, control suitable for the state of the input signal can be performed without increasing the number of rules and the number of membership functions.

(G) Embodiment FIGS. 6 and 7 show a block diagram and a processing procedure of a control unit constituting the antecedent scale determining means 1 and 2 and the consequent scale determining means 7 shown in FIG. It is a flowchart.

In FIG. 6, the temperature sensor 10 detects the liquid temperature and outputs a signal V1.
Occurs. Differentiating circuit 11 differentiates the signal to generate signal V2. The CPU 20 controls the control of the entire control unit. R
The program is written in OM21 in advance. RAM2
2 is used as a working area such as a buffer. The A / D converter 24 converts the signal V1 into digital data, and the CPU 20 reads the data via the I / O port 23. The DA converters 26, 27, and 29 convert various scale data output from the I / O ports 25 and 28 into voltage signals and output them.

As shown in FIG. 7, at the sampling timing, the CPU 20 reads the signal V1 and performs a differential operation together with the signal V1 to obtain a temperature change (data corresponding to the signal V2) (n1 to n3). Subsequently, the scales SV1 and SV2 of the antecedent part are obtained by fuzzy inference using V1 and V2 as input variables, and output (n4 → n5). As a fuzzy inference rule here, for example, “temperature is past n
If the temperature was "around the set temperature" but the temperature change was "somewhat large" this time (because the temperature is still likely to change from the set temperature), the temperature scale SV1 is "slightly widened". ] Can be described. Here, "○○
“○” is a language value, and the target state can be naturally expressed by such a rule.

Subsequently, in step n6 in FIG. 7, the scale SD of the consequent part is obtained by fuzzy inference using the scales SV1 and SV2 already obtained as input variables, and this is output in n7.

FIG. 8 shows a fuzzy inference rule in the operation of step n6 shown in FIG. Where the scale SV1,
For the SV2 and the consequent part (SD) of the matrix, each symbol means S: Narrow scale width N: Medium scale width B: Scale width is wide.

In this way, when the temperature scale SV1 is narrow (it is near the set temperature) but the temperature change scale SV2 is medium (because a sudden change in temperature is caused by disturbance), The disturbance response becomes faster by making the scale SD wider. When the temperature scale SV1 is medium (converging or vibrating) and the temperature change scale SV2 is wide (because it is vibrating or exceeding the set value), the operating amount scale SD is medium. To When both the temperature scale SV1 and the temperature change scale SV2 are narrow (because they are almost settled), the scale of the manipulated variable SD is narrowed to perform fine control near the set temperature.

By expressing the fuzzy inference rule using the scale of the antecedent part as an input variable and the scale of the consequent part as an output variable, it is possible to cope with the dynamic correlation of the control system.

Next, FIG. 9 shows a specific configuration of the antecedent part basic membership function generating means 3, 4 and the antecedent part membership value calculating means 5, 6 shown in FIG. The concrete configurations of the consequent part basic membership function generating means 8 and the fuzzy relation calculating part 9 are described below.
Figure 10 shows.

As shown in FIG. 5, the antecedent membership value calculation unit for generating the membership value in the antecedent part calculates the membership value in the antecedent part in order to output the inference result for each rule according to each inference rule. It has a membership function generator for calculating. Since each antecedent part membership value operation part is provided for each rule, a total of seven are provided. FIG. 9 shows an antecedent part operation unit for obtaining a specific antecedent part membership value. As shown in the figure,
It has membership function generators 50 and 51, and a label NS corresponding to the temperature V1 and a label PS corresponding to the temperature V2 are input to each membership function generator. Membership function generator
When this label is input, 50 and 51 generate a basic membership function corresponding to the label. The scale changing circuits 52 and 53 provide the input signals V1 and V2 in response to (for example, in proportion to) the scale signals SV1 and SV2 with respect to the input signals V1 and V2.
By changing the level of V2, a signal corresponding to the proportion (%) of the antecedent scale is generated. In other words, the scale change circuits 52 and 53 convert the input signals V1 and V2 to the scale signal SV.
1. Normalize to a predetermined scale by SV2. The antecedent AND circuit 54 selects a smaller membership value according to the mini rule of the mini-max rule.

The consequent part operation part is a label corresponding to the smaller membership value X1 to X7 of the antecedent part for each rule as shown in FIG.
ZR, NS ... PB is the consequent part basic membership function generator 60,
61 ... 62 are input. Consequent part AND circuit 63, 64 ...
Numeral 65 truncates the head based on the membership values X1, X2,... Xn output from the membership function generators 60, 61,.
The logical sum circuit 66 is a part for calculating the max rule of the mini-max rule. The logical sum of each trapezoidal output (inference result) is obtained, and the determinate value calculating circuit 67 obtains the position of the center of gravity of the logically summed area.
This final value is obtained as a ratio (%) on the consequent part scale. The scale changing circuit 68 outputs the control amount Y based on the final value obtained as the “ratio” and the consequent part scale signal SD.

In the embodiment described above, the antecedent scale was determined by performing fuzzy inference from the history of the magnitude of the input signal.However, when there is no correlation between the input variables, determination using a method other than fuzzy inference is performed. can do.

For example, the antecedent scale is set so that the largest value of the absolute value among the past n sampling times of the input signal is 60% of the antecedent scale. However, if the current input is larger than the previous scale,
Set to 0%. This corresponds to, for example, a case where t is a sampling period in FIG. 4 and T is a period of the past n sampling times. This is expressed by the following equation.

By determining the scale of the antecedent part in this way,
While the control system is settling, the scale gradually narrows, and fine judgment and control can be performed near the set value. Also, when the state changes greatly due to disturbances or characteristic fluctuations, the scale rapidly increases with that,
Sufficient response is possible.

In the embodiment, the scale of the consequent part is determined by fuzzy inference based on the scale of the antecedent part. However, if the object has a certain degree of linearity, other methods are possible. is there.

For example, the average of the ratio of the current value to the initial setting value of the antecedent part scale is taken, and the average is set to the ratio to the initial setting value of the consequent part scale.

 This can be expressed by the following equation.

In addition, a weighted average of the ratio of the current value to the initial setting value of the antecedent part scale may be obtained and determined as the ratio to the initial value of the consequent part scale. This can be expressed as follows.

In the embodiment, as a method of calculating the antecedent part membership value, the level fluctuation is given to the input signal based on the antecedent part scale signal. It is also possible to change the antecedent scale by controlling the variable axis of the basic membership function.

Further, in the embodiment, the level is changed with respect to the final value based on the consequent part scale signal. Conversely, the variable axis of the consequent part basic membership function is changed based on the consequent part scale signal. You may make it fluctuate.

(H) Effects of the Invention According to the invention described in claims (1) and (6), the scale of the antecedent part is determined based on the history of the input signal during the past predetermined time, and the scale of the antecedent part is determined from the scale of the antecedent part. Because the scale of the subject is determined, when setting the knowledge, it is possible to automatically reconfigure the knowledge according to the situation simply by setting the basic membership functions and rules,
Since it is not necessary to use a large number of rules and membership functions, the entire device can be downsized.

According to the inventions described in claims (2), (3) and (7), the membership function is automatically corrected and used according to the situation, so that the number of membership functions to be stored is reduced. can do.

According to the invention described in claims (4) and (5), knowledge can be hierarchized by using fuzzy inference at the stage of determining the scale of the antecedent part and the scale of the consequent part, It can be expressed more naturally as higher-order knowledge, and setting and adjustment of control system performance becomes easier.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration example of the present invention. 2 and 3 are diagrams showing examples of the basic membership function according to the present invention. FIG. 4 and FIG. 5 are diagrams showing a response example and a fuzzy inference rule of the process control system for explaining the operation of the present invention. 6 and 7 are a block diagram of a control unit corresponding to the antecedent part scale determining means and the consequent part scale determining means according to the present invention, and a flowchart showing the processing procedure thereof. FIG. 8 is a diagram showing a fuzzy inference rule used when determining the consequent part scale according to the embodiment. FIG. 9 and FIG. 10 are block diagrams showing examples of the antecedent operation unit and the consequent operation unit. 50,51… Antecedent basic membership function generator, 60 to 62… Consequence basic membership function generator, SV1, SV2… Antecedent scale signal, SD… ) Scale signal.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 9/44 G05B 13/02

Claims (7)

(57) [Claims] 1. An antecedent part basic membership function generating means for generating an antecedent part basic membership function for each input signal, and an antecedent part basic function according to a history of the input signal during a predetermined past time. Antecedent scale determining means for determining the scale of the variable axis of the membership function or the input signal to the antecedent basic membership function; and based on the antecedent basic membership function, antecedent scale and the input signal. Antecedent part membership value calculating means for obtaining antecedent part membership value; consequent part basic membership function generating means for generating a basic consequent part basic membership function for each input signal; antecedent part scale Consequent part scale determining means for determining the scale of the variable axis or the scale of the final value of the consequent part basic membership function based on the consequent part basic membership function Fuzzy inference apparatus comprising a fuzzy relation computing section for obtaining the determined value based on the consequent scale and antecedent membership values, the. 2. An antecedent part basic membership function generating means for generating an antecedent part basic membership function for each input signal, and an antecedent part basic function in accordance with a history of the input signal during a predetermined time in the past. Antecedent scale determining means for determining the scale of the variable axis of the membership function or the input signal to the antecedent basic membership function; and based on the antecedent basic membership function, antecedent scale and the input signal. An antecedent part membership value processing means for obtaining an antecedent part membership value; 3. An antecedent part basic membership function generating means for generating an antecedent part basic membership function for each input signal; and an antecedent part basic function according to a history of the input signal during a predetermined time in the past. An antecedent scale determining means for determining the scale of the variable axis of the membership function or the input signal to the antecedent basic membership function, and after generating the consequent basic membership function as a basis for each input signal Consequence part membership function generation means, Consequence part scale determination means for determining the scale of the variable axis or fixed value scale of the consequent part basic membership function based on the antecedent part scale, Consequence part basic member A fuzzy relation operation unit for obtaining a definite value based on the ship function, the consequent part scale and the antecedent part membership value; 4. The fuzzy inference means according to claim 1, wherein said antecedent scale determining means is constituted by fuzzy inference means for performing fuzzy inference from a history of input signals during a predetermined time in the past to determine an antecedent scale. Inference device. 5. An antecedent part scale determining means comprising fuzzy inference means for performing fuzzy inference from a history of input signals during a predetermined time in the past to determine an antecedent part scale. Case part membership function processing unit. 6. A scale of a variable axis of an antecedent part basic membership function or an input signal scale for an antecedent part basic membership function for each input signal is stored in a history of the input signal during a predetermined time in the past. Then, the antecedent part membership value is determined based on the antecedent part basic membership function, antecedent part scale, and input signal. Change the scale of the variable axis or the scale of the confirmed value.
A fuzzy inference method which is determined based on an antecedent scale and obtains a final value based on a consequent basic membership function, a consequent scale and an antecedent membership value. 7. A scale of a variable axis of an antecedent part basic membership function or an input signal scale to an antecedent part basic membership function for each input signal is stored in an input signal for a past predetermined time as a history. And determining an antecedent membership value based on an antecedent basic membership function, an antecedent scale and an input signal.