Study and Application on Dynamic Modeling Method based on SVM and Sliding Time Window Techniques

Abstract

The paper introduced a kind of dynamic modeling method based on support vector machine and sliding time window techniques. Aiming at the composition-estimated problem of the azeotropic distillation column, an appropriate industry soft sensor model was built by support vector machine based on least square (LS-SVM). The sliding time window techniques were used to update modeling database. For improving estimate precision, the industry model was corrected on-line by the error between analyzed value and estimated value and was updated automatically by the dynamic modeling database. The industry model was successfully applied to the butadiene distillation equipment to estimate the water content of the azeotropic column. The results of research show that the LS-SVM soft sensor modeling method based on the sliding window is an effect method of the soft sensor modeling method