Predicción mediante rna de la concentración de co2 y h2s del proceso de endulzamiento de gas natural

Abstract

The aim of this Curriculum Integration Work is to predict by means of ANN the concentrations of CO2 and H2S of the natural gas sweetening process, for which the sweetening process was simulated in the free software DWSIM, the same one that served us to generate the base of essential data for the design of the artificial neural network (ANN) suitable for the prediction of the molar fractions of interest of the natural gas that enters the process. For the ANN development, the database generated from DWSIM was used, which is composed of 130 data with three inputs: temperature, pressure and molar flow of the sour gas, and their respective outputs: molar fraction of CO2 and H2S of the sweet gas flow exiting the top of the absorption tower. The artificial neural network was designed in MATLAB using 170 hidden neurons in its architecture and trained with the Bayesian Regularization algorithm; where a mean square error (MSE) of 1.6317e-05 and a total regression coefficient of 0.99882 were obtained. The ANN was validated by means of a comparative statistical analysis, obtaining a confidence percentage of 95%. The simulation was able to reduce the concentrations of CO2 and H2S to molar fractions of 0.0019 and 0.00138, respectively, and regenerated a molar fraction of 0.11081967 of the MDEA amine used. It is recommended that for the design and training of the ANN the data to be used for inputs and outputs be normalized, that is, that the network works in a range of 0-1 and not with very large values, in order to make the prediction of the molar fractions of CO2 and H2S more precise and reliable.