The goal of this research is to create a new machine
learning (ML)-centric
methodology for process modeling, designing, and evaluating hydrogen
production based on the water–gas shift reaction (WGSR). The
approach evaluates the one-pass conversion of catalyst and process
overall conversion, as well as the economics of the catalytic conversion
process, without the use of kinetics and a process model that requires
much trial and error. To accomplish this, an ML model was developed
to predict the catalyst performance based on critical catalysis descriptors
like catalyst composition, operating conditions, and feed composition.
We developed a surrogate model for the hydrogen production process
based on the predicted results to determine the mass and energy information
on the process, which includes multiple unit operations and recycling.
Finally, we assessed the hydrogen production process using different
technical and economic metrics such as hydrogen amount, energy consumption,
and unit energy cost. The approach can perform a kinetics-free simulation
of hydrogen production processes using predicted catalyst performances
and evaluate early-state catalysts from an industrial perspective
by identifying the optimal operating conditions and the catalyst structure
for economic and energy-efficient hydrogen production. As a result,
the processes over Pt/Co(10 wt %)/Al2O3, Pt/Co(20
wt %)/Al2O3, and Pt/Ce(5 wt %)/TiO2 show the best performance to produce high-purity hydrogen.