Soft sensor modeling based on the soft margin support vector regression machine

This paper focuses on regression applications of the Support Vector Machine (SVM) in the process industry. The support vector regression machines are employed to build soft sensing models in the paper. Soft sensor modeling, in a sense, is a kind of regression problems in industrial processes. First we review the development history of the Vapnik Chervonenkis (VC) theory and SVM. And then, the basic idea behind the SVM is introduced and some famous SVM regression algorithms are talked about. After that, the standard QP and SMO implementations to Vapnik's soft margin epsilon-SVM regression algorithm are discussed in detail. Using these two implementing methods, we perform some experiments, to predict pulp Kappa numbers, over a real-life dataset retrieved from a kraft pulp cooking process. Some useful conclusions are drawn finally.