Application of machine learning algorithms to predict removal efficiency in treating produced water via gas hydrate-based desalination

Abstract

The integration of machine learning (ML) with gas hydrate-based desalination (GHBD) presents a significant advancement in the produced water treatment with special focus on efficient prediction of removal efficiency. GHBD operates by forming gas hydrates under controlled thermodynamic conditions, selectively encapsulating gas within water molecules while excluding dissolved ions. However, the stochastic nature of hydrate formation, is influenced by gas composition, temperature, pressure, and ion concentration, makes it difficult to predict accurately removal efficiency. In this context. ML algorithms provide powerful data driven means to model complex relationship within experimental datasets to improve process optimisation This study systematically evaluated several supervised ML models, including Random Forest (RF) Support Vector Machines (SVM), Ridge Regression, Lasso Regression, Decision Tree, Extra Tree Regression, Gradient Boost, and XGBoost, to predict removal efficiency in GHBD system. Among these, the SVM model showed the best predictive accuracy, R² of 0.98, with the lowest AIC (56.75), RMSE (1.50), and MAE (1.22) values, highlighting its robustness in capturing the intricate dependencies between operational parameters and removal performance. Additionally, graphical analysis confirmed that the predictive accuracy of SVM model is superior, compared to other models. Furthermore, sensitivity analyses validated SVM's robustness in capturing the nonlinear relationships governing ion removal efficiency. These findings demonstrate that integration of ML with GHBD significantly improved predictive capabilities, enabled real time application, reduce experimental effort, as well as improve the development of intelligent, sustainable, and scalable water treatment technology.