Time-domain Elastic Full Waveform Inversion With GPU Parallel Computing

As an effective method,full waveform inversion with a high resolution,high signal ratio and high fidelity result can take advantage of seismic wave information.At the same time,the potential advantages of full waveform inversion are more evident than other methods when applied to the underground media with complex structure and lithology.Elastic wave equation inversion in the time domain,as a highly nonlinear inversion method,obtains the parameters of subsurface medium by calculating the minimum of the objective function which is built by the residuals between the observed and simulated seismic data.This paper starts with the reserch status of full waveform inversion investigating the finite difference method and the perfect matching layer(PML)boundary conditions for acoustic wave equation and elastic wave equation.Then,we use Lagrange multiplier to derive the adjoint equation of the first-order velocity-stress elastic wave equation.This paper chooses the conjugate gradient method to optimize the objective function.In order to overcome the objective function falling into local minimum,we apply multiscale inversion to our algorithm,and we obtain satisfactory results.Finally,this paper researches on GPU parallel computing based on CUDA and the efficient boundary storage strategy to enhance the efficiency of the full waveform inversion.