JPH0256602A - Fuzzy inference device - Google Patents

Abstract

PURPOSE:To output an operation value by a circuit constitution of a small scale by leading digitized observation data to an address input, providing a memory part for showing an operation value by outputting its data, and storing in advance data corresponding to a method for a fuzzy inference and a method for a reverse fuzzy conversion in the memory part. CONSTITUTION:By leading digitized observation data through A/D converters 14, 15 to an address input of a memory part 11 in which data corresponding to a method for a fuzzy inference and a method for a reverse fuzzy conversion is stored in advance, an operating value corresponding to the observation data appears in an output of a memory part 11. The observation data in the fuzzy inference and on inference value correspond to each other to one-to-one, there fore, by the memory part 11 in which the data corresponding to the method for the fuzzy inference and the method for the reverse fuzzy conversion is stored in advance, a data conversion of the observation data is executed, and also, a result of its conversion is used as an operation value. In such a way, even when a kind of the observation data and the number of rules are increased, the operation value can be obtained from the observation data by a circuit constitution of a small scale.

Description

[Detailed Description of the Invention] The present invention relates to a fuzzy inference device that performs fuzzy inference based on observed data, and more specifically, the present invention relates to a fuzzy inference device that performs fuzzy inference based on observed data, and more specifically, it relates to a fuzzy inference device that performs fuzzy inference based on observed data, and more specifically, it performs fuzzy inference and de-fuzzification using a memory unit. This invention relates to a fuzzy inference device.

In recent years, it has become possible to respond extremely quickly to changes in events by determining the output of a sensor that detects speed or angle, for example, with a range of speeds such as fast, slightly fast, etc. Development of a fuzzy inference device has begun as a device that makes it possible to do this.

Generally, in the above-mentioned fuzzy inference devices, hardware is used to generate membership values determined in response to observation data, so each sensor has hardware to generate membership values corresponding to each sensor. It is provided.

In addition, in order to perform an inference called the MAXMIN method, for example, among the outputs sent from each of the above-mentioned hardware prepared above, a plurality of memberships derived for each of the plurality of rules is required. A processing circuit is provided for outputting the minimum value of the values, and a processing circuit is provided for outputting the maximum value of the outputs of the plurality of processing circuits,
This maximum value is taken as the inference value in fuzzy. A configuration is adopted in which an arithmetic circuit is provided to perform defuzzification based on this inferred value, and control is performed by the output of this arithmetic circuit.

In the above device, one processing circuit is required for each type of observation in order to obtain a membership value based on one rule. On the other hand, when performing fuzzy inference,
Since membership values based on multiple rules are required and it is common to use two or more types of observed data, membership values are generated by multiplying the number of rules by the type of observed data. The indicated number of processing circuits are required.

Also based on these membership values MAXMIN
When the types of observed data and the number of rules increase, the hardware for executing the method can respond to this increase by increasing the number of rules.
The scale of the circuit that performs fuzzy inference has increased, and a problem has arisen in that the scale of the arithmetic circuit that performs de-fuzzification has to be similarly increased.

The present invention was created to solve the above problems, and its purpose is to be able to output manipulated values with a small-scale circuit configuration even when the types of observation data and the number of rules increase. The object of the present invention is to provide a fuzzy inference device.

In order to solve the above-mentioned problems of Ruta, the fuzzy inference device of the present invention is configured to perform fuzzy inference based on two or more pieces of observation data indicating movement, and output a manipulated value for movement. The system is equipped with a memory section in which the converted observation data is guided to an address input and the operation value is indicated by the output of the data, and data corresponding to the fuzzy inference method and the de-fuzzification method are stored in advance in the memory section. .

, when focusing on the relationship between the value of observation data indicating the work and motion and the operation value indicating the value corresponding to the motion, the number of rules, the method of generating membership values, the fuzzy inference method such as the MAXM T N method, or No matter what method of calculation is used for de-fuzzification, the relationship between observed data and manipulated values is a one-to-one correspondence. This means that once the observed data value is determined, the operation value is uniquely determined. This means that a fuzzy inference device can be regarded as a conversion device that converts the value of observed data according to a certain method.

Therefore, by leading the digitized observed data to the address input of the memory section in which data corresponding to the fuzzy inference method and the de-fuzzification method are stored in advance, the output of the memory section corresponds to the observed data. The operation value will appear.

Figure 1 is a block diagram showing the electrical configuration of a fuzzy controller to which an embodiment of the present invention is applied.

In the figure, an inverted pendulum 31 is rotatably supported on a support base 33 by a support shaft 32, and the output of an angle sensor 12 that detects the angle of the inverted pendulum 31 with respect to the support base 33 is determined by the support of the inverted pendulum 31. The output of the angle sensor 12 and the output of the differential circuit 13 are provided in correspondence with each other, and the input signal is converted into an 8-bit signal. A/D converter 14 that converts into a digital signal of
．． It is guided by 15.

The outputs of these A/D converters 14.15 are led as observation data 141.151 to the upper 8 bits address input and the lower 8 bits address input of the memory section 11 constituted by a 64 kilobyte memory element. ing.

The 8-bit digital output of the memory section 11 is the operation value 1.
11. Then, via the D/A converter 16, the drive circuit 17
The output of this drive circuit 17 is led to a drive unit 18 for moving the support base 33 in the direction of arrow A.

Regarding the operation of one embodiment of the present invention having the above configuration,
This will be explained below.

As shown in FIG. 2, the angle of the inverted pendulum 31 with respect to the support base 33 is O when it is vertical, and it becomes a negative value when it is tilted towards the figure (31b, 31c), When it tilts to (31d), it becomes a positive value. Further, the angular velocity takes a positive value when the inverted pendulum 31 rotates in the direction of arrow B, and takes a negative value when it rotates in the opposite direction to arrow B. These values are converted into observation data 1, which is an 8-bit digital signal, by the angle sensor 12, the differential circuit 13 for detecting angular velocity, and the A/D converter 14.15.
41.151, it is led to the memory section 11. Then, the memory unit 11 performs fuzzy inference and de-fuzzification based on the inferred value.

FIG. 4 is an explanatory diagram showing a method for obtaining data to be stored in the memory section 11. In the following, the operations of fuzzy inference and de-fuzzification performed by the memory unit 11 will be explained with reference to FIG. 4 as necessary.

In this embodiment, in order to prevent the inverted pendulum 31 from falling, the support stand 33 is
The configuration is such that five types of control are performed: four types of control to move the support base 33 quickly and a little faster, and a control that does not move the support base 33.

Therefore, the labels corresponding to each control are
The fifth labels 91 to 95 are named, and the first label 91 performs fuzzy inference when it is necessary to move the support base 33 quickly in the direction of arrow C, and the second label 92 performs Fuzzy inference is performed when it is necessary to move the table 33 somewhat quickly in the direction of arrow C. The third label 93 is fuzzy reasoning when there is no need to move the support base 33, the fourth label 94 is fuzzy reasoning when the support base 33 needs to be moved a little faster in the direction of the arrow, and the fifth label is 95 performs fuzzy inference when it is necessary to move quickly in the direction of the arrow.

Therefore, the inference value of the fuzzy inference in the third label 93 increases as the inverted pendulum moves toward a position perpendicular to the support base 33 and has a stronger tendency to stop at the vertical position. , approaches the value 1, which is the maximum value.

To explain this in detail with reference to FIG. 2, when the inverted pendulum 31 is at position 31a and the angular velocity is extremely close to 0, or when it is at position 31b and rotates at a slightly faster speed in the direction of arrow B. (In this case, when the inverted pendulum 31 rotates to the position 31a, the angular velocity becomes extremely close to O), or when it is at the position 31c and is rotating rapidly in the direction of arrow B. (A similar relationship holds true even when the inverted pendulum 31 is located on the right side in the drawing).

Therefore, one of the fuzzy inference rules in the third label 93 is that the change in membership value with respect to the angle is
When the membership value changes as shown in a of FIG. 3 (this shows that the inverted pendulum 31 is in a position very close to the perpendicular to the support base 33), the change in membership value with respect to the angular velocity changes as shown in b. (This means that the angular velocity is extremely close to 0). and these 2
This is a method in which the smaller of the species membership values is set as the output value of the first rule.

The second rule is that when the membership value changes with respect to the angle as shown in FIG. ), the change in the membership value with respect to the angular velocity is the change shown in d (this means that the angular velocity is rotating at a slightly faster speed in the direction of arrow B). Similarly to the first rule, the smaller of the two membership values is set as the output value.

The third rule is that when the change in membership value with respect to angle changes as shown in e, this is the inverted pendulum 31.
is located at a position extremely close to 31c with respect to the support base 33), and the change in membership value with respect to the angular velocity is the change shown in f (this is because the angular velocity rotates at a high speed in the direction of arrow B). ).

And the fourth and fifth rules are that the inverted pendulum 31 is 31
The rules for the case where the position is to the left of c and the case where the position is to the right of 31a are shown.

As described above, in each rule, if the MAXMIN method is adopted from the membership value for angle and the membership value for angular velocity, the smaller value is output first. Since there are rules up to the Nth rule, in each rule, the membership value for the angle is an1, the membership value for the angular velocity is b
If n is the output value and Cn (1≦n≦N), then the value Cn is expressed as Cn=MIN (anSbn).

Also, since all the rules are executed in parallel, these values Cn are output as N values, and the maximum value of these values becomes the fuzzy inference value, so if that value is M3, then M3 =MAX (CI, C2, . . . CN).

This value M3 is the value of the two A/
This is the only value determined by the value of the output of the D converters 14 and 15. Moreover, this value M3 approaches 1 as the degree of inverted pendulum 31 toppling decreases.

Similarly, in the first label 91, the support stand 33
A fuzzy inference tag indicating the degree to which the inverted pendulum 31 will fall if the support base 33 is not moved quickly in the direction of the arrow C is obtained, and in the second label 92, the inverted pendulum 31 will fall if the support base 33 is not moved somewhat quickly in the direction of the arrow C. A fuzzy inference price tag is obtained indicating the degree to which the fourth and fifth labels 94.
95, fuzzy inference values M4 and M5 are obtained that indicate the degree to which the inverted pendulum 31 will fall if the support stand 33 is not moved slightly or quickly in the direction of the arrow.

These fuzzy inference values ML M2, old, and M5 are uniquely determined values corresponding to the outputs of the A/D converters 14 and 15, similar to the fuzzy inference values M3. Therefore, A/
Corresponding to the output of the D converter 14.15, 5 of M1 to M5
Two values are obtained by fuzzy inference for each label 91-95.

Next, based on these values M1 to M5, inverse fuzzification, which is a calculation of the speed and direction for moving the support stand 33, is performed.

In the de-fuzzification, first, for the fuzzy inference values M1 to M5 for each label 91 to 95, those values M1 to M5 are 0 to 0.
Since the values are in the range of l, weighting required for control is performed by multiplication and summation, and then the weighted fuzzy inference values are added to each fuzzy inference value Ml −
If the weighting value for M5 is indicated by -1 to W5, and the addition result is indicated by X, the value X is X=WIXM1+W2xM
2...+W5XM5.

At this time, the fuzzy inference values M1 to M5 are values uniquely determined for the output of the A/D converter 14.15, and the values -1 to -5 are constants corresponding to weighting. , the value
．． This is the only value determined corresponding to the output of 15.

Furthermore, when adding 97 is performed to the fuzzy inference values OLD to Y of each label 91 to 95 and the result is indicated by Y, the value Y becomes Y=M1+M2...+M5. Then, perform calculations based on these values X and Y.
Denoting the result as D, the value is shown as D=X/Y-128 (the value 128 means the median value of the 8-bit data).

This value, similar to the value above, is calculated by the A/D converter 14.1.
Since this is the only value that can be determined corresponding to the outputs 141 and 151 of 5, the various operations for the above-mentioned fuzzy inference and de-fuzzification are performed for each address input value of the memory unit 11 using a computer. The result is stored in the memory section 11.

Since the data shown above is stored in the memory section 11 in advance, the output 141.1 of the A/D converter 14.15
51, when the angle and angular velocity of the inverted pendulum 31 with respect to the support base 33 are given to the memory section 11, the output of the memory section 11 is as follows corresponding to the angle and the angular velocity.
A value indicating the result of fuzzy inference and de-fuzzification appears, and that value is sent out as the operation value 111.

Since this manipulated value 111 is guided to the drive circuit 17 via the D/A converter 16, the drive circuit 19 receives the manipulated value 111.
The drive unit 18 is driven according to the following. As a result, the support stand 33
moves in the direction of arrow A in accordance with the state of the inverted pendulum 31.

Through the above operations, the inverted pendulum 31 is rotatably supported.
This state is maintained at an angle close to perpendicular to the support base 33, although it sways slightly without falling over.

Note that the present invention is not limited to the above embodiments, and the memory unit 11 stores two types of observation data 141.15.
Although the case where 1 is derived has been described, the present invention can be similarly applied to the case where 3 or more types of data are derived.

In addition, although the number of bits indicating observation data 141.151 has been explained in the case of 8 bits, it is possible to use other bit numbers, such as 6 bits or 10 bits, depending on the accuracy of the required observation data. be.

Furthermore, as the types of observation data have increased, when the number of bits in the memory section must be extremely increased, as shown in FIG. For each of the two types of observation data 811.812 and 821.822, first layer fuzzy inference is performed using the 64 kilobyte memory element 81.82 provided correspondingly, and the fuzzy inference value 813.823 is On the other hand, the memory element 83 can be configured to perform 2N-th fuzzy inference and de-fuzzification. In this structure, it is also possible to have a multilayer structure.

In addition, in response to an increase in observation data, as shown in Figure 6, observation data 841 to 8, each represented by 8 bits, are
44 through the address controller 84, for example, the 20 bit signal line in the 32 bit observation data 〇 is connected to the RAM 85 with a storage capacity corresponding to the 20 bit input.
and through the memory controller 86,
This RAM 85 is configured as a virtual memory that guides data (data corresponding to the fuzzy inference method and the de-fuzzification method) stored in advance in the auxiliary storage device 87, and the values indicated by the observed data 841 to 844 are stored in the RAM 85. When indicating a request for data not found in the RAM 85, the address controller 84 can be configured to switch the signal line leading to the RAM 85 and move the corresponding data from the auxiliary storage device 87 to the RAM 85.

In addition, to prevent the inverted pendulum 31 from falling, there are five ways to move the support stand 33 (movement at two speeds, fast and slightly fast, in each of the two directions indicated by arrow A).
Although a configuration has been described in which the inverted pendulum 31 is prevented from overturning by any type of movement method, for example, seven types of movement methods, it is possible to prevent the inverted pendulum 31 from falling. It is.

Further, although the fuzzy inference device according to the present invention is applied to a controller that prevents the inverted pendulum 31 from falling, it may also be applied to any other control device, such as a control device that controls the running of a train. is possible.

The fuzzy inference device according to the present invention has a one-to-one correspondence between observed data and inferred values in fuzzy inference, so it is possible to obtain data corresponding to the fuzzy inference method and the de-fuzzification method in advance. The stored memory unit performs data conversion of observation data and uses the conversion results as operation values, so even when the types of observation data and the number of rules increase, observation data can be easily converted using a small-scale circuit configuration. In addition, the time required from inputting observation data to outputting the manipulated value is only the delay time of the memory section, making it possible to output the manipulated value at extremely high speed, and it is possible to use an unlimited number of rules. use of labels and labels, free selection of defuzzification methods, no restrictions on precision in calculations during the process of deriving operational values from observed data, simple circuit configuration, and easy use of clocks. It has the advantage of improved reliability due to not using a prerequisite CPU, and ease of linking with computer systems because it handles digitized data, as well as the ability to prevent inferred data from electromagnetic interference. Even if a part of it is missing due to
The effect is that there is no decisive difference in the final inference conclusion value, and when the operation value is fed back to the memory unit via the data conversion memory, the shape of the membership function is modified by the feedback in an evolutionary manner. It also has the effect of being able to create the rules themselves.

[Brief explanation of the drawing]

- Fig. 1 is a block diagram showing the electrical configuration of a fuzzy controller to which an embodiment of the present invention is applied, Fig. 2 is an explanatory diagram showing the state of an inverted pendulum, and Fig. 3 is a diagram showing members for angle and angular velocity. Figure 4 is an explanatory diagram showing the state of ship values, Figure 4 is an explanatory diagram showing a method for obtaining data to be stored in the memory section, Figures 5 and 6 are diagrams showing how the memory unit is used to accommodate an increase in observation data. FIG. 2 is a block diagram showing the configuration of FIG. 11...Memory section 111...Operation value 141...Observation data 151...Observation data

Claims (1)

[Claims] (1) In a fuzzy inference device that performs fuzzy inference based on two or more pieces of observation data indicating a movement and outputs an operation value for the movement, the digitized observation data is guided to an address input, and the A fuzzy inference device, comprising a memory section that indicates the operation value by outputting data, and data corresponding to a fuzzy inference method and a de-fuzzyization method are stored in advance in the memory section.