Evaluation of phase equilibrium conditions of clathrate hydrates in natural gas binary mixtures: Machine learning approach

Hydrate formation temperature (T) is an important parameter for any industrial process that deals with natural gas hydrates. In this study, the Group Method of Data Handling (GMDH) approach is used to predict hydrate formation Tin natural gas binary mixtures. A comprehensive database containing 728 data samples is compiled from 46 published experimental works. To find the best combination of input variables, different sets of input variables were assessed. A total of seven models were developed using different sets of input variables. Compared to the correlations proposed in the literature, the developed models in this study performed better and the model developed based on input Set #7 was the most accurate: RMSE values of 1.6381 and 1.5499 for the training and testing datasets, respectively. All models also were evaluated using a blind dataset-that was not included in testing or training-to check model applicability to wider data. Similarly, all GMDH models performed excellently for the external dataset, where the developed model based on input Set #5 showed the best performance: RMSE values of 1.3482. The findings of this study contribute to our understanding of hydrate formation conditions in natural gas binary mixtures. As pure and binary mixtures of natural gas main constituents are studied, the results can be especially useful for purification and energy transport applications.