Global method for learning an integrated temperature prediction model in a slab reheating furnace

Abstract

Temperature prediction in a slab heating furnace is an important problem for steel production processes. However, the problem is complicated owing to process complexity and industrial noise. These factors make it difficult to obtain precise prediction results by a mechanism model in practical production. In this article, an integrated modelling approach is proposed through combining mechanism and data-driven models. This method constructs a generalized-kernel support vector regression (SVR) on new search space to improve the predictive performance of a mechanism model, in which the kernel matrix is a combination of multiple single kernels. The learning problem can be solved globally by a semi-definite programming (SDP) problem. Numerical experiments using actual data from an iron and steel enterprise in China are performed to illustrate the effectiveness of the proposed integrated method.