JPH02165237A - Rule evaluating device for fuzzy processor - Google Patents

Abstract

PURPOSE:To evaluate and select a rule properly in long-period view by providing a means which sets a time zone wherein the rule is evaluated and a means which calculates an evaluated value by integrating condition adaption for this set time zone. CONSTITUTION:A front condition part processing part 27 outputs condition adaption qi indicating the adaption of input to an integrator 41. A time section wherein the integrator 41 is operated is set in a time scheduler 42 and the integrator 41 is controlled by the scheduler 42 to integrate a voltage input indicating the condition adaption qi and outputs a rule evaluated value (v) to a comparator 43. Then the ineffectiveness reference value of a rule is set in the comparator 43 in advance and the input evaluated value (v) is compared with the reference value; when the evaluated value (v) is smaller than the reference value, the ineffectiveness decision output of the rule is generated. Consequently, the rule is evaluated and selected properly from the point of long-period view.

Description

[Detailed Description of the Invention] Summary of the Invention In a fuzzy processing device that performs fuzzy inference based on a plurality of rules, the degree of importance, importance, etc. of each rule can be determined by evaluating the suitability of each rule to the human power value. Allows selection of rules, etc.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rule evaluation device in a fuzzy processing device such as a fuzzy controller or fuzzy computer.

BACKGROUND OF THE INVENTION In recent years, fuzzy theory has been in the spotlight, research and development have been conducted on the application of fuzzy processing devices, and in some cases, their practical use is progressing. Fuzzy processors typically perform an inference called modus ponens. This is the so-called I
It is expressed by a rule of the l', then format (II', then rule). These rules are often set based on human experience. Unless you have set only rules that have been thoroughly examined, or even in such cases, when you actually run the inference, unnecessary rules, redundant rules, etc. will be found among the important rules. are often mixed. These unnecessary or unimportant rules require extra hardware in devices that perform inference in hardware, creating penalties in terms of cost, power consumption, and system size; In devices that run software, problems arise such as the program becomes bloated, processing time is long, and real-time performance is lacking. Therefore, rule evaluation is an important issue for improving system efficiency.

Conventional devices merely detect the degree of conformance of a rule at a certain point in time and exclude it from the objects of inference calculation (for example, Japanese Patent Application Laid-Open No. 1983-226304). However, with this conventional method, it has been difficult to appropriately evaluate rules from a long-term, comprehensive perspective.

SUMMARY OF THE INVENTION OBJECT OF THE INVENTION An object of the present invention is to provide a device that can evaluate rules from a long-term perspective.

Structure of the Invention The present invention provides means for setting a time period in which a rule should be evaluated in a fuzzy processing device that performs inference based on a rule including a fuzzy proposition consisting of a preset antecedent part and a consequent part, and a fuzzy proposition. Means for calculating an evaluation value by integrating the degree of condition suitability expressed by the processing result regarding the antecedent part over the time period set in the setting means, and comparing the calculated evaluation value with a predetermined evaluation standard for rule evaluation. The invention is characterized in that it includes means for outputting information regarding.

Here, the fuzzy processing device means any device that performs fuzzy inference or fuzzy operation, and
It includes computers, fuzzy controllers, fuzzy inference devices, fuzzy inference calculation devices, etc. In addition, if only a dedicated device (regardless of analog φ type or digital type) for fuzzy inference (for example, [Nikkei Electronics J July 27, 19g7, pp. 148-152, 2 Nikkei McGraw-Hill, Inc.] It includes a binary type computer, processor, etc., which is programmed to perform fuzzy inference.

One or more evaluation criteria may be provided.

In a fuzzy processing device given multiple rules,
By setting at least one of the above evaluation criteria from evaluation values for a plurality of rules, it is possible to perform a relative evaluation of each rule. For example, the evaluation standard may be determined based on the average value of evaluation values for a plurality of rules.

Functions and Effects of the Invention According to the present invention, there is provided a means for setting a time period in which a rule should be evaluated, and a means for calculating an evaluation value by integrating the condition conformity over the set time period. . Since the evaluation value is an integral value over a certain period of time, it is possible to evaluate rules over a long period of time rather than at a certain moment. This makes it possible to select a more appropriate evaluation 2 rule.

The time period for evaluating rules can be arbitrarily set, or may be divided into a plurality of discontinuous time periods.

DESCRIPTION OF EMBODIMENTS Generalized Modus Bonens refers to inferring a conclusion from given premises when one or more implications exist. The most common form of inference that has multiple implications and includes multiple fuzzy propositions in the antecedent part of the implication is expressed as follows.

Implications 1 x-A, y=B, z-C-Q-Q.

1 1, 1 Implication 2 x-A, y-B, z-C-* q-Q2 Implication n: x=A, y=B, z-C-* q-
Qlln n
n premises: x-A'', y-B', z=c' Conclusion:
q-Q' where A
, B1. C1, Q, (i-1~n).

A'B'C'Q' is a fuzzy set.

Represented by a membership function.

Implication i has the following meaning.

It' x is A, andlor y
is B, andlor zlsC, thcnq
is Q.

This is called the If, then rule.

x-A, y-B, z-C, are antecedent parts,
q-I Q is called the consequent part.

An example of a fuzzy controller that performs fuzzy inference based on the above rules will be briefly described with reference to FIG. This fuzzy controller is of a type that expresses membership functions by voltage distributions appearing on a plurality of signal lines, and performs fuzzy inference using the old N/MAX calculation.

Inference units 11 to 1n are provided in a number corresponding to the number of set rules (implications). Each inference part 1
1 (i=1 to n) is provided with a membership function circuit (hereinafter referred to as MPC) 21.22°23 for the number of fuzzy propositions in the antecedent part (three here).

These NPCs 21 to 23 output voltage signals representing membership functions representing the fuzzy set (A, B, , C,) described in the antecedent part of the corresponding rule. The above-mentioned premise becomes an input, but since it is a controller, the input is represented by fixed values x, yl, zl, and the NFCs 21 to 23 use these input values x, y, .

Output the membership function value corresponding to Zt. NP
The outputs of C21 to C23 are input to the old N circuit 24, and the old N
An operation is performed. The output of the old N circuit 24 is q, (i-1
~ n).

On the other hand, a circuit (hereinafter referred to as MFG) 25 that generates a membership function representing a fuzzy set described in the consequent part of a rule (implication) is provided, and a plurality of (m) output lines are provided from this MFG 25. The membership function represented by the voltages distributed above is output and applied to the old N circuit (truncation circuit) 26. The MIN circuit 26 calculates the MIN of each of the voltage values representing the membership function given from the MFG 25 and the calculation result q output from the MIN circuit 24.
Perform calculations.

The membership function representing the inference result is output in the form of voltage signals distributed on m lines (output Q ”:
i Kiri~n).

In each reasoning unit 11, NPCs 21 to 23. M! 2
It goes without saying that the membership function in G25 is preset in a predetermined form according to the rules. The control forces for setting the membership functions in the desired form are represented in FIG. 1 as labels.

The inference results Qi output from the inference units 11 to 1n are then M
After being applied to the AX circuit 3I and subjected to MAX calculation, the final inference result Q is obtained as a voltage signal similarly distributed over m lines. A centroid circuit 32 is provided to obtain a definitive output q from the inference result Q.

FIG. 2 shows an embodiment of the invention. An integrator 41 is provided in the inference section II described above. A time scheduler 42 and a comparator 43 are newly provided.

The antecedent processing section 27 is composed of MPCs 21 to 23 and the old N circuit 24 shown in FIG. Its output q
, represents the degree of conformity of the input to the antecedent part (condition), so this will be referred to as the degree of condition conformity. The consequent part processing unit 28 to which this output ql is given is Ml''G shown in FIG.
25 and the old N circuit 2B.

Condition conformity q1 is given to an integrator 41. Integrator 41
is controlled by time scheduler 42. In the time scheduler 42, the time interval (which may be given as a time or the elapsed time from a predetermined reference time) in which the integrator 41 should be operated is preliminarily set using an appropriate input device or computer. has been done. When the integrator 41 is given a start command from the time scheduler 42, it starts integrating the voltage input representing the condition compliance degree q1, and continues the integration operation until a stop command is given from the time scheduler 42. The integral value of the integrator 41 when a stop command is given is the rule evaluation value V
is output as

It goes without saying that the integrator 41 performs an integration operation while the fuzzy controller is performing a predetermined fuzzy inference. The integral output, that is, the rule evaluation value V, indicates a larger value as the degree of conformity to the input of the rule set in the inference section is higher. In other words, the lower the suitability of the rules set in the inference section for human power, the smaller the value, and the lower the suitability means that the rule is less important.

The rule evaluation value V is also input to the comparator 43.

The invalidity reference value TH of the rule is set in the comparator 43, and the rule evaluation value V to be input is set to this reference value T.
If it is less than or equal to H, a rule invalidity determination output f is generated. The comparator 43 performs the above comparison process when the integration operation by the integrator 41 is completed. In this sense, time
The stop command from the scheduler 42 may be given to the comparator 43 as a comparison start command, as shown by the broken line.

The invalid reference value TH is shown as the most extreme example, but
A plurality of different reference values may be provided to the comparator 43, and the rule evaluation value ■ may be divided into degree levels to provide information regarding the rule evaluation.

FIG. 3 shows the above-mentioned integrator 41. It shows a plurality of inference units each provided with a time scheduler 42 and a comparator 43. An average value of the rule evaluation values V, -Vo outputted from each inference section 11 to 111 is calculated by an average value calculator 44, and this average value is given to a comparator 43 of each inference section as a reference value TH. In this way, the comparator 43 obtains a determination output f1 (i-1 to n) indicating whether the rule evaluation value is above or below the average value. This judgment output f represents a so-called relative evaluation of each rule. It goes without saying that a plurality of different base $ values may be created based on the average value of the rule evaluation values.

The functions of the integrator, time scheduler and comparator are
It goes without saying that the present invention can be implemented with other hardware or software depending on the configuration of the fuzzy processing device to which it is applied.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the overall configuration of an example of a fuzzy controller. FIG. 2 shows an embodiment of the present invention, and is a block diagram showing the configuration of one inference section. FIG. 3 is a block diagram showing another embodiment including a circuit for calculating the average value of rule evaluation values of a plurality of inference units. 43... Comparator. 44...Average value calculator. Patent J1 applicant Power Reactor and Nuclear Fuel Development Corporation (one other person) Agent Patent attorney Kenji Ushiku (
1 person outside) 11, 11, in - Reasoning Department. 27...Antecedent part processing part. 28...Consequent part processing part. 41... Integrator. 42...Time scheduler.

Claims (2)

[Claims] (1) In a fuzzy processing device that performs inference based on a rule including a fuzzy proposition consisting of a preset antecedent part and a consequent part, means for setting a time period in which a rule should be evaluated, and an antecedent part of a fuzzy proposition. means for calculating an evaluation value by integrating the degree of condition suitability represented by the processing result in the setting means, and comparing the calculated evaluation value with a predetermined evaluation standard to obtain information regarding the rule evaluation. A rule evaluation device in a fuzzy processing device, comprising a means for outputting. (2) A fuzzy processing device given a plurality of rules, comprising a criterion setting means for setting at least one of the evaluation criteria from evaluation values for the plurality of rules;
A rule evaluation device in a fuzzy processing device according to claim (1).