Time-shift Multiscale Full Waveform Inversion Based On Low-frequency Reconstruction And Research On Source Encoding Migration

Nowadays,with the development of seismic exploration,the target area is gradually developing towards deeper and more complex structures.The velocity model construc-tion and imaging are the key processes of the overall seismic data processing.They are closely related.High-precision migration imaging depends on high-precision velocity model.Full waveform inversion is a potential way to construct an accurate velocity model by minimizing the residuals of the synthetic data and the observed data.Local optimization algorithms are often used to solve the minimization problem to recover the underground velocity model.However,due to the problem of cycle skipping caused by the poor initial velocity model,applying full waveform inversion in field examples remains a challenging task.To deal with the problem of cycle-skipping,starting from the multiscale method,this paper introduces the time-shift multiscale method based on the time-shift imaging conditions.Different from the traditional method of directly preprocessing the data,the time-shift method extracts the low wavenumber components in the gradient by stacking the weighted time-shift cross-correlations,and the overall workflow is more natural.However,for the low efficiency of the gradient calculation of the time-shift method,an improved high-efficiency solution is also introduced,which can greatly improve the efficiency of the method while ensuring the extraction of low wavenumber components.Multiple numerical experiments also prove the reliability of the method.As a result of the time shift method is hard to deal with the problem which missing the low-frequency data,we introduce two methods to construct the low-frequency data.One solution is based on the sparse inversion of underground reflection coefficients for low-frequency data construction.Otherwise,combined with time-shift multiscale methods,a new full waveform inversion process to deal with the problem of missing the low-frequency data is proposed.The second solution is constructing the low-frequency data based on the sliding time window down-sampling method.The low-frequency data and the high-frequency data are connected through the optimal sampling points in the time window,and the low-frequency data is constructed by re-interpolation of multiple sampling points.Several examples in related chapters also illustrate the reliability of the first method for low-frequency construction and the effectiveness of the full waveform inversion.The second method is better for the recovery of the shallow and the middle background velocity model.However,due to some amplitude and phase errors in the reconstructed data,the constraint for the deep background velocity is poor.At the end of this paper,we also addressed the problem of low efficiency of the convolutional reverse time migration.We introduced the multi-source RTM,which can reduce the number of migrations of reverse-time migration by stacking multiple sources as a synchronized source.However,the multi-source strategy has a disadvan-tage in that crosstalk artefacts arise from interactions between different sources in the cross-correlation imaging condition.To deal with the problem,we introduced the super virtual shot(SVS)encoding method.This scheme modulates the wavefields of multiple sources to fit the wavefield of a suspended SVS,which can eliminate crosstalk artefacts as they are absent in single super virtual shot migration.We deduced the encoding func-tions of the method and the related parameter selection of the method through detailed theory.Based on three numerical experiments,the reliability,error introduction and high efficiency of the method are illustrated.