Structure based model reduction in real-time optimization

Abstract

Process model with high fidelity plays an extremely important role in the design and operation for chemical plant. Yet the growing complexity along with fidelity increases the computational time, causing decision delay in RTO (real-time optimization). This study demonstrates a systematical method of model reduction in the application of RTO. The reduced model could get the optimization results much faster without breaking the model structure and keep the model accuracy under an acceptable error tolerance. In this method, a chemical model is first classified as layers according to its structure and the inner layer which is kinetics and thermodynamics layer is mainly chosen for reduction since it takes the most CPU time during simulation and optimization. In the reduction of kinetics, the DRG method is used, and in the reduction of thermodynamics, a Kriging method is used to map the input and output. A numerical case study on an industrial P-Xylene oxidation process problem is applied to validate the method.