A comparative study of hybrid models combining various kinetic and regression models for p-xylene oxidation

Abstract

The hybrid modeling approach, which combines kinetic and regression module parts to an overall process model, is an attractive process modeling approach. However, the selection of various kinetic and regression models affects the hybrid model performance. As an illustrating example, we investigated the p-xylene (PX) oxidation process and summarized the published results of the PX oxidation kinetic model. The kinetic parameters of three kinetic models (i.e., three frequently used kinetic models of PX oxidation) were estimated and the fitting results were evaluated. Six hybrid models were then developed based on these three kinetic models and two regression models (artificial neural network and support vector regression). Afterwards, a comparative study of the six hybrid models was carried out based on various kinetic and regression models. The performances of these kinetic and regression models on the hybrid models were also evaluated. Finally, the best suitable hybrid model was obtained for the PX oxidation process.