JPH06314266A - Modeling supporting device and method therefor - Google Patents

Abstract

PURPOSE:To automatically select a modeling technique suited to the purpose of a user. CONSTITUTION:In a modeling parts storing device 13, the utilizing conditions of plural kinds of modeling techniques are previously registered. A selection condition for a desired model is inputted from a model condition input device 11. A model adaptability arithmetic unit 12 calculates adaptability between the inputted selection condition and the utilizing conditions of plural kinds of modeling techniques stored in the storing device 13. The modeling technique of the highest adaptabilty or that of adaptability over a prescribed threshold value is selected by a model selection device 14 and outputted from a model output device 19.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to a modeling support apparatus and method for selecting a modeling method suitable for a user's purpose.

[0002]

2. Description of the Related Art Conventionally, as a modeling method of an actual system, there are various modeling methods such as a fuzzy modeling method (Takagi Sugano model), a modeling method applying a fuzzy measure, and a modeling method using a neural network.

However, only the user who knows what model is suitable for each modeling method must make a model by trial and error until a suitable model is found. A sample for making a model・ There were many problems in model creation, such as not knowing how much data was needed, and the problem was that expert know-how was required.

[0004]

DISCLOSURE OF THE INVENTION An object of the present invention is to make it possible to easily find an optimum modeling method which meets the purpose of a user.

The modeling support device according to the present invention is a modeling / parts storing means in which use conditions relating to the above-mentioned modeling method are stored in advance for each of a plurality of modeling methods, and a model condition input for inputting model selection conditions regarding a model to be modeled. Means, a model goodness-of-fit calculation means for calculating a model goodness of fit for each of the modeling methods based on the input model selection conditions and the usage conditions for each modeling method stored in the modeling parts storage means, And model selection means for selecting a modeling method suitable for the input model selection condition based on the calculated model suitability.

In the modeling support method according to the present invention, the usage conditions for the modeling method are stored in advance in the modeling parts storage means for each of a plurality of modeling methods, and the input model selection conditions for the model to be modeled are accepted. Calculating the model suitability for each modeling method based on the input model selection condition and the usage condition for each modeling method stored in the modeling parts storage means, and calculating the model suitability Based on the degree, the modeling method suitable for the inputted model selection condition is selected.

According to the present invention, the model conformity between the use condition of each modeling technique stored in advance and the inputted model selection condition is calculated, and the modeling technique is selected based on this model conformance.

Therefore, even a user who is not familiar with the modeling method can select the optimum modeling method by inputting the model selection conditions for performing the modeling. In addition, since it is not necessary to carry out modeling by trial and error to select the optimal modeling method, the model construction becomes faster.

In a preferred embodiment of the present invention,
Preprocessing storage means in which data relating to a plurality of types of preprocessing is stored in advance, and prestored in the preprocessing storage means based on at least one of the input model selection condition and the selected modeling method. It further comprises a pre-processing selection means for selecting a suitable one from the plurality of types of pre-processing.

Therefore, it is possible to automatically select the kind of preprocessing suitable for the input selection condition or required for the selected modeling method.

In a preferred embodiment of the present invention,
Post-processing storage means in which data relating to a plurality of types of post-processing are stored in advance, and pre-stored in the post-processing storage means based on at least one of the input model selection condition and selected modeling method. It further comprises a post-processing selection means for selecting a suitable one from among the plurality of types of post-processing that are performed.

Therefore, the kind of post-processing suitable for the input model selection condition or required for the selected modeling method is automatically selected.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing the overall configuration of a modeling support device.

The modeling support apparatus can be realized by a hardware architecture or a programmed computer system. Further, part of the modeling support device can be realized by hardware and the other part can be realized by software.

The model condition input device 11 is used by a user to input selection conditions of a system for modeling, and is realized by a keyboard or the like.

The input selection conditions include the type of application model, presence / absence of process output, model size (input / output number), quantity and quality of sample data, required processing speed, and the like.

The application model indicates the type of model of the actual system that the user is going to model, and specifically, the evaluation model, the diagnostic model, the prediction model,
There are control models, recognition models, and so on.

Whether or not the process output is present is not limited to the final output of the model when the calculation is performed using the created model.
Indicates whether or not to request the output of various values in the calculation process of the model.

The numbers of inputs and outputs represent the number of types of input variables input to the model and the number of types of output variables obtained from the model.

The quantity and quality of sample data relates to the degree and quantity of sample data provided for modeling purposes.

The processing speed is the calculation speed (time required for one calculation) required for the created model.

An example of selection conditions input from the model condition input device 11 is shown in FIG. The "processing speed" is given by fuzzy information. In order to deal with such fuzzy information input, a membership function is preset in the model fitness calculation device 12. An example of a membership function for "fast" information about processing speed is shown in Fig. 2 (B). Data quality
It is given as a number between 0.0 and 0.1.

Selection conditions do not have to be entered for all conditions. Alternatively, a plurality of selection conditions may be set in advance and one set may be selected from the selection condition sets. Regarding the number of inputs and outputs and the amount of data, actual sample data may be used as the selection condition. Further, as the selection condition, it is possible to input whether the know-how of the expert can be input, the correlation between the input and the output, or the like.

The input selection conditions are given from the model condition input device 11 to the model conformity calculation device 12, the pre-processing device 15 and the post-processing device 17.

The model conformance calculation device 12
With reference to the parts storage device 13, the selection condition given from the model condition input device 11 and the modeling parts storage device
This is to calculate the model compatibility with the modeling method registered in advance in 13.

In the modeling parts storage device 13, program modules for realizing a plurality of modeling methods and usage conditions for each modeling method are registered in advance. The modeling parts storage device 13 is realized by a memory such as a hard disk or a ROM.

Examples of modeling techniques include the following. 1) Modeling method using fuzzy inference method 3 2) Modeling method using fuzzy measure 3) Modeling method using fuzzy linear programming 4) Modeling method using neural network 5) Automatic synthesis of membership functions and rules Modeling method using functions 6) Learning method using sequential membership function tuning 7) Learning method using member data to tune membership function

There are several types of fuzzy inference methods, and for the sake of convenience, fuzzy inference methods 1, 2, 3,
They are named 4, ... etc. The fuzzy inference method 3 of 1) is one of these types.

FIG. 3A shows an example of usage conditions for the "modeling method using the fuzzy inference method 3" and the "modeling method using the neural network".
And FIG. 4 (A) respectively.

The usage conditions of each modeling method include an applied model, presence / absence of process output, number of inputs, number of outputs, and processing speed.

The usage condition of the applied model indicates how suitable the modeling method is for the evaluation model, the diagnostic model, the prediction model, the control model and the recognition model, and the degree thereof. This degree is 0.0
Expressed as a number between ~ 1.0. For example, the application model usage conditions of the "modeling method using the fuzzy inference method 3" are 0.2, 0 for the evaluation model, the diagnostic model, the prediction model, the control model and the recognition model, respectively.
3, 0.7, 0.8, and 0.0.

The presence / absence of process output is a usage condition of whether or not a calculation process can be output in each modeling method. The usage conditions include “output possible”, “output conditionally”, “cannot output”, etc. There is. For example,
In the “modeling method using the fuzzy inference method 3,” the process output is “can be output”.

The number of inputs is a condition for the number of types of input variables that can be input when the system is modeled by each modeling method. This input number is generally represented by a positive integer, but it may be represented by a fuzzy number. For example, in the "modeling method using the fuzzy inference method 3," the usage condition of the number of inputs is set to "about 10 or less", and in the "modeling method using the neural network", "about 2 to about 20" It is set.

The number of outputs is a condition for the number of kinds of output variables that can be output when the system is modeled by each modeling method. This output number is also generally represented by a positive integer, but it may be represented by a fuzzy number. For example, in the "modeling method using the fuzzy inference method 3," the usage condition of the number of outputs is set to "about 5 or less", and in the "modeling method using the neural network", it is set to "about 5 or less". ing.

As described above, when the usage condition is expressed by fuzzy information or a fuzzy number, a membership function representing the fuzzy information or the fuzzy number is registered in association with it. 3 (B) and 3 (c), the membership function representing the usage condition "about 10 or less" for the number of inputs of the "modeling method using the fuzzy inference method 3" and the usage condition "about 5" for the number of outputs. Less than"
Membership functions are respectively represented.
Similarly, in FIGS. 4 (B) and 4 (C), the usage condition “about 2 to about 20” for the number of inputs and the usage condition “about 5 or less” for the “modeling method using a neural network” are shown. Membership functions are shown to represent

The usage condition for the processing speed represents the calculation speed when each modeling method is used. The usage condition of this processing speed is generally expressed by an actual speed. For example, the processing speed of the “modeling method using the fuzzy inference method 3” (FIG. 3 (A)) is “1 msec”.
The processing speed of the "modeling method using a neural network" (Fig. 4 (A)) is expressed using the mathematical expression "10 µsec x (number of inputs x number of outputs) ^2. "
In addition to this, the usage condition of the processing speed can be expressed and registered by using fuzzy information (language information), fuzzy number, fuzzy label and the like.

The model goodness-of-fit calculating device 12 performs matching processing between the selection condition input from the model condition input device 11 and the use condition of each modeling method stored in the modeling parts storage device 13 (use condition). ) For each item and calculate the fitness for each item. The model suitability of each modeling method is calculated by calculating the average value of the calculated suitability for each item.

When the selection condition shown in FIG. 2 (A) is input, the matching process result of the input model condition with respect to the use condition (FIG. 3 (A)) of the "modeling method using the fuzzy inference method 3" is displayed. 5 is shown.

Regarding the item of the applied model, the usage condition “0.7” registered for the input selection condition “prediction model” is selected and adopted as the goodness of fit. Regarding the presence / absence of process output, the compatibility is 1.0 because the input selection condition “output” and the registered usage condition “output possible” match. If they do not match, the goodness of fit is
It becomes 0.0. As for the number of inputs, the degree of conformity of the input selection condition “number of inputs: 12 inputs” to the membership function for the number of inputs set in connection with the usage conditions (FIG. 3 (B)) is obtained. In this example, the goodness of fit is 0.6. Similarly, regarding the number of outputs, the fitness of the input selection condition “number of outputs: 1 output” with respect to the membership function shown in FIG. 3 (C) is calculated as 1.0. It is not necessary to calculate the amount of data and the quality of the data, because the usage conditions are not set. Regarding the processing speed, the usage condition "1 m" is used for the membership function (Fig. 2 (B)) that represents the input processing speed selection condition "fast".
sec ”conformity is required. In this example, the goodness of fit is 0.7
Becomes

As for the model goodness of fit, the average value of the goodness of fit of each item obtained as described above is (0.7 +1.0 +0.5 +1.
It is calculated as 0 +0.7) / 5. Rounding off the second decimal place, the model goodness of fit is 0.8.

The model conformity calculation device 12
The model conformance is calculated for all the modeling techniques registered in the parts storage device 13, and the modeling technique having the maximum model conformance is selected.

This modeling method is selected by presenting a modeling method having a model matching degree of a predetermined threshold value or more and the model matching degree to the user (for example, displaying it on a display device), and based on the presented model matching degree. The user may select (specify and input the selected modeling method).

The name of the selected modeling method (eg, "modeling method using fuzzy inference method 3") and its use conditions are given to the preprocessing selection device 15, postprocessing selection device 17 and model output device 19. . A program module that implements the selected modeling method is provided to the model output device 19.

The preprocessing selection device 15 is a model condition input device.
Required for modeling the system from among a plurality of types of pre-processing preset in the pre-processing storage device 16 based on the selection condition input from 11 and / or the modeling method selected by the model selection device 14. Select the appropriate pretreatment.

The types of preprocessing set in the preprocessing storage device 16 include noise identification processing using normal distribution, data processing processing, and statistical processing. In the preprocessing storage device 16, program modules for executing these preprocessings are registered. Known techniques can be used for these processes.

First, a method for selecting the necessary preprocessing based on the input selection conditions will be described.

Preselection rules for preprocessing and membership functions related thereto are registered in the preprocessing storage device 16 in advance. The preprocessing storage device 16 is realized by a memory such as a hard disk or a ROM.

The pre-processing selection rules (know-how) include the following rules. Rule 1 IF data volume / number of inputs = large TheN data selection processing is required Rule 2 IF data volume = large AND Data quality = bad THEN Noise identification processing is required Rule 3 IF inputs = large number THEN Effective feature quantities are extracted It is necessary to limit the number of inputs

Regarding the variables of the antecedent part, FIGS. 6 (A) and 6 (B)
Membership functions are provided as shown in and (C). Figures 6 (A), (B) and (C) show
Three membership functions are shown for "data volume / number of inputs", "data volume" and "data quality".

The preprocessing selection device 15 performs matching between the input selection condition and the antecedent part of the preprocessing selection rule in accordance with the preprocessing selection rule registered in advance,
Find the suitability of the antecedent part. When there are multiple variables in the antecedent part, the average of the goodness of fit of them is calculated, and this is the goodness of fit of the antecedent part. The pre-processing set in the consequent part of the rule in which the suitability of the antecedent part is equal to or greater than the predetermined threshold is selected as the required pre-process.

Looking at the selection condition shown in FIG. 2 (A), "data amount / input number" is 5000/12 = 417, and it is applied to the membership function of FIG. 6 (A). The suitability of the antecedent part of Rule 1 described above is "1.0". Also, the input data amount is 5000, and by applying this to the membership function of Fig. 6 (B), the quality of the input data is
Applying the membership function of Fig. 6 (C) to 0.3, the goodness of fit of "data volume" and "data quality" in rule 2 are "1.0" and "0.4", respectively.
, And taking the average of these two goodnesses of fit,
(1.0 + 0.4) / 2 = 0.7. This is the suitability of the antecedent part of rule 2.

Here, if the threshold value is set to 0.7, the antecedent conformity of rule 1 and rule 2 is equal to or greater than the threshold value. Therefore, "data selection processing" and "noise identification processing" are performed.
Is selected.

Similar to the selection of the modeling method, the usage condition is set for each kind of preprocessing, and the kind of preprocessing required by matching the input selection condition with this usage condition. Can be determined.

Since the modeling method selected by the model selection device 14 may have an indispensable pre-processing, the pre-processing selection device determines the kind of pre-processing required for the selected modeling method. You may do it.

In any case, each display showing the type of pre-processing selected or determined and the program for executing it.
The modules are from the pre-processing selection device 15 to the model output device 19
Given to.

The post-processing selection device 17 is a model condition input device.
Based on at least one of the selection condition input from 11 and the modeling method selected by the model selection device 14, a system is modeled from a plurality of types of pre-processing preset in the post-processing storage device 18. Select the required post-treatment.

The types of post-processing registered in the post-processing storage device 18 include a process for evaluating the effectiveness of knowledge, a process for detecting a deviation between the inference result and the actual measurement data, and a process for translating into knowledge that is easy for humans to understand. There is. The post-processing storage device 18 stores program modules that execute these post-processings. Known techniques can be used for these processes.

In the post-processing storage device 18, post-processing selection rules and membership functions related thereto are registered in advance. The post-processing storage device 18 is realized by a memory such as a hard disk or a ROM.

The post-processing selection rules (know-how) include the following rules. Rule 1 IF Data volume / number of inputs = small THEN Post-processing required to evaluate effectiveness of knowledge Rule 2 IF modeling method = learning function THEN Post-processing required to detect deviation between inference result and measured data Rule 3 IF modeling method ＝ Neural network AND Presence / absence of process output ＝ Output THEN Post-processing is required to translate into human-friendly knowledge.

The post-processing selection device 17 performs matching between the input selection condition and the selected modeling method and the antecedent part of the post-processing selection rule according to the above-mentioned post-processing selection rule registered in advance. Find the suitability of the subject section. When there are multiple variables in the antecedent part, the average value of the goodness of fit is taken as the antecedent part adaptability. The post-processing described in the consequent part of the rule having the suitability of the antecedent part equal to or larger than the predetermined threshold is selected.

For example, when the selection condition shown in FIG. 2 (A) is applied to rule 1, the matching degree of "data amount / number of inputs" is "0.0", so if the threshold value is 0.7, "knowledge Post-processing to evaluate effectiveness will not be selected.

The name indicating the type of post-processing selected and its execution program module are as follows:
To the model output device 19.

If the modeling method selected by the model selection device 14 is the "modeling method using a neural network" and "the presence / absence of process output = output" is set in the input selection condition, the rule 3 is hit, and "post-processing for translating into human-friendly knowledge" is selected as necessary.

The model output device 19 displays the name of the modeling method given from the model selection device 14, the kind of pre-processing given from the pre-processing device 15, and the kind of post-processing given from the post-processing device 17. Or output by printing.
When multiple modeling methods are selected, it is preferable to output all of them together with their goodness of fit. Regarding the pre-processing and post-processing, it is preferable to clearly indicate whether the processing is necessary for the input selection condition or the processing required for the selected modeling method.

Simultaneously with or after these outputs, or in response to a module output command, implements program modules, selected or designated pre-processing and post-processing that implement a selected or designated modeling approach. It also outputs the program module. These program modules may be transmitted to a computer system or stored on a recording medium such as a floppy disk.

If the interfaces of the program modules are unified, a model can be created and evaluated according to the selection conditions input by the modeling support device.

This modeling support device is required to be able to select at least the type of modeling method that meets the input selection conditions, and to select the type of pre-processing or post-processing as necessary, and preferably the selected modeling method. , Outputs the program module that executes pre-processing and post-processing.

FIG. 7 is a flow chart showing the processing procedure of the modeling support device. This is especially true if the functionality of the modeling aid is implemented in software.

From the model condition input device 11, the applied model,
Selection conditions regarding process output, number of inputs, number of outputs, amount of data, data quality and processing speed are input (step 2
1).

The model conformity between the input selection condition and the usage condition of each modeling method registered in advance in the modeling parts storage device 13 is calculated (step 22).

The modeling method with the highest model goodness of fit is selected (step 23). Multiple modeling methods are selected and displayed as needed.

It is determined whether preprocessing is necessary based on at least one of the input selection condition and the selected modeling method (step 24). If it is determined that the preprocessing is necessary, the required preprocessing type is selected from the plurality of types of preprocessing registered in the preprocessing storage device 16 (step 25).

Based on at least one of the input selection condition and the selected modeling method, it is determined whether post-processing is necessary (step 26). When it is determined that the post-processing is necessary, the necessary type of post-processing is selected from the plurality of types of post-processing registered in the post-processing storage device 18 (step 27).

The selected modeling method, pre-processing and post-processing are output (step 28).

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of a modeling support device.

FIG. 2A shows an example of a selection condition, and FIG. 2B shows an example of a membership function relating to the processing speed in the selection condition.

[FIG. 3] (A) shows an example of usage conditions, and (B) and (C)
Shows membership functions related to the number of inputs and the number of outputs of the usage conditions, respectively.

[FIG. 4] (A) shows an example of usage conditions, and (B) and (C)
Shows membership functions related to the number of inputs and the number of outputs of the usage conditions, respectively.

FIG. 5 shows an example of a matching result of selection conditions and usage conditions.

6 (A), (B) and (C) are graphs respectively showing examples of membership functions relating to “data amount / input number”, “data amount” and “data quality”.

FIG. 7 is a flow chart showing a software processing procedure of the modeling support device.

[Explanation of symbols]

 11 Model condition input device 12 Model conformity calculation device 13 Modeling parts storage device 14 Model selection device 15 Pre-processing selection device 16 Pre-processing storage device 17 Post-processing selection device 18 Post-processing storage device 19 Model output device

Front Page Continuation (72) Inventor Yoichi Tamamaki, 10 Hanazono Todocho, Ukyo-ku, Kyoto City, Kyoto Prefecture Omron Co., Ltd. Within

Claims (8)

[Claims] 1. A modeling parts storage means in which use conditions related to the modeling method are stored in advance for each of a plurality of modeling methods, a model condition input means for inputting model selection conditions related to a model to be modeled, and the input condition described above. Model suitability calculating means for calculating the model suitability for each modeling method based on the model selection condition and the use condition for each modeling method stored in the modeling part storage means, and the calculated model A modeling support device comprising model selection means for selecting a modeling method suitable for the inputted model selection condition based on the goodness of fit. 2. The modeling support device according to claim 1, wherein the modeling part storage means stores a program module for executing each of a plurality of modeling methods. 3. The modeling support device according to claim 1, further comprising output means for outputting the modeling method selected by the selection means. 4. The preprocessing based on at least one of preprocessing storage means in which data on a plurality of types of preprocessing is stored in advance, and at least one of the input model selection condition and the selected modeling method. The modeling support apparatus according to claim 1, further comprising preprocessing selection means for selecting a suitable one from a plurality of types of preprocessing stored in advance in the storage means. 5. The post-processing based on at least one of a post-processing storage means in which data regarding a plurality of types of post-processing is stored in advance, and at least one of the input model selection condition and the selected modeling method. The modeling support device according to claim 1, further comprising post-processing selection means for selecting a suitable one from a plurality of types of post-processing stored in advance in the storage means. 6. The use condition related to the modeling method for each of a plurality of modeling methods is stored in advance in the modeling parts storage means, the input model selection condition related to the model to be modeled is accepted, and the input model is input. A model goodness of fit for each modeling method is calculated based on the selection condition and a use condition for each modeling method stored in the modeling parts storage means, and input based on the calculated model goodness of fit. Modeling support method for selecting a modeling method suitable for the selected model selection condition described above. 7. Pre-processing data is stored in advance in a pre-processing storage means, and the pre-processing is performed based on at least one of the input model selection condition and the selected modeling method. 7. The modeling support method according to claim 6, wherein a suitable one is selected from a plurality of types of preprocessing stored in advance in the storage means. 8. The post-processing is pre-stored in a post-processing storage means, and the post-processing is performed based on at least one of the input model selection condition and the selected modeling method. The modeling support method according to claim 6, wherein a suitable one is selected from a plurality of types of post-processing stored in advance in the storage means.