Predicting three phase (hydrate–liquid–vapour) equilibria of mixed hydrates in guest gas swapping: AI-based approach versus physical modelling

Abstract

Prior to investigating the guest gas replacement characteristics, the estimation of equilibrium condition for the coexisting hydrate–liquid–vapour (HLV) phases is crucial. For this, there are various studies which have reported the physical thermodynamic model for equilibrium estimation. In this contribution, a data-driven formulation is developed as an alternative approach within the framework of artificial intelligence (AI) to predict the three-phase equilibrium of binary and ternary mixed hydrates associated with guest swapping at diverse geological conditions. For this, we use the experimental data sets related to guest (pure and mixed CO₂) replacement in hydrate structures with and without salts (i.e., single and multiple salts of NaCl, KCl, and CaCl₂). Various training algorithms, namely Levenberg–Marquardt (LM), scaled conjugate gradient (SCG), Broyden–Fletcher–Goldfarb–Shanno (BFGS) quasi-Newton, and Bayesian regularization (BR), are employed to formulate the artificial neural network (ANN) model. Performing a systematic comparison between them, we select the best option suited for the hydrate system. The best performing ANN model is compared with an existing physical thermodynamic model for predicting the equilibrium condition in pure water. It is observed that the ANN (BR) model consistently secures the lower percent absolute average relative deviation (i.e., %AARD <2%) than the latest physical model. Finally, the developed AI model is extended to predict the three-phase HLV equilibrium in presence of salt solutions.