Prediction of coiling temperature on run-out table of hot strip mill using data mining

Abstract

The application of data mining (DM) on hot strip rolling was introduced to improve the prediction accuracy of a model for estimating coiling temperature on a run-out table. Due to nonlinear and time-variation characteristics of coiling temperature control, conventional methods with simple mathematical models and a coarse adaptation scheme are not sufficient to obtain a good prediction of coiling temperature. A new method establishing a control model of coiling temperature is proposed based on the on-line information processing technology, which adopts DM to mine the database of laminar cooling process. A linear regression model and BP neural network are used for control of coiling temperature. Combination of regression analysis for model parameters and neural network for predicting the error of mathematical model was conducted successfully. Off-line simulation results and on-line application in hot strip mill verify the effectiveness of the proposed method. This method can improve the prediction accuracy of coiling temperature by 20%.