Hybrid Modeling Approach in Chemistry

摘要

A hybrid approach to modeling of chemical processes is proposed. The hybrid model consists of an analytical model and a neural network, connected in parallel. The purpose of a neural network is to improve the prediction accuracy in the domain where enough measured data exist. A radial basis function type neural network is applied for this task due to its local approximation property. A case study "Modeling of the liquid-phase methanol synthesis" is presented where an analytical and a hybrid model are constructed for the prediction of methanol production rate. Significant improvement in the prediction accuracy is demonstrated by using the hybrid modeling approach compared to the analytical one.