Optuna-LightGBM : An Optuna hyperparameter optimization framework for the determination of solvent components in acid gas removal unit using LightGBM

Abstract

Acid gas removal unit (AGRU) serves as an essential process in gas treatment, specifically designed to eliminate acid gases like hydrogen sulfide (H${}_2$S) and carbon dioxide (CO${}_2$) from natural gas. Absorption-based AGRU are extensively employing chemical solvents because of their strong performance and effectiveness. Nonetheless, using various solvents with unique properties significantly affects AGRU performance. The determination of this solvent component primarily relies on experimental or simulation-based trial and error, with minimal research dedicated to classification methods aimed at identifying the optimal solvent under specific conditions. In addressing the research problem, this study systematically evaluated several supervised machine learning models, including Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Decision Tree, and Artificial Neural Network (ANN), to identify the optimal solvent component from a selection of six different solvents used in AGRU. Data were gathered from the verified flowsheet utilizing the Aspen HYSYS software. The results indicate that LightGBM algorithms surpass the performances of all algorithms, with accuracy (98.4%) and training time (0.7 s). Hyperparameter optimization using Optuna was employed to increase the performance of the LightGBM model, with an increment of 0.4% and a training time reduction of over 50%. Additionally, hyperparameter importance and sensitivity analysis confirmed that the number of boosting rounds and CO${}_2$ composition are the key parameters affecting the Optuna-LightGBM model in predicting solvent components. These findings offer valuable perspectives for enhancing solvent components to boost AGRU efficiency in industrial settings.