Artificial neural networks application in combined carbon dioxide reforming and partial oxidation of methane

Abstract

A catalytic carbon dioxide reforming of methane with the effect of oxygen was carried out on 1 wt% of Rhodium (Rh) on Magnesium Oxide (MgO) and ZSM-5 catalysts. The effect of parameters on the methane conversion, synthesis gas selectivity and H2/CO ratio were studied. Three main parameters: temperature, O2/CH4 ratio and catalyst weight in 100 ml/min total feed flow rate, have been identified as the major factors that control the process. The results indicated that Rh/MgO showed better catalyst reactivity and stability even though at temperature higher than 800 C. Thus, the optimization of the combined CORM and partial oxidation of methane over Rh/MgO catalyst was carried out. The optimization study was performed with the help of experimental design and two mathematical approaches: empirical polynomial and artificial neural network. Empirical polynomial models were employed to analyze the effect of parameters on the response factor and the correlation coefficient, r, was above 85%. However, the feed forward neural network correlation coefficient was more than 95%. The feed forward neural network modeling approach was found to be more efficient than the empirical model approach. The optimum condition for maximum methane conversion was obtained at 850 C with O2/CH4 ratio of 0.14 and 141 mg of catalyst resulting in 95% methane conversion. A maximum of 40% hydrogen selectivity was achieved at 909 C, 0.23 of O2/CH4 ratio and 309 mg catalyst. The maximum H2/CO ratio of 1.6 was attained at 758 C, 0.19 of O2/CH4 and 360 mg catalyst. The utilization of neural network in predicting the reaction for other catalyst was also tested by introducing other reaction data in the network. The result showed a feed forward neural network was able to predict the output of the reaction even for different reaction or catalysts.