A Comparison of LS-based Steel Thickness Prediction Methods for a Hot Rolling Mill Process

Abstract

This paper reviews the prediction methods of multiple linear regression models least squares (LS), Partial least squares (PLS), and higher order partial least squares (HOPLS) and compares the characteristics of these three methods. The methods are applied to the hot rolling mill process. Three kinds of methods are used to predict the exit thickness of finishing rolling steel plates with different thickness specifications. The mean absolute error (MAE), root mean square error (RMSE), and the percentage of the number of samples whose prediction error is within ±3% of the measured value in the total number of predicted samples are used as indices of performance to compare the thickness predicted performance. The experimental results show that HOPLS has better prediction accuracy and generalization performance compared with the other considered methods.