(126c) Explore the Potential of Machine Learning in Building Reaction Models for Chemical Industry

Process optimization has been widely considered to maximize the benefit of chemical manufacturing plants, either from a technical and/or economic point of view. Machine learning techniques like neural network have been used to approximate equation-based first-principle reaction models [1] or numerical simulation models with excellent accuracy and efficiency. We have developed a standard framework to combine neural network based reaction model in process optimization model and demonstrated its excellent performance. However, unlike equation-based models and simulation models, machine learning models heavily rely on training data. This limitation requires machine learning models to be re-trained for different reactions and any major changes in the reactors. The goal of this research is to explore the potential of developing a generalized machine learning model that can cover a series of reactions and a large reactor geometry space. Success of this work will facilitate knowledge transfer among similar plants and similar reactions with even limited production data or experimental measurements. Potentially, the advanced machine learning model will help plant diagnosis and new plant design.

References

[1] C. A. O. Nascimento, R. Giudici, R. Guardani; Computers and Chemical Engineering 24 (2000) 2303–2314