A Practical Application of Simulation-based Surrogate Modeling for Prereformer Reactor

In this work, a practical example of surrogate modeling in process engineering is demonstrated in the field of hydrogen production: Modeling of a prereformer reactor. The main motivation is to show the potential of this approach for performance increase of simulations in process design and operation optimization. In a first step, sample points were generated following a computational design of experiments procedure. A prereformer reactor model was built in Aspen Plus (R), and corresponding simulation results were collected based on the sample points. The resulting dataset, i.e. sample points and simulation outputs, were used for surrogate model development for the reactor outlet temperature and concentrations, via built-in artificial neural networks (ANN) feature of JMP (R). It was observed that resulting surrogate models, with a single layer ANN of 100 hidden nodes, satisfied the expected accuracy limits, e.g. the outlet temperature was predicted with an RMSE of less than 0.04 degrees C. The predictive behavior of the models was also examined via a blind test set, and the speedup in computation was shown. While this study constitutes a successful proof-of-concept, further improvements are possible via e.g. customization of model development, employment of adaptive sampling methodology.