Minireview and Outlook on Integrated Inversion of Sediment Heterogeneity and Geomechanical Stability during Gas Hydrate Decomposition

Abstract

Natural gas hydrate is a novel clean energy source with high development potential. It occurs as solid deposits within the pores of unconsolidated seabed sediments. Hydrates occupy fluid flow pathways, provide skeletal support and natural cementation between particles, and thus govern the seepage and mechanical properties of host sediments. Exploitation disturbs hydrate phase equilibrium, inducing decomposition into gaseous methane and liquid water; methane is then produced via recovery wells. This process involves multifield coupling of thermal (T)-hydraulic (H)-mechanical (M)-chemical (C) processes. Integrated inversion of sediment heterogeneity and geomechanical stability during hydrate decomposition represents the core scientific issue restricting safe exploitation. High-dimensional heterogeneous parameters cannot be precisely retrieved via conventional numerical and experimental approaches, while deep learning is recognized as a viable alternative. The heterogeneous evolution of seepage and mechanical parameters in hydrate-bearing sediments is systematically reviewed herein, together with inversion methodologies and optimization strategies for physics-based and data-driven models. Existing inversion technologies are classified into three mainstream categories: pure data-driven inversion, single physics-guided inversion, and coupled physics-data hybrid inversion. Their merits, computational overheads, and intrinsic defects for hydrate THMC coupled systems are quantitatively compared. Three typical physics-informed neural network variants among hybrid frameworks are elaborated, and training drawbacks of conventional soft-constraint PINNs for marine hydrate reservoirs are clarified. Two critical bottlenecks are identified, namely insufficient precision in coupled heterogeneity-stability inversion and the scarcity of mature physics-data hybrid architectures. Three targeted solutions are proposed, including refined reservoir heterogeneity characterization, multisource data fusion inversion construction, and hybrid PINN model development. Theoretical references and technical guidance are supplied for safe and efficient natural gas hydrate exploitation.