Combined approach using mathematical modelling and artificial neural network for chemical industries: Steam methane reformer. (1st December 2019)

• Record Type: Journal Article
• Authors: Vo, Nguyen Dat; Oh, Dong Hoon; Hong, Suk-Hoon; Oh, Min; Lee, Chang-Ha
• Is Part Of: Applied energy. Volume 255(2019)
• Publication Date: 2019-12-01
• Subjects: Steam methane reforming -- Multiscale modelling -- Dynamic simulation -- Artificial neural network -- Stochastic simulation; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2019.113809
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A rigorous SMR model with a multiscale reactor, wall and furnace was developed. The temperature, pressure, heat flux and mole fraction agree well with a reference. A combined method using an SMR dynamic model and ANN was suggested for industries. The developed approach predicts the output with 98.91% accuracy in a few seconds. The method can be used for the design and online optimization of H2 production. Abstract: The steam methane reformer (SMR) has become more attractive owing to the increasing importance of hydrogen production using natural gas. This study developed a rigorous dynamic model for an SMR including sub-models for a multiscale reactor, wall, and furnace. The developed SMR model was validated within a small error (lower than 4%) using the reference data such as temperature, pressure, mole fraction, and average heat flux. The results predicted by changing the catalyst parameters and operation conditions confirmed the reliability of the model. Therefore, the developed model was used to generate the SMR performance data using a deterministic and stochastic simulation with four main operating variables: the inlet flow rate, temperature, S/C ratio of the reactor side, and the inlet flow rate of the furnace side. To reduce the data dimensionality, the resultant dataset was analyzed using the principle components based on a singular value decomposition method. Artificial neural network (ANN) trained through 81 datasets was applied for the feed-forward back propagation of a neural network to map the relationship between the operating variables and predicted outputs. And the ANN relation predicted the outputs (temperature, velocity, pressure, and mole fraction of components) with higher than 98.91% accuracy. Furthermore, the computational time was significantly reduced from 1200 s (dynamic simulation) to 2 s (ANN). The developed methodology can be applied not only for the online operation and optimization of a reformer with high accuracy but also for the design of a hydrogen production system at low computational cost.