Modeling of nitrogen solubility in unsaturated, cyclic, and aromatic hydrocarbons: Deep learning methods and SAFT equation of state

Background: An accurate understanding of the nitrogen solubility in hydrocarbons is crucial for developing enhanced oil recovery (EOR) by nitrogen injection and the design of thermal separation and chemical conversion processes in chemical industries and oil refineries. Unsaturated, cyclic, and aromatic hydrocarbons as important compounds of crude oil are common solvents in the chemical industry that have received less attention compared to normal alkanes. Methods: In this paper, four deep learning models, including recurrent neural network (RNN), long shortterm memory (LSTM), deep belief network (DBN), and convolutional neural network (CNN) were developed for estimating the nitrogen solubility in unsaturated, cyclic, and aromatic hydrocarbons. To this end, 673 experimental nitrogen solubility data for 23 various hydrocarbons were collected from the literature in a wide ranges of operating pressure (0.03 similar to 100.1 MPa) and temperature (78 similar to 662.8 K). The input parameters to the models were considered critical temperature, critical pressure, and molecular weight of hydrocarbon solvents accompanying operating conditions of temperature and pressure. Also, the performance of deep learning models was compared with Soave-Redlich-Kwong (SRK), Peng-Robinson (PR), and statistical associating fluid theory (SAFT) equations of state (EOSs). Significant findings: The CNN model is able to estimate the experimental values of nitrogen solubility with a root mean square error (RMSE) of 0.0211. Also, SAFT EOS outperformed the two cubic EOSs. Based on sensitivity analysis, pressure has the greatest impact on nitrogen solubility in unsaturated, cyclic, and aromatic hydrocarbons, followed by temperature and the critical temperature of hydrocarbons. Nitrogen solubility in unsaturated, cyclic, and aromatic hydrocarbons increases with the increase in pressure and temperature, and it lowers by the increase in the critical pressure, critical temperature, and molecular weight of hydrocarbon solvents. Finally, high reliability of the experimental data and statistically high validity of the CNN model were proved by the Leverage approach. The findings of this study can have implications in increasing the efficiency of gas processing units, thermal separation, and chemical conversion processes. (C) 2021 Published by Elsevier B.V. on behalf of Taiwan Institute of Chemical Engineers.

Automated summary

This study developed four deep learning models to estimate the nitrogen solubility in unsaturated, cyclic, and aromatic hydrocarbons, revealing the diminishing impact of pressure, temperature, and the critical temperature of hydrocarbons on nitrogen solubility. The findings demonstrated the excellent performance of the CNN model in estimating nitrogen solubility, contributing to the efficiency improvement of gas processing units, thermal separation, and chemical conversion processes.