Dynamic Modeling of Chemical Reaction Systems with Neural Networks and Hybrid Models

Abstract

Data-driven methods, such as neural networks, offer an interesting alternative to conventional physical modeling of chemical kinetics. However, there are disadvantages, primarily high data requirements for model adaptation. A suitable combination of physical and data-driven methods leads to hybrid models that utilize the advantages of both approaches while avoiding their disadvantages. This paper explains the procedure of hybrid modeling of integral data by using examples from chemical kinetics and describes the benefits of hybrid models in comparison to the limiting cases of purely physical and purely data-driven models.