MODELLING OF CO2 SOLUBILITY IN DIETHANOLAMINE, NMETHYLDIETHANOLAMINE AND THEIR MIXTURES USING ARTIFICIAL NEURAL NETWORK

Natural gas has a wide range of acid gas concentrations, from parts per million to 50
volume percent and higher, depending on the nature of the rock formation from which it comes.
Because of the corrosiveness of H2S and CO2 in the presence of water and because of the toxicity
of H2S and the lack of heating value of CO2, sales gas is required to be sweetened to contain no
more than a quarter grain H2S per 100 standard cubic feet (4 parts per million) and to have a
heating value of no less than 920 to 980 Btu/SCF, depending on the contract. The most widely
used processes to sweeten natural gas are those using the alkanolamines, and of the
alkanolamines the two most common are n-methyldiethanolamine (MDEA) and diethanolamine
(DEA).
In this research, data from Khalid Osman et al (2012), A. Benamor et al (2005) and
Zhang et al (2002) will be used to simulate the solubility of CO2 in MDEA + DEA aqueous
solution using ANN model and the performance will be compared to show which model is better
for CO2 absorption. Besides, the study of CO2 solubility in MDEA and DEA aqueous solution
respectively will be using data from Jou et al (1982) and Lee et al (1972) works and simulation
of ANN model was used to compare the performance between ANN model and the reference
research works mentioned earlier.
Developed model has an absolute relative deviation (δAAD) of 8.71% while δAAD for data
from Khalid Osman et al (2012), A. Benamor et al (2005) and Zhang et al (2002) are 17.06%,
12.09% and 9.82% respectively. In terms of pure amine prediction, ANN model of CO2
solubility predicted in pure MDEA has δAAD of 8.29% while the reference paper which is A.
Benamor et al (2005) has absolute relative deviation of 10.76%. For prediction in pure DEA, the
model has δAAD of 3.33% compared to reference paper which is also from A. Benamor et al
(2005) with 4.72%.
ANN has great ability to predict CO2 solubility in pure MDEA, DEA, and their mixtures
only by developing models for each situation and condition due to the limitation of ANN itself
which cannot simulate the new input data if they do not have same patterns with the one that has
been used to develop the model.