Deep Learning of Evolution of Hydrate Dissociation Front during Gas Production Considering THCM Parameters of Hydrate Reservoirs

Abstract

Natural gas hydrate has the potential to be an alternative source of energy, as demonstrated by several recent field trials. Its commercial viability depends on the long-term operational efficiency of gas production from these reservoirs, which requires a sustained rate of dissociation of the hydrate with time. Gas production from hydrate reservoirs involves coupled thermo-hydro-chemo-mechanical (THCM) processes, and accurate predictions of such rely on numerical simulations with fine spatial and temporal discretization. Meanwhile, the input parameters required for these simulations often involve inherent variability arising from site conditions. These make it difficult to determine the optimal production strategy, including the depth of the production zone and the drawdown pressure, as this entails numerous time-consuming simulations. This study proposes a deep learning (DL)-based approach to substantially reduce the computational demands in determining the evolution of hydrate saturation throughout the operation, which is one of the key spatiotemporal variables that influence production efficiency. The eXtreme Gradient Boosting (XGBoost) model, which is a tree structure-based DL model, is adopted to learn the correlations between the diverse variations of input parameters and the resulting temporal changes in locations of hydrate dissociation fronts. The XGBoost model is significantly more efficient than the coupled THCM numerical simulator and showed excellent performance in predicting the dissociation fronts with varying degrees of dissociation, from 20%, 50% to 80%. Furthermore, the model identifies five key parameters that influence the evolution of the dissociation fronts. These parameters are in situ hydrate saturation, temperature, absolute permeability, effective permeability, and the rate of depressurization. The first four represent site-specific conditions that can be determined through site investigation, while the fifth is an operational control that can be adjusted to achieve the desired hydrate dissociation outcomes and, hence, gas production efficiency.