Intelligent frequency-dependent nonlinear seismic 3D inversion for fluid-matrix decoupled fluid factor based on scattering theory

Abstract Fluid identification is essential for geophysical exploration. Although the Zoeppritz equation and its linear approximations underpin seismic inversion, the exact equation involves complex calculations, and linear approximations are limited to small angles. Wave-induced fluid flow introduces frequency-dependent reflection behaviour, necessitating higher-order scattering formulations and viscoelastic effects for accurate characterization. Moreover, purely data-driven deep learning approaches are increasingly inadequate for complex exploration scenarios. To overcome these challenges, this study proposes an intelligent frequency-dependent nonlinear seismic 3D inversion method for fluid-matrix decoupled fluid factor based on scattering theory. First, we derive a novel second-order reflection formulation with explicit frequency dependence to characterize fluid effects. This equation accurately captures viscoelastic dispersion and remains valid at large angles beyond conventional linear approximations. Second, we integrate the derived physics-based formulation and convolutional model into the proposed workflow of physics-constrained GRU-attention network to enable frequency-dependent nonlinear inversion with improved physical interpretability. Application to model test and real field data demonstrates that the proposed method can reliably identify fluids, outperforming conventional linear and purely data-driven approaches.