JP2001075611A - Automatic setup model construction method and apparatus - Google Patents

Abstract

(57) [Summary] [Problem] When constructing a model, a series of operations from data collection to processing, evaluation, analysis, and modeling are automatically performed. SOLUTION: A necessary number of required result data is automatically collected from a huge amount of process data, abnormal data is eliminated from the collected result data, stable data necessary for analysis is automatically extracted, and the extracted stable data is extracted. Based on the evaluation results, the model features are automatically extracted and the model accuracy is automatically evaluated.Based on the evaluation results, error analysis of many influencing factors related to the model accuracy is performed from the collected actual data, and the model structure and parameters are automatically decide.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for automatically constructing a setup model, collecting necessary data from various process data, and performing theoretical or statistical analysis using the collected data. The present invention relates to a method and an apparatus for automatically constructing a setup model for constructing a setup model used for presetting an apparatus.

[0002]

2. Description of the Related Art Conventionally, in general, the work of constructing a setup model is performed based on the experience of a designer and technical know-how.
It is performed by theoretical or statistical analysis. Specifically, for example, the model calculation value is compared with the actual value, and if the accuracy is poor, a correlation analysis of the control variable to be modeled and a factor considered to affect the model is performed, and the result is obtained. The model is constructed or modified based on this, and the accuracy is evaluated by comparing the model calculated value with the actual value again. Therefore, the addition of new varieties and changes in operating conditions require human intervention, which requires a large amount of data analysis and trial and error, and requires a great deal of human power each time.

As a function to support this model construction work, Japanese Patent Publication No. 6-61566 regression-analyzes a model error with respect to a setup model of a rolling process, and makes a main variable of a control variable a factor. There has been proposed a method of automatically adjusting a weighting factor of a correction function represented by a weighting factor indicating a degree of influence of an error on an error.

[0004]

SUMMARY OF THE INVENTION
The automatic model adjustment method according to 66 is effective when the structure of the model (independent variables and function forms) is clear, but when the optimum structure is not known, sufficient improvement in model accuracy cannot be expected. It cannot be applied to other processes. In general, model construction work also has problems such as the need for experience and know-how even in the data processing stage until data analysis.

[0005] The present invention has been made to solve the above-mentioned conventional problems, and it is an object of the present invention to automatically perform a series of operations from data collection to processing, evaluation, analysis, and modeling when constructing a model. And

[0006]

SUMMARY OF THE INVENTION The present invention provides a setup for collecting necessary data from various process data, and using the collected data for theoretical or statistical analysis for presetting to each control device. When building a model, the required number of actual data required was automatically collected from a huge amount of process data, abnormal data was eliminated from the collected actual data, and the stable data required for analysis was automatically extracted and extracted. Based on the stable data, the characteristics of the model are automatically extracted, the model accuracy is automatically evaluated, and based on the evaluation result,
This problem has been solved by performing error analysis of a number of influencing factors related to model accuracy from collected performance data and automatically determining the structure and parameters of the model.

The present invention also provides a means for automatically collecting necessary data from a huge amount of process data in an automatic setup model construction apparatus, and eliminating abnormal data from collected actual data to perform analysis. Means for automatically extracting stable data, means for automatically extracting the characteristics of the model based on the extracted stable data, and means for automatically evaluating model accuracy. The above problem is also solved by providing means for performing error analysis of a number of influential factors and automatically determining the structure and parameters of the model.

In the present invention, a series of tasks of data collection, model accuracy evaluation, model error evaluation, and modeling conventionally performed by humans by trial and error are automated to construct a high-accuracy model and model construction. Enables quick and efficient work.

[0009] Specifically, first, process data is collected. When collecting this performance data, it is necessary to determine the collection period and collection timing, but in the present invention,
It has an automatic data collection function that collects the required number of required results data from a huge amount of process data by an appropriate method.

Next, the collected data is processed. That is, the data is selected (filtered), the data is structured (meshed), etc. from the collected data. The present invention excludes abnormal data from the collected raw data and uses it for analysis. It has an automatic data stability determination function for automatically extracting stable data that can be obtained.

Third, model accuracy is evaluated based on the processed data. Here, the evaluation items are determined and the characteristics of the model are grasped. The present invention finds regularity of hidden data from collected data,
For example, using a technique such as data mining, a function is provided for automatically analyzing a cause of model accuracy deterioration caused by a change in operating conditions, a change in manufacturing conditions, and the like.

Fourth, based on the model accuracy evaluation, model error analysis and modeling based on the model error analysis are performed. Analyze the influence of each influential factor on a large number of models based on theory and experience, select multiple or singular factors that have a large effect as independent variables, and use a nonlinear function approximation method that can estimate the structure by, for example, genetic programming. And has a function of automatically determining a model.

According to the present invention described above, a series of model building operations that require a great deal of manpower can be automated based on the experience and know-how of a designer, so that anyone can efficiently build a model. Will be able to

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings.

In this embodiment, as shown in FIG. 1, a data collection device 10 having a function of automatically collecting a required number of required result data from a huge amount of process data, and eliminating abnormal data from the collected result data. Data stability determination with the function of automatically extracting stable data required for analysis.
Based on the processing device 12, a model accuracy evaluation device 14 having a function of automatically extracting model features based on the extracted stable data and automatically evaluating model accuracy,
A model determining device 16 having a function of performing error analysis of a number of influencing factors related to model accuracy from the collected performance data and automatically determining the structure and parameters of the model.

Hereinafter, the operation of the present embodiment will be described using an example applied to the construction of a rolling load model in a cold rolling process.

First, a specific rolling cycle and a large number of rolling process data are extracted by the data collecting device 10 from the huge rolling process data.
Automatic collection of rolling performance data, such as rolling load, rolling speed, plate thickness, tension, etc., is performed during a collection period and timing designated in advance, such as low speed / high speed rolling data.

Since the collected performance data also includes abnormal data that is inappropriate for use in the analysis, the data stability determination / processing device 12 next filters the abnormal data by filtering the data. Elimination takes place.
Further, in the analysis, data structuring (mesh classification) is performed according to steel type, size, and the like.

The data processed by the data stability determination / processing device 12 is processed by a model accuracy evaluation device 14.
The accuracy evaluation (load prediction accuracy, advanced rate prediction accuracy, etc.) of the rolling model set in advance in advance is performed, the characteristics (bad in a certain steel type, speed range, etc.) are grasped, and the factor of the model accuracy deterioration is analyzed.

In response to the evaluation result, the model determination unit 16 performs (1) model error analysis and (2) modeling (correction) of the rolling model.

As an example, a description will be given of an example in which a model of a friction coefficient μ, which is one parameter of the rolling load model, is constructed in order to improve the accuracy of the rolling load prediction model.

First, the friction coefficient is calculated back from the rolling load,
Analysis of a number of influential factors, such as the rolling speed V, the rolling distance L, and the rolling reduction γ, is performed. Of these, the factor with the greatest effect is selected as a variable, and the friction coefficient is modeled. At the time of modeling, for example, by performing modeling using a nonlinear function approximation method capable of estimating a structure by genetic programming or the like, a model with higher accuracy can be obtained as compared with a conventional method such as regression analysis. In the example of FIG.
On the other hand, since the rolling speed V has a negative correlation and the rolling distance L has a positive correlation, the rolling reduction γ has no correlation, and thus is excluded during modeling.

When the coefficient of friction is modeled, the predicted rolling load is calculated again using the model, and the model accuracy evaluation device 1 is used.
By (4), model accuracy evaluation is performed again. If the accuracy is within the target range, the model building is finished and applied online. On the other hand, if it is outside the target range, the model error analysis is performed again by the model determination device 16, and the selected variables are changed by, for example, a method of increasing the variables in descending order of correlation, and the modeling is performed again. .

In the above description, the construction of the rolling load model of the cold rolling process has been described as an example. However, it is obvious that the present invention can be applied to the construction of another model or a model in another process. .

[0025]

According to the present invention, when constructing a setup model used for presetting to each control device, data collection, model accuracy evaluation, model error analysis, model A series of operations for automation are automated, and at the time of modeling, it is automatically determined from the model structure aiming at high-precision automatic model refining, so a high-precision model can be constructed efficiently and by anyone. And can be adjusted.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an automatic setup model construction apparatus according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Data collection device 12 ... Data stability determination / processing device 14 ... Model accuracy evaluation device 16 ... Model determination device

Claims (2)

[Claims] When collecting necessary data from various process data and constructing a setup model used for presetting to each control device by theoretical or statistical analysis using the collected data, an enormous amount of data is generated. Automatically collect the required number of required performance data out of the necessary process data, eliminate abnormal data from the collected performance data, automatically extract the stable data required for analysis, and model features based on the extracted stable data. And automatically evaluates the model accuracy, and based on the evaluation result, performs error analysis of many influencing factors related to the model accuracy from the collected actual data, and automatically determines the structure and parameters of the model. How to automatically build a setup model. 2. A setup for collecting necessary data from various process data and constructing a setup model to be used for presetting to each control device by theoretical or statistical analysis using the collected data. A means for automatically collecting the required number of actual data from a huge amount of process data in an automatic model construction device, and a means for automatically extracting stable data required for analysis by eliminating abnormal data from the collected actual data Means for automatically extracting model features based on the extracted stable data and automatically evaluating model accuracy. Based on the evaluation results, error analysis of a number of influential factors relating to model accuracy from collected actual data is performed. Means for automatically determining the structure and parameters of the model. Apparatus.