Predicción de la composición de syngas mediante redes neuronales a partir de la simulación de su proceso de producción

Abstract

Syngas composition using artificial neural networks (RNA) was predicted from the simulation of their production process. It simulated the process of obtaining syngas in the DWSIM software, where we obtained all the data to design the RNA, the simulation was used to obtain a percentage error of less than 4%. The RNA was designed in the MATLAB software, using the temperature, pressure, and mole fractions of CH4, CO2, and N2 as input parameters. The RNA architecture is a multilayer feedforward type and was trained with the Levenberg-Marquardt (LM) algorithm, obtaining an RNA with 7 input layers, 20 hidden layers, and 4 output layers. The data predicted by the RNA were validated using the performance indicators, mean squared error (MSE), and the correlation coefficient (R), and analyzed with the ANOVA test. The prediction results indicated that RNA is highly successful in predicting the composition of syngas. The RNA predicted values similar to those obtained in the simulation, indicating an R equal to 0.99767 and an MSE equal to 0.00088232, validating its design of the same based on the minimum resulting error. It is recommended to use this technological proposal, as a guide, to predict the composition of natural gas, obtained in the oil extraction process in Ecuador.