Prediction of variables involved in TEG Dehydration using hybrid models based on boosting algorithms

Abstract

The extraction of gas from fields often involves impurities, necessitating natural gas processing to separate these impurities. This process typically entails the removal of acid gases (such as carbon dioxide and hydrogen sulfide) and dehydration, commonly achieved through absorption using triethylene glycol (TEG). Efforts to minimize BTEX emissions and maintain optimal dry gas water content are pivotal for enhancing the economic and environmental sustainability of natural gas processing. In this study, the accurate prediction of BTEX and dry gas water contents is aimed by boosting-based methods and hybridization techniques. Four hybrid models, including XGBoost, CatBoost, HGBR, and LightGBoost, were developed in conjunction with the Arithmetic Optimization Algorithm (AOA) meta-heuristic algorithm to optimize and fine-tune hyperparameters. Through a comprehensive case study and comparison of various evaluation indices, the XGBoost-AOA hybrid model emerged as the most accurate in predicting both outputs. Hence, we recommend the XGBoost algorithm optimized with the AOA algorithm as the preferred approach for predicting BTEX and dry gas water contents in natural gas processing.