Simulación y validación de un sistema de destilación para la separación de azeótropos de co2-etano en procesos mejorados de recuperación de petróleo

Abstract

The purpose of this research was to replicate and validate an alternative extractive distillation system for CO2-ethane azeotrope separation during the natural gas treatment process, which provides a vital basis for the design of an artificial neural network (ANN) which can predict the mole fractions of the main products obtained. For the development of the ANN, a database was generated from a simulation in the free software DWSIM. The sample consists of 130 data with three inputs: pressure, temperature, and solvent/feed ratio, and their corresponding six outputs: the molar fraction of distilled CO2 and residual ethane in the extraction column, the molar fraction of distilled ethane and residual propane in the solvent recovery column, and the molar fraction of distilled ethane and residual ethane in the concentrator column. The network was designed in MATLAB using 80 hidden neurons in its architecture and the Bayesian regularization algorithm for training. A mean squared error (MSE) value of 0.0036 and a total regression coefficient of 0.95546 were obtained. The network was validated employing a comparative statistical analysis obtaining 95% reliability. The simulation allowed the removal of 95.6% of CO2 present in natural gas and 91.56% of ethane was recovered. Moreover, part of the Natural Gas Liquids (LGN) from the recovery column was used as a solvent without the need to resort to an external one. This proposal proved to be more efficient than the conventional model reducing 10% of total annual and operating costs without compromising the desired purification and reducing energy consumption. It is recommended to extend the ANN learning degree with new input and output parameters so that it becomes a more didactic complete computational tool and serves as a case study of various chemical and/or industrial processes.