Prediction Model of Wear Amount of Work Roll and Replacement Moment in Finishing Rolling Based on Lévy's Improved Arithmetic Optimization Algorithm Twin Support Vector Regression

Abstract

For the control of the wear amount of work rolls and replacement moment in finishing rolling, most of the traditional models are unable to accurately predict the optimal finishing wear amount and replacement moment of work roll in advance, which may lead to the disruption of the production rhythm, and even cause product quality defects. This research describes a Lévy's improved arithmetic optimization algorithm twin support vector regression (LAOA-TSVR) prediction model for wear amount of work roll and replacement moment in a finishing mill. Firstly, the research group initially employed real production data from a hot strip finishing mill to identify influential factors of wear amount of work roll through correlation analysis using SPSS. Subsequently, to validate its predictive performance, the model was compared against three classical algorithms: Back Propagation (BP), Radial Basis Function (RBF), and Support Vector Machine (SVM), confirming LAOA-TSVR's superior accuracy. Finally, the model underwent practical production testing with a dataset totaling 200 sets. The findings reveal that the model attains a 95.2 pct hit rate for predicting wear amount of work roll within ± 0.5 pct. Likewise, it achieves a 98.3 pct hit rate for predicting the replacement moment of work roll for finishing mill.