Estimation of atmospheric 3rd line diesel oil solidifying point via Adaptive kernel based Relevance Vector Machine

Atmospheric 3rd line diesel oil solidifying point is an important quality index, which cannot be measured in real time, in petroleum industry. Due to the great nonlinear characteristic of distillation columns, common statistic methods, such as PCR and PLS, based on linear projection, are not able to estimate such a quality index effectively. In this paper, Adaptive kernel based Relevance Vector Machine (aRVM) is introduced to build a nonlinear soft sensor model. This soft sensor is then applied to a real solidifying point estimation experiment, with comparison to other nonlinear models such as KPLS, SVM and typical RVM. The result reveals that aRVM shows better performance than KPLS, SVM and models a much sparser representation than SVM and typical RVM.