JPH03225428A - Fuzzy computing element - Google Patents

Abstract

PURPOSE:To speed up membership function calculation by determining the shape of a function according to a value loaded in a memory which is provided previously and calculating the membership function by a dedicated function generator. CONSTITUTION:The fuzzy computing element is equipped newly with an external memory 8, a function parameter memory 9, a function generator 10, and a parameter pointer decoder 11. Namely, the dedicated function generator 10 and the memory 9 for holding a value determining the shape of the function are provided in the computing element and when the membership function is calculated, the shape of the function is determined according to the value loaded in the memory 8 which is provided previously to calculate the function by the dedicated function generator 10. Consequently, the membership function can be calculated with a single instruction and the membership function can be calculated fast.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a fuzzy arithmetic unit, and particularly to a fuzzy arithmetic unit that calculates membership functions in fuzzy arithmetic at high speed.

[Conventional technology]

To perform fuzzy inference, for example, "Introduction to Fuzzy Systems" by Hisabe Terano et al. (Ohmsha, April 30, 1986)
As described on page 150 of the Japanese publication, it is necessary to perform various calculations on the function f(2). In digital calculation, the number f(2) is a discrete value X (x=0.1, 2.-
, n-1, n) integer values f (0), f (1)
, f(2) , ・=, f (n-1), f(
n), but since there was no conventional dedicated hardware to directly calculate this type of function, in order to find this f (x) from the value It was necessary to calculate (x).

For example, Intel's microprocessor 8088 has a third
In order to calculate the membership functions as shown in the figure, it was necessary to run a program as shown in FIG.

Below, FIG. 6 will be briefly explained.

Figure 6 is a diagram showing a program that calculates f (X) in the AX register from the X in the BX register. In Figure 0, the first and seventh lines are instructions to prepare constants in the DX register. . The second line is an instruction to set f(2) to 0 in advance, and the tenth line is an instruction to set f(2) to 255 in advance. First, in the 3rd and 4th lines, it is checked whether or not X is within the range of p in Figure 3, and if so, f(x) = O, and then 0, then in the 5th and 6th lines, Check whether X is within the range of q in Figure 3 in the line, and if so, branch to line 13 to calculate the straight line A.0 Next, in the 8th and 9th lines, Check whether it is within the range of t in the figure, and if so, f(→
0th order ending as =O, line 11.

In line 12, it is checked whether is q in Figure 3 (DX register is 32 at this time)
or S (at this time, the DX register is 192). Here, straight line A or straight line B
is calculated and the program ends, but the 13th line is an instruction to subtract X from DX. If this result is negative, then X is within the range of q, so it is inverted to positive in the 14th and 15th lines. The 16th and 17th lines are for multiplying the subtraction result by four, and by adding the DX register itself twice, the contents are multiplied by four. The 18th line is an instruction to move the result to AX.

[Problem to be solved by the invention]

In conventional membership function generation, functions are calculated by software using an ordinary computer as described above, which poses a problem in that calculations take time.

The present invention was made to solve the above-mentioned problems, and it is an object of the present invention to provide a fuzzy arithmetic unit that can calculate membership functions at high speed.

[Means to solve the problem]

The fuzzy arithmetic unit according to the present invention includes a dedicated function generator and a memory that holds values that determine the shape of the function inside the arithmetic unit, and when calculating membership functions, loads the functions into the pre-provided memory. The shape of the function is determined based on the values given, and the function is calculated by the dedicated function generator described above.

[Effect]

In this invention, when calculating a membership function, the shape of the function is determined based on values loaded in advance into internal memory, and the value of the function is calculated using a dedicated function generator. Therefore, the membership function can be calculated with a single instruction, and the membership function can be calculated quickly.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a fuzzy arithmetic unit according to an embodiment of the present invention. In FIG. 0, 1 is a general-purpose register group, which is composed of eight general-purpose registers. 2 is an internal data bus, 3.4 is a latch, 5 is an ALU, 6 is an instruction decoder, 7 is an external control unit, and 8 is an external memory. Reference numeral 9 denotes a function parameter memory, which is composed of 12 parameter registers that hold parameters indicating the shape of the function.

Further, 10 is a function generator, 11 is a parameter pointer decoder, and 12, 13, and 14 are control signals.

Note that the control signal 12 is sent to the general-purpose register 1. Latch 3° Latch 4. ALU5. Although this is a control signal for controlling the function generator 10, the connection state thereof is omitted in FIG. 1 because it makes the drawing unsightly.

Next, the operation will be explained.

The fuzzy arithmetic unit shown in FIG. 1 has an external memory 8. Function parameter memory9. Function generator 10. The components except for the parameter pointer decoder 11 have the same configuration as a normal microprocessor. Then, it repeats the operations of fetching instructions from the external memory 8, decoding and executing them in the same way as a normal microprocessor. However, the first
The fuzzy arithmetic unit shown in the figure has an instruction set that includes instructions such as addition and subtraction operations, nine logical operations, and other instructions that a normal microprocessor has, as well as instructions that calculate the membership function described above and perform operations on it.

FIG. 2 shows an example of this command.

In the figure, the obcode part is a part indicating this instruction, and indicates that this instruction is an instruction to calculate a membership function and store the value in a general-purpose register. The parameter designation section designates one of the 12 parameter registers in the function parameter memory. General-purpose register specification section 1
．． The general-purpose register specifying unit 2 specifies one register containing the contents of each of the eight general-purpose registers and one register to store the membership function value.

Next, the instruction execution procedure will be explained. In the fuzzy arithmetic unit shown in FIG. 1, an instruction decoder controls the overall operation. When the instruction decoder finishes executing one instruction, it controls the control signal 13 and causes the external control unit 7 to fetch the next instruction from the external memory 8. The instruction decoder 6 decodes the fetched instruction and sends the control information necessary to execute the instruction to the control signal 12. control signal 13
Output to. At this time, if the decoded instruction does not use the function generator 10, general register 1. Latch 3. Latch 4. The AlO2 performs predetermined processing corresponding to commands such as addition/subtraction operations and logical operations according to the control information of the control signal 12. At the same time, the external control section 7 performs external control as necessary according to the control information of the control signal 13. When data is read from the memory 8 and outputted to the internal data bus 2, or conversely data from the internal data bus 2 is written to the external memory 8, the decoded instruction is the one shown in FIG. The instruction decoder 6 receives the control signal 12. In addition to the control signal 13,
Control information for selecting the parameter register designated by the parameter designation section shown in FIG. 2 is output as the control signal 14. Then, in the first step, the parameter pointer decoder 11 receives the control signal 14, decodes its contents, and transfers the contents of the designated parameter register to the function generator 1.
The function parameter memory 9 is controlled so that the output is zero. At the same time, the general-purpose register group 1 outputs the contents of the general-purpose register specified by the general-purpose register specifying section 1 shown in FIG. 2 to the internal data bus 2 in response to the control signal 12. Then, latch 3 takes it in and latches it. Furthermore, the function generator 10 calculates the value of the function shown in FIG. do. General-purpose register group 1 is controlled by control signal 12.
Accordingly, the contents are taken into the general-purpose register specified by the general-purpose register specifying section 2 shown in FIG.

Note that, as shown in FIG. 4, the parameter register contains the slope of straight line A (
2) in Figure 3, the X-intercept of straight line A (32 in Figure 3),
Values indicating the slope of straight line B (2 in FIG. 3) and the 5-line box intercept (192 in FIG. 3) are stored in advance.

FIG. 5 is a diagram showing a specific configuration example of the function generator 10. In FIG. 0, 21.22 is a subtracter, 23.24 is a multiplier, and the subtracter 21 is an , and the multiplier 23 multiplies the result by the slope to the straight line.

In other words, the value of straight line A is calculated. Similarly, the subtracter 22 subtracts X from the straight line BOX intercept, and the multiplier 24 multiplies the result by the slope of the straight line B. In other words, the value of straight line B is calculated.

25 is a range detector, which checks in which range of pt in FIG. 3 X lies based on the calculation results of multipliers 23 and 24. That is, the calculation result (Xl) of the multiplier 23 is 0≧X,
When the relationship is, it is determined that X is in the range of p, and O<
When the relationship x, <255 exists, it is determined that X is within the range of q, and xa.

When the calculation results (X, ) of the multiplier 23 are both 255 or more, it is determined that X is in the range of r, and Xb is O<x.
When there is a relationship of b<255, it is determined that X is within the range of S, and when there is a relationship of 0≧Xb, it is determined that X is within the range of t, and the detection result is output.

26 is a selector, which selects the output of the multiplier 23 (value of straight line A) and outputs that value if X is in the range of q according to the detection result of the range detector 25, and if X is S If X is in the range of P or t, select the output of the multiplier 24 (the value of straight line B) and output that value; if X is in the range of P or t, output the value O,
If is within the range of r, the value 255 is output.

As a result of the above, if the fetched instruction is the instruction shown in FIG. 2, the membership function will be calculated and stored in the general-purpose register. In other words, the membership function is calculated with one instruction.

Note that the above example shows the case where an instruction is executed to store the calculated value of the membership function in a general-purpose register, but in the fuzzy operation, the value of the membership function is compared in magnitude with another value, and the value of the membership function is compared with another value. In some cases, a process of selecting the larger or smaller one among them may be performed. In such a case, the first step should be the same as above, and the second step should be the same as above.
In this step, the contents of the function generator 10 are latched into the latch 3 via the internal data bus 2. Then, in the third step, the contents of the general-purpose register specified by the general-purpose register specification 2 are transferred to the latch 4 via the internal data bus 2, and the AL
Compare the two using U5.

Finally, only when the contents of latch 3 are larger or smaller than latch 4, the contents of function generator 10 are transferred to the general-purpose register specified by general-purpose register specification 2 via internal data bus 2 in the fourth step. do it.

Furthermore, although the above embodiment has shown an example in which the function parameter memory is composed of registers, it may be composed of a simple ROM instead of registers. In this case, the contents cannot be changed during operation, but the hardware is simpler than in the case of registers.

〔Effect of the invention〕

As described above, according to the present invention, in a fuzzy arithmetic unit, when calculating a membership function, a dedicated function generator and a memory for storing values that determine the shape of the function are provided inside the arithmetic unit. The shape of the function is determined based on the value loaded in the provided memory, and the function is calculated by the dedicated function generator mentioned above, so membership function calculations can be calculated with a single instruction. This has the effect of making it possible to obtain a fuzzy arithmetic unit that calculates membership functions at high speed.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a fuzzy arithmetic unit according to an embodiment of the present invention, FIG. 2 is a diagram showing membership function generation instructions of a fuzzy arithmetic unit according to an embodiment of the present invention, and FIG. 3 is a membership function FIG. 4 is a diagram showing the contents of the parameter register included in the fuzzy arithmetic unit shown in FIG. 1, and FIG. FIG. 6 is a diagram showing an example of a program when a membership function is generated by a conventional microprocessor. 1 is a general-purpose register, 2 is an internal data bus, 3゜4 is a latch, 5 is an ALU, 6 is an instruction decoder, 7 is an external control unit,
8 is an external memory, 9 is a function parameter memory, 10 is a function generator, 11 is a parameter pointer decoder, 12.
13.14 is a control signal, 21.22 is a subtracter, 23.2
4 is a multiplier, 25 is a range detector, and 26 is a selector.

Claims (1)

[Claims] (1) In a fuzzy arithmetic unit that calculates a membership function, a storage means for holding a plurality of pieces of information regarding parameters of the function, and one of the plurality of pieces of information held in the storage means is selected, and its storage contents are selected. and a function generation means that generates a function according to the stored content of the storage means outputted by the control means, and calculates the membership function with a single instruction. A fuzzy arithmetic unit characterized by: