JP3101742B2 - Minimum detector - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for detecting a minimum value from a plurality of numerical values.
The present invention relates to a minimum value detecting means which is indispensable when constructing a fuzzy expert system or the like using a ral network.

[0002]

2. Description of the Related Art In recent years, attempts have been made to realize a super-parallel operation by simulating the function of an electronic circuit by using a human brain or a nerve cell of a living body as a model, and it is generally called a neural network. It is expected to be effective in performing a process for executing many simple operations like a combination problem in a short time. On the other hand, the fuzzy theory has been introduced in the case of realizing smooth braking in the operation control of trains and trains, etc. As a typical example of fuzzy inference, the method of Mamdani is known. ing. In this method, it is necessary to detect the maximum value and the minimum value from among a large number of numerical values having a value between 0 and 1, but conventionally, an operation for judging the magnitude relationship between two numbers by a logical method is repeated. Was. However, the conventional method has a drawback that when the numerical value to be compared is large, the amount of calculation is dramatically increased and quick processing becomes difficult. Therefore, an efficient minimum value detecting means has been desired.

[0003]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and detects a minimum value from a large number of numerical values at an extremely high speed. It is an object of the present invention to provide a minimum value detector suitable for.

[0004]

SUMMARY OF THE INVENTION The present invention provides the following to achieve the above object.
Take such a configuration. That is, two values having a value between 0 and 1
Means for selecting a smaller value from the numerical values,
Two input layer intervals where each of the two values is 1
Linear element U₀₁, U₀₂And the weights are -1 and 1
One two-input intermediate layer section linear element U ₁₁And weights are 0 and 1
Of the second intermediate layer section linear element U₁₂Each of the input layer sections
Linear element U₀₁And U₀₂And the intermediate layer
Interval linear element U₁₁And U₁₂Output is weighted -1 and 1
Output layer section linear element U_{twenty one}By entering
Force layer section linear element U_{twenty one}Is smaller than the above two numbers
Output. Also, inputting each of the two numerical values
Linear elements U₀₁, U₀₂And the two sections
Linear element U₀₁, U₀₂Output is weighted by -1 and 1 respectively
And the intermediate layer section linear element U₁₁In the middle layer section
Linear element U₁₁And the second input layer section linear element U₀₂Out of
Output layer interval linear element U with weighting -1 and 1_{twenty one}Enter in
The same function with fewer elements.
It is also possible to perform
Output element interval linear element of one minimum detector
Child U_{twenty one}Is the other input layer section linear element U₀₁To be shared with
By connecting to, any number of numbers
Can be extended to detect the minimum value.

[0005]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments. First, a section linear element used in the present invention will be described. 1 (a) is a conceptual diagram illustrating a general piecewise linear elements, the piecewise linear element U is n input terminals _{i 1, i 2, ···,} i n and one output terminal in operation element having a phi, numerical I _1, I _2, I ₃ between the n 0 and 1 to the input terminal, ..., the I _n is input, w to the input of the respective _{1, w 2, w 3,} ···,
w _{n is} weighted and all the other values are added to the internal state N. The output terminal φ is set to 0 if the result of the addition is less than 0 as shown in FIG. If the number is between 1, the number is output as it is, and if it is larger than 1, 1 is output.

To explain this using mathematical formulas, the internal state N of the section linear element U is as follows: N = Σ w _i I _i (1) Using the internal state N, φ = f (N)... (2) where the internal state N and the output φ = f (N)
1 (b), that is, f (N) = 0; N <0 = N; 0 ≦ N ≦ 1.

FIG. 2 is a block diagram showing an embodiment of a minimum value detector according to the present invention using the section linear element U. In the drawing, U ₀₁ and U ₀₂ are input layer section linear elements. Here, the above-described section linear element is used as one input element, and the weight w for the input is 1
It is what it was. U ₁₁ and U ₁₂ shown in FIG. 1A are used as two-input elements as the intermediate-layer section linear elements, and their weights are w ₁₁₁ = −1 and w ₁₁₂ = 1.
And w ₁₂₁ = 0 and w ₁₂₂ = 1, and the connection is required to input both outputs of the input layer section linear elements U ₀₁ and U ₀₂ . Furthermore,
U ₂₁ is a fifth arithmetic unit as an output layer section linear element, and the weight of each of the two inputs is w _{211 as described above.}
= -1, w ₂₁₂ = 1 and connected so that the outputs of the two intermediate layer section linear elements U ₁₁ and U ₁₂ are input, and a desired signal is obtained from the output terminal of the elements.

The operation in this configuration will be described. Assuming that the inputs are I ₁ = 0.5 and I ₂ = 0.8, the internal states N ₀₁ and N ₀₂ of the interval linear elements U ₀₁ and U _{02 in} the input layer have a weight of 1 respectively. The values are 0.5 and 0.8 as they are, and their outputs also have the same value. Next, when this numerical value is input to the intermediate-layer section linear elements U ₁₁ and U ₁₂ , the internal states N ₁₁ and N ₁₂ of the elements become N ₁₁ = 0.5 × (−1) + 0.8 × 1 = 0.3 (3) N ₁₂ = 0.5 × 0 + 0.8 × 1 = 0.8 (4) According to the characteristic diagram of FIG. , Elements U ₁₁ , U ₁₂
Are 0.3 and 0.8, respectively. Further, when these two output values are weighted by −1 and 1 respectively and input to the output layer section linear element U ₂₁ , the internal state N ₂₁ is N ₂₁ = 0.3 × (−1) +0. .8 × 1 = 0.5, and a small value of 0.5 among the inputs is obtained as an output. According to this configuration, if the input numerical value is a value in the range from 0 to 1, the smaller numerical value is always output and functions as a target minimum value detector.

[0009] Figure 2 is a diagram showing a modified circuit of the embodiment described above, the fourth piecewise linear element U ₁₂ is removed, the second element U ₀₂
If it is possible just as the minimum value detection as in Example even by this modification circuit which is configured so as to connect the fifth element U ₂₁ directly. That is, FIG. 1 (a) in the leading to the second from the input layer elements U ₀₂ via the second intermediate layer element U ₁₂ of the output layer element U ₂₁ Le - In bets, weighting of each element are both 1, is not affected because the weighting of the bets is 0 be omitted said second intermediate layer element U ₁₂ results - further first input layer elements Le extending from U ₀₁ to the second intermediate layer element U ₁₂ of. Therefore, the minimum value can be similarly detected by the circuit shown in FIG. The present invention can include a plurality of minimum value detectors for the two numerical values described above and detect the minimum value from an arbitrary number. FIG. 4 is a block diagram showing a modified embodiment of the present invention. The minimum value detector S shown in FIG. 3 is provided with n-1 units for the number n of detection targets, and as shown in FIG. The output elements of the output layer section linear elements of the minimum detector are sequentially connected in cascade so as to be input to the first input layer section linear elements of the adjacent minimum value detector. According to this configuration, among the numerical values input to the respective minimum value detectors, the numerical value having the smaller value is sequentially output and compared with the next numerical value, and the numerical value finally output is n numerical values. Is the smallest of the

[0010] FIG. 5 is obtained by further simplifying the circuit shown in FIG. 4, a first input layer piecewise linear elements to be next connected to the output layer piecewise linear element U ₂₁ of one of the minimum value detector U
₀₁ is shared, and by this connection, the same function as that of FIG. 4 is achieved. In the above description, the numerical value to be compared is limited to a numerical value between 0 and 1. However, a numerical value larger than 0 is divided by, for example, the other numerical value by the maximum value in the numerical value group. From 1
After performing the pre-processing for converting to the numerical values up to the above, it is input to the above-described circuit, and the finally obtained numerical value may be multiplied by the maximum value to obtain the original numerical value. In the embodiment, the input layer elements U ₀₁ and U ₀₂ are section linear elements. However, since both have a weight of 1 and function only as an input circuit, they are not necessarily section linear elements.
Further, in general, many piecewise linear elements are provided with a bias means. However, in practicing the present invention, by setting the bias to zero, the element can be used in the same manner as described above. In the piecewise linear element used in the present invention, there is no need to limit the output value of one or more inputs. Therefore, it is not necessary to limit to the characteristics of FIG.

[0011]

Since the present invention is constructed and functions as described above, it is possible to detect the minimum value among a large number of numerical values with an extremely simple circuit configuration. Is very effective when fuzzy control is performed using Further, it goes without saying that the present invention is not limited to fuzzy control but can be widely applied.
Further, since all the elements can have the same configuration, it is convenient for forming an integrated circuit.

[0012]

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram showing an example of a section linear element used in the present invention.

FIG. 2 is a configuration diagram showing one embodiment of the present invention.

FIG. 3 is a configuration diagram showing a modified embodiment of the present invention.

FIG. 4 is a schematic configuration diagram showing another modified example of the present invention.

FIG. 5 is a schematic configuration diagram showing still another embodiment of the present invention.

[Explanation of symbols]

U ₀₁ , U ₀₂ Input layer section linear element U ₁₁ , U ₁₂ Middle layer section linear element U ₂₁ Output layer section linear element U Section linear element I ₁ I ₂ I ₃ ··· Numerical value w ₁ w ₂ w ₃ ··· Weighting value

Claims (3)

(57) [Claims] 1. Two numerical values having a value between 0 and 1 are:
In the means for selecting a small value, each of the two numerical values is weighted by two input layer interval linear elements U ₀₁ ,
U ₀₂ , the outputs of the elements U ₀₁ and U ₀₂ are weighted by −1 and 1, respectively, and the first two-input intermediate layer section linear element U
To _11, and the element U ₀₁ and the second intermediate layer piecewise linear element U ₁₂ respectively with respectively input, intermediate layer piecewise linear element U ₁₁ by weighting 0 and 1 at the output of the U ₀₂ and U ₁₂ by weighting the output is input to the output layer piecewise linear element U ₂₁ -1 and 1, the minimum value, characterized in that from the output layer piecewise linear element U ₂₁ outputs the smaller of the two numbers Detector. 2. Two numerical values having a value between 0 and 1 are:
In the means for selecting a small value, two input layer interval linear elements U ₀₁ and U ₀₂ for inputting each of the two numerical values are provided, and outputs of the two interval linear elements U ₀₁ and U ₀₂ are respectively −
By weighting 1,1 enter the intermediate layer piecewise linear element U _11, the output layer weights -1 and 1 the output of the the intermediate layer piecewise linear element U ₁₁ second input layer piecewise linear element U ₀₂ by entering the piecewise linear element U _21, the minimum value detector, characterized in that it outputs the smaller of the two numbers from the output layer piecewise linear element U ₂₁ 3. A linear element U ₂₁ comprising a plurality of minimum value detectors according to claim 1 or 2, wherein the output layer section linear element U _{21 of} one of the minimum value detectors.
Are sequentially connected so as to serve also as the other input layer section linear element U ₀₁ , whereby a minimum value is detected from arbitrary numerical values.