Comparison of ANFIS and ANN modeling for predicting the behavior of a catalytic methane reformer

Abstract

Adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN) were applied in modeling of methane mixed reforming in a packed bed catalytic reactor. These methods were developed by use of data collected from a methane reforming pilot plant using CO2 and steam and in process conditions near to MIDREX reforming plant in sponge Iron production. Different reaction temperatures from 700 to 1100 C with different values of carbon dioxide, steam, hydrogen, methane and carbon monoxide, were randomly selected and used to generate around 5000 data set of input- output data. Both networks achieve quite satisfying scientific results with acceptable deviations. However, it is hard to say which one is better as they have close output values but ANN marginally outperformed ANFIS in predicting the reaction outputs by varying the inputs. The prediction performances of these models are compared. The accuracies of the two models were evaluated in terms of square correlation coefficient (R2) and mean square error (MSE).