The Research Of Reflection Full Waveform Inversion

The traditional seismic data processing flow usually includes: modeling the velocity by travel time and waveform tomography,to provide an accurate velocity field for subsequent processing,and performing the imaging based on the known background velocity field in order to accurately describing the main underground layer.Accurate velocity field modeling is the premise and guarantee for high-precision imaging.With the increasingly complex exploration conditions,higher requirements are also put forward for the velocity field modeling method.Full waveform inversion(FWI)can theoretically make full use of effective information such as amplitude,phase,travel time and waveform contained in the seismic data and has made great progress in different domains and different media.In the global inversion of the structure and exploration geophysics,FWI has achieved good results.However,the full waveform inversion is a very strong nonlinear problem.One of the most effective solutions is to iteratively obtain the optimal solution through the local optimization method.In this case,an initial model that is very close to the global optimal solution is required.The conventional full waveform inversion method lacks the ability to recover the medium and long wavelength components,so the initial velocity field is required to provide accurate long-wavelength information.In recent years,the rise of the reflected wave full waveform inversion played an increasingly important role in terms of recovering the long and middle wavelength component,providing an accurate initial model for conventional FWI and migration.Reflected Waveform Inversion(RFWI)provides a more accurate initial velocity field for traditional full waveform inversion(FWI)by alternately updating the low and high wavenumber components of the model.Due to the lack of low-frequency information and the complexity of the amplitude information,the reflected waveforms inversion is also a highly nonlinear problem.The separation of amplitude and phase information can effectively improve the linearity of RFWI.However,the traditional phase calculation method has serious problem of phase wrapping.In this paper,we first give a method to calculate the more stable quasi-phase information,using the envelope information of seismic data,which is capable of avoiding the problem of the traditional phase warping issue.Then,the objective function is constructed using the information of the quasi-phase,and the corresponding ad-joint source and gradient are given by the ad-joint method.Theoretical analysis proves that the proposed method can ensure the stability and accuracy of the phase information,and the misfit function based on the quasi-phase information has a wider global convergence domain.In order to effectively reduce the nonlinear of full waveform inversion of reflected waves,an initial model can be established using the travel time information in the seismic data,based on the idea of inversion of seismic information subsets.Traveling-time inversion method based on wave equation can get the travel time information through cross-correlation method,Wave equation reflection travel time inversion based on DIW which utilizes the traveltime information is an efficient tool to build the initial model for conventional FWI.However,instability can be observed when the wavelets of de-migrated data and observed data are significantly different.This paper introduces a double difference wave-equation reflection travel time inversion(DWERTI)method and modifies the conventional DIW method with a regularization term to improve its performance.This paper discusses in detail the limitations of traditional full waveform inversion,such as the lack of low frequency data,the inaccurate initial model and the limited offset distance,and proves the advantage of the reflection waveform inversion by the comparison of kernel function.At the same time,aiming at the nonlinear problem of the waveform inversion of the reflected wave,the optimization of RFWI is carried out by modifying the misfit function and using the travel time information.The proposed methods are applied to the synthetic data to verify the efficiency and accuracy.Finally we make a discussion and draw some conclusions based on the research.