Rolling model learning method

Abstract

The present invention addresses the problem of improving prediction accuracy while suppressing the time required to stabilize learning coefficients of a learning table from becoming huge when predicting a target variable of a phenomenon in a rolling step by referring to the learning table. A rolling model learning method of the present invention calculates a plurality of main components by decomposing the main components of a past data set having a correlation variable relating to a target variable. A first main component among the main components is added to the explanatory variable, and a learning table is extended using a weight coefficient of the first main component. A main component of a data set of a correlation variable when a predicted value of a target variable is calculated is calculated, and a prediction learning coefficient is extracted from a learning table. A predicted value of the target variable is calculated using a rolling model determined by the learning coefficient during prediction, and an error between actual values of the target variable corresponding to the predicted value is calculated. The prediction-time learning coefficient in the learning table is updated using an update learning coefficient calculated on the basis of the error.

Description

technical field [0001] The present invention relates to a method of learning a rolling model used in a rolling process. Background technique [0002] In the rolling process, various controls are performed in order to roll a rolled material having desired dimensions, temperature, mechanical properties, and the like. As various controls, setting control and dynamic control are exemplified. In the setting control, a rolling model is used to predict phenomena such as deformation and temperature of the rolling material in the rolling process. In the setting control, based on the prediction result, the setting values ​​of the control variables of the control objects such as the roll gap of the rolling mill, the rolling speed, and the amount of cooling water are calculated. [0003] Rolling models are typically constructed based on theoretical and experimental methods. In order to realize the manufacture of high-quality products and stable rolling, the prediction accuracy of the...

AI Technical Summary

Problems solved by technology

However, there is a problem that the processing for constructing the predictive formula is easily complicated and the calculation load is large

Images