Izvorni znanstveni članak
Prediction of Water Activity Coefficient in TEG-Water System Using Diffusion Neural Network (DNN)
H. Karimi
; Chemical Engineering Department, Yasouj University, Yasouj 75914-353, Iran
N. Saghatoleslami
; Department of Chemical Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
M. R. Rahimi
; Chemical Engineering Department, Yasouj University, Yasouj 75914-353, Iran
Sažetak
Accurate determination of activity coefficients of water in a binary triethylene glycol (TEG)-water system, is of prime concern in designing the natural gas dehydration process. In this work, a hybrid model (a combination of information diffusion theory and neural network) and a so-called diffusion neural network (DNN) have been developed for the prediction of activity coefficients of a binary TEG-water system. Owing to the insufficient experimental data available in the literature for binary mixtures, and in particular
for infinite dilution, we have employed the information diffusion technique as a tool in extrapolating data points from the original data. By means of this technique, a new dataset has been trained and optimized for the DNN model with more nodes in the input
and the output layers. The result of this study reveals that DNN model is superior to the conventional neural nets in predicting the activity coefficient of water in the range of temperature (293–387.6 K) and mole fractions with mean absolute error of 0.31 %
(MAE = 0.31 %), and high correlation coefficient of 0.999 (r = 0.999). Furthermore, the results of this work using DNN have also been compared with Parrish’s correlation. The findings of this work demonstrate that the DNN model exhibits better results over Parrish’s correlation in predicting the activity coefficients of water in a binary triethylene glycol-water system with a mean absolute error of 5.03 percent (MAE = 5.03 %).
Ključne riječi
Triethylene glycol; water; diffusion neural network; activity coefficient; gas dehydration
Hrčak ID:
55023
URI
Datum izdavanja:
1.7.2010.
Posjeta: 1.458 *