Development of a Hybrid Model for Industrial Ethylene Oxide Reactor

Abstract

A hybrid modeling approach for an ethylene oxidation reactor with silver catalyst is proposed. Under the form of mechanistic model, support vector regression is used to construct the catalyst deactivation model with the operating data coming from real plant. Prior knowledge is extracted to enhance the generalization of the deactivation model. With the hybrid model, the prediction error is less than 5% for the prediction of industrial reactor. The approach is shown to predict the production more accurately and have more reliable extrapolation.