Simulación y validación de un sistema de fraccionamiento para la separación de los componentes del crudo en procesos de refinación

Abstract

ABSTRACT The objective of this research work was the simulation and validation of the crude oil fractionation unit, which was used as the basis for the creation of an artificial neural network (ANN) capable of predicting the fractions of the products obtained after crude oil distillation. In the creation of the ANN, a data bank of 150 values was used for each variable: crude oil flow, furnace temperature, light naphtha, heavy naphtha, kerosene, diesel oil and fuel oil; obtained from the simulation created. The ANN architecture is divided into: an input layer (2 neurons), three hidden layers (50 neurons) and an output layer (6 neurons); this architecture is based on the Feed-forward backprop network type and has a training algorithm based on Levenberg-Marquardt (TRAINLM). In this network, a mean square error of 0.0016801 was obtained with a correlation level of 0.9954; validating it by means of comparative statistical tests such as the Kruskal-Wallis test and the Friedman test where the real and predicted data were compared, taking as a rule that the test values obtained should not exceed the value of 3.8415 to accept the null hypothesis, resulting in that for all the products the null hypothesis that dictates the existence of a similarity between these two groups of values is accepted. It is advisable to raise the range of scope of ANN as it can be applied to new tools and research in crude oil refining processes, ensuring greater efficiencies in these processes.