Predicting semiclathrate hydrates dissociation pressure using a rigorous machine learning approach

Abstract

In this study, a supervised learning algorithm known as multi-layer perceptron neural network (MLP-NN) is employed to predict the dissociation pressure of semiclathrate hydrates of CO₂+N₂, CH₄+N₂, CO₂+CH₄, CO₂, CH₄, and N₂ systems in the presence of different concentrations of TBAB. The hydrate dissociation pressure was assumed as a function of the molar concentrations of the gases, the weight percent of TBAB tetra n-butyl ammonium bromide (TBAB) and the system temperature. A data set of 349 data points gathered from the open literature and divided into 70%, 15%, and 15% ratio for three subsets of training, validation and testing, respectively. The optimum network structure was selected through a trial and error procedure with the overall R² of 0.97, MSE of 0.16, and AARD% of 6.19%. While both graphical and statistical measures approved the robustness and performance of the network, the applicability of the model and quality of experimental data was investigated through Leverage approach.

Graphical Abstract

Keywords:

• Semiclathrate hydrates
• gas hydrates
• TBAB
• artificial neural network