JP2767608B2 - Membership function analyzer - Google Patents

Description

Description: BACKGROUND OF THE INVENTION The present invention relates to an apparatus for analyzing a membership function for fuzzy inference.

Fuzzy controller, fuzzy computer,
Devices for performing fuzzy inference called various names, such as fuzzy inference operation devices, are being developed. Many such devices perform fuzzy inference according to the Modus-Ponens fuzzy inference rules, which are so-called If, then rules. Fuzzy inference rules include a membership function.

In the application of fuzzy inference devices, it is important to set appropriate inference rules and membership functions for the controlled objects. What role the rules and membership functions once set play in fuzzy inference?
Analyzing how the set rules and membership functions are related to each other is also essential for improving and improving fuzzy inference control.

However, research on the application of fuzzy inference has only just begun, and at present it has not been sufficiently studied.

SUMMARY OF THE INVENTION The present invention provides a novel device capable of analyzing the relationship between a plurality of set membership functions.

The membership function analysis device according to the present invention includes a plurality of membership functions set in advance.
Means for calculating the similarity between two different membership functions by comparing the shapes of the membership functions between the two different membership functions, and comparing the calculated similarity with a predetermined reference value to determine at least the similarity. Means for outputting the membership function determined to be.

If necessary, there is provided means for deleting one membership function while leaving one membership function among the membership functions determined to be similar to each other. The worker can view the output data on the similarity of the membership function, input a deletion command when the worker judges that the data should be deleted, or delete the data automatically. Good.

According to the present invention, since an objective similarity between two membership functions is calculated, unified handling is always possible, regardless of the subjectivity and feeling of the worker.

Also, by leaving one of the membership functions determined to be similar to each other and deleting the others, the number of stored membership functions is reduced, so that the storage area of the membership functions can be saved. . Since a minimum number of necessary membership functions can be set, fuzzy inference can be wasted.

The present invention, when setting rules for fuzzy inference and when modifying the rules set, automatically checks whether there are any similarity in the membership functions used in these rules. Useful in cases. In particular, when modifying rules, similar membership functions may be set, and in such a case, it will be effectively used as an automatic rule check function.

Furthermore, when considering a fuzzy inference device having a learning function and automatically setting or modifying rules by learning, the analysis device according to the present invention is indispensable to avoid duplication of membership functions created by this device. It will be.

FIG. 1 shows an embodiment of a membership function analyzing apparatus according to the present invention. This analyzer is realized by a computer device including a central processing unit (including a microprocessor), for example.

Two members as shown in FIG. 2 membership functions F _A and F _B
The analysis processing of the similarity will be described.

First these membership functions F _A in the similarity calculation unit 11, the similarity I of F _B (tau) is calculated by taking the correlation according to the following equation.

Then by similarity determination section 12 whether or not the these membership functions F _A and F _B can be said to be similar or not. One membership function F _A is a reference. The autocorrelation I _{0 of the} reference membership function F _A is calculated by the following equation.

The maximum value I _{max of the} similarity I (τ) given by the equation (1)
And τ _max that gives this maximum value are obtained (see FIG. 3). Further, γ is obtained from the following equation using I ₀ and I _max .

_{γ = I max / I 0 ...} (3) γ is and large | tau _max | it can be said that the more membership functions similarity F _A and F _B is large or small.

The criterion for similarity determination (similarity criterion) is γ _L , τ _L. This criterion is externally provided to the determination unit 12. Then, τ and τ _max given by equation (3) are γ> γ _L and If both are satisfied, the membership function F _A and
The F _B is judged to be similar.

Similarity determination result of such membership functions F _A and F _B is, CRT display device, a liquid crystal display device, is displayed on the display device 13 such as a plasma display device. Or printed by a printer if necessary.

If there are three or more membership functions F _A , F _B , F _j , whether the other membership functions F _B , F _j are similar to the membership function F _A centering on the membership function F _A Can be determined and displayed in the manner described above.

Finally, the deletion processing section 14, leaving the one membership function is judged to be similar to each other (for example, leaving the membership functions F _A, which became the center) are all deleted.

Assuming that the membership function is stored in the memory, the deletion process may clear that area.
The deletion signal may be automatically generated inside a digital computer or the like, or may be input by an operator while watching the display.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention. FIG. 2 is a graph showing an example of the membership function, and FIG. 3 is a graph showing the correlation of the membership function. 11 ... Similarity calculator, 12 ... Similarity determiner, 13 ... Display device.

Claims (2)

(57) [Claims] A means for calculating a similarity between two membership functions by comparing the shapes of the membership functions between two different membership functions among a plurality of membership functions set in advance;
And a means for comparing the calculated similarity with a predetermined reference value and outputting at least a membership function determined to be similar. 2. A means for calculating the similarity between two membership functions by comparing the shapes of the membership functions between two different membership functions among a plurality of membership functions set in advance. Means for comparing the determined similarity with a predetermined reference value and outputting at least a membership function determined to be similar, and leaving at least one membership function from the membership functions determined to be similar to each other. A membership function processor comprising: a means for deleting other membership functions.