JP2549441B2 - Membership function processor - Google Patents

Description

DETAILED DESCRIPTION [Overview] Regarding the processing of membership functions used in so-called fuzzy operations, an object is a membership function processing device capable of collectively managing membership functions corresponding to a plurality of events. Then, when outputting a predetermined event and a ratio representing the degree of the event to the input value, an identification value is assigned from the predetermined event value for each required event, and two different events are assigned. As specified, a closed curve consisting of line segments connecting each point of each of the two sets of coordinate groups, each of which is a required point having the input value, the ratio value, and the identification value as three-dimensional coordinate values. And a discriminant value closest to the event value that maximizes the ratio value obtained from the function with respect to the input value by providing a three-dimensional function defined by a curved surface formed by a straight line trajectory connecting corresponding points between The input value and the identification The configured to output the said ratio value determined from the function number as input.

[Industrial applications]

The present invention relates to a membership function processing device used for so-called fuzzy operations.

As is well known, the membership function is a function for obtaining a value for fuzzy calculation, which indicates the degree of an event in a certain event classification indicating a process state or the like, from an input value such as a measured value by a sensor.

[Conventional technology]

Fuzzy (Fu), which is known as an operation that can handle values that include so-called ambiguity, such as values that correspond to human senses
The zzy) operation is also adopted in expert systems and the like for diagnosing process states in the process processing field.

In that case, for example, the value of the temperature measured by a suitable sensor is classified into several stages of events from "very low""slightlylow" to "very high", for example, "80% Membership function is used to convert the value to a fuzzy operation such as "a little lower than."

That is, as shown in FIG.
When an input value such as temperature is input and a function name is input, a ratio represented by a value of 0 to 1.0, for example, is output as the function value of the function name, and the degree in the event classification corresponding to the function is expressed. .

For this purpose, the function calculation unit 1 has a function definition table 2, and according to the definition information separately input in advance, the coordinate value (x _i ,
y _i ) is held, and the input value is processed as described above by using a curve approximated by a line segment connecting these points as a designated function.

As shown in the above temperature example, generally, there are many event classifications for one category of input value, but one function is required for each classification. For example, regarding temperature, it is necessary to have about 5 classifications, such as very low, slightly low, normal, high, and very high.
As shown in Fig. (A), 2 for each of the 5 functions
~ Specify 3 points to define.

When obtaining a value for a certain input value, for example, the third
In the case of the input value indicated by a in FIG. 7A, the function having the largest value among the function values is determined to be “normal at about 70%”.

[Problems to be Solved by the Invention]

As described above, although it is necessary to provide a plurality of functions for one sensor or the like, the system is generally arranged to process sensor inputs of a plurality of categories to perform comprehensive diagnosis and so on. There is a problem that the number of membership functions becomes extremely large, and the work of modifying the functions for system optimization and adjustment becomes complicated.

An object of the present invention is to provide a membership function processing device capable of collectively managing membership functions corresponding to a plurality of events.

[Means for solving the problem]

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

The figure shows the configuration of the membership function processing device. In the membership function processing device that outputs a predetermined event and a ratio representing the degree of the event with respect to an input value, the function operation unit 12 and each required A first table storing an identification value table 10 in which identification values are assigned from predetermined event values for each event, and the input value, the ratio value, and the identification value so as to specify two different events. Defined by a curved surface that is a locus of straight lines connecting the corresponding points between the closed curves formed by the line segments that connect each point of each coordinate group of the two sets of coordinate groups that have the respective required points And a second table storing a three-dimensional function definition table 11 for maximizing the ratio value obtained from the function definition table 11 of the second table for the input value by the function operation unit 12. The event value is defined to define the event value in the first table. It is configured to output an identification value closest to the event value from the different value table 10 and the ratio value determined from the function definition table 11 of the second table by inputting the input value and the identification value. There is.

[Work]

With the above processing method, it is not necessary to provide and manage a function for each individual event of the same category, and one function that is grouped for each category is defined by a relatively small number of coordinate points, and thus multiple conventional Can be replaced with a function definition.

〔Example〕

The function calculation unit 12 in FIG. 1 uses a group of coordinates (x _1i , y _1i , z ₁ ) and coordinates (x _2i , y _2i ,
Two sets of three-dimensional coordinate groups of the group z ₂ ) are held in the function definition table 11. Here, the x and y coordinates are made to correspond to the input value and the ratio value as in the conventional case. The z coordinate is associated with an event value, and a specific event value is assigned to each required event and held in the identification value table 10 as an event identification value.

The two sets of coordinate groups are respectively composed of coordinate values defining a membership function for _two different events z ₁ and z ₂ selected from a plurality of events.

The function calculation unit 12 defines a three-dimensional solid defined by the two coordinate groups, which is defined by a curved surface formed by a locus of straight lines connecting the corresponding points between the closed curves formed by the line segments connecting the points of each coordinate group. As a matter of course, the input value is processed as follows using the point on the curved surface as a designated function.

That is, as shown in FIG. 2, in a processing step 20, for a certain input value x, for example, a function value z determined from the input value x and the proportional value y = 1.0 is obtained. This processing step is a pre-processing for selecting one membership function corresponding to a certain phenomenon from among the plurality of membership functions expressed by the above-mentioned three-dimensional curved surface. That is, the y coordinate is, for example, y =
Fixed to 1.0 (proportional value = 100%), the function value z of the z coordinate with respect to the input value x of the x coordinate is obtained.

Next, in processing step 21, the discriminant value z _0k closest to the calculated function value z is obtained from the discriminant value table 10. This processing step 2
1 is a processing step from among the identification values z _{01 to} z ₀₅ corresponding to each event stored in the identification value table 10 of the first table.
The identification value z _0k closest to the function value z obtained in 20 is selected. As a result, the z-coordinate is used as the identification value z for the subsequent processing.
_It is possible to perform a process of obtaining the ratio value of a specific membership function of a normal fuzzy operation, such as fixing the value to _0k and _obtaining the ratio value y of the y coordinate with respect to the input value x of the x coordinate.

Next, in a processing step 22, a ratio value y determined from the input value x and the event value z _0k is obtained, and in a processing step 23, an identification value z _0k representing a specific event with respect to the input value x and its ratio value y are output.

FIG. 3 (b) is a diagram for explaining a three-dimensional membership function that replaces the five membership functions illustrated in FIG. 3 (a). For example, a function with events at both ends in (a) is defined. Use each xy to define those functions
A three-dimensional function is defined by the coordinates and the coordinate group in which the identification values z ₀₁ and z ₀₅ assigned to the “very low” event and the “very high” event are z ₁ and z ₂ described above.

Also, z ₀₂ , z _03, and z ₀₄ are calculated so as to divide the section from z ₀₁ to z ₀₅ on the event value axis into four equal parts, and the values of “slightly low”, “normal”, and “slightly high” events are calculated. The identification value.

As a result, the function calculation unit 12 assumes that the triangle specified by the conventional function and the function is a cross section parallel to the xy plane, and the sideways triangular prism as shown in the figure is defined as the function. Process the input value to z ₀₁
~ Z ₀₅ and the ratio value y are output.

 This FIG. 3 will be described in correspondence with the processing of FIG.

The processing step 20 in FIG. 2 is to project the line having the proportional value y = 1.0 in FIG. 3 (b), that is, the line of the vertices of the triangular prism on the xz plane to obtain the function value z of the z coordinate with respect to the input value x. Calculated by a simple proportional calculation. In FIG. 3 (b), the calculated function value z
_Lies between z ₀₃ (normal) and z ₀₂ (slightly lower), z
It is close to ₀₃ (normal). In processing step 21, the discrimination value closest to the obtained function value z, that is, z ₀₃ (normal), is discriminated in the discrimination value table 1
Calculate with 0. The function name is now specified.

In the processing step 22, the laterally-displaced triangular prism shown in FIG. 3 (b) is cut in parallel with the xy plane at the position z ₀₃ (normal), and the proportional value y to the input value x (about 3 in the example of FIG. 0.7) is obtained by a simple proportional calculation.

In processing step 23, the obtained function value z = z ₀₃ (normal)
And a proportional value y = 0.7 is output.

In FIGS. 3A and 3B, when the input value x is extremely low and extremely high, for example, when the present invention is applied to the process processing field, it is an abnormal state and is a target of processing. Not.

By doing this, as is apparent from FIG. 3, conventionally, two or three defining points were required for each of the five functions, and a total of 13 points had to be managed. By using the three-dimensional function of, the function can be defined with only 6 points.

In the above description, the function is defined in two sections, but if necessary, the function corresponding to one category may be defined by dividing into two or more sections each defined in two sections. Of course, even in that case, the processing in each section may be performed in the same manner as in the above example.

〔The invention's effect〕

As is clear from the above description, according to the present invention, in processing membership functions used for so-called fuzzy operations in a computer, a relatively small number of membership functions corresponding to a plurality of events in the same category are collectively processed. Since it can be managed by the defined value, there is a remarkable industrial effect of improving the efficiency of management and processing of the membership function.

[Brief description of drawings]

 1 is a block diagram showing a configuration of the present invention, FIG. 2 is a flow chart of processing of the present invention, FIG. 3 is an explanatory diagram of a membership function, and FIG. 4 is a block diagram showing a conventional configuration example. In the figure, reference numerals 1 and 12 denote a function calculation unit, 2 and 11 denote a function definition table, 10 denotes an identification value table, and 20 to 23 denote processing steps.

Claims (1)

(57) [Claims] 1. A membership function processing device for outputting a predetermined event and a ratio representing the degree of the event with respect to an input value, a function operation unit, and a predetermined event value for each required event. A first table storing an identification value table to which identification values are assigned respectively, and the input value, the ratio value, and the identification value are three-dimensional coordinate values so as to specify two different events. A three-dimensional function definition table defined by a curved surface formed by a locus of straight lines connecting corresponding points between closed curves formed of line segments connecting each point of each coordinate group of two sets of coordinate groups including required points And a stored second table, and the function calculation unit determines the event value that maximizes the ratio value obtained from the function definition table of the second table for the input value, and determines the event value of the first table. From the identification value table, the identification value closest to the event value The membership function processing apparatus is configured to output the ratio value determined from the function definition table of the second table by inputting the input value and the identification value.