Reactor design of methanol steam reforming by evolutionary computation and hydrogen production maximization by machine learning

A numerical model is developed to predict the methanol steam reforming for H2 production. This research designs an methanol steam reforming reactor and uses the Nelder-Mead algorithm to find an equivalent steam tube radius by minimizing the error between the simulation and experimental data. The effects of three operating parameters (ie, inlet temperature, S/C ratio, and Reynolds number) on CH3OH conversion and H2 yield are discussed. Finally, the predictions of CH3OH conversion and H2 yield in terms of the operating parameters through neural networks are performed for finding the best combination of the operating parameter to maximize the H2 yield. After finding the equivalent radius from the simplified reactor, the evolutionary computation improves the prediction accuracy by 42.69%. For the operating parameters, an increase in temperature or S/C ratio intensifies the reforming performance, whereas the Reynolds number of 50 is more suitable for H2 production. A three-step training and test of the database by the neural networks is adopted to evaluate the influence of the number of data sets and find the best combination of the parameters. The best combination poses the highest H2 yield of 2.905 mol (mol CH3OH)−1, and the error between the prediction and simulation is merely 0.206%.