A time-domain multi-parameter elastic full waveform inversion with pseudo-Hessian preconditioning

Based on waveform fitting, full waveform inversion (FWI) is an important inversion method with the ability to reconstruct multi-parameter models in high precision. However, the strong nonlinear equation used in FWI presents the following challenges, such as low convergence efficiency, high dependence on the initial model, and the energy imbalance in deep region of the inverted model. To solve these inherent problems, we develop a time-domain elastic FWI method based on gradient preconditioning with the following details: (1) the limited memory Broyden Fletcher Goldfarb Shanno method with faster convergence is adopted to improve the inversion stability; (2) a multi-scaled inversion strategy is used to alleviate the nonlinear inversion instead of falling into the local minimum; (3) in addition, the pseudo-Hessian preconditioned illumination operator is involved for preconditioning the parameter gradients to improve the illumination equilibrium degree of deep structures. Based on the programming implementation of the new method, a deep depression model with five diffractors is used for testing. Compared with the conventional elastic FWI method, the technique proposed by this study has better effectiveness and accuracy on the inversion effect and con-vergence, respectively.