Optimization of hydrogen production via catalytic autothermal reforming of crude glycerol using response surface methodology and artificial neural network

Abstract

In this work, the central composite design of the response surface methodology (RSM) and artificial neural network (ANN) were deployed to investigate process variables and optimize hydrogen gas production via autothermal reforming of crude glycerol using 5%Ni/CeZrCa catalyst in a packed bed tubular reactor. The input variables studied are reaction temperature, feed flow rate, steam to carbon ratio (S/C), oxygen to carbon ratio (O/C), and catalyst weight. The neural network analysis of these input variables measured their individual significance and contribution in the autothermal reforming process prior to the development of a quadratic polynomial model with RSM for predicting hydrogen yield as well as optimization. Statistical analyses showed that the mathematical model developed excellently represents the data with high model adequacy and no significant lack of fit. The results of the neural network input variable significance ranking approach showed that the reaction temperature contributed more than any other independent variable at a ranking weight of 53% while the O/C ratio showed the least impact at 3%. The results of the optimization gave the optimum parameters for hydrogen gas yield as follows; Temperature: 650°C, crude glycerol flow rate: 3.3 mmol C/min, S/C: 2.34, O/C: 0.052, and catalyst weight of 0.15 g with a maximum feed conversion of approximately 92% and 97% H2 yield. ANN provided a better regression with excellent statistical test values as follows R2: 0.999, MSE: 10−8, AAD: 0.34%, and RMSE: 3.5 × 10−4%. It is clear from the results of the model adequacy and error calculations that ANN is superior and has shown a better model fit than the RSM.