ISSN print edition: 0366-6352 ISSN electronic edition: 1336-9075 Registr. No.: MK SR 9/7 Published monthly

Development of supervised machine learning model for prediction of TEG regeneration performance in natural gas dehydration

Amin Hedayati Moghaddam and Abdellatif Mohammad Sadeq

Department of Chemical Engineering, Central Tehran Branch, Islamic Azad University, Tehran, Iran

E-mail: ami.hedayati_moghaddam@iauctb.ac.ir

Received: 26 July 2023  Accepted: 25 September 2023

Abstract:

In this study, triethylene glycol (TEG) regeneration process, which is a critical step in natural gas (NG) dehydration, was investigated. Machine learning (ML) approach was used to develop robust models that could assess the impacts of operative variables on TEG regeneration. A supervised multilayer feed-forward neural network was employed to develop the models, and the k-fold cross-validation technique was used during the training phase. The impacts of TEG flowrate, pressure of distillation column, and temperature of reboiler on energy consumption and TEG purity were investigated. The optimal conditions for TEG regeneration was found using a genetic algorithm (GA) based on the developed models. The R² values of test dataset were 0.9998 and 0.9989 for TEG purity and reboiler duty, respectively, demonstrating the reliability of optimally tuned models. Overall, this study sheds light on the factors that affect TEG regeneration and provides a useful framework for optimizing the NG dehydration process.

Keywords: Natural Gas; Dehydration; TEG regeneration; Machine learning; Genetic algorithm

Full paper is available at www.springerlink.com.

DOI: 10.1007/s11696-023-03113-7

Chemical Papers 78 (1) 587–597 (2024)