JPH05289880A - Fuzzy knowledge extracting device and fuzzy knowledge detailing/simplifying device and fuzzy inference device - Google Patents

Abstract

PURPOSE:To constitute the device so that a man can understand and correct easily, even if the quantity of fuzzy knowledge increases. CONSTITUTION:A parameter PFZ of a fuzzy inference is inputted to a knowledge setting device 2 through an input device 1, and it is set to a fuzzy inference device 4 and a fuzzy knowledge data base 5. In a fuzzy adjustment data storage device 6, learning input data Xi for executing a recollection processing, and expected output data for evaluating a result of inference are stored, and by giving its learning input data to the fuzzy inference device, a fuzzy inference is executed by using an MIN-MAX method, etc., a calculated output data train Yi is sent to an evaluation value calculating device 7 of the post-stage, the output data and the expected output data are compared therein, an error is derived, a rule of a large error is extracted, and detailing, etc., are executed. The parameter is stored hierarchically, and in accordance with requested approximate accuracy, necessary knowledge is read in the fuzzy inference device.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fuzzy knowledge extracting device, a fuzzy knowledge detailing / simplifying device and a fuzzy inference device.

[0002]

2. Description of the Related Art As is well known, in a fuzzy reasoning device,
Fuzzy knowledge consisting of a predetermined number of fuzzy rules shown in the "IF ... THEN ..." format and a membership function that defines a fuzzy label that is input to the conditional part that constitutes the IF part of the first half of the fuzzy rule is set. When the input signal (variable) is given, the membership value (fuzzy label) for the input signal is obtained from the membership function, and the fuzzy label is substituted into the conditional part in each fuzzy rule to calculate, MIN-MAX
I try to draw conclusions by inferring by law. Then, when making the inference, it calculates and processes all fuzzy rules.

[0003]

In the conventional fuzzy inference apparatus described above, when the number of labels of the input signal (variable) or the membership function is increased in order to perform control with higher accuracy, the number of fuzzy rules is increased accordingly. Will increase exponentially, making it difficult for humans to understand fuzzy knowledge.

And, among the set fuzzy knowledge,
There may be unnecessary rules (which do not contribute much to the inference result), the rule settings may not be sufficient, and more detailed rule settings may be required. Therefore, it is desirable to perform correction work such as addition, correction, deletion, etc. as appropriate. However, since it is difficult to understand the fuzzy knowledge as described above, it becomes impossible to appropriately perform the correction work.

Therefore, if the correction work is carried out, the rules and the like are exclusively added, and there is a possibility that unnecessary rules and the like are increased. Moreover, the increase of fuzzy knowledge causes a problem that the inference time becomes long.

The present invention has been made in view of the above background, and an object of the present invention is to enable a person to easily understand the fuzzy knowledge even if the amount of the fuzzy knowledge increases. In addition, it is possible to freely and easily refine and simplify the knowledge by appropriately judging the goodness / badness of the fuzzy knowledge, and to obtain the fuzzy knowledge suitable for the required approximation accuracy. Fuzzy knowledge extraction device and fuzzy knowledge detailing /
It is to provide a simplification device and a fuzzy reasoning device.

[0007]

In order to achieve the above-mentioned object, a fuzzy knowledge extracting apparatus according to the present invention is a fuzzy inference apparatus for executing fuzzy inference using fuzzy knowledge consisting of fuzzy rules and membership functions. A fuzzy inference applying a fuzzy knowledge, which is provided with a storage means for storing the learning input data to be the target of the fuzzy inference and the expected output data corresponding to the learning input data, and receiving the input of the learning input data. A fuzzy inference unit to be executed, a range extraction means for comparing the data output from the fuzzy inference unit with the expected output data, and extracting a range of values of the learning input data having an error larger than a reference value; Fuzzy rules included in the fuzzy rule group having the range of the learning input data extracted by the extraction means in the antecedent part It was constructed and means for extracting.

Further, the fuzzy knowledge refinement apparatus according to the present invention comprises means for replacing the fuzzy rules extracted by the fuzzy rule extraction means with a plurality of fuzzy rules, and the fuzzy knowledge simplification apparatus further comprises: Range extraction means for comparing the data output from the fuzzy inference unit with the expected output data and extracting a range of values of the learning input data whose error is smaller than the reference value; and the learning purpose extracted by the range extracting means. It is composed of a fuzzy rule group extracting means for extracting a fuzzy rule group having a continuous range of input data in the antecedent part, and a means for replacing the fuzzy rule group extracted by the fuzzy rule group extracting means with a single fuzzy rule. ..

Further, in the fuzzy reasoning apparatus according to the present invention, means for hierarchically storing the fuzzy knowledge and knowledge of the fuzzy knowledge stored in the means that satisfies the approximation accuracy required for the inference result are The fuzzy inference device is read and the knowledge is used to infer the input data.

[0010]

[Operation] The set fuzzy knowledge is subjected to the recall process using the learning data and the expected output data, the error of the output data is obtained, and the learning input data larger than the reference value is included in the antecedent part. Extract fuzzy rules. This extracted fuzzy rule is a rule that reduces the approximation accuracy. Then, the detailing device is used to hierarchically refine the rule to improve the approximation accuracy. Further, by replacing the fuzzy rule group including the learning input data in which the error of the output data is smaller than another reference value with one rule,
Fuzzy knowledge is simplified and the contents are easier to understand without significantly reducing the approximation accuracy (error).

Further, in order to actually perform fuzzy inference using the fuzzy knowledge set in this way, of the fuzzy knowledge, the fuzzy knowledge necessary to obtain knowledge of desired approximation accuracy necessary for control or the like. Is set to the fuzzy inference device, and the given knowledge is used to infer the given input.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of a fuzzy knowledge extracting device, a fuzzy knowledge detailing / simplifying device and a fuzzy inference device according to the present invention will be described in detail with reference to the accompanying drawings.
FIG. 1 shows the configuration of a fuzzy inference system capable of executing various devices according to the present invention. As shown, the fuzzy inference parameter P _FZ is input to the knowledge setting device 2 via the input device 1 such as a keyboard and a mouse. As shown in FIG. 2, the parameter P _FZ is composed of a fuzzy knowledge hierarchical structure L and fuzzy knowledge contents P _{FZi'j 'of} each hierarchy. The parameter P _FZ is displayed on the liquid crystal display 3 connected to the knowledge setting device 2 until the processing is completed. In this example, the liquid crystal display 3 is used for displaying, but it goes without saying that it may be replaced with a display device such as a CRT.

Here, the above-mentioned parameter P _FZ will be described in detail. First, in the present invention, as shown in FIG. 3, each fuzzy knowledge content P _{FZi'j '} is constructed in layers, and the fuzzy knowledge contents P _FZi'j' The approximation accuracy increases (high-precision inference can be performed) as the hierarchy becomes deeper, such as the second layer, the third layer, .... The second layer and the third layer are detailed versions of a part of the fuzzy knowledge of the first layer. However, the second layer is a detailed version of the fuzzy knowledge of the first layer.
As is apparent from the figure, the third layer is not necessarily a detailed version of a part of the fuzzy knowledge of the second layer, but may be a detailed version of a part of the fuzzy knowledge of the first layer.
That is, the fuzzy knowledge of a certain hierarchy is a detailed version of a part of the fuzzy knowledge of a predetermined hierarchy located in a higher hierarchy. Such a hierarchical structure makes it easy for humans to understand the content of knowledge even if the amount of knowledge such as the number of rules increases.

Based on such a premise, the hierarchical structure L is specifically as shown in FIG. Here, the approximation accuracy E _{pi '} and the accuracy of the output value obtained as a result of performing fuzzy inference using the fuzzy knowledge are shown. That is, the lower the approximation accuracy E _{pi '} is, the larger the allowable error of the output value is, but the time required for the inference is shortened, and the outline can be known relatively roughly. Further, the higher the approximation accuracy E _{pi '} is, the smaller the allowable error of the output value becomes and the more accurate control is possible, but the time required for inference becomes longer. The approximation accuracy E _{pi '} is not the approximation accuracy of only the fuzzy knowledge set to the i-th layer, but the approximation accuracy when all the fuzzy knowledge set from the first layer to the i'th layer is used. Therefore, the value calculated during the recall process is input. The specific calculation process will be described later.

Further, the above fuzzy knowledge contents P
_{FZi'j '} is the number of inputs / outputs, the number of its membership functions, and the coordinate values of each standardization constant of the antecedent part and each output (singleton) of the consequent part. The data structure is as shown.

On the other hand, the fuzzy knowledge parameter P _FZ given to the knowledge setting device 2 can exchange data with the fuzzy inference device 4 which carries out fuzzy inference, and further, this fuzzy inference. The device 4 is adapted to transmit / receive fuzzy knowledge, that is, write / read, also to / from the fuzzy knowledge database 5 in which the created fuzzy knowledge is stored.

The fuzzy adjustment data storage device 6 composed of a ROM or the like stores learning input data Xi for performing recall processing and expected output data Di for evaluating inference results. .. Then, the learning input data Xi is given to the fuzzy inference device 4,
There, the learning input data Xi is input to the fuzzy knowledge set in the fuzzy inference apparatus 4 and M
Fuzzy inference is performed using the IN-MAX method or the like, and the calculated output data string Yi is sent to the evaluation value calculation device 7 in the subsequent stage.

Here, the learning input data string Xi is
Means the i-th input data string, specifically (xi1, x
i2, ..., xiIN), and the expected output data string Di means the i-th output data string, specifically, (di1, di2, ..., DiOUT). Further, the i-th output data string Yi is (yi1, yi2, ..., yiOUT)
Has become. Further, the above i means the order of the input / output data strings, is an integer of 1 to n (where n is the total number of input / output data strings), and IN, which is a subscript of x, indicates the number of inputs.
OUT, which is a subscript of y and d, indicates the number of outputs.

Then, the evaluation value calculation device 7 uses the given output data string Yi and the expected output data Di sent from the fuzzy adjustment data storage means to evaluate the output data string Yi. The (error) Ei is calculated by the following calculation formula.

That is, the evaluation value data eir for each output r of the i-th output data string is obtained by eir = dir-yir, and the evaluation value Eir of the i-th output data string is calculated as follows from the calculation result. Become.

Ei = (ei1, ei2, ..., eiOUT) Then, in the evaluation value calculating device 7, the evaluation value Ei obtained as described above, the output data string Yi and the expected output data string Di are displayed on the liquid crystal display. 3, the liquid crystal display 3 also displays the learning input data string Xi sent from the fuzzy adjustment data storage device 6. Further, in the present example, the evaluation value calculation device 7 uses the evaluation value Ei to obtain the approximation accuracy Ep.

Next, an example in which the apparatus having the above-mentioned configuration is used as the fuzzy knowledge extracting apparatus according to the present invention will be described. First, a description will be given of the input and correction work of fuzzy knowledge which is a prerequisite for operating the device of the present invention including the present device. As shown in FIG. 6, when the present system is operated and the fuzzy knowledge is initialized. , Input a new fuzzy inference parameter P _FZ from the input device 1,
The input parameter P _FZ is displayed on the liquid crystal display 3 until the initial setting work is completed (S101 to 10).
3). Then, the input work and the display process are repeatedly performed until the input of all parameters is completed (S1).
04) When the initial setting is completed, the set parameter P _FZ is set in the fuzzy inference device 4 and further set in the fuzzy knowledge database 5 (S105, 10).
6). This completes the initial setting.

When the already set fuzzy knowledge is to be corrected, in step 101, the process jumps to step 201 shown in FIG. 7 and the parameter P _{FZ of} the inference to be corrected is _executed.
Is read from the fuzzy knowledge database 5 into the knowledge setting device 2 (S202). Next, if it is necessary to move the layer of the parameter P _{FZ in} order to correct the fuzzy knowledge, the layer move is executed and the layer moved parameter P _FZ.
And the current parameter P _FZ is displayed on the liquid crystal display 3 (S203, 204). Then, the content of the parameter P _FZ is appropriately modified using the input device 1, and the modified parameter P _FZ is displayed on the liquid crystal display 3 each time the modification is performed (S205, 206). Then, the above work is repeated, and when the correction is completed, the corrected parameter P _FZ is set in the fuzzy inference apparatus 4 and the fuzzy knowledge database 5, and the work is completed (S207 to 209).

Next, a processing procedure for actually performing the recall process will be described. Since both of the steps 101 and 201 are "N", the process proceeds to step 301 in the flowchart shown in FIG. That is, the parameter P _FZ used for recall processing is read from the fuzzy knowledge database 5 (S302). Then, it is determined whether knowledge is selected with approximation accuracy (S303),
When selecting, the target (required) designated approximation precision E _FZ is input, and the high-precision approximation precision Epi (the numerical value representing the precision is small) that satisfies the designated approximation precision E _FZ (i-th i)
It is judged whether or not there is the approximation accuracy of the th layer (S3).
04, 305). If there is no hierarchy having the corresponding approximation accuracy, an error message is displayed on the liquid crystal display 3 and the process ends (S306).

If there is a corresponding one, the hierarchy of the fuzzy knowledge having the lowest precision among the corresponding approximation precisions is selected, and the parameters necessary for calculating the hierarchy of the selected parameter P _FZ are selected. _PFZ is a fuzzy inference device 5
Is set to (S307, 308). That is, if the third
Assuming that the approximation accuracy of the layer is applicable, the parameters of the third layer and the upper layers, that is, the first layer and the second layer are set.

The fuzzy adjustment data storage device 6
The learning input data Xi stored in is sequentially input from i = 1 to n, is inferred using the knowledge set above, the output data string Yi is calculated, and the calculation result is calculated by the evaluation value calculation device 7
Set to. When obtaining the evaluation value, the expected output data Di is read and the evaluation value Ei is calculated (S3).
09-317).

When the above processing is performed for all the input data strings, the recall result is read from the evaluation value calculation device, displayed on the liquid crystal display 3, and the processing ends (S318, 319). The recall result displayed on the liquid crystal display 3 includes the evaluation value Ei obtained by the evaluation value calculation device 7 and the approximation accuracy Ep calculated from the evaluation value Ei. This approximation accuracy Ep can be obtained using, for example, the sum of error squares or the maximum error, and in this example, the sum of error squares obtained by the following formula is used.

[0028]

[Equation 1] On the other hand, when the recall process is performed by designating the highest approximation precision, in step 303, jump to step 320 in FIG.
The smallest evaluation value among the parameters P _FZ (highest approximation accuracy)
After selecting the fuzzy knowledge of the hierarchy (S320, 32)
1), jumping to step 308, the above inference processing and the like are performed.

Further, when the fuzzy knowledge is selected by designating the hierarchy of the fuzzy inference parameter P _FZ , the step 320 further jumps to the step 322, and the fuzzy knowledge of the designated hierarchy is selected (S322, S322). 32
3), jumping to step 308, the above inference processing and the like are performed.

Then, in the present extraction device, the evaluation value Ei obtained by the above-mentioned recall processing is sent to the knowledge setting device 2, where it becomes an evaluation value Ei (a large error) equal to or greater than a certain value (reference value). The learning input data corresponding to the output data is extracted, and the fuzzy rule having the extracted learning input data in the antecedent part is extracted. That is, the range setting means for performing the above two processes and the means for extracting the fuzzy rules are arranged in the knowledge setting device 2. Then, in this example, the extracted fuzzy rule is displayed on the liquid crystal display 3.

As a result, it is possible to extract a fuzzy rule that causes a problem (has an adverse effect) in the range of the input having a large error in the fuzzy inference, and correct or delete the extracted fuzzy rule. By doing so, it is possible to set highly accurate fuzzy knowledge.

Next, a case where the above-mentioned apparatus shown in FIG. 1 is applied to the fuzzy knowledge detailing apparatus according to the present invention will be described. This device has basically the same configuration as the above-mentioned extraction device, and extracts fuzzy rules that cause a large evaluation value (error) and cause a large inference error by performing recall processing and the like. be able to. Further, in this example, a rule replacement means for replacing the extracted fuzzy rules with a plurality of subdivided fuzzy rules is provided, and this rule replacement means is also arranged in the knowledge setting device 2.

Next, the operation of this refining device will be described. First, in this example, it is assumed that the fuzzy rule used for convenience is a 2-input 1-output fuzzy knowledge, as shown in FIG. Then, in order to refine, the part of the fuzzy knowledge to be refined must first be determined. However, the criterion for this decision is that the error (evaluation value) of the output data obtained as a result of performing the recall process becomes large. , Extract the part that causes the approximation accuracy to deteriorate. Here, when there is existing fuzzy knowledge A, the consequent part of that fuzzy knowledge A

[0034]

[Outer 1] The error in the part of the rule using is large, and the refinement of this part is executed. Then, only the part of the fuzzy knowledge to be refined is extracted, and the detailed fuzzy knowledge is defined as the fuzzy knowledge of the lower hierarchy. That is, as shown in FIG. 11, the fuzzy knowledge B obtained by refining a part of the fuzzy knowledge A set in the first layer is set in the second layer.

Further, when the error of the predetermined portion of the fuzzy knowledge B is large, the fuzzy knowledge which further refines the portion is provided and set in the third layer. In this way, fuzzy knowledge can be constructed sequentially in hierarchical layers, and the created fuzzy knowledge is stored in the fuzzy knowledge database 5.

Further, when the fuzzy knowledge formed in this way is subjected to the recall processing, the above-mentioned FIG.
It can be executed by the same processing procedure as in FIG. 11, but here, for example, when the approximation accuracy to be used is designated, the second
When it becomes necessary to use the fuzzy knowledge B of a layer, in this example, inference processing is performed using not only the fuzzy knowledge B but also the rules of the detailed part of the fuzzy knowledge A. (Of course, other fuzzy knowledge A that has not been refined is also used). If the designated approximation accuracy is low and the inference with a relatively large error is acceptable, the inference processing is performed using only the fuzzy knowledge A before the detailing.

On the other hand, in the case of modifying the fuzzy knowledge constructed as described above, basically, the fuzzy knowledge of the upper hierarchy is modified, and based on this, the fuzzy knowledge of the lower hierarchy is automatically It is supposed to be fixed. That is, for example, among the standardization constants of the membership function of the antecedent part of the fuzzy knowledge A,

[0038]

[Outside 2] Is corrected by Δr. Then, if the modified part of the fuzzy knowledge of the upper hierarchy is detailed by the fuzzy knowledge of the lower hierarchy, the standardization constant of the membership function of the antecedent part of the fuzzy knowledge B of the lower hierarchy.

[0039]

[Outside 3] Is corrected in proportion to the existing distance as shown in the table below. The correction is also automatically performed by the knowledge setting device 2.

[0040]

[Table 1] As described above, in the refining device according to the present invention, it is possible to reduce the error by refining the fuzzy knowledge that causes the error in the higher hierarchy. In addition, if fuzzy knowledge with a large error is basically left as it is, when the required approximation accuracy is low, the inference processing is performed without using detailed knowledge by dividing the fuzzy knowledge with a predetermined hierarchy that satisfies the requirement. Therefore, the amount of knowledge used is small and high-speed processing can be achieved. In addition, since the systematic details have been gradually developed, it is possible to perform the modification work only on the upper hierarchy where the number of rules etc. is small and humans can easily understand, and the modification work of the fuzzy knowledge once set. Can be done easily and accurately.

Next, a case where the above-mentioned apparatus shown in FIG. 1 is applied to the fuzzy knowledge simplifying apparatus according to the present invention will be described. The basic configuration and functions are the same as those of the above-mentioned devices, but the judgment criteria in the range extraction means are
Contrary to that of the above apparatus, learning input data corresponding to output data smaller than a predetermined reference value (with a small error) is extracted. Then, in place of the fuzzy rule extracting means, a fuzzy rule group extracting means for extracting a fuzzy rule group having a continuous range of the learning input data extracted by the range extracting means in the antecedent part is arranged in the knowledge setting device 2. Further, the rule replacing means provided in the knowledge setting device 2 is also arranged to replace a fuzzy rule group consisting of a plurality of fuzzy rules with a single fuzzy rule, contrary to the simplification device. It has a function.

Next, the operation of this refining device will be described with reference to FIG. First, the part of the fuzzy knowledge to be simplified is determined, but this decision is made by extracting the part of the fuzzy knowledge that does not increase the error of the output data as much as possible and does not deteriorate the approximation accuracy. Change the resolution of the singleton value of the subject part (1
It is possible to change the resolution into two steps to make it coarser) and simplify the place where all the singleton values of the consequent part for a plurality of membership functions of a certain input match.

Specifically, for example, when there is existing fuzzy knowledge A as shown in the figure, the resolution of the consequent part is reduced. At this time, if the values of the singletons of the consequent part with respect to the membership functions of the first part and the second part of the antecedent part of the input 2 of the fuzzy knowledge A are all equal as shown in the table below, this part is simplified. It will be.

[0044]

[Table 2] Then, only this simplified part is extracted, and a new simplified fuzzy knowledge is defined over the entire range of the original fuzzy knowledge. That is, in this example, fuzzy knowledge A
The fuzzy knowledge B formed by simplifying a predetermined part of the above is defined over the entire range of the fuzzy knowledge A.

Then, this simplified fuzzy knowledge B
Is defined as the upper layer fuzzy knowledge, and the lower layer fuzzy knowledge is defined so that only the simplified part is extracted to realize the fuzzy knowledge before the simplification. That is, the simplified fuzzy knowledge B is used as the fuzzy knowledge of the first layer, and in order to realize the original fuzzy knowledge A, the simplified part of the fuzzy knowledge B before the simplification is changed to the fuzzy knowledge C. The fuzzy knowledge C (which is a detailed version of the fuzzy knowledge B) is defined as the fuzzy knowledge of the second layer, and the fuzzy knowledge database 5
To store.

In this way, the simplified fuzzy knowledge is set in the upper layer, and the original detailed fuzzy knowledge is defined by the upper layer (fuzzy knowledge B) and the newly created hierarchy (fuzzy knowledge C). Therefore, when performing simple control, inference is performed using only the upper layer, and when performing detailed control, inference is performed using the upper layer and the newly created layer. It will be.

Further, the fuzzy knowledge created by executing the present apparatus is processed in the reverse of the above-described refinement apparatus (in the refinement apparatus, the knowledge is constructed toward the hierarchy, while the simplification is simplified). Although the device is created by building knowledge toward the upper layer), the knowledge that is finally formed is created while the lower layer is detailed while being associated with the upper layer. Since they are the same, the recalling process and the correcting process in the refinement device are also performed in the same procedure in the simplification device. However, (A) and (B) in the table showing the correction process are (B) and (B), respectively.
It shall be replaced with (C).

As described above, in the simplification device according to the present invention, it is possible to reduce the amount of fuzzy knowledge required and the time required for inference. Moreover, since the multiple fuzzy rules before simplification extracted the singletons whose outputs are equal, the evaluation value (error) and the approximation accuracy in the inference result are not much lower than those before simplification. ..

Further, in the present invention, a detailed / simplified device can be constructed by appropriately combining the above-mentioned devices. By doing this, first, roughly create fuzzy knowledge, then perform recall processing using the created fuzzy knowledge, evaluate this knowledge with approximate accuracy, and finish creating if satisfactory However, if you are not satisfied, you can successively refine the knowledge (at this time, you can focus on the details with large errors and prevent unnecessary refinement (creating similar rules)). By
It is possible to efficiently and accurately create fuzzy knowledge. On the contrary, when the accuracy is higher than necessary, it can be simplified and the fuzzy knowledge having a desired approximation accuracy can be efficiently created.

Further, by using a simplification device in addition to the above-described usage examples, correction processing can be easily performed. That is, the fuzzy knowledge that has been created and stored is read, and the fuzzy knowledge is simplified.
Liquid crystal display 3 with rough fuzzy knowledge
To display. Therefore, a person can easily understand the contents by looking at the rough fuzzy knowledge and can easily perform the correction work. Then, the original fuzzy knowledge is corrected in accordance with the correction made to the rough fuzzy knowledge. Then, the recall process is performed on the original detailed fuzzy knowledge, and it is evaluated with approximation accuracy. Then, the above processing is repeated until the desired approximation accuracy is obtained. Therefore, the correction performed by the human is performed on the simplified knowledge, and the original detailed evaluation is performed by the apparatus, so that the correction processing can be easily performed.

Further, in the fuzzy inference apparatus according to the present invention, the fuzzy knowledge obtained by executing each of the above-mentioned apparatuses or the like (which may be set hierarchically from the beginning) is stored. A fuzzy knowledge database 5 is provided, and the approximation accuracy for each hierarchy is also stored. Therefore, when actually performing fuzzy inference for various controls, etc., it is not necessary to use all the fuzzy knowledge stored in the database 5, but it is necessary to obtain the knowledge of the desired approximation accuracy required for the control, etc. Such fuzzy knowledge is set in the fuzzy inference device, and the set knowledge is used for inference with respect to a given input, so that a desired approximation accuracy can be obtained with a short inference time.

[0052]

As described above, in the fuzzy knowledge extracting device, the fuzzy knowledge detailing / simplifying device and the fuzzy inference device according to the present invention, even if the amount of fuzzy knowledge is large, they are appropriately detailed. The fuzzy knowledge can be easily understood by humans because it is hierarchically constructed by making it simple or simplifying. And by properly judging whether the fuzzy knowledge is good or bad,
Knowledge can be refined and simplified freely and easily. Moreover, since the inference processing is performed using the fuzzy knowledge suitable for the required approximation accuracy, the speed required for the inference is improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing an example of a fuzzy inference system for executing various devices according to the present invention.

FIG. 2 is a diagram showing a data structure of a fuzzy parameter.

FIG. 3 is a diagram showing an example of a hierarchical structure of fuzzy knowledge.

FIG. 4 is a diagram showing a hierarchical data structure of fuzzy knowledge.

FIG. 5 is a diagram showing a data structure of fuzzy knowledge contents of each layer.

FIG. 6 is a flowchart showing a process of initializing fuzzy knowledge.

FIG. 7 is a flowchart showing a process of performing a fuzzy knowledge correction work.

FIG. 8 is a flowchart showing a part of a recall process.

FIG. 9 is a flowchart showing a part of a recall process.

FIG. 10 is a flowchart showing a part of a recall process.

FIG. 11 is a diagram showing a configuration of a 2-input 1-output fuzzy.

FIG. 12 is a diagram for explaining the operation of the fuzzy knowledge refining device according to the present invention.

FIG. 13 is a diagram illustrating an operation of the fuzzy knowledge simplification device according to the present invention.

[Explanation of symbols]

2 knowledge setting device 4 fuzzy inference device (fuzzy inference unit) 5 fuzzy knowledge database (means for hierarchically storing fuzzy knowledge) 6 fuzzy adjustment data storage device (learning input data and expected output data are stored as a set) Storage means) 7 Evaluation value calculation device

Claims (5)

[Claims] 1. A fuzzy inference device for executing fuzzy inference using fuzzy knowledge consisting of fuzzy rules and membership functions, and sets learning input data to be the target of the fuzzy inference and expected output data corresponding thereto. A fuzzy inference unit that receives the learning input data and executes fuzzy inference to which the fuzzy knowledge is applied, and compares the data output from the fuzzy inference unit with the expected output data. Range extraction means for extracting a range of values of the learning input data whose error is larger than the reference value, and the range of the learning input data extracted by the range extraction means is included in the fuzzy rule group having the antecedent part. Fuzzy knowledge extraction device comprising means for extracting fuzzy rules. 2. A fuzzy inference device for executing fuzzy inference using fuzzy knowledge composed of fuzzy rules and membership functions, and sets learning input data to be fuzzy inference and expected output data corresponding to the learning input data. A fuzzy inference unit that receives the input of the learning input data and executes a fuzzy inference to which the fuzzy knowledge is applied, and compares the data output from the fuzzy inference unit with the expected output data. A range extraction means for extracting a range of values of the learning input data whose error is larger than the reference value, and a range of the learning input data extracted by the range extraction means are included in the fuzzy rule group having the antecedent part. Fuzzy rule extracting means for extracting the fuzzy rules, and the fuzzy rule extracting means for extracting the fuzzy rules. Fuzzy knowledge detailing device composed of a means for replacing a Iruru multiple fuzzy rules. 3. A fuzzy inference device for executing fuzzy inference using fuzzy knowledge composed of fuzzy rules and membership functions, and sets learning input data to be fuzzy inference and expected output data corresponding to the learning input data. A fuzzy inference unit that receives the input of the learning input data and executes a fuzzy inference to which the fuzzy knowledge is applied, and compares the data output from the fuzzy inference unit with the expected output data. A range extracting means for extracting a range of values of the learning input data whose error is smaller than the reference value, and a fuzzy rule group having a continuous range of the learning input data extracted by the range extracting means in the antecedent part. Fuzzy rule group extracting means for extracting, and fuzzy rule group extracted by the fuzzy rule group extracting means Fuzzy knowledge simplified device comprising means for replacing a single fuzzy rule. 4. The set fuzzy knowledge is refined by using the apparatus according to claim 2 as necessary, or the knowledge is refined by using the apparatus according to claim 3. A fuzzy knowledge detailing / simplification device having knowledge setting means for executing simplification. 5. Means for hierarchically storing the fuzzy knowledge, and among the fuzzy knowledge stored in the means, knowledge that satisfies the approximation accuracy required for an inference result is read into the fuzzy inference apparatus and A fuzzy inference device that uses knowledge to infer input data.