TY - JOUR
T1 - Modeling the n-Hexane Isomerization over Iron Promoted Pt/WOx-ZrO2 Catalysts Using Artificial Neural Networks
AU - Hernández-Pichardo, Martha Leticia
AU - Macías-Salinas, Ricardo
N1 - Publisher Copyright:
© 2016 American Chemical Society.
PY - 2016/8/17
Y1 - 2016/8/17
N2 - An artificial neural network approach was presently used to model the hydroisomerization reaction of n-hexane over platinum supported on tungstated zirconia (Pt/WOx-ZrO2) catalysts doped with different amounts of iron. Four different multilayer feed-forward neural network arrangements were then devised and trained using previously reported experimental catalyst activity in terms of selectivity (S) and yield (Y) of 2,2-dimethylbutane measured at several Fe/W weight ratios and surfactant/zirconia molar ratios (SMR). The performance of the four neural networks during the training process was acceptable despite the fact that a limited database was used for such a purpose; the catalyst synthesis variables chosen as neural network inputs (SMR, %W, and %Fe) played a very important role in successfully correlating the catalyst activity (in terms of S and Y of 2,2-dimethylbutane) in all cases. The predictive capabilities of the trained neural networks were further verified by computing some selectivities and yields of 2,2-dimethylbutane not considered in the training process. The agreement between predicted and observed catalyst activity data was highly acceptable thus demonstrating the abilities of the four neural networks (particularly, the 3-3-2-2 arrangement) in predicting suitable values of catalyst activity within the present "gray-box" modeling work.
AB - An artificial neural network approach was presently used to model the hydroisomerization reaction of n-hexane over platinum supported on tungstated zirconia (Pt/WOx-ZrO2) catalysts doped with different amounts of iron. Four different multilayer feed-forward neural network arrangements were then devised and trained using previously reported experimental catalyst activity in terms of selectivity (S) and yield (Y) of 2,2-dimethylbutane measured at several Fe/W weight ratios and surfactant/zirconia molar ratios (SMR). The performance of the four neural networks during the training process was acceptable despite the fact that a limited database was used for such a purpose; the catalyst synthesis variables chosen as neural network inputs (SMR, %W, and %Fe) played a very important role in successfully correlating the catalyst activity (in terms of S and Y of 2,2-dimethylbutane) in all cases. The predictive capabilities of the trained neural networks were further verified by computing some selectivities and yields of 2,2-dimethylbutane not considered in the training process. The agreement between predicted and observed catalyst activity data was highly acceptable thus demonstrating the abilities of the four neural networks (particularly, the 3-3-2-2 arrangement) in predicting suitable values of catalyst activity within the present "gray-box" modeling work.
UR - http://www.scopus.com/inward/record.url?scp=84983451323&partnerID=8YFLogxK
U2 - 10.1021/acs.iecr.6b01821
DO - 10.1021/acs.iecr.6b01821
M3 - Artículo
SN - 0888-5885
VL - 55
SP - 8883
EP - 8889
JO - Industrial and Engineering Chemistry Research
JF - Industrial and Engineering Chemistry Research
IS - 32
ER -