Prediction of methane hydrate formation conditions in salt water using machine learning algorithms

Abstract Predicting formation conditions of gas hydrates in salt water is important for the management of hydrate in processes such as flow assurance, deep-water drilling, and hydrate-based technology development. This paper applied and compared five machine learning algorithms to develop prediction tools for the estimation of methane hydrate formation temperature in the presence of salt water. These machine learning algorithms are Multiple Linear Regression, k-Nearest Neighbor, Support Vector Regression, Random Forest, and Gradient Boosting Regression. In total, 702 experimental data points in literature from 1951 to 2020 were collected for modeling purposes. The experimental data span salt concentrations up to 29.2 wt% and pressures up to 200 MPa. Among these five machine learning methods, Gradient Boosting Regression gives the best prediction with R 2 = 0.998 and AARD = 0.074%. Thus, the methods of Gradient Boosting Regression function as an accurate tool for predicting the formation conditions of methane hydrates in salt water.