Machine-Learning-Based Prediction of Gas Hydrate Dynamics: A Comparison with a Fundamental Model against Experimental Data

Gas hydrate has emerged as a potential source of energy (i.e., natural gas) and is one of the prime factors in climate change. In this work, we formulate the complex formation and growth dynamics of gas hydrates occurring in pure and saline water with/without porous media. For this, four different modeling approaches under the framework of machine learning (ML), random forest (RF), gradient boosting (GB), extra trees (ET), and decision tree (DT), have been formulated by using a bank of 560 data sets. The ML algorithms are developed with the physical understanding of hydrate formation with (i) nth-order phase transformation, (ii) surface renewal, and (iii) the temperature-dependent (Arrhenius type) kinetic factor, and are employed to optimize the hyperparameter and thermokinetic parameters related to these practically relevant issues. To investigate their satisfactory predictability at a wide variety of geological conditions, we further move to compare them with the latest thermokinetic model that considers the hydrate formation at interstitial space between uneven porous particles and inside their nanometer-sized pores, along with the above-mentioned practical issues. It is examined that the simple ML approaches show competitive performance with the fundamental model to predict the formation and growth dynamics of clathrate hydrates under various reservoir-mimicking conditions.