The Study On Optimization Algorithm Of Prestack Full Waveform Inversion

Full waveform inversion(FWI)is a technique that aims to find physical property of Earth's subsurface,which is based on the wave-equation modeling method to get modeling record,and matches modeling record with raw wave-field dataset,and then uses the non-constrained iterative local optimization methods to get high resolution result.Compared with traditional inversion method,FWI has higher resolution,because it uses all amplitude and phase signal information of raw wave-field data.FWI is widely concerned by many scholars because of that the theoretical framework is easy to extend and it can get highest resolution of inversion result.In this paper,it begins with the wave-equation modeling,and then deduces the finite difference equation of modeling in time domain and frequency domain,establishes the objective function of wave-field error,and deduces the gradient equation of full waveform inversion object function,and then uses the formula of L-BFGS algorithm based on the time domain and frequency domain FWI.In order to solve the problem in the theory of FWI,this paper proposes some optimization algorithm: as to the low efficiency forward modeling in frequency domain,using the director solver and iterative method with pre-condition to improve the efficiency,as to the large amount of data to calculate in time domain FWI,proposing the CPU/GPU heterogeneous parallel framework,using the high performance computer to do the parallel computation.And as to the large amount of date to input and output(I/O),presenting a new method for the time-domain FWI's gradient to improve the efficiency and the new algorithm is named the excitation main energy gradient equation.In the last,this paper analysis the difficulties of field seismic data inversion and applies the new method to invert the theoretical model and actual data.