JPS61264411A - Reasoning abnormality detecting system for fuzzy reasoning arithmetic unit - Google Patents

Abstract

PURPOSE:To display the generation of the reasoning abnormality by detecting and treating the reasoning abnormality due to an error of a control rule for fuzzy reasoning operation or the incomplete designing during an arithmetic operation. CONSTITUTION:A fuzzy reasoning arithmetic unit 3 performs a fuzzy reasoning operation by means of the input values given from a measured value input device 1 and a process state input device 2 and fetching plural control rules of a control rule storage device 5 via a control rule input device 4. Thus the output value obtained from a membership function for output is delivered from a control output device 6 in the form of the control output and also displayed on a control state display command device 7 and a control state display device 8. Then a reasoning abnormality detection parameter input device 9, a reasoning abnormality detector 10, a reasoning abnormality state display device 13 and a reasoning abnormality state display command device 12 are added to a fuzzy control device. Thus an output membership function given from the unit 3 is collated with the parameter given from the device 9 for study. Then the abnormality of the reasoning operation is detected and displayed.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention provides a method for inputting either or both of process measurement values and process status values that are the results of intuitive judgments of process status by operators. An inference abnormality in a fuzzy inference calculation device that performs fuzzy inference calculations according to predetermined control rules, creates an output membership function as a result, determines an output value from the function, and outputs it as a control output. This relates to the detection method.

The concept of fuzzy, which can be seen in terms such as fuzzy inference operation and fuzzy adjustment, is a relatively new concept that appeared around 1960, so it will be briefly explained below to the extent necessary for understanding the present invention.

Fuzzy is an English word that means "ambiguous." So-called "fuzzy mathematics" such as fuzzy set theory, fuzzy logic, and fuzzy algebra have appeared in recent years, but the control performed by applying this fuzzy logic is called fuzzy control, and it is inferior to conventional process control. It is said that there is no

The aim of this ambiguous control is to use a computer to reproduce, even in complex processes for which a mathematical model cannot be created, a control that a human (operator) would be able to successfully achieve based on their in-depth knowledge of process behavior and control experience. It's there where you try.

Therefore, as a method for this ambiguous control, in addition to the process measurement values, the results of the operator's subjective and intuitive judgment of the process status (in other words, whether the process is large, small, medium, or high) are used. Or low.
(these are called ambiguous variables), inference (estimation calculation) is performed according to a predetermined control rule, and the result is used as a control output. Here, for ambiguous variables, a membership function is predetermined that represents the degree to which the same process situation is determined to be "large" if it is "large" (for example, the proportion of the number of people who make such a determination). The ambiguous variable is digitized using the membership function (herein, this digitized value will be referred to as a process status value).

Since the present invention is not aimed at ambiguous control itself, further explanation will be omitted.
2. Papers P150-160 "Fuzzy sets and their application to logic", also "Measurement and Control J VOL, 22゜N11
．． Please refer to the paper "Ambiguous Control" on pages 84-86).

Hereinafter, in this specification, we have decided to use the term "fuzzy" for all explanations, as we believe that the term "vague" may still cause misunderstandings and is inappropriate.

[Prior art]

In general, inference operations in a fuzzy inference operation device are performed based on control rules (also referred to as estimation rules). The antecedent part (IF part) and the consequent part (THEN
When the quantitative expression in (section) is expressed by fuzzy variables (ambiguous variables such as large or small), the fuzzy inference calculation device uses the given input value to find the output membership function and calculates its members. The output value (operation value) is determined by the ship function.

FIG. 4 is a graph showing an example of an output membership function obtained by the fuzzy calculation device as a result of inference calculation.

In the same figure, the horizontal axis is the normalized value of the output value (for example, if the valve opening is the output value, the normalized value is 1 for the valve opening of 100%, and the normalized value for the valve opening θ%) The vertical axis shows the grade for the output value, that is, the degree of fit when the output value is y, and the grade 1
This means that the degree of applicability is 100%, and grade 0 means that the degree of applicability is 0.
%.

For example, if the consequent proposition of a control rule is "Make the valve opening to a medium level," the grade that best applies to that proposition is 1, so the output corresponding to the peak value of 1 is The value y will be between 1 and 0 on the horizontal axis (in other words, the middle), and other output values generally have a lower grade as they move away from the middle. good.

In reality, grade often refers to the degree to which it applies to multiple antecedent propositions at the same time, rather than the degree to which it applies to a single antecedent proposition. The grade is l only if it is completely applicable to every proposition. Since there are few cases in which an input value completely applies to any of the plurality of control rules, there is rarely an output value with a grade of 1 in the function waveform of the output membership function.

Now, assuming that the output membership function has been determined as shown in Fig. 4, the method of determining the output value that is actually output as the control output from this membership function is as follows.
A method is often used in which the output value at a point that bisects the area formed by the function waveform of the membership function and the horizontal axis is used as the output value (yo in Figure 4), but this method works. An implicit precondition for this is that the waveform of the membership function for the output value does not exhibit multimodality.

For example, if the waveform of the membership function is as shown in Figure 4, it can be said that it is the result of good inference, but a multimodal function waveform as shown in Figure 5 was obtained. This cannot be said to be the result of good reasoning. The reason why the membership function waveform shown in Fig. 5 appears is because there is an error in the control rule used in the fuzzy arithmetic device, or the control rule is insufficiently designed. In such cases, it is necessary to modify the control rules.

However, in conventional fuzzy arithmetic control that uses fuzzy variables as quantitative expressions for both the antecedent and consequent parts that make up the control rule, the membership is close to unimodal, as shown in Figure 4. Because it was assumed that the control rules were designed so that a functional waveform could always be obtained, there was no method for detecting errors in control rules or insufficient design.

[Problem that the invention seeks to solve]

Therefore, the problem that the present invention aims to solve is that when an abnormality in the inference operation occurs due to an error in the control rule or insufficient design during the operation of the fuzzy inference operation device, it is possible to detect and deal with the abnormality. It can be said that it is to do something. Therefore, an object of the present invention is to provide an inference anomaly detection method in a fuzzy inference arithmetic device that makes the above possible.

[Summary of the invention]

This invention provides that when an inference operation in a fuzzy inference operation device is performed on an input value based on a control rule, if there is an error or insufficient design in the control rule, the output value obtained as the result of the inference operation Focusing on the fact that the waveform of the membership function has multimodality, we judge whether the multimodality has appeared or not using the maximum peak value, the second highest peak value, and the two peaks of the membership function waveform. This is performed during control operation using the three minimum values of the valley between the two, and if it is determined that multimodality has appeared, the occurrence of the inference abnormality is displayed on the display device, and the date and time of occurrence are displayed. , display and memorize the input values at that time, the control rules mainly used, and the output membership functions, and display the contents memorized in the past to detect inference anomalies and deal with them. This is what made it possible.

〔Example〕

Next, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. The embodiment shown in the figure is different from the conventional fuzzy adjustment device.
- It can be said that the circuit part S surrounded by the broken line is added.

That is, conventionally, input values from the measurement value input device 1 and the process status input device 2, which express and input the process status as a rough quantity such as "large" or "small 1", are used.
The control rule input device 4 inputs the control rule storage device 5.
A fuzzy inference calculation device 3 that takes in a plurality of control rules stored in the computer and performs fuzzy inference calculations based on these rules.
and a control output device 6 that outputs an output value obtained from the output membership function obtained as a control output.
and a control status display device that displays input values, control rules mainly used for fuzzy inference, output values, and output membership functions that are the basis for obtaining them, according to commands from the control status display command device 7. A fuzzy adjustment device consisting of 8 has been realized. Parameters input from an inference anomaly detection parameter input device 9 are given to this device, and output membership functions obtained by fuzzy inference are sent to an arithmetic device 3. an inference anomaly detection device 10 that detects an inference anomaly in the arithmetic device 3 by examining the membership function in light of the parameters, and when an inference anomaly is detected;
When an inference abnormality is detected by the inference abnormality detection device 10 based on commands from the inference abnormal situation storage device 11 that stores related data at that time as inference abnormal situation data and the inference abnormal situation display command device 12, , that an inference error occurred immediately, the date and time of occurrence, the input value at that time,
Mainly used control rules, output values, and output membership functions are displayed as inference abnormal situation data, and
An inference abnormality situation display device 13 is added which can also display the above-mentioned data regarding inference abnormalities that occurred in the past, which are stored in the inference abnormality situation storage device 11.

Now, the important point in the present invention is that in the inference anomaly detection device 10, by examining the output membership function given from the arithmetic device 3 in light of the parameters input from the inference anomaly detection parameter input device 9, The objective is to detect an abnormality in the inference calculation performed in the calculation device 3. The examination method for this will be described later, but
In order to contribute to understanding the validity of the examination method described later, we provide the following explanation (.

First, it will be described that each peak value in the function waveform of the output membership function corresponds to the effectiveness of the control rule used (the degree to which the input value satisfies the proposition of the antecedent part). Control rules are generally expressed in the following format.

IF P THEN Q (If P, then Q)...(1) P is called an antecedent proposition, and it consists of multiple propositions for conditions, and one proposition takes the following form, yx, xf,・・・・・・
(2) Here, x is the variable name of the input value, and the input value and l
Corresponds to pair l. f, is a fuzzy variable (large.

(small, medium, etc.) and is expressed by a membership function defined as a range in which the value of xl changes.

~ indicates the relationship between xl and f (equal, greater than, smaller than, etc.). Therefore, the antecedent proposition given by equation (2) can be expressed, for example, in words as ``if the input value X is equal to the fuzzy variable f.''

In the case of multiple inputs, the above equation (2) is expressed as a plurality of equations (3).

Xl””+ + Xt ~ft * Xl ~f3
``'(3) Therefore, in this case, similarly expressed in words, for example, ``If the input value X is equal to the fuzzy variable f, and x2 is greater than f2, and X is less than r3.'' It is expressed as follows. Q is called a consequent proposition and has the following form.

yI''+C,...(4) This indicates that the manipulated variable y should be set to C, and C
3 is a fuzzy variable indicating a control output value, and is expressed by a membership function. In some cases, the above equation (4) is made into a plurality of equations as shown in equation (3) to describe a plurality of operations at the same time. Therefore, when the number of input values is n and the number of control rules is m, the entire control rule is generally expressed as the following equation (5) (when there is one manipulated variable).

Generally, the fuzzy inference performed by the fuzzy inference calculation device 3 is performed on input values based on the control rule expressed by the above equation (5). As a method for this, an inference synthesis agent is often used, but there is a method of partially modifying it and performing the calculation shown in the following equation (6).

Qo = MAX ((M I N p ij) *
Qi) ... (6) 1≦i≦m, 1-5j≦
n However, Qo: Grade of output membership function pij + antecedent part imperative B P 1 = (expressed as 0 to 1 as the degree to which the proposition is satisfied) *: Control when the antecedent part of control rule i is completely satisfied Calculation to uniformly multiply the membership function Q, which indicates the output value, by a constant MIN: Calculation to take the minimum value MAX: Calculation of the union of multiple membership functions (take the maximum) Calculation to calculate the output value from the output membership function Q0 To determine this, a point on the horizontal axis that bisects the area of a closed figure consisting of the membership function waveform represented by Qo and the horizontal axis is found.

From the above equation (6), each peak in the waveform of the output membership function Q0 corresponds to the M control rule used in the calculation in a one-to-l ratio, and its height is the height before the corresponding control rule. You can see that there is a window.

In this way, when performing fuzzy inference operations using control rules in which the quantitative expressions in both the antecedent part and the consequent part are expressed as fuzzy variables, no matter which inference method is used,
The process of finding the output membership function is involved. Since the peak in the waveform of the membership function is determined by the degree to which the antecedent proposition of the corresponding control rule is satisfied, that is, the effectiveness of that control rule, it exhibits remarkable multimodality as shown in Figure 5. What the output membership function waveform indicates means that control rules that command different operations on the same input value are simultaneously valid, and multiple judgments that are given equal importance are made.

This means that the general idea that one decision should be given importance when a certain situation is given does not hold, and if there is an error in the control rule (for a certain situation This indicates a case where there are multiple control rules that make different judgments or a simple input error) or a case where the design is insufficient (a control rule that makes a judgment between two peaks is not prepared).

Therefore, by determining the presence or absence of multimodality in the waveform of the output membership function, it is possible to detect an abnormality in the inference calculation that is the source of the multimodality.

As mentioned earlier, the inference anomaly detection device 10 in FIG.
is given an output membership function which is the inference result of the fuzzy inference arithmetic device 3, and detects an inference abnormality in the arithmetic device 3 by determining whether or not there is multimodality in the waveform.

Next, a method for determining the presence or absence of multimodality in the waveform of a given output membership function will be described.

Assume now that the waveform of the output membership function as shown in FIG. 5 is given. In the same waveform, find the maximum peak value and the second highest peak value, and convert those values into gs.
Let ++g□. Next, find the minimum value of the membership function waveform that exists between these two peak values! , and so on.

To obtain this gsr*gszyItI, even if the output membership function is expressed as a function of the output value y (the following equation (7)), or by discretizing the output value y and calculating the grade value for that discrete value. Even if the output values are expressed as a table (see FIG. 2) with a one-to-one correspondence, it is possible to search for the change tendency of the grade value by sequentially searching the change range of the output value.

g=f (y)...
(7) In this method, it is determined that there is multimodality when the following equation (8) or (9) holds using g□*gsz*Jl.

g14I=gMz (8) 8M vote-gHz Here, rg is a parameter input by the inference abnormality detection parameter input device 9 and is normally set to 1.0.

According to the method described above, when the inference anomaly detection device 10 determines that the output membership function given from the arithmetic device 3 has multimodality, it is determined that there is an inference anomaly in the arithmetic device 3, and the inference anomaly is detected. Notification 9
The inference abnormality status display device displays the date and time of occurrence, input values at that time, control rules and output values mainly used in fuzzy inference calculations, and output membership functions based on instructions from the inference abnormality display command device 2. 13. It is also provided with an inference abnormal situation storage device 11 that stores these display contents, and it is also possible to display the contents (that is, past data) by command.

FIG. 3 is a flowchart showing the processing flow of the above-described multimodal presence/absence detection operation performed by the inference anomaly detection device 10. Although the flow is shown as consisting of steps S1 to Sll, there is no need to explain this further.

Modifications of this invention include the following.

Instead of setting the parameter rg in equation (9) as a constant value, we set it as a function of the distance between the maximum peak value and the re-output value that gives the second peak value, that is, )'+-f (g
s+),)'z -f-'(gHt)...(10
), and let rg be a function as shown in the following equation (11). f″′ is the inverse function of the function f.

rg” f (l 7t −Yz l)
...” (11) There may be cases where such a method is reasonable.

Further, the calculation and storage functions in the inference anomaly detection apparatus 10 shown in FIG. 1 may be realized by a digital computer.

〔Effect of the invention〕

According to the present invention, in an arithmetic device that performs fuzzy inference calculations based on control rules that use fuzzy variables in quantitative expressions in antecedent and consequent propositions, errors in the control rules or insufficient design (The advantage is that it is possible to detect and deal with inference abnormalities during calculation operations.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of the present invention, Fig. 2 is an explanatory diagram of a correspondence table between output values and grade values obtained for a given output membership function waveform, and Fig. 3 is a multimodal Flowchart showing the processing flow of the presence/absence detection operation, Part 4
The figure is a graph showing an example of the output membership function obtained by the fuzzy arithmetic unit as a result of the inference calculation, and FIG. 5 is a graph showing another example. Explanation of symbols 1... Measured value input device, 2... Process situation input device, 3... Fuzzy inference calculation device, 4... Control rule input device, 5... Control rule storage device, 6... ... Control output device, 7... Control situation display command device, 8... Control situation display device, 9... Inference abnormality detection parameter input device, 10... Inference abnormality detection device, 11... Inference Abnormal situation storage device, 12... Inference abnormal situation display command device,
13... Inference abnormal situation display device agent Patent attorney Kiyoshi Matsuzaki Figure 1 Figure 2 IEA Figure 3

Claims (1)

[Claims] 1) Fuzzy inference according to predetermined control rules using either or both of the process measurement value and the process situation value, which is the result of an operator's intuitive judgment of the process situation, as an input value. In an inference anomaly detection method in a fuzzy inference arithmetic device that performs an operation, creates an output membership function as a result, determines an output value from the function, and outputs it as a control output, the output membership function a determining means that examines whether the function waveform given is normal or abnormal, and determines that there is an inference abnormality when it is abnormal; a storage device that captures related data that is the basis of the fuzzy inference calculation performed in the arithmetic unit and stores it as inference abnormality situation data; An inference anomaly detection method in a fuzzy inference arithmetic device, comprising display means for extracting inference anomaly situation data from the storage device and visually displaying the inference anomaly situation data. 2) In the inference anomaly detection method in the fuzzy inference calculation device according to claim 1, the determining means:
An inference anomaly detection method in a fuzzy inference arithmetic device, comprising means for determining whether a function waveform of an output membership function is multimodal in light of a given input parameter.