Wavefield Reconstruction Inversion via Physics-Informed Neural Networks

Wavefield reconstruction inversion (WRI) formulates a PDE-constrained optimization problem to reduce cycle skipping in full-waveform (FWI). WRI often requires expensive matrix inversions reconstruct frequency-domain wavefields. Physics-informed neural network (PINN) uses the underlying physical laws as loss functions train (NN), and it has shown its effectiveness solving Helmholtz equation generating Green's functions, specifically for scattered wavefield. By including data-constrained term function, trained NN can wavefield that simultaneously fits recorded data satisfies given initial velocity model. Using predicted wavefields, we rely on small-size predict using reconstructed In this prediction NN, spatial coordinates are used input is define function. After network, able domain of interest. We develop PINN-based method demonstrate potential part Sigsbee2A model modified Marmousi The results show invert reasonable with very limited iterations frequencies, which be subsequent FWI application.