Prediction of Head Thickness of Hot Rolled Strip Based on Tensor CP Decomposition

Abstract

In order to improve the control accuracy of strip head thickness in rolling process, a head thickness prediction model based on tensor CP decomposition is proposed in this paper. In the prediction model, the input data are composed of a tensor index matrix composed of rolling speed, rolling force, roll gap, inlet temperature and thickness parameters, and the output is the factor matrix composed of various rolling parameters that affect the thickness trend. To solve the problem of tensor sparseness in rolling data, a gradient-based CP-WOPT method is used to decompose tensor weights. The head thickness prediction and simulation of strip with rolling thickness of 3mm and 6mm were carried out, and the relative errors were 88.25% and 84.9% within ±2%, respectively. The prediction accuracy meets the actual demand, and verifies the feasibility of the prediction method.