JPH02260041A - Fuzzy multistage inference device - Google Patents

Abstract

PURPOSE:To attain a fuzzy multistage inference and to perform an inference which is more approximate to a human inference than a single stage inference by transferring the intermediate result of a fuzzy inference to the next inference in a real number or a fuzzy label. CONSTITUTION:A fuzzy number storage part 105 stores the fuzzy variables showing the value of the controlled variables such as 'negative and large', 'zero', 'positive and small', etc., which are used at the anteceedent and consequent parts of a fuzzy inference arithmetic. An inference rule control part 106 takes an inference rule out of a fuzzy inference rule storage part 103. A fuzzy inference arithmetic part 107 performs a fuzzy inference in accordance with the inference rule against the initial data received from a work area 102 or the intermediate data on an inference. Then the part 107 outputs the result of inference. A fuzzy number/real number arithmetic part 108 converts the fuzzy number of the result of inference into a real number. If the inference is not finished yet, the converted real number value is stored in the area 102 as an intermediate variable. Then an inference mechanism part 104 performs the next inference. When the inference is completed, the real number value is inputted to a result output part 109 as the result of inference.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a fuzzy multi-stage inference device that performs multi-stage inference by using the inference results obtained by inference using fuzzy inference rules as input again.

BACKGROUND OF THE INVENTION A conventional fuzzy inference device has a large number of IF to THE
The inferences indicated by the N~ rules are simultaneously processed in parallel on certain input data, and the inference operation is completed in one step, which is used as the control operation amount of the control device (for example, Toshiba L/H, -Vo
i43 No.4 PP3GON303).

A conventional method for determining the control amount based on fuzzy inference will be explained.

For example, in the case of simple temperature control to keep the room temperature constant, the deviation e1 between the room temperature and the target temperature is the rate of change d
The relationship between the two human power of e and the control output (operation amount) U, which is the amount of heating power of the heater, can be described as the following IF~DHEN... rule.

IF e Is Approxlmate Zero
and do l5Posltlye Medium THEN u is Negative Me
dium, -...-(1) The inference rule for equation (1) is ``If the room temperature is close to the target temperature and the room temperature rises moderately (slowly), then set the heat to medium. Indicates the rule, ``Be loose.'' Fuzzy control (inference) prepares a plurality of inference rules such as equation (1). For example, as the second inference rule, IF e Is Approximate Zero
and de ism pprox1mate Zer.

TO): N u is legat Ive S
mall.

Prepare etc. Here, the part IF~ is the antecedent part, T
HE N-... is the consequent, and e Is Ap
proxlmato Zero” etc. as an antecedent proposition, “u
1s Negative) Iedlu+s” is called the consequent proposition. Approximate Zero, Po
sit1veMe old ul, etc. are labels representing fuzzy numbers of inputs and outputs used to describe rules, and are called fuzzy variables. An example is shown in FIG. Usually, fuzzy variables are triangular and symmetrical membership functions. 16BtiveB1g (
FIB), Negative Medlus (IN)
, Negative Small (Is), Ap
proxlmte Zero (20), Po5i
tive Small (PS), Po5itive
Medlum(PM), Po51tlve Big
(PR) (see Figure 7).

Next, the fuzzy inference process will be explained. A normal real value of 611° de- is measured as an actual input value from various sensors, etc., and the fuzzy number of the conclusion of the inference rule (first rule) obtained from equation (1) is determined as follows.

μ+(u):μzs Ce")Aμpn(de")A
μ5n(u)” (2) where A represents mln,
The fuzzy data for one day is a triangular fuzzy number similar to the fuzzy variable.

FIG. 8 shows the fuzzy variable μm(u). As shown in Figure 8, μm(u) compares the degree (membership value) μzs(e”) of 69 belonging to the fuzzy number ZO in the antecedent part and the degree μPN(Δe9) of de” belonging to the fuzzy number PM. Then, it is obtained by cutting (win) the fuzzy variable NM of the consequent part with the degree μzi(e@) having the smallest value.

Since there are multiple control rule equations (1), the fuzzy numbers that combine the fuzzy numbers of all the conclusions are μv(u)”μ+(u)Vua(u)V, , , using equation [2). ,,V
It is calculated as μn(u). Here, V represents WaX.

As an example, in the case of the first rule and the second rule (n = 2), μv(u):μ+(u)VH is shown in FIG.
2(u). This fuzzy number T is a fuzzy number of the conclusion that represents the amount of control operation, but the actual amount of control operation U@ is not a fuzzy number but a real value, so it is shown below (3)
By adopting the weighted center of gravity of the formula, the final result (control operation amount) u
Determine e (see Figure 9).

U・ 8 dings (u) du u@= ... (3) formula % formula %) Although Mamdanl's method was explained here as the inference method, there are other inference methods, and the formula (3) There is also a method of taking the median value or maximum value of the fuzzy number T of the conclusion as the weighted center of gravity.

The manipulated variable U- obtained in this way is applied to the controlled object as a controlled variable based on fuzzy inference.

Problems to be Solved by the Invention However, the above configuration has the following problems. Because fuzzy inference is processed in one step,
For example, it cannot handle processing that requires multi-stage inference, such as performing fuzzy inference based on input data that occurred at a certain time t, and inferring the next action based on the inference result.

Furthermore, it cannot handle processing that requires multi-stage inference, such as inference using an inference result at a certain time t and input data at time t+1.

Another problem is that multi-stage inference cannot be performed when input data is input as real values or fuzzy labels.

In view of this point, the present invention processes intermediate results of fuzzy inference using real numbers or fuzzy labels.
The purpose of this invention is to provide a device capable of fuzzy multi-stage inference.

Means for Solving the Problems The present invention performs fuzzy inference on real-valued input data using the fuzzy inference rules, converts the fuzzy number intermediate result obtained by the fuzzy inference into a real-valued form, The feature is that multi-stage inference is performed again based on this real value data.

Further, the present invention is characterized in that multi-stage inference is performed from intermediate result data of inference and input data.

Furthermore, the present invention is characterized in that the fuzzy numbers obtained by fuzzy inference are replaced with fuzzy labels, and the fuzzy labels are input as fuzzy data to the inference mechanism section to perform multi-stage inference.

Further, the present invention is characterized in that the fuzzy numbers obtained by fuzzy inference are replaced with fuzzy labels, and this fuzzy label is input as an intermediate result to the inference mechanism unit along with the input data at this point to perform multi-stage inference. It is something.

Effect of the Invention With the above-described configuration, the present invention can perform multi-stage fuzzy inference in order to process the inference result produced in -stage fuzzy inference as an intermediate result as an input for the next inference using a real number or a fuzzy label.

EXAMPLE An example of the first invention will be described below with reference to FIG.

101 is a data input unit into which real values obtained from sensors and the like are input. 102 is a work area section and a data input section 10
1 or intermediate results of inference are stored. 1
03 is a fuzzy inference rule storage unit which stores a large number of inference rules used in fuzzy inference. The rule used here is expressed in the form "rIF...(antecedent part)...THEN...(consequent part)", where the antecedent part is an inferential proposition containing fuzzy variables.

Furthermore, the consequent part is expressed as a function containing ordinary real numbers.

Reference numeral 104 denotes an inference mechanism section, which is composed of the following sections. 10
5 is a fuzzy number storage unit, which is the antecedent part of the fuzzy inference operation,
It stores fuzzy variables used in the consequent that represent the magnitude of the control amount, such as ``negative and large,''``zero,'' and ``positive and small.'' 106 is an inference rule management unit that retrieves inference rules from the fuzzy inference rule storage unit 103; A fuzzy inference calculation unit 107 performs fuzzy inference on the initial data or inference intermediate data from one work area 102 in accordance with inference rules, and derives inference results. 108 is an arithmetic unit that converts the fuzzy numbers resulting from this inference into real numbers. If the inference is in progress, the converted real value is stored in the work area 102 as an intermediate variable, and the inference mechanism unit 104 performs the next inference. When the inference is completed, this real value is inputted to the result output section 109 as the inference result.

The operation of the fuzzy multi-stage inference of the present invention configured as described above will be explained with reference to FIG.

When driving a car, the inter-vehicle distance (OL) at time t is 30 m.
1 Infer the braking strength (BR) when the running speed (OS) is 80/m using the -th inference. Next (time t
+1) Based on the determined braking strength, a voice synthesizer or the like infers how strongly the warning sound should be emitted to the passenger.

The fuzzy inference rule storage unit 103 stores the relationship between the OL, OS, and BR in the form of the following two rules as a knowledge base 1.

R1: IF CL ls Small (P
S) And CS 1s Bfg(PR)T
HEN BR18Big(PB) R2: IF CL Is Blg(PR)And C
S Is Small (PS) THEN BRis S
mall (PS) - Inference rule management unit 10 in the inference of the stage
6 takes out the above-mentioned knowledge base 1 from the fuzzy inference rule storage unit 103. And initial value data (0L=3h+
, 0S=80Ks8), the fuzzy inference calculation unit 1
In step 07, the 1n operation is performed according to the above-mentioned equation (1), and the goodness of fit of the antecedent part and the consequent part is calculated for each rule.

These results are combined using the max operation in equation (2), and the inference result T
, get . Although it is possible to pass the inference result T as an intermediate result to the next inference, there are the following two problems.

Since the initial data of the fuzzy inference calculation unit 107 is a real value, an inference configuration for inputting a real value is required.

On the other hand, intermediate results are input as fuzzy numbers, so
The fuzzy inference calculation unit 107 requires an inference configuration for inputting fuzzy numbers and is complicated.

Furthermore, if intermediate results are passed to multiple stages as fuzzy numbers, n+ax operations are performed, so the height of the fuzzy numbers in the inference results is low and the range is widened, resulting in inferences that differ from reality.

Therefore, the center of gravity of the inference result T+ is calculated by the fuzzy real number calculation unit 108 according to the above-mentioned equation (3), and the real value u+"
is stored in the work area 102 as an intermediate result. Needless to say, in addition to the center of gravity, it is also possible to use the height method to take the highest point of the fuzzy number.

In the second stage of inference, the inference rule management unit 106 takes out the following three rules stored in the fuzzy inference rule storage unit 103 as the knowledge base 2.

R1: Summer F BRIs Small (PS) T
HEN AL Is Small (PS) R2: IF
BRis Medium (PM) THEN AL i
s Nedlum (PM) R1: IF BRIs B
ig (PR) THEN AL Is Big (PR) Next, the fuzzy inference calculation unit 107 performs inference similar to the first stage by inputting the intermediate result U, and obtains the inference result Te as a fuzzy number. The fuzzy number real number calculation unit 108 converts it into a real number u2°, and inputs it to the speech synthesizer to generate a warning sound to the fellow passenger. Thereafter, the third and subsequent stages of inference are performed in the same manner, allowing multi-stage inference.

Furthermore, as the first invention, an example in which only intermediate results are used in the second-stage inference has been described, but as a second invention, it is possible to transfer the time from the work area 102 in FIG. Input data at time t+1 may be inferred together with intermediate results of inference. Since the other configurations are similar to those of the first invention, detailed explanations will be omitted. This configuration enables multi-stage inference using the results of inference from past data and current input data, and is especially convenient for learning and predictive inference.

Next, an embodiment of the third invention will be described with reference to FIG. The difference from the first invention is that the input is not a real value but a fuzzy number. That is, 301 is a data input unit into which fuzzy data input from a control panel or the like is input. A work area 302 stores initial value data or intermediate results of inference. A fuzzy inference rule storage unit 303 stores a large number of inference rules used in fuzzy inference. The rules used here are rIF... (antecedent part).
...THEN... (consequent part)" is expressed in the form.

The antecedent part is expressed by an inferential proposition containing fuzzy variables, and the consequent part is expressed by a function containing ordinary real numbers.

Reference numeral 304 denotes an inference mechanism section, which is composed of the following sections.

A fuzzy number storage unit 305 stores fuzzy variables representing the magnitude of the control amount, such as "negative and large,""zero," and "positive and small," used in the antecedent and consequent parts of the fuzzy inference calculation. 306 is an inference rule management unit that retrieves inference rules from the fuzzy inference rule storage unit 303; 307 is a fuzzy inference calculation unit that applies inference rules to the initial data or inference intermediate data from the work area 302,
Perform fuzzy inference and derive inference results. 308 is an arithmetic unit that converts the fuzzy number resulting from this inference into a fuzzy label. If inference is in progress, the converted fuzzy label is stored in the work area 302 as an intermediate variable, and the inference mechanism unit 304 performs the next inference. When the inference is completed, this real value is inputted to the result output unit 309 as the inference result.

The operation of the fuzzy multi-stage inference of the present invention configured as described above will be explained with reference to FIGS. 4 and 5.

When driving a car, the braking strength (B
R) is inferred using the -th stage of reasoning. Next (time t+1),
Depending on the strength of the brakes determined, the right side of the display indicates caution, and the left side indicates safety, and the system infers in what range the lamp should be lit to warn passengers.

Similar to the first invention, the fuzzy inference rule storage unit 303 stores the relationship between CL, OS, and BR in the form of the following two rules as knowledge base 1.

R1: IF CL Is Small (PS) An
d CS is 81g (PB) THEN BRIs
Big(PH) R2: IF CL Is Blg(PB) And
CS 1s Small (PS) THEN BRIs
Small (PS) - Inference rule management unit 3 for stage inference
06 takes out the above-mentioned knowledge base 1 from the fuzzy inference rule storage unit 303. And initial value data (OL=Med
1 nm.

OS: Medlum), fuzzy inference calculation unit 3
In step 07, m1n calculation is performed according to the formula shown below, and the degree of suitability of the antecedent part and the consequent part is calculated for each rule.

When the fuzzy numbers (fuzzy data) F'' and Fl of ``the inter-vehicle distance is medium'' and ``the traveling speed is medium'' are input from the control panel, etc., the fuzzy number of the conclusion of the above inference rule (first rule) is The numbers are as follows.

μ+(u)"μps(F") Aμpa (F') However, μps(F"): 1aX(μps(f) AμplI(f
))μpe(F'to:n+ax(μpe(f)Aμp'
(f)) Here, A indicates 1n, and the fuzzy data door,
1'l is a triangular fuzzy number similar to the fuzzy variable. FIG. 4 shows the fuzzy variable μI(u). As shown in Fig. 4, μ, (U) is the degree (membership value) of F- belonging to the fuzzy number PS of the antecedent part μps (F”), and the degree μpe (F') of Fl belonging to the fuzzy number PR is obtained by comparing the values and cutting (1n) the fuzzy variable PB of the consequent part at the degree μps (F'') having the smallest value.

Since there are multiple control rule equations (1), the fuzzy number T3 that combines the fuzzy numbers of all the conclusions is II yz(11)”μ+ (u)V Is (u)V using the above-mentioned equation (2).
is required. This fuzzy number T3 is a fuzzy number of the inference result representing the control operation amount, but for the same reason as the first invention, the fuzzy label calculation unit 308 generates a fuzzy label
Convert to 3.

As a method of converting to fuzzy labels, the obtained inference result T3 and Small 1 (PS), led lung
(PM) and B fg (PR) so that the difference in area between the respective fuzzy labels is minimized. In the example shown in Figure 4, PM is selected and the fuzzy label F3 is assigned to work area 3.
02 and use it as the second stage inference data.

In the second stage of inference, the inference rule management unit 308 takes out the following three rules stored in the fuzzy inference rule storage unit 303 as the knowledge base 2.

R1: IF ORIs Sa+all(PS)THE
N AL 1s Small (PS) R2: IF B
Ris Medium (PM) THEN AL Is
Medlum(PM)R1: IF BRi8 Big
(PR) THEN AL is Big (PB) Next, in the fuzzy inference calculation unit 307, the intermediate result F2 is input and the same inference as in the first stage is performed, and the inference result T4 is obtained as a fuzzy number as shown in FIG.

The fuzzy label calculation unit 368 converts it into a fuzzy label F4, which is displayed on the left and right positions of the panel surface as a warning to fellow passengers. Thereafter, the third and subsequent stages of inference are performed in the same manner, allowing multi-stage inference.

In addition, as a third invention, an example was described in which only intermediate results are used in the second stage of inference, but as a fourth invention, the time as shown by the dotted line from the work area 302 in FIG. Input data at time t+1 may be inferred together with intermediate results of inference. Since the other configurations are similar to the third invention, detailed explanations will be omitted. This configuration enables multi-stage inference using the results of inference from past data and current input data, and is especially convenient for learning and predictive inference.

Further, as a sixth invention, as shown in FIG. 6, the fuzzy inference rule storage unit 103 and the fuzzy inference rule storage unit 303 that store the knowledge base are unified into one, and the fuzzy inference mechanism 104 and the fuzzy inference By having each mechanism 304, multi-stage inference can be performed whether the input is a real number or a fuzzy number, the same knowledge base can be used, and the effort and memory required for creating inference rules can be halved.

As described in detail, according to the present invention, fuzzy multi-stage inference is possible by passing intermediate results of fuzzy inference to the next inference using real numbers or fuzzy labels, which is even more efficient than -stage inference. Can make inferences similar to human reasoning.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a fuzzy multi-stage inference device according to an embodiment of the first and second inventions, FIG. 2 is an explanatory diagram of the fuzzy inference calculation process in the first invention, and FIG. FIG. 4 is an explanatory diagram of the first stage fuzzy inference calculation process in the third invention, and FIG. 6 is a block diagram of the second stage fuzzy inference in the third invention. An explanatory diagram of the calculation process, FIG. 6 is a block diagram of a fuzzy multi-stage inference device according to an embodiment of the fifth invention, FIG. 7 is an explanatory diagram of fuzzy variables in the present invention, and FIG. 8 is a fuzzy inference process in the inference rule. FIG. 8 is an explanatory diagram of the frequency of the inference results. 101...Data input section, 102...Work area section, 10
3... Fuzzy inference rule storage unit, 104... Inference mechanism unit, 105... Fuzzy number storage unit, 108... Inference rule management unit, 107... Fuzzy inference calculation unit, 108... Fuzzy number real number calculation unit, 109 ...Result output section, 301...
Data input section, 302...Work area section, 303...Fuzzy inference rule storage section, 304...Inference mechanism section, 305.
-Fuzzy number storage section, 308...Inference rule management section. 307...Fuzzy inference calculation unit, 308...Fuzzy label calculation unit, 308...Result output unit. Name of agent Patent attorney Shigetaka Awano and 1 other person Time t (-sth m1) -J Shenwangshi℃↑1 (Second-seat m1) Kg) Figure antecedent part Minato sweat stroke Figure Figure Figure Figure Figure ■

Claims (7)

[Claims] (1) An inference rule storage unit that stores fuzzy inference rules, an inference mechanism unit that performs fuzzy inference using the fuzzy inference rules on real-valued input data, and a task that stores intermediate results of inference and input data. The inference mechanism section stores the intermediate results of fuzzy numbers obtained by fuzzy inference in the form of real numbers in the work area, and inputs this real value data to the inference mechanism section again to perform multi-stage inference. A fuzzy multi-stage inference device that performs the following. (2) An inference rule storage unit that stores fuzzy inference rules, an inference mechanism unit that performs fuzzy inference using the fuzzy inference rules on real-value input data, and a task that stores intermediate results of inference and input data. The inference mechanism section stores the intermediate results of fuzzy numbers obtained by fuzzy inference in the form of real numbers in the work area, and again combines the intermediate result data of real numbers with the input data at this point. A fuzzy multi-stage inference device is characterized in that both are input to an inference mechanism section to perform multi-stage inference. (3) The fuzzy multi-stage inference device according to claim 1 or 2, characterized in that the center of gravity of intermediate results of fuzzy numbers obtained by fuzzy inference is taken, and the result is stored in the work area in the form of real numbers. (4) An inference rule storage unit that stores fuzzy inference rules, an inference mechanism unit that performs fuzzy inference using the fuzzy inference rules on input fuzzy data, and a work area that stores intermediate results of inference and input data. It consists of
The inference mechanism unit replaces the fuzzy numbers obtained by fuzzy inference with fuzzy labels, stores the fuzzy labels in the work area, and inputs the fuzzy labels to the inference mechanism unit again as fuzzy data to perform multi-stage inference. A fuzzy multi-stage inference device. (5) An inference rule storage unit that stores fuzzy inference rules, an inference mechanism unit that performs fuzzy inference using the fuzzy inference rules on input fuzzy data, and a work area that stores intermediate results of inference and input data. It consists of
The inference mechanism section replaces the fuzzy numbers obtained by fuzzy inference with fuzzy labels, stores them in the work area as fuzzy labels, and inputs both the fuzzy label of this intermediate result and the input data at this point into the inference mechanism section again. , a fuzzy multi-stage inference device characterized by performing multi-stage inference. (6) Claim 4 or 5, characterized in that the fuzzy label that minimizes the difference in area between the fuzzy number intermediate result obtained by fuzzy inference and the fuzzy label is selected and stored in the work area as the intermediate result. Fuzzy multi-stage reasoning device described. (7) An inference rule storage unit that stores fuzzy inference rules, an inference mechanism unit that performs fuzzy inference using the fuzzy inference rules on real-valued input data, intermediate results of this inference, and input data at this time. This inference mechanism section stores the intermediate results of fuzzy numbers obtained by fuzzy inference in the form of real numbers in the work area, and inputs this real value data to the inference mechanism section again. an inference mechanism section that performs fuzzy inference on input fuzzy data using fuzzy inference rules from the inference rule storage section; and an input of intermediate results of this inference and fuzzy labels at this time. This inference mechanism section replaces the fuzzy numbers obtained by fuzzy inference with fuzzy labels, stores them in the work area with fuzzy labels, and then transfers the fuzzy labels to the inference mechanism section again as fuzzy data. A fuzzy multi-stage inference device comprising a means for inputting and performing multi-stage inference.